Morphosyntactic Expression in Functional Grammar 9783110920833, 9783110183658

Citation preview

Morphosyntactic Expression in Functional Grammar

W G DE

Functional Grammar Series 27

Editors

Casper de Groot J. Lachlan Mackenzie

Mouton de Gruyter Berlin · New York

Morphosyntactic Expression in Functional Grammar

edited by

Casper de Groot Kees Hengeveld

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter G m b H & Co. K G , 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

Morphosyntactic expression in functional grammar / edited by Casper de Groot, Kees Hengeveld. p. cm. — (Functional grammar series ; 27) Includes bibliographical references and index. ISBN 3-11-018365-X (hardcover : alk. paper) 1. Grammar, Comparative and general — Morphosyntax. 2. Functionalism (Linguistics). I. Groot, C. de (Casper) II. Hengeveld, Kees, 1 9 5 7 III. Series. P290.M67 2005 415'.9—dc22 2005001606

Bibliographic information published by Die Deutsche

Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at .

ISBN 3-11-018365-X © Copyright 2005 by Walter de Gruyter G m b H & Co. K G , 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. Printed in Germany.

Preface

Over the last decade, morphological and syntactic issues have received relatively little attention in Functional Grammar. This is partly due to the fact that this grammatical model, given its functional orientation, assigns pride of place to pragmatics and semantics. A further reason is that given the developments within the theory during this period increasing attention had to be paid to discourse structure on the one hand and the lexicon on the other. Yet any fully-fledged theory of language needs a well-defined morphosyntactic component, and there is a revived awareness in the FGcommunity of the urgency of the task of developing this component of the theory in more detail. Since the original formulation of Functional Grammar in Dik (1978), a distinction has generally been made between three sets of expression rules: one set responsible for the form of linguistic units, one for the order of linguistic units, and one for the prosodic features of linguistic units. The expression rule component, which contains these three sets of expression rules, takes abstract pragmatic and semantic underlying clause structures as its input, and converts these into actual linguistic expressions. Much of the work on expression rules in Functional Grammar has therefore been dedicated to finding out the exact relationship between underlying pragmatic and semantic categories on the one hand, and the syntactic, morphological, and phonological manifestations of these categories on the other, in a wide variety of languages. Several of the basic assumptions concerning the organization of the expression rule component in Functional Grammar have recently been challenged. First of all, in a number of publications, Bakker (1999, 2001) emphasizes the problems posed to the approach summarized above by the complex interactions that one finds in many languages between the sets of expression rules that have to account for form on the one hand and those that establish order on the other. He proposes to deal with these interactions in a radically different integrated implementation of the expression rule component in FG, called the Dynamic Expression Model. Secondly, in Functional Discourse Grammar (Hengeveld 2003a, 2003b; Mackenzie and Gomez Gonzalez eds. 2003; Hengeveld and Mackenzie fc.) a further step is taken by considering morphosyntactic and phonological representations to

vi

Preface

be part of the underlying structure of the grammar rather than as the output of that grammar. In view of the issues mentioned in the preceding paragraphs, this book contains three groups of papers. The first two articles further develop the idea of the Dynamic Expression Model in general terms. Dik Bakker's paper concentrates on agreement phenomena, John Connolly's on word order. The next five articles look at morphosyntactic expression in Functional Discourse Grammar. Kees Hengeveld shows how the idea of a dynamic implementation may be applied within Functional Discourse Grammar. Niels Smit proposes a way to handle noun incorporation within this model. Casper de Groot stresses the importance of a separate morphosyntactic level of analysis, and Francis Cornish gives a detailed analysis of the phenomenon of locative inversion. In their analysis of agreement and crossreference Anna Siewierska and Dik Bakker explore the possibility of using Functional Discourse Grammar as a model of the speaker. The remaining articles provide detailed analyses of a range of semantic and pragmatic categories and their morphosyntactic expression in a wide variety of languages. Two papers concentrate on non-verbal predication: Eva H. van Lier shows that typological hierarchies uncovered in crosslinguistic research using Functional Grammar can be applied equally well to developmental language data. Angel Herrero-Blanco and Ventura Salazar-Garcia investigate the appropriateness of the copula support rule in Functional Grammar for the analysis of Spanish Sign Language. The next two papers deal with operators of various types: Annerieke Boland provides a new analysis of operators of tense, aspect, and event quantification in English, while Ahmed Moutaouakil studies exclamative sentences in Arabic. The next paper is concerned with the analysis of term phrases: Evelien Keizer gives a new classification of close appositions in English using the Functional Grammar framework. The issue of syntactic, semantic, and pragmatic functions in Functional Grammar are the topic of the last four papers. Arok Wolvengrey discusses inversion in Plains Cree, and Ole Nedergaard Thomsen obviation in Mapudungun. Johan Lotterman and Lachlan Mackenzie study the reasons behind the unexpected occurrence or non-occurrence of an absolutive marker in Tanggu, and Lisbeth Falster Jakobsen the form of pronominal expressions in Danish. In all, the contributions to this volume show that the issue of morphosyntactic expression in Functional Grammar is very much alive and moving into promising new directions, while at the same time contributing to a

Preface

vii

better understanding of a large number of morphosyntactic phenomena in a wide variety of languages. We are grateful to a large number of anonymous colleagues for their willingness to act as a referee in selecting the papers for this volume. We are also very much indebted to Lachlan Mackenzie, who went through the entire manuscript and provided many detailed and helpful comments. Finally, we would like to thank Niels Smit, who very efficiently and in an extremely short time span prepared the final version of this book. March 2005

Casper de Groot Kees Hengeveld

Contents

Agreement: More arguments for the dynamic expression model Dik Bakker

1

Constituent ordering in the expression component of Functional Grammar John H. Connolly

41

Dynamic expression in Functional Discourse Grammar Kees Hengeveld

53

Noun incorporation in Functional Discourse Grammar Niels Smit

87

Morphosyntactic templates Casper de Groot

135

A crosslinguistic study of 'locative inversion': Evidence for the Functional Discourse Grammar model Francis Cornish

163

The agreement cross-reference continuum: Person marking in FG Anna Siewierska and Dik Bakker

203

The explanatory power of typological hierarchies: Developmental perspectives on non-verbal predication Eva H. van Lier

249

Non-verbal predicability and copula support rule in Spanish Sign Language Angel Herrero-Blanco and Ventura Salazar-Garcia

281

χ

Contents

A new view on the semantics and pragmatics of operators of aspect, tense and quantification Annerieke Boland

317

Exclamation: Sentence type, illocution or modality? Ahmed Moutaouakil

351

Close appositions Evelien Keizer

381

Inversion and the absence of grammatical relations in Plains Cree Arok Wolvengrey

419

Direction diathesis and obviation in Functional Grammar: The case of the inverse in Mapudungun, an indigenous language of south central Chile Ole Nedergaard Thomsen

447

Unexpected insertion or omission of an absolutive marker as an icon of a surprising turn of events in discourse Johan Lotterman and J. Lachlan Mackenzie

483

Pronominal expression rule ordering in Danish and the question of a discourse grammar Lisbeth Falster Jakobsen

503

Index of names Index of languages Index of subjects

525 529 531

Agreement: More arguments for the dynamic expression model Dik Bakker

1.

Introduction

In Bakker (1999; 2001) it is argued that the way in which the Functional Grammar expression rules (ER) are traditionally organized leads to two kinds of fundamental problems: undergeneration and overgeneration. There are a number of arguments for this position. In this first section I will briefly discuss what I think are the more important ones. For more arguments I refer to the contributions mentioned above. Undergeneration implies that certain forms that actually occur in languages cannot be produced systematically by the expression rules. Thus, the strict organization of ER into two stages, i.e. first the generation of forms and then their linearization, makes all constellations in which form is in some way dependent on order impossible sui generis. The interdependency between form and order is in fact quite common in the languages of the world, above all in the case of agreement phenomena in languages with a non-rigid constituent order. 1 1 will give four examples of this. In (1), from Basque, in prenominal relative clauses the auxiliary verb agrees with the antecedent in person and number (la); however in postnominal ones it agrees in defmiteness, number and case (lb). In (2), from Selknam, a language spoken on Tierra del Fuego, now probably extinct, the order of the Subject and Object agreement markers on the verb is the mirror image of the order of the corresponding constituents vis ä vis the verb while both orders are allowed. In (3), from Koegu, a Surmic language from Ethiopia, the verb agrees optionally with the subject when it is postverbal (3a/b), but obligatorily when it is preverbal (3c/d). In (4), from Yagua, a language from the Amazon area, there is an agreement marker on the verb in the case of a postverbal subject (4a) but not in the case of a preverbal subject (4b).

2

(1)

Dik Bakker

Basque (Saltarelli 1988:36) a. Gizon-ari haragi-a sal-du Man-SG.DAT meat-SG.ABS sell-PRF d-io-0-n harakin-ari 3.ABS-3.SG.DAT-3.SG.ERG-COMP butcher-SG.DAT hortz-ak eror-i za-izk-io. tooth-PL. ABS fall-PRF AUX-ABS.PL-3.SG.DAT 'The teeth have fallen out on the butcher that has sold the meat to the man.' b. Galtzerdi hori-ek soin-ean Stocking that-PL.ABS body-SG.LOC d-a-u-z-ka-zu-n-ak 3.ABS-PRS-have-PL.ABS-have-2.SG.ERG-COMP-PL.ABS polit-ak d-i-ra. beautiful-PL. ABS 3.ABS-PRS-be 'The stockings which you have on, are very pretty.'

(2)

Selknam (Nachlis 1973:17) a. P'ejj kah-jqe-en mer Knife M-see-F PST

na? woman

b. Na? tah-jqe-enn mer p'ejj. Woman 3.SG-see-M PST knife 'The woman saw the knife.' (3)

Koegu (Hieda 1998:367) a. A-mat-i-yaa aan. l.SG-drink-3.SG l.SG b. 0-mat-aa drink-3.SG

aan. l.SG

c. Aan a-at-i-yaa. l.SG l.SG-drink-3.SG d. *Aan 0-at-i-yaa. l.SG drink-3.SG Ί drank it.'

More arguments for the dynamic expression model

(4)

3

Yagua (Payne 1990:30) a. Sa-juuy Anita. 3.SG-fall Anita b. Anita 0-juuy, Anita 3.SG-fall 'Anita fell.'

Though there may be solutions for such phenomena based on information contained in underlying representations only, i.e. purely on semantic or pragmatic factors, examples from Arabic to be discussed later show that not all problems of form may be solved before the linear order rules are executed.2 A second source of undergeneration is the fact that there are no clear principles for the expansion of operators apart from centripetal expansion, i.e. operator application according to the inside-out order in which they are are found in underlying representations (cf. Dik 1997:381f). If this iconicity-driven principle of expression were to be the only device for the generation of grammatical material, we would have a problem in dealing with underlying representations like the one in (5). (5)

(int Ei : (X] : (past ej : (smile [V] (dl Xj : John [PN]) Ag ))))

If we expanded the operators in (5) in a straightforward centripetal fashion, we would first derive a past form for the verbal predicate before we generated the auxiliary do on the basis of the Interrogative operator at the illocutionary level, which would lead to the ill-formed utterance in (6a) instead of the correct one in (6b). (6)

a. *Do John smiled? b. Did John smile?

So, in this case the illocutionary operator should trigger before the tense operator does, in contrast to their centripetal order in the underlying representation. Obviously, there must be more principles at work here to decide which combinations of what operators might be selected at what stage of the form generation process, be they universal, typological or language specific.3 To my knowledge, no suggestions for other such principles are to be found in the FG literature to date, nor do we find any complete example

4

Dik Bakker

of the derivation of a sentence based just on centripetality. The fact that there is no worked-out strategy for combining expression templates at the respective syntactic levels probably plays an important role in this respect. In other words: there is no theory of syntactic constituency. As such, undergeneration threatens the most elementary of adequacy requirements of a linguistic theory, i.e. observational adequacy. Turning to overgeneration, this is essentially due to the lack of constraining principles and formal structures to apply them to. Since there seems to be no way of controlling which (combination of) operators operate on which operands, which outputs may and may not be recycled to later rule applications, etc., the impression at least is that 'anything goes'. 4 Without enough constraining principles on the semantics-syntax interface and given the ban on filtering devices in FG we will in all likelihood be left with a large number of potentially meaningful but formally unacceptable utterances apart from the well-formed ones. This in turn threatens the cognitive and typological adequacy of the theory; it also casts doubt on the learnability of a grammar operating with the traditional expression rules. In order to solve these problems, at least in principle, Bakker (2001) formulates a version of the expression component which integrates the form generation and linearization modules, thus providing a solution for cases such as those illustrated above. Furthermore, in this newly shaped ER component, the rules are assumed to operate in a dynamic fashion. They produce tree-like structures with phonological strings for their terminal nodes. As in the case of the original ER component the point of departure is that the expression rules are triggered by the interaction between the semantics and pragmatics of the underlying representation. The wellknown templates and placement rules reappear, albeit in a slightly modified form. According to Bakker & Siewierska (2002) the dynamicity of rule execution is governed by five principles of tree construction. Three principles determine the way expression trees come into existence, viz. via top-down, left-to-right, and depth-first expansion. Two more principles determine the way in which linguistic information is passed on during the dynamic expansion process, viz. inheritance and percolation. The overall view will be that the shape of the expression rules in a synchronic grammar of a language L is dictated not only by the requirements of underlying representations, but also by the direct and indirect influence of non-linguistic factors of a cognitive nature, such as iconicity and economy. Is is also assumed that the dynamic process itself plays an active role in instances of

More arguments for the dynamic expression model

5

grammaticalization, as an active factor or, diachronically, as an independent ground for explanation.5 As has been argued in the contributions referred to above a large class of problems of the type demonstrated by examples (1) - (5) may well be solved within this dynamic ER framework. In section 3 below this will be demonstrated for a domain where form and order interdependencies are rather common in the languages of the world, i.e. the shape of subject agreement marking on the verb. However, certain problematic cases from this domain remain unsolved if only the three tree expansion principles top down, left to right, and depth first are available. On the basis of a case study of subject agreement in Arabic it will be argued, in section 4, that a fourth tree expansion principle is called for, i.e. a limited amount of look ahead. Before we launch into this in section 2, there will be a short resume of the dynamic expression rule component as it has been developed to date. However, a remark on linguistic modelling is called for first. According to Dik (1997:1), the ultimate question FG wants to answer is how the language user works. From this it follows that the requirement of cognitive adequacy which the theory imposes on itself should be taken very seriously.6 Interestingly, however, the models that have been presented so far within FG theory model the grammar rather than the speaker and/or the addressee (cf. Dik 1997:60; Hengeveld 2004). I would argue that such models may serve a purely organizational purpose, such as structuring the theory, ordering the respective components of the grammar in relation to each other, clarifying a linguistic discussion or even planning the linguistic enterprise itself. But grammar models, which were not constructed directly on the basis of insights from the psychology of language cannot by implication serve to test the theory on its cognitive adequacy. I would therefore propose the development of a model of the language user, arguably even distinguishing between a model of the spreaker and one of the addressee.7 Elsewhere (Siewierska & Bakker, this volume) it is shown that the dynamic expression rule model fits well into a dynamic discourse-oriented model of the speaker based on FG theory. Having said this, for the current discussion we will assume, as in the traditional transformational grammar models, including the standard FG grammar model, that a fully specified underlying representation, with all the semantic and pragmatic aspects filled in, is available before the first expression rule is triggered. Discourse representations will play no significant role here.

6 2.

Dik Bakker The dynamic ER component: A short introduction

The idea of dynamic expression hinges on the integration of the form generation rules and the templates and placement rules of the standard theory into a tree-like structure that represents the history of the expression of a specific underlying representation (URj). Such a tree should be seen as the result of a dynamic process rather than as a static representation of the (morpho-)syntactic structure of the corresponding URj as in most syntactic theories. The tree consists of a number of nodes, which should be interpreted as the constituents of the tree. Each node contains a specific part of the URj under expression including its lexical material, operators, functions and layered structure. Examples are the subject term, the main predicate, an adjectival restrictor or a relative clause. A node explicitly specifies the primary and auxiliary μ operators relevant at the corresponding stage of the expression process. In fact, a node may only contain μ operators and no lexical material at all. On the basis of its constellation of lexical and grammatical material, a node selects a template from the grammar, which specifies the linear order of the actual elements of the URj element handled by the node. For each of the slots in that template a new daughter node is created, in a left-to-right fashion, which then will deal with the element of the mother node that has been assigned to it, and so on recursively. This process continues until each lexical element resulting from the original URj has been assigned its own terminal node. Unlike in the case of the standard model of expression, grammatical elements such as articles, auxiliaries and plural suffixes are not created before but rather during the process of linearization, at the moment at which precisely enough information is available in order to select the right form and in time for them to introduce features that derive from them, such as case. Typically, grammatical markers are the expression of a terminal node which contains only μ operators and no lexical material whatsoever. The corresponding grammatical element is then selected by the grammar. Some terminal nodes may even be prespecified for a form, which is then considered to be completely grammaticalized. This may be the case for bound forms such as plural suffixes, which only have phonologically determined allomorphs. This top-down process starts with the creation of a node for the fully specified underlying representation under expression, typically a clause, but it may also be a term or any other well-formed underlying construct. For the English sentence in (7) below the top node for its expression may look as in (8). In node specifications, Slot is the lable that identifies the

More arguments for the dynamic expression model

7

node in the grammar. It corresponds to the notion of functional slot in the traditional order templates. Cat contains the morpho-syntactic category of the constituent to be expressed via this slot. Config is the field that prespecifies the type of configuration to be expressed by this type of node (in this case a clause). The set of all possible configurations that fit this description could be seen as the (recursive) defmition-by-enumeration of the Slot lable. FncFtrs and FrmFtrs are the functional (i.e. primary) and formal (i.e. auxiliary) μ operators relevant for this node. SubCat determines the subcategories of Cat, and will contain the template that determines the order in which the subconstituents will be expressed. It could be seen as the (recursive) definition-by-enumeration of the Cat lable, and therefore as the definition of morphosyntactic constituency in the corresponding language.8 (7)

[decl E,:[Xi:[pres e^'smart'[A] (rempi 'Those girls are smart.'

(8)

Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE [ PRED CAT ARGi ]]]] FncFtrs: ILLOC, TENSE, CAT, PERSON.subject, NUMBER, subj ect FrmFtrs: SubCat:

Xi:'girl'[N]) z e r 0 , s u b j e ct ] ] ] ]

(8) gives the node in its abstract, uninstantiated form, the way in which it is part of the language-specific grammar. It is selected on the basis of certain formal aspects of the UR under expression, in this case a clause. The Configuration field of this node will now be filled by the UR of (7), which leads to the partially instantiated version of it in (9) below. Note that the two types of feature sets are instantiated at the same time, to the extent that their values are available.

8

DikBakker

(9)

Node 1 (partially instantiated) Slot: clause Cat: s Config: [decl Ei:[Xi:[pres ei:['smart'[A] (rem pi x^'girl' [N])zero,subject ]]]] FncFtrs: decl, pres, [A], [-sp,-hr], pi FrmFtrs: SubCat:

In the following step, the right template will be selected for the expression in the Config field of Node 1. This will give us its fully instantiated version in (10). (10) Node 1 (fully instantiated) Slot: clause Cat: s Config: [decl Ei:[Xi:[pres ei:['smart'[A] (rempi x^'girl' [N] )zero, subject ]]]] FncFtrs: decl, pres, [A], [-sp,-hr], pi FrmFtrs: SubCat: PI, subject, vfin, mainpred The process now continues in a left-to-right, top-down fashion with the expansion of the PI position. For this slot several alternative nodes will be available, a situation comparable to the alternative placement rules for PI in the standard theory (cf. Dik 1997:421). The right candidate node will then be selected, and the corresponding element of the Config field of node 1 will be inserted into the Config field of this newly created node 2 - in this case the subject term. The process will continue depth-first by selecting the right template to express the subject term, here a noun phrase template. Subsequently the leftmost element of this template - in this case a determiner position - will be expanded. The final result of this process will be a tree as in figure 1 below. The figures next to the branches indicate the order in which node expansion takes place. Note that the tree is not just a representation of the syntactic distribution of the material in the UR but extends into the morphological level.

More arguments for the dynamic expression model

'girl'

's'

9

'smart'

Figure 1. Full tree expansion

The typical way in which the functional and formal features of nodes get their values is by inheritance. These values are either derived directly from the UR material in the configuration field or copied from the Config field of the mother node. However, feature values may also be acquired via processes that take place in daughter nodes of the node under discussion. In those cases, features will percolate bottom upwards. The percolation mechanism is necessary for those cases in which formal information which results from early and very 'deep' stages of expression, such as case assignment by an adposition, certain order phenomena or the phonological form of some grammatical marker, will remain available for later stages in the expression process, where it may have some impact on the form.9 Trees such as those in figure 1 may also be seen as static constituent trees, as in Role and Reference Grammar (Van Valin & La Polla 1997) or Government and Binding theory (Haegeman 1994). However, unlike the former, dynamic ER trees are not stored as ready-mades in the fund, and selected on the basis of underlying semantic representations. And unlike

10

DikBakker

the latter, they are not the result of a transformational process that leads from a base-generated structure to a surface structure and which is in fact unrelated to the actual process of uttering a sentence. FG expression trees are the dynamic result - one could say: the side-effect - of the actual expression process, and exist only after an utterance has been completed. In a way, they are the utterance itself. This short introduction should be enough for our purposes here. For a more comprehensive treatment we refer to the articles mentioned. In the sections below, several finer details will be discussed, as well as a necessary extension of the 5 principles mentioned above.

3. Attempting to solve the form-order problem Form-order problems for which a unified underlying factor may be found can be handled in a more or less straightforward fashion. This situation arguably applies to the following case from Konjo, an Austronesian language from Sulawesi. (11) Konjo (Friberg 1996:141, 146) a. Na-peppe'-i Amir asung-ku. 3.ERG-hit-3.ABS Amir dog-POSS.l 'Amir hit my dog.' b. I-nai ang-kanre-i PI-who TRANS.DEF-ate-3.ABS 'Who ate my sweet potatoes?' c. I-Ali ang-kanre-i PI-Ali TRANS.DEF-ate-3.ABS 'Ali ate your sweet potatoes.'

lamejaha-ku? sweet.potatoes-POSS.l

lamejaha-ta sweet.potatoes-POSS.2

In (1 la) we have the pragmatically neutral declarative VSO order in which the verb displays agreement with both the Agent and the Patient. In (1 lb) a question - and (11c), the answer to that question, the focalized constituent is placed in preverbal position. At the same time the agreement marker for the Agent is replaced by a marker which codes transitivity and the definiteness of the preverbal focus constituent. So, we could reason that in this case the word order and the type and form of the Agent agreement prefix

More arguments for the dynamic expression model

11

are determined by the same underlying factor, and therefore in some sense are the expression of it, i.e. of the pragmatic status (the information focus) of the Agent, among other features. A somewhat more complex form-order relationship is illustrated in the following example from Fore, a language from Papua New Guinea. (12) Fore (Foley 1986:171) a. Yaga: wä a-egü-i-e. pig man 3SG.PAT-hit-3SG.AG-DECL 'The man hits the pig.' b. Yaga:-wama wä a-egü-i-e. pig-ERG man 3SG.PAT-hit-3SG.AG-DECL 'The pig hits the man.' c. Wa mäsi a-ka-i-e. man boy 3SG.PAT-see-3SG.AG-DECL 'The man saw the boy.' d. Mäsi wä-mä a-ka-i-e. boy man-ERG 3SG.PAT-see-3SG.AG-DECL 'The man sees the BOY.' The order of the preverbal constituents in this verb final language is relatively free: both APV and PAV orders are acceptable, and probably reflect pragmatic differences. Often the verbal morphology indicates which constituent is the Agent - the suffix - and which the Patient - the prefix. If this is ambiguous, an animacy hierarchy determines what is the more likely Agent, as in (12a). If the speaker wants to assign the Agent role to the referent which is lower on the hierarchy, as in (12b), then explicit ergative marking of the Agent is necessary. When both arguments score equal on the hierarchy, and there is no ERG marker, constituent order is decisive, as in (12c), which has only one interpretation. If pragmatic reasons make APV order undesirable, then again case-marking may be called in to disambiguate, as in (12d). Example (12) makes clear that the interplay between form and order phenomena may sometimes be quite complex, and that both are not necessarily the direct expression of the same underlying factor. This makes cases like the one from Fore at best very problematic for the standard expression

12

DikBakker

rules, especially when order has an impact on form rather than the other way around. To find out to what extent the dynamic expression rules can deal with such matters we will have a look at subject-verb agreement in Arabic, which presents us with a number of relevant problems. Let us first consider the relatively simple examples in (13) below. (13) Arabic (Mohammad 2000:109)10 a. ?al-?awladu dza?-u. the-boysTOp came-3PL.M b. Dza?-a l-?awladu. came-3SG.M the-boysFoc 'The boys came.' In (13a) the subject is topic and we have SV order. In this case there is full Person/Number/Gender agreement marking on the verb. In (13b), however, with focal subject and VS order, there is no Number agreement, only Person and Gender. Potentially, this case could be handled by the standard rules. We would first calculate the form of the verb, taking into consideration the pragmatic function of the subject term. We might have the - simplified and very partial - expression rules for the form of the finite verb in Arabic as in (14) below. They take the form suggested in Dik (1997:383), with the pragmatic function of the subject as a condition. (14) Arabic; form of finite verb a. TENSE, PERSON, NUMBER, GENDER [PRED [V]] = FORM Condition: PragF.subject = topic b. TENSE, PERSON, NUMBER sg, GENDER [PRED [V]] = FORM Condition: PERSON = 3, PragF.subject=focus The arrow in rule (b) is a shorthand for a redundancy rule which inserts 'sg' for the value of Number in the set of μ operators. Subsequently, placement rules which also take the pragmatic function of the subject into consideration would then locate the subject term either in the PI template slot (PragF. subject=topic) or in a postverbal S or Focus slot (PragF. subj ect=focus).

More arguments for the dynamic expression model 13 Although this looks quite straightforward, there is a fundamental problem with this approach. It includes the pragmatic function of the subject in the formulae that calculate the form of the verb. If we do not interpret expression rules as being mere description, but rather in terms of the underlying elements they really express - provided that they are not completely arbitrary - , then having a rule like (14) would imply that in the grammar of Arabic the form of the subject agreement marker on the verb actually expresses the fact that the subject is topic or focus, and is not just an indirect side effect. This does not seem to be a very satisfactory interpretation of this constellation, the more so since it works only for third person plural subjects and not for first and second person, as shown in (15) below.11 In (15a) we have a preverbal first person plural pronoun. In (15b) it is postverbal, a rather marked but not impossible constellation, according to Olmsted Gary & Gamal-Eldin (1982:54). In both cases, the marker on the verb is the one for 1st plural. The constellation in (15c), with 1st singular on the verb is not acceptable. (15)

a. niHin 1PL

yii-na. come.PST-IPL

yii-na come.PST-IPL

niHin. 1PL

*yii-t niHin. come.PST-lSG 1PL 'We came.' Let us now see how the examples of (13) would be treated in the dynamic expression rule framework introduced in section 1. In the representations given below no attempt is made at completeness. They just serve to clarify how the dynamic expression rules might handle such cases. Any feature that does not seem to be relevant for the current discussion will therefore be left out. The uninstantiated node in (16) might be a candidate node for the sentence level.

14

DikBakker

(16) Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE E-VAR [ PRED CAT ARGi ]]]] FncFtrs: TENSE, PERSON, NUMBER, GENDER Instantiation for the sentence with subject-topic in (13a) gives the version in (17). (17) Node 1 (partially instantiated) Slot: clause Cat: s Config: [decl [ [past ej : [dza? [V] [def pi X] : ?awladu [ N ] ] a g e n t .subject, topic ] ] ] ]

FncFtrs:

past, PERSON, NUMBER, GENDER

Full instantiation of this node will give us (18). (18) Node 1 (partially instantiated) Slot: clause Cat: s Config: [decl [ [past ei : [dza? [V] [def pi Xi : ?awladu [ N ] ]agent,subject,topic ] ] ] ]

FncFtrs: SubCat:

past, PERSON, NUMBER, GENDER PI, vfin, focus

This node is still only partially instantiated. Three of the four functional features still have no value.12 The next step is the expansion of the PI node. Assuming that there are several candidate fillers for this node, one of the alternatives for it may look like (19) below. Note that the configuration is prespecified for the subject-topic combination. (19) Node k (uninstantiated) Slot: PI Cat: np Config: [DEF NUM X j : PRED [N] ]SEMF,subject,topic FncFtrs: DEF, PERSON, NUMBER, GENDER

More arguments for the dynamic expression model

15

After instantiation we get (20). It is assumed that there exists a lexical rule which finds the value for the 'invisible' Gender feature by looking into the lexical entry of the head noun.' 3 (20) Node 2 (instantiated) PI Slot: Cat: np Config: [def pi xj: '?awladu' [N] ]agent,subject,toPic FncFtrs: def, 3, pi, m SubCat: art, noun In a few more steps, which we will not explore further here, this will lead to the expression of the subject term, including the cliticized definite article and the plural suffix. What is crucial is that at node number 2 the relevant features of the subject term are uniquely determined, and therefore become available. Through percolation the corresponding features of Node 1 will be replaced by these values, which gives us a fully instantiated version for this node, as in (18') (18') Node 1 (fully instantiated) Slot: clause Cat: s Config: [decl [ [past ei : [dza? [V] [def pi Xi : ?awladu FncFtrs: SubCat:

past, 3, pi, m PI, vfin, focus

The process continues with the selection of a node for the functional vfin slot. This vfin slot will inherit the values for its Person, Number and Gender features from node 1, and then pass them on to the actual nodes that express the finite verb form. Since after this step all material in the Config field of node 1 will be expressed the process is complete and comes to a halt.14 Although this schema seems to work for the SV case, it is not adequate for the VS one. Person and Gender information is necessary to get the right verb form, and will not be available to the vfin node if the subject term has not been expressed yet. This implies that the values for these features should be available at the clause level, before the vfin node is created. This calls for a redefinition of the clause node. In (16') a modified version is to be found which provides them.

16 Dik Bakker (16') Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE [ PRED CAT ARG, ]]]] FncFtrs: TENSE, PERSON.subject, NUMBER, GENDER, subject With the extra instruction that the values for Person and Gender should be retrieved from the subject term the values for these features will become available at the clause level as soon as Config is instantiated. For the expression of (13a) this has no repercussions. The same steps apply as the ones given in (17)-(20) above. The only difference is that there will be no percolation of the Person and Gender values; they are simply available from the start. However, for the expression of (13b) these subtle changes are crucial. Since in the grammar of Arabic there is no PI node which is specified for a focal subject, or indeed for any of the material in the underlying representation of (13b), expression will start with the instantiation of the vfm node. This node will inherit the values for its Person and Gender features from Node 1. The fact that the Number value remains uninstantiated will trigger a default rule which assigns singular to it.15 This will give us the right form for the finite verb. The default assignment process is local, i.e. there is a barrier to the percolation of the values that were assigned. After expression of the finite verb, for which there may be two more nodes to be expanded, one for the verb root and one for the agreement suffix, expression will continue with the focus node. One of the possible alternatives for this node will look like (21) below. (21) Node k (uninstantiated) Slot: focus Cat: np Config: [DEF NUM X j : PRED [N] ]SEMF,SYNTF,FOCUS FncFtrs: DEF, PERSON, NUMBER, GENDER Since there is an unexpressed candidate for this node, it will be instantiated as in (22) below. On this occasion, the Number feature will get its value from the inserted term, i.e. plural. The Person and Gender features inherit their values from Node 1. The latter values will harmonize with the local values for these features for obvious reasons.16

More arguments for the dynamic expression model (22)

17

Node 5 (instantiated) Slot: focus Cat: np Config: [def pi x j : '?awladu' [N] ]agent,subject,focus FncFtrs: def, 3, pi, m SubCat: art, noun

For the examples of agreement in (13) this treatment seems to be satisfactory. It is no longer necessary to assume that form and order phenomena are the expression of the same underlying factor, in this case topicality versus focality of the subject term. Rather, we can say that the constituent order is pragmatically determined while the form of the agreement marking may be seen as a side-effect of the dynamics of expression. If we express the subject first, as in (13a), its Number feature becomes available for its own expression. As a result it will percolate upwards, and become available for the rest of the expression. This is not the case, however, for (13b). As for the Person and Gender features of the subject, I will assume that they are so prominent that they are accessible at the level of the clause. This accessibility of Person and Gender is coded in (16') via the set of features that may be instantiated on the basis of the local Config field. As a result, the corresponding values are available under any of the order variants, as opposed to the value for Number, which is assigned via a local default rule and does not percolate. 17 Interestingly, this analysis seems to get support from a related aspect of verbal agreement in Arabic. Unlike many other languages, in case of the presence of an auxiliary verb, both the latter and the main verbal predicate agree with the subject. However, when the subject is expressed after the auxiliary but before the main verb there is person agreement on the latter but not on the former. This is exemplified in (23a) and (23b), respectively. (23)

Arabic (Bahloul & Harbert 1993:16) a. Al-bint-aani kaan-ataa ta-ktub-aani The-girl-3.DU was-3F.DU 3F-write-DU b. Kaan-at al-bint-aani ta-ktub-aani was-3F.SG the-girls-3.DU 3F-write-DU 'The two girls were writing their lesson.'

darsa-humaa. lesson-F.DU darsa-humaa. lesson-F.DU

18

DikBakker

These facts may well be explained on the basis of the assumption that number only becomes available for reasons of agreement after the subject has been expressed and the number feature of the term has been brought into play.18 Though possibly satisfactory for the above cases, the solutions presented so far will not suffice to handle other aspects of subject verb agreement in Arabic. Further provisions need to be made, to which we now turn.

4. The necessity of looking ahead In section 3 we saw that the dynamic top-down, left-right, depth-first expansion of nodes, combined with the inheritance and percolation of features was adequate for the representation of the form-order interaction phenomena in example (13). The dynamicity aspect even provides a possible explanation for them. However, this is certainly not the full picture of subject-verb agreement in Arabic. Complications arise in the case of subjects that consist of conjoined noun phrases with different genders, a problem labelled 'resolution' in Corbett (1991:261), where it is discussed extensively. Let us first consider the preverbal situation as exemplified in (24). Unless stated otherwise, the Arabic examples below are from the discussion in Mohammad (2000:112ff). (24)

a. ?al-?awladu w-al-banatu qara?u the-boys and-the-girls read.3PL.M 'The boys and the girls read a book.'

kitaban. book

b. ?al-banatu w-al-?awladu qara?u the-girls and-the-boys read.3PL.M 'The girls and the boys read a book.'

kitaban. book

As these examples show, gender conflicts are resolved in favour of masculine, independent of the order of the conjuncts. Only when all conjuncts are feminine do we get feminine agreement on the verb. With VS order, however, the situation is more complex. In such a case, the order of the conjuncts is decisive, as shown in (25) below. Note that, unlike the situation with preverbal subject we have agreement with the conjunct closest to the verb, masculine for (25a) and feminine for (25b). In the latter case masculine agreement is even ungrammatical, as shown in (25c). Note furthermore

More arguments for the dynamic expression model 19 that also in these cases we have the singular rather than the plural form on the verb. Apparently, the conjunction is not an extra factor for the assignment of plural number in VS sentences. (25) a. Qara?a ?al-?awladu w-al-banatu read.3SG.M the-boys and-the-girls 'The boys and the girls read a book.' b. Qara?at l-banatu w-al-?awladu read.3SG.F the-girls and-the-boys 'The girls and the boys read a book.' c. *Qara?a l-banatu w-al-?awladu read.3SG.M the-girls and-the-boys 'The girls and the boys read a book.'

kitaban. book

kitaban. book

kitaban. book

This situation changes, however, when there is a constituent between the verb and the subject. In the sentences of (26) we have VXS order, with a feminine subject. Although (26a) is acceptable, (26b) is observed more often. (26) a. qad kan-at yaskun-ut PRT was-3SG.F settle-3SG.F l-iraq-a Pumamun muxtalifatun. the-Iraq-ACC people.F.PL different b. qad kan-a yaskun-u PRT was-3SG.M settle-3SG.M l-iraq-a Pumamun muxtalifatun. the-Iraq-ACC people.F.PL different 'Different peoples had settled in Iraq.' Apparently, the distance between the expression of the verb form and the (first conjunct of the) subject determines the amount of semantic correspondence between the agreement marker and the source of the agreement. The greater the distance, the less the semantics of the source are accessible. Other semantic factors may play a role in such processes as well. Corbett (1991) analyses examples of subject-verb agreement in Classical Arabic VS sentences as determined by the (in)defmiteness of the subject term.

20

DikBakker

Below, in (27a) there is feminine agreement followed by a definite feminine subject; in (27b) there is masculine agreement followed by an indefnite feminine subject. Note, however, that in the latter case there is also an intervening (definite) object as in the examples of (26) above. (27)

Classical Arabic (Corbett 1991:125) a. Jä-at Hindun. came-F.SG Hind (female proper name) 'Hind came.' b. Hadar-a l-qädiya ('i)mra'atum. came.before-M.SG the-judge woman Ά woman came before the judge.'

The examples in (24)-(27) are highly suggestive of the following scenario for agreement resolution with conjoined subjects in Arabic. In SV situations there is no real resolution problem. The agreement is masculine, provided that at least one of the conjuncts is masculine, otherwise it is feminine. In VS situations, apart from the general unavailability of Number as an agreement factor for third person subjects, the conjunct to be expressed first determines the gender of the agreement marker on the verb, unless there is intervening material. In that case Gender becomes inaccessible as well and masculine gender serves as the default. Finally, definiteness may add to the accessibility of a subject term. That definiteness may also play a role in the form of agreement marking in other languages may be illustrated on the basis of the example in (28) from Hua, a Papuan language. In (28a), with an object marker on the verb, the corresponding referent should be interpreted as definite; in (28b), without such a marker, we are dealing with an indefinite interpretation. (28) Hua (Haiman 1980:371) a. Vedemo p-go-e. men 3PL-see-lSG Ί saw the men.' b. Mna-vza-mo bird-COLL-PL Ί saw birds.'

ko-e see-lSG

(*p-go-e). (3PL-see-lSG)

More arguments for the dynamic expression model

21

Assuming that this analysis of the agreement facts in Arabic is correct and that it should be dealt with within the expression component, the set of principles which determine what information is available at what stage during expression needs to be modified. The combination of depth first and inheritance which constrained the agreement marking in section 3 does not suffice to code the facts of this section. What we need is a mechanism that allows the expression rules to look ahead, i.e. to inspect features of a constituent that has not been expressed yet but is about to be expressed. This mechanism should represent the accessibility of features in a more subtle way than the yes/no fashion of node 1 in example (16'), repeated here for convenience. (16') Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE [ PRED CAT ARGi ]]]] FncFtrs: TENSE, PERSON.subject, NUMBER, GENDER, subj ect In the case of (16'), the Person and Gender values are available immediately upon instantiation of the clause node; they are maximally accessible. The value for Number becomes available only upon the expression of the element that contains it, here the subject. However, given the facts of Arabic, we have to cater for (at least) three types of accessibility of semantic and pragmatic information in underlying representations. -

-

Global accessibility: the information is available immediately after the expression process has started. Relative accessibility: the information is available some time before the expression of the UR constituent to which it belongs but not for the complete expression of the clause concerned Local accessibility: the information is available only upon the expression of the UR constituent to which it belongs.

In terms of subject-verb agreement in Arabic, Person would be an example of global accessibility, Gender of relative accessibility and Number of local accessibility. Obviously, relative accessibility is a dynamic notion since it introduces time into the expression schedule. Rather than measuring relative accessibility in time units, such as milliseconds, however, it seems

22

Dik Bakker

to be more adequate in our model to measure it in terms of a 'look ahead window'. This window represents the maximum number of nodes to the right of the node currently under expression that may be taken into consideration in order to find the relevant feature values for the current node.19 An adequate representation of accessibility in expression schemata may be a scale, with values between 0 (local accessibility) and oo (global accessibility). In order to constrain the accessibility of the respective features in node representations, the accessibility window may then be added to the features involved. In an extended version of the formalism, there might be features that co-determine the width of the accessibility window of some feature, such as defmiteness in the case of gender in Arabic, as in example (27) above.20 Taking all this into consideration, we may modify the definition for the clause node as in (16") below. (16") Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE [ PRED CAT ARG, ]]]] FncFtrs: TENSE, PERSON.subject , GENDER.subject , NUMBER, subj ect The descriptor is an instruction to find a value for this feature in the Config, whatever its level of embedding.21 The descriptor implies the opposite. The corresponding feature value should be present locally, i.e. it should be found at the top level of the Config constituent. For a clause, this would only concern the illocutionary operators and satellites; for a term it would at least concern its pragmatic, semantic and syntactic functions, the ω operators, and some features of the nominal head, such as its Gender and its nominal aspect in terms of Rijkhoff (2002). Since is the strongest constraint, it may well be the best candidate for the default in node descriptions. Any descriptor , where 0 < η < oo, implies that the corresponding feature, if not readily available, should become available after up to a maximum of η nodes have been created to the right of the current node at the level of the same SubCat. The and descriptors may be handled adequately under the original 5 principles, but calls for an extension. In order to implement the model has to allow for a limited amount of look ahead. Thus, to ensure the correctness of a form, it may be necessary to access information from a constituent which will be expressed either immediately after the

More arguments for the dynamic expression model

23

current node or to a maximum of η nodes away. This is indeed a special example of the intertwining of form and order. For the model this means that, at least in these cases, the development should be breadth first, at least one step ahead, rather than depth first. Arguably, this is the reflection in the model of parallel processing by speakers. In a really parallel model, while one processor works on the expression of the constituent in node k, another process is busy in the background selecting the constituent for node k+1. Some information from the latter will then be available for the process under node k before it terminates and node k+1 moves to the foreground. If we translate this parallelism into a limited amount of breadthfirst node expansion, we get a dynamic tree that looks more or less like the one in figure 2 below. Of course, such a 'flat' representation can never reveal the real intricacies of parallel processing.22 In such a setup it must be assumed that, after the subject node 3 has been created and partially instantiated, the top-down process for node 2 has progressed, however not to the extent that the verb form has been expressed fully. Matters might be even more complex in the sense that there may be more than two parallel processes.

Figure 2. Partial tree with breadth-first development Two points need some extra attention here. Firstly, on several occasions there has been mention of default-value assignment when no 'real' value had been found for a feature at the time of expression. With the (16") ver-

24

DikBakker

sion of the clause node above this is still the necessary procedure for Number in all VS cases, and for Gender in VXS cases. This mechanism should receive more general attention since it is probably relevant for other modules in the FG model. An example is the insertion of the operator value definite for terms that have a demonstrative operator like proximate or remote in URs. Typically, but not necessarily, default values will be the least marked values for some feature in a language, such as masculine and singular for Arabic. To the extent that languages apply the same values for such defaults, the mechanism should be treated at a more universal level of the theory. For lack of such a default theory, and since the discussion above focusses on one phenomenon in one language only, we will have to supply the default values in an explicit and more or less ad hoc fashion here. It will be assumed that in a more comprehensive fragment of the grammar they will be provided by a general procedure which operates independently in all relevant instances. For the time being, (16"') will be an updated version of the clause node for Arabic. (16" ')Node k (uninstantiated) Slot: clause Cat: s Config: [ILLOC [ [TENSE [ PRED CAT ARG, ]]]] FncFtrs: TENSE, PERSON.subject < oo, default=3 >, GENDER.subject < 1, default=m >, NUMBER.subj ect < 0, default=sg > A second point that should be made here concerns constituency. It could be argued that in a more or less indirect manner constituency in our dynamic expression rule component is defined by the templates, or rather the fillers for the SubCat fields of the respective nodes. So, in the case of the clause node in (18) above, we can say that the Cat fields of the fillers of the PI, vfin and focus slots are the constituents of the s category of its own Cat. For a language, the complete set of possible fillers of all Cat nodes dynamically and recursively defines the constituency of that language. It is crucial, however, that the dynamics of the tree expansion process are in tune with any static interpretation of constituency. More concretely, for the Arabic data above we should check whether the breadth-first strategy has implications for the way in which we represent compound noun phrases. For the computation of the right form for (25a-b) - repeated below for convenience - it seems to be attractive to assume that both conjuncts are

More arguments for the dynamic expression model

25

constituents at the clause level in their own right, expressed via serialized slots, rather than as two subconstituents of a higher conjunct node. Finding the right value for the Gender feature would then simply be taking the value of the first constituent that comes in, i.e. masculine for (25a) and feminine for (25b). (25) a. Qara?a ?al-?awladu w-al-banatu read.3SG.M the-boys and-the-girls 'The boys and the girls read a book.' b. QaraPat l-banatu w-al-?awladu read.3SG.F the-girls and-the-boys 'The girls and the boys read a book.'

kitaban. book

kitaban. book

However, such a 'flat' structure seems to make it problematic to cater for the agreement in example (29) below where plurality is a feature of the combined terms rather than the individual ones. Singular agreement is ungrammatical here. (29) Moroccan Arabic (Corbett 2000:202) ?umar w-?ali msaw/*msa. Omar and Ali left.PL/left.SG.M Omar and Ali left.' Although proposals have been made for coordination in FG, there exists no canonical underlying representation for coordinated terms, nor for coordinates at any other level of grammatical description.23 It is therefore somewhat speculative to discuss them here, let alone to introduce a formal structure for their expression. What follows is therefore at best tentative. The most straightforward way to represent the compound noun phrase the girl and the boy seems to be the term structure in (30), which contains the coordination operator '&'. (30)

((def sg Xl : girl (x,)) & (def sg x2 : boy ( x 2 ) ) )

A disadvantage of this representation may be the fact that its plurality in the syntax should be inferred by the expression rules since it is not coded in the underlying representation in a direct way. We also need, however, the sg operators in order to get the right forms for the respective conjuncts.

26

Dik Bakker

An alternative would be to embed the conjuncts under a common Number operator, as in (31), with its own variable. (31)

(pi x 3 : ((def sg x, : girl ( X l ) ) & (def sg x2 : boy ( x 2 ) ) ) )

The assumption underlying (31) is that coordination creates a new referent in the discourse, which is the union of the referents of the conjunct terms. An extra complication for (31) is the question whether we also need to insert the other term operators for the x3 variable, such as definiteness, and what we would do in that respect with combinations like the one in (32). (32)

That boy and a girl entered the store.

For this discussion I will assume that (30) is at least a good candidate for the representation of compound nouns. It has the advantage of simplicity, but leaves part of the work to the expression rules. However, this would not be a novelty altogether. For Gender the eventual determination of the value for a compound term has to be left to the expression rules anyway, since this feature is not coded as a term operator in the first place. Interestingly, in Arabic there are other examples of the interaction between gender and constituent order. A way of making yes-no questions is to prefix the sentence with a form of the 3rd person pronoun. According to Olmsted Gary & Gamal-Eldin (1982) the number and gender of this prefixed pronoun are based on those of the subject. In the case of a coordinated subject, the gender of the first conjunct is chosen, analogous to the situation with postverbal subject agreement discussed above. This is shown in (33) below. (33a) has a masculine first conjunct (proper name Sami) and (33b) a feminine one (proper name Muna). (33) Arabic (Olmsted Gary & Gamal-Eldin 1982:5) a. Huwwa Sami wi Muna maguush? 3.M.SG Sami.M and Mona.F came.not.3PL 'Haven't Sami and Mona come?' b. Hijja Muna wi Sami maguush? 3.F.SG Mona.F and Sami.M came.not.3PL 'Haven't Mona and Sami come?'

More arguments for the dynamic expression model 27 In general, therefore, resolution seems to be a rather natural part of (dynamic) expression rather than (static) semantic representation. As for the corresponding underlying representation, we would like to claim as little semantic structure as is strictly necessary.24 A more complex term structure such as the one in (31) could then be claimed for cases where there is both semantic and formal evidence for it. This is most probably the case for examples such as the ones in (34) below, where the conjuncts are quantified by zuuz, 'both' and plural agreement is also obligatory for VS orders. (34) Arabic (Aoun et al. 1994:211) a. ?umar w Sa?id msaw Omar and Said went.3PL

b-zuuz with-both

b. *Msa ?umar w Sa?id b-zuuz went.3SG Omar and Said with-both Omar and Said both went to school'

l-l-medrasa. to-the-school l-l-medrasa. to-the-school

For such utterances I would postulate the more complex term representation in (31). The binding variable x3 and its plural operator are presumably introduced by the constituent containing the zuuz quantifier, which is probably best analyzed as a satellite term at the level of the core predication. This term is in focus, therefore highly accessible and always expressed in the postverbal focus position. It could be said to bind both conjuncts via its term variable.25 Against this background, verb agreement with compound terms in Arabic may then be determined in the following way. Person is a feature of the whole compound term. It will get value 3 if there is no [+S] or [+A] among the conjuncts of the term. As a globally accessible feature it is established at the sentence node and is available by inheritance. Both Gender and Number are derived in a more dynamic sense from the subject term, when it is processed itself. Both feature values will be established on the basis of the values of the respective conjuncts. In a left-to-right manner, the overall value will be updated under the application of the relevant resolution schema. For Number, the initial value will be established as the value of the first conjunct. If any conjuncts follow, this value will be overridden by plural. In Arabic this process only has effect on agreement in SV cases, when it reaches the vfin node via percolation and inheritance. It will be assumed that Number cannot be retrieved from terms that are still being processed in the background in parallel with the expression of an earlier

28

DikBakker

node, in other words it is inaccessible. Therefore all VS cases will get default singular agreement. For Gender, on the other hand, it works rather differently. The initial value for Gender is also established as the gender of the first conjunct term. As the conjuncts are processed and expressed sequentially their gender value will be taken into consideration by the resolution rule for [-S,-A] terms. In the case of Arabic this means that masculine will override feminine. As a result the value for terms like the one in (30) will end up being masculine. This works for SV cases. For VS constellations we have already assumed above that Gender as opposed to Number is accessible in the background process that runs in parallel with the Vfm expression process. However, under the 'flat' term constellation proposed earlier only gender information from the first conjunct of a compound term in the background will be available in time for it to be taken into consideration for agreement purposes. This explains why only the gender of the first conjunct will be coded in the agreement marker on the verb.

5. Discussion and conclusions On the basis of data from subject verb agreement in Standard Arabic it was argued above that the five principles of dynamic expression proposed on earlier occasions do not provide the complete framework necessary. In order to explain why there is only a certain amount of gender agreement with postverbal subjects rather than full agreement or none at all it was assumed that during the expression of some underlying constituent there is access to a parallel process that selects the next constituent. Only a subset of the features of this constituent becomes available, i.e. the more accessible ones. These may be selected on the basis of their more prominent role in reference tracking, or for other functionally motivated reasons.26 For Arabic it appears that Gender is more outstanding than Number, at least in the context of subject verb agreement. Access is also limited in another sense viz. only the constituent selected to directly follow the one under expression is available; in the case of compound constituents this concerns only the leftmost conjunct. In the dynamic expression rule model and its treelike representations this parallelism may be implemented in terms of a restricted amount of breadth-first node development ('look ahead') coming before any expansion in depth. A slightly different way of looking at the role of parallelism, which in my opinion would not affect the position taken here, is to see it as an ex-

More arguments for the dynamic expression model

29

planatory device in a diachronic sense rather than a mechanism which is active in a synchronic grammar of Arabic. In other words, gender agreement with the first conjunct would then be a grammaticalized result of this parallellism and simply a rule applied by speakers of today's Arabic, who learned it and internalized it at the language acquisition stage. Corbett (2000:208) assumes that this may well be the case in Modern Standard Arabic. Other varieties, including Moroccan Arabic according to Aoun et al. (1996:196), have grammaticalized plural agreement in both orders, i.e. there is no difference between the two situations in terms of agreement marking. In yet other varieties, and notably Cairene Arabic, there actually is variation in plural versus singular marking for 3 rd person referents even for SV patterns. The crucial factor appears to be the score on the animacy scale for the subject. On the basis of recorded interviews with 26 speakers of Cairene Arabic Belnap (1999:174) observes the following for plural preverbal subjects (my table is an aggregation of Belnap's table 1).

Table 1. Animacy and agreement in Cairene Arabic Semantics of head noun human animal inanimate

Number of instances in corpus 174 20 335

Percentage plural marking on V 91% 35% 3%

This variation in Cairene Arabic may be indicative of the dynamic role of the animacy parameter rather than its being grammaticalized completely into fixed rules for this variety. The respective varieties of Arabic (and other languages) could then be placed on the following number agreement continuum. Languages to the far left have grammaticalized number agreement for 3 rd person subjects in VS orders to singular, the languages to the far right to plural, while the languages in the middle determine it on the basis of its accessibility in the discourse situation. Animateness and Focality are factors which co-determine accessibility.

ISG Standard Arabic

-

variation Cairene Arabic

-

PL | Moroccan Arabic

Figure 3. Grammaticalization cline for 3rd person number agreement

30

Oik Bakker

But even if we assumed full grammaticalization for this agreement phenomenon in Standard Arabic, we would still be left with the problem of how to code this in the FG model. Even under the full grammaticalization scenario we need a way to inspect the order in which conjunts will be expressed before the agreement parameters are determined. Of course, we might make the '&' operator order-sensitive, just like the restrictor operator ':'. Or we might assign a pragmatic function to the first conjunct in order to code its 'prominence'. Agreement could then be made sensitive to such devices, and look ahead could be disposed of. We would, however, still need a processing approach since they apply in VS constellations but not in SV ones. And a look-ahead device or something similar is still necessary for independent reasons i.e. if we want our model to explain why these types of grammatical patterns have come about in the first place. We also need look ahead if we want to explain why certain speech errors such as so-called Spoonerisms are much more common than others. Of course these observations have been based on a limited amount of data, stemming mainly from one language only. It is unlikely, however, that speaker models of other languages would work in a completely different way as far as their dynamic operating principles are concerned. In this respect it is interesting that form-order interactions are certainly not uncommon in the languages of the world, as shown in examples (1) - (4). This means that a look-ahead mechanism might well turn out to be part of the speaker model irrespective of the language, and that we could predict that traces of it may be found in all spoken varieties, and not only in speech errors. In details, however, languages may of course vary considerably. For instance, the three agreement parameters Person, Number and Gender may play a different role in different grammars and there may be different window sizes for different stages of the expression process. For Standard Arabic the following accessibility hierarchy seems to apply, at least to subject verb agreement: (35) Person

>

Gender

>

Number

The hierarchy in (35) seems to be only partially in tune with some typological observations that were made concerning the synchronic and diachronic behaviour of these three parameters in subject-verb agreement (Siewierska & Bakker, fc.). In our sample of 426 languages of the world subject-verb agreement is relevant in some 78% of the cases. In over 99% of these languages Person is one of the agreement features, which points

More arguments for the dynamic expression model

31

towards its prominence over the other features. In 94% of these cases Speaker and Addressee are formally contrasted while non-discourse participants (i.e. 3rd persons) are often homophonous with one of these forms, or unmarked (zero). This confirms the prominent status of Speaker and Addressee; the frequency of zero expression confirms the status of 3rd person as an 'else' category or a default. Diachronically, going from free pronouns to verbal affixes, Person is almost never lost altogether. For the two other features the picture is a bit different. Overall, Number plays a role in subject agreement in 93% of the relevant languages, while in only 35% Gender is manifest. The same relative distribution was found for the semantic parameters of free pronominal forms, although the absolute numbers are higher. Furthermore, when Number gets lost from subject agreement markers this is typically only partial (e.g. dual gets lost but not plural), while Gender typically gets lost completely. It remains to be seen whether a conclusion should be that the relationship between the overall role of Number and Gender in grammars may differ from their role in the dynamics of subject agreement. It may also be the case that Arabic deviates from the mainstream in this respect in the sense that Gender plays a more prominent role in this language than in many other languages. This view may get some support from the fact that Arabic has gender not only in the third person but also in the second person pronouns and in subject agreement markers, which is relatively rare in our database. Gender is manifest in both free pronominal forms and bound ones in only 16 languages, or less than 5% of the languages that have subject agreement in the first place. The value of a model of grammar is determined to a high degree on the basis of the predictions that it makes, because these render the theory verifiable and falsifiable in terms of real language data. Whatever the merits of the expression scheme presented above, a number of predictions might be derived from it concerning subject marking which, all other things being equal, would potentially be different for languages of varying constituent order types, for both their basic and marked orders. The same holds for object agreement marking. It is not clear what types of predictions may be derived on the basis of the standard model of expression in FG. A final remark concerns underlying representations in Functional Grammar. According to the theory, the criteria on the basis of which the precise contents of UR's are determined are twofold. They should be precise enough to get at the right interpretation of the utterance in the discourse and to trigger the right expression rules. As in the case of the representation of compound terms above, the latter requirement may well work

32

Dik Bakker

as a constraining factor on the former. Indeed, it is not always clear how to determine what belongs to the pure semantics and pragmatics of a sentence on the one hand and which implications and inferences might be derived from them in the context on the other hand. Not subjecting oneself to the exercise of working out the admittedly tedious step-by-step process that eventually leads to the final form may both result in the unnecessary introduction of some elements and the neglect of others. Therefore, even if the priority in FG obviously lies with the functional side of language, work on the more mundane aspects of grammar, i.e. morphosyntax, may well provide a better insight into the more invisible conceptual aspects which are often invoked in order to explain the visible side of linguistic behaviour. There is some risk involved, however, in that we may find that expression has its own principles which cannot be reduced to functional explanation.

Abbreviations 0 1 2 3 A ABS AG AUX COLL COMP DAT DECL DEF DU ERG

null morpheme first person second person third person addressee absolutive agent auxiliary collective complementizer dative declarative definite dual ergative

F FOC LOC Μ PAT PI PL POSS PRF PST

feminine gender focus locative masculine gender patient person indicator plural possessive perfective past tense speaker s singular SG TOP topic TRANS transitive

More arguments for the dynamic expression model

33

Notes 1.

2.

3.

This was observed long ago. Corbett (1983:74) concludes that 'the further a target is from its controller, the more frequently semantic agreement will occur [rather than grammatical agreement - DB]'. Of course in the formal sense there is no problem if we assume that all order phenomena can be predicted on the basis of functional information. But even if this were the case, the cost would be to prerun the order rules before form is calculated, i.e. by reversing the order of the two expression modules. This would in turn endanger any case where order is determined by form. Dik (1997:341) seems to acknowledge the form-order problem by giving an example from Dutch, cited in (i) below. (i) a. Jij zie-t een kat in de tuin. 2SG see-2SG a cat in the garden 'You see a cat in the garden.' b. In de tuin zie je een kat. In the garden see-0 2SG a cat 'In the garden you see a cat.' In (ia), with a preverbal second person singular subject, there is an agreement marker on the verb; in (ib), with a postverbal subject, there is no such marker. No solution is provided for such constellations, however, by Dik (1997). Note that a solution based on phonological reasoning, as suggested by an anonymnous reviewer, makes matters in fact maximally complex since the decision about the form should then be postponed until the very end of the expression process: the determination of the phonological form. In that case we would be dealing with a deletion, a device not desired for the expression component. Apart from that, there would be complications for such phonologically related utterances as in (ii). (ii) Je/Hij maak-t je broek vuil. 2SG/3SG make-2SG your pants dirty 'You are/He is dirtying your pants.' If we base the phenomenon of (i) on phonological diachrony rather than an active phonological rule, then again we do not have a solution, since a diachronic rule can have no place in a synchronic grammar. In that case it should be treated as a completely arbitrary rule in the sense of Croft (1995), with order as the only 'explanatory' factor. This is not to say, of course, that iconicity is not one of the more important factors that determine linguistic form in general. As such, it is frequently and

34

4.

5.

6. 7.

Dik Bakker convincingly invoked in functional and typological linguistic work. See Haiman (1983) for a first discussion; see Croft (1990: 164f) for an overview; and see Hengeveld (1989) for an early example in FG, motivating the hierarchical structure of underlying representations. However, the complex and varied ways in which the different types of primary and auxiliary operators project onto form across languages calls for a more comprehensive approach, involving multiple explanatory principles. Also, even if iconicity is assigned a very central explanatory role, as is done in the case of the centripetal principle in Dik (1997), it may be assumed to have shaped the expression rules together with other grammar external principles, such as economy and processing, and that they are still active in continuously reshaping them. This still leaves the theory with the task of providing the precise rules for getting at the right syntactic and morphological elements in the synchronic grammar of some language. This is more or less suggested by the way the so-called contextual retrieval principle is defined. In order to assemble the necessary primary operators for some expression rule 'the context in which it occurs in [the] underlying clause structure is searched for relevant agreement features' (Dik 1997:356). Before expression takes place the underlying representation is then extended precisely such that copies of relevant operators are locally available. It is not clear from this how the searching process would take place other than taking precisely the set of operators that are necessary for the job on hindsight, thereby becoming circular. In a way, this would locate aspects of expression in the semantics of the theory, and could be interpreted as 'syntactic smuggling', as a correlate to the 'semantic smuggling' which is sometimes observed in the practice of formal syntacticians, who turn [+animate] or [+topic] into syntactic features. This is a mild form of autonomy of morphosyntax which following Croft (1995) may be called 'arbitrairiness'. The 2SG phenomenon of Dutch as discussed in note 2 may be an example of this. A stronger form of assigning autonomy to the formal component of the grammar is assuming that there are inherent ('innate') principles at work, such as those of Chomsky's Universal Grammar. One could reserve the notion 'autonomy' for this type of independence of morpho-syntax. Our five principles and the constraints they create for expression may be seen as aspects of such autonomy. The notion 'cognitive adequacy' replaces the traditional notion of psycholinguistic adequacy. Currently, authors in the field seem to disagree as to whether two more or less independent models should be distinguished or only one. Brown & Hagoort (1999) devote two separate chapters to the two modes of language processing,

More arguments for the dynamic expression model

8.

9. 10.

11. 12.

13.

14.

15.

35

written by different authors. Others argue that the overlap between the two modalities is such - the speaker is also an addressee and the other way around - that an integrated model is called for (Gerard Kempen p.c.). This does not affect our argument for a cognitively adequate model, however. In the formalizations below, upper case notions indicate feature names (TENSE), and lower case notions indicate feature values (past). Quotes indicate predicates of the object language ('girl'). When there may be ambiguity among feature names, they are made more specific by way of an extension: e.g. NUMBER.subject means the Number value of the Subject term. All these notations are relatively informal, and meant only to illustrate the points under discussion. A formally more rigourous version, based on feature-value logic and unification, has been introduced in Bakker (1989). A computer implementation is discussed in Bakker (1994). In the domain of phonology backward vowel harmony would be an example. Mohammad (2000) discusses and contrasts a number of Arabic dialects. Unless stated otherwise I will only use examples here which are labelled as Standard Arabic by this author. See Siewierska (fc.) for cross-linguistic differences between first and second person versus 3rd person forms. The three labels Person, Number and Gender are not uniquely determinable at the clause level since there may be more terms which have operator values that satisfy them. They therefore remain uninstantiated. Such lexical priority rules are also necessary for the insertion of irregular forms during the expression process. Their role in expression is discussed in somewhat more detail in Siewierska & Bakker (this volume). For this discussion I will assume that a monitor will keep track of the contents of the Config, will mark all material that has been expressed, and will send a signal when there is no unmarked material left. An alternative way would be to 'do away' with material that has been expressed. This would be more in keeping with the literal meaning of the notion of 'expression', but it would fail for cases where material is expressed more than once. Furthermore, there is still the Focus slot to be filled. The problem of how to deal with empty slots, and whether they should be generated in the first place, will not be discussed here. See Bakker (2001) for some remarks on empty slots. Some authors in the generative literature, and notably Mohammad (2000) and van Gelderen (1996), argue that the VS order should be interpreted as a covert expletive, which explains for the 3rd person singular agreement. Although a case for a 'pro-dropped' expletive could technically also be made within an FG approach, this does not seem to be a suitable solution for the gender agreement phenomena to be discussed below, which are in need of a funda-

36

16.

17.

18.

19.

20.

21.

DikBakker mentally different treatment in my view. For lack of an explicit expletive pro form I would prefer to interpret the third person singular verb form in these cases as a somewhat underrepresented version of 'the subject is a non-speech participant'. Aoun et al. (1994:200f), who take a generative perspective, reject an analysis based on expletives on theory-internal grounds. For an idealized model of the speaker, who makes no speech errors, I will assume that harmony between inherited and local features is always the case. However, for a more realistic model of the speaker, and for one of the addressee, there should of course be provisions for the recognition (and creation!) of ill-formed utterances. It may look rather arbitrary to postulate this for the respective features concerned. However, apart from the fact that the empirical data point in the direction of Number being less prominent than Person and Gender in these utterances of Arabic, it is known from work in typology that Person is the most prominent feature in verb agreement markers and pronouns in the languages of the world. Its major function is the identification of speech participants, i.e. 1st and 2nd person, versus non-participants, or 3rd person. The major function of Gender and Number seems to be the identification of discourse referents, which is mainly a subdistinction of the currently relevant 3rd person referents. Cf. Bhatt (to appear) for an in-depth discussion of this. These examples seem to support the 'underrepresented 3rd person subject' interpretation that I suggested above in note 15 as opposed to the expletive interpretation of some authors. As a technical term, look ahead was introduced in the theory of parsing natural language rather than the generation of sentences, as is the case here (cf. van der Steen 1987). It remains to be seen whether in a model that represents both the Speaker and the Addressee it may in fact turn out to be one and the same notion in both modalities. For this discussion it suffices to assume that the upper limit for global accessibility is the information contained in the underlying clause (but see Siewierska & Bakker (this volume) for a discourse approach to agreement and crossreference phenomena). The notion 'accessibility' is used as an English word rather than a technical term. While it is interpreted here from the perspective of a model of the Speaker, other authors - most notably Ariel (2000) - use it from the perspective of a model of the Addressee. Undoubtedly there will be lower limits also to such a process. For a discussion of this point see Bakker (1994:283ff). For the current discussion it will be assumed that any higher level of underlying representations that is recursively invoked within a lower level creates such a barrier.

More arguments for the dynamic expression model

37

22. Technically, we are forced to make a choice between depth first and breadth first. The latter seems to win out in the case at hand. However, since in the model we are dealing with the mapping of a no doubt very complex parallel process onto the much simpler dynamic model, this is only an either/or choice on first sight. In fact we have a mixed situation here, in which partial breadthfirst development alternates with extension in depth. 23. This is extra remarkable since a functional theory of grammar was first suggested in Simon Dik's (1968) PhD thesis entitled 'Coordination: Its implications for the theory of general linguistics'. 24. Aoun et al. (1994:207), who try to explain the agreement phenomena of Arabic discussed above in terms of structural relationships such as specifier-head and government, represent what they call the prominence of the first conjunct in postverbal coordinated structures by putting the first element in the specifier position of the NP where it governs the other conjuncts. This could be seen as a static version of a 'look-ahead' mechanism. Note that this representation is only assumed for VS constellations; in other words constituent order is relevant in these cases, as admitted by Aoun et al. (1994:219). 25. My expectation would be that plural would also emerge in VS orders under the presence of other types of first-order satellites which predicate something about the subject rather than about the verbal predicate, such as 'sick' or 'with pleasure'. A representation as in (31) could also be claimed for collective as opposed to distributive readings, and for predicates that force a collective reading onto their arguments, such as intransitive 'meet'. 26. This may be true of information processing in general. In the minimalist framework (Chomsky 1992:44) a distinction is made between strong and weak inflection, and correspondingly between strong and weak NP-features, precisely to cater for agreement differences due to constituent order, as in Arabic. This may be the formal translation of what I consider to be functionally determined agreement phenomena.

References Aoun, Joseph, Elabbas Bennamoun and Dominique Sportiche 1994 Agreement, word order, and conjunction in some varieties of Arabic. Linguistic Inquiry 25-2: 195-220. Ariel, Mira 2000 The development of person agreement markers: from pronouns to higher accessibility markers. In Usage-based models of language,

38

DikBakker

Michael Barlow and Suzanne Kemmer (eds.), 197-260. Stanford: CSLI Publications. Bahloul, Maher and Wayne Harbert 1993 Agreement Asymmetries in Arabic. In The proceedings of the eleventh West Coast conference on formal linguistics, Jonathan Mead (ed.), 15-31. Los Angeles: University of California Press. Bakker, Dik 1989 A formalism for Functional Grammar expression rules. In Functional Grammar and the computer, John H. Connolly and Simon C. Dik (eds.), 45-64. Dordrecht: Foris. 1994 Formal and computational aspects of Functional Grammar and language typology. Amsterdam: IFOTT. 1999 FG expression rules: from templates to constituent structure. Working papers in Functional Grammar 67, University of Amsterdam. 2001 The FG expression rules: a dynamic model. Revista Canaria de Estudios Ingleses 42: 15-54. Bakker, Dik and Anna Siewierska 2002 Adpositions, the Lexicon and Expression Rules. In New perspectives on argument structure in Functional Grammar, Ricardo Mairal Uson and Maria J. Perez Quintero (eds.), 125-178. Berlin: Mouton de Gruyter. Belnap, Robert K. 1999 A new perspective on the history of Arabic variation in marking agreement with plural heads. Folia Linguistica XXXIII/2: 169-185. Bhatt, D.N.S. fc. Pronouns: a crosslinguistic study. Oxford: Oxford University Press. Brown, Colin M. and Peter Hagoort (eds.) 1999 The neurocognition of language. Oxford: Oxford University Press. Chomsky, Noam 1992 A Minimalist Program for Linguistic Theory. MIT Occasional Papers in Linguistics. Cambridge: MIT Press. Corbett, Greville G. 1983 Hierarchies. Targets and controllers: Agreement patterns in Slavic. London: Croom Helm. 1991 Gender. Cambridge: Cambridge University Press. 2000 Number. Cambridge: Cambridge University Press. Croft, William C. 1990 Typology and universals. Cambridge: Cambridge University Press. 1995 Autonomy and functionalist linguistics. Language 71-3: 490-532.

More arguments for the dynamic expression model

39

Dik, Simon C. 1997 The Theory of Functional Grammar 1: The structure of the clause. 2nd, revised edition, edited by K. Hengeveld. Berlin: Mouton de Gruyter. Foley, William A. 1986 The Papuan languages of New Guinea. Cambridge: Cambridge University Press. 1991 The Yimas Language of New Guinea. Stanford: Stanford University Press. Friberg, Barbara 1991 Konjo's peripatetic person markers. In Papers in Austronesian linguistics, Hein Steinhauer (ed.), 137-171. Pacific Linguistics A84. Gelderen, Elly van 1996 Parametrizing agreement features in Arabic, Bantu languages, and varieties of English. Linguistics 34: 753-767. Haegeman, Liliane M. 1994 Introduction to government and binding theory. Oxford: Basil Blackwell. Haiman, John 1980 Hua: a Papuan language of the Eastern Highlands of New Guinea. Amsterdam: John Benjamins. 1983 Iconic and economic motivation. Language 54: 781-819. Hengeveld, Kees 1989 Layers and operators in Functional Grammar. Journal of Linguistics 25.1: 127-157. 2004 The architecture of a Functional Discourse Grammar. In A new architecture for Functional Grammar, J.Lachlan Mackenzie and Maria A. Gomez Gonzalez (eds.). Berlin: Mouton de Gruyter. Hieda, Osamu 1998 A sketch of Koegu grammar. In Surmic languages and cultures, Gerrit J. Dimmendaal and Marco Last (eds.), 345-373. Köln: Rüdiger Köppe Verlag. Mohammad, Mohammad A. 2000 Word order, agreement and pronominalization in standard and Palestinian Arabic. Amsterdam: John Benjamins. Najlis, Elena 1973 Lengua Selknam. Filologiay lingiiistica. Buenos Aires: Universidad del Salvador.

40

DikBakker

Olmsted Gary, Judith and Saad Gamal-Eldin 1982 Cairene Egyptian colloquial Arabic. Amsterdam: North-Holland. Payne, Doris L. 1990 The pragmatics of word order: Typological dimensions of verbinitial languages. Berlin: Mouton de Gruyter. Rijkhoff, Jan 2002 The noun phrase. Oxford: Oxford University Press. Saltarelli, Mario 1988 Basque. London: Croom Helm. Siewierska, Anna and Dik Bakker this vol. The agreement cross-reference continuum: Person marking in FG. fc. The semantic parameters of person agreement: synchrony and diachrony. Steen, Gert van der 1987 A program generator for recognition, parsing and transduction with syntactic patterns. Ph. D. diss., University of Utrecht. Van Valin, Robert D. and Randy J. LaPolla 1997 Syntax. Cambridge: Cambridge University Press.

Constituent ordering in the expression component of Functional Grammar John H. Connolly

1.

Introduction

In the treatment of syntax and morphology in FG, the expression component has a vital role to play. Within the FG framework, every clause that is generated is assigned an abstract, semantically and pragmatically oriented representation known as an underlying clause structure, in which the elements are as yet unordered in respect of the syntactic form of the clause. The underlying clause structure is then realised as a linguistic expression by the application of rules, which are (naturally) called expression rules. These rules serve to specify the form, order and prosodies of the resulting expression, and collectively they constitute the expression component of the grammar. In two recent papers Bakker (1999, 2001) has proposed a new approach to the construction of the expression component. According to this new approach, which he terms the 'dynamic model', the expression rules assign an explicit hierarchical syntactic structure to the clauses and other units generated by the grammar in a way that does not occur in the standard model of FG introduced in Dik (1978) and elaborated in Dik (1997a,b). One of the purposes of the FG expression component is to deal with constituent order. Accordingly, it is natural to enquire how this aspect of syntax may be handled in the dynamic model. It is to this question that the present chapter is addressed.

42

John Η. Connolly

2. Constituent ordering in the standard model 2.1. The Basic Linearisation Mechanism Let us begin by reminding ourselves of the mechanism by which constituent order is handled in the standard model. Dik (1978, 1980) proposed that the linearisation of elements should be accomplished by means of placement rules, working in conjunction with schemata such as: (1)

PI

S

V

Ο

The purpose of the placement rules is to insert constituents into the appropriate positions within the schema. Examples of placement rules are as follows: (2)

PI-constituent GivenTopic Subject Verbal-constituent Object

PI PI -» s -> V ο

(A PI-constituent is one that normally stands at the beginning of the clause, regardless of its syntactic function, for example interrogative nominale such as 'who' or 'whom' in English.) Clearly, with such an arrangement it is possible for positions within the schema to remain unfilled. For instance, if the clause contains no object then the Ο position will stay empty. Similarly, if the clause contains neither a PI-constituent, nor a constituent bearing the requisite pragmatic function (such as Given Topic, though the precise conditions of eligibility for PI vary from language to language), then the PI position will remain unoccupied. However, what matters is that the constituents that are present should end up in the right order. It was pointed out in Connolly (1983, 1991) that in the case of languages in which the order of major (or nuclear) constituents, namely subject, verbal-constituent (or predicator), object and complement, is free, a slightly modified treatment is preferable. Consider, for instance, a language like Latin, in which all six permutations of the constituents {S, V, 0 } are possible. If we were to postulate a schema for Latin of the style

Constituent ordering in Functional Grammar (3)

PI

S

Ο

43

V

then in order to generate the sequence SVO we would need to have rules like (4)

Verbal-constituent Object

-> Ο V

which appear bizarre. Instead, it was proposed to adopt a more abstract schema, along the lines of (5)

PI

N1

N2

N3

(N for 'nuclear constituent'), enabling us to formulate rules such as (6)

Verbal-constituent Object

-> N2 -> N3

In Connolly (1983) the term 'syntactic template' was employed to refer to a schema, such as (5), into which constituents are inserted by means of placement rules. This usage is adopted in Dik (1997a: 393). Of course, this does not prevent functional patterns like (1) and (3) from being used to encapsulate the basic ordering conventions of clauses, when describing particular languages.

2.2. Some Refinements A fairly detailed attempt was made in Connolly (1991: 45-129) to describe the constituent order of English within the standard model of FG. The coverage included both major constituents and minor constituents (adverbials and vocative elements), and encompassed both clause-rank ordering (involving subject, predicator, object, adverbials and so on) and phrase-rank ordering (involving, for example, determiner, adjectives and noun within the noun phrase). It was found that basically the standard model was equal to the task, though certain refinements were needed. Three such refinements will now be outlined. Firstly, it was found necessary to add numerous conditions to placement rules. For instance, suppose we have a template of the form

44

John Η. Connolly

(7)

PI

N1

N2

N3

N4

N5

N6

N7

to accommodate the major elements of the English clause. N2 is the normal position for the subject, N1 and N3 for the predicator (N1 accommodating the finite verb in instances of interrogative inversion, for instance), and N4 and N5 for objects and complements. N6 and N7 need not concern us here, but are explained in Connolly (1991: 68-69). In order to achieve the normal relative order of direct and indirect object, it is necessary to attach conditions to the relevant placement rules, along the lines of (8)

Indirect Object

-> or

Direct Object or

N4(if not introduced by a preposition) N5 (if introduced by a preposition) N4(if N4 is not already occupied) N5

Secondly, it was necessary to allow for the possibility that more than one constituent with the same syntactic function (such as adverbial) may be placed in the same position within a template, and that the constituents concerned will then have to be ordered relative to one another by means of a further stage in the placement process. Thus, in order to cater for the placement of adverbiale it is necessary to extend (7) as follows: (9)

MO PI Ml N1 M2 N2 M3 N3 M4 N4 M5 N5 M6 N6 M7 N7 M8

(M for 'minor constituent'), reflecting the fact that it is possible in principle to fit adverbials in between any pair of nuclear elements (admittedly, sometimes resulting in a marked order). For examples, see Connolly (1991: 78ff). Of course, not every slot will be filled (and indeed, all the Mpositions will remain empty if the clause happens to contain no minor elements). On the other hand, in order to generate an expression like (10)

He rarely completely neglected his duties.

it is necessary to insert two adverbials ('rarely' and 'completely') into M3, between the subject in N2 and the predicator in N3. Within M3 the two adverbials then have to be ordered relative to each other. Thirdly, it was necessary to introduce the notion of an 'elastic template'. In contrast to a template like (9), which contains a fixed number of positions, an elastic template contains as many positions as are needed to

Constituent ordering in Functional Grammar

45

accommodate the constituents to be linearised. For instance, it is not possible to predict in advance how many adverbials will need to be linearised within a particular M-position such as M3. Therefore, the requisite template needs to be elastic: (11)

M3.1 M3.2 ... M3.i

or, more generally (12)

Mn.l Mn.2 ... Mn.t

where t represents the total number of adverbials within a given Apposition, MM, that need linearising within that overall position. Finally, it may be noted that the placement rules in Connolly (1991) make frequent reference to syntactic functions such as subject or adverbial. In order to facilitate this it is proposed that these functions are assigned to constituents as part of the expression process, after the basic form of the constituents has been established but (obviously) prior to linearisation; cf. Connolly (1991: 46-49). This means, for example, that a unified treatment can be given to the placement of the syntactic subject whether it is the underlying subject or not. For instance, in a sentence like (13) It is snowing. the syntactic subject it is semantically empty, but this does not affect its placement.

3.

Constituent Ordering in the Dynamic Model

3.1. The Basic Syntactic Generation Mechanism Bakker (2001: 17) states that in the dynamic model the processes of constructing morphosyntactic forms and of linearisation are conflated, but that as far as constituent order is concerned, the mechanism based on placement rules and syntactic templates is preserved, though in a disguised manner. Let us examine this approach to the handling of constituent order. When the dynamic model is employed to derive the linguistic expression for a simple sentence, the first step is to generate a node labelled ' sen-

46

John Η. Connolly

tence', accommodating relevant information such as illocutionary force. The node is constructed on the basis of a frame supplied by the grammar, containing uninstantiated variables, for instance ILLOC. These variables are then instantiated by matching them with corresponding information in the underlying clause structure of the sentence being generated. Thus, if the underlying clause structure specifies that the illocutionary force of the clause is declarative, then the ILLOC variable in the frame is instantiated with the value 'decl'. One of the slots in the frame is named 'SubCat' (subcategorisation), and this is now filled by means of a template supplied by the grammar. This template contains a sequence of entries for all the possible constituents of the frame concerned, in the correct order, along the lines of the following: (14) PI

V-fin

S

V-nonfm

Ο

When the template has been copied into the SubCat frame of the node, its entries are matched with the corresponding parts of the underlying clause structure, filling PI first, then V-fin, then S and so on. If a slot in the template does not match with any material in the underlying clause structure then it is skipped. For example, if there is no non-finite verb then the Vnonfin slot in the template will remain empty and the matching procedure will move on and try to fill the Ο slot. The leftmost filled slot in the template then acts as a trigger for the generation of a new node, which will be a child of the current node within the hierarchical structure that is now starting to develop. For instance, supposing that the S node is the leftmost filled node, then a new node will be created, labelled 'subject' and containing variables such as NUMBER. The values of these variables, such as 'singular', will again be determined with reference to the underlying clause structure of the current sentence. Next, a template of a suitable type is supplied by the grammar, copied into the SubCat slot of the node frame and used to linearise the internal constituents of the subject term. And so the process continues, in a left-to-right depth-first manner, until the entire linguistic expression has been derived.

3.2. Remarks on the Handling of Constituent Ordering Some comments will now be offered in relation to the treatment of constituent order within the dynamic model. First of all, Bakker (2001: 30)

Constituent ordering in Functional Grammar

47

states that the slots provided by syntactic templates may be dynamically tailored to the actual contents of the underlying clause structure. In other words, he would like to see an element of elasticity being a property of all templates. He envisages (2001: 36) that unfilled entries will usually either be deleted or else not generated in the first place. Presumably, the motivation for this lies in the avoidance of overgeneration. But in fact, templates with unfilled entries do not do any harm in the generation of syntactic structure in the dynamic model. This is because the slots in the templates are not themselves structural nodes. Rather, they act as triggers for the generation of further nodes. But they can easily be constrained to do so if and only if they are filled. If they are empty, then they can simply be skipped and will thus not give rise to spurious child nodes. It is therefore proposed that full templates should be employed within the dynamic model. There is no need for any pruning apparatus to eliminate unfilled entries or for any mechanism for suppressing the generation of entries that are not going to be filled (a mechanism which sounds as though it would be difficult to formalise in any case). Matters are slightly complicated by the fact that Bakker (2001: 36) wishes to allow for the possibility of generating nodes which are present in the syntactic structure but are not expressed, yet need to be acknowledged because they exert effects that are sometimes described in terms of 'trace' phenomena. A possible mechanism to handle this situation is to generate the nodes concerned, but to insert an empty string into their 'Config' (configuration) slots. As a way of representing such a configuration, we may adopt the following notation: (15)

Config: Δ

Here, the symbol Δ represents the empty string. Of course, it is somewhat unfortunate to have to use any symbol at all to denote the empty string, since in reality the latter has no observable manifestation! However, it is essential to have a means of indicating that the Config slot has been filled by the empty string, rather than being unfilled and hence ineligible to generate a child node. The advantage of deploying the empty string is that it will not be expressed by any sound or character in the spoken or written expression, which is precisely the result desired. Moreover, the effect is achieved without having to delete any element of structure.

48

John Η. Connolly

With regard to the internal form of the syntactic templates in the dynamic model, Bakker (2001: 43) presents examples such as the following: (16) (17)

PI Det

Subject Num

Vfin Adj

Mainpred Nomhead

RelCl

These templates resemble the functional patterns found in Dik (1978). They contain entries which specify particular elements in particular positions, and therefore suffer from the disadvantage (mentioned earlier) that they do not readily generalise to languages in which constituent order is free. It would therefore be better to replace them with the type of templates employed in Connolly (1991) and Dik (1997a: 393), exemplified in (9) above and repeated here as (18): (18)

MO PI M l N1 M2 N2 M3 N3 M4 N4 M5 N5 M6 N6 M7 N7 M8

The placement rules, correspondingly, need to be formulated in such a way as to assign elements to the appropriate slots. Such rules will generally need to have conditions attached, as demonstrated earlier. However, the templates in (16) and (17) carry some useful grammatical information, including syntactic functional categories such as Subject and 'Nomhead' (nominal phrase head), which we would not want to lose. A possible means of dealing with this, while retaining the spirit of the dynamic model, is as follows. First, whenever an element is about to be inserted into a template slot, assign it a functional category, such as Subject, Object, Modifier or Head. Second, place the element in the appropriate slot. If desired (though this is not essential), the filled template could be annotated to show the functions associated with its occupied slots. For example, in the case of the sentence (19)

Then who found it?

the annotated template would read: (20)

MO PI A S M5

N5

Ml

N1 M2

M6

N6

N2

M7

M3

N7

N3 V M8

M4

N4 Ο

Constituent ordering in Functional Grammar

49

(A for 'adverbial'; here occupying MO and thus preceding the Plconstituent 'who'). Third, whenever a child node is generated, copy the syntactic function into its 'Lab' (label) slot, for instance (21) Lab: V in the case of the verbal constituent. In this way, whenever a placement rule needs to refer to a syntactic function, the relevant function of the element concerned will already be explicitly indicated. Furthermore, once the whole tree has been generated, syntactic-functional structures (such as ASVO) can be read off from the Lab slots of the relevant nodes (which is why the annotation of the template is not essential). The proposal just outlined implies that, contrary to Bakker (2001: 43), labels such as PI should not be copied into the Lab slot of a child node. Such purely positional labels are useful for the placement of elements and there is therefore no objection to including them as entries within templates. However, once they have served this purpose they do not appear to have any further use, in which case it is preferable to label the nodes in positions like PI with their functional category (such as Subject or Object), as this is more informative. (If positional markers like PI do, after all, turn out to be relevant to further processing, then it would be better to add them to the node rather than to use them in place of the syntactic functions). Another issue concerns the handling within the dynamic model of the situation where more than one constituent is placed in a single slot, for example two or more adverbiale in M8. In fact, this can be treated in much the same way as outlined previously. The grammar needs to make available an elastic template, along the lines of (12), repeated here as (22): (22)

ΜηΛΜη.2

...Mn.t

into whose slots the relevant elements are inserted by means of appropriately formulated placement rules. The variable η should be instantiated to match the value of the host slot, so that if the host slot is N8 then η is assigned the value 8, while t again represents the total number of constituents to be inserted into (22). At first sight, this might appear to involve inserting an additional layer of structure into the clause, without any syntactic justification. However, this is not the case. Templates do not, in themselves, constitute sequences of nodes in the actual structure of sentences, and so the insertion of further

50

John Η. Connolly

templates into a slot inside a node does not have the effect of adding an extra syntactic layer. This technique merely makes possible a two (or more) stage pass in the linearisation of constituents within a single node, and thus helps to simplify the formulation of the placement rules.

4.

Conclusion

The enhancement of the expression component of FG through the incorporation of a more elaborate syntactic generation mechanism, as accomplished through the dynamic model, is to be welcomed. This is because it enriches the information (including functional information of a syntactic nature) that is specified by the grammar in relation to the expressions which it generates. Of course, it is essential that in modifying the generative mechanism of the grammar, we do not inadvertently lose the good points of the standard model. With this in mind, the linearisation process within the dynamic model has been examined in the present chapter. The conclusion we have reached is that by making certain amendments to the dynamic model, while still maintaining its essential spirit, we are able to claim that we still have an adequate mechanism for the handling of constituent order in the expression component of FG.

References Bakker, Dik 1999 FG expression rules: from templates to constituent structure. Working Papers in Functional Grammar 67, University of Amsterdam. 2001 The FG expression rules: a dynamic model. Revista Canaria de Estudios Ingleses 42: 15-53. Connolly, John H. 1983 Placement rules and syntactic templates. In Advances in Functional Grammar, Simon C. Dik (ed.), 247-266. Dordrecht: Foris. 1991 Constituent Order in Functional Grammar: Synchronic and Diachronie Perspectives. Berlin: Foris. Dik, Simon C. 1978 Functional Grammar. Amsterdam: North-Holland. 1980 Studies in Functional Grammar. London: Academic Press.

Constituent ordering in Functional Grammar 1997a

1997b

51

The Theory of Functional Grammar 1: The Structure of the Clause. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. The Theory of Functional Grammar 2: Complex and Derived Constructions. Edited by Kees Hengeveld. Berlin: Mouton de Gruyter.

Dynamic expression in Functional Discourse Grammar Kees Hengeveld

1. Introduction The aim of this paper is to investigate to what extent the idea of a dynamic expression model, as proposed in Bakker (1999, 2001, this volume) in the context of Functional Grammar, can be applied within Functional Discourse Grammar (FDG) (Hengeveld 2004a, 2004b; Mackenzie & GomezGonzalez 2 0 0 4 ) T h e conclusion will be that a dynamic implementation, proposed by Bakker for the expression rule component, is not only relevant to the expression rule component as such, but can be integrated into the model of the grammar as a whole, i.e. including both the process of formulating underlying pragmatic and semantic representations and the process of converting these into morphosyntactic and phonological representations. These four types of representation distinguished within FDG thus all engage in a single dynamic implementation of the grammar. This point will be illustrated through the analysis of three different one-word utterances. The paper is organized as follows. In section 2 I describe the main properties of FDG that distinguish it from Functional Grammar (FG) (Dik 1997). A significantly updated version of the general architecture of FDG exhibiting these properties is presented in section 3. Subsequently, section 4 defines a number of general principles for the dynamic implementation of FDG, which are then illustrated for the aforementioned one-word utterances in section 5. The paper is rounded off with some general conclusions in section 6.

54

Kees

Hengeveld

2. From FG to FDG 2.1. Introduction Functional Discourse Grammar is a new version of Functional Grammar. FDG shares its functional-typological orientation with FG but differs from it with respect to a number of important basic properties, which are discussed in the following sections: FDG has a top-down organization (2.2); FDG takes the discourse act as the basic unit of analysis (2.3); FDG systematically interacts with a conceptual, a contextual, and an output component (2.4); FDG includes morphosyntactic and phonological representations as part of its underlying structure, rather than as the output of the grammar (2.5).

2.2. Top-down FDG incorporates the layered hierarchical structure of the clause characteristic of FG, but at the same time is radically different, in the sense that in generating utterances it starts with the encoding of the speaker's intention and then works down to articulation. FG, on the other hand, starts with the selection of lexical items and then gradually expands the underlying structure of the clause. This radical shift is motivated by the assumption that a model of grammar will be more effective the more its organization resembles language processing in the individual. Psycholinguistic studies (e.g. Levelt 1989) clearly show that language production is a top-down process, which starts with intentions and ends with the articulation of the actual linguistic expression. The grammatical production model reflects this process and is organized in a top-down fashion. This does not mean that FDG is a model of the speaker: FDG is a theory about grammar, but one that tries to make use of psycholinguistic evidence in its basic architecture. Two major operations have to be distinguished in the top-down construction of utterances: FORMULATION and ENCODING. Formulation concerns the rules that determine what constitute valid underlying pragmatic and semantic representations in a language. The rules involved in formulation may be called MAPPING RULES. Encoding concerns the rules that convert these pragmatic and semantic representations into morphosyntactic and phonological ones. The rules involved in encoding may be called EXPRESSION RULES. Encoding in FDG thus roughly corresponds to the

Dynamic expression in Functional Discourse Grammar

55

expression rule component in FG. Formulation in FDG has no counterpart in FG, since in FG the validity of underlying representations is generally presupposed. The top-down organization of the model is a precondition for a grammatical theory that aims at describing discourse units rather than sentences. In a discourse-oriented model the sentence is just one of the options that the speaker has to contribute to the ongoing discourse, hence mapping rules have to precede expression rules. This is the topic of the next section.

2.3. Discourse grammar 2.3.1. Units larger than a sentence There are many grammatical phenomena that can only be interpreted in terms of units larger than the individual clause or sentence. Examples of this are narrative constructions, the use of discourse particles, anaphorical chains, and tail-head linkage.2 By way of example, consider the following instance of tail-head linkage in Tidore: (1)

Tidore (van Staden 2000: 275) ... turus jafa cahi saloi ena=ge turus ... then Jafa carry.on.the.back basket 3.NH=there then ena=ge paka ine. Ine una oka koi... 3.NH=there ascend upwards upwards 3.SG.M pick banana '...then Jafa carried the saloi and went upwards. Went upwards he picked the bananas ...'

In many Indo-Pacific languages there are several grammatical phenomena that are a faithful and direct reflection of discourse organization. In Foley's (1986: 176) words: "A text is a coherent linking of clauses and sentences, and this coherence is achieved by rules of the language which state how clauses and sentences can be joined". Example (1) illustrates one of these linking devices offered by the grammar of Tidore. Episodes within stories are often realized as single sentences containing strings of clauses. The sentences (or rather: the episodes contained within them) are linked to each other by means of tail-head linkage: the last verb of the one sentence is repeated as the first verb of the next sentence, as illustrated in (l). 3

56

Kees Hengeveld

The crucial point here is that, as stated in the quote from Foley (1986), phenomena such as tail-head linkage are governed by rules of the language and thus form part of the grammatical system as it applies to narratives. Grammatical phenomena like these thus clearly show the need for a grammatical model that allows for the treatment of units larger than the individual sentence and of the (discourse) relations that obtain between and within these units.

2.3.2. Units smaller than a sentence As argued in Mackenzie (1998), the need for a discourse-oriented grammar also becomes apparent when units smaller than a sentence are considered. The following examples illustrate some types of non-sentential utterances, or HOLOPHRASES:

(2) (3) (4)

(What are you eating?) A donut. Congratulations! Oh John!

The answer in (2), the exclamation in (3), and the vocative expression in (4) all take a non-sentential form. Yet in the appropriate circumstances they all count as full and complete contributions to the discourse. In fact, any further elaboration of e.g. (2) would lead to an unnatural exchange. Given that in FG (and FDG) the deletion of specified material is disallowed, each of these utterances has to be taken as non-sentential at the level of the underlying representation too, and cannot be interpreted as the reduced version of an underlying complete sentence. The model should thus find a way of dealing with non-sentential utterances which recognizes the fact that they constitute fully grammatical discourse units.

2.3.3. The discourse act The conclusion that I draw from the preceding discussion is that the basic unit of discourse is not the sentence but the discourse act. Discourse acts combine into moves, which in turn may enter into larger discourse structures. These larger structures account for the units larger than the individual clause or sentence discussed in 2.3.1 above. On the other hand, dis-

Dynamic expression in Functional Discourse Grammar

57

course acts may be manifested in language as sentences, but also as fully grammatical sentence fragments, phrases or words. This is the way in which the units smaller than the individual sentence discussed in 2.3.2 can be handled. The latter point is a crucial one: it requires the grammatical model to be capable of mapping the unit of discourse act onto morphosyntactic units of various kinds. This mapping procedure in turn requires a topdown approach. Moves and discourse acts are notoriously difficult to define. I here stick to the definitions offered in Kroon (1995: 65-66), who, following Sinclair & Coulthard (1975) defines a move as 'the minimal free unit of discourse that is able to enter into an exchange structure' and a discourse act as 'the smallest identifiable units of communicative behaviour'. Note that a move consists of a single central act, which may be supported by one or more subsidiary acts. For an extensive discussion of the notion of discourse act see Hannay & Kroon (fc.).

2.4. Conceptual, contextual and output components The grammatical component of FDG is linked to a conceptual component, an output component, and a contextual component. The CONCEPTUAL COMPONENT is not part of the grammar but is the driving force behind the grammatical component as a whole. In fact, one way of interpreting the operation of formulation referred to in 2.2 is that it represents the conversion of a prelinguistic conceptual representation into the linguistically relevant semantic and pragmatic representations that are allowed by the grammar of the language concerned. The conceptual component is responsible for the development of both a communicative intention relevant for the current speech event and the associated conceptualizations with respect to the relevant events in the external real or imaginary world. Depending on the modality chosen, the OUTPUT COMPONENT generates acoustic, signed, or orthographic expressions. It does so in the operation of ARTICULATION, which is external to the grammatical component as such, but is fully dependent on the information provided by the grammatical component. In the acoustic modality, the operation of articulation takes a phonological (both segmental and suprasegmental) representation as its input and converts this into an acoustic signal, applying the necessary phonological rules.

58

Kees Hengeveld

The CONTEXTUAL COMPONENT contains a description of the discourse domain as it has been built up during the current discourse to the extent that this is relevant to the form that subsequent utterances may take. It does not only contain a description of the content and form of the preceding discourse, but also of the actual perceivable setting in which the speech event takes place. In this way the contextual component helps trigger utterances like the following: (5) (6)

I visited Peter in jail yesterday. He had just been arrested. He did it (pointing at suspect).

Once a participant has been introduced into the discourse, it can be referred to anaphorically within the same episode. This is illustrated in (5). A participant present within the speech situation can be referred to deictically, as in (6). In both cases a specification of the participant involved is available in the contextual component: in the first case in the description of the preceding discourse, in the second case in the description of the setting of the speech event. The contextual component is relevant for the description of all languages, but in some languages there are phenomena that demonstrate its relevance quite directly. Some random examples are the use of logophoric pronouns in many African languages, the distinction between anaphoric ai 'there' and deictic 'inä 'there' in Samoan (Mosel & Hovdhaugen 1992: 133), and the use of narrative constructions of various types, reflecting chronological progress within a narrative, in many different languages. For a rather specific example of how the content of the contextual component may affect grammatical choices consider the following example from Spanish: (7)

jQue päiid-a est-äsl what pale-F.SG COP-IND.PRES.2.SG 'How pale you look!'

The sentence in (7) is appropriate only when the addressee is female. Hence the contextual information about the identity of the speech participants is directly relevant to the form the predicative adjective takes.

Dynamic expression in Functional Discourse Grammar

59

2.5. Levels of representation In FG, underlying structures contain an interpersonal (pragmatic) and a representational (semantic) level of representation. FDG adds additional structural and phonological levels to these underlying representations, which contain morphosyntactic and phonological representations, respectively. One of the reasons to do so is that anaphoric reference is possible to all of these levels of linguistic organization. This means that these levels should be available as potential antecedents in underlying representations. Consider the following examples: Interpersonal level (8) A. Get out of here! B. Don't talk to me like that! Representational level (9) A. There are lots of traffic lights in this town. Β. I didn 't notice that. Structural level (10) A.I had chuletas de cordero last night. B. Is that how you say 'lamb chops' in Spanish? Phonological level (11) A.I had /tJuletas#de#kordero/ last night. B. Shouldn 't that be /tfuletas#de#0ordero/? In (8B) the anaphoric element that refers back to the communicative strategy chosen by A, which is indicative of the presence of an interpersonal level in the underlying representation of (8A). In (9B) that refers back to the situation in the external world that is described within (9A). This purely semantic reference shows that the underlying representation of (9A) contains a representational level of organization. The anaphoric references in (10B) and (IIB) are of a different nature since they are metalinguistic in nature. They are instances of REFLEXIVE LANGUAGE (Lucy ed. 1993) or MESSAGES ABOUT THE CODE (Jakobson 1971). In (10B) that does not refer to the entity described by chuletas de cordero but to the phrase 'chuletas de cordero' as such. This phrase is a morphosyntactic unit, hence the conclusion must be that this phrase is pre-

60

Kees Hengeveld

sent in underlying structure and can therefore function as an antecedent for anaphoric reference. A similar line of reasoning can be set up for the anaphoric reference in (IIB), the only difference being that here the antecedent is a phonological rather than a morphosyntactic unit. From these facts it may be concluded that the underlying representation of an utterance contains four levels of organization: an interpersonal (pragmatic), a representational (semantic), a structural (morphosyntactic), and a phonological level. Note that all these levels are purely linguistic in nature. This holds for the interpersonal and representational level too: these levels describe language in terms of its functions, but only in so far as these functions are encoded in the grammar of a language. Thus the interpersonal level represents a linguistic unit in terms of its communicative function, and the representational level in terms of its ontological status.

3. General architecture of FDG 3.1. Overall organization The general architecture of FDG may now be represented as in figure 1 at page 61, in which the grammatical component is presented in the centre, the conceptual component at the top, the output component at the bottom, and the contextual component to the right. Within the various components, circles contain OPERATIONS, boxes contain the PRIMITIVES used in operations, and rectangles contain the LEVELS OF REPRESENTATION produced by operations. In line with the topdown organization of FDG, I start my discussion of figure 1 at the top. As mentioned in 2.4, at the prelinguistic conceptual level a communicative intention (e.g. issuing a warning) and the corresponding mental representations (e.g. of the event causing danger) are relevant. Through the operation of formulation these conceptual representations are translated into pragmatic and semantic representations at the interpersonal and representational level, respectively. The mapping rules (see 2.2) used in the operation of formulation are language-specific, i.e. FDG does not presuppose the existence of universal pragmatic and semantic notions. As a result, similar conceptual representations may receive different pragmatic and semantic representations in different languages. To give just one example: warnings

Dynamic expression in Functional Discourse Grammar

Figure 1. General layout of FDG

61

62

Kees Hengeveld

are in some languages encoded as a distinct type of speech act, whereas in others they receive the same treatment as orders. This type of crosslinguistic variation may be expected to be governed by typological hierarchies, just as morphosyntactic and phonological variation. Mapping rules make use of a set of primitives that contains frames, lexemes, and primary operators. Primitives will be discussed in more detail in 3.3. below. The representations at the interpersonal and representational levels are translated into a morphosyntactic representation at the structural level through the operation of morphosyntactic encoding. The morphosyntactic expression rules draw on a set of primitives containing templates, auxiliaries, and (morphological) secondary operators (see 3.3). Similarly, the representations at the pragmatic, semantic, and structural level are translated into a phonological representation at the phonological level. The phonological expression rules draw on a set of primitives containing prosodic patterns, morphemes, and (phonological) secondary operators (see 3.3). By organizing the grammatical component in this way, FDG takes the functional approach to language to its logical extreme: within the top-down organization of the grammar, pragmatics governs semantics, pragmatics and semantics govern morphosyntax, and pragmatics, semantics and morphosyntax govern phonology. The phonological level of representation is the input to the operation of articulation, which, in the case of an acoustic output component, contains the phonological rules necessary for arriving at an adequate phonetic utterance. Articulation takes place outside the grammar proper. The various levels of representation within the grammar feed into the contextual component, thus enabling subsequent reference to the various kinds of entity relevant at each of these levels once they are introduced into the discourse. The contextual component feeds into the operation of formulation, so that the availability of antecedents and visible referents may influence the composition of (subsequent) discourse acts. Note that the representation of these feeding relations in figure 1 is a simplification when looked at from the perspective of the language user. In order to create a contextual specification, the addressee has to reconstruct all the levels of representation within the grammar on the basis of the actual output of that grammar, i.e. the phonetic utterance. Since in this paper I restrict myself to the perspective of language production and concentrate on the grammatical component, I abstract away from this complication by provisionally assuming direct feeding relationships between the grammatical and the contextual components.

Dynamic expression in Functional Discourse Grammar

63

3.2. Levels and Layers 3.2.1. Introduction Each of the levels of representation distinguished within the grammatical component in figure 1 is structured in its own way. What all the levels have in common is that they have a hierarchically ordered layered organization. They differ in the sense that at each level a linguistic expression is analyzed in terms of the distinctions relevant to that level. It should be stressed again that the representations at all levels are purely linguistic in nature, so that only those distinctions are provided that are actually reflected in the grammar of the language involved. Note furthermore that the representations below are not exhaustive: there are higher levels of linguistic organization not captured here.

3.2.2. The interpersonal level At the interpersonal level the hierarchical structure given in figure 2 applies.

(M,: [(A,: [ILL (P,)s (P2)A (C,: [...(T,) (R,)...] (CO)] (A,))] (M,)) Figure 2. The interpersonal level

As argued in 2.3.3. one of the units of analysis at the interpersonal level is the move (M), which may contain one or more discourse acts (A). Each discourse act is organized on the basis of an illocutionary frame (ILL), which has two speech act participants (P, the speaker S and the addressee A) and the communicated content C as its arguments. The communicated content contains a varying number of ascriptive (T) and referential (R) acts. Note that the latter two units are operative at the same layer, i.e. there is no hierarchical relation between them. In general, then, at the interpersonal level units are analysed in terms of their communicative function.

64

Kees Hengeveld

3.2.3. The representational

level

At the representational level the layers presented in figure 3 are relevant.

(ep,: [(p,: [(e,: [(f.) (x,)] (e,))] (Pi))] (ep,)) Figure 3. The representational level At this level of analysis linguistic units are described in terms of the entity type they designate (see Hengeveld 1989, 2004; Mackenzie fc.). These entity types are of different orders: third-order entities or prepositional contents (p); second-order entities or states of affairs (e); first-order entities or individuals (x); and zero-order entities or properties (f). In narratives, propositions may furthermore be joined into episodes (ep). Note that firstorder and zero-order entities belong to the same layer, i.e. there is no hierarchical relation between them. 4 The nature of an entity type is not indicative of the way the linguistic unit describing that entity is used within a discourse act. Entity types are categories, not functions. The functional analysis is given at the pragmatic level. Thus, the same property (f) may be either ascribed (T) to an entity, or it may be referred to (R). The following examples illustrate this point: (12)

a. Sheila is tall. (Ascription of zero-order entity: T/f) b. Tallness impresses me. (Reference to zero-order entity: R/f)

Similarly, a first-order entity type may be ascribed or referred to: (13)

a. Sheila is my best friend. (Ascription of first-order entity: T/x) b. My best friend visited me last night. (Reference to first-order entity: R/x)

Dynamic expression in Functional Discourse Grammar 3.2.4. The structural

65

level

The more one moves down through the model, the more language-specific the levels become. At the structural level, for instance, adpositional phrases will be relevant for some languages, but not for others; some languages will be of the isolating morphological type, others agglutinative. Figure 4 serves as an example of a simple constituent structure representation. In section 5 below some specific examples of morphosyntactic representations will be given.

[[[[lexeme A dj]ModP l e x e m e N ] R e f P [ l e x e m e v [lexemeAdv]Modp]predp]cL]s

Figure 4. The structural level (example) There is no necessary one-to-one mapping between semantic and pragmatic units on the one hand and morphosyntactic units on the other. As argued earlier, discourse acts may be expressed as sentences, clauses, phrases, or words. To give another example: semantic predications consisting of a unit designating a (zero-order) relation and two units designating (first order) individuals may be realized in one language as a clause with three constituents and in others as a single word. Consider the following examples: (14)

I made shirts.

(15)

Southern Tiwa (Gerdts 1998: 88)5 Te-shut-pe-ban. 1 .SG>PL-shirt-make-PAST Ί made (the) shirts.'

The English sentence in (14) can be subdivided into three constituents corresponding to the three semantic units mentioned earlier: a unit designating a relation (made) and two units designating individuals (I, shirts). The same semantic configuration is expressed in Southern Tiwa as a single word. The Agent argument is expressed by means of a prefix on the verb and does not have to be expressed independently. The Patient argument is incorporated into the verb. The fact that the patient is cross-referenced on the verb shows that it is really an argument of that verb. These examples

66

Kees Hengeveld

thus clearly show that there are many possible mappings between the semantic and the structural level. I refer to Smit (this volume) for an in-depth discussion of noun incorporation in FDG, and to section 5 below for a full analysis of example (15).

3.2.5. The phonological level The phonological level is equally language-specific, and contains both the segmental and the suprasegmental phonological representation of a construction. In figure 5 a simple example is given, which contains segments (x), accent positions (capitals) and an indication of the prosodic contour 0).

/ xxx#XXX#xxx#XXX \ / Figure 5. The phonological level (example)

Again, there is no necessary one-to-one mapping between pragmatic, semantic, and morphosyntactic units on the one hand, and phonological units on the other. Thus, in some languages subordinate clauses are set off from the main clause by means of an intonation break, whereas in others they form a single intonation unit with the main clause. To give another and perhaps more pervasive example: the foot structure relevant at the phonological level does not respect word or constituent boundaries at the morphosyntactic level.

3.3. Primitives 3.3.1. Introduction The various operations creating the levels just discussed make use of sets of primitives which serve as the building blocks for their respective levels of application. Together these sets of primitives constitute the FUND6 of the grammar. The rules that constitute the operations within the grammar com-

Dynamic expression in Functional Discourse Grammar

67

bine these primitives in order to produce the various levels of representation.

3.3.2. Primitives used in Formulation The operation of formulation has to produce two different levels of representation: the interpersonal and the representational levels. For each of these levels, similar primitives are relevant. First of all, the set of primitives contains FRAMES which define the possible combinations of elements at the interpersonal and representational levels for a certain language. Despite their language-specific nature, the inventory of frames is expected to be partly predictable in terms of crosslinguistically valid typological hierarchies. Relevant distinctions captured by frames at the interpersonal level include coded illocutions and coded information structure configurations. At the representational level possible quantitative and qualitative valencies and possible modification structures are relevant, among others. These issues have been addressed briefly in 2.2. Secondly, this set of primitives contains LEXEMES. Within the set of lexemes a distinction is made between those that function at the interpersonal level (e.g. interjections, proper names, illocutionary adverbs, performative expressions, etc.) and those that function at the representational level. Lexemes are not stored in predicate frames, as they are in FG, but are independent units that have to be associated with the aforementioned frames through their meaning definitions (see Garcia Velasco & Hengeveld (2002) for further discussion of this issue). In the implementation of the grammar the frames are selected first, and only after that are lexemes inserted. This reflects the choice the speaker often has in describing one and the same entity through a variety of lexemes with different connotations and/or denotations. Thirdly, this set of primitives contains PRIMARY OPERATORS, which represent grammatical expressions purely in terms of their pragmatic or semantic content. The classification of operators in terms of the layer at which they apply is common in FG, and is also relevant for FDG. The classification of these operators will be addressed extensively in Hengeveld & Mackenzie (fc.). Here it will suffice to give a number of examples. At the interpersonal level, MITIGATION is an operator at the level of the illocution, REPORTATIVE is an operator at the level of the communicated content,

68

Kees Hengeveld

APPROXIMATION ('sort-of) is an operator at the level of the ascriptive act, and DEFINITENESS operates at the level of the referential act. At the representational level, examples are SUBJECTIVE MODALITY at the third-order layer, OBJECTIVE MODALITY and TENSE at the second-order layer, NUMBER at the first-order layer and e.g. PHASAL ASPECT at the zero-order

layer.

3.3.3. Primitives used in Morphosyntactic Encoding The structural level is organized on the basis of templates for words, phrases, clauses, and sentences, which are stored as part of the set of primitives relevant for the operation of morphosyntactic encoding. The inventory of templates has to be specified for each language individually, although again the expectation is that cross-linguistically valid generalizations will make this inventory largely predictable on the basis of a limited number of parameters. The second set of primitives relevant at the morphosyntactic level consists of free grammatical morphemes, i.e. words that express a grammatical meaning, such as auxiliaries and grammatical particles. These free grammatical morphemes have to be introduced at the structural level, since, unlike bound grammatical morphemes, they occupy slots in the syntactic configuration, which is determined at this level. For instance, in Dutch the main verb normally occurs in second position in a clause, but when an auxiliary verb is present, this auxiliary occupies the second position and the main verb occurs in final position, as illustrated in (15)-(16): (15)

Karel won de wedstrijd. Karel win.PST.SG DEF game 'Karel won the game.'

(16) Karel heeft de wedstrijd Karel have.PRS.3.SG DEF game 'Karel has won the game.'

gewonnen. win.PST.PRT

Examples like these clearly show that it is impossible to determine the order of constituents without taking free grammatical morphemes into account.

Dynamic expression in Functional Discourse Grammar

69

The third set of primitives relevant at the structural level consists of (morphosyntactic) secondary operators. 7 These operators anticipate morphological means of expression, the form of which, in languages in which they are relevant, will eventually be selected at the phonological level. Morphosyntactic secondary operators are by their very nature not identical to primary operators. Often, various semantic distinctions captured by primary operators map onto a single morphosyntactic primary operator. For instance, the Accusative case in a certain language may be triggered by the semantic function patient, but also by various types of adjunct, or it may be lexically triggered by certain verbs or adpositions. The other way around, a single semantic category may map onto various morphosyntactic categories, as when the form of the accusative when expressing the Patient argument is dependent on the noun class of the head of that Patient argument. Secondary operators at the morphosyntactic level can thus be considered to be the names or addresses of actual forms or sets of forms. In assigning names to (sets of) forms I will follow the convention proposed in Comrie (1976), in which general semantic categories are written in lower case, whereas forms in specific languages start with a capital letter. Thus, the imperfective past is expressed in French by means of the Imparfait. It is important to realize that the names of forms could just as well be represented by numerical codes, like 581, for the Imparfait First Person Plural. Since this would not enhance readability, more mnemonic labels are chosen, but this should not obscure the fact that at this level we are only interested in labelling specific forms.

3.3.4. Primitives used in Phonological

Encoding

The first set of primitives relevant for the constitution of the phonological level consists of prosodic patterns, which organize the phonological information coming in from higher levels into coherent blocks. The second set of primitives consists of the actual phonemic bound morphemes that correspond to the primary or secondary operators that have been specified at the higher levels of organization. Bound grammatical morphemes are introduced at the phonological level since in many languages the form of grammatical morphemes may be affected by the syntactic configuration in which they occur. Bakker (this volume) cites various examples of this phenomenon. I repeat his example (4) here as (17):

70

Kees Hengeveld

(17) Yagua (Payne 1990:30) a. Sa-juuy Anita. 3SG-fall Anita 'Anita fell.' b. Anita 0-juuy. Anita 3.SG-fall 'Anita fell.' In Yagua, the subject-agreement prefix on the verb is sa- when the subject term occurs in postverbal position (17a), but it is 0 - when the subject occurs in preverbal position (17b). This means that in this language the form of the secondary operator Subj.3.Sg can only be determined after establishing the constituent order of the sentence. A third set of primitives potentially relevant at the phonological level consists of (phonological) secondary operators. These anticipate acoustic (signed, orthographic) means of expression that are not a direct reflection of a primary operator. A good example of the kind of phenomenon for which such secondary operators are necessary is unit accentuation in Danish noun incorporation (Nedergaard Thomsen 1992). In cases of noun incorporation in Danish the last element retains its inherent stress ('), but the first element is realized with reduced stress (0), as in the following exam-

(18) Danish (Nedergaard Thomsen 1992: 182) De /angede 'SOMMERFLUGE they catch.PAST butterfly.PL 'They caught butterflies.' This example shows that the process of syntactic noun incorporation has phonological effects. These effects are captured by means of the introduction of (phonological) secondary operators at the phonological level.

3.3.5. Generalizations There are certain correspondences across the three sets of primitives. Within each set there is a subset of units with a structuring function: the frames used in formulation, the templates in morphosyntactic encoding,

Dynamic expression in Functional Discourse Grammar

71

and the prosodic patterns used in phonological encoding all serve the purpose of providing an overall organizing structure for their respective levels. Within each set of primitives there is furthermore a subset of units in phonemic form: the lexemes used in formulation, the auxiliaries used in morphosyntactic encoding, and the bound morphemes used in phonological encoding all contribute to the cumulative segmental specification of the underlying representations. Finally, within each set of primitives there is a subset of operators: primary operators are relevant to the operation of formulation, secondary operators to the operation of encoding.

3.4. Levels and primitives For a simple illustration of how a single constituent gets different representations at each level, using different sets of primitives, consider the example in (19). The constituent these apples is represented in four different ways within FDG, as indicated in (20). (19)

(I like) these apples.

(20)

a. b. c. d.

(id Ri) (prox m X;: /£epl/N (x,)) [[/öis/-Plural]Det [/ffipl/-Plural]N]Ref?hr /Si:z#£eplz/

At the interpersonal level (20a), the constituent is characterized as having a referential function (R). The referent is furthermore assumed to be identifiable (id) by the addressee. At the representational level (20b) the constituent is characterized as designating a first-order entity (x) located near the speech location (prox), and consisting of more than one unit (m). The nominal (N) lexeme /aspl/ specifies a basic property of this entity. At the structural level the constituent is characterized as being a Referential Phrase (RefPhr), which consists of a determiner (Det) and a head noun (N). At this level the free grammatical morpheme /8is/ is introduced, since it has to occupy a syntactic slot. The primary operator m is converted into the secondary operator Plural, which occurs twice, since it has to be expressed on each of the two words making up the Referential Phrase. At the phonological level the appropriate Plural forms of the words are introduced, in

72

Kees Hengeveld

the case of the noun by adding the plural suffix, in the case of the determiner by selecting the appropriate suppletive form.

3.5. Functions In FG functions play an important role: semantic functions are part of FG predicate frames, pragmatic and syntactic functions are assigned to constituents. In FDG functions are part of the various structuring devices mentioned in the previous paragraph: pragmatic functions are part of interpersonal frames, semantic functions are included in representational frames, and syntactic functions, in languages in which they are relevant, are part of the morphosyntactic clause templates. Syntactic functions are thus no longer considered to be perspectivizing in nature, as they are in FG. Rather, they are matched to pragmatic and semantic units as part of the encoding operation. The pivotal nature of syntactic functions can thus be attributed to the semantic and pragmatic factors that trigger their occurrence.

4. Dynamic implementation 4.1. Introduction The idea of a dynamic implementation of the grammar, as proposed in Bakker (2001, this volume) is highly compatible with the basic architecture of Functional Discourse Grammar, particularly its top-down organization. I propose that in a dynamic implementation of FDG the two principles described in the following sections be taken into account. These principles are not sufficient to account for the entire process of language generation, but I focus on these principles here since they are specific to FDG.

4.2. Depth first The depth-first principle was proposed in Bakker (1999) but gets a somewhat different interpretation in FDG. In defining its role within the grammar, recall that a basic assumption in FDG is that a grammatical production model will be more efficient the more it resembles language production in the individual. In accordance with this assumption, informa-

Dynamic expression in Functional Discourse Grammar

73

tion from a certain level is sent down to a lower level as soon as the necessary input information for that lower level is complete. The grammar would slow down considerably if first the interpersonal level had to be fully specified, and then the representational level had to be filled in completely, so that only then could the morphosyntactic configuration be determined, which after that would be mapped onto a phonological configuration. This is not how language production in the individual works, and it would therefore, given the basic assumption mentioned above, not lead to a very efficient model of grammar either. As an example, consider the effect of specifying an illocutionary value at the interpersonal level. As soon as an imperative (IMP) frame has been selected for the discourse act, there are potentially important consequences at all subsequent levels of representation: (i) at the representational level, the event frame will have to designate a controlled event, and the first argument will have to include the addressee; (ii) at the structural level, in some languages a specific constituent order is used, or there may be special imperative auxiliaries or morphological markers; (iii) at the phonological level, there may be specific prosodic patterns that are used with imperatives. All this means that the selection of an imperative frame at the interpersonal level may trigger a whole range of specifications at subsequent levels, both in terms of formulation and of encoding, irrespective of the specification of further elements at the interpersonal and lower levels.

4.3. Maximal depth The principle of maximal depth states that only those levels of representation that are relevant for the building-up of (a certain aspect of) an utterance are used in the production of that (aspect of the) utterance. This principle, too, is meant to speed up the implementation of the grammar. It avoids the vacuous specification of levels of representation that are irrelevant to the production of the utterance at hand. Following up on the example in the previous section, this means that in a certain language there may be a direct connection, circumventing the representational level, between the interpersonal and the structural levels in those cases in which the imperative frame has to be mapped onto a specific clausal template. Similarly, there may be a direct connection, circumventing the representational and the structural level, between the interpersonal and the phonological levels when the imperative frame is mapped onto a

74

Kees Hengeveld

specific prosodic pattern. In this way, superfluous steps in passing on information within the top-down procedure are avoided. Looking at this from a bottom-up perspective, it means that the expression of underlying structures is potentially based on information from all higher levels, not just from the next higher one.

4.4. Pathways through the grammar In view of the principles discussed in the preceding sections, the pathways through the grammar may be represented as in figure 6 at page 75. The horizontal arrows 1, 8, and 11 concern the consultation of the sets of primitives by the various operations. The dynamic implementation of the model is shown vertically. Examples of possible (partial) pathways through the grammar are the following. 1 - > 2 - > 5 - > 11 12: This pathway will be used when, for instance, the illocutionary value of a discourse act is expressed through prosodic means. Note that in this case the representational and structural levels do not play a role in the generation process. l-»2-»4-»8-»9-»10: Similarly, this pathway is used to map the illocutionary value of a discourse act onto a specific morphosyn tactic sentence template. In this case the representational level is circumvented. 1 —> 3 —> 6 —> 8 —> 9 —> 10: This partial pathway will for instance be used when a two-place predication frame has to be mapped onto a transitive clause template. Here the interpersonal level is irrelevant. 1 3 7 11 12: This pathway will be used in those cases in which a semantic distinction is expressed through a distinct prosodic pattern.

Dynamic expression in Functional Discourse Grammar

Figure 6. Pathways through the grammar

75

76

Kees Hengeveld

5. One-word utterances in FDG 5.1. Introduction Let me now turn to the analysis of some concrete examples. The ones I present below have in common that they all consist of a single word: a verb form constituting a full main clause in Spanish (5.2); a verb form with incorporated object constituting a full main clause in Southern Tiwa; and an interjection constituting an independent utterance in English.

5.2. Spanish Spanish is a language with a strong tendency towards morphological fusion. The subject of a sentence does not have to be expressed lexically when it is topical in nature. It is, however, expressed on the verb, often in a portmanteau expression together with mood, tense and aspect. These properties are illustrated in (21). (21)

Lleg-ό. arrive-IND.PAST.PF.3.SG 'He/she/it arrived.'

Note that the suffix -o in (21) is special in the sense that it attracts word stress. The main challenges here are (i) the fact that two semantic constituents are expressed in a single syntactic constituent, (ii) the fusional nature of the suffix, and (iii) the stress pattern. Figure 7 at page 77 is a static representation of the various steps that have to be taken in order to generate (21). In a dynamic description of figure 7, the following steps may be distinguished.8 1 —» 2: Selection and insertion of the declarative act frame with topical R. Such a predefined frame is relevant for Spanish given the effects topicality has on the expression of the subject.

Dynamic expression in Functional Discourse Grammar

(A 1 :[Decl(P 1 )s(P 2 )A(C 1 :

11

Formulation

[(Τ,) (R,)TOP] (CO)] (A,))

(e,: [(f,) (xi)Ag] (e,)) //Ceg/v

(A,: [Deel (P,) s (Pj)A (C[:[(Tr) (R,)Top] (CO)] (A,))

past pf (past ei : [(pf f i : /Xeg/V (f|)) (1 x,: 3 (Xi))Ag] (eO) 6 [Stem v -TamSubj Sx ] IndPastPßSg

9

ιr

[/Xeg/-IndPastPf3Sg]

10

ιr

/..../ \

Phonological Encoding 12

/-O/ /Xeg-O/ \

Figure 7. Analysis of example (21)

7

78

Kees Hengeveld

5 —» 11 —» 12: Selection and insertion of the prosodic pattern with falling intonation used for declarative sentences, triggered by the presence of the declarative frame at the interpersonal level. 1 —> 3: Selection and insertion of one-place agentive event frame. 1 —» 3: Selection and insertion of lexemes. Note that I use '3' as a shorthand representation of the abstract semantic content 'third person'. The formulator selects this abstract item rather than a lexical specification of (x,) in view of the topicality of R at the interpersonal level. 1 —> 3: Selection and insertion of the primary operators for past tense (past), perfective aspect (pf) and singularity (1). 6 —» 8 —» 9: Selection and insertion of the word template for verbs. Note that this template can only be selected at this stage, after the insertion of a verb rather than a non-verb in the property slot at the representational level, given that in Spanish verbal and non-verbal constructions have different formal properties. 6 —» 9: Insertion of the verb specified at the representational level into the Stem slot in the word template for verb forms. 4/6 —» 8 —» 9: Selection and insertion of the secondary operator IndPastPßSg into the TamSubj slot in the word template for verb forms, triggered by the decl illocution at the interpersonal level, the operators past, perfective, and singular at the representational level, and the third person specification of the single argument at the representational level. The fact that a whole range of pragmatic and semantic distinctions has to be mapped onto a single form characterizes fusional languages.

Dynamic expression in Functional Discourse Grammar

79

1 0 - > 12: Insertion of the verb form into the prosodic pattern waiting to be filled at the phonological level. 11 12: Selection and insertion of the morpheme I-01 into the suffix slot of the verb form. Note that this suffix brings along inherent stress, indicated by capitalization.

5.3. Southern Tiwa Southern Tiwa is a language that is both polysynthetic and agglutinating. The polysynthetic nature of the language is manifested, among other things, in the possibility of incorporating nouns into verb forms. Example (22) was discussed earlier in 3.2.4. (22)

Te-shut-pe-ban. 1 .SG>PL-shirt-make-PAST ' I m a d e (the) shirts.'

The main challenge here is the fact that three semantic constituents are expressed as a single syntactic constituent. The various steps that have to be taken are indicated in figure 8 at page 80 and can be listed as follows. 1 ->2: Selection and insertion of declarative act frame with one ascriptive and two referential subacts. 1 ->3: Selection and insertion of the two-place agentive event frame. 1 ->3: Selection and insertion of lexemes. Note that I use S here as a shorthand representation for the first person. 1 —» 3: Selection and insertion of the primary operators for past tense (past) and plurality (m).

80

Kees Hengeveld

( Α , : [Deel (P,)s ( P 2 ) A ( C , :

Formulation

[(TO (R,) (R2)] (C,))] (A,)) ( e , : [ ( f O ( x , ) A g (x2)p»t] ( e O )

/pe/v

( A , : [Deel (P,)s (P,)A (C,:[(Ti) (R,) (R,)] (CO)]

/shut/N

(A,))

S past m

( p a s t e i : [(f i : /pe/v ( ί ) )

S (χΟ)α 8

/ s h u t / N (Xj))pat] ( e O )

[Agrpx-[Pat]RefPhr-StemvTenseSx] 9 lSg>Pl Past

[ 1S g>Pl- [/shut/]

r

Refphr-/pe/-Past]

10 ι Phonological

12

/ti-/ /-ban/ /ti-shut-pe-ban/

Figure 8. A n a l y s i s

of example (22)

r

Encoding

Dynamic expression in Functional Discourse Grammar

81

6 —» 8 —» 9: Selection and insertion of the word template for verb forms with incorporated patient argument. 6 —» 9: Insertion of the verb specified at the representational level into the Stem slot in the word template for verb forms, and of the patient argument at the representational level into the patient slot within the same template. 6 —» 8 —» 9: Selection and insertion of the secondary operator 1 Sg>Pl into the Agr slot and Past in the Tense slot within the word template for verb forms. The secondary operator 1 Sg>Pl is triggered by the presence of a first singular agent argument and a plural patient argument at the representational level. The secondary operator past is triggered by the presence of a past operator at the representational level. Note that, unlike in Spanish, there is a straightforward matching between the primary operator past and the secondary operator Past. This characterizes agglutinating languages. 10 —> 12:

Insertion of the verb form into the prosodic pattern9 waiting to be filled at the phonological level. 11 -> 12: Selection and insertion of the morphemes /ti-/ and /-ban/ into the appropriate slots of the verb template.

5.4. English The last example is quite the opposite of the two preceding ones. In the examples from Spanish and Southern Tiwa the challenge was to explain the complex internal structure of words, i.e. the fact that various pragmatic and semantic units are mapped onto a single word form. In this last example the challenge is to explain the fact that a word that is capable of functioning as a complete and independent discourse act has no internal complexity at all. The example is given in (23), and the analysis in figure 9. (23)

Congratulations!

82

Kees Hengeveld

(A^ExprCP.MP^CC,)] (A,)) /kangratjuleijanz/in,

Formulation

(A,: [Expr(POs ( P , ) a ( Q : [/kangrstjuleijanz/int] (Q))] (AO)

9 ιr

10 ιr

/

/! 12

/kangraetjuleijanz/!

Figure 9. Analysis of example (23)

Dynamic expression in Functional Discourse Grammar

83

In figure 9 at page 82 the following steps are represented. 1 —> 2: Selection and insertion of an expressive act frame. 5 —» 11 —» 12: Selection and insertion of the distinctive prosodic pattern that characterizes expressive discourse acts. 1 ->2: Selection and insertion of the lexeme /IrangraetjuleιJsnz/ into the C slot of the act frame at the interpersonal level. 5 —> 12: Insertion of this lexeme into the prosodic pattern.In this analysis interjections like congratulations are analyzed as ready-made lexical contents of discourse acts which do not have semantic content, only pragmatic content. Since they have no morphosyntactic structure either, this means that only the interpersonal and the phonological levels of representation are relevant for their analysis.

6.

Conclusion

In this paper, after giving an outline of an updated version of FDG, I have argued that the idea of a dynamic implementation, as proposed in Bakker (1999, 2001, this volume) within the context of FG, can be fruitfully applied within the context of FDG as well. An advantage of FDG over FG is that mismatches between pragmatic, semantic, morphosyntactic and phonological units of analysis can be handled relatively easily, due to its modular organization. In order to demonstrate this, I have analyzed a number of one-word utterances which illustrated various matches and mismatches between the various levels of linguistic organization in a dynamic implementation of FDG·

84

Kees Hengeveld

Notes 1. 2. 3.

4. 5. 6.

7. 8.

9.

I am grateful to Matthew Anstey, Lachlan Mackenzie and Gerry Wanders for comments on an earlier version of this paper. For a more elaborate discussion of these constructions in the context of FG see Hengeveld (1997). In several languages this construction type grammaticalizes further in the sense that a generic verb is used as the head of a new paragraph. See e.g. de Vries (1989: 207) on generic verb linkage in Kombai. Although at the representational level there is a valency relation that is absent at the interpersonal level. The gloss 1 ,SG>PL should be read as ' 1 .SG acting on plural patient'. The notion of 'fund' was introduced in the context of FG in Dik (1980), where it was used as a notion with a wider meaning than 'lexicon', including processes of term formation and predicate formation. Here the notion is stretched a bit further to include all building blocks relevant to the grammar of a particular language. Secondary operators correspond to Dik's (1997) 'μ-operators' and de Groot's (1990) 'secondary triggers'. Note that I follow the standard FG convention in which non-instantiated variables have a subscripted number, and instantiated variables a subscripted letter. Since the actual prosodic patterns of Southern Tiwa are unknown to me, I refrain from representing it in figure 8.

References Bakker, Dik 1999 FG expression rules: from templates to constituent structure. Working Papers in Functional Grammar 67, University of Amsterdam. 2001 The FG expression rules: a dynamic model. In Challenges and Developments in Functional Grammar. Revista Canaria de Estudios Ingleses, Maria Jesüs Perez Quintero (ed.), 15-54. La Laguna: Universidad de La Laguna. Comrie, Bernard 1976 Aspect. Cambridge: Cambridge University Press.

Dynamic expression in Functional Discourse Grammar

85

Dik, Simon C. 1980 Studies in Functional Grammar. London: Academic Press. 1997 The Theory of Functional Grammar 1: The Structure of the Clause. edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Foley, William A. 1986 The Papuan Languages of New Guinea. Cambridge: Cambridge University Press. Garcia Velasco, Daniel and Kees Hengeveld 2002 Do we need predicate frames? In New perspectives on argument structure in Functional Grammar, Ricardo Mairal Uson and Maria Jesüs Perez Quintero (eds.), 95-123. Berlin: Mouton de Gruyter. Gerdts, Donna B. 1998 Incorporation. In The handbook of morphology, Andrew Spencer and Arnold M. Zwicky (eds.), 84-100. Oxford: Blackwell. Groot, Casper de 1990 Morphology and the typology of expression rules. In Working with Functional Grammar, Mike Hannay and Elseline Vester (eds.), 187201. Dordrecht: Foris. Hannay, Mike and Caroline Kroon fc. Acts and the relationship between discourse and grammar. Hengeveld, Kees 1989 Layers and operators in Functional Grammar. Journal of Linguistics 25.1, 127-157. 1997 Cohesion in Functional Grammar. In Discourse and pragmatics in Functional Grammar, Christopher S. Butler, John H. Connolly, Richard A. Gatward and Roel M. Vismans (eds.), 1-16. Berlin: Mouton de Gruyter. 2004a The architecture of a Functional Discourse Grammar. In A New Architecture for Functional Grammar, J. Lachlan Mackenzie and Maria Ä. Gomez-Gonzälez (eds.). Berlin: Mouton de Gruyter. 2004b Epilogue. In A New Architecture for Functional Grammar, J. Lachlan Mackenzie and Maria A. Gomez-Gonzälez (eds.). Berlin: Mouton de Gruyter. fc. State-of-Affairs concepts. To appear in Morphology: A handbook on inflection and word formation, Geert Booij, Christian Lehmann and Joachim Mugdan (eds.). Berlin: Mouton de Gruyter. Hengeveld, Kees and J. Lachlan Mackenzie fc. Functional Discourse Grammar. Oxford: Oxford University Press.

86

Kees Hengeveld

Jakobson, Roman 1971 Shifters, verbal categories, and the Russian verb. In Selected Writings, Vol. II., Roman Jakobson, 130-147. The Hague: Mouton. Kroon, Caroline 1995 Discourse particles in Latin. Amsterdam: Gieben. Levelt, Willem J.M. 1989 Speaking. Cambridge, MA: MIT Press. Lucy, John A. (ed.) 1993 Reflexive language: Reported speech and metapragmatics. Cambridge: Cambridge University Press. Mackenzie, J. Lachlan 1998 The basis of syntax in the holophrase. In Functional Grammar and verbal interaction, Mike Hannay and A. Machtelt Bolkestein (eds.), 267-295. Amsterdam: Benjamins. fc. Entity concepts. To appear in Morphology: A handbook on inflection and word formation, Geert Booij, Christian Lehmann and Joachim Mugdan (eds.). Berlin: Mouton de Gruyter. Mackenzie, J. Lachlan and Maria A. Gomez-Gonzälez (eds.) 2004 A new architecture for Functional Grammar. Berlin: Mouton de Gruyter. Mosel, Ulrike and Even Hovdhaugen 1992 Samoan reference grammar. (Instituttet for Sammenlignende Kulturforskning B85.) Oslo: Scandinavian University Press. Nedergaard Thomsen, Ole 1992 Unit accentuation as an expression device for predicate formation: the case of syntactic noun incorporation in Danish. In Layered structure and reference in a functional perspective, M. Fortescue, P. Harder and L. Kristoffersen (eds.), 173-229. Amsterdam: Benjamins. Payne, Doris L. 1990 The pragmatics of word order: Typological dimensions of verbinitial languages. Berlin: Mouton de Gruyter. Sinclair, John M. and Richard M. Coulthard 1975 Towards an analysis of discourse: The English used by teachers and pupils. London: Oxford University Press. Staden, Miriam van 2000 Tidore: A linguistic description of a language of the North Moluccas. Ph. D. diss., University of Leiden. Vries, Lourens de 1989 Studies in Wambon and Kombai. Ph. D. diss., University of Amsterdam.

Noun incorporation in Functional Discourse Grammar Niels Smit

1.

Introduction

A strictly modular view of grammar is hard to maintain1. A model in which lexicon, pragmatics, semantics, syntax and phonology are completely independent sub-systems that only process the completed output of one another, incapable of interaction, cannot account for many linguistic phenomena in a satisfactory manner. This becomes especially clear in the study of noun incorporation (NI), where all modules interact to produce complex multilexemic words or word-like constructions. Previous research has mostly centred around the contribution of the syntactic module, leading to several formalist proposals for comprehensive treatment of incorporation phenomena (Baker 1988; 1996; Sadock 1985). Except for some isolated studies (Mithun 1984; Velazquez-Castillo 1995) which focus on the contribution of pragmatics and semantics, a comprehensive functional treatment of noun incorporation is a desideratum. In part, this is due to the lack of a suitable functional framework, capable of dealing with the intricate problems of multi-level analysis that the process calls for. The present article aims to prove that Functional Discourse Grammar (FDG) might exactly be the 'tool' that functionalism needs to arrive at a comprehensive treatment of NI. By providing alternative analyses for the four types of deliberate NI mentioned in Mithun (1984), I will show that FDG offers a good way to capture their semantic and pragmatic peculiarities, attributing the various existing types, which involve different constituents at the semantic and pragmatic level, to interactions between different levels of the grammar. Moreover, it will become apparent that an alternative typology of incorporating constructions is called for, different from the one that Mithun proposes. Nevertheless, also this alternative typology

88

Niels Smit

seems to obey Mithun's empirical findings regarding the presence of an implicational hierarchy of NI. The article is divided into eight sections. In section two, the framework of Functional Discourse Grammar will be introduced. Section three contains general properties of noun incorporation, whereas section four presents an overview of previous accounts of the phenomenon, with special attention begin paid to Mithun's typology (1984). In section five, an FDG approach to incorporation is proposed. This approach is then tested in section six, where FDG analyses of each separate type of noun incorporation in Mithun's typology are presented. In section seven, it will be argued that FDG allows for an alternative typology of NI, which has several advantages over Mithun's proposal. A conclusion is reached in section eight.

2. Functional Discourse Grammar Functional Discourse Grammar (FDG) is a functionalist framework, characterised by an intensive usage of formalism at four different levels of analysis. Merging the best features of other frameworks, FDG combines a top-down modular approach with procedures of upward layering within each separate level.

2.1. An outline of FDG architecture FDG regards linguistic competence as intimately linked to three other modules of cognition: these are the conceptual component, the contextual component, and the acoustic component. The latter is the most straightforward, and serves to lend phonetic shape to the output of the linguistic component. The contextual component (in its current conception) contains both linguistic and perceptual information relating to the ongoing and immediately preceding discourse, thus facilitating the usage of anaphora and the distribution of pragmatic functions, inter alia. The conceptual component is the place where communicative intentions, which serve as input for the linguistic component, are formed. As such, the conceptual component contains very diverse kinds of information, all of which are to be thought of as pre-linguistic, however. Here we find a speaker's encyclopaedic world knowledge, his communicative competence, his knowledge regarding interlocutors and preceding conversations, etc.

Noun incorporation in Functional Discourse Grammar

89

The internal organisation of the linguistic component encompasses four levels of analysis: these are the interpersonal level, the representational level, the structural level and the phonological level, all of which will be discussed separately below. All levels are structured into several layers. A major improvement in comparison with preceding models of Functional Grammar, and one crucial to the FDG framework, is the fact that not all levels of analysis need necessarily be invoked to arrive at a linguistic expression. That is, expressions that only have interpersonal meaning, but no semantic content or structure, like the exclamation Damn!, can 'skip' all irrelevant levels of analysis and pass directly to the phonological level to be given their shape. This immediate processability of higher-order structures is in keeping with findings from psycholinguistics. Speakers do not prepare an entire utterance before speaking; rather, they divide it up into chunks, uttering each chunk as soon as it is ready, a procedure which FDG successfully captures. At all levels of analysis in Functional Discourse Grammar, it is the fund that supplies the necessary building blocks, both lexical and structural. As such, the fund in FDG is considerably more comprehensive than the lexicon in most other theories, which generally contains only lexemes and their subcategorisation frames. The structural items in the fund are diverse in nature, depending on the level of analysis: at the interpersonal and representational levels, the fund supplies frames to order the items at each level, and operators, which convey meanings that are expressed grammatically instead of lexically. At the structural level, the fund supplies templates, which govern word order and the distribution of syntactic functions, as well as secondary operators that, for instance, regulate the expression of agreement. At the phonological level, the fund supplies the prosodic patterns. The insertion of lexical items in the appropriate slots (henceforth termed 'instantiation') is assumed to happen at the highest level possible. That is, as soon as a slot becomes fully specified, the actual lexeme is inserted into it, which is then carried on to the next level. Interjections, for instance, which only serve interpersonal purposes, are inserted in their lexical form at the interpersonal level. The lexical shape of an auxiliary, on the other hand, can only be determined at the structural level, and can hence only be inserted there. Nonetheless, none of the lexical items receives phonological shape before being processed by the phonological component.

90

Niels Smit

2.1.1. The interpersonal level The central unit of analysis at the interpersonal level is the discourse act (A), which Kroon (1995:66) defines as "... the smallest identifiable unit of communicative behaviour." Every act may be characterised in terms of its illocution, and is structured by means of an illocutionary frame, for example 'command' or 'question'. The frame has slots for the discourse participants of the act, usually the speaker (P s ) and the addressee (PA), and the communicated content (C). Communication in FDG is conceived of as ascribing properties to referents; hence, C contains acts of ascription (T) and reference (R). Discourse acts combine into moves (M), which Kroon considers the smallest free unit of discourse that is able to enter into an exchange structure. Thus, a minimal structure on the interpersonal level looks as follows:2 (Mi: [(A,: [ill (P,) s (Pn)A (Q: [(T,) (R,)] (CO)] (AO)] (M,)) Figure 1. The Interpersonal Level (Hengeveld 2004a)

Note that all variables (M, A, C, T, R, P) at the interpersonal level may be modified by operators. Their exact nature has yet to be investigated; in the present article, the only relevant operators at the interpersonal level are d(efiniteness) and spec(ificity) for referential acts.

2.1.2. The representational level The representational level is concerned with semantics. Hengeveld (2004a) notes that "... in transmitting his communicative intention, the speaker in most cases will have to fill his utterance with basic semantic content". This basic semantic content may involve entities of different nature. Following FG (Dik 1997), FDG distinguishes third-order entities (propositions, variable (pO), second-order entities (states of affairs, variable (eO), first-order entities or (individuals, variable (xj) and zero-order entities (properties, variable (f,)). These entity types are ordered strictly hierarchically, so that a maximal representation contains a proposition which contains a state of affairs which contains individuals that have certain properties. As an alternative, however, all entity types may also be expressed non-hierarchically

Noun incorporation in Functional Discourse Grammar

91

through lexical items. The simplest maximal structure at the representational level looks as follows: (π 3 pi: [(7γ2 ej: [π, T (f,) ±R ( Xi ) σ,] (

unitary NI

Figure 4. The Hierarchy of NI (Mithun 1984)

The hierarchy predicts that a language which displays classifier NI necessarily displays all other types as well; a language displaying discoursemanipulative NI must also have case-manipulative and unitary NI, but not necessarily classifier NI, and so on. Mithun motivates her hierarchy of incorporation from a developmental angle; according to her, each type is a natural extension of its predecessor. Unitary NI can be explained by "... a tendency among languages for Vs to coalesce with indefinite direct objects" (Mithun 1984:872; cf. also Hopper & Thompson 1984), which is also present in non-incorporating languages such as Turkish and Hungarian. As shown below, the difference between

102

Niels Smit

NV-coalescence in Turkish (6) and unitary NI in Kusaiean (7) is minimal, especially since both involve bare, unspecified Ns: (6)

Turkish (Mithun 1984:873) Ahmet hergün pipo ic-iyor. ahmet every.day pipe drink-AOR 'Ahmet smokes pipe every day.'

(7)

Kusaiean (Lee 1975:271) Nga twetwe mitmit-lah. lSG sharpen- knife-PRES.PF Ί have knife-sharpened.'

In the intimate compounding that Kusaiean shows, the IncN loses its individual salience and syntactic role. A next step in the development is the creation of a morphological bond between the V and the IncN, as is the case in Yucatec Mayan (1). In case-manipulative NI, the bond between V and IncN has become so strong that the vacated syntactic role can be redistributed to a salient peripheral participant. This process may also be used in discourse to background a given topic (discourse-manipulative NI). Finally, a classificatory system may arise in which the IncN is reduced to a mere qualifier that has lost almost all of its nounhood, while an external Noun refers to the implied participant.

5. Noun incorporation in FDG FDG enables a functional-syntactic approach to NI. In the framework, nouns can head three types of constituents. Inserted in the representation in its minimal form, a noun is associated with a predicate frame (fj), denoting a zero-order entity. When used in its prototypical function to denote a firstorder entity, these property-denoting predicates are embedded in a term (x;) as first restrictor. The full representation of a term is thus (xj: [(fj: [lexeme] (f;))] (xO), with the possibility of modifying xf grammatically (through cooperators) and/or lexically (through 0-restrictors). Zero-order entities cannot be used to refer, and a noun that is not used as head of a term hence cannot be referential. But first-order entities are not necessarily referential, either. That is, referentiality is not assigned at the level of semantics, to which (xj) pertains, but at the interpersonal level

Noun incorporation in Functional Discourse Grammar 103 instead. Thus, while a term typically corresponds with a referential act (Rj) at the interpersonal level, terms also occur that denote a first-order entity, but lack a corresponding referential act. Predicatively used nouns in English are a good example of this. Consider the following: (8)

a. Tom en Ron zijn arts. (PRES e,: [T(IMPF f, : [arts] (f,)) R(x2: [torn] [ron] (x2))0] (e,)) 'Ton and Ron are doctor.' b. Tom and Ron are fine doctors. (PRES e,: [T(IMPF f,: [-R(mx,: [(f2: [doctor] (f 2 ): [fine] (f2))] (x,))]

(f,)) r(X2: [torn] [ron] (x2))J (e,))

c. Tom and Ron hate doctors. (PRES ei: [T(lMPF f,: [hate] (fi)) R(mx2: [torn] [ron] (x2))Ex R (mxj: [doctor] (xi))Go] (e0) (8a) is a Dutch example of a noun used as a predicate. Since it is not embedded in a term, it cannot be pluralized or modified; arts is a singular form. English does not allow for the use of bare nouns as predicates (Mackenzie 1987); instead, they have to be embedded in terms, which are then embedded in a second predicate frame. Because the noun doctor in (8b) is head of a term, it can be pluralized and modified, unlike the noun arts in (8a). However, neither of them is referential, witness their incompatibility with deictic pronouns, for example (Tom and Ron are (*those) doctors). In (8c), the constituent doctors is referential. Consequently, a sentence like Tom and Ron hate those doctors is perfectly acceptable. Since FDG distinguishes between three types of noun-headed constituents ((f), (Xj) and (Rj)), it follows that three types of NI can be distinguished as well, each with its own properties. Incorporation of a predicate noun (fNI) leads to a VN-complex in which the IncN is neither modifiable nor referential, roughly meeting Mithun's criteria for unitary NI. Incorporation of a noun heading a term (XNI) yields a VN-complex in which the IncN is not referential, but can be externally modified.14 Incorporation of a noun heading a referential act (RNI), finally, creates a VN-complex in which the IncN is both modifiable and referential. In all cases, the actual incorporation takes place through the selection of a specific template at the structural level. Thus, f NI and XNI must be viewed as an interaction between the representational and the structural levels, while RNI comes about through in-

104

Niels Smit

teraction between the interpersonal and the structural levels, leaving the representational level completely unaffected. In the next section, some examples of NI will be discussed, following the classification given by Mithun (1984). It will become apparent that an alternative typology, firmly based in FDG theory, is possible. This typology not only accounts for the range of phenomena attested and gives insight into their semantic and pragmatic motivation, but also offers a more theory-inspired explanation for Mithun's implicational hierarchy, which may help to gain understanding about the path along which NI seems to evolve.

6.

Analysis

6.1. Unitary noun incorporation Consider the example below, in which the VN-complex designates a coherent, name-worthy State of Affairs (SoA) in which only a single argument is present: (9)

Kusaiean (Lee 1975:271) a. Nga twem-lah mitmit lSG sharpen-PRES.PF knife Ί have sharpened the dull knife.'

sahfiht dull

sac. DEF

b. Nga twetwe mitmit-lah. lSG sharpen- knife-PRES.PF Ί have knife-sharpened.' The incorporation in (9b) satisfies Mithun's (1984) semantic criteria for unitary noun incorporation. Most importantly, mitmit 'knife' does not function as an argument of the verb twe-, 'sharpen'; instead, the IncN is used as a kind of qualifier, meant to narrow down the meaning of the verb. The VN-complex thus designates an intransitive, de-actualised State of Affairs (SoA).15 Although the verb and the noun do not engage in a morphological bond, there are clear indications that the above is nevertheless a case of incorporation. First, the verb stem in (9b) displays reduplication, which is indicative of a derived intransitive verb (Lee 1975: 264). Second, the verbal suffix -lah is no longer attached to the verb itself, but to mitmit,

Noun incorporation in Functional Discourse Grammar 105 'knife', the noun directly following it, signalling that they form a single constituent. Finally, nothing can be placed between the verb and the noun, and the noun is neither referential 16 nor modifiable. Consider (9c) and (9d): (9)

c. Nga twetwe (*upac) mitmit-lah lSG sharpen diligently knife-PRES.PF Ί have knife-sharpened diligently.'

(upac).

d. Nga twetwe mitmit-lah (*sahfiht). lSG sharpen- knife-PRES.PF dull Ί have knife-sharpened (*the dull).' In (9c), a manner adverb is placed between the verb and the IncN, while (9d) contains an modifying adjective. Both constructions are rejected by speakers of Kusaiean (Lee 1975:271ff).

6.1.1. The interpersonal level In cases of unitary NI, the IncN does not have any interpersonal reality of its own, witness its inability to be bear focus, topic, or any other pragmatic function. 17 It cannot bear (focal) stress, cannot be questioned separately, nor can it surface in cleft constructions or ellipsis. Hence, it is safe to assume that the IncN does not correspond to a referential act at the interpersonal level.18 Instead, only two entities are present in the communicated content (C): a 'rich' ascriptive act (Ti), and a referential act (Ri) that will be used to refer to the agent of the action, in this case the speaker. Since this referent only contains interpersonal, and no semantic information, it can be instantiated immediately: Ri is restricted to the set of discourse entities that meets the selection restriction . By virtue of his presence in the discourse act, the Speaker is definite: therefore, the operator d(efinite) will be added to R,. The final result is (dR,: S (R,)). The ascriptive act (T]) is not further specified at the interpersonal level. The communicative intention of the speaker, which is coded in the Move and can be paraphrased as "Communicate T, about Rj to A ", does not require any specification now. Instead, Ti will be further instantiated at the representational level. Topic-focus distribution among the constituents at the interpersonal level, finally, is irrelevant to an explanation of f NI, since the incorporation

106

Niels Smit

process is not triggered by interpersonal considerations, but at the representational level instead. Hence, Ri can bear topic while T, bears focus, or the other way around, depending on the discourse. Summarising, the interpersonal analysis of (9b) looks as follows:

(INFORM M , : [ ( D E C L A , : [ ( P , ) S (P2)A ( Q :

(dR,: S (Ri)) t 0 p/foc] (C,))]

[(T,)lop/tac

(MO)

Figure 5. Interpersonal Level in Unitary NI

6.1.2. The representational

level

Mithun characterises unitary NI in terms of a clearly defined semantic relationship between the verb and the incorporated noun. According to her, "... incorporated objects indicate unitary, institutionalised activities [...] they simply modify the type of activity under construction." (Mithun 1984:850; emphasis NS). Put differently, IncNs in unitary NI function as modifiers of the verbal predicate. In what follows, the possibilities are explored for a representational analysis of unitary NI in terms of lexical predicate modification.19 In most theoretical accounts, unitary NI is typically analysed as a lexical operation, instances of which are processed by the grammar as ready-made, internally opaque elements. However, two counterarguments against this viewpoint can be raised, as Mithun & Corbett (1999:51ff) point out. First, the process is highly productive. The Kusaiean verb twe-, 'sharpen', for example, can incorporate any noun that satisfies the selection restriction 'sharpen-able'. Second, unitary NI is completely semantically transparent: all IncNs entertain the same semantic relationship with the V in which they are incorporated, namely that of specification or modification. 20 If we want grammar to be a system that accounts for all productive and systematic processes in a language with the use of the smallest possible number of primitives, these are compelling reasons to remove unitary NI from the lexicon, and account for it in the grammar itself. The analysis proposed here advocates a fund consisting of basic lexical primitives, while reference is made to the grammar to account for the formation of productive and semantically transparent complex words.

Noun incorporation in Functional Discourse Grammar 107 Predicate modification in FDG is an undecided matter. Original proposals (Dik 1978; 1997) have put lexical predicate modifiers on a par with arguments, with the exception that modifiers do not fill obligatory slots in a predication frame, but rather are attached to it as 'satellites'. Later proposals (Vet 1986; Hengeveld 1992b) advocate analysing predicate modifiers as secondary restrictors which further delimit the set of potential SoAs that the predication designates, parallel to the function of attributive adjectives in the term phrase. This approach will be followed here. 21 Note that predicates can be modified by means of operators, too. Operators do not qualify to account for NI, however, given the lexical nature of the modifying element: Dik (1997) explicitly restricts the use of operators to the anticipation of grammatical, non-lexical elements. A second issue concerns the layer in which the f-constituent is attached as a restrictor. The literature on the structure of the representational level offers a range of possible locations, depending on the number of layers that is distinguished, and the question whether or not restrictors may only operate on layers, or on predicates as well. 22 Keizer (1992) makes a proposal for the representational level that seems to qualify best for a correct treatment of unitary NI. According to her, it is organised as follows: (tt2 e,: [(π! f,: [π0 [pred]: σ 0 ] (f,): [σ,] (f,)) (χ,) (xn)]: [σ2] (e,)) Figure 6. Representational Level according to Keizer (1992) Keizer diverges from the traditional model in that she assumes the presence of an explicitly non-referential zero-layer, directly attached to the predicate. Since this layer has no variable of its own, but falls inside the scope of the f-variable, reference can only be made to the predicate with its zerolayer operators and restrictors. This is clearly the case for unitary NI: whenever a unitary VN-complex is used, separate reference to either of its constituents is impossible. Keizer's approach needs some modification, however, to make it fit the data on unitary NI. Consider (9c) again; it becomes apparent from Kusaiean word order restrictions that the IncN must be attached to the predicate at a lower level than the manner adverb, given the fact that they are not freely interchangeable. 23 Keizer (1992) assumes that restrictors of manner and degree apply at the zero-level: I will follow Cuvalay instead, who reserves the zero-level for "productive derivational processes" (Cu-

108

Niels Smit

valay 1997: 67), and will claim that restrictors of manner and degree apply at the next higher level.24 Applying the above remarks to the original representation of Keizer (1992) results in the following, which represents the semantics of (9c): (PRES e,: [(PFTf,: [[twe]: ["R(f2: mitmit (f2))]] (f,): [f3: upac ( f 3 ) ] M a n „ e r (fi)) R1 (lxi: 1 (xi))Ag] (e.) Figure 7. Representational Analysis of (9c)

The above representation makes clear that unitary NI can thus be viewed in FG terms as fNI. Both mitmit and upac are attached as restrictors, but on different levels. Due to the former's lack of referentiality, mitmit cannot bear pragmatic functions; due to the absence of an Xj variable, it cannot be modified or pluralized either. Note that the speaker has selected an intransitive predication frame for his representation; it is this intransitivity that will cause the verb to display stem reduplication at the expression level, resulting in the form twetwe. The operators pf and pres anticipate the structural expression of perfect aspect and present tense, while the constituent (xi) will be instantiated as the personal pronoun nga at the structural level. Now, contrast figure (7) with the representational analysis of (9a), the transitive counterpart of (9c): (PRES e,: [(PF f,: [twe] (f,): [upac] (f,)) R1 (lx,: 1 (x,))A| R2 (lx 2 : [(f2: [mitmit] (f 2 ))] (x2))Go] (e,)) Figure 8. Representational Analysis of (9a)

In this representation, mitmit is not a restrictor, but an argument of the verb. As an argument, mitmit is referential (which presupposes the presence of a term, hence the term variable (x2)), and as such is eligible for pragmatic function assignment. Since a transitive predication frame is chosen here, the verb will not display stem reduplication. How to account, though, for the occurrence of a nominal σ0 restrictor on the predicate? This is not in keeping with the functional definition of the lexical category noun, given in Hengeveld (1992a). According to Hengeveld, a noun without additional morphological or syntactic modification can only occur as the first restrictor of a term phrase, and not as a

Noun incorporation in Functional Discourse Grammar 109 (secondary or lower) restrictor on a predicate. In order for it to assume a modifying function, a noun must be subjected to derivational modification^), as in the case for English foolN - fool-ishA - foolish-lymAdv. This is clearly not the case in unitary NI, however, where the IncN is stripped of most of its categoriality features instead. Yet, serious doubt can be cast on the 'nouniness' of the IncN in unitary NI. Unitary NI is most ubiquitous in Oceanic languages, which we generally know to have highly flexible lexical categories. That is, the same lexeme can often be used as a prototypical verb, a prototypical noun and a modifier in the exact same shape. When used as a verb, such a lexeme takes on verbal inflectional morphology; when used as a noun or a modifier, it displays nominal, adjectival or adverbial inflection. Viewed as such, unitary NI might not be noun incorporation at all. Rather, it is more likely analysed as an instance of predicate modification through a flexible lexeme which assumes the function of modifier by virtue of its incorporation into a predicate. In that way, there is considerable resemblance to compounding, where we find items of the same lexical category modifying each other, as in table leg, computer lab, etc. Moreover, the dependent Ν in N-N compounding is also low in categoriality, non-referential and not eligible for modification by ω-operators or φ-restrictors.

6.1.3. Structural

level

As I mentioned in section two, the structural level of grammar is languagespecific, unlike the interpersonal and the representational levels. The constituents and the secondary operators it contains differ across languages. At this point, consider again (1), here repeated as (10): (10)

a. kINCOMP-

inIS-

c'akchop-

0-

ik

3S-

IMPF(TR)

c'akchop-

ce'- 0. tree- I M P F ( I N T R )

ce'. tree

Ί chop a tree.' b. k-

inIsΊ wood-chop.'

INCOMP-

Just like the Kusaiean example, the above is an instance of unitary NI. This time, however, the V and the IncN engage in a tight morphological bond:

110

Niels Smit

they form a single constituent that meets Sadock's criteria of wordhood. The representational analysis of (10), however, is identical to that of unitary NI in Kusaiean: (PRES e,: [(INCOMP IMPF f,: [[c'ak]: [ce']] (f,)) (lx,: 1 (xi))Ag] (e,))25 Figure 9. Representational Analysis of (10b)

Apparently, the difference between the two languages must be captured in terms of differences at the structural level. Where Kusaiean apparently uses a template in which IncN and V have separate word slots, Yucatec Mayan moulds the two elements together in a single word. The possible structures that both languages employ would then look something like the following: Kusaiean (CI,: [(RP,: [S] (RP,))] [(PrP,: [[verb]-[±TRANS]] [[σ0]-[ΤΜΑ]] [σ,] (PrP,))] (CI,)) (CI,: [(RP,: [S] (RP,))] [(PrP,: [[twe]-[lNTR]] [[mitmit]-[PRES.PF]] [upac] (PrP,))] (CI,)) Ψ Nga twetwe mitmitlah upac

Yucatec Mayan (CI,: [(PrP,: [[±cOMP]-[S]-[verb]-[a0]-[TMA]] (PrP,))] (CI,)) (CI,: [(PrP,: [[-COMP]-[lSG]-[c'ak]-[ce']-[lMPF(lNTR)]] (PrP,))] (CI,)) Ψ kinc'akce' Figure 10. Structural Analyses of (9c) and (10b)

In the case of Kusaiean, all the elements that make up the predicate phrase are instantiated in separate words, indicated by the square brackets. In the case of Yucatec Mayan, however, the hyphenated elements are expressed together in one word, indicated by the fact that the group of elements constituting the predicate phrase is closed off by an extra set of brackets. Furthermore, note the presence of the secondary operator [±trans] in the Kusaiean example. The value of this operator (intr) is triggered by the

Noun incorporation in Functional Discourse Grammar 111 presence of only a single argument at the representational level; at the expression level, this operator will cause the stem of the verb to reduplicate. The hyphen between the verb and the transitivity operator indicates that the two will be realised in a single word. 26

6.2. Case-manipulative noun incorporation According to Mithun (1984), instances of NI in which case manipulation plays a role must be understood as a natural extension of unitary NI. Consider the following: (11)

Yucatec Mayan (Mithun 1984:858) a. kinc'ak0- k

ce'

INCOMP- lSG- chop- 3SG-IMPF tree Ί c h o p the tree in m y c o r n f i e l d . '

b. kinc'akce'INCOMP- lSG- chop- treeΊ tree-chop my cornfield.'

icil

in-

in

m y - cornfield

tik TRANS- IMPF

kool.

in- kool. my- cornfield

Mithun claims that (1 lb) is comparable to unitary NI in that the IncN functions as a qualifier of the verb in both cases. The difference between the two is that case-manipulative NI allows another, peripheral participant 27 to occupy the Obj. By virtue of this 'promotion' to a non-oblique syntactic function, the peripheral participant comes to bear greater pragmatic salience. If Mithun is right, the interpersonal analysis of the above example would be similar to that of (9), while the analysis of the representational level would look as follows: (PRES ei: [(INCOMP IMPF f x : [[c'ak]: [ce']] (fi)) ( l x , : 1 (xi)) A g ]: [ ( l x 2 : [(f 2 : [kool] (f 2 ))] (x 2 )) L o c ] (e,))

Figure 11. A Possible Representational Analysis of (1 lb) As can be seen, in-kool 'my cornfield' is attached as a level-2 satellite at the level of the core predication, the appropriate place for operators and satellites that locate a SoA in space and time (cf. Dik 1997a:243ff). Repre-

112

Niels Smit

sented as such, it seems that Mithun is right, and that noun incorporation which leads to a manipulation of the default distribution of syntactic roles is indeed an extension of unitary NI. The promotion of the locative PP inkool to Obj can be assumed to be triggered by a focus-bearing referential act at the interpersonal level. This can be done without any difficulty, since there is no other candidate for Obj assignment; no second argument is present. However, Mithun's analysis does not take all the facts into account. First, her view predicts that the IncN in case-manipulative NI denotes a vague, unspecified referent to which subsequent reference cannot be made at later stages in discourse, given its lack of referentiality. This is not always the case, however: although the IncN may lack both definiteness and specificity, many cases are also attested in which it refers to a particular referent, and can be referred to independently in subsequent discourse. Second, the verbs that can be used for type Π NI are not necessarily semantically empty, as in unitary NI. Moreover, the VN-complex does not denote a unitary concept, but rather seems to be a deliberate, structural choice by speakers to lend syntactic prominence to a participant that is particularly affected by a certain state of affairs, by assigning it Obj function. In other words, case-manipulative NI must be pragmatically inspired. Consider the following Mohawk example: (12)

Mohawk (Mithun 1984:868) a. *Wa- hi'sereht- ohare. PAST- 3SGS/1SG01- carwash 'He car-washed me.' b. WaPAST-

hi3SGS/1SG01-

'sereht- anvhsko. steal car-

'He car-stole me.' This example shows how case-manipulative NI in Mohawk is pragmatically conditioned. (12a) is not accepted, since 'washing' affects the car more than the owner of the car; in (12b), however, the focus is on the speaker himself, being the participant most affected by the stealing of the car. For that purpose, he is expressed through Obj function instead of a PP, which would be the default construction for a beneficiary or possessor. The above considerations call for an alternative analysis of casemanipulative NI. Rather than choosing a bottom-up perspective, as Mithun

Noun incorporation in Functional Discourse Grammar 113 does in assuming that incorporation at the semantic level facilitates foregrounding of an otherwise peripheral participant, I will pursue an analysis that starts on the other side of the matter: given a deliberate choice by the speaker at the interpersonal level to foreground a peripheral participant, 28 a more central participant is denied expression through the syntactic function Obj. As a consequence, the latter is incorporated. This analysis has the advantage that it allows us to account for case-manipulative NI without making reference to the representational level, because we can attribute the whole phenomenon to the interaction between the interpersonal and the structural levels. In other words, the speaker uses certain structural means to convey decisions at the interpersonal level, while the semantics are left unaffected.

6.2.1. The interpersonal

level

Let's turn to (1 lb), again. At the interpersonal level, we find a speaker with the communicative intention Μ to 'inform A about C', where C conveys that 'Ri did R 2 such that R3 is of primary importance'. What we have, then, is one ascriptive act T 1 ; accompanied by three referential acts. One of these bears focus or another 'foregrounding' function, and will be instantiated at the representational level as the locative satellite. The other two are the agent and patient of the SoA. Thus, the interpersonal level looks as follows: (INFORM M,: [(DECL A,: [(P,) s (P2)A (C,: [(T,) ( d R , : S ( R , ) ) ( R 2 ) (R 3 )FOC] ( C O ) ]

(M,))

Figure 12. Interpersonal Analysis of (1 lb) As we saw in previous examples, R] is immediately instantiated at the interpersonal level since it has no semantic, but only interpersonal content; it represents the speaker. Note that the IncN (the patient) is present with a referential act R 2 of its own, unlike what happens in instances of unitary NI; this accounts for the fact that the IncN is referential, and not a mere qualifier. Because of this, also, (1 lb) must be seen as an instance of R NI, in which (R 3 ) represents the focused locative constituent.

114

Niels Smit

6.2.2. The representational level As was pointed out before, case-manipulative NI does not have relevance at the level of semantics, given the fact that it is a mere structural choice to convey a decision made by the speaker at the interpersonal level. Thus, the organisation of (1 lb) at the representational level mirrors that of any nonincorporated ditransitive expression. For completeness' sake, the representation of (1 lb) is given in the figure below: [c'ak] (fi)) R1 (lxj: 1 (xi))Ag (-slx 2 : [(f2: [ce'] (f2))] (x2))Go]: R3(x3: [(f3: [kool] (f3))]: (Xl)poss (x3))LOC ( e l))

(PRES ei: [T(LNCOMP IMPF fi: R2

Figure 13. Representational Analysis of (1 lb) In the above figure, ce' is the second argument of the predicate, rather than a qualifying σ0 satellite as was the case in unitary NI. Note that ce' in this case is non-specific, although this is not necessarily the case (witness (12)). The locative satellite is attached at the level of (ei). Case-manipulative NI shows considerable similarity with a choice that speakers can make in English to express a Beneficiary by means of a prepositional phrase or as an indirect object. As the following shows, the two English sentences John gives Mary the book and John gives the book to Mary share a single semantic representation:29 (PASTei: [(IMPF fi: give] (fi): [(lxi: Mary ( χ , Χ Μ (fi)) (lx,: John (Χ,))Α 8 (lx 2 : [(f3: [book] (f3))] (x2))Go] (e,)) Figure 14. An Example from English The difference between the two actual expressions only arises under the influence of choices by the speaker as to which part of the information conveyed in the expression is of prime importance to the Addressee; when the book is in focus, the speaker will probably use indirect object to express Mary, while focus on the Beneficiary will probably lead to expression of the latter by means of a PP.

Noun incorporation in Functional Discourse Grammar 115

6.2.3. Structural level The interpersonal decision to focus on a peripheral participant that normally would be expressed by means of a prepositional phrase leads to incorporation of the patient at the structural level. Notice that the IncN is not necessarily stripped of most of its categoriality features, as is the case in unitary NI. Since we are dealing with incorporation of a referential participant, the presence of a term is presupposed. Therefore, we may also expect the IncN to exhibit properties of termhood, such as modifiability through operators and restrictors. These may be incorporated alongside the head noun, or strand verb-externally: what is, and what is not susceptible to incorporation, ultimately depends on the structural level of a language. Rosen (1989:297ff) points out that languages allow different types of NPs to follow an IncN. That is, whereas terms consisting of a single noun incorporate rather easily, some languages (like Southern Tiwa) also allow the incorporation of a first restrictor of an otherwise more elaborate term, leaving the 'remainder' of that term stranded as an external modifier (cf. (4)). Still other languages, like Mohawk, allow not only for stranding, but also for external reference to an IncN, as in: (13)

Mohawk (Baker 1988: 288) ThetAre' wa '-ke-nakt-a-hninu-'. yesterday

I-k-her-e'

FACT-lsS-bed-0-buy-PUNC

0-lsS-think-IMPF

Uwari Λ-ye-nühwe '-ne'. Mary FUT-FsS-like-PUNC 'Yesterday I bought a bed. I think Mary will like it.' This article will leave these structural phenomena out of further consideration; it is important to stress, however, that the semantico-pragmatic nature of case-manipulative NI does allow for a wide variety of structural realisations of the incorporation, whereas stranding in the case of unitary NI is excluded a priori, given that unitary NI is f NI and ( f ) is not modifiable. The representation of (1 lb) at the structural level looks as follows: note that the locative function at the representational level leads to the realisation of the prefix in- at the structural level.

116

Niels Smit

(Cl,: [(PrP,: [[±comp]-[S]-[verb]-[Obj]-[TMA]] (PrP,))] [(RPi: in-[kool] (RP,))] (Cl,)) (Cl,: [(PrP,: [[-COMP]-[S]-[C'ak]-[CE']-[IMP(INTR)]] (PrP,))] (Cl,)) \

kinc'akce 'tik inkool Figure 15. Structural Analysis of (lib)

6.3. Discourse-manipulative NI Distinguishing between case-manipulative NI and discourse-manipulative NI, as Mithun (1984) does, is disputable. That is, there is no real structural or semantic difference between the two, since both are instances of incorporation of a participant, the only (claimed) difference being the pragmatic motivation on the part of the speaker. As is shown above, casemanipulative NI involves focusing on a participant that is not by default in a focus position. Consider the following example, a fragment of a Gunwinggu folk tale: (14)

Gunwinggu (Oates 1964:99ff) Benewam benebebmey gure manguq They-went they-came-up where honey .bag galug Then yende limb

dulge? dulgendi (tree) was marked

gargayn bidbom magnadji, chicken.hawk he-climbed-up he-cut-limb-off mangay dja it-fell-down and

dabu-bagmey honey.comb-broke

dolgarj dabu-mey galug he-got-up he-honey.comb-took then

galug then

wiriwiriya:g Wiriwiriyag

dabu-yuney he-honey.comb-ate

'... They went to the tree which was marked where the honey bag was. Then the chicken hawk climbed up and cut off the limb. The limb fell down and the honeycomb broke. Wiriwiriyag got up and took the honeycomb and ate it.'

Noun incorporation in Functional Discourse Grammar 117 In the example above, we come across the participant mangurj, 'honey bag'. On its first introduction, this honeybag is new information in the ongoing discourse, and hence is expressed by means of a non-incorporated noun in Obj function. In the subsequent discourse, the honeybag is no longer a pragmatically salient piece of information; consequently, it is incorporated in the verb (note that Gunwinggu has many different noun stems for free nouns and IncNs). The only structural difference with respect to case-manipulative NI would be, then, that the 'vacated object position' is not necessarily used to foreground another participant; it can also remain unfilled, 30 as is the case in the Gunwinggu example above. The analysis of (14) at the interpersonal level looks as follows: Dabubagmey ' honeycomb-broke' (INFORM M,: [(DECL A,: [(P,)s (P2)A (C,: [(T.) (dROoivTop] (CO)] (M,)) Figure 16.

Interpersonal Analysis of (14)

As we see, discourse-manipulative NI is also a type of r N I in terms of FDG; the honeycomb has the status of a separate referential act (R]). It differs from case-manipulative NI in that no 'foregrounding' function is necessarily assigned, but a backgrounding function instead. At the level of semantics, incorporation again plays no role, as we saw in casemanipulative NI. For completeness' sake, the representational level is given in the figure below: (PASTe,: [(iMPFf,: [bagme] (f,)) R1 (x,: [(f 2 : [dabu] (f 2 ))] ( x O W ] (e,)) Figure 17.

Representational Analysis of (14)

At the structural level the first argument eventually is incorporated under the influence of the selection of the pragmatic function GivTop at the interpersonal level:

118

Niels Smit

(Cl,: [(PrP,: [[S]-[verb]-[TMA]] (PrP,))] (Cl,)) (Cl,: [(PrP i: [[3SGS]-[dabu]-[bagme]-[PAST]] (PrP,))] (CIO) Figure 18. Structural Analysis of (14)

6.4. Classifier NI The fourth type of incorporation that Mithun distinguishes has received much attention in the formalist literature on NI, since it appears to support Baker's analysis of the phenomenon in terms of head movement. Consider the following example: (15) Tuscarora (Mithun-Williams 1976:60) nehrataskwähkw- hwa? ha? tsi:ir. DUAL- 3SGM- domestic.animalpick.up SERIAL EMPH dog 'He animal-picks dogs he is a dog-catcher.' The above example shows something that is unattested in the other types of NI, namely coreferentiality between the IncN and an external object NP. Mithun mentions that this type of construction typically occurs at the start of a new part of discourse; the external NP is used to narrow down the meaning of the general IncN, after which it suffices to use the IncN alone to refer to the previously identified discourse participant. Although she makes no explicit claim about it, Mithun seems to assume that the external NP in classifier NI has argument status, while the IncN functions as a mere classifier. In fact, her analysis of classifier NI could be restated in FDG terms as unitary NI with a two-place predication frame, which would make (15) look as follows: (PRES SERIAL ei: [(IMPF f,: [DUAL [ähkw]taskw] (fi)) ( l x , : 3 M ( x 0 ) A |

(-smx2: [(f3: [tsi:ir] (f3))] (x2))Go] (e,)) Figure 19. Representational Analysis of (15) according to Mithun (1984)

In the above, -taskw- is attached to the predicate as a σ 0 satellite at the representational level, while the specific NP ts:ir 'dog' functions as the second argument.

Noun incorporation in Functional Discourse Grammar 119

This at first sight plausible analysis is untenable, however. Most importantly, Mithun claims that the IncN lacks the capacity to refer. Although this cannot be proven convincingly in those cases where an external NP is present, it is downright unlikely in the light of subsequent clauses in the same discourse, where the IncN, referring to the same discourse participant, very clearly is referential. Furthermore, if the above analysis of classifier NI were correct, Mithun could not account anymore for the evolution of NI, which she claims develops from unitary NI, via case-manipulative and discourse-manipulative NI to classifier NI (Mithun 1984:872ff). Since the above analysis shows much more similarity with unitary NI than with the other types, it is difficult to show how classifier NI could ever have developed from manipulative NI. Third, a 'semantic' analysis of classifier NI in terms of predicate modification is not in keeping with Mithun's claim that the process, just like manipulative NI, is used for pragmatic reasons; if that is indeed the case, we must attribute its existence to the interpersonal rather than the representational level of the grammar. An alternative analysis is offered by Baker (1996). He interprets the external NP as a modifier of the IncN which has argument status. This analysis is a lot more attractive; it can account for the pragmatic motivation of classifier NI, it allows us to see the developmental parallel between classifier NI and manipulative NI, and it accounts for the referentiality of the IncN. Translation of Baker's analysis of cases like (15) to FDG would yield the following at the representational level:

(PRES SERIAL e ^ [(IMPF f,: [ D U A L [ ä h k w ] ] ( f i ) ) R2

R1

( l x i : 3 M (X]))Ag

( - s m x 2 : (f2: [ t a s k w ] ( f 2 ) ) (x2): (f3: [tsi:ir] ( f 3 ) ) ( x 2 ) ) G o (e,))31

Figure 20. Representational Analysis of (15) according to Baker (1988)

As can be seen, the first restrictor (head) of the patient is now the general Ν taskw 'domestic animal', which is modified by the specific Ν tsi:ir 'dog'. At the structural level, the head of the argument will be incorporated under the influence of some pragmatic operator or a specific structural template, while the remainder of the argument term will 'strand' as an external NP. Note that the IncN is now the head of a term which corresponds with a referential act (R2) at the interpersonal level; in other words, classifier NI is a third type of RNI.

120

Niels Smit

Another advantage of the head incorporation analysis of classifier NI is that the lexical relationship between the IncN and the external NP can be controlled with much more ease, since the external NP falls under the scope of the selection restrictions of the IncN. That is, as the first restrictor of a term, taskw can be understood to impose selection restrictions on lower restrictors, such that they meet the criterion of hyponymy. If the IncN and the external NP did not belong to the same constituent at the representational level, as is the case in Mithun's analysis, such lexical control would be hard to conceive of.

7. A revised typology of NI The preceding section has shown how FDG can account for the four kinds of non-obligatory NI that Mithun distinguishes in her 1984 typology (unitary NI, case-manipulative NI, discourse-manipulative NI and classifier NI). Unitary NI can be analysed as f NI, brought about by an interaction between the representational level and the structural level of the grammar; case-manipulative, discourse-manipulative and classifier NI are all forms of KNI, stemming from an interaction between the interpersonal level and the structural level. The tables below summarise the path that all types of NI travel through the levels of analysis in FDG: Interpers. Level Repr. Level Struct. Level

τ. (f,: [verb]:[(f 2 : [IncN] (f 2 ))lao(f,)) (PrP,: [[verb]-[f 2 ]] (PrP,))

Ri X

1

RP l(Subj)

Figure 21. An FDG Account of Unitary NI Interpers. Level Repr. Level Struct, level

T, (f,: [verb] (f,)) (PrP,: [[verb][R211 (PrP,))

R, x, RP l(Subj)

Figure 22. An FDG Account of Case-manipulative NI

R2 X2

R3(Fore grounded) X3 RP2(Obj)

Noun incorporation in Functional Discourse Grammar 121

Interpers. Level Repr. Level Struct, level

T, Ri (f,: [verb! (f,)) x, (PrP,: [[verb]- R P 1 ( S u b j ) [R211 (PrP,))

R2(Backgrounded)

-

Figure 23. An FDG Account of Discourse-manipulative NI

Interpers. Level Repr. Level Struct, level

T, R, (f,: [verb] (f,)) x, (PrP,: [[verb]- RP1(Subj) [R2]] (PrP,))

R2(Backgrounded)

X2 R2(Stranded modifier; Obj)

Figure 24. An FDG Account of Classifier NI

As was already suggested in section four, an alternative typology of NI follows from the above analyses. As the figure below shows, FDG allows for a three-way subdivision, distinguishing between non-referential, nonmodifiable FNI, non-referential, modifiable "NI, and referential, modifiable R NI. Possible modifiers, numerals and deictic/pronominal elements typically strand verb-externally, although they may be incorporated along with their head N. This does not really affect the typology, though. Type

Level

Modifiable

Referential

(obligatory NI) f NI X NI R NI

Structural Repr. Repr. Interpers.

Lang.-dependent No Yes Yes

Lang.-dependent No No Yes

Figure 25. An FDG Typology of NI

The above relies heavily on the hierarchical ordering of the levels inside linguistic competence. At the lowest level, we find instances of obligatory, non-deliberate NI. These are motivated neither semantically nor pragmatically, but are the simple result of the interaction of a representational string and a language-specific set of templates. An example would be Southern Tiwa, in which all (heads of) inanimate objects obligatorily incorporate. At the next level, we find instances of unitary NI, which are motivated semantically. At the semantic level, a noun is attached as a zero-level restrictor to

122

Niels Smit

a predicate; it incorporates at the structural level. The next types involve the incorporation of (heads of) discourse participants; while XNI cannot be motivated pragmatically by lack of a corresponding referential act at the interpersonal level, the three type of RNI are motivated pragmatically, either by foregrounding peripheral participants (case-manipulative NI) or by backgrounding given topics (discourse-manipulative NI). The representational level is not affected by RNI; there is merely interaction between the interpersonal and the structural levels. Classifier NI can be interpreted as a subtype of discourse-manipulative NI, anticipating the argument's backgrounding in subsequent discourse. Stranding, finally, may occur at all levels, with the exception of fNI, for the obvious reason that a zero-level satellite cannot be modified. Recall that σ0 is attached to the predicate without a variable of its own (cf. figure (7)); hence, it cannot exhibit an internal structure of its own. Now, let's see how this alternative typology can be brought into accordance with Mithun's claims regarding the existence of an implicational hierarchy of noun incorporation. The figure below repeats the hierarchy mentioned in section three: · ry Classiner XTT ΝI

Discourse, . ~> manipulative NI

.

case. .. XTT manipulative NI

>

unitary NI

Figure 26. Mithun's (1984) hierarchy of NI, based on a 100 language sample

At first sight, it seems difficult to reconcile the FDG typology with Mithun's hierarchy. Her upper three types receive the same FDG analysis, the only difference being the pragmatic consideration that underlies the process (foregrounding vs. backgrounding). How could these three types of R NI ever form a hierarchy? Moreover, Mithun notes that all languages displaying NI at least have unitary NI. But, as I argued earlier, unitary NI pertains exclusively to languages with flexible lexical categories, while many of the languages concerned (esp. Amerindian languages) are rigid, not allowing at all for the direct expression of modification. Those difficulties can be overcome, however, if we take into account that at least two classes of examples that Mithun mentions may be given an alternative analysis in terms of X NI. 32 First, instances are attested of what seems to be unitary NI in non-flexible languages. Consider the aforementioned case of Yucatec Mayan, which is repeated here as (16):

Noun incorporation in Functional Discourse Grammar 123 (16)

a. kinINCOMP- ISΊ chop a tree.'

c'akchop-

0- ik ce'. 3S- IMPF(TR) tree

b. kinINCOMP- ISΊ wood-chop.'

c'akchop-

ce'- 0. tree- IMPF(INTR)

In section five, (16) has been treated as a case of f NI. However, an alternative analysis in terms of XNI is also possible, witness the following figure where the IncN is analysed as the head of a non-referential term: k-in- c'ak- ce'- 0, Ί tree-chop' interpersonal level f

NI analysis X

NI analysis

(INFORM M,: [(DECLA,: [(P,) s (P2)A (C,: [(T,)top/foc (dR,: S (R.))top/foc] (C,))] (M,)) (PRES e,: [(INCOMP IMPF fi: [[c'ak]:[ce'] v ] (fi)) (lx,: 1 ( Xi ))AJ (e,)) (PRES βρ [(INCOMP IMPF fi: [c'ak] (fi)) (lxi: 1 (xi))Ag "R(X2: [ce'] (x 2 ))] (e,))

Figure 2 7. An Analysis of (16) in Terms of fNI and XNI Obviously, there should be some semantic difference between the f NI and the XNI analysis, given the difference at the representational level. The motivation for this kind of incorporation is clear, however: it stems from a clash of the interpersonal level (no expression of the participant 'tree') with the representational level (presence of the participant 'tree'), which is then solved by incorporating it in the verb. Since the head of a term is incorporated, we also expect the IncN to show Xj-characteristics, such as ω-operators and -restrictors. This seems to be the case indeed: although examples of pluralization and modification of non-referential participants are scarce in the literature, and seem absent for Yucatec, at least one is offered by Dixon (1988), who mentions a Fijian VN-complex that can be translated as 'eat-rolled-taro.leaves'. In this example, the IncN is both pluralized and modified, a clear indication that f NI cannot be the case.

124

Niels Smit

'eat-rolled.taro.leaves' (f,: [eat] (f,)) -R(mx,: [(f2: [taro.leafs] (f2))] (x,): [(f3: [rolled] (f3))] (χ,)) αο Figure 28.

X

NI in Boumaa Fijian (Austronesian: Dixon 1988)

Another class of incorporation phenomena that may be best explained in terms of XNI concerns those cases that display possessor raising. Mithun analyses them as instances of case-manipulative NI, which would make them r NI in FDG terms; however, there is a fundamental difference between the two, in that the IncN seems to be non-referential. Furthermore, a peripheral participant is assigned Obj function in case-manipulative NI, while in the case of possessor raising, a term modifier (second or lower restrictor) is assigned Obj function. I would argue that an analysis as XNI does justice to both considerations. A distinction must be made, however, between possessor raising of alienably and inalienably possessed nouns. A possible and plausible analysis with regard to the former is to deny its existence altogether. In most cases, after all, the 'raised possessor' can equally well be interpreted as a beneficiary or another semantic function. Consider (12) again; although the incorporated pronoun hi is often interpreted as a possessor, I see little objection to treating it as a (negative) beneficiary. Hence, the interpretation of the sentence would be 'He stole the car from me\ rather than 'he stole my car'. As a result, cases of 'possessor raising' can be analysed in the same way as any other instance of case-manipulative NI. Inalienably possessed nouns cannot be treated in this fashion, probably because inalienability causes the possessor to be more severely affected by the action denoted in the predicate. Consider the following: (17) Mohawk (Rosen 1989:301) kwiskwis y-a'-t-ho -Tnyukwal-ihsta-?. pig

TRS- AOR-DPL-3 Μ. 3M-snout-grab-PUNC

'He snout-grabbed the pig

he grabbed the pig's snout.'

I suggest to analyse possessor raising of inalienably possessed nouns as an example of XNI. That is, 'snout' in the above sentence is claimed to be nonreferential, while it is modified by a possessor which is referential. The IncN can be seen as the head of a non-referential term, so that the representational analysis of (17) would look as follows:

Noun incorporation in Functional Discourse Grammar

125

Ί snout-grabbed the pig' (e,: [(f,: [grab] (f,)) (x,)Ag "R(x2: [(f2: [snout] (f2))] (x2): [R(X3: [(f3: [pig] (f3))] ( x 3 ) ) P o s s ] ( x 2 ) ) g o ] (e,)) Figure 29.

X

NI in Mohawk

The stranded possessor then assumes the grammatical function Obj, and can be referred to separately in subsequent discourse. This example also rather clearly explains a semantic difference between unitary NI (fNI) and X NI. Whereas the verb and the IncN form one, inseparable whole in the case of the former, evidenced by the close bond at the representational level, the IncN in the latter example stands in a verb-argument relation to the verb. With the reanalysis of these two classes of NI-phenomena (unitary NI in non-flexible languages and possessor raising NI of inalienably possessed nouns) as instances of XNI, the gap that seemed to exist in the FDG typology of NI is now filled; moreover, it turns out that some members of Mithun's class of case-manipulative NI can be analysed as such, as well as some cases of unitary NI. What we have, then, is an in-between class that nicely fits the hierarchy, a preliminary version of which can now be formulated in FDG terms. Mithun's observation that all incorporating languages exhibit unitary NI can be restated as 'all incorporating languages exhibit XNI, while those with flexible word classes may also display f NI'. This reformulation also accounts nicely for the genetic/geographical spread in incorporation phenomena; as several authors have noted, there is sharp distinction between Oceanic languages and Amerindian languages with regard to the kinds of incorporation they allow for. The two higher types in her hierarchy (casemanipulative and discourse-manipulative NI) can be taken together as a single category of RNI, the presence of which presupposes the presence of X NI. I do not see any real objections to collapsing these categories, given their considerable similarities. Mithun (1984: 874) notes, moreover, that the extension of case-manipulative to discourse-manipulative NI is not so much another process as it is an alternative usage of that process which is reserved for polysynthetic languages.

126

Niels Smit

The above points to an implicational hierarchy, though not in its classical sense. It seems that enough support is found to assume the following hierarchy: R

NI

>

X

NI

>

f

NI

Figure 30. An Alternative Hierarchy of NI Only the presence of the lowest type of NI is not guaranteed. Rather, languages seem to select a part of the hierarchy as their domain. Amerindian languages and some Australian languages choose the 'upper' part, which includes incorporation of referential acts and terms. Oceanic languages, on the other hand, seem to choose the lower part, exhibiting incorporation of terms and incorporation of predicate modifiers.

8.

Conclusion

This article has shown how FDG can account for deliberate noun incorporation, focusing on its semantic and pragmatic motivation. It turned out in section five that the framework of FDG can accommodate all existing varieties of the phenomenon by attributing them to interactions between different modules of the linguistic competence, which involve the incorporation at the structural level of constituents of different types. Pragmatically inspired types of NI stem from the interpersonal level, and involve the incorporation of nouns heading referential acts. R NI exhibits all properties of R-constituents; consequently, the IncN is referential and (optionally) modifiable. Semantically inspired NI stems from the representational level, and divides into two subtypes. XNI involves the incorporation of the head noun of a non-referential term; hence, it is non-referential and modifiable. On the other hand, f NI involves the modification of the verbal predicate by a non-verbal restrictor which is neither modifiable nor referential. The latter type only occurs in languages with flexible lexical categories, since genuine nouns cannot be used as modifiers. The FDG-based typology also supports the implicational hierarchy of incorporation processes that Mithun advocates, although a thorough reclassification of phenomena is involved. If the FDG analysis is adopted, the hierarchy divides into three types of deliberate NI (to which obligatory NI

Noun incorporation in Functional Discourse Grammar 127

can probably be added as a fourth category): languages then either take the left side or the right side of the spectrum. As a consequence, it is not fNI that is the lowest category, but it is XNI that must be present in all incorporating languages as a central category, while the flexible languages can extend their inventory of incorporating constructions to fNI, and rigid languages to r NI. The research presented here is only a first impulse towards the treatment of complex words in the framework of Functional Discourse Grammar. Many issues have been left unattended and need to be researched in the future. First and foremost, it is crucial that new data be collected in order to reach more solid conclusions. Most data available in the literature has been stripped of its surrounding discourse, which makes it difficult to assess its pragmatic motivation and (lack of) referentiality. Moreover, my proposal assumes that referentiality and modifiability do not necessarily co-occur. Specific data in support of such a claim is needed as well, also from other areas than NI. With the help of data collected through guided elicitation (questionnaires) and the collection of larger pieces of (ideally spoken) discourse, our perspective on matters of incorporation can be widened considerably.

Notes 1. 2.

3.

I am greatly indebted to Casper de Groot and Kees Hengeveld (University of Amsterdam) for their invaluable and extensive comments on earlier draft versions of this article. For details on notational conventions in FDG, cf. Dik (1997). On all levels, numerical labels are attached to variables to facilitate their identification, while entities are labelled with their function (either interpersonal, representational or structural). Cf. Evans & Sasse (2002). Sadock's (1980: 302ff) criteria for wordhood are: 1. Words are continuous stretches of morphemes that, as wholes, can occur in isolation 2. Words are subject to phonological 'merging' through sandhi processes 3. Phonological rules that apply within words are lexically controlled 4. The order of elements within words is semantically fixed 5. Explicit conjunction never occurs word-internally

128

Niels Smit

6. Words cannot be interrupted with pauses or parenthetical material 7. Error correction cannot take place word-internally 4. The glosses in the examples are adapted to EUROTYP standards as much as possible. For deviant abbreviations, cf. the author(s) concerned. 5. Sadock (1980; 1986) argues in favour of a treatment of West-Greenlandic verb formation in terms of NI. His views are challenged by Mithun (1984), Sapir (1911) and Johns (1988), among others. 6. In this respect, Hopper and Thompson (1984:711) point out that, in the case of NI "... often the Ν is truncated in some way, and it invariably loses the ability to take determiners and inflections. Incorporated N's are very low in categoriality." Their claim that categoriality reflects discourse prominence implies that considerations from the domain of discourse affect syntax, something which is found in FDG as well. 7. This particular finding Mithun reports on may well be influenced by a wrong interpretation of the data. That is, non-specialised verbs are far more frequent than specialised ones in languages anyway. 8. For a considerable number of languages, descriptive grammarians have pointed out that this semantic opacity really must have developed diachronically (for example Matisoff (1981), for the Sino-Tibetan language Lahu). 9. The term (discourse) participant is used here in the widest sense possible, to refer not only to arguments of the verb, but also to other semantic slots in the discourse that usually are less central, such as location, source, destination, path, etc. Although she does not explicitly say so, the claims in Mithun (1984) offer an explanation for the fact that these are generally less susceptible to incorporation, given that, in most cases, they do not constitute independent elements in the information structure of the discourse, but rather serve as specification of the 'actions' depicted. Also, the fact that these constituents cannot bear syntactic roles is indicative of their subordinate nature in the pragmatic information structure of the clause. 10. Baker accounts for doubling by stating that the external NP is added through adjunction to the trace ti; left behind by the incorporated noun; this structure could be paraphrased either as "I animal r catch xi; such that X; equals [the ferocious dog]NP", or as "I animalj-catch [the ferocious Xi]Np, such that X( equals [dog]N". 11. The name is chosen on account of semantic restrictions that govern the selection of an overt external object NP; that is, the latter must be in a hyponymic relation with the IncN. 12. As a solution to this problem, Thomas Rosen (1989:316) suggests that only two types of NI out of a possible four have yet been explored: in addition to

Noun incorporation in Functional Discourse Grammar 129

13.

14.

15. 16. 17.

18.

19.

20. 21.

22.

compounding NI without, and classifier NI with null pro-forms, her framework allows for the converse types as well. This is not in accordance with findings from Baker and others, who note that doubling occurs with non-generic IncNs as well. Baker's analysis, which assumes the IncN to be the head while the external NP is added as a modifier/specifier, accounts for the facts in a more satisfactory way. Depending on the structural level of the language involved, it is sometimes also possible to incorporate a noun with its modifier, witness examples from Boumaa Fijian (Dixon 1988) and Gunwinggu (Oates 1964). Cf. Dik(1997b:14). The claimed lack of referentiality is also supported by the incapacity of IncN to be questioned separately (Lee 1975:276). 'Other pragmatic functions' comprises not only all sub-types of focus and topic, but also functions which pertain to the general importance and relevance of information, irrespective to the knowledge that S assumes A to have. Such functions can be foreground, background, etc. As a consequence, I would claim that derivational analyses of unitary NI in terms of verb-internal argument saturation (Thomas Rosen 1989; Anderson 2000) are fundamentally erroneous: if unitary NI were argument incorporation at the structural level, the incorporated argument would retain its interpersonal relevance and would correspond to a referential act, which enables the assignment of pragmatic functions. As is in the example, however, that is not the case Although theorists have different opinions about the status of the IncN in cases of unitary - or 'compounding' - NI (Thomas Rosen 1989; Anderson 2000), most descriptive grammarians concur with Mithun, and describe the semantic relationship between V and IncN in terms of head - modifier, rather than verb - argument. This and the complete lack of referentiality of IncNs in unitary NI led me to reject the analysis of IncNs as arguments. This clear semantic relationship between the constituents of the VN-complex may be obscured over time by lexicalisation. For detailed argumentation, cf. Vet (1986) and Hengeveld (1992b). Analysing predicate modifiers as restrictors enhances the structural symmetry between predicates and terms; moreover, it allows us to abolish the class of satellites in its entirety, whose structural anchoring in the representation has always been a troublesome matter. Cf. Dik (1989); Hengeveld (1992b); Keizer (1992); Cuvalay-Haak (1997) for discussion. The proposed structures for the representational level are the following (Vet's 1986 remarks on the status of satellites are assimilated herein):

130

23. 24.

25.

26.

27.

28.

29. 30.

31.

32.

Niels Smit τγ2 e,: [τ, [pred] (χ,) (χ„) σ,] σ2 Dik (1989) (χ 2 β!: [(π, f,: [pred] (f,): σ, (f,)) (χ,) (χη)] (e,): σ2 (e,)) Hengeveld (1992b) (x2 e,: [(π, f,: [x0 [pred]: σ„] (fi): σ, (fi)) (*i) 0 0 ] : σ2 (e,)) Keizer (1992) (x2 e,: [(x, c,: [[(x0 f,: [pred] (f,): σ0 (f,))] (x,) (x„)] Cuvalay (1997) (ci): ffi (cO)] (eO: σ2 (e,)) Cf. Dik's Specific Principle IX on constituent ordering (Dik 1997a:414). Cuvalay (1997) only mentions zero-level operators, which she uses to account for regular changes in the inherent lexical aspect (Aktionsart) of verbs. I see no principled objection against the existence of zero-level restrictors, however, given that the IncNs in unitary NI function as restrictors in the true sense of the word, narrowing down the verb's denotation. Only few publications about the structural level in FDG are yet available. Hence, the formalisms concerned should be understood in spirit, not as attempts to elaborate theory. Furthermore, only those parts of the structural level that are relevant to the discussion at hand are worked out in detail. It seems that the inventory of structural primitives that a language has at its disposal is controlled to a large extent by the morphological type. Thus, isolating languages will show many templates that correspond to a single element at the structural level, whereas polysynthetic languages will mould many elements together in a single 'word' or word-like constituent. Mithun uses the notion argument. I prefer to use participant in cases like these, given that non-arguments (location, instrument, etc.) can be promoted to Obj as well. The term 'foregrounding' is used here in the widest sense possible, since a peripheral participant may be promoted syntactically for a variety of reasons, ranging from pragmatic considerations of topic-focus distribution to expressing empathy of the speaker. Beneficiaries are represented in FDG as level-1 satellites. Bear in mind that Obj in the FDG is not 'vacated', since movement or other transformations are not allowed in FDG. It would be more correct to say that the Obj function is simply not assigned. Note that the operator dualic is treated as a 7i0 operator here, following Cuvalay (1997). Mithun-Williams also points out that the dualic "...reflects a common semantic element in complex predicates with which they occur [...] the dualic must be considered a component of certain complex predicates at the time of lexical insertion." (Mithun-Williams 1976:128). A strong reservation must be made with regard to xNI. Although its existence can be given a firm theoretical basis in the FDG framework, there is little explicit evidence of heads of terms which are modifiable yet not referential.

Noun incorporation in Functional Discourse Grammar 131 Since separating the two properties is not generally done in most other frameworks, the available data from the literature is biased against such an analysis. It goes without saying that specific data will have to be elicited in order to substantiate the claims made here.

References Anderson, Stephen R. 2000 Lexicalism, incorporated (or incorporation, lexicalized). CLS 3 6 - 2 : 13-34. Baker, Mark C. 1988 Incorporation: A theory of grammatical function changing. Chicago: Chicago University Press. 1996 The polysynthesis parameter. Oxford: Oxford University Press. Charney, Jean O. 1989 A grammatical sketch of the Comanche language. Ph. D. diss., University of Colorado. Cook, Eung-do & Andrea Wilhelm 1998 Noun incorporation: New evidence from Athapaskan. Studies in Language 22 - 1: 49-81. Cuvalay, Martine 1995 The Ε-structure in Functional Grammar: Towards a consistent treatment of tense, mood, aspect and illocutionary force. Working Papers in Functional Grammar 59, University of Amsterdam. Cuvalay-Haak, Martine 1997 The verb in literary and colloquial Arabic. Functional Grammar Series 19. Berlin: Mouton de Gruyter. Dik, Simon C. 1989 The theory of Functional Grammar. Part 1. Dordrecht: Foris. 1997a The theory of Functional Grammar 1: The structure of the clause. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. 1997b The theory of Functional Grammar 2: Complex and derived constructions. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Dixon, Robert M. W. 1988 A grammar of Boumaa Fijian. Chicago: University of Chicago Press.

132

Niels Smit

Evans, Nicholas & Hans-Jürgen Sasse 2002 Introduction: Problems of polysynthesis. In Problems of polysynthesis, Nicholas Evans & Hans-Jürgen Sasse (eds.), 1-13. Berlin: Akademie-Verlag. Frantz, Donald G. 1971 Toward a generative grammar of Blackfoot. SIL publications 34. Norman: University of Oklahoma. Garcia Velasco, Daniel & Kees Hengeveld 2002 Do we need predicate frames? In New perspectives on semantic representation in Functional Grammar, Ricardo Mairal Uson & Maria Jesüs Perez Quintero (eds.), 95-122. Berlin: Mouton de Gruyter. Gerdts, Donna B. 1998 Incorporation. In The handbook of morphology, Andrew Spencer and Arnold M. Zwicky (eds.), 84-100. Oxford: Blackwell. Hengeveld, Kees 1992a Non-verbal predication. Theory, typology, diachrony. Functional Grammar Series 15. Berlin: Mouton de Gruyter. 1992b Parts of Speech. In Layered structure and reference in a functional perspective, Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), 29-55. Pragmatics & Beyond 23. Amsterdam: John Benjamins. 2003 Functional Discourse Grammar. Presentation held at the Amsterdam Functional Discourse Grammar colloquium, spring 2003. 2004a The architecture of a Functional Discourse Grammar. In A new architecture for Functional Grammar, Mackenzie, Lachlan J. & Maria Ä. Gomez-Gonzalez (eds.). Berlin: Mouton de Gruyter. 2004b Epilogue. In A new architecture for Functional Grammar, Mackenzie, Lachlan J. & Maria Ä. Gomez-Gonzalez (eds.). Berlin: Mouton de Gruyter. Hopper, Paul J. & Sandra A. Thompson 1984 The discourse basis for lexical categories in Universal Grammar. Language 60 - 4: 703-752. Keizer, M. Evelien 1992 Predicates as referring expressions. In Layered structure and reference in a functional perspective, Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), 1-27. Pragmatics & Beyond 23. Amsterdam: John Benjamins.

Noun incorporation in Functional Discourse Grammar 133 Kiparsky, Paul 1982 From cyclic phonology to lexical phonology. In The structure of phonological representations, Harry van der Hulst and Norval Smith (eds.), 131-176. Dordrecht: Foris. Kooij, Jan & Maarten Mous 2002 Incorporation: A comparison between Iraqw and Dutch. Linguistics 4 0 - 3 : 629-645. Kristoffersen, Lars 1992 Derivation and inflection in a functional grammar of West Greenlandic. In Layered structure and reference in a functional perspective, Michael Fortescue, Peter Harder and Lars Kristoffersen (eds.), 143-171. Pragmatics & Beyond 23. Amsterdam: John Benjamins. Kroon, Caroline 1995 Discourse particles in Latin: a study of nam enim, autem, vero and at. Amsterdam Studies in Classical Philology 4. Amsterdam: Gieben. Launey, Michel 2002 Compound nouns vs. incorporation in classic Nahuatl. In Problems of Polysynthesis, Nicholas Evans & Hans-Jürgen Sasse (eds.), 1 Π Ι 34. Berlin: Akademie-Verlag. Langacker, Ronald W. 1987 Foundations of Cognitive Grammar. Stanford, CA: Stanford University Press. Lee, Kee-dong 1975 Kusaiean reference grammar. PALI language texts: Micronesia. Honolulu: The University Press of Hawaii. Mackenzie, J. Lachlan 1987 The representation of noun predicates in the fund. Working Papers in Functional Grammar 25, University of Amsterdam. Mithun-Williams, Marianne 1976 A grammar of Tuscarora. Garland Studies in American Indian Linguistics. New York: Garland. 1984 The evolution of noun incorporation. Language 60 - 4: 847-894. 1986 On the nature of noun incorporation. Language 6 2 - 1: 32-37. Mithun, Marianne and Greville G. Corbett 1999 The effect of noun incorporation on argument structure. In Boundaries of Morphology and Syntax, Lunella Mereu (ed.), 49-71. Amsterdam: John Benjamins.

134

Niels Smit

Nespor, Marina 1993 Fonologia. Le strutture del linguaggio 1. Bologna: II Mulino. Nichols, Joanne 1986 Head-marking and dependent-marking grammar. Language 6 2 - 1 : 56-119. Oates, Lynette F. 1964 A tentative description of the Gunwinggu language of Western Arnhem Land. Oceania Linguistic Monographs 10. Sydney: University of Sydney. Rijkhoff, Jan 2002 The noun phrase. Oxford: Oxford University Press. Rosen, Sara T. 1989 Two types of noun incorporation: A lexical analysis. Language 65 2: 294-317. Sadock, Jerrold M. 1980 Noun incorporation in Greenlandic: A case of syntactic word formation. Language 5 6 - 2 : 300-319. 1985 Autolexical syntax: A proposal for the treatment of noun incorporation and similar phenomena. NLLT3 - 4: 379-439. 1986 Some notes on noun incorporation. Language 62 - 1: 19-31. Velazquez-Castillo, Maura 1995 Incorporation and object placement in Guarani. In Word order in discourse, Pamela Downing and Michael Noonan (eds.), 555-579. Amsterdam: John Benjamins. Vet, Co 1986 A pragmatic approach to tense in Functional Grammar. Working Papers in Functional Grammar 16, University of Amsterdam. Wasow, Thomas 1977 Transformation and the lexicon. In Formal Syntax, Peter Culicover, Thomas Wasow and Adrian Akmajian (eds.), 327-360. New York: Academic Press.

Morphosyntactic templates Casper de Groot

1.

Introduction

This paper is devoted to the status of templates at the structural level in the Functional Discourse Grammar model. I will argue that morphosyntactic templates form a necessary part of expression rules and that they are relevant to the structure of words, NPs and clauses. I will show that specifications in the fund, at the interpersonal level, as well as the representational level may trigger morphosyntactic templates at the structural level. Furthermore I will show that in production different configurations even at different levels may converge in the expression rules by triggering the same morphosyntactic template. For this purpose, data from Dutch will serve to illustrate the necessity of a morphosyntactic template to account for the structure of words. A morphosyntactic template relevant to the expression of terms will be based on data from Hungarian, whereas Oro Nao, a dialect of Wari will provide data which suggest the relevance of a morphosyntactic template in expressing clauses. For an outline of Functional Discourse Grammar, I refer to the contribution by Hengeveld (this volume).

2. Templates and the expression of words 2.1. Dutch participles Regular so-called past participles in Dutch are formed by the verbal stem together with the prefix ge- and the suffix -t/d.1 For instance:

136

Casper de Groot

(1)

druk- 'print' a. ge-druk-t PX-print-SX 'printed' b. Jan heeft een boek gedrukt. John has a book print.PAST.PART 'John has printed a book.'

Past participles, i.e. elements with the shape of past participles, may also be employed as attributes, as in: (2)

een (door Jan) ge-druk-t boek. a (by John) PX-print-SX book 'a book printed by John, a printed book'

Forms such as gedrukt in (2) which function as modifiers of the head noun of the term are analyzed as instances of verbal restrictors. Verbal restrictors come in different forms, for instance as finite (relative) clauses, present and past participles, etc. Past participles may also arise as derived intransitives {Er wordt gedrukt 'There is printing going on') and in passive constructions (Het boek werd gedrukt 'The book was printed'). The examples (1) and (2), however, will be sufficient for the discussion here. Expression rules which produce the two examples with gedrukt in Dutch, will contain at some stage in production the following kind of specifications: (3)

a. Jan heeft een boek gedrukt. b. Pres Perf druk-[V] (xi:Jan)Ag(ilXj: boek[N])Go c. Perf => heb-[V] gedrukt Pres, 3sgAg => heeft

(4)

a. een (door Jan) gedrukt boek. b. (ilxj: boek[N]: druk-[V] (Xi:Jan)Ag(Xj)Go) c. 0 TNS; open pred. in A2 => gedrukt

Note that the participle and the attribute take the same form. They are, however, derived on the basis of two different expression rules. In other

Morphosyntactic templates 137 words, (combinations of) different grammatical operators may yield the same expression. In the case of (3) it is the operator Perfect, whereas in (4) it is the absence of a tense operator in the embedded predication in combination with the fact that the embedded predication is an open telic predication in A2 (second argument). If two or more distinct expression rules produce the same articulated form and the rules contain both the introduction and placement of morphemes, expression rules will contain quite some redundancy. For that reason, De Groot (1990) argued for a distinction between trigger rules (primary and auxiliary) and placement rules. Dik (1997) further developed the notions of triggers, which he called 'morpho-syntactic operators' or μoperators. Instead of two distinct rules such as (5), it may be argued that the paths in which two expressions are made to converge at some point. This is illustrated in (6) below. (5)

a. Perf => Aux + ge-druk-t b. TNS = Zero, where the embedded predication is an open telic predication in x2 => ge-druk-t

(6)

Morphosyntactic encoding a. Perf =>Aux; TEMPLATE [μ1-ΡΓβά[ν]-μ2] b. TNS 0 , Pred. in A2 => TEMPLATE |>l-Pred[V]^2] c. Pred[V] =>druk Aux => hebStructural level a. heb- [ß\-druk[V]-ß2] b. [ß\-druk[V]-ii2] Phonological encoding a. μΐ =>ge b. μ2 =>d/t Phonological level a. heb-gedrukt b. gedrukt

138

Casper de Groot

The location of the rules given in (6) are conform to the model proposed by Hengeveld (this volume). The rules in (6) are not fully specified. The abbreviated rules illustrate the following steps in the production of words (see figure 1 on the next page), (i) Based on information at the representational level, a template, an auxiliary, and two secondary operators (μΐ and μ2) are taken from the middle box of the grammatical component. The auxiliary will take the form heb-. The lexical form druk-[V] will be inserted in the template; (ii) The results of the morphosyntactic encoding are the configurations at the structural level; (iii) In the phonological encoding component the (bound) morphemes ge- and -d/t are introduced; (iv) The results of these rules are the forms found at the phonological level. In the following section we will see that the template suggested in (6) is also relevant in the formation of new words. Moreover, the analysis of synthetic compounds in Dutch reinforces the idea that a distinction should be made between μ-operators and templates.

2.2. Dutch derived modifiers and synthetic compounds In Dutch new Modifiers 2 can productively be derived from Nouns as in: (7)

Ν a. bochel 'hunch' b. laars 'boot'

-> Modifier —» ge-bochel-d 'hunchbacked' —» ge-laars-d 'wearing boots'

Note that the derived modifiers take a form similar to the participles discussed in section 2.1 above, i.e. a lexical stem with a prefix ge- and a suffix -t/-d. Two major differences between the participle on the one hand and the derived modifier on the other are that the participle has a verbal stem and the form arises through a grammatical operation (inflection) and that the derived modifier has a nominal stem and the form arises through predicate formation (derivation). The modifiers may also be derived on the basis of nominal compounds, as in (8):

Morphosyntactic templates

Figure 1. General layout of FDG

139

140

Casper de Groot

(8)

a. Ν + Ν bont+mantel fur+coat 'fur coat'

Modifier ge-bontmantel-d PX-furcoat-SX 'wearing a fur coat'

b. Modifier + Ν dun+doek 'bunting'

Modifier ge-dundoek-t 'covered with bunting'

The following formations differ from the formation of modifiers from nominal compounds. Note that the derived modifiers also contain a noun and a modifier. However, the relation between the two lexical elements is that of head [N] and modifier [Mod]. Also note the order of the morphemes, in particular the position of the modifier and the prefix ge-. Consider: (9)

a. Ν : Modifier broek: kort trousers: short

Modifier kort-ge-broek-t short-PX-trousers-SX 'wearing shorts'

b. schouder: breed shoulder: broad

breed-ge-schouder-d broad-PX-shoulder-SX ' broad-shouldered'

Note that in all three examples (7) through (9) the base for the derivation is nominal, whereas the output takes a form very similar to that of the participle, which has a verbal base. Let us now consider some examples in which a verbal element is combined with a modifier or noun. There are several types, two of which I will discuss further. 3 The first type is the derived verb. Note the form of the participle: (10)

a. Mod + N kort+wiek short+wing

V kortwiek-en shortwing-INF 'to clip wings'

/ / /

participle ge-kortwiek-t PX-shortwing-SX 'wing-clipped'

Morphosyntactic templates

b. V + V —»V / brand-schilder brandschilder-en / burn-paint burnpaint-INF 'stain' /

141

participle ge-brandschilder-d PX-burnpaint-SX 'stained'

In the second type a modifier is derived from a combination of verb and modifier, where the modifier modifies the verbal stem. Consider: (11) a. V : Modifier breien : net knit: proper

b. doeken : dun cloth : thin

-»

Modifier net-ge-brei-d proper-PX-knit-SX 'well knitted'

—>

dun-ge-doek-t thin-PX-cloth-SX 'thinly clothed / covered with cloth'

Interestingly, de-nominal modifiers such as for instance gemanteld (lit. coated, 'wearing a coat') may also form the input for a predicate formation rule similar to (11), in that the output is articulated in the same way. Note that the stem in gemanteld is the nominal mantel, that there is nothing verbal in the semantics, and that the form looks like a participle. The form may be modified as if it were verbal. Consider: (12) a. Ν mantel 'coat' b.

-> —>

Modifier ge-mantel-d PX-coat-SX

bont-gemanteld colourfully-coated

Note that (8b) and ( l i b ) as well as (8a) and (12b) form minimal pairs, in that the position of dun 'thin' and bont 'fur' or 'colourful', contributes to a significant difference in meaning: (13) a. ge-dun-doek-t dun-ge-doek-t b. ge-bont-mantel-d bont-ge-mantel-d

'covered with bunting' 'thinly-covered' 'wearing a fur coat' 'colourfully coated'

142

Casper de Groot

The second forms in (13) pattern along with the so-called separable complex verbs in Dutch, where a verb combines with a particle, as in: (14)

particle-V a. op-bellen up-call 'to call/phone'

participle op-ge-bel-d up-PX-call-SX 'called/phoned'

b. in-smeren in-smear 'rub'

in-ge-smeer-d in-PX-smear-SX 'rubbed'

Table 1. A morphological typology of modifiers/participles in Dutch example

infl

1 2 3

gedrukt gedrukt gebocheld

X X

4 5 6 7 8 9 10

gebontmanteld gedundoekt kortgebroekt

type

der

px-X-sx

X

X X X

X-px-X-sx

SIMPLEX

V > participle V > attribute Ν > Mod COMPLEX

N+N > Mod A+N > Mod N:Mod > Mod N?Mod > Mod Mod+V > V

v+v>v V:Mod > Mod Particle-V

11

goedgemutst gekortwiekt gebrandschilderd netgebreid opgebeld

X X X

X X X X X X X

X X X X X X X X

Morphosyntactic templates

143

2.3. Discussion The data of the former two sections may on the basis of their morphology be summarized as in table 1, where the plus (+) indicates composition, the colon (:) indicates modification, and the question marker (?) indicates that the relation between the two lexical categories is unclear. In section 2.1. above, we saw that types 1 and 2 arise through different types of expression rules which make use of grammatical operators. Type 3 takes the same structural form as types 1 and 2. This type, however, arises through predicate formation. Predicate formation takes place in the upper box of the grammatical component (see figure 1). I assume that the output of the formation rule will contain a primary operator. Let's call this operat o r PROPERTY.

(15) input: Pred[N]

OUTPUT:

Pred[Mod] + PROPERTY meaning: 'have/wear Pred[N]'

The primary operator PROPERTY will trigger the introduction of a template at the structural level. We may stipulate the following kind of expression rules: (16) a. Perf b. 0 TNS, pred in A2

TEMPLATE

C. "PROPERTY"

TEMPLATE

d. Pred[V] Pred[N] e. μΐ μ2

druk bochel ge-d/-t

TEMPLATE

|>l-Pred[V]^2] |>l-Pred[V]^2] [μ1-ΡΓεά[Ν]-μ2]

Types 4 and 5 are in fact the same as type 3, the only difference being the fact that types 4 and 5 consist of nominal compounds. The final forms are derived by rule (15) and the expression rules (16). Types 8 and 9 are verbal compounds. That is why they are marked X for derivation. They behave like ordinary verbs (marked X for inflection) and are sensitive to rules (16) Type 6 is different. Let us have a look at the form kortgebroekt (shortpx-trousers-sx) "wearing short trousers". The form gebroekt 'with trousers' is a correct word in Dutch derived on the basis of rule (15). The form kortgebroekt cannot be analyzed as a compound of kort + gebroekt, because if

144

Casper de Groot

it were a compound it would rather have the meaning "wearing trousers for a short time". The modifier kort in kortgebroekt can only be associated with the nominal broek 'trousers', as a modifier as in (17) Jan draagt een korte broek. John wears a short pair.of.trousers 'John is wearing shorts.' When we have a look at type 10, we see that there is also a relation between a head and a modifier, hi this case the head is a verbal stem brei 'knit' and has net 'proper' as its modifier. The semantic relations between head and modifier in types 6 and 10 could be captured in the follwing representations: (18)

a. (fi: Pred[N] (f,): Pred[Modifier] (f,)) b. (fi: Pred[V] (fi): Pred[Modifier] (fi))

When we have a look at type 7, we see that there is no such relation between the noun and the modifier as in type 6. The modifier goed does not modify the nominal element muts. The form gemutst (as the output of rule (15)) means 'wearing a cap'. Goedgemutst does not mean 'wearing a good cap', but rather 'wearing a cap well', meaning 'good humoured'. In other words, the semantics of the output is not completely predictable. Type 11 is added to the list to illustrate that so-called separable complex verbs in Dutch also take a sequence of morphemes similar to types 6, 7 and 10. In contrast to these types which are derivations, type 11 is inflectional. Thus, like types 1 through 3, which involve both inflection and derivation and using the same morphosyntactic template (16), it seems that another morphosyntactic template is relevant to the formation of the so-called synthetic compounds (types 6, 7 and 10) and the inflection of separable complex verbs (type 11), namely: (19)

TEMPLATE

[stem-μΐ -stem[HEAD]^2]

A predicate formation rule which accounts for the morphosyntactic properties of types 6, 7 and 10 may take the following form, where again PROPERTY is a primary operator:

Morphosyntactic templates

(20)

Input Predl[Mod] Pred2[N]/[V]

output: Predl Pred2 [Mod] +

145

"PROPERTY"

Different expression rules will introduce the template as given in (19), for instance: (21)

a.

Perf

b. "PROPERTY"

TEMPLATE

[stem^l-stem[HEAD]^2]

- > TEMPLATE [ s t e m - / x l - s t e m [ H E A D ] ^ 2 ]

Simplex predicates will take template 1 as given in (16), whereas complex predicates will take template 2, as given in (19). As for the semantic relation between the two lexical items involved, the predicate formation rule should also specify which relations may hold between the head and the modifier. There are three possibilities, namely (22)

a. netgebreid b. kortgebroekt c. goedgemutst

(V : Mod) (N : Mod) (N ? Mod)

The addition of the representations given in (18) as frames (upper box of the grammatical opponent, see figure 1) neatly account for the semantics of the examples in (22a) and (22b). Consider: (23)

Input Predl [Mod] -> output: Predl Pred2 [Mod] + PROPERTY Pred2[N/V] FRAMES: i. (f,: Pred[N] (f,): Pred[Modifier] (f,)) ii. (f,: Pred[V] (f,): Pred[Modifier] (f,))

Example (22c) is different. It contains lexical elements which fit frame (i), i.e. there is a Noun muts 'cap' and a Modifier goed 'good'. The semantics, however, fit frame (ii), i.e. goed modifies as if the head were a verb, which is not the case. This misfit which speakers of Dutch allow may be explained on the basis of the consideration that words which take the form of a participle (but are not participles and are not verbal), are interpreted as verbal elements. I cannot oversee what consequences this view might have for the FDG model. Future research may reveal these consequences. For now, there are already interesting theoretical implications, namely that frames in the lexicon may take the form of those given in (23), and that predicates may be associated with different frames at the same time. 4

146

Casper de Groot

A last word on the order of morphemes. In table 1 we see that forms in which two lexical elements are involved may take two different expression formats: (24)

a. px-stem+stem-sx b. stem-px-stem-sx

In those cases where the two lexical elements are together surrounded by a prefix and suffix (as in (24a)), the two elements form a compound. In the other case where one of the lexical elements precedes the prefixal element, that lexical element behaves as a modifier of a head. In that respect the two Dutch complex constructions show similarities with different cases of incorporation as discussed by Smit in this volume.

3. Templates and the expression of terms 3.1. Non-adjectival/verbal modifiers in Hungarian In Hungarian, adjectives 5 can be used to modify the head of a term. The adjectives take in these cases the non-inflected form, as for instance in: (25) a. az okos läny the clever girl 'the clever girls' b. egy nepszerü irö a popular writer 'a popular writer' Verbal restrictors receive the expression of finite relative clauses or nonfinite participles.6 If other elements then adjectives and verbs are employed to modify the head these elements will be marked by the suffix -i, or the participle of copula van 'be'. First consider the following examples with a nominal modifier in (26a), an adverb in (26b), and a term in (26c): (26) a. a tavasz-i fesztiväl the spring-SX festival 'the spring festival'

Morphosyntactic templates 147 b. a fent-i tenyek the above-SX facts 'the above-mentioned facts' c. α härom ora-i elöadäs the three hour-SX performance 'the three o'clock performance' Interestingly, the marker -i also applies if the modifier - or an argument of the head as in (27a) - consists of a postpositional phrase. Consider: (27) a. α Ιιάζ Anna ältal-i felipitis-e the house Anna by-SX construction-3.SG 'the construction of the house by Anna' b. az asztal alatt-i kalap the table under-SX hat 'the hat under the table' Postpositional modifiers may also be expressed as a kind of non-finite construction with the present participle form of the auxiliary van 'be' (cf. Van Lier, this volume). Consider (28), which is an alternative expression of (27b): (28) az asztal alatt levo kalap the table under being hat 'the hat under the table' The postpositional modifier expressed after the head is ungrammatical. In present-day Hungarian, however, that type of expression is as a new alternative gaining more and more ground.7 If the application of suffix -i were unconditioned, the following expression format would hold: (29)

[restrictor] Ν —» [restrictor]-/ Ν where restrictor τ^Α or V

148

Casper de Groot

There are, however, a number of restrictions on the application of the suffix -i. The restrictions are the following (cf., E.Kiss et al. 1998, Kenesei et al. 1998). Restriction 1. The suffix attaches only to a number of semantic classes of nouns: (30)

a. place names: b. names of institutions or locations: c. d. e. f.

nouns expressing professions: personal relations: time: abstract notions:

anglia-i 'of England' egyetem-i 'of (a/the) university' orvos-i 'medical' anya-i 'motherly' tegnap-i 'yesterday's' gazdasäg-i 'economic'

The following classes do not allow the suffix: (31)

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

mass noun: product: result of mental work: collection: animal, plant: 'situation'nouns:

*arany-i 'of gold' *auto-i 'of a car' *festminy-i 'of a painting' *csoport-i 'of a group' *tigris-i 'of a tiger' *beke-i 'of peace'

Restriction 2. The suffix combines with place adverbs and directional adverbs 'from', but not 'towards': (32)

a. place adverb: b. direction'from'adverbs: c. direction 'toward' adverbs:

mögött-i 'behind-SX' mögül-i 'from behind-SX' *möge-i 'toward behind-SX'

Restriction 3. The suffix attaches to postpositions but not to case endings: (33)

a. a pole mögött-i könyv the shelf behind-SX book 'the book behind the shelf b. *a polc-on-i könyv the shelf-LOC-SX book 'the book on the shelf

Morphosyntactic templates 149 The standard expression for terms where the modifier is marked by a case is the construction with the participle form of van 'be'. 8 Compare: (34) a. a pole mögött levö könyv the shelf behind being book 'the book behind the shelf b. a polc-on levö könyv the shelf-LOC being book 'the book on the shelf Restriction 4. Modifiers with the suffix -i may be stacked onto each other, whereas just one modifier may be expressed by the participle of van 'be' (Cf. Chisarik 2005, Laczko 1995a,b). Consider example (35) for the stacking of modifiers with -i. (35)

a. α väros-on kivül-i, erdö mellett-i nyaralo the city-LOC outside-SX forest near-SX summer.house 'the outside-the-city near-the-forest summer house' b. az erdö mellett-i, α väros-on kivül-i nyaralo the forest near-SX the city-LOC outside-SX summer.house 'the near-the-forest outside-the-city summer house'

The following example illustrates the possibility of combining the two different expressing devices, -i and a participle of van 'be': (36)

a. a to-hoz közel-i erdö mellöl valo gyogynövenyek the lake-DIR close-SX forest near being herbs 'the herbs close to the lake near the forest' b. az erdö mellöl valo to-hoz közel-i gyogynövenyek the forest near being lake-DIR close-SX herbs 'the herbs near the forest close to the lake'

The following example is ungrammatical, because the participle form occurs more than once in one term:

150

Casper de Groot

(37)

*a templom mellett levö väroshäzä-val szemben levö iskola the church near being town.hall-INSTR opposite being school 'the school being near the church and opposite the town hall'

Restriction 5. The last restriction is a phonological restriction. Nouns, assuming that they belong to one of the classes mentioned under (30) above, ending in -i do not allow the suffix -i. For instance: (38) a. ferfi ruha man cloth 'gentleman's clothes' b. *ferfi-i man-sx

ruha cloth

3.2. Discussion Given the description of the Hungarian data, we may conclude that a number of restrictors of the head of a term in Hungarian mould into one pattern, namely (39) Representational level a. (xi: Pred[N] (x,): PredfAdv^/so^e] (χ,)) b. (χ,: Pred[N] (χ,): PredfNcl] (χ,)) c. (χ,: Pred[N] (x,): (x2: Pred[N] (x2)) (x,)) d. (x,: Pred[N] (x,): (x2: Pred[N] (x2))Sf (x,)) (40)

Structural level TEMPLATE [[X]-i N] where X is the modifier of N, specified as Pred[AdvLoc/Source], Pred[Ncl], (x2: Pred[N] (x2)), or (x2: Pred[N] (x2))Sf in the representational level.

The specification of Loc/Source selects the appropriate adverbs. The subclasses (cl) of the Noun should be specified in order to avoid incorrect expressions later. Given the template in (40) it thus seems that unlike other markers in Hungarian, the suffix -i is a phrase marker.

Morphosyntactic templates 151 Given the restrictions mentioned in section 3.1 above, the following information should also be accessible: (41)

a. adverb specifies location or direction from (not direction towards) b. nominal subclass c. if X is N, it may not end in -i. d. semantic function is expressed by a postposition

Information concerning (41a) and (41b) will be available at the representational level (see 39a-b)). Information concerning (41c) and (41d) will arise through the expression rules. (42) Restrictions on the application of expression rules a. Structural level Sf —> free morpheme i.e. the expression of the semantic function will be realised as a free morpheme. b. Phonological level The expression rule that marks the modifier phrase with -i does not apply if the modifier consists of a Noun ending in -i. There remains one other matter, namely the conditioned expression of modifiers with a participle form of van 'be'. Only one, and no more than one modifying phrase marked by a semantic function may be expressed by the participle valo or levö. For the sake of convenience I repeat the two expression devices for non-finite restrictors (i) levo/valo and (ii) the suffix -i, together with the finite expression (information we need below). Compare: (43) a. az asztal alatt levö kalap the table under being hat 'the hat on / under the table' b. az asztal alatt-i kalap the table under-SX hat 'the hat under the table'

152

Casper de Groot c. a kalap amely az asztal alatt van the hat REL the table under COP.3.SG 'the hat which is under the table'

Restrictors with a function expressed by a case do not allow type (43b). In order to derive correct term phrases in Hungarian the expression rules must first check whether the semantic function of a restrictor will be expressed by a bound morpheme (case ending) or a free morpheme (postposition). 9 If the semantic function will be expressed by a case, the participle levo/valo will be introduced. If the semantic function will be expressed by a postposition there are two possibilities, either the participle levo/valo, or the suffix -i. Where there are two restrictors, each with a semantic function, one of the restrictors must be expressed by a finite relative clause (which will have further implications for the assignment of operators within the restrictor, such as the presence of a tense operator). The distribution of markers will be as in table 2, where restrictor 1 and 2 merely indicate the number of restrictors, not necessarily the order of restrictors as first or second restrictor. The marking is not sensitive to the order of restrictors. 10 Table 2. Distribution of markers of restrictors with a semantic function in terms Restrictor 1

Expression

Case Case Postposition Postposition

levö/valö finite relative clause -i or levö/valö -i or levö/valö

Case

levö/valö

postposition

Restrictor 2

Expression

Case

levö/valö

postposition

-i (or levö/valö if restrictor 1 takes —i) -i

More technically, in the component of morphosyntactic encoding the choice will be made whether a semantic function will be expressed by a postposition or a case. If the expression will be a case (bound morpheme) the expression rule will introduce a secondary //-operator and the auxiliary or copula levö/valö. Consider:

Morphosyntactic templates

(44)

a.

hat (x,: kalap[N] (xj):

table (x2: asztal[N]

153

[the hat on the table] (xi))

(x 2 ))supEss

i

b. c.

Aux + asztal + μ3 Aux -» levo

These rules will result in the following configuration at the Structural level: (45) asztal^3 levo The secondary μ-operator will in the phonological component be expressed as -on, the case marker for the 'superessive': (46)

a. μ3 -> -on b. asztal-on levo

If the expression will be a postposition, we will get the following picture, where (47b) and (47c) are alternatives. Consider: (47) a.

hat (xi: kalap[N] (χ,):

table (x2: asztal[N]

[the hat under the table] (x 2 )) S ubEss(xO)

I

b. c.

asztal + alatt + μ4 asztal + alatt + levo

These rules will result in the following configuration at the Structural level: (48) a.

[Χ]-μ4 [asztal alatt]-μ4 b. asztal alatt levo TEMPLATE

The secondary μ-operator will in the phonological component be expressed as -i (49)

a. μ4 -i b. asztal alatt-i

154

Casper de Groot

In order to avoid ungrammatical expressions with levo/valo twice, the expression rules which end up at the structural level must generate the correct structural forms. If the rules contain the condition that Aux levo/valo may be assigned just once within a term restrictor, no more than one configuration of the type in (44) will be generated. Other restrictors, if present, will be expressed by other rules, such as (47b) and (48a), or rules which will lead to the expression of finite relative clauses. In this section on data from Hungarian, it has been shown that four different specifications of term restrictors at the representational level fit into a morphosyntactic template at the structural level. Furthermore it has been shown that the rather complex distribution of different expression devices involved could also be handled at the structural level. I conclude this section by claiming that, like in Dutch, a relation may be stipulated between the representational level (lexical elements, term structure, semantic functions) and the expression format on the structural level.

4.

Templates in the expression of clauses

4.1. Question word questions in Oro Nao For a template relevant to the order of constituents at the clause level, we will have a look at question word questions in Oro Nao, a dialect of Wari' spoken along the Pacaas Novos river in Rondönia (Western Brazil). The data and the interpretations are all from Everett & Kern (1997). A quite striking fact about Oro Nao is that the language does not have question words. Question-word questions are formed by means of a special use of demonstratives. 11 Oro Nao has a very rich person-marking system on verbs, nouns and prepositions. Overt expression of nouns which are coreferential with the person-marking elements is rare. Most utterances consist of just the verb and a lot of verbal morphology / clitic elements etc. 12 That is why Everett & Kern claim that Oro Nao does not have a basic word order, because one does not find many utterances with the overt expression of two, three or even more NPs. However, if noun phrases are overtly expressed, we find the following pattern: (50)

VOSX

Morphosyntactic templates

155

Constituents questioned are not in their normal position, and instead there is a string containing a spatial deictic element in sentence initial position, thus: (51)

(.. that..) (V) (O) (S) (X)

Just one element may be questioned. For the presentation of the data, I will use the notions of subject and object in the traditional sense. When the predicate is questioned, we schematically get (52a), the object (52b), the subject (52c), and when a satellite is questioned, the type of expression is (52d). Consider: (52)

a. b. c. d.

(.. that..) (.. that..) (.. that..) (.. that..)

Ο V V V

SX SX ΟX ΟS

[Note, V is missing, therefore V is questioned] [ Note, Ο is missing, therefore Ο is questioned] [ Note, S is missing, therefore S is questioned] [ Note, X is missing, therefore X is questioned]

The following examples illustrate these patterns.13 (53) Predicate is being questioned a. Verbal predicate cain' cain' caca mon tarama'? that.N.REM that.N.REM 3.P.M COLL man 'What did the men (do)?' b. Non-verbal predicate cain' cain' ne wixi-con tarama' that.N.REM that.N.REM 3.Ν name-3.S.Mman cara ne? that.recently.absent 'What was that recently absent man's name?' In cases where the predicate is questioned, the spatial deictic element occurs twice. This could be seen as a specific expression device to indicate that the predicate is being questioned and not one of the term positions.

156

Casper de Groot

(54) Object is being questioned ma' carawaca pa' caca mon tarama? that:PROX animal INFL.N.RP/P kill 3.P.M COLL man 'What (thing/animal) did the men kill?' Everett & Kern gloss the element carawa in (54) as animal, or thing/animal. They do not consider this element to be lexical in this kind of example. (55)

Subject is being questioned ma' co pa' nana hwam that.PROX EMFL.M/F.RP/P kill 3.P.RP/Pfish xec ca' ne? day this REC.PAST 'Who killed fish this morning?'

pain PREP.3.N

The following two examples illustrate the questioning of satellites, one with the function of Beneficiary in (56a) and Instrument in (56b): (56)

Satellite is being questioned a. Beneficiary ma' co pa' mi' caca that.PROX INFL.M/F.RP/P kill give 3P.M tarama'? man 'For whom did the men kill fish?'

hwam fish

mon COLL

b. Instrument cain' ca mam pa' cocon hwam that.N.REM INFL.N.RP/P INSTR kill 3.P.M./3.S.M fish mon tarama'? COLL man 'How (with what) did the men kill fish?' Parts of NPs may also be questioned. In that case, the head of the NP is placed together with the deictic element in sentence-initial position. Modifiers of the head, if present, are placed in the pattern position of the NP ('remain in situ'). Consider (57), where the head of the term panxi-ne occurs together with the spatial deictic element as a Q-word question device

Morphosyntactic

templates

157

in PI, whereas cwa' the demonstrative associated with the head is in the pattern position of the term: (57) ma' panxi-ne co mi' pin na cwa'? that.PROX child-3.N INFL give completely 3.S.RP/P this.M/F 'Whose son died?' (give completely = to die) Note that the speaker of Oro Nao when using a 'question-word question' often has to anticipate the number and gender of the entity given in the answer. The estimation of the speaker is expressed in the choice of the verbal inflectional clitics.

4.2. Discussion The data for question-word questions in the Wari' dialect of Oro Nao suggest that a specification at the interpersonal level will trigger a morphosyntactic template at the structural level. If the interrogative mood has been chosen at the interpersonal level, or more precisely the information-seeking type (an open question or question-word question), the structural level will provide the following unique template: (58)

TEMPLATE [ A u x ( V ) ( O ) ( S ) ( X ) ]

Aux will be spelled out at the structural level as one of the spatial demonstratives together with one or more secondary operators. These operators, which relate to gender and agreement, will be spelled out as bound morphemes at the phonological level. Interestingly, some of the other positions must be lexically filled in to make clear which position has not been taken (or element is missing). The syntactic configuration will then be the expression of the question-word question.

5.

Conclusions

In this paper I have argued that morphosyntactic templates form a necessary part of expression rules and that they are relevant to the structure of words, NPs and clauses.

158

Casper de Groot

On the basis of data from Dutch I have argued that specifications in the fund may take the form of frames which are relevant to the semantic interpretation of derived complex words. Furthermore primary operators, which also apply in the fund, may at the structural level introduce morphosyntactic templates. Similarities in the structure of inflected and derived words in Dutch, i.e. same forms, are taken to illustrate that morphosyntactic templates may accommodate lexical and grammatical morphemes coming from different sources. Data from Hungarian reinforces the claim that different configurations may lead to one type of expression. It was argued that four different specifications of term restrictors at the representational level in Hungarian fit into one morphosyntactic template at the structural level. The structural level also seems to be the appropriate level to account for the rather complex distribution of the different devices involved in the expression of restrictors in Hungarian. Finally, on the basis of data from Oro Nao it was shown that specifications at the interpersonal level may introduce a morphosyntactic template at the structural level. The template suggested concerns the order of constituents on the level of the clause.

Notes 1.

The choice of using - d or - t in Dutch spelling is determined by whether the final element of the stem is voiced or voiceless respectively. In speech the distinction between final - d and - t is neutralised as [t]. The combination of the prefix ge- together with the suffix -d/-t could be analysed as one expression, therefore as a circumfix. I will not treat the elements as a circumfix, because under certain morpho-phonological conditions the prefix or suffix may not be expressed, as for instance in: (i) gebruiken'use' -gebruik-t *ge-gebruik-t (ii) praten 'talk' - ge-praat *ge-praat-(e)t Were it to be analysed as a circumfix, part of the circumfix would have to be deleted in the expression rules to arrive at the correct form. Deletion is a very disfavoured operation in Functional Discourse Grammar. An alternative rule which avoids deletion could be formulated on the basis of having suppletive forms, i.e. three alternative circumfixes: ge.. .d/t, ge.. .0; and 0 . . .d/t.

Morphosyntactic templates 2.

3. 4.

5. 6.

7.

159

The Parts of Speech system of Dutch consists of three lexical categories: V, N, Modifier. Unlike English and Hungarian for example, Dutch does not make a distinction between the lexical categories Adjective and Adverb. One single lexical category may be employed to modify the Verb in an adverbial sense and the Noun in an adjectival sense. Compare: (i) een mooi lied. a beautiful song 'a beautiful song.' (ii) Zij zingt mooi. She sings beautiful 'She sings beautifully.' I will refer to elements such as mooi in these examples as the lexical category of Modifier (Mod) - with a capital Μ - in this paper. For the sake of simplicity, I will not include cases of Goal and Instrument incorporation here. De Groot & Meijs (1986) have attempted to account for the formation of Dutch synthetic compounds within the Functional Grammar framework. The paper shows that the FG framework, like other linguistic theories at that time, could not satisfactory account for the derivation. The view mentioned here that two frames may be associated with one predicate offers an interesting new way to look at synthetic compounds. Futhermore it could also be helpful in understanding morphological productivity and creativity in word formation. Unlike Dutch, Hungarian does make a distinction between the lexical categories Adjective and Adverb. Examples of finite and non-finite verbal restrictors are for instance: (i) a läny aki könyvet olvas the girl REL book.ACC read.3SG 'the girl who is reading a book' (ii) a könyvet olvas-ό lany the book.ACC read-PRES.PART girl 'the book reading girl' (iii) az olvas-ott könyv the read-PAST.PART book 'the book read' The occurrence of the modifier in postnominal position may be due to language contact. The modifier in postnominal position is marked neither by the suffix - i nor by a participle form of van 'be'. For that reason, I will not include the construction in the discussion of this paper. Consider:

160

Casper de Groot (i)

a kalap az asztal alatt the hat the table under 'the hat under the table' 8. Copula van 'be' has two (present) participle forms: valo and levö. See Laczko (1995a, 1995b) for the distribution of the different forms. 9. Case endings and postpositions are in complementary distribution. If each case and postposition corresponds to a distinctive semantic function, we must assume that more than 40 semantic functions are relevant to the grammar of Hungarian. 10. For a detailed analysis of the structure of the NP in Hungarian and an account for the distribution of the markers I refer to Laczko (1995a,b) and Chisarik (2005). 11. Oro Nao has two types of demonstratives: (i) spatial cwama' and cwain (relevant to Q-word formation) (ii) temporal paca, cara ne, and cara pane (not relevant to Q-word formation) The deictic elements can have agreement clitics: Spatial Proximate to Proximate to Distal speaker hearer Masculine co cwa' co ma' co cwain singular Feminine cam cwa' cam ma' cam cwain singular Neuter 'i ca' 'i ma' 'i cain Plural caram cwa' caram ma' caram cwain Temporal Masculine singular Feminine singular Neuter Plural

Seen/not heard co paca'

Recently absent

Long absent

co pacara ne

co pacara pane

cam paca

cam pacara ne

cam pacara pane

'i paca' caram paca'

'i cara ne caram pacara ne

'i cara pane caram pacara pane

12. According to Everett & Kern, the verbal inflectional clitics apply as follows. "Although case is not overtly marked on nouns, the verbal inflectional clitic is sensitive to the semantic role that the noun phrase bears in the clause, rigidly observing the following hierarchy for object agreement (subject agreement is always obligatory, regardless of the semantic role of the subject): Goal > circumstance > theme > benefactive > comitative > location > time

Morphosyntactic templates

161

13. Abbreviations used in the examples are: M/F/N = masculine/feminine/neuter 3.S/3.P.M = third person singular/third person plural masculine RP/P = realis past/present REX.PAST = recent past

References Chisarik, Erika Ζ. 2005 The structure of Hungarian noun phrases: A lexical-functional approach. Ph. D. diss., University of Manchester. E.Kiss, Katalin, Ferenc Kiefer & Peter Siptär 1998 Uj magyar nyelvtan [A grammar of modern Hungarian]. Budapest: Osiris. Everett, Daniel L. and Barbara Kern 1997 Wari'. London: Routledge. Groot, Casper de 1990 Morphology and the typology of expression rules. In Working with Functional Grammar: descriptive and computational applications, Mike Hannay and Elseline Vester (eds.), 187-201. Dordrecht: Foris. Groot, Casper de and Willem Meijs 1986 Synthetic compounding in Functional Grammar. Paper read at the 14th International Congress of Linguists, Berlin. Laczko, Tibor 1995a The Syntax of Hungarian noun phrases: A lexical functional approach. Frankfurt: Peter Lang. 1995b On the status of valo in adjectivized constituents in noun-phrases. In Levels and structures (approaches to Hungarian, Vol. 5.), I. Kenesei (ed.), 125-152. Szeged: JATE.

A crosslinguistic study of 'locative inversion': Evidence for the Functional Discourse Grammar model Francis Cornish

1.

Introduction

In this chapter,1 I will try to isolate the internal syntactic, semantic and discourse-pragmatic properties of the "locative inversion" construction in six languages, representing three constituent order types: English, French, Italian, European Portuguese,2 Turkish (rigid SOV) and Arabic (flexible VSO). The term 'inversion' is purely descriptive within this context and is not to be taken as indicating a "movement" derivation of such constructions;3 and in any case, we shall see later that the F(D)G approach in terms of the subsequent placement of underlyingly unordered constituents in positions within a template is perfectly suited to a proper treatment of this construction. In any event, locative inversion is a distinct construction in its own right, and its properties do not derive from those of the canonical, unmarked construction with which it can be correlated. My aim, then, is to show that the properties specific to the construction flow from its essential function within discourse, and from its informationstructural value. Thus a proper description of the construction presupposes a top-down organisation of the grammar, as in Hengeveld (2004a,b). It will be shown in what follows that the different types of function of each of the main components of the construction under study need to be determined essentially via specifications at the Interpersonal Level. This level precedes the Representational Level at which the subject function would be assigned, as well as the Structural Level where the various constituents would be placed within the positional template available at this level; but in standard FG (Dik, 1997: ch. 10), this level's equivalent follows the Representational Level's counterpart (the extended predication) at which syntactic functions are assigned. Hence the initial locative or temporal term,

164

Francis Cornish

the non-predicating verb and the subject-cum-Presentational Topic function of postposed subject terms could not be adequately specified within the standard model of FG (see section 4 for discussion). Section 2 outlines four general properties of the construction, in terms of English and French examples; section 3 examines those of the equivalent forms in Italian, Portuguese, Turkish and Arabic in the light of this initial characterization; and section 4 draws together the common features isolated from the specifics of each type of realization, and proposes an account in terms of the Functional Discourse Grammar model (Hengeveld, 2004a,b). Finally, section 5 reexamines the subject-verb agreement patterns evident in the data, particularly where the verb is unaccusative, 4 the subject term occurring to its right rather than to its left, as canonically in the SVO langages. Independently of this factor, a number of other properties characteristic of subjects are shown to be lacking in the construction. It is suggested that, at least where the verb is unaccusative, the construction may actually be an impersonal one.

2.

Some syntactic, semantic and discourse-pragmatic properties of the locative inversion construction in English and French

The discourse function of the locative inversion construction as a whole is to present a referent which is discourse-new 5 within the framework of a locative or temporal context. At the same time, the construction as a whole serves to establish the setting for a new discourse unit, indicating that the state, process or event it denotes is to be backgrounded as context for a more foreground event to come (see Huffman, 2002). The essential properties of the construction flow from this discourse function: the presence in clause-initial position of a locative or temporal adverbial which is either subcategorized by the verb of the construction, or is a level-1 or level-2 satellite, designating the place or time in which this new referent is to be located within the discourse model being co-constructed by the speech participants; a locative or existential non-predicating verb whose inherent semantics corresponds or is reduced to this type of denotation, and which is tightly connected syntactically and semantically to the preposed constituent; and a postposed rhematic subject term occurring in predicate focus position. Sections 2.1 - 2.4 below attempt to isolate these essential properties, on the basis of some French and English examples.

A crosslinguistic study of 'locative inversion' 165 2.1. Presence vs. absence of a preposed adverbial of certain types Before we examine an initial set of French data involving preposed adverbials, it will be useful to present a recent version of the FG languageindependent constituent order template, which standardly forms part of the Expression component, and enables particular constituent orders to be imposed upon the underlyingly unordered clause representations. See (1) below for this template, which I have augmented by a clause-final position PO, proposed for Polish by Siewierska (1998) and for Bulgarian by Stanchev (1997), but which is evidently more relevant cross-linguistically. (1)

P2, PI (S) (Ο) (Χ) V (S) (Ο) (Siewierska 1998: 246, slightly adapted)

(X)

(PO) (Χ),

P3

S, Ο, X and V symbolize 'subject', 'object', 'argument/satellite' and 'verb' respectively, and parentheses indicate optionality (not all languages require description in terms of the subject or object functions, for example). Standard FG (Dik, 1997: 70-71, 391-392) uses a set of placement rules within the Expression component of the grammar for assigning particular constituents bearing specified properties and functions to given positions in the ordering template appropriate for the language in question (that given in (1) is intended as a generalization from such language-specific templates). Subject and object expressions which are unmarked for pragmatic function will go to their respective pattern positions in the template relevant for the language at hand, while constituents marked for either the topic or the focus function will go to PI position. See (21) for a proposed set of placement rules relevant to the "locative inversion" construction in the six languages under study, in terms of a revised version of template (1) (template (14), section 4 below). In the case of an SVO language such as French, where, unlike in English, predicate ("completive") Focus cannot simply be marked via a nuclear pitch-accent on the non clause-final constituent concerned in situ, subject-verb inversion - as in (2a-5a) below - is needed to place the (Presentational) Focus constituent (the subject expression) in clause-final position, which is where the unmarked predicate-focus function is assigned in French. Thus, the subject expressions in the (a) examples below will be placed in PO position in template (1). Borillo (2000: 88), Ono (2001), Fournier (1997: 98-102), and other linguists argue that those verbs which subcategorize a PP complement (locative or temporal in the present case) invert with their subject when this

166

Francis Cornish

complement is preposed.6 Compare in this respect examples (2)-(5), which involve the French verbs se trouver 'to be found', s 'ouvrir 'to open',/tamper 'to blaze', and gagner 'to win', respectively: (2)

a. Dans I 'armoire se trouvaient les chaussures. 'In the wardrobe were (to be) found the shoes.' b. *Dans 1'armoire les chaussures se trouvaient. 'In the wardrobe the shoes were (to be) found.' (Nolke, 1995, ex. (36))

(3)

a. Avec cette cle s 'ouvre touteporte. 'With this key opens every door.' b. ?Avec cette cle, toute porte s 'ouvre. 'With this key, every door opens.'

(4)

a. Dans une large cheminie flambait un grand feu. 'In a broad chimney blazed a big fire.' b. Dans une large cheminie, un grand feu flambait. 'In a broad chimney, a big fire blazed.' (Borillo, 2000, ex. (10))

(5)

a. *D 'apres mon frere, gagnera ce cheval. 'According to my brother, will win this horse.' b. D'apres mon frere, ce cheval gagnera. 'According to my brother, this horse will win.' (ex. (14) from the Fournier & Fuchs (2002) corpus)

Since in cases like (2a), where a preposed PP is subcategorized by a verb,7 inversion is virtually obligatory (see (2b), in contrast to (2a)), one can postulate that the preposed locative PP is placed in the structural position PI in the standard FG universal constituent-order template as given in (1) - the initial position within the clause, reserved for expressions given special treatment, particularly for discourse-pragmatic reasons.8 Now, given that PI in (2b) is occupied by the locative PP as a quasitopic, the subject term as clausal topic cannot also be placed there. Level-1

A crosslinguistic study of 'locative inversion' 167 satellites such as avec cette cle 'with this key' in (3a,b), make subject-verb inversion a "preferred" option, in Fournier's (1997: 97) words. This would not be the case if the PP were to occupy the extra-clausal position P2, as in (4b). As the verb flamber 'to blaze' does not subcategorize the locative PP dans une large cheminee 'in a broad chimney', the inversion here is in no way obligatory (see example (4b), which is well-formed). The PP dans une large cheminee, a level-2 satellite in FG terms, is thus placed in P2 (at least in the case of (4b)). The choice of inversion or of non-inversion is thus equally well-formed with preposed level-2 satellites, as expected. Finally, propositional level-3 satellites, as in the evidential satellite D'apres mon frere 'according to my brother' in (5a,b), rule out subject-verb inversion completely.9 So we see here that, as the tightness of the grammaticalsemantic relation between verb and preposed adverbial decreases as we move from argument to level-1, then level-2 and finally level-3 satellite, the "attraction" it exerts on the verb lessens, and the pressure for subjectverb inversion within the construction as a consequence decreases. The presence in initial position of the locative adverbial is motivated by the fact that, to introduce a new referent into a discourse, that referent must be located somewhere, existence and location being two sides of the same coin, as is well known - see, for example, Lambrecht (1994: 179). However, while the nature of the preposed adverbial is one determinant of the inversion following it - and no doubt the most important such determinant - , it is not the only one, by any means, as Fournier (1997) also points out. Whether or not there is subject-verb inversion within the construction is a function of the interaction of several factors, in particular, the information-structural value of the utterance as a whole, the syntactic and semantic nature of the verb, and the internal structure of the subject term. Let us briefly examine each of these factors in turn.

2.2. Information-structure influences on the construction A central factor explaining the unnatural status of examples like (2b) is the positioning in clause-final position (unmarked Focus position) of an informationally "light" predicate: that is, not involving any specific semantic feature capable of being highlighted via this syntactic and prosodic treatment. This is also the reason for the difficulty, or even impossibility, of the non-inverted variants of (7a, b, g) and (7a', b', g') in section 2.3 below,

168

Francis Cornish

where the verbs are be able to be seen / se voir, stand / se dresser, and be / etre. Nelke (1995) argues that the assignment of the Focus function in French always involves the setting up of a paradigm of entities established as a function of the context - entities which are potential focus targets and the marking of a contrast between one member of this paradigm and all the others. He suggests that, where the focus is assigned to a single lexeme rather than to a whole phrase, the element on which the focus operates is the specific semantic feature which makes it possible to differentiate that element from all the other elements within the same lexical or semantic field. That feature is thereby highlighted, and the predicate has its full semantic value (see the (a) examples under (6) below); where inversion occurs, on the other hand, as in the (b) examples under (6), the verb's specific semantic feature is not highlighted, and only the generic semantic feature^), which establish(-es) the sortal category of entity to which the predicate may be applied, is/are in evidence. The verbs or past participles in the (b) examples below have a purely existential or locative, and not predicating, interpretation in this inversion context. In (6a-f), syllables in upper case indicate nuclear pitch accent. ((6c-f) are my examples; see Cornish, 2002: section 4 for fuller discussion of this issue). (6)

a. Dans I'armoire, les chaussures etaient ranGEES. 'In the wardrobe, the shoes were neatly arranged.' (Ex. (33a) in Nolke, 1995; originally presented in Borillo, 1990, ex. (a), p. 80) b. Dans I'armoire etaient rangees les chauSSURES. 'In the wardrobe were stored the shoes.' (Ex. (33b in Nolke, 1995) c. A I'horizon, un orage couVAIT. 'On the horizon, a storm was brewing.' d. A I'horizon couvait un oRAGE. 'On the horizon was brewing a storm.' e. ? #Dans la vallee, une riviere COULE.10 'In the valley, a river flows.'

A crosslinguistic study of 'locative inversion'

169

f. Dans la vallee coule une riVIERE. 'In the valley flows a river.'

2.3. The type of verb involved in locative inversions In general, locative inversion is only possible in English and French when the verb involved is intransitiveInvolved here are, in particular, intransitive verbs denoting the existence, localisation, appearance/emergence or disappearance of an entity, like stand in (7b), hang in (7c), run in (7e) and work in (7h); furthermore, "unaccusative" or "ergative" verbs,12 like sink in (7d)); passive verbs, especially those interpretable as stative-resultative, like be seen in (7a), be engraved in (7f); pronominal verbs: se voir 'be seen' in (7a') and se dresser 'stand' in (7b'); attributive or copular verbs, like be in (7g); and more generally, verbs which are interpretable in context as "light", from the point of view of their informational contribution.13 Birner (1994: 254), on the basis of the data she collected to carry out her study of subject-verb inversion in English, argues that the constraint involved here is pragmatic in nature, and reflects the need for the verb in the construction to "contribute no new (i.e discourse-new) information". Thus the verbs in her data represented only evoked or inferable information. We may add to the explanation the fact that, if the verb in question were to contribute new information in context, then it would necessarily be predicative, and the referent of the postposed subject term would then be potentially topical, a value which is excluded in this type of construction. The parallel sets of illustrative examples from English, followed by their French counterparts under (7), form a standard group of examples which are used as a baseline on which to view the locative inversion facts in the four other languages examined here: (7)

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

In a clearing could be seen a charming cottage. At the end of a courtyard stood some ageing huts. On the wall hung an antique chimney hook. In the distance was slowly sinking a patched-up oil tanker. Into the garden suddenly ran three young barefoot boys. On the door was engraved in gothic letters a mysterious inscription. g. Nestling in a pile of dead leaves was a tiny reddish bird. h. ?In this office work four people.

170

Francis Cornish

(7)

a'. Dans une clairiere se voyait une coquette chaumiere. b'.Au fond d'une cour se dressaient de vetustes baraquements. c'. Sur le mur pendait une antique cremaillere. d'. Dans le lointain coulait lentement un petrolier rafistole. e \ Dans le jar din accoururent soudain trois jeunes garqons aux pieds nus. f . Sur la porte etait gravee en lettres gothiques une inscription mysterieuse. g'. Tapi dans un tas de feuilles mortes etait un oiselet rougeätre. h'.Dans ce bureau travaillent quatrepersonnes.

So we can say, with Levin & Rappaport-Hovav (1995: 230) in the case of English, and Fournier (1997: 116) in that of French, that it is the presence of the inversion which gives these verbs an informationally "light" interpretation.14 These authors argue, moreover, that, in order for the inversion to be acceptable, the sense of the inverted verb should characterize the referent of the postposed subject term: in other words, there should be an effect of redundancy in the subject-verb relation, the verb expressing an activity or state which is characteristic of the type of entity represented by the subject referent. This property is obviously closely connected to the criterion of "informational lightness" which the verb needs to assume in context.

2.4. The syntactic form and semantic content of the postposed subject term, and its relation to the verb First of all, the postposed subject term does not have to be placed immediately to the right of the verb involved: (7d / d'), (7e / e') and (7f / f ) show clearly that an adverb ((7d / d') and (7e / e')) or a modifying PP (7f / f ) may be closely dependent on the verb. Thus it is the subject and the verb group or V', in the X' notation,15 which invert. The postposed subject in the eight examples under (7) (both the English and the French series) is morphologically indefinite, since a form of the indefinite article, or a non-definite numeral determiner, in (7e,h), is used in each case. This is not obligatory, though, the important factor being that the referent of this term should constitute discursively new and not topical information once it is brought into relation with its immediate context (see Birner, 1994: 252-253). This phrase may therefore be morphologically

A crosslinguistic study of 'locative inversion'

171

definite, but may certainly not be expressed by an unaccented definite 3rd person pronoun, which carries the presupposition that its potential referent is already highly active, psychologically, for the addressee (a GivTop, in FG terms).16 Birner (1994: 253) points out that the postposed subject may be formally definite, since it is not hearer-oldness which that subject's referent is prohibited from expressing, but discourse-oldness. A postposed subject may well refer to an entity with which the addressee is familiar (of which s/he has knowledge), but so long as that entity is construable as evoking discourse-new information in context, the postposed subject which expresses it will be felicitous. The longer and more complex the subject term is, the more likely it will be to be positioned to the right rather than to the left of its verb. This is predictable in terms of discourse, since if the information contribution associated with a nominal term is relatively more substantial, then it will tend to be interpreted as rhematic (introducing new information relative to the context) and not as thematic; and the default position for placing such constituents is generally towards the end rather than at the beginning or the middle of the clausal utterance. In addition, as Daniel Garcia Velasco pointed out to me, another motivation for subject-verb inversion in such cases is the placing of complex terms in clause-final position in order to facilitate processing (as in the case of extrapositions). Finally, as far as semantic content is concerned, the animacy of the subject term (human-denoting or inanimate) will cause a potentially movement verb to tend to be interpreted as dynamic and agentive when it is humandenoting, or as stative when it is inanimate. In the first case, the subject will tend not to be postposed, while in the second, it is likely to be.

2.5. Summary of the properties of the locative inversion construction, based on the English and French data To summarize the situation established so far on the basis of the English and French data we have seen in section 2.1-2.4, the presence in PI of an argument of the predicator, or of a level-1 (locative or temporal) satellite, tends to result in the subject being placed to the right of the verb, in PO position on template (1). The effect of this is that the verb in question (an intransitive verb in both languages) is non-predicating, and simply denotes the existence, localisation, appearance or disappearance on the scene of the referent of the subject expression. The subject expression thereby receives

172

Francis Cornish

presentational focus and is marked both by its non-canonical position in the clause, and by a pitch-accent. All these properties follow directly from the discourse functionality of the construction as a whole: to present a discourse-new entity in terms of a locative or temporal framework set up via the argument or level-1 satellite in PI. This constituent serves also as a link with the preceding discourse (cf. Birner & Ward, 1998).

3. Locative inversion in four other languages 3.1. Italian and European Portuguese 3.1.1. Italian Let us now compare the situation in relation to English and French with that which characterizes the Southern Romance languages, Italian and European Portuguese. In Italian, as in European Portuguese, lexical subject-full verb inversion seems to be completely free (unconstrained). It can involve all kinds of verbs - unaccusatives, unergatives and transitives, even without the copresence of a preposed overt locative or temporal PP (according to Pinto 1997). However, when transitive verbs are involved, only the order VOS is possible, and never VSO, according to Joäo Costa (p.c.). The only verb (here, a quasi-verb) which is excluded from this construction is the copula essere 'to be', which when it occurs induces the selection of the existential construction with ci 'there (is)' (see example (7g") below). The translation into Italian of the English examples of locative inversion given under (7ah) is presented under (7a"-h"). As in the case of the English and French examples, the verb of each example is in italics:17 (7)

a". b". c". d". e". f'.

In una radura si vedeva una casetta graziös a. Nel fondo di un cortile sorgevano delle baracche vecchie. Sul muro pendeva una catena di camino antica. In lontananza stava affondando lentamente una petroliera rappezzata. In giardino accorsero subito tre giovani ragazzi scalzi. Sulla porta era incisa in lettere gotiche un 'iscrizione misteriosa.

A crosslinguistic study of 'locative inversion' 173 g". h".

Rannicchiato in un mucchio difoglie morte c'era un uccellino rossiccio. In questo ufficio lavorano quattro persone.

Note that, with the exception of (7g"), where the copula essere is obligatorily accompanied by the locative clitic expletive ci (Carminati, 2001: 66), Italian expresses this construction in the same way as French. Maria Nella Carminati informs me that in many cases, preposing of the locative adverbial cannot co-occur with the absence of subject-verb inversion - that is, in the absence of inversion, the adverbial must be placed in its canonical position to the right of the verb. Among the examples (7a"-h") above, the last two, (7g") and (7h"), would not be possible without inversion - in the last-mentioned case, even if the preposed locative is not a governed complement of the verb lavorare 'to work': *In questo ufficio quattro persone lavorano 'In this office four people work'. She proposes the example of the verb nascere 'to be born' (an unaccusative verb like morire 'to die'), where we find the locative inversion construction: In questa casa e nato un poeta famoso 'in this house was born a famous poet', as well, as in English and French, as the variant without preposing and without inversion: Un poeta famoso e nato in questa casa, but not the variant with locative preposing and no inversion: *In questa casa un poeta famoso e nato. Nascere, morire, arrivare etc. are unaccusative, intransitive verbs, though possessing a locative argument at the level of their lexical-semantic structure; this argument may be analyzed as a "default argument" in Pustejovsky's (1995: 63-67) terminology. We may consider that such is also the status of the locative PP in questo ufficio in relation to the unergative verb lavorare 'to work' which we have just seen. The generalization which seems to apply here is that when the preposed constituent is either subcategorized syntactically, or is a "default argument" of the predicate to which the verb corresponds at the lexicalsemantic level, subject-verb inversion is either obligatory or strongly preferred in relation to the canonical order. In fact, it would be incorrect to say that inversion in Italian is completely free: as Pinto (1997) stresses, the inversion involving the majority of unaccusative verbs may have either a wide or a narrow focus, whereas that involving unergative or transitive verbs only allows narrow focus. For Carminati (and Pinto), the postposed subject is focal, rhematic, whereas the preverbal subject is topical or thematic.

174

Francis Cornish

Pinto (1997: 8) suggests that in general, verbs allowing inversion with wide focus interpretation in Italian are characterized by the presence in their argument structure of a locative or temporal argument denoting the spatio-temporal context in which the proposition expressed is asserted as true or false (cf. Erteschik-Shir's (1997) notion "stage topic"). Verbs not allowing inversion with wide focus interpretation18 would lack this extra argument (the unergative verb telefonare in (8e) below is exceptional in this respect). For Pinto, this phonetically null argument is basically deictic in character, with its interpretation being oriented towards the speaker. As an illustration, let us look at the pairs of examples (8a-b), (8c-d), and (8ef): note in particular the differences apparent in the English translations of the members of each pair: (8)

a. E' entrato Dante. has entered Dante 'Dante has come in (here/in this place).' b. Dante e entrato. 'Dante has entered (somewhere).' c. E' morto Fellini. has died Fellini 'Fellini has just died (.. .I've just heard it).' d. Fellini e morto. 'Fellini has died (at an unspecified time).' e. Ha telefonato Beatrice. has telephoned Beatrice 'Beatrice telephoned (here, at this place).' c. Beatrice ha telefonato. 'Beatrice telephoned (somewhere, has made telephone calls).' (Pinto, 1997, ex. (19)-(21), ch. 1)

When the subject occurs in postverbal position with 'invertible' verbs (i.e. verbs where wide focus interpretation correlates with inversion), the verb selects an extra locative or temporal argument. In fact, this argument is present even in the non-inverted variant, the two "objective / subjective"

A crosslinguistic study of 'locative inversion'

175

readings also being possible. Pinto concludes then that it is the implicit presence of this locative/temporal argument - which she symbolizes as LOC - which motivates the verb-subject order in the case of those verbs, which allow it. Subject-verb inversion in Italian would therefore only be marked where it can only give rise to a narrow-focus interpretation.

3.1.2. European Portuguese According to the description put forward by Costa (2001 a,b), in terms of subject-verb inversion, Portuguese would appear to behave like Italian rather than French, within the Romance group of languages - as in fact would be expected. This language allows inversion with all sorts of verbs, unccusative as well as unergative or transitive. According to Costa (2001a: 2), in "wide focus" sentences (thetic utterances19), the subjects of unaccusative verbs may be either pre- or postverbal. See (9a) vs. (9b) below: (9)

a. What happened? Ο Paulo chegou. the Paulo arrived b. Chegou ο Paulo. arrived the Paulo 'Paulo arrived.' (Costa, 2001a, ex. (2a, b))

On the other hand, with intransitive verbs other than unaccusatives, only subject-verb order, as in (9a), would be possible as a realisation of a thetic sentence - in response, then, to a question of the type What happened?. Costa (2001b) confirms the inference one can draw from this, on the basis of the situation characterizing Italian, that in Portuguese, verb-subject order for verbs other than unaccusative ones can only give rise to a narrow-focus interpretation. One factor connected with verb-subject order in Portuguese which we have not observed in the case of Italian concerns subject-verb agreement. According to Costa (2001a: 8-9), in familiar spoken discourse,20 agreement of an unaccusative verb with apostvzxbdX subject in number may not occur, whereas it is obligatory when the subject precedes the verb. This possibility is absolutely ruled out with the other categories of invertible Portuguese

176

Francis

Cornish

verbs (unergative and transitive). Interestingly, Maria Nella Carminati (p.c.) informs me that her native Bergamasco dialect in northern Italy also shows this possibility under exactly the same conditions. See the pairs (10a/a'), (10b/b') and (10c/c') below: a. Chegaram ο Pedro 21 arrive-PAS -3PL the Pedro a'. Chegou 0 Pedro arrive-PAS-3SG the Pedro 'Pedro and Paulo arrived.' b. Fecharam muitas close-PAS-3PL numerous b \Fechou muitas close-PAS-3SG numerous 'Many factories closed.'

e ο and the e ο and the

Paulo. Paulo Paulo. Paulo

fäbricas. factories fäbricas. factories

Chegaram as cadeiras. arrive-PAS-3PL the chairs Chegou as cadeiras. arrive-PAS-3SG the chairs 'The chairs arrived.' (Costa, 2001a, ex. (19a-c)) I will not follow Costa in his attempt to explain this phenomenon (in Minimalist terms, as a function of alternative Case assignment mechanisms), but will look at it rather in terms of information structure. I suggest the following hypothesis: Amongst the different types of verbs, only the unaccusatives in pre-subject position allow a "thetic" reading of their containing clause. Now, in familiar spoken Portuguese, postverbal subjects in such conditions may not trigger number agreement with their preceding verb. Why? Because in this case, neither the referent of the subject, nor the denotatum of the verb would be topical. And agreement in number (at least in the colloquial, spoken language) is sensitive to the informational status of the potential controller (in this case, the term fulfilling the subject function), and not solely to its purely grammatical status (that of subject, in opposition to object or to other grammatical functions). This combination of statuses - preposed focal verb and postposed focal subject - is what distinguishes preposed unaccusative verbs in relation to the other types of

A crosslinguistic study of 'locative inversion' 177 verbs in Portuguese: indeed, whereas transitive and unergative verbs may be preposed in relation to their subject, in this case, only the "narrowfocus" interpretation is possible - in other words, where only the referent of the subject will constitute contextually new information once the content expressed by the sentence is integrated with its discourse context. The denotatum of the preposed verb will correspond in this case to presupposed, anaphoric information. However, one must treat this generalization somewhat carefully, as Costa cautions, since there is wide variation in current usage. See section 5 below for further development of this analysis.

3.2. Turkish Turkish, an "agglutinating" language which is a member of the Altaic group, is a rigid SOV language. Let us see then how it deals with the locative inversion sentences in the SVO languages which we have been examining up to now. The Turkish equivalents of (7) are given below: 22 (7)

a'".

Agag-siz bir alan-da tree-without one zone-LOC göriin-üyor-du. see-REFL-PROG-PAS

b"\

Bir avlu-nun uc-u-nda one courtyard-GEN end-POSS.3-LOC kulübe dur-uyor-du. huts be-immobile-PROG-PAS

c"\

Duvar-da antik bir wall-LOC antique one dur-uyor-du. stand-PROG-PAS Uzak-lar-da far-PL-LOC bir tanker one tanker

sirin pretty

ocak chimney

defa-lar-ca time-PL-ADVR yavas yavas slow slow

bir one

kulübe cottage

birkag some

yikkin old

gengel-i asili hook-POSS.3 hanging

tamir ed-il-mis repair make-PASS-PART bat-iyor-du. sink-PROG-PAS4

178

Francis Cornish

e"\

Birden yalinayak üg genggocuk suddenly feet.bare three young.child bahge-ye gir-di. garden-ALL enter-PAS

f".

Kapi-da door-LOC gizemli Mysterious

g"\

Bir yigin kuru yaprak-lar-da one heap dry leaf-PL-LOC kus yuval-iyor-du. bird nestle-PROG-PAS

kos-arak run-PART

gotik harf-ler-le kaz-il-mis gothic letter-PL-COM engraved-PASS-PART bir yazit bul-un-uyor-du. one inscription find-REFL-PRQG-PAS kügük kirmizimsi small reddish

bir one

h " ' . Bu biiro-da dort kisi galis-iyor. D1 office-LOC four persons work-PROG Among these examples, it is worth mentioning example (7e"'), which cannot work in the same way as the parallel examples (7e), (7e') and (7e"); the reason is that bahge-ye 'in the garden', since it is a complement of the verb girmek, cannot be separated from it by being preposed clauseinitially.23 This is interesting in the sense that it reinforces an observation made by Fournier & Fuchs (2002): namely, that when the term preposed clause-initially in French is tightly governed by the verb, the latter is strongly "attracted" to it, a feature which is expressed via its placement in second position in the clause, in front of the subject. Since in Turkish, the verb is constrained to occupy the final position in the clause, preposing of the locative phrase which it governs, and hence the separation of these two constituents, is impossible. The immediate observation to make on the basis of the Turkish data is that there is no subject-verb inversion in this language; the most we can say is that there is "inversion" between subject and preposed locative/temporal adverbial, the verb remaining in its canonical position clause-finally. We thus have the following order of constituents: locative/temporal adverbial + subject + verb. Turkish being, as already noted, an SOV language, the unmarked canonical order of the constituents at issue in these examples would be Subject + locative/temporal adverbial + finite verb. A constant feature in relation to the examples of subject-verb inversion which we have

A crosslinguistic study of 'locative inversion' 179 seen so far in terms of the SVO languages would be the placement of the rhematic, focal subject in the position reserved for this purpose in the type of language at issue: postverbal in the SVO languages, and immediately preverbal in SOV languages such as Turkish (see also Hungarian in this respect). If the constituent placed in this position carries the nuclear pitchaccent, it is assigned contrastive focus. When the subject is in second position, but not the immediately preverbal one,24 it is most often interpreted as part of the rheme, even if it does not bear contrastive focus in this position (Gerd Jendraschek, p.c.). Moreover, in Turkish - as also in Arabic and Russian, amongst other languages - subject-adverb order signals the definiteness and hence potential topicality of the subject (no articles existing in Turkish - apart from the indefinite article bir): Qocuk-lar yer-de yat-iyor child-PL ground-LOC liePROG 'The children are lying on the ground', in contrast to Yer-de gocuklar yat-iyor ground-LOC child-PL lie-PROG 'Children are lying on the ground' (examples due to Gerd Jendraschek). So all the examples ( 7 a ' " h ' " ) , which show adverbial-subject order, would contain indefinite subjects - with the exception of example ( 7 e " ' ) with the adverb birden 'suddenly', according to Gerd Jendraschek, which is a sentence adverb, and so is more autonomous in relation to the clause. On the other hand, when there is no adverbial phrase in front of the subject, the sentence is ambiguous with respect to definiteness.

3.3. Classical and modern Arabic 25 Arabic is a basic VSO language (cf. Dahlgren, 1988, Moutaouakil, 1984); however, Ahmed Moutaouakil informs me that it is currently changing from a more or less flexible VSO order towards a rigid SVO one. Like Turkish, Arabic is a case-language, a property which in principle allows for flexibility in word order. According to Albert Abi Aad (p.c.), when the sentence is marked for past tense, a quasi-auxiliary morpheme is used, namely käna26 with the lexical verb occurring later on in the sentence in the neuter imperfective form. Still according to Abi Aad, in this case, the subject would occur to the left of this second verb, since it will have been thematised by käna. According to Moutaouakil, the Arabic sentence requires two special positions clause-initially rather than one (as in (1) above), since PI contains the complementizers and PO the interrogative pronouns, or a constituent marked for contrastive focus, or (optionally), a

180

Francis Cornish

non-subject topic.27 See (11) below, the positional template proposed specifically for Arabic by Moutaouakil (1989:170) (the two clause-initial positions here are to be seen as solely relevant to the facts of Arabic): (11) P2, PI PO V S 0 Χ, P3 (Moutaouakil, 1989:170. positional template required for Arabic) (12) a. halidan (PatObj) 'absartu ('la 'amran), Halid-ACC have-seen-I (not Amr-ACC) 'It's Halid that I saw (and not Amr).' b. ??? zaydun (AgSuj) gä'a (la 'amrun). Zayd-NOM has-come (not (Amr)' 'It's Zayd who came (and not (Amr).' (Moutaouakil, 1989: 62, ex. (182) and (183)) As in the other languages analyzed so far, the (canonical, in Arabic) positioning of the subject to the right of the verb signals its rhematic, and not thematic, status - though according to Moutaouakil (1989:76-77), topic constituents may remain in their canonical position rather than being placed in P0. However, he also notes (1989: 77) that the preferred position of topic constituents is towards the beginning rather than the end of the clause. See (7a" " - h " " ) below for translations into modern Arabic28 of examples (7a-h), which we have already seen in English, French, Italian and Turkish: (7)

a " " , fi fushatin min algabati in an.opening of DEF-forest kuhun 'aniqun. cabin-NOM pretty b " " . dahila alfina'i kanat inside-LOC DEF-courtyard was masakinu qadimatun. hut-PL.NOM ageing

kana yatara'a was visible-V

taqumu stand-V-3SG.F

A crosslinguistic study of 'locative inversion' 181

c " " . 'ala alha'iti kanat on DEF-wall was atlqatun. old-NOM

tatadalla usratun hang-IMPERF chimney-hook-NOM

d " " . fi alb'ädi kanat naqilatu naftin baliyatun in DEF-distance was tanker patched-up tagraqu bibut 'in. sink-IMPERF-3 SG slowly e " " . fag'atan ada fi alhadiqati Jalatatu suddenly run.up-3SG.PERF in DEF-garden three subbanin hufatu al'aqdami. boys bare-GEN DEF-feet-GEN f".

c

ala albabi kanat cqad' nuqisat on DEF-door was-3SG.F engraved-PASS.PERF bihhurufin gutiyyatin 'ibaratun in-letters-F.NOM gothic-F.PL formula-NOM gämidaiun. obscure.F.NOM

g " " . fi kawmati 'awräqin maytatin kana yarbudu in heap leaves-GEN dead was nestled-PERF tayrun ma 'ilun ila alhumrati. bird-NOM veering towards redness-GEN h " " . fi hada almaktabi ya'malu in this-F. DEF-office-F. there-work-3SG ; arba catu 'ashäsin. four-NOM persons-PAR Apart from example (7d""), all the Arabic examples translated from (7ah) show the same basic structure: LOC/TEMP ADV - (kana(t)) - V ( + ADV) - SUBJECT. Arabic prefers the order theme-rheme to the opposite according to Abi Aad, when the verb is clause-initial in the canonical VSO order, it is itself thematised, in fact. Thus the preposed locative or temporal adverbial, as in all the examples (7a" " - h " " ) here, would necessarily be understood as thematic. In this case, a definite subject (also thematic, then)

182

Francis Cornish

could not precede the verb. According to the positional template given in (11), we might say that the initial locative adverbial occupies either PO, or P2. In PO (as in the examples here), it would be interpreted as topical, in the strict sense - which would prevent a definite, also topical, subject from occurring therein. And in P2, it would have a "framing" type of interpretation, rather than a topical one. Regarding subject-verb agreement in classical Arabic (see also Bakker, this volume), (13) below gives the rules formulated by Neyereneuf & AlHakkak (1996): (13)

Rules for subject-verb agreement in Arabic 1. in the first and second persons, .... the verb agrees in gender and in number with the subject; 2. in the third person, when the verb precedes the subject, it remains in the singular and agrees with the subject if the latter is animate (in both singular and plural) or inanimate (in the singular); 3. the verb assumes the feminine form if the subject is inanimate and plural.

Indeed, as far as the second rule is concerned, we see in examples ( 7 b " " ) , ( 7 e " " ) and ( 7 h " " ) , where the postverbal subject is plural, that the verb takes the 3 rd person singular form. The situation is similar to what one finds in Russian, according to Corbett (2001: 17), where a verb with a quantified subject may take either 3 rd person singular form, or 3 rd person plural - the order subject-verb favouring plural agreement, while verb-subject order prefers singular 'agreement'.

4.

Towards an account of the locative inversion construction within Hengeveld's FDG model

First of all, we note the existence of certain common constraints within the six languages represented here, in three of the four major features of the construction, initially isolated on the basis of a consideration of English and French in section 2. First, the preposed locative or temporal phrase - in initial position in all the six languages - fulfils a thematic function, assuring the link with the discourse context upstream and also the anchoring 'ground' in which the new subject referent is to be located (Hannay, 1991: 145-6 proposes the term "Stager" as a characterization of the function of

A crosslinguistic study of 'locative inversion'

183

such constituents within PI; but his examples of this (see (19) below) involve predicates and not locative or temporal adverbials, which I would argue are not only "topical", but (Sub)Topics; yet since they do nonetheless perform a "stager" function with respect to the discourse-new entity to be presented, they share these two functions). Next, the Subject occupies a rhematic position in relation to the verb, occurring either to its right in the four basic SVO languages, and in the one VSO one (Arabic), or to its left, but nonetheless in a special focus position P4 immediately in front of the verb in the SOV one (Turkish) (see the extended template under (14) below); it may also occur to the left of this position but belonging to the rheme, in the case of Turkish. Finally, from the point of view of information structure, the order VS is indicated for unaccusative verbs in five of the six languages, expressing by default a thetic interpretation (wide focus, including the verb); for the unergative verbs, in Italian and Portuguese this order normally expresses a narrow, contrastive focus, excluding the verb, which would in this case be presupposed. Both types of reading - thetic and contrastive focus - are possible in this case only for Arabic and Turkish: in the latter case, the subject does not invert with the verb but with the preposed adverbial. We can see, then, that the common properties in the realization of this construction across the six languages, representing three basic order types, far outstrip the differences. (14) below presents an extended version of the positional template given under (1), which takes account of Turkish and Arabic. Note that I have replaced the position PO proposed by Moutaouakil (1989) in the template for Arabic under (11) by P5, since a position PO has independently been proposed for the Slavic languages by Stanchev (1997) and Siewierska (1998) clause-finally. This latter position was already included in template (1). In addition, position P4 immediately in front of the position available for the verb, would be needed for Turkish as well as for other SOV languages. According to this extended template, the subject may be placed within this construction, depending on the language at issue, either in its pattern position for the language in question, otherwise in position P4, or else in the special position PO clause-finally. Given that Turkish is a strict V-final language, however, this latter position would not be available for subjects in that language. (14) P2, PI (P5) (S) (Ο) (X) (P4) V (S) (Ο) (X) (PO) (Χ), P3

184

Francis Cornish

Several properties of the locative inversion construction are specific to it, as we have seen: the locative or temporal preposed argument, level-1 or level-2 term placed in PI rather than P2 (which position may well be filled by level-2 satellites, cf. section 2.1); the non-predicating, purely locative or existential verb in second position (more or less equivalent here to a copula, as is suggested in section 5); and the rhematic subject term in P4 or PO (according to the language type), signalling new information relative to the context. Within a wider discourse context, the construction as a whole serves to background the situation evoked, including the new subject referent (cf. Huffman, 2002, Borillo, 2000, Cornish, 2001). What we see from our examination of the construction across six languages is that there are a number of interacting dependencies holding between the various components of the construction: If (as is the most common situation) the preposed locative or temporal term is closely linked to the verb, then the fact that it is preposed and does not occur in its canonical position to the right of that verb (in the five languages besides Turkish examined here), attracts the verb into second position, so that it is adjacent to it - again, Turkish is the one exception to this tendency. The initial position (PI) occupied by the locative or temporal term is determined by the discourse functionality of the construction as a whole (cf. Birner, 1994, Birner & Ward, 1998, Huffman, 2002). The fact that, when the verb is tightly connected with this preposed term, it inverts with (or, in the basic SVO languages, occurs to the left of) its subject, and, as a corollary, the subject is rhematized, entails that the verb is non-predicating, but simply points towards the existence of the entity denoted by the subject. In order to account adequately for all these properties, it is necessary to have available a level of analysis at which the determining discourse properties of the construction may be represented. Such a level is provided in Hengeveld's (2004a,b) Functional Discourse Grammar (FDG) model in the shape of the Interpersonal Level. Unlike standard FG, FDG is a top-down, hierarchical, modular discourse (and not sentence) grammar model of language, which interfaces with a Cognitive and a Communicative Context component. These latter components are in part non-linguistic in nature. As Hengeveld (2004a) notes, "decisions at higher levels and layers of analysis determine and restrict the possibilities at lower levels and layers of analysis" (this property reflects the top-down character of the model). Each of the levels or modules within the grammatical Component connects with the module immediately below it via a set of mapping or interface rules, which convert the representation issuing from the higher module into that appro-

A crosslinguistic study of 'locative inversion' 185 priate at the lower level. At the Interpersonal level, then, message management strategies would be planned and structured (in terms of potential moves, and the discourse, referential and ascriptive acts which may implement them). The representations established at this level then feed the Representational Level at which the grammatical and semantic structuring of the construction may be specified, prior to the Structural Level (morphological form, syntactic function assignment and constituent ordering) and Expression Level, at which matters of phonological and prosodic form are specified. The highest Interpersonal Level also feeds (and is in turn fed by) the Conceptual and Contextual components (see fig. 3 in Hengeveld 2004b), assignments from which are also clearly necessary for an adequate and complete analysis of the construction under study. See Cornish (2003: §4) for brief discussion of the possible treatment of the four FG Topic functions within the FDG model. At the initial Interpersonal Level in the FDG model, we may specify the speaker's discourse intentions in terms of Hannay's (1991: 148) five "modes of message management" (All-New information, Topic, Reaction, Neutral and Presentative). Hannay envisages these types of message mode as being subcategories of the DECL illocution (mood) operator (DECL-A: All-new mode, DECL-N: Neutral mode, DECL-P: Presentative mode, DECL-R: Reaction mode, DECL-T; Topic mode (Hannay, 1991: (21), p. 148)). Examples illustrating Hannay's message management modes are as follows: (15) All new mode (= thetic utterances) A bomb exploded yesterday in Armington Valley high street. (15) is a so-called 'thetic' utterance (see also examples (9a,b) from European Portuguese presented in section 3.1.2). Here, the context is minimal, no constituent being singled out for special treatment by means of word order (at least in the case of English). (15) might be the first utterance of a Radio or TV news bulletin. Such utterances respond to the very general question What happened?, and serve to create context for what is to come. In particular, the 'starting point' constituent has no special importance relative to the remainder of the clause. I therefore disagree with Hannay (1991: 146-7) when he claims that the subject of such utterances will be placed in PI position, simply because this is "in keeping with the idea that clause-initial position is the unmarked position for Subject constituents in English" (1991:147). PI, after all, is a position which marks constituents

186

Francis Cornish

singled out for special treatment, in particular, for pragmatic reasons; and yet in thetic utterances such as (15), no constituent is singled out in this way (this is the hallmark of thetic utterances, precisely). I would argue, then, that in All-new (thetic) utterances, the clause constituents will each be assigned to their pattern positions. (16) Topic Mode Q: Have you thought of going to London? a. No, I hadn't considered London actually. b. No, London I hadn't considered actually. (Hannay 1991, ex. (la,b)) Here, in both (16a and b), there is a relation of aboutness holding between 'London' and the pragmatically-construed proposition (Lambrecht, 1994) - particularly so in (16b), where London is in PI, since this constituent's referent is singled out for special treatment as topic, according to Hannay. Here, the speaker has clearly chosen to anchor his message to the common ground presented in the question ('going to London'). The Focus would then be the subject (which would be 'topical', but not Topic, in (16b)) and the verb complex hadn't considered. (17) Reaction mode Q: Did you get wet? A: Wet? Bloody soaking I was. (Hannay 1991, ex. (2b)) The Reaction mode involves the preposing of immediate, expressivesubjective material bearing the Focus function. Here, the foundation chosen by the speaker for his message is "the most important focal information" (bloody soaking in the second pair-part utterance in (17)), in conformity with Givon's (1988) "first-thing-first" principle, as Hannay points out (1991: 151). There is no topic here, according to Hannay (p. 143) - in fact, the highly predictable (and thus presupposed) subject-copula segment following the adjective phrase in (17) may be analysed as a Tail constituent: indeed, it has the intonational markings of Tails. Moreover, it could be replaced in (17) with very little effect by a corrective phrase such as ..., more like it!, which would clearly be a Tail constituent.

A crosslinguistic study of 'locative inversion' 187 (18)

Neutral mode It was surprising that the Greens won the election.

The Neutral mode is said not to involve an initial Topic or Focus expression and comprises a dummy subject (it or there). (18) is not constructed from the standpoint of a Topic (there is none); rather the speaker "just builds up the Focus" (Hannay 1991: 147). Now, as in the case of examples of type (15) (in the All New mode), impersonal constructions such as (18) are clearly thetic utterances, since no constituent is singled out for special treatment, and the entire clause is in Focus. The only significant difference between (15) and (18) is that in the first case, the subject is a term phrase (a bomb), and in the second, it is an impersonal pronoun. I would thus analyse both (15) and (18) as exemplifying subtypes of the All New mode. Hannay also places dummy it in examples like (18) in PI - presumably on the same basis as for lexical subjects like a bomb in (15), i.e. the fact that it is a subject expression, and "the normal position for subjects in English is PI". But there is even less justification for this treatment than in the case of a bomb in (15), since a dummy element has no reference at all, and is simply a "place-holder" - English main clauses requiring a subject in the syntax. Thus such constituent types can hardly be said to be "singled out for special treatment" in any conceivable way, and so should not be placed in PI, but in the pattern position for subjects in English. (19)

Presentative mode CONTEXT: I will now turn my attention to various implications of the analysis. Particularly interesting is its influence on Topic assignment. (Hannay 1991, ex. (19b))

The predicate constituent particularly interesting in (19) is neither Topic nor Focus, according to Hannay, but corresponds to what he calls a "staging device" (Stager). The information it conveys in context is clearly inferable from the context clause, and as such is topical. Thus it is used by the speaker to introduce a new discourse topic. The definite subject term phrase its influence on Topic assignment is therefore assigned the pragmatic function Focus. Clearly, the constituent bearing the Stager function will be placed in PI. As for the locative inversion construction as analysed in sections 2 and 3 above, it is clear that this is a particular subcategory of the Presentative mode of message management.

188

Francis Cornish

Now, clearly, each message management mode, with the exception of the first and the fourth (the All New and Neutral modes), presupposes a specific kind of context. This must be represented in the Communicative Context component, whose task is to provide a representation of the discourse developed prior to the utterance about to be produced, as well as of the prevailing situational context. A very important part of this discourse representation must be (a) an indication of the current (local and global) macro-topics and (b) the structure of this discourse (i.e. the level or 'unit' of the discourse under development which the discourse has reached). The Presentative, Neutral and All New modes may serve to initiate a new discourse unit, since their essential function is to introduce new referents or states of affairs. In the case of the Presentative mode, as we saw in example (19) as well as those illustrating the locative inversion construction in sections 2 and 3 above, there is often a preposed thematic locative, temporal (or more generally, circumstantial) phrase introducing the constituent manifesting it; in which case, the content of the clause as a whole is presented as tightly connected with the preceding co-text. Thus, such clauses may serve either as transitions between one discourse unit and the next, or as links between utterances within the same unit. The structure of the discourse corresponding to a particular Move is thus established at the Interpersonal Level in the new model, in terms of a coherence relation with respect to the last such Move, and to the wider discourse and thematic structure as specified by the Communicative component. The mode of message management adopted at this level will in part determine at the lower Representational Level a particular underlying clause structure, with Topic and Focus assignments marked (as appropriate) as a function of the message structure indicated at the Interpersonal Level. Finally, the Structural and Expression components will convert these specifications into an actual object language expression Let us focus now particularly on Hannay's (1991) Presentative mode. Let's take the English example (7c) from section 2.3 as a basis for the demonstration (repeated as (20) for convenience). (20)

On the wall hung an antique chimney hook.

The first factor to be specified is the communicative context presupposed by the construction - to be provided, then, by the Contextual component. This is that there is an inferential relation in terms of the place or time at which the state of affairs is situated, with respect to a place or time acces-

A crosslinguistic study of 'locative inversion' 189 sible within the prior context of the occurrence of the construction. This relationship may be one of identity (coreference), of part in relation to whole, of "ingredient" within a frame, or simply of an association of some kind (e.g. via metonymy). Birner & Ward (1998: 17) call this a "po-set" (partially-ordered set) relationship.29 This information then is fed from the Contextual component to the Interpersonal Level. At this level, the speaker's move is one of wishing to present a discourse-new referent to the addressee within such a staging context. In terms of the ascription of a property to an argument, there is none here, as we have seen (cf. section 2.3), the verb in the construction being nonpredicating. There are however two referential acts, involving the establishment of the existence of a discourse-new referent within a given locative or temporal context. According to these values, then, the locative argument or level-1 or 2 satellite in the construction will be assigned (by hypothesis) the pragmatic function Topic (more specifically, I would argue, SubTopic as well as Hannay's 1991 Stager function); and the argument whose referent is to be introduced in this context, the function Presentational Focus. At this level, then, we derive (20a) (note that the locative adverbial on the wall in (20) is an argument of hung, and not a level-2 satellite): (20)

a. Interpersonal Level [M: (A, [DECL-T(P,)S(P2)A (C,: [(R, ( A 2 [ D E C L - P (PI)S

(loc/temp)Sub.Top/st)])]),

(P 2 ) A (C 2 : [(R 2 (XWFOC)])])]

(20a) says that we have a single move consisting of two acts, the first involving reference to a locative or temporal entity bearing the dual function SubTopic/Stager, and the second, to an entity bearing the function Presentational Focus. The symbols used in (20a) include the following: 'M' = 'Move'; Ά ' = 'Discourse Act'; 'DECL-T' = 'Declarative illocution in Topic mode'; 'DECL-P' = 'Declarative illocution in Presentative Mode'; '(Pi)s' = 'the Speaker of A n '; '(P 2 ) A ' = 'the Addressee of A n '; 'C 1/2 ' = 'the communicated content associated with Discourse Act 1/2 '. At the following level in the derivation, the Representational Level, a predication whose verb merely denotes existence or localisation (here, hang) is chosen, in conformity with the absence of property-ascription specified at the preceding Interpersonal Level. This verb, as we have already seen, has two arguments - an entity - and a locative-denoting one, respectively. This yields (20b):

190

Francis Cornish

(20) b. Representational Level [DECL-P Ej : [POS pi: [PAS e,: [hang [V] (ilx;: chimney-hook [N]0: antique [A]0Pres.Foc) (dlxj: wa//[N]LocSubTop/St)]]]] Furthermore, at the next stage in the derivation, that of the Structural Level (Hengeveld 2004b), the first argument will be assigned the syntactic function Subject: (20)

c. Structural Level [DECL-P Ej : [POS pi: [PAS e;: [hang [V] (ί1χ(: chimney-hook [N]0: antique [A] 0 p res -FocSubj) (dlxj: v r a / / [ N ] L o c S u b T o p / s t ) ] ] ] ]

Finally,

still

within the Structural Level, the locative argument or alternatively the level-1 OOVoc/TempSub-Top/st or level-2 2 satellites (yi) LoC/TempSub-Top/st (whichever of these form types is chosen in any instance) will be placed in PI position30 on the constituent order template in (14), the locative or existential verb in V, and the first argument (^l)pres-FocSubj will go to P0. In the case of the SOV language represented here (Turkish), the latter will be placed under P4, there being no 'P0' position available for such languages (cf. Herring & Paolillo, 1995). (21) gives the placement rules relevant for the constituents of the locative inversion construction in the six languages represented, in terms of template (14). (*2)Loc/TempSub-Top/st,

(21)

Placement rules relevant for the locative inversion construction 1- A Loc/Temp Pl> else 2. σ LocATemp PL else 3. a^oc/Temp "> PI (if "SubTop/St"), and P2 (if "Theme") 4. V - > V 5. If any of (21.1-21.2) or the first part of (21.3) have applied, then SubjPres_Foc -> P0 (if the basic constituent order is SVO or VSO), else -> P4 (if basic c.-order is SOV)

A crosslinguistic study of 'locative inversion'

191

5. Concluding remarks: the agreement properties of the construction in certain languages, the subjecthood of postposed 'subject' terms, and the impersonal status of locative inversion Finally, let us turn to the question of agreement, and whether the postposed subject is still a genuine Subject. We have noted at several points in the comparison of the locative inversion construction in the six languages up to now that agreement in number could be lacking between preposed verb and postposed subject: this was the case with the unaccusative verbs in European Portuguese (examples (10a'-c')), but only in the context of the colloquial, spoken language;31 as well as in the case of Arabic, whatever the type of verb involved.32 Arabic goes even further than colloquial European Portuguese, in that it allows non-agreement in number even in the case of the order subject-verb: this would be motivated by the fact that the subject is inanimate and plural. This state of affairs, that is, the absence of agreement between a verb and its subject, where agreement is otherwise regularly expressed between these two items in the language in question, is not rare in the world's languages, as Lazard (1994) reminds us. One factor which would motivate this, according to Lazard, is what he calls the speaker's "communicative goal" ("la visee communicative") - in other words, the thematic or rhematic status assigned either to the verb, or to the subject, or to the subject+verb (or verb+subject) unit: ".. .En regle generale, l'accord tend a ne pas se faire quand l'actant concerne et le verbe appartiennent au meme membre (rheme ou, plus rarement, theme) de la structure de visee. Au contraire, un actant thematique tend ä etre coreferencie dans la forme verbale, qui est par nature typiquement rhematique." [As a general rule, agreement tends not to occur when the argument concerned and the verb belong to the same unit (rheme, or more rarely, theme) of the information structure. On the other hand, a thematic argument tends to be cross-referenced on the verb, which is by nature typically rhematic] (Lazard, 1994: 212)

Although it does not involve subject-verb inversion, one of the examples presented by Lazard is particularly instructive. See the pair of Persian sentences taken from Lazard (1994: (2a,b), p. 174) under (22a and b):

192

Francis Cornish

(22)

a. mehmän-hä ämad-and. guest-PL come/PAS-3PL 'The guests have come.' b. mehmänhä ämad-e. guest-PL come/PAS-3SG 'There have come a number of guests.'

As Lazard (1994: 214) explains, from the point of view of information structure (la "visee communicative"), (22a) includes two units, a theme, mehmänhä 'the guests', and a rheme, ämadand 'have come', and the verb agrees in number and person with the theme constituent. According to Lazard (1994: 214), "the sentence is pronounced with two pitch-accents". In (22b), on the other hand, there is only a single pitch accent, which falls on the nominal term, the sentence is "entirely rhematic", and the verb does not agree (at least in number). Lazard's comment (p. 234) on the rhematic subject which no longer determines an agreement marking on the verb in this configuration, is eloquent: "...can we still call it a subject?"; for indeed, one of the cross-linguistic properties of the subject is the determination of an agreement marking in the verb with which it is in construction. It is precisely this state of affairs which characterizes the examples mentioned a moment ago where the subject is postverbal, bearing a rhematic function, then, and where agreement with the subject (in number, at least) is not expressed. It is in fact no accident that, in every case, it is the 3rd person singular form of the verb which appears; for it is precisely these values which are the unmarked feature values of the morphosyntactic categories involved - namely, 3rd for the category of person, and singular for that of number. Perlmutter (1983), in a very interesting article in relation to constructions with and without inversion in Italian, claims that inversions involving unaccusative verbs (whose subject corresponds to an internal but not external argument, in relation to the verbal predicate) are equivalent to an impersonal construction, where the subject is an unexpressed expletive pronoun. Whenever the verb does not agree with its (lexical) subject, as in the examples of spoken Portuguese and standard Arabic we saw earlier, the true impersonal status is manifest, via the "default" verbal inflections, i.e. 3rd person, singular number and neuter gender, where relevant - the subject then being an unexpressed expletive pronoun. This is manifest overtly in French with the impersonal construction which is possible with unaccusa-

A crosslinguistic study of 'locative inversion'

193

tive verbs, as in: II est arrive/venu quinze touristes 'There arrived/came 15 tourists'. The impersonal status which inversion constructions would represent - where there is a choice with respect to an uninverted variant would be entirely congruent with the presentational function of inversion constructions generally. Moreover, the postposed subjects of unaccusative verbs in Italian cannot control the reference of the unexpressed subject (PRO in Generative terminology) of an embedded infinitival or participial clause, nor even act as antecedent for a 3rd person pronoun within the sentence of which the preposed verb is main verb, according to Carminati (2001: ch. 3). I would explain these "defective antecedent" properties in terms of the following consideration: a definite unaccented pronominal expression is ipso facto thematic, in other words, its referent is assumed to be familiar to the addressee; yet at the same time, the referent of the postposed subject (in SVO languages) is rhematic, focal, hence presented as "discourse-new" to the addressee: whence the contradiction and the impossibilities observed. The following constructed example of defective cataphora illustrates this: (23) #While he\ was leaving the house hex had just broken into, the alarm was accidentally set off by a Dover-based burglarj.33 For Relational Grammar (the framework chosen by Perlmutter for his study), postposed subjects of unaccusative verbs would have the status of grammatical chömeurs, since their prime function of acting as pivot at an earlier stratum in the derivation would have been usurped by another nominal constituent at a later stratum - in the present case, that of the null expletive, which we can identify with the abstract LOC constituent proposed by Pinto (1997) (see section 3.1 above). If the locative inversion construction is at base an impersonal construction, then it should be possible to insert in preverbal subject position, in those SVO languages which do not allow the non-realization of a pronominal subject, an expletive pronoun (e.g. there in English and impersonal il in French). Let us try to apply such insertions first of all in the canonical English, then the French examples presented earlier in (7a-h) and (7a'-h'): (24) a. b. c. d.

In a clearing there could be seen a charming cottage,34 At the end of a courtyard there stood some ageing huts. On the wall there hung an antique chimney hook. In the distance there was slowly sinking a patched-up oil tanker.

194

Francis Cornish e. Into the garden there suddenly ran three young barefoot boys. f. On the door there was engraved in gothic letters a mysterious inscription. g. Nestling in a pile of dead leaves there was a tiny reddish bird. h. ?In this office there work four people.

(25) a. b. c. d. e.

Dans une clairiere il se voyait une coquette chaumiere. Au fondd'une cour il se dressait de vitustes baraquements. Sur le mur il pendait une antique cremaillere. Dans le lointain il coulait lentement un pitrolier rafistole. Dans le jardin il accourut soudain trois jeunes garqons aux pieds nus. f. Sur la porte il etait gravee en lettres gothiques une inscription mysterieuse. g. Tapi dans un tas de feuilles mortes il y avait un oiselet rougeätre. h. Dans ce bureau il travaille quatre personnes.

On the whole, such insertions do not greatly affect the grammaticality of the inversion sentences, nor the type of interpretation to which they give rise - though this is more clearly the case in English with expletive there than with the impersonal pronoun il in French. As far as the latter realizations are concerned, we may observe that the more the preposed verb is empty semantically (tending towards a mere copula, as in (25a,b) - cf. Blinkenberg (1928: 105) - , the less the insertion of impersonal il is likely to modify the meaning and grammaticality of the original example. In (25g), the copula etait 'was' would be replaced by the existential presentational form ily avait 'there was'. (25e) and (25h) would not be candidates for the impersonal construction, since their verbs (respectively, accourir 'to run up' and travailler 'to work') are unergatives and not unaccusatives. As such, their sole argument is not internal, but external. This supports Perlmutter's (1983) analysis, then, according to which it is only unaccusative verbs which give rise to an impersonal construction when their 'subject' 35 is postposed in relation to the verb. It is for the same reason that (24h) is not fully acceptable in English (as was already its inverted counterpart (7h)). Yet, rather than subscribing entirely to Perlmutter's (1983) hypotheses, it would seem preferable to conclude from these non-subject properties of postposed subjects of unaccusative verbs,36 that these NPs are still subjects, but not canonical ones (see note 34). This is also the position defended by

A crosslinguistic study of 'locative inversion'

195

Lambrecht (1994: 22). It is in the last analysis the subject's marked status as bearer of the pragmatic function Presentational Focus, and not the unmarked one of Topic, strongly expected in the case of canonical subjects, which seems to lie behind all these non-subject properties. This would be yet another example of a pragmatic property overriding a syntactic one.

Notes 1.

2. 3. 4.

5. 6.

The present chapter is an expanded and revised version of a paper presented at the 10th International Conference on Functional Grammar (26-29 June, 2002, University of Amsterdam), entitled "Locative inversion in 8 languages: syntax, semantics, discourse-pragmatics and functional position". I would like to thank John Connolly, Joäo Costa, Daniel Garcia Velasco and Ahmed Moutaouakil for their help in revising the original paper, as well as an anonymous external reviewer and the editors of the present volume for their advice on revising the originally submitted version. I am also grateful to Christian Molinier, Maria-Nella Carminati, Gerd Jendraschek, Ahmed Moutaouakil, Joseph van den Reysen and Albert Abi Aad, for their help with the French, Italian, Turkish and Arabic data presented in the chapter. Any remaining errors are my sole responsibility. All basic SVO languages, at least in main clauses. Functional Grammar rules out movement rules, and in fact transformational rules in general. Below are some technical terms used in other frameworks than the FG one, and which I will be using in what follows, together with their definitions: unaccusative (or ergative) verbs: verbs having an "internal argument" (i.e. A2) which is a Patient or Zero in terms of semantic function, but no "external" argument (i.e. Al). In their (basic) intransitive use, these verbs have an inchoative or stative value; they may also be used transitively, however, in which case they acquire an Agent or Force Al, which induces a causative sense. Examples from English are verbs like burn, melt, cook, sink... Unergative verbs are intransitive verbs with an Al (i.e. "external argument" in Generative frameworks), but which may not be used transitively. Examples are work, stroll, laze, think... A NewTop, in standard FG terms. See also in this regard Birner & Ward (1998: 31), who note that the preposing of lexically governed constituents is more restricted than that of other types of (merely modifying) constituents.

196 7. 8. 9.

Francis Cornish Witness *Les chaussures se trouvaient, 'The shoes were to be found'. See also Connolly (1991: 78) for further evidence of the placement in PI position giving rise to subject-verb inversion in English. Presumably because the grammatico-semantic connexion between verb and satellite is too tenuous to attract the verb into second position. Nonetheless, Joäo Costa (p.c.) suggests to me that it would be prudent to test for the possibility that it might be the type of verb (here unergative) rather than the nature of the preposed PP itself as such, which rules out the inversion here. However, if we take an unaccusative verb in place of the unergative one (gagner), the result is little better: ?*Selon mon frere, sont arrives une dizaine de garfons 'According to my brother, have arrived about ten boys'. It would seem, then, that it is clearly the level-3 (and not argument, level-1 or level-2) status of the preposed adverbial phrase, and not the nature of the verb per se, which is at issue here. If we replace the level-3 adverbial here by a level-2 one, the result is acceptable: Avec une heure de retard, sont arrives une dizaine de garyons 'With an hour's delay, have arrived about ten boys'.

10. (6e) is pragmatically odd, since there is no (obvious) predicate with which the verb couler 'to flow' could contrast, occurring as it does in end-focus position: after all, 'flowing' is the only normal thing which rivers can do. 11. For exceptions to the intransitive condition concerning French, see Korzen (1996) and Hobaek Haff (2000). Joäo Costa points out to me that locative inversion and the inversion one finds with transitive verbs in the Romance languages other than French, are two different phenomena, with distinct properties. 12. For an in-depth study of this type of verb in English, see Levin & RappaportHovav (1995). 13. It is exactly this situation which characterizes the possibility of inversion of subject and verb in the locative inversion construction in the Bantu language Chichewa, according to Bresnan & Kanerva (1989). 14. Christian Molinier (p.c.) points out that in the following pair of examples with the verb travailler 'work', subject-verb order, as in (i), can easily be interpreted as bearing an actualised temporal value (where the people involved are actually at work at the time of utterance). The felicity of the progressive aspect in the English gloss reflects this interpretation. However, such an interpretation is ruled out in the case of (ii), which manifests verb-subject order. Here, only the habitual interpretation seems possible. This factor is no doubt closely connected to the need for preposed verbs in this construction to be semantically light, and non-predicating (cf. (7h') in the text): (i) Dans ce bureau quatre personnes travaillent. 'In this office four people work/are working'

A crosslinguistic study of 'locative inversion'

15. 16.

17. 18. 19.

20.

21.

22. 23. 24. 25.

26.

197

(ii) Dans ce bureau travaillent quatre personnes 'In this office ?work/#are working four people' What Korzen (1996) calls "the predicative unit". In this respect, lexical subject-full verb inversion is clearly distinct from pronominal subject-auxiliary or - verb inversion - as in French interrogative structures, for example. Thanks are due to Maria Nella Carminati for having helped me with these translations. Unergative and transitive verbs, as well as some unaccusative verbs, according to Pinto. A thetic utterance is the use of a clause which presupposes no particular prior context, and in which no constituent is singled out for special treatment in pragmatic terms - the entire clause being in Focus. See the European Portuguese examples (9a,b). An English example is (15) in section 4. Costa (2001a: 8, n.4) points out that the data involving verbs which do not agree in number with their postposed subject would not be acceptable by normative grammars of Portuguese, despite the fact that such forms are "quite often" produced spontaneously by speakers. He also points out (p.c.) that when native speakers see such utterances written down, they often experience a (mild) shock, even though such forms would pass unnoticed in their spoken realization. The abbreviations used in the glosses for the examples from now on are as follows: ABS = absolute; ACC = accusative case; ADVR = adverbialiser; ALL = allative case; COM = comitative; D1 = 1st person demonstrative (i.e. "proximal"); DEF = definite article; F = feminine; GEN = genitive case; IMPF = imperfective; INDIC = indicative; LOC = locative; Μ = masculine; NEG = negation; NEUT = neuter; NOM = nominative case; PAR = partitive; PART = participle; PAS = past tense; PASS = passive; PERF = perfective; PL = plural; POSS = possessive; PRES = present tense; PROG = progressive aspect; 1 = 1st person; 3 = 3rd person; REFL = reflexive; SG = singular. I thank Gerd Jendraschek for translating these examples with a native speaker. Thanks go to Gerd Jendraschek for this information. A situation which can arise when an adverbial or a direct/indirect object is placed between subject and verb. This sub-section is essentially composed from information kindly provided by Joseph van den Reysen, Ahmed Moutaouakil and Albert Abi Aad, and derived from Moutaouakil (1984), (1989). An aspectual operator or even 'past time adverbial', according to Abi Aad (p.c.): see the examples ( 7 a " " - d " " ) and ( 7 f ' " - g " " ) below.

198

Francis Cornish

27. The fact that zaydun is a focused subject in (12b) accounts for the tension caused by its being placed in PO, according to Moutaouakil; this would also be the case if the subject term were topic. 28. My thanks are due to Joseph van den Reysen and Albert Abi Aad for having translated these examples. 29. The authors define this as "any set defined by a transitive partial ordering relation R such that R is either reflexive and antisymmetric, or irreflexive and asymmetric" (1998: 17). 30. In fact, in the case of the choice of a level-2 locative or temporal satellite, it would seem that there are two possible positions for such a form type to occupy: either PI (as for level-1 satellites), in which case the verb will follow without interruption (see example (4a) in section 2.1), or P2, in which case it will have a Theme-type value (in the standard FG sense) and not a (Sub)Topic one (see example (4b)). In such a situation, the verb will not need to follow the satellite immediately, and the subject term can then occupy PI - not as Topic, but as Focus. 31. An identical situation obtains with preposed unaccusative verbs in the Bergamasco dialect of northern Italy, according to Carminati (p.c.), as already noted. 32. Examples (7b""), (7e"") and (7h""), the first two containing an unaccusative verb, and the last, an unergative one. 33. The co-indices here indicate the identity of the intended referents, and the crosshatch, the pragmatically defective character of the example as a potential utterance. 34. Connolly (1991:68; and p.c.) regards sentences such as (25a) as containing two syntactic subjects: a 'anticipatory subject' {there) and a 'subject-proper' (α charming cottage) which represents the underlying subject in the FG sense. This would appear to correspond exactly to the traditional distinction between "apparent subject" and "real subject". However, an immediate objection which may be raised against this analysis is this: how can a given verb be said to have two subjects? A solution to this question would appear to be available within the FG framework, however, as follows: given the essential distinction drawn within this framework between function and position (part of the formal properties of a term bearing a given function), we might then say that the Subject function is, under certain circumstances - namely, where a dummy expletive occurs in the clause, within impersonal constructions of the kind examined here - , able to be "shared" between distinct constituents. Where, for example, a lexically-filled term is called upon to perform a pragmatic function other than the one it carries by default (here that of Presentational Focus for subject terms), then part of the properties associated with this function may be

A crosslinguistic study of 'locative inversion'

199

borne by a dummy expletive (position associated with subject function; case marking and, according to the language, determination of an agreement marking on the verb), and part by the lexically specified term occurring in PO or P4 position. This would then explain the absence of certain typical subject properties in the postposed Subject term, as discussed in this section. In terms of the extended template given under (14) in the text, English there and French il would occupy the S position to the left of V. 35. In fact, internal argument, since it is made a chömeur by the unexpressed expletive which fulfils the subject function in this case. 36. Position to the right and not the left of the verb, possible absence of agreement in number expressed by the verb, impossibility of controlling the reference of the unexpressed subject of a non-finite subordinate clause, etc.

References Birner, Betty J. 1994 Information status and word order: an analysis of English inversion. Language 70.2: 233-259. Birner, Betty J. and Gregory Ward 1998 Information status and non-canonical word order in English. Amsterdam: John Benjamins. Blinkenberg, Andreas 1928 L'ordre des mots en frangais contemporain. Copenhagen: Munksgaard. Borillo, Andree 1990 A propos de la localisation spatiale. Langue Franqaise 86: 75-84. 2000 Le complement locatif et le genre descriptif. In Studia linguistica in honorem Lilianae Tasmowski, Martine Coene, Walter de Mulder, Patrick Dendale and Yves d'Hulst (eds.), 85-95. Padua: Unipress. Bresnan, Joan and Jonni M. Kanerva 1989 Locative inversion in Chichewa: a case study of factorization in grammar. Linguistic Inquiry 20.1: 1-50. Carminati, Maria Nella 2001 The processing of Italian subject pronouns. Ph. D. diss., Department of Linguistics, University of Massachusetts. Connolly, John H. 1991 Constituent order in Functional Grammar. Berlin: Mouton de Gruyter.

200

Francis Cornish

Corbett, Greville G. 2001 Principles of the SMG agreement data-base draft of 30 November 2001. University of Surrey, UK (MS: 1-24). Cornish, Francis 2001 L'inversion "locative" en frantjais, italien et anglais: proprietes syntaxiques, semantiques et discursives. Cahiers de Grammaire 26: 101-123. 2002 "Downstream" effects on the predicate in Functional Grammar clause derivations. Journal of Linguistics 38.2: 247-278. 2004 "Focus of attention" in discourse: a comparison between the four FG topic functions and the systems of focus and deixis in the Columbia school of Linguistics. In J. Lachlan Mackenzie and Maria A. GomezGonzälez (eds.), 117-150. Costa, Joäo 2001a Postverbal subjects and agreement in unaccusative contexts in European Portuguese. The Linguistic Review 18: 1-17. 2001b Marked and unmarked inversion in Optimality Theory. In Subject inversion in Romance and the theory of Universal Grammar, Aafke C.J. Hulk and Jean-Yves Pollock (eds.), 91-106. Oxford: Oxford University Press. Dahlgren, Sven-Olof 1998 Word order in Arabic. Göteborg: Acta Universitatis Gothenburgensis. Dik, Simon C. 1997 The theory of Functional Grammar 1: the structure of the clause, 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Erteschik-Shir, Nomi 1997 The dynamics of focus structure. Cambridge: Cambridge University Press. Fournier, Nathalie 1997 La place du sujet nominal dans les phrases a complement prepositionnel initial. In La place du sujet en franqais contemporain, Catherine Fuchs (ed.), 97-132. Louvain-la-neuve: Duculot. 2001 Expression et place des Constituante dans l'enonce en francjais classique: la relation sujet-verbe et la relation verbe-objet. Langue Frangaise 130: 89-107.

A crosslinguistic study of 'locative inversion'

201

Fournier, Nathalie and Catherine Fuchs 2002 Les enonces ä complement prepositionnel initial et ä sujet nominal postpose. Talk given at the LATTICE Seminar Les expressions introductrices de cadres de discours et leur portee, Universite de Paris III, February 1st, 2002. Givon, Talmy 1988 The pragmatics of word order: Predictability, importance and attention. In Studies in syntactic typology, Michael Hammond, Edith A. Moravcsik and Jessica Wirth (eds.), 243-284. Amsterdam: John Benjamins. Hannay, Mike 1991 Pragmatic function assignment and word order variation in a functional grammar of English. Journal of Pragmatics 16: 131-155. Hengeveld, Kees 2004a The architecture of a Functional Discourse Grammar. In J. Lachlan Mackenzie and Maria A. Gomez-Gonzalez (eds.). 2004b Epilogue. In J. Lachlan Mackenzie and Maria A. Gomez-Gonzalez (eds.). Herring, Susan C. and J.C. Paolillo 1995 Focus position in SOV languages. In Word order in discourse, Pamela Downing and Michael Noonan (eds.), 163-198. Amsterdam: John Benjamins. Hobaek Haff, Marianne 2000 Regard sur l'inversion du sujet en franfais moderne. Revue Romane 35.1: 21-32. Huffman, Alan 2002 Cognitive and semiotic modes of explanation in functional grammar. In Signal, meaning and message: Perspectives on sign-based linguistics, Ricardo Ο. Otheguy, Wallis Reid and Nancy Stern (eds.), 311-337. Amsterdam: John Benjamins. Korzen, Hanna 1996 L'unite predicative et la place du sujet dans les constructions inversees. Langue Frangaise 111: 59-82. Lambrecht, Knud 1994 Information structure and sentence form: topic, focus, and the mental representations of discourse referents. Cambridge: Cambridge University Press. Lazard, Gilbert 1994 L 'actance. Paris: Presses Universitaires de France.

202

Francis Cornish

Levin, Beth and Malka Rappaport Hovav 1995 Chap. 6: The problem of locative inversion. In Unaccusativity: at the syntax-lexical semantics interface. Cambridge: Cambridge University Press. Mackenzie, J.Lachlan and Maria Ä. Gomez-Gonzälez (eds.) 2004 A new architecture for Functional Grammar. Berlin: Mouton de Gruyter. Moutaouakil, Ahmed 1984 Le focus en arabe: vers une analyse fonctionnelle. Lingua 64: 115176. 1989 Pragmatic functions in a functional grammar of Arabic. Dordrecht: Foris. Nolke, Henning 1995 Utterance focus: elements of a modular theory. Copenhagen Studies in Linguistics 18:74-108. Ono, Noriko 2001 On the interaction between lexical and constructional meanings. In Proceedings of the first international workshop on generative approaches to the lexicon, Pierrette Bouillon and Kyoko Kanzaki (eds.). Ecole de Traduction et d'Interpretation, Universite de Geneve, 26-28 April 2001. Perlmutter, David M. 1983 Personal vs. impersonal constructions. Natural Language and Linguistic Theory 1: 141-200. Pinto, Manuela 1997 Licensing and interpretation of inverted subjects in Italian. Ph. D. diss., University of Utrecht. Pustejovsky, James 1995 The generative lexicon. Cambridge, MA: MIT Press. Siewierska, Anna 1998 Polish main clause constituent order and FG pragmatic functions. In Functional Grammar and Verbal Interaction, Mike Hannay and A. Machtelt Bolkestein (eds.), 243-266. Amsterdam: John Benjamins. Stanchev, Svilen B. 1997 Pragmatic functions and special sentence position in Bulgarian. In Discourse and pragmatics in Functional Grammar, John H. Connolly, Roel M. Vismans, Christopher S. Butler and Richard A. Gatward (eds.), 121-135. Berlin: Mouton de Gruyter.

The agreement cross-reference continuum: Person marking in FG Anna Siewierska Dik Bakker

1.

Introduction1

Despite the fact that linguists and linguistic theories may fill in the details in vastly different ways, most of them would be relatively happy with the following more or less general definitions of agreement and crossreference. Agreement could be defined as the systematic co-variation of one linguistic element (the target) with another element (the controller) in some linguistic domain on a set offeatures which originate from the target (cf. Corbett 1983). The clause or the sentence is often, but not necessarily assumed to be the largest domain for agreement phenomena. Agreement features which can be found in varying combinations are person, number and gender, and according to some also in/exclusivity, definiteness, animacy, specificity, honorificity, and possibly others. Cross-reference, on the other hand is often taken to be the property of one linguistic element - an anaphor - to refer to another element in the context - its antecedent - on the basis of certain features. The context for cross-reference is typically, but not necessarily anything above the clause or sentence level. The set of features of the antecedent coded on the anaphor overlaps to a large extent with those mentioned above for agreement. Under the above definitions, in theories which take the sentence as the highest level of linguistic description agreement is about phenomena which should be accounted for in the grammar while the organizational aspects of cross-reference fall mainly outside the scope of in depth description. While notionally distinct, in practice agreement and cross-reference are not so easy to differentiate. Therefore a given instance of person marking may be treated as a manifestation of agreement in one theory, but of crossreference in another. Particularly controversial in this respect have proved

204

Anna Siewierska and Dik Bakker

to be cases of so-called pro-drop, that is constructions involving no overt controller as in (lb) as compared to (la) from Gumawana, an Oceanic language spoken in the Mine Bay province of New Guinea. (1)

Gumawana (Olson 1992:326, 308) a. Kalitoni i-paisewa. Kalitoni 3SG-work 'Kalitoni worked.' b. I-situ vada sinae-na. 3SG-enter house inside-3SG 'He entered the inside of the house.'

In Lexical Functional Grammar (LFG; Bresnan & Mchombo 1987; Bresnan 2001) for example, the person marker in (la), in the presence of the overt controller Kalitoni, is considered to be an agreement marker, the one in (lb) a bound cross-referencing pronoun. In the various versions of Chomskian theory (Chomsky 1981; 1995) the person markers in both (la) and (lb) are considered to be agreement markers, the only difference being that in (la) the agreement is with the overt subject Kalitoni, while in (lb) it is with the covert subject known as 'pro'. In contrast to both of the above analyses, in standard FG (Dik 1997a: 373f; 388f) the person marking not only in (lb) but also in (la) is considered to be a bound cross-referencing pronoun. Thus neither (la) nor (lb) are instances of agreement. Agreement in FG is restricted only to cases where the controller is always overtly present in the relevant domain. Since in pro-drop languages this is by definition not so, all languages manifesting pro-drop emerge in FG as displaying cross-reference and not agreement. Accordingly, coreferential NPs accompanying person markers in such languages as for example Kalitoni in (la) are treated as being in an appositional relation with the person form on the verb. This approach is closely reminiscent of the pronominal argument analysis of radical head-marking languages in the sense of Nichols (1992) as developed within GB syntax by Jelinek (1984) and Speas (1990), and in Role and Reference Grammar (RRG; Van Valin & La Polla 1997). The traditional FG position, however is more extreme in that it is applied not only to head-marking languages such as Dakota or Mohawk, but also to dependent-marking languages such as Latin or Spanish. Taking a discourse perspective instead of a sentence perspective we would like to suggest a reinterpretation of agreement and cross-reference

The agreement cross-reference continuum

205

within FG, seeing them as two extremes of a continuum rather than two more or less independent notions. Unlike Dik (1997a) we will therefore side with authors such as Moravcsik (1978), and more particularly with Lehmann (1982) and Barlow (1988). Using Seller's (1979) notion of Apprehension, Lehmann (1982:233) assumes that the same conceptual object - in this context: a nominal concept or referent - may manifest itself in various places in one and the same sentence, via different formal devices (full NP, pronominal form, agreement marker). So, rather than assuming a controller and its possible targets, all lexical and morphosyntactic elements in a sentence related to some underlying conceptual object together establish the 'apprehension' of that object in that sentence. Agreement in the strict sense is a problematic notion for some languages in the first place, since features of targets may in fact contradict those of the controller, as exemplified in (2) below. In (2a), from Arabic, the non-animate masculine plural subject 'agrees' with the verb marked feminine singular. In (2b), from Chichewa politeness forces a plural marker on the possessive pronoun (examples from Barlow 1988:95; 124). Obviously, there is competition between grammatical and semantic features in such cases. 2 (2)

a. ?al-jimaaly naam-at. DEF-camel.M.PL sleep-F.SG 'The camels slept.' b. Bambo anga. father.SG SG.POSS.PL 'My father.'

Barlow (1988) steers away from the traditional sentence perspective on agreement. In his Discourse Linking Theory (DLT), he locates agreement at the discourse level. According to DLT, instead of essentially being redundant elements in a sentence, agreement markers function as indentifiers of discourse referents or as topic continuity markers in their own right. As such they create cohesion in stretches of discourse. Radical Construction Grammar (RCG; Croft 2001) adopts Barlow's approach to agreement to a large extent. As a result, in RCG person markers are interpreted functionally as referents to cognitive/semantic entities rather than markers of syntactic relations. Since they are not seen as fillers of unique argument slots there is no constraint on the number of times person markers may appear in

206

Anna Siewierska and Dik Bakker

a single clause, as under the theta criterion of GB or the uniqueness principle of LFG. The proposal we will make below shares the multi-expression aspect with Lehmann's (1982) proposal and the discourse perspective taken by Barlow (1988). Furthermore it unites the notions of agreement and crossreference into one framework. Since we think that the linguistic phenomena in this area of linguistic description point to 'superficial' morphosyntactic differences between languages in the way they express discourse referents rather than to 'deep' semantic or pragmatic differences, we will pay more attention to aspects of expression than to underlying representations. In order to demonstrate how our rules work we will implement them on the basis of the dynamic expression component we have proposed elsewhere (Bakker 2001, this volume; Bakker & Siewierska 2004) rather than via the standard FG expression rules. Functional Grammar aims at pragmatic adequacy, and at modelling the real behaviour of the language user rather than the Ideal Speaker/Hearer (cf. Dik 1997a: If). In our view this implies that aspects of discourse should be included in all descriptions for which it is relevant and appropriate. In fact, FG should in principle first try to solve linguistic problems at the level of discourse before seeking explanations within the boundaries of a sentence. As we hope to show below, this point of departure puts FG more or less naturally in the position to take a unified perspective on agreement and cross-reference, and more specifically on the whole range of what are generally called person markers, irrespective of their morpho-phonological form: free, weak, clitic or affix. Ours seems to be a realistic point of departure, from the perspective of the speaker. Obviously, the speaker always 'knows' who or what he is talking about, even if the referent is introduced for the first time in the ongoing discourse. Therefore, restricting ourselves to a model of the speaker, we will propose that in underlying representations terms always refer to the discourse representation of their referent via the unique term variable. Depending on the speaker's estimate of the pragmatic knowledge of the addressee and the morphosyntactic constraints of the language concerned the right expression form will then be selected from the full range of term expression forms available in that language. Thus, term representations may run from a full indefinite nominal term with all kinds of extra restrictors where the speaker assumes that the referent is completely new for the addressee to a mere Same Subject marker or even zero expression for a Given Topic term. All features that are accessible via the term variable are available for this purpose, albeit not necessar-

The agreement cross-reference continuum

207

ily to the same extent for all different markers. Also, there is no principled upper limit to the number of times elements expressing the features of a certain term may be present in an utterance in some form or other. The article is organized as follows. In section 2, we will discuss several problems with the way FG currently deals with cross-reference and agreement, and with the treatment of person markers in general. Some of the arguments were already presented in Siewierska (2001). In section 3 we will explore which parameters play a role in person marking on the verb, and outline a unified approach to this domain. In section 4 we will implement these ideas for FG. A dynamic model of the speaker as was presented in Bakker & Siewierska (2004), and the set of expression rules that were introduced in Bakker (2001) will serve as a point of departure. Speech errors made in bilingual situations provide insight into the logistics of the speech process and the differences between languages. Section 5 will briefly discuss some such regularly observed errors related to agreement, which in our view give support to the analyses in the earlier sections. Finally, in section 6 we will see to what extent our proposal counters the problems with the standard solutions as discussed in section 2.

2. Agreement and cross-reference in FG: Problems with the current approach In our interpretation, the standard FG position towards agreement and cross-reference as formulated in Dik (1997a), and inspired by De Groot & Limburg (1986), is as follows. -

-

-

-

A bound marker on a target is an agreement marker if and only if its controller is obligatorily present in the clause, else it is a cross-reference marker. The latter, but not the former is the expression of the argument position with its full referential potential. Agreement markers are produced by the expression rules; they derive their features from the controller term in the relevant argument position (e.g. subject), from the first restrictor, etc. Bound cross-reference markers, such as subject markers on the verb in pro-drop languages, are coded in the relevant argument positions on the basis of the features which they express. When a cross-reference marker which occupies an argument position is accompanied in a sentence by a co-referential noun phrase or pronoun, the latter is assumed to be in (grammatical) apposition with the marker. The

208

-

Anna Siewierska and Dik Bakker appositional slot in the underlying representation is created by copying the argument position of the cross-reference marker, including its semantic function. Free pronouns are represented in underlying representations in the form of terms with abstract predicates for their first restrictors and which are retrieved from the lexicon.

The position taken above implies that only non pro-drop languages can have subject or object agreement, and can have terms rather than a set of features in the (original) argument positions of the main predicate. Prodrop languages will always have features in the argument positions and appositional slots for the corresponding nominal and pronominal terms in case of their presence in the sentence.3 This would give us the following rather different underlying representations. For English, in (3), we have an underlying representation with both terms in the argument positions; for Tauya, a Trans New Guinea language, which is pro-drop, we have the appositional representation in (4). (3)

The woman saw the men. [PST see [V] (dl x^ woman [N])Ag (dm x2: man [N]) Go ]

(4)

Tauya (MacDonald 1990) Fena?-ni fanu-0 nen-yau-a-? a. Woman-ERG man-ABS 3PL-see-3SG-IND 'The woman saw the men.' [yau [V] (dl x,: )Ag [N])Ag (dm x2: fanu [N]) Go ]

(dm x2: )

Go

, (dl Xi: fena?

The first thing that strikes one is the fact that the representations for what is most likely the same State of Affairs are fundamentally different. This would be relatively unimportant if the FG underlying representations (UR) were seen as a purely linguistic device, foreshadowing the eventual form. However, under the current view of URs as having language-independent, even cognitive status, to accept differences as those between (3) and (4) would either amount to assuming a rather Whorfian position or to giving up the universal status of underlying clauses to a large extent, making them rather language-type-dependent.

The agreement cross-reference continuum

209

A second objection to the above analysis is of a more typological nature. In our database on person agreement which currently contains data on 428 languages of the world we found the distribution for the respective types of Subject agreement as given in table 1 below. Table 1. Agreement types (n = 428) Type of Subject Agreement

Percentage of languages

No agreement Pro-drop 'Pure' agreement

31.5% 65.9% 2.6%

The figures in table 1 show that the vast majority of the languages which do have subject agreement are in fact pro-drop: 96.2%. This means that complex representations such as the one under (4) above are the rule while the more or less canonical representation under (3) is representative of just a very small number of languages with 'pure' agreement.4 In fact, 'pure' agreement seems to be restricted to a few languages in Western Europe, West Africa and New Guinea. There may be sentences for which multiple representation of some referent may be argued for. Constructions with left and right dislocation of constituents are a case in point. According to several authors, these are even a possible starting point for the development of agreement markers out of free pronominal forms (cf. Givon 1976; Ariel 2000). However, such constructions have a full term expressed outside the clause proper, typically in the Theme position, and a coreferential pronominal within the clause. In such constellations there may be differences in case form between the two elements. Over time, they may grammaticalize, which may result in the intonational break disappearing, both forms receiving the same case marking, etc. Appositions, on the other hand, are different in the sense that they consist of two nominal rather than pronominal expressions of the same referent, belonging to the same argument position, while the second element serves to further specify or identify the referent. This makes apposition as such a less likely candidate for the start of a diachronic scenario eventually leading to agreement marking in the first place.5 A third objection is that in many pro-drop languages - at least in 20% of the cases in our database - there is a condition on the presence of the agreement marker. In Hua, for example, an Indo-Pacific language from

210

Anna Siewierska and Dik Bakker

Papua New Guinea, objects are marked on the verb only when they are definite and human, as is shown in example (5) below. (5)

Hua(Haiman 1980:371) a. Vedemo p-go-e. men 2/3PL-see-lSG Ί saw the men.' b. P-go-e. 2/3PL-see-lSG Ί saw them.' c. Mna-vza-mo bird-COLL-PL Ί saw the birds.'

ko-e see-lSG

(*p-go-e). (2/3PL-see-lSG)

One of the consequences of the appositional approach would be that for such languages we would have two types of underlying representations, one with appositional copies for objects (for (5a)) and one without (for (5c)), depending on certain semantic features of the filler term. This does not seem to be very appealing.6 A special case of conditional agreement marking is found in about 2% of the languages in our database. For these, there is a marker either for the subject or the object, depending on which of the filler terms is higher on some hierarchy, typically animacy or person. So, in Camling, a SinoTibetan language from Nepal, speech act participants are always marked on the verb in transitives when the other argument is a third person, whatever the respective functions may be. The Inverse marker on the verb appears when the argument marked on the verb is not the first argument. When both arguments are non-speech act participants we find (ergative) third person marking. (6)

Camling (Ebert 2003:538; Ebert 1997:33) a. Phlod-uya. help-lSG Ί helped him.'

The agreement cross-reference continuum

211

b. Pa-phlod-uqa. INV-help-lSG 'He helped me.' c. Tyuko m-cheikuma that 3P0SS-sister 'Let him kill his sister.'

set-yi-nyo. kill-3 .PAT-OPT

In these cases, the mechanism which creates the extra argument position in URs should compare the relevant features of both the first and the second argument in order to pick the right one for the copy. In the case of Camling, and other so-called Inverse languages, this will be the one highest on the relevant person hierarchy. In our database there are 7 languages which go counter to the prediction, made by De Groot & Limburg (1986), that the copying of second argument positions always implies the copying of a first argument.7 The following examples are cases in point. In example (7) from Ani, a Khoisan language from Botswana, it is shown that there is a marker on the verb for the second argument, both when it is a full noun phrase (7a) and a pronoun (7b). And in example (8) from Nakanai, an Austronesian language from Papua New Guinea we have a marker on the verb for the recipient while the agent is not marked on the verb. Both languages are pro-drop. (7)

Ani (Heine 1999:33) a. Ti mün-m-te I see-3 SG.M-PRES Ί see a big lion.'

xdm-mä lion-3SG.M

jäü. big

b. Tö ti mün-tö-te. you.PL.COMG 1SG see-2PL.COMG-PRES Ί see you (men and women).' (8)

Nakanai (Johnston 1980:27) Ε Baba abi-a-gite la valua la NMRKBaba give-TRNS-3PL NMRK men NMRK 'Baba gave the men areca nuts.'

bua. areca.nut

In themselves these are just counterexamples to what could be an elegant generalization, and therefore do not affect the apposition proposal in a

212

Anna Siewierska and Dik Bakker

direct sense. But they point in the direction of a more problematic aspect of the proposal. Copying takes place on the basis of the underlying representation; more specifically, one or more argument positions are copied. However, in at least some languages which have a Subject in terms of FG, it is not so much the semantic function which is decisive of pro-drop but the assignment of the Subject function. Compare the following examples from Spanish. (9)

Spanish a. Ve-o a ti. see-lSG OBJM 2SG.ACC Ί see you.' [PRESver [V] (dl x,: [ + S , - A ] ) A g , S u b j e c t (dl x2: [-S,+A]) Go ] b. Eres visto por mi. Be.2SG seen by 1SG.ACC 'You are seen by me.' *Τύ eres visto por. 2SG be.2SG seen by [PRESver [V] (dl x,: [+S,-A])Ag (dl x2: [-S,+A])Go>Subject ]

In (9a) the Agent/Subject is dropped and the Goal argument is expressed. In (9b), with Subject assigned to the Goal argument, pro-drop affects the Goal rather than the Agent, as shown by the ungrammaticality of the second version. For these languages the proposal thus needs reformulation with the inclusion of the Subject as the factor which determines the creation of a copy. This has far-reaching implications for the process of Subject assignment, which could be argued not in fact to be part of the semantics of the clause but rather of its expression. Another complication for the appositional approach is introduced by languages in which the expression form of an argument depends on the semantics of the main predicate. For example in Hdi, a Chadic language from Northern Cameroon, some verbs take free pronominal objects while others take agreement suffixes, as shown in (10) below. In such cases, the argument-copying strategy must be made sensitive to the meaning of the verb, or at least to a lexical subcategory defining it.

The agreement cross-reference continuum

213

(10) Hdi (Frajzyngier & Shay 2000:111) a. Dgä-nä-dgä tä xen. divide-DEM-divide OBJM them 'He divided them.' b. Ne tä ζά-xen What IPFV eat-3PL 'What is eating them?'

(na)? Q

A further problematic point is how many copies should be made of each argument position. In many languages, arguments may be expressed a number of times, not just via a noun phrase and/or a cross-reference / agreement marker on the verb. A frequently occurring instance of this phenomenon is so-called clitic doubling, as in the examples from Spanish in (11) and Greek in (12). (11)

Spanish (Anagnostopoulos 2003:235) Α los alumnos les To DEF.PL pupils 3PL.DAT 'The pupils liked the book.'

gusta like

el libro. DEF book

(12) Greek (Anagnostopoulos 2003:207) I mitera tu to sinodhepse DEF mother 3SG.POSS 3SG.ACC accompanied pedhi. child.ACC 'Every child was accompanied by its mother.'

to kathe DEF every

Anderson (1993) devotes an article to obligatory multiple expression of morphological markers. He discusses examples from a variety of languages, e.g. Old Georgian, Winnebago, Fox, Crow and Tunica. The example in (13), from Ket, an isolate from Siberia, has a double expression on the verb of both relevant person markers. Note that the two markers for the third person agent are from different paradigms, called D and Β respectively. The markers for the second person are from the same paradigm, D.

214

Anna Siewierska and Dik Bakker

(13) Ket (Anderson 1993:10) Di-bu-k-ku-xos. 3D-3B-2D-2D-carry 'He carries you.' An extreme case of multiple agreement may be observed in Sandawe, a Khoisan language from Southern Tanzania, which expresses the subject on virtually every constituent except on the verb itself, as in example (14). (14)

Sandawe (Daglish 1979) Ci mancha-s liwaka koo-na-s hapu-me-s I food-lSG bring house-to-lSG you-for-lSG Ί am bringing food to the house for you.'

?ia. PROG

It is not directly clear to us what type of copying strategy should take care of this kind of agreement marking. Another remarkable instance of subject agreement is from Bartangi, an Iranian language spoken in Tadzhikistan and Afghanistan. In this language the subject marker may be affixed to any constituent, preferably the first one of the clause. This may in fact be the controller itself, as shown in example (15). (15) Bartangi (Payne 1980: 163) Äz-um tä-r kitob vuj. I-1SG you-to book bring.PRF Ί have brought you a book.' It seems rather extreme to use the notion 'agreement' here in its more or less direct sense, implying that the controller is its own target and agrees with itself. Interestingly, in one of the few 'real' agreement languages, Dutch, a similar phenomenon may be observed. Optionally, speakers may produce a demonstrative copy of the subject, provided that it is in PI position. These demonstratives may appear if there is a certain type of term in PI position which is pragmatically marked (contrastive focus; new topic). These demonstratives, which we will analyze here as a kind of focus marker, are typically unstressed, and there is no intonational break between the nominal and this pronominal version of the subject.8

The agreement cross-reference continuum (16) Dutch (colloquial) Mijn zuster die woon-t My sister DEM.REM live-3SG 'My sister lives in Duivendrecht.'

in in

215

Duivendrecht. Duivendrecht

A point directly related to the discussion in this section is the way FG deals with the underlying representation of person markers. For this purpose we will distinguish between four types: two types of free person forms, strong and weak, person clitics and person affixes. As will be discussed in more detail in the next section, these forms can be represented both on a formal, a functional and a semantic cline, which do not necessarily coincide in the same way for all languages. In standard FG, all three clines play a role in the way person forms are found in underlying clauses (cf. Dik 1997a: 152,157,186; Dik 1997b:215; De Groot & Limburg 1986:46f). So, forms that have some referential potential are represented in underlying clauses by a term consisting of the relevant term variable plus the anaphoric operator A, as in (17) below. (17)

(AxO

In cases where argument positions are expressed by pronouns, terms as in (17) are replaced by abstract predicates. These correspond to the meaning definitions of the personal pronouns in the lexicon. So, (18a) has the (partial) underlying representation given in (18b), and the lexicon contains an entry as in (18c). During expression, the pronominal form he will be retrieved. (18)

a. He works. b. work [V] (dlx, : [-S,-A,+M])Ag,Subj c. he [PRO], ... , MD = [-S,-A,+M]

In the case of pro-drop, argument slots which are expressed by way of bound verbal markers contain abstract representations based on precisely those semantic features which play a role in the verbal paradigm. They are very similar to the abstract predicates as shown in (18b), which reflects their referential, pronoun-like status. During expression they are not, how-

216

Anna Siewierska and Dik Bakker

ever, replaced by pronominal forms but they play a role in the selection of the right form from the verbal paradigm. Example (19) is from Hungarian. (19)

a.

Olvas-ok. read-lSG Ί read.'

b. olvas [V] (dlxi : [+S,-A]) Ag Finally, for nominal expression the argument position (or its appositional copy in the case of pro-drop) will be filled by the full term. In case of 'real' agreement as in English the expression rules will retrieve the relevant agreement features from this argument slot. This, too, is not an unproblematic scenario. Whether or not a language is pro-drop has a direct bearing on the way person forms are coded in underlying clauses. Compare the English example in (18) above with its Spanish translation in (20a) and the emphatic version in (20b) with the pronoun in grammatical apposition. (20)

a.

Trabaj-a. work-3SG 'He/she/it works.' trabaj- [V] (dlx, :

[-SrA])Ag,Subj

b. Trabaj-a el. work-3SG he-EMPH 'He works (not I).' trabaj-[V] (dlx,: [-S,-A]) Ag ,ContrFoc,Subj (dlx,: [-S,-A,+M]) Ag,ContrFoc,Subj

>

It seems that the different analyses for English and Spanish of what is arguably the same state of affairs is mainly motivated by differences in expression form, not by semantic or pragmatic distinctions. Note that, in the case of Spanish, the representations for the verbal marker and the pronoun in (20b) are virtually the same. They differ only to the extent that certain features are relevant for the two forms, i.e. masculinity. Note also that under these analyses there must be a fundamental difference between the two types of languages in the way the expression rules handle agreement phenomena. In the case of pro-drop languages such as Spanish the relevant

The agreement cross-reference continuum

217

features are readily available for expression in the corresponding argument positions of the UR. In 'real' agreement languages such as English, however, they have to be recovered by the expression rules from the relevant argument positions. This might seem a negligible difference from the perspective of a model of the grammar for these languages. However, if our goal is modelling what speakers do then the consequences seem to be rather far-reaching. A further point not very clearly stated in the theory so far is what determines the precise set of features for representations such as (17)-(20). Languages may differ internally as to the semantic features necessary for the expression of the respective person forms. This is clearly the case for verbal paradigms, where verb class or type and also 'external' factors such as mood, tense and aspect may play a role. For instance, in Southern Sierra Miwok, a language from California (Broadbent 1964:43), there exist four classes of person suffixes. Their use is dependent on the type of predicate they are affixed to and the presence or absence of case markers. A major difference is that classes 3 and 4 have dual forms, and distinguish between inclusive and exclusive while classes 1 and 2 do not. Furthermore, classes 2-4 have portmanteau forms for their Subject and Object markers while class 1 has independent suffixes for each of them. Since not all factors relevant for the determination of the precise set of features will be available in the underlying representation, the best that can be done under the current approach is always to have the maximum set, which will necessarily be an overspecification, in at least some of the cases. A last point is at what stage and under what conditions underlying term positions contain anaphoric operators as in (17). In cases like the above they are replaced by abstract predicates at some later stage, arguably before expression. In some representations, however, they are maintained well into the expression stage. This is the case for non-first order variables, as in the propositional term (AX,), which is directly handled by the expression rules. We are not aware of any criteria for this difference in the treatment of anaphora being mentioned in the literature.

3. Person marking: a unified multidimensional approach Before proposing an alternative FG analysis of person markers in section 4, we will first sketch the most important parameters determining the differ-

218

Anna Siewierska and Dik Bakker

ent ways in which languages give expression to terms in underlying representations. This sketch will then be used as a background for our proposal. A major function of all forms which code the notion of person in language is to refer to a term at the discourse level. Alternatively, such forms may be seen as some kind of expression of the term they refer to. As we mentioned already briefly in the introduction, it is our assumption that there is by definition a mental representation in the mind of the speaker of any term that plays a role in the ongoing discourse, whether it is new or old, before any utterance relating to it is produced. 9 It is to these mental constructs that reference is made by all person markers rather than to elements of the current or previous utterances. In that sense full noun phrases do not differ fundamentally from person forms, free or bound. The difference between the forms lies mainly in the amount of information they carry, and which supports their reference tracking capacities, inter alia. The only exception that probably should be made are in regard to terms with an inherent or explicit negation, such as 'nobody' or 'no person in their right mind'. According to MacDonald & Just (1989) such terms do not introduce referents into the discourse representation. As a result, they are only available in the working memory, and may be referred to only via devices that work at short distance. One consequence of this view is the blurring of the distinction between cross-reference and agreement. What remains is the optionality or obligatoriness of some types of expression form in some linguistic domains, and the way in which different forms can or must be combined. Another consequence is that in our opinion no principled distinction should be made in a grammar between anaphoric and deictic reference. Person forms do not refer directly to elements of the preceding text nor to entities in the speech situation, only indirectly, via mental representations e.g. terms. If languages do make a distinction between forms for deictic and anaphoric use, we will consider this to be a semantic rather than a pragmatic distinction. 10 Finally, there are no differences in the referential sense between specific and non-specific or generic terms, or for cases where there is no manifestation for some term in the 'real world', as in the familiar 'donkey' and 'unicorn' sentences of formal semantics. When languages make distinctions in form for such categories in person markers we will assume that these are meant to support the reference tracking function of the form vis ä vis the addressee, just as gender and person do, or to instruct the addressee that this is a new discourse entity and that a referent term should be created, as in the case of indefinite person forms.

The agreement cross-reference continuum

219

Against this general background let us now consider which are the parameters that determine the expression of terms. Not much is new about the parameters themselves that will be put forward here. They may be found, in isolation or combination in virtually all work on the synchrony and diachrony of person markers and related topics (see Siewierska 1999; to appear for references). What we want to stress here however is that together they may be seen as construing a multidimensional space. Within the limits of this space they determine, in cooperation and competition, the ways languages (or rather: speakers) give expression to the mental constructs of terms. Furthermore, we expect that there will be constraints on the values for and interactions between the different parameters which, in their turn, may form the basis for the establishment of typologies. Finally, the interaction between the different parameters and their value settings co-determine what variation we may find in a language, and how a language may eventually change in this respect. Within a model of the speaker three dimensions for the parametrization of person markers may be distinguished: formal, pragmatic and semantic. The formal dimension involves the way person markers are realized in a language, and is mainly about their morphosyntactic category and about the formal aspects of the syntactic domain in which they may or may not (co)occur. Although we see the categorial dimension fundamentally as a continuum, we will discuss it in terms of the major prototypes distinguished in (21) below. The left-to-right order displayed in (21) is indicative of their score on several other parameters, such as the syntactic domain in which they may appear and the type and number of semantic features coded in them. It also indicates the order of (possible) grammaticalization. (21)

noun phrase - demonstrative - free pronoun - weak pronoun - clitic - affix - zero

Within the noun phrase category, we may further distinguish between 'full' noun phrases and more or less fixed hyperonymic ones like 'that man', 'your servant' etc. The latter may constitute a step towards development into pronominal forms. By 'weak pronoun' we mean not just unstressed versions of independent forms but rather forms that differ from free forms both phonologically and in terms of their syntactic distribution. Finally, 'zero' stands for the absence of a marker, i.e. non-expression. 11 In a more elaborate version of the continuum in (21) a place might be allocated to pronominal forms which are not necessarily marked for per-

220

Anna Siewierska and Dik Bakker

son. Among these are reflexive pronouns. Another, relatively rare type of pronoun that might be considered is the so-called logophoric pronoun, as exemplified in (22) below from Mupun, a West-Chadic language from Nigeria. (22) Mupun (Frajzyngier 1993:108) a. Wa sat ne ta de dee She say COMP stop she.LOG stay 'Shej said that shej stopped over at Jos.' b. Wa sat ne wa ta She say COMP she stop 'She; said that shej stopped over at Jos.'

n-Jos. at-Jos

dee stay

n-Jos. at-Jos

In (22a) the logophoric third person feminine form de in the subordinate clause is coreferential with the subject of the main clause, i.e. with the person that originally produced the utterance. In (22b), with the 'regular' pronoun in the subordinate clause there is no such coreferentiality. We will treat reflexivity and logophoricity as semantic features of the corresponding pronominal forms. The syntactic domains relevant for person marking are discrete entities. Their precise properties are language-specific, at least to some extent. Syntactic domains form a hierarchy in the sense that an entity more to the right in (23) below is a subcategory of the one to the left of it. For this exercise we will distinguish the following domains. (23) discourse - paragraph - sentence - clause - noun phrase Noun phrases are the expression form of a term; they may be discontinuous. Clauses are the expression form of an Act as defined in Functional Discourse Grammar (FDG; Hengeveld 2004; for details see Kroon 1995:65f). Both phrase and clause are defined recursively. A sentence consists of one or more coordinated clauses and/or extra-clausal elements, such as Themes and Tails. It is the expression form of a Move as in FDG. A paragraph will be loosely defined here as a sequence of sentences which form the expression of a series of coherent Moves. Its relevance will be assumed for certain discourse phenomena such as Resumed Topic, but will not play a role in our discussion here. Finally, a discourse consists of one or more such paragraphs. Depending on the syntactic domain, and on

The agreement cross-reference continuum

221

pragmatic factors, grammars determine whether terms are expressed at all, and if so, in which form and whether there will be simple or multiple expression. The second dimension into which the space of person marking may be parameterized is the pragmatic one. The major contributor here is the amount of accessibility that can be ascribed to a discourse referent. Depending on the accessibility of a referent R in the discourse representation of the speaker, represented by Acc R S, and the speaker's assessment of the accessibility of the intended referent for the addressee, ( A C C R . A ) S , some form is selected by the speaker to give expression to R. The more accessible the speaker assumes R to be for the addressee the simpler its expression form may be. Value 0 for ( A C C R A )S will typically lead to the selection of a full, indefinite noun phrase; the maximum value for ( A C C R A ) S may give rise to zero expression. Our ( A C C R A ) S is equivalent to the type of hearer directed accessibility as defined in Ariel (1988), and, to a lesser extent, to Gundel et al.'s (1993) Givenness Hierarchy, and Prince's (1981) Familiarity Scale. Of course, the grammar may play a role here in that the selection of the right form may be grammaticalized to the extent that it depends on the domain in which the form will appear or to the extent that it is optional or obligatory. Apart from the more or less obvious cases, such as a completely new referent, the speaker's assessment of some A C C r a will typically be derived from the accessibility of R in his own discourse representation, ACC rs . Furthermore, a relatively low value for AccR.s may blur some of the less salient features of the referent. As a result of this, such features may be generalized to default values or remain non-expressed in the speaker's output. The phenomena of subject agreement in Arabic as discussed in Bakker (this volume) may be a good example of this second effect of Acc R .s in language production. Our Acc R S is more or less equivalent to the activation level of a referent in the speaker's working memory, as discussed by Kibrik (1999). From this author we adopt a second contributor to the pragmatic dimension of person marking, i.e. the filter (Kibrik 1999:49). This device operates in situations in which referential conflicts may arise because of the occurrence of two referents with more or less the same (estimated) accessibility. In such cases, speakers may revert to more explicit expression forms than the ones that would have been applied in cases without possible ambiguities. Languages may develop devices to find a way out of such conflicts in a systematic fashion; the availability of logophoric pronouns may be an example of this.

222

Anna Siewierska and Dik Bakker

The third dimension for the representation of person markers is the semantic one. The only subdimension relevant for all languages is person, by definition. 12 It has values [+S], [+A], and [-S,-A]. For the current article we will furthermore distinguish the subdimensions given in table 2 below, with their most frequently occurring values. The abbreviations used here are in conformity with the ones proposed in the EuroTyp Guidelines (Bakker et al. 1994). Given their general usage and transparency, we suggest that they replace the conventions employed so far for term representations in FG publications to the extent that they deviate from them. Table 2. Semantic dimensions of person marking Semantic Dimension

Values

Definiteness Demonstrative Gender (grammatical) Honorificity In/Exclusiveness Logophoricity Number Referentiality Reflexivity Specificity Switch Reference

DEF, INDEF PROX, DIST M, F, NT INFRM, FRM INC, EXC, AUGM LOG, NLOG SG, PL, DU, PAU ANA, DEIC REFL, NREFL SPEC, NSPEC, GNR SS, DS

Taking our discussion in section 2 of the current state of affairs of person marking in FG as a point of departure, we will now make an attempt at the implementation of the ideas of the current section in the framework of FDG. For the expression aspects we will revert to the dynamic expression rules. A brief overview of these may be found in Bakker (this volume).

4. The representation and expression of terms in FDG Our proposal to generalize over agreement and cross-reference in the case of person marking within the FDG model of the speaker leads us to the following stipulations:

The agreement cross-reference continuum

223

A. Terms are fully represented at the discourse level, with the relevant predicational restrictors and operators. They are accessible via their unique term variable, which is the actual referent. The traditional notions of antecedent and controller are unified in this discourse referent. B. Different terms may be linked to one term variable, provided that they are semantically compatible. C. Term representations in discourse may be underspecified for some operators, notably definiteness. The value for such operators is determined by the pragmatic component during expression. D. Argument positions in fully specified underlying representations are bound by term variables only. There are no predicates or operators present. E. Any instantiation of an argument-term variable pair is unique in an underlying clause. F. The pragmatic component determines the form and the number of appearances of a term in a clause, within the constraints set by the expression rules of the language. Each instantiation of a term in a linguistic expression unifies the traditional notions of anaphor and a target. For a first example of the implementation of this, let us look at the Dutch sentence in (16) above, repeated here as (24), which contains three different ways of expressing the same referent. (24)

Dutch (colloquial) Mijn zuster die woon-t My sister D E M . R E M live-3SG ' M y sister lives in Duivendrecht.'

in in

Duivendrecht. Duivendrecht

This sentence has the semantic representation given in (25a) below, under the assumption that we have the term representations under (25b) on the discourse level. On the basis of the foregoing discourse, Contrastive Focus has been assigned to X]; term x 2 is new. Since all terms have unique referents they are coded for definiteness from the outset.

224

Anna Siewierska and Dik Bakker

(25) a. [X, : [pres e,: [ woon [V] (χ,)ρ05 ] b. x0 X! x2

New

FncFtrs: FrmFtrs: SubCat:

( d e f s g x 2 : Duivendrecht

]]

ILLOC=decl, TENSE=pres, VAR.subject=x 1 , VAR.P1 VOICE=active PI, pre_v, v_fin, subject, y, predicate

Since Subject and Focus are assigned to the same term in (26), this constituent will be selected for expression in PI. The corresponding node for PI will be retrieved from the grammar. In its turn it will select and express the subject term in the form of a full noun phrase. In the process, the VAR.P1 feature will be set to the corresponding term variable, i.e. Xi. This value percolates upwards to set the corresponding feature of Node l. 1 5 The expression process now has reached the subcategorial position immediately in front of the final verb slot labelled pre_v, and will subsequently retrieve an uninstantiated node for that slot. It is prespecified for the required pragmatic and semantic properties of the constituent in PI, as indicated above when we discussed example (24). In its uninstantiated form it may look as in (27) below. Note that the values of some of the functional features have been prespecified here, in order for them to operate as constraints on the binding of the PI term variable to this node. In the case of the pragmatic function the specification consists of a list of alternative values. The values for gender and animacy are unspecified, and have to be retrieved from the lexicon entry of the nominal head of the term once it has been bound. (27)

Node k (uninstantiated) Slot: pre_v Cat: focus_marker Config: VAR.P1

The agreement cross-reference continuum FncFtrs:

227

PRAGF=(NewFoc, ContrFoc, ...), ANIM=yes, GENDER, NUMBER

FrmFtrs: SubCat: Phrasing all this a bit differently we could say that a preverbal node as under (27) is created to give extra expression to a pragmatically highlighted term. When the prespecified conditions indeed apply - as in the case of our example - we will find the node instantiated as in (27'). The assumption now is that the features relevant for this node are indeed accessible via the discourse variable. Since the full term has just been expressed, it may be supposed that information pertaining to it is optimally accessible in the discourse representation almost by definition. Given the nature of the referent, it may now also be decided that the subcategory of expression is a demonstrative pronoun rather than e.g. a locative adverb. (27') Node k (fully instantiated) Slot: P r e_v Cat: focus_marker Config:

VAR.Pl=Xi

FncFtrs:

PRAGF=ContrFoc, ANIM=yes, GENDER=f, NUMBER=sg

FrmFtrs: SubCat:

dempro

Thus, this node will get further expression via its only subcategory, dempro. That slot will be instantiated by a node that looks as in (28) below. (28)

Node k+1 (uninstantiated) Slot: dempro Cat: dem Config: [FORM [DEM], GENDER, NUMBER] FncFtrs: GENDER, NUMBER SubCat: FORM

This node will inherit the corresponding features from its mother node, k. The result will be as in (28').

228

Anna Siewierska and Dik Bakker

(28') Node k+1 (uninstantiated) Slot: dempro Cat: dem Config: [FORM [DEM], GENDER=f, NUMBER=sg] FncFtrs: GENDER=f, NUMBER=sg SubCat: FORM Since the configuration field contains the definition of a lexical entry, this node will induce the expression rules to retrieve a corresponding form from the lexicon.16 This will give us the following result. The subcategorization field now contains the phonological string which is the eventual expression form of this node. (28") Node k+1 (fully instantiated) Slot: dempro Cat: dem Config: ['die' [DEM], GENDER=f, NUMBER=sg] FncFtrs: GENDER=f, NUMBER=sg SubCat: 'die' Finally, we will look at the expression of the slot for the final verb form, v_fin. There will be a number of alternatives for this slot in the grammar of Dutch, including the insertion of the several types of auxiliaries which may be triggered by the respective ρ operators, the auxiliary in the case of a passive, and the copula in the case of a non-verbal main predicate. Since none of these apply here, the main verb will be selected as the default choice. This node may have the characteristics given in (29). Since we are dealing here with a regular predicate rather than a grammatical element, a distinction is made in the specification of the lexical entry between the lemma and the form that will eventually give expression to it, and which in Dutch depends on the features TENSE, PERSON and NUMBER. (29) Node η (uninstantiated) Slot: v_fin Cat: ν Config: [LEMMA [V], FORM, TENSE, PERSON, NUMBER] FncFtrs: TENSE, VAR.subject, PERSON, NUMBER SubCat: FORM

The agreement cross-reference continuum

229

First, the Config field and the inherited features will be instantiated, leading to (29')· (29') Node η (partially instantiated) Slot: v_fm Cat: ν Config: ['woon' [V], FORM, TENSE=pres, PERSON, NUMBER] FncFtrs: TENSE=pres, VAR.subject=x b PERSON, NUMBER SubCat: FORM Next, the values for the remaining features are inferred from the subject term via the term variable. Because of the recency of the activation of the subject term and its functional features for its (double) expression, it will be assumed that these features are readily available. 17 This will give us (29"). (29") Node η (partially instantiated) Slot: v_fin Cat: V Config: ['woon' [V], FORM, TENSE=pres, PERSON=3, NUMBER=sg] FncFtrs: TENSE=pres, VAR.subject=Xi, PERSON=3, NUMBER=sg SubCat: FORM The last step is the computation of the right verbal form. The first operation is a check for lexical priority, i.e. whether there is an irregular verb form for the constellation as specified in the configuration field. If this turns out to be the case, then this form will be retrieved from the lexicon and inserted for the FORM feature. If not, as in this case then a morphological process will start up which will produce the right form, leading to (29"'). 1 8

230

Anna Siewierska and Dik Bakker

( 2 9 " ' ) Node η (partially instantiated) Slot: v_fin Cat: ν Config: ['woon' [V], 'woont', TENSE=pres, PERSON=3, NUMBER=sg] FncFtrs: TENSE=pres, VAR.subject=xl, PERSON=3, NUMBER=sg SubCat: 'woont' Having sketched three different ways of expressing a term variable in one and the same grammar, we will now have a look at some other types of term expression. Our assumption is that the principles presented at the outset of this section may serve as the basis for any type of single, multiple or non-expression of a term in any language. Let us first look at the typical pro-drop situation. We will reconsider the case of Spanish, as exemplified in section 2 by (20a) and (20b) respectively, repeated here as (30) and (31) with what we think are the relevant underlying representations. (30)

a.

Trabaj-a work-3SG 'He/she/it works'

b. trabaj-[V] (XOAS c. Xi (31)

(def sg Xi : Pedro)

Topic

a. Trabaj-a έΐ work-3SG he-EMPH 'He works (not I)' b. trabaj- [V]

(Χ,)Α8

c. Xi

(def sg X] : Pedro)

(Topic) ContrFoc

The variable x b as in (30c) and (31c), is assumed to have been bound earlier on in the discourse by the term 'Pedro', which was possibly the form by which it has been expressed first. The (simplified) underlying clauses in b. are identical for both utterances, underlining the fact that there is no

The agreement cross-reference continuum

231

difference in the semantics between the two cases. As long as referent Xi is topical in the ongoing discourse and subject of the sentence it may be expressed on the verb only, as in (30a). However, when it bears contrastive focus function - which will make it no less topical - the expression rules will give it alternative expression in the form of a postverbal pronominal. There will be a postverbal slot in the Spanish sentence level template which optionally gets expression via a pronoun, provided that the right pragmatic conditions apply. The corresponding uninstantiated filler node is given in (32) with the value for the pragmatic function prespecified; after instantiation it will look like (32'). (32) Node ρ (uninstantiated) Slot: emph_pro Cat: perspro Config: [FORM [PERSPRO], PERSON, GENDER, NUMBER] FncFtrs: PRAGF=ContrFoc, SYNTF, VAR.subject PERSON, GENDER, NUMBER FrmFtrs: CASE SubCat: FORM (32') Node ρ (uninstantiated) Slot: emph_pro Cat: perspro Config: [el [PERSPRO], PERSON=[-S,-A], GENDER=m, NUMBER=sg] FncFtrs: PRAGF=ContrFoc, SYNTF=subject, VAR.subject=x1, PERSON=[-S,-A], GENDER=m, NUMBER=sg FrmFtrs: CASE=nom SubCat: Pro-drop utterances are also to be found in spoken variants of non pro-drop languages. Compare the following examples of elliptic answers. In the Dutch in (33) the object in the answer to the question may optionally be left out, and typically is. In the English answer in (34) both the subject and object are dropped; expressing the latter would be rather marked. (33) Q: Hoe laat is het? A: (Dat) weet ik niet.

('What time is it?') ('(That) know I not.')

232

Anna Siewierska and Dik Bakker

(34)

Q: Do you know what time it is? A: Dunno.

Pro-drop takes place routinely at the sentence level, when two clauses with the same subject are coordinated, as in (35). (35) Bob came on stage and immediately started playing 'All along the watchtower.' In these cases we assume that all argument positions are bound by the right variable - in a model of the speaker there can be no doubt about this. An empty first argument slot would give us an impersonal passive. As in Spanish, discourse-oriented motivations lead to different forms of expression, including non-expression. In non pro-drop languages there are simply more constraints on non-expression in terms of type of discourse pragmatics and type of syntactic domain, although classical pro-drop may not be ruled out altogether for these languages, especially in informal usage, as shown in (33) and (34). There are even examples of agreement drop despite the presence of a subject, as shown in (36). Instead of subject agreement markers the verbs occurs in the unmarked infinitival form. (36) En die kinderen maar zeur-en en and those children PRT nag-INFIN and 'And those children kept on nagging all the time.'

zeur-en. nag-INFIN

It could be claimed that non pro-drop is a final stage in the grammaticalization process of person markers, where the reference tracking potential of the verbal markers in erstwhile pro-drop languages was assumed by speakers to be so scanty that they started expressing the referent via free pronominal forms more often than not, eventually leading to their becoming obligatory.19 In that sense it might be better to speak of 'pro-support' than of pro-drop. The phenomena of (33)-(36) are all the more reason to assume the same underlying representation at the semantic level for all languages, independent of their expression parameters and their pro-drop character. Another phenomenon that should get some attention in this context is that of switch-reference systems. They come in two types. In type I, exemplified in (37) by the Papuan language Ono, all verbs in a sentence, or a sequence of sentences with the same subject have zero expression but for the last in the series, which has a person suffix, and possibly other markers,

The agreement cross-reference continuum

233

e.g. for tense or aspect. This marker announces the switch to a different subject. Type Π, exemplified by Kobon in (38), also from New Guinea, has the same full expression for the last verb in the row, but it has a separate marker for the same subject cases, which is typically a reduced form with fewer semantic distinctions than the different subject one.20 (37)

Ono(Haiman 1983:108) Ngauk ne-0 ari-mai-ke. Tobacco smoke go-PROG-3SG 'He had a smoke and went.'

(38) Kobon (Comrie 1983:29) Hoi bi kaj pak-ul ram We.two man pig strike-SS.lDU house 'We two killed a pig and took it home.'

ud take

ar-bul. go-DS.lDU

We propose that in all these cases there is a term variable in the subject argument position, the same one for the SS and the DS cases. The pragmatic component will be planning ahead a sequence of discourse acts which will be semantically coded in a sequence of propositions with the same topical referent in an argument position, typically that of the Agent or another first argument function. The whole sequence will be expressed as one complex sentence. The pragmatic component will keep track of the respective steps, inserting the feature SS for all non-final clauses and DS for the last one. The same control mechanism will take care of the prosodic pattern. A special problem is introduced by serial verbs such as ud in the Kobon examples in (38) above and in (39) below. (39) Kobon (Davies 1981: 203) Nipe wanib si He string.bag illicitly 'He stole the string bag.'

ud take

ar-öp. go-3SG.PF

The infinite verb form ud 'take' combines semantically with the following finite verb form ar 'go' to code one state of affairs. This is rendered in the translation by the presence of just one clause with one verbal main predicate. Although typical serial verb constructions have grammaticalized to some extent, and exhibit quite a few morphological and syntactic con-

234

Anna Siewierska and Dik Bakker

straints in comparison to fully inflected verbs, in the case of Kobon there seems to be enough reason to assume that the respective verbs have all their argument positions bound by the variables for the intended referents. For the example in (40) it seems to be obvious that ne 'you' serves as a filler for the first argument of both verbs, while gasi 'thought' relates to the second argument of ηδή 'perceive', the non-finite verb. (40) Kobon (Davies 1981: 204) Ne gasi ποη aij You thought perceive good 'You must think carefully.'

gi-mön. do-2SG.DEONT

Both verbs, however, code one state of affairs. We will formalize this by letting them form one predication, as in (41) below. (41) a. [DEONT X, : [PRES e, : [c, : ηοη [V] : aij [ADV] (Χ,)Ρ08ΕΧΡ (X2>Go] : [ c 2 : g l [ V ] ( x i ) p 0 s E x P ( C I ) G O ] ] ]

b. xi x2

ascriptive [locative > adjectival > nominal > possessive] (Hengeveld 1992)

The explanatory power of typological hierarchies

253

These hierarchies read as follows: if in a language (some type of) ascriptive predication is predicable, then equative predication is predicable as well. Furthermore, if one of the sub-types of ascriptive predication is predicable in a certain language, then all sub-types of acriptive predication preceding it in the sub-hierarchy are also predicable in that language (and so are, by the previous implication, equative predications). In other words, for every element in the schema the following prediction can be made: If a type of non-verbal predication at a certain point in a (sub)hierarchy is predicable in a certain language, then all predication types preceding it in that (sub)hierarchy are also predicable in that language. Having thus established the hierarchical system of (part of) the domain of non-verbal predication, the next question is: when a certain type of nonverbal predication is predicable in a language, then what grammatical means are employed to express it? Two main groups of such expression formats can be distinguished: those without a copula and those with one. The first group, lacking a copula, consists in turn of two formats, termed zero-1 and zero-2 (Hengeveld 1992). In the case of a zero-1 expression format, the non-verbal predicate shows the same morphosyntactic behaviour (marking for tense, mood, aspect and person) as does an intransitive verbal predicate in a particular language. By contrast, in the case of a zero-2 format there is no such marking; the non-verbal predicate and the argument are simply juxtaposed. The second group of expression formats, those with a copula, can be subdivided into constructions with a predicativizing copula on the one hand and those with a discriminating copula on the other. Predicativizing copulas can carry the morphosyntactic distinctions characteristic of main predicates in a language. Typically, this type of copula is a verb. By contrast, discriminating copulas cannot carry 'main predicate-like' morphosyntactic distinctions. Their function may therefore be defined as the mere signalling of non-verbal predication. Elements used as discriminating copulas are typically pronouns and particles. The distribution of the different expression formats across types of nonverbal predication types can be captured in two implicational hierarchies. These are presented in (8) and (9): (8)

Zero-1 hierarchy Bare>Referential>Relational (Hengeveld 1992:199)

254

Eva Η. van Lier

(9)

Zero-2 hierarchy Equative >Ascriptive (Hengeveld 1992:201)

Notably, the first hierarchy (8) shows that the type of predicate involved predicts the distribution of the zero-1 strategy. This hierarchy must therefore be read as follows: if a language uses a zero-1 strategy in constructions with a predicate type on a certain point in the hierarchy, then that language uses a zero-1 strategy for constructions with al preceding predicate types in the hierarchy. The form of the zero-1 hierarchy is not surprising, if one compares the underlying representation in Functional Grammar of the various predicate types in the hierarchy with the underlying representation of a verbal predicate, as in (10) below. Bare predicates have the same underlying structure as verbal predicates and therefore they behave more easily like a verb than the other predicate types, which show an increasingly deviant structure: (10)

Verbal: Bare (non-verbal): Referential: Relational: (Hengeveld 1992: 200)

(f;: pred v (£)) (ή: predA/N ( i ) )

(fj: (Xi: i : predN (£)) ( Xi ) 0 ) (ζ)) (f): (Xj: i : predN (£)) (Xi)Sem) (fj))

Interestingly, the hierarchy (9) shows that the distribution of the zero-2 strategy is identical to (part of) the hierarchical system of predicability in (7). In the present context this hierarchy predicts that if ascriptive predication employs a zero-2 strategy (i.e. does not require predicativisation by means of a copula), then the predication type preceding it in the hierarchy, i.e. equative predication, employs it as well. This means that more easily predicable prediaction types are more likely to be expressed by means of a zero-2 strategy. In many cases, however, two different formats are used in a language for the expression of a single type of non-verbal predication; one containing a copula and one lacking it. In these cases, the presence of a copula is triggered by the need to express tense, mood, aspect and person. This means that in what may be called the default (or most unmarked) situation the copula is not present. Such a default situation could be the present tense, the third person singular or any situation when no explicit expression of time reference, aspectual meaning or person is required.

The explanatory power of typological hierarchies

255

To summarise, this section has provided a general overview of the typology of non-verbal predication, based on the notion of copula support as put forward in the theory of Functional Grammar. It discussed the hierarchical predicability of the different semantic types of non-verbal predication and the hierarchical distribution of their expression formats, both with and without a copula. On the basis of this typological system in the next section the hypotheses will be presented concerning the predicability and expression of non-verbal predication in developmental contexts involving variation, change, contact, and acquisition.

3.

Hypotheses

The general hypothesis of this study is that data obtained in studies on language development can be explained in terms of the implicational hierarchies presented in section 2. This hypothesis follows from the assumption that implicational hierarchies represent linguistic universale, which are applicable to all types of linguistic systems. Particularly, both intra-linguistic variation and internally motivated diachronic change concerning the system of non-verbal predication are expected to follow the typological hierarchies. As such variation should be interpretable as a synchronic attestation of change in progress. Contactinduced change is hypothesised to proceed in accordance with the same universals, unless exceptional speaker attitudes play a role (cf. Thomason 2001). Similarly, the order of acquisition of non-verbal predication is expected to correspond to the order of adjacent hierarchical levels. In the case of second language acquisition, the learner's native language may possibly interfere with the relative speed of acquisition, but not with the universal order of the process. These hypotheses are tested in the next section.

4. Developmental applications 4.1. Synchronic variation: change in progress One of the most extensively studied phenomena in the field of synchronic variation is copula deletion in African-American Vernacular English (AAVE). Rickford et al. (1991) provide a critical overview of the methods

256

Eva Η. van Lier

used and the data obtained by the various authors that have engaged in this research. Among the most important factors that turned out to influence the patterns of copula deletion were tense, predicate type (linguistic factors) and age (social factor). Regarding the first factor, according to Rickford et al. (1991: 105) virtually all researchers agree that past tense forms of the copula are almost invariably present. This is in line with Hengeveld's finding that if a language employs two types of expression formats for the same type of non-verbal predication, the zero-2 strategy is used mostly in "default" situations where no supportive device is necessary to carry morphosyntactic distinctions (NON-PAST, in this case). The influence of the second factor, predicate type, was assessed for predications formed with locatives, adjectives and nouns. Various researchers studied the deletion of either is or are or both forms for social groups differing (among other factors) in geographic location and age. The outcomes of these studies are assembled in table 1:4 Table 1. Copula absence rates by predicate type in various AAVE studies study

copula form Locative Adjective

Noun

Labov 1969 Labov 1969 Baugh 1979 Wolfram 1969 Wolfram 1969 Mitchell-Kernan '71 Baugh 1979 Baugh 1979 Bailey & Maynor '87 Bailey & Maynor '87

Is Is Is is + are is + are Is Is Are is + are is + are

.23 .32 .14 .37 .02 .09

.36 .52 .31 .44 .13 .14 .29 .69 .19 .15

.48

.36 .72 .47

.04 .20 .56

.35 .25

.14

.32

.25 .12 .09

What table 1 shows is that, with one exception (the bold printed .32 in the noun-column), the deletion rates are consistently higher for locative predicates than for nominal ones. Adjectival predicates sometimes trigger deletion rates that are higher than for locatives (those are printed in bold), and sometimes lower. In all cases, however, the rates are higher for adjectives than for nouns. Unfortunately, Rickford et al. do not explicitly differentiate between predications based on bare nominals on the one hand (ascriptive, status-

The explanatory power of typological hierarchies

257

assigning predications) and those based on referential predicates on the other (equative, classifying predications). In AAVE (as in Standard English), the use of a bare nominal without an article is ungrammatical (*John is teacher.) Therefore, it is not possible to infer from the examples whether the one or the other or both categories of predications based on a noun are taken into account. However, assuming for the moment that the noun category in table 1 includes ascriptive predications only, the pattern of copula deletion in AAVE shows a considerable degree of similarity with the sub-hierarchy of predicability (ascriptive predication): locatives are more easily predicable then nominals, and so are adjectives. The relative position of adjectives in the deletion pattern is less clearly comparable: sometimes they are in between locatives and nominals, as they are in the predicability hierarchy of ascriptive predication, but sometimes they are not. Moreover, the distinction between ascriptive and equative predications based on a noun becomes irrelevant if the copula deletion pattern in AAVE is explained in terms of the time-stability scales suggested by Givon (1984) and Stassen (1992, see also 1997). These scales order predicates as to the degree of changeability over time of the properties or relations they denote. Verbs and adverbial expressions (in this case locatives, such as in the kitchen) generally denote properties or relations that can easily change over time. Nouns, by contrast, usually denote time-stable properties (such as a man). Adjectives occupy an in-between position; some of them denote relatively time-unstable properties (such as tired or angry), whereas others refer to more time-stable properties (such as intelligent), like those denoted by nouns.5 A plausible explanation for Rickford et al.'s data would therefore be the following: of the predicate types studied, locatives designate the least stable properties. As such they are closest to verbs and function most easily as predicativizers. By contrast, nouns designate the most stable properties; they are in that respect the least 'verby'. The adjective category occupies a less clear position in the sense that it does show a more 'verby' behaviour than do nouns, but not always a less 'verby' behaviour than locatives. This means that AAVE speakers appear more inclined to delete the verbal copula be in locative predications, where a 'verby' predicate is already available. As the predicate becomes less 'verby', i.e. when it is a noun and in some cases when it is an adjective, the deletion of be decreases, presumably because be is more needed to fulfil the function of predicativizer.6

258

Eva Η. van Lier

This explanation is compatible to what Pustet calls "a cline of predicate-worthiness of lexeme types" (Pustet 2003: 187). As she notes, verbs tend to function as predicates, combining with grammatical categories that denote person, tense, aspect, modality and voice. Nouns, on the other hand, tend to function as arguments, combining with grammatical categories that denote case, number, (in)definiteness and gender or noun class. In between verbs and nouns are adjectives, which tend to function as attributes and do not show particular categorical preferences (Pustet 2003: 11). According to the cline of predicate worthiness, a lexeme with a high frequency of occurrence in predicate position (i.e. a verb) is less likely to be accompanied by a copula than a lexeme which displays a low frequency of occurrence in predicate position (i.e. a noun or an adjective). Although the position of locatives on the cline of predicate worthiness would appear to be between verbs and adjectives, according to their time-stability semantics, they are not included in Pustet's analysis. The relevance of the comparison between the AAVE copula deletion pattern on the one hand and both the sub-hierarchy of predicability (ascriptive predication) and the time-stability or predicate-worthiness scale on the other becomes even more meaningful if we consider the third factor influencing deletion, namely the age of the AAVE speaker. This factor had a clear-cut effect on the deletion pattern, at least within the population of East Palo Alto (California), where Rickford et al. carried out their research. Table 2 shows that young people delete copulas much more often than do older persons, whereas the middle-aged group occupies an in-between position (Rickford et al. 117, 119): Table 2.

Copula deletion in East Palo Alto by age group

Age group

Deletion rate

Young

.83

Middle

.42

Old

.22

This significant effect for age makes Rickford et al. wonder whether they are facing stable age-grading or language change in progress. Although the authors are not conclusive as to the answer to this question, I would argue that a diachronic process is indeed involved here. That is, as will become

The explanatory power of typological hierarchies

259

clear in the next section, the sub-hierarchy of ascriptive predication is relevant as well for the diachronic development of non-verbal predication systems in many languages. The driving force behind this development is sought in the semantics of time-stability, as discussed above.

4.2. Language change 4.2.1. Internally motivated linguistic change Through the course of time, changes can occur in the non-verbal predication system of a certain language. Such a change can either mean the development of a copula in constructions where the language formerly lacked one, or it can mean the development of an alternative copula in constructions where one was already available. Hengeveld (1992) has identified two pathways along which these developments take place, crosslinguistically. These are schematically represented in (11) and (12) below: (11)

Diachronic pathway 1 Verb Locative (Hengeveld 1992:201)

Adjective

(12) Diachronic pathway 2 Pronoun -> identifying -> (Hengeveld 1992:251)

Noun

classifying

Possessive

Adjective/Noun

The first pathway involves a postural verb that comes to be used as a copula in locative predications. From there, the use of such a locative copula can be further extended along the domain of non-verbal predication: first to predications based on adjectives, and later to ones based on a noun or a possessive element. Notably, from the locative construction onwards, this pathway corresponds to the sub-hierarchy of ascriptive non-presentative predications. Hengeveld (1992: 136-139) explains the ordering of the steps of pathway 1 in terms of the time-stability scale discussed in the previous subsection. As he notes, it is not immediately evident that possessive phrases express properties that are more time-stable than those expressed by nouns. Firstly, however, nouns often denote non-inherent properties (such as a teacher) and, secondly, in many languages the predicative use of posses-

260

Eva Η. van Lier

sive phrases is restricted to the expression of permanent possession. Whether or not these arguments make the position of possessives on the pathway acceptable is not of primary interest now, since they hardly play a role in the data to be discussed. The implementation of diachronic pathway 1 may be illustrated with the entry of the Latin positional verb stare, 'to stand' into the systems of nonverbal predication of the Ibero-Romance languages (Hengeveld 1991: 8790). The moment when stare became used as a copula differs for the individual languages, resulting in a pattern that shows corresponding differences in the advancement of the new copula along the various steps of the pathway. This is represented in table 3: Table 3. Use of stare as a copula in the Ibero-Romance languages language

locative

Judeo-Spanish Catalan Spanish Galician Portuguese

+

-

+

+

+ +

+

+

+

adjective

noun

possessive

+ +

The second diachronic pathway starts off with an anaphoric pronoun in topic-comment constructions. The intonation break between the theme and the topic clause gradually disappears and the sentence is reanalysed as a subject-predicate construction, with the pronoun as its copula. Taking the resulting equative identifying predication as its starting point, the use of the new copula can extend towards classifying equative predication and further to ascriptive predications based on an adjective or a noun. As was the case with pathway 1, the second pathway can be related to the hierarchical system of predicability: after the reanalysis as a copula, the development follows the hierarchical system from the equative to the ascriptive level. Diachronic pathway 2 was first reported by Li & Thompson (1977). They document the pronominal origin of copulas in various languages such as Palestinian Arabic, Hebrew, Wappo, Zway and Manderin Chinese. The copula shi in the latter language functioned as a demonstrative pronoun in Archaic Chinese literature, as is shown in (13):

The explanatory power of typological hierarchies

261

(13) Chinese (Li & Thompson 1977: 423) zi yü shi ri ku. Confucius at this day cry 'Confucius cried on this day.' However, shi also occurred as an anaphoric demonstrative pronoun in equative identificational sentences with a topic-comment construction, as in (14): (14) quiong yu jian, shi ren zhi süo wit ye poverty and debasement, this people GEN NR dislike DECL 'Poverty and debasement, that is what people dislike' If the intonation break (represented by the comma) that is typical of this kind of utterance becomes weakened, the topic-comment construction is reanalysed as a subject predicate construction, in which shi is the copula. In (15) the extension of the copular function of shi in equative identificational predications is exemplified: (15) Chinese (Li & Thompson 1977: 426) Υύ shi suö jiä fü-ren zhi fü ye 1SG COP NR marry woman GEN father DECL Ί am the father of the married woman' The relevance of the two diachronic pathways discussed in this section and of the hierarchies presented in section 2 is further tested with regard to language contact in the next subsection.

4.2.2. Contact-induced language change De Groot (2005) observes various differences between Hungarian as it is spoken in Hungary and varieties of Hungarian spoken outside Hungary. These changes in the varieties outside Hungary are assumed to be the result of contact with adjacent languages. One of the differences has to do with the copular system. In non-contact Hungarian some of the predicable types of non-verbal predication are expressed by means of a zero-1 strategy and some with the copula van:

262

Eva Η. van Lier

(16) Hungarian (De Groot 2005) a. Piter diäk-0. Peter student-3SG 'Peter is a student.' b. Mari okos-0. Mary clever-3SG 'Mary is clever.' c. Jänos α tanär-0. John DET teacher-3SG 'John is the teacher.' d. A macska a kert-ben DET cat DET garden-INES 'The cat is in the garden.' e. Tamäs-nak könyv-e Tom-DAT book-3SG.POSS 'Tom has a book.'

van- 0. COP-3SG

ναη-0. COP-3SG

As these examples show, non-contact Hungarian follows the zero-1 hierarchy: predications based on bare and referential predicates (16a-c) are expressed without a copula, while morphosyntactic distinctions are marked on the non-verbal predicate, whereas predications with a relational predicate (16d and e) use the verbal copula van. According to De Groot, this situation also holds in contact Hungarian, at least in main predications. Now, Hengeveld (1992) claims the validity of the zero-1 hierarchy for main predications only. However, De Groot argues that (at least in the case of Hungarian) it holds also for embedded predications. In non-contact Hungarian the situation is the same in embedded as in non-embedded predications: bare and referential predicates do not have a copula, whereas relational predicates do. This is exemplified in (17):7 (17)

a. Janos, aki a tanar... John REL DET teacher 'John, who is the teacher...'

The explanatory power of typological hierarchies

263

b. Peter, aki diäk... Peter REL student 'Peter, who is a student....' c. az okos läny. DET clever girl 'the clever girl.' 8 d. a kert-ben levö macska. DET garden-INESS COP.PRES.PRT cat 'The cat in the garden.' (lit. 'the in the garden being cat.') However, in the contact varieties outside Hungary the expression of nonverbal predication in embedded clauses differs in the following way: contact Hungarian does not obligatorily use a copula in embedded predications with a relational predicate: (18) a kip a fal-on. DET picture DET wall-SUPER.ESS 'the picture on the wall.' In non-contact Hungarian this construction would look as in (19): (19) a fal-on levö kep. DET wall-SUPER.ESS COP.PRES.PRT picture 'the picture on the wall.' (lit. 'the on the wall being picture.') This difference between non-contact and contact Hungarian can be described as a shift of the cut-off point in the hierarchy from between referential and relational predicates to after relational predicates, in accordance with the zero-1 hierarchy. This change does not appear to be related in a clear way to either of the two diachronic pathways, but does constitute evidence that a synchronic hierarchy for expression strategies, the zero-1 hierarchy, can also account for (at least this instance of) contact-induced language change. De Groot makes no reference to the role played in this shift by the specific typological features (regarding non-verbal predication) of the languages with which Hungarian is in contact. These are Slavonic languages, Rumanian and German (De Groot, personal communication). In fact, at

264

Eva Η. van Lier

least some of these languages appearently have the possibility of leaving out the copula in embedded non-verbal predications with a relational predicate, as is illustrated below for Rumanian, German and the Slavonic language Slovak: (20) Rumanian (Mallison 1986:5) Am vazut sobolanul din PERF.AUX. 1SG. see.PTC.PERF. rat.DET REL.in Ί saw the rat in the kitchen.' (21)

(22)

German Der Mann im Zimmer had DET man in.DET room had 'The man in the room had not spoken.'

bucatarie. kitchen

nicht gesprochen. not spoken

Slovak (Short 1993:567) vel 'ky dom na konci ulice. big house at end.LOC street.GEN 'The big house at the end of the street.'

A different kind of contact situation has been studied by Arends (1986, 1989): he discusses the historical development of the copula system in the Creole language Sranan. This language has two copulas: de and da. The first copula, de, has a secondary function as a preverbal durative aspect marker with non-stative (or non-time stable) verbs: (23)

Sranan (Smith, p.c.) a. mi sabi. Ί know.' [+stat] b. mi naki en. Ί beat him.'[-stat] c. mi de naki en. Ί beat him.' [+stat] Ί was (continually) beating him.'

Thus, although de is not a full verb, because of its association with nonstative verbs (cf. the time-stability scale) it is semantically suitable to be-

The explanatory power of typological hierarchies

265

come used as a copula, following diachronic pathway 1. Indeed, this is what Arends claims to have happened; first de became used as a copula in locative predications and later it advanced to adjectival predications.9 The second copula, da, also has a secondary function, namely as a determiner and in older Sranan as a demonstrative pronoun. In the latter function, da was used in topic-comment constructions, as in (24). (24)

Saramaccan (Arends 1986:107 [Schumann 1783:46]) 'adjabre', da Djutongo. adjabre that Jew.language "Adjebre', that is Saramaccan.'

In accordance with diachronic pathway 2, the pronoun was reanalysed as a copula and became used as such in equative predications of both the identifying and the classifying type, as is shown in examples (23a, b), respectively. Also, according to Arends da became used in sentences like "That man is a teacher." 10 (25)

Saramaccan (Arends 1989:155 & 1986:107) a. Joe da mi lobbi pikien (...) You COP my love child 'You are my beloved child.' b. Da somma da wan That person COP a 'That is a good person.'

boen good

somma. person

Furthermore, Arends shows that between 1800 and 1850 a shift occurs in this system: ascriptive predications based on nouns, in which Sranan formerly used da, develop a predilection for de, and this change is later consolidated. To summarise, the development of the copular system in Sranan manifests both diachronic developments mentioned in section 4.2.1. The aspect marker with non-stative verbs, de, has moved along pathway 1 to location and to adjectives, whereas the demonstrative pronoun (and later determiner) da has moved (nearly) until the end of pathway 2 (it is used with nouns, but not with adjectives). It seems then, that Sranan's two copulas have come from both extremes of the system and have met in the middle.11 Around this meeting point, instability arose. Before 1800 ascriptive predi-

266

Eva Η. van Lier

cations with a nominal predicate made use of da, whereas ascriptive predications with non-nominal predicates used de. However, after 1850 in all ascriptive predications de comes to be used as the copula, resulting in a neat formal opposition between equative {da) and ascriptive predications (de). These developments are schematically represented in table 4: Table 4.

Development of the copula in Sranan

Predication type

Predicate type

Historical development —> aspect marker with non-statives de

Ascriptive

Locative Adjective Ν (bare)

0 0 0

de 0 0

de de da

de de de

Equa-tive

Ν (classifying) Ν (identifying)

0 0

da da

da da

da da

(dem, pronoun +) determiner da

Regarding the influence on this development of Sranan's substrate languages, Arends notes that the systems of non-verbal predication of Kwa languages also seem to be crucially determined by relative degrees of timestability (Arends 1986: 114-116). However, the system in Sranan does not seem to have started out from such Kwa-like systems, nor does it currently manifest the same oppositions. It seems more plausible that the situation in which Sranan arose gave rise to simplification (a zero-2 strategy in all predicable predication types) and that subsequently the copula system developed independently from the substrate languages, following universal diachronic pathways.12 McWhorter (1992) studied the development of the copular system of Modern Swahili. In Early Modern Swahili, a verbal copula existed, which was -li. According to McWhorter, this copula was used in locative predications.13 In equative and ascriptive predications based on a bare predicate (adjective or noun), 14 -li was used in non-present tense, whereas in the present tense a zero-2 strategy was used. Furthermore, in identifying equative predications a resumptive pronoun was used, which became reanalysed as a copula when -li was not present:

The explanatory power of typological hierarchies

267

(26) Modern Swahili (McWhorter 1992) Wewe u mkurugenzi. 2SG 2SG/COP director 'You are the director.' Finally, a focus particle ni existed in Early Modern Swahili, which was also reanalysed as a copula, when used in equative sentences with a zero-2 expression format: (27) Rahayi ni shida na taabu. handmill FOC/COP distress and trouble 'This treadmill is distress and trouble.' In the course of the twentieth century, the expression of non-verbal predication in Swahili has undergone a number of changes. The most drastic of these was the disappearance of -li as a copula, due to its reanalysis as a past tense affix. This was the result of the usage of -li before full verbs to express past continuous or imperfective aspect: (28) A -li meketi nyumba-ni. 3SG/IMP -be sit house-LOC 'He was sitting in the house.' The double function of a- in such constructions as in (27), had the effect that -li began to be processed as the marker of imperfective aspect, rather than a-, a reanalysis that was made even more likely by the fact that Swahili also had (and has) a present tense marked by a-. This confusing situation finally led to the complete drop-out of -li as a verbal copula. In nonpresent locative predications -li was replaced by the verb -wa, originally meaning 'to become'. In present tense locative predications no verb was needed to replace -li, and this type of construction was marked only by an already existing locative suffix -ko. In equative predications, the former focus particle ni (in the present tense alternating with a zero-2 strategy) came to be used as a copula, at the expense of the copula that originally was a resumptive pronoun. Also, ni extended its copular function to ascriptive predications. Finally, according to data reported by Myers-Scotton (1985), in non-standard Swahili the locative suffix -ko is also used in predications based on an adjective denoting a temporary or acquired property, as in (29):

268

Eva Η. van Lier

(29)

Swahili (Myers-Scotton 1985:290) U -ko mgonjwa? 2SG -LOC sick? 'Are you sick?'

The developments in the copular system in Swahili are schematically presented in table 5. Table 5. Development of the copula system in Swahili( r.p. = resumptive pronoun, various forms) Predica- Predicate tion type type

Ascript.

Equat.

Locative Adjective Ν (bare) Ν (class.) Ν (ident.)

Historical development —» pres -pres pres verb -wa ('to become") -li -li -li 02 -li 02 02 -li 02 02 -li 02/ni 02 -li 02/ni/rp rp/foc.part. ni Φ

-pres pres -li -li -li -li -li

-pres pres

-ko -wa 02/ni ni 02/ni ni 02/ni ni 02/ni ni

-ko -ko 02/ni 02/ni 02/ni

-pres -wa ni ni ni ni

Both McWhorter and Myers-Scotton claim that the process outlined above is one of simplification. To account for this simplification, they refer to the fact that, due to its status as a lingua franca, for many speakers Swahili is a second language. Indeed, it seems that the reanalysis o f - / / and the marginalisation of the resumptive pronoun copula are the results of resolving ambiguous situations in Swahili and can as such be characterised as simplifications. However, the reanalysis of the focus particle ni as a copula and its extension through the system of non-verbal predication is clearly an implementation of diachronic pathway 2. Furthermore, the later extension of -ko from locatives to time-unstable adjectives is an example of pathway 1. It seems then, as was the case with Sranan, that the development of Swahili's copula system is a combination of simplification resulting from a contact situation on the one hand and universal diachronic processes on the other hand.

The explanatory power of typological hierarchies

269

4.3. Language development in individuals 4.3.1. First language acquisition The acquisition by children of non-verbal predication and its expression in their first language has not been studied extensively. The only larger-scale research, to my knowledge, is Becker's dissertation on the acquisition of copulas in child English (Becker 1999, 2000). She examined patterns of expression versus non-expression of the copula be in different types of predication, namely those based on a noun (example 30),15 on a locative expression (31) and on an adjective (32). Within the latter type of predication she distinguishes those formed with time-stable adjectives (32a) on the one hand and those with non-time stable adjectives (32b) on the other: (30) John is a man. (31)

The book is on the table.

(32) a. Sally is tall. b. Bill is sick. (Becker 2000: 81) Becker reports the average rate of expression of the verbal copula be (as opposed to a zero-2 format) in these types of predications four children, from the age of about two years onwards, during periods lasting two to seven months. Her findings are summarised in table 6, which shows that the expression rates of all children increase from locative to adjectival and then to nominal predications. Furthermore, within the category of adjectival predications, all expression rates are lower when a non-time-stable adjective is used then when a time-stable adjective is used. These findings correspond directly to the sub-hierarchy of ascriptive predications and diachronic pathway 1, and as such to the time-stability scale. Notably, Becker's data are very similar to the AAVE copula deletion pattern discussed in section 4.1: where the deletion rates in table 1 decreased, the expression rates in table 6 increase. An advantage of Becker's study appears to be her distinction between time-stable and time-unstable adjectives: whereas the status of the adjective category in the AAVE-study was not very clear-cut, table 6 shows that the adjectives consistently occupy their position in between the locatives and the nouns, with the time-

270

Eva Η. van Lier

unstable ones behaving more like the locatives and the time-stable ones more like the nouns. Possibly, a similarly clear pattern would have emerged in the AAVE-case if the distinction between time-stable and timeunstable adjectives had been applied. Table 6. Average rate of copula expression in locative, adjectival and nominal predications in child English (Becker 2000) Child

Nina Peter Naomi Adam

age range

2;0-2;2 2;0-2;3 2;0-2;7 2;7-3;4

locative

14,0% 26,7% 38,1% 4,9%

adjective non-stable

stable

total

noun

49,5% 39,8% 52,0% 43,3%

75,2% 60,0% 93,5% 44,4%

53,5% 42,2% 59,8% 42,9%

74,1% 81,2% 89,7% 44,4%

The similarity between the patterns in table 1 and table 6 calls for a parallel explanation in terms of the time-stability scale: in the case of a most timeunstable, 'verby' predicate like a locative, children acquiring English as a first language seem to feel less need to express a verbal copula. As the predicate becomes less 'verby' (from time-unstable adjectives, via timestable adjectives and finally to nouns), this need for a copula to fulfil the function of predicativizer becomes stronger and the expression rates increase. Again, this analysis is in line with the cline of predicate-worthiness as discussed in section 4.1. According to this interpretation, the sub-hierarchy of predicability, as reflected in diachronic pathway 1 and explained in terms of the timestability scale, also accounts for the pattern of copula introduction (i.e. the selection of expression strategies) in first language acquisition, just as it accounted for synchronic social variation in copula deletion. Finally, although Becker's study provides evidence that first language acquisition develops along typologically universal lines, it is important to point out that much more research on different languages, different parts of the system of non-verbal predications and other typological hierarchies is needed to be more conclusive in this respect. Also, it would be necessary to see how the expression patterns of children in various age groups relate to this issue.

The explanatory power of typological hierarchies

271

4.3.2. Second language acquisition As is the case with first language development, there are not many studies about the acquisition of non-verbal predication systems and their expression formats in a second language. One relevant study in this respect was carried out by VanPatten (1987) and is concerned with the acquisition of the copula by adult, English-speaking second language learners of Spanish. As is well known, Spanish has two copulas; ser and es tar. The difference in meaning and use of these copulas is related to the factor of time-stability and has been extensively studied. Generally speaking, ser is used with time-stable predicates and estar with unstable ones. Going into the details of the discussion about this (rather rough) division falls outside the scope of this article. Therefore, I will simply take over Hengeveld's (1986: 395379) classification, as outlined below: (33)

Use of ser and estar in Spanish Equative predications (identifying and classifying): Ascriptive predications: Locative: Adjectival: unstable property: time stable property: Nominal: Possessive:

ser estar16 estar ser ser ser

VanPatten's hypothesis is that all second language learners have to pass through certain stages when acquiring grammatical structures. He reports such a natural order of acquisition for the Spanish copular system, based on oral data and grammaticality judgements. The stages he finds are defined as below: 17 (34) Stages of acquisition of the copula in Spanish as a second language 1. Absence of copula in learner speech. 2. Selection of ser to perform most copula functions. 3. Appearance of estar with locatives. 4. Appearance of estar with time unstable adjectives (VanPatten 1987: 64)

What can be immediately observed is that the transitions from stage 2 to stage 3 and from stage 3 to stage 4 show a pattern that can again be related

272

Eva Η. van Lier

to the sub-hierarchy of predicability (ascriptive predications), diachronic pathway 1 and the time-stability hierarchy. That is to say: estar is acquired first in constructions with the most unstable predicates: the locative constructions. From there it extends towards time-unstable adjectives. The remaining data are harder to interpret. For one thing, only part of the predicable non-verbal predication types is covered; no reference is made to equative predications or remaining types of ascriptive predications (nominal and possessive). Furthermore, it is unclear how the transition from stage 1 to stage 2 develops. It appears to be the case that ser suddenly "pops up" in all types of predications (rather than getting selected earlier in certain types than in others). It is probable that this is due to influence from English, the native language of the participants, which has only one copula. To test this idea, one would need to investigate whether the acquisition of ser and estar develops quicklier in second language learners with a native language in which time stability also crucially determines the use of certain expression formats or different copulas. Such research would also allow a more precise comparison with data from processes of first language acquisition. Clearly, the scarcity of available data on the acquisition of copula systems by second language learners makes it impossible to draw any firm conclusions about the relevance for such learning processes of the typological generalisation in the domain of non-verbal predication. However, a fair amount of research has been done on the explanatory power of other synchronically attested typological hierarchies for second language acquisition. In fact, these studies do suggest that the interlanguage systems of second language learners are governed by the same hierarchical systems that govern primary languages, while at the same time interacting with the specific characteristics of the learners' native language (Van Lier 2002, Verhagen 2003, Giacalone Ramat 2003).

5. Discussion and conclusion To summarise the evaluations of the developmental data in section 4, they all indeed appear to follow the hierarchies of non-verbal predication attested in synchronic typological research.18 This conclusion strongly supports the idea that the implicational hierarchies, accounted for in terms of copula support as put forward in Functional Grammar, indeed represent universal characteristics of human language systems. Moreover, the ex-

The explanatory power of typological hierarchies

273

planatory power of a number of these hierarchies has been shown to be extendable in various new ways to various new domains. First, the study on copula deletion in AAVE (section 4.1.) showed that the sub-hierarchy of predicability (concerning ascriptive predications) can also account for expression strategies used in certain varieties of English. Moreover, the analysis of the deletion pattern in terms of the time-stability scale, combined with the age-effect, suggested that this language-internal variation should indeed be interpreted as a case of language change in progress. This idea was strengthened in section 4.2.1, where two universal pathways of change were introduced, both directly related to the synchronic hierarchies discussed in section 2, and, in the case of pathway 1, to the time-stability scale. These pathways were attested for languages that were relatively little influenced by contact with others, namely the IberoRomance languages and Chinese. Regarding contact-induced language change, the pathways also turned out to be at work in Sranan and Swahili, in combination with (earlier) processes of simplification due to the more or less extreme contact situations. Furthermore, the Hungarian case suggested that the zero-1 hierarchy, originally describing synchronically attested expression strategies, could also account for contact-induced language change. Finally, the two remaining developmental studies, those concerning first and second language acquisition, turned out to be interpretable in terms of the predicability hierarchy, the semantics of the time-stability scale and diachronic pathway 1. These outcomes are represented in (34) below: This summary should be interpreted as follows: The various types of non-verbal predications (first column) are determined in terms of (i) predicate type and (ii) predication type (second column). Both types of characteristics are relevant for various typological hierarchies (third column), which account in turn for various types of developmental data in a number of different languages (fourth column). First, predicate type predicts the distribution of expression formats as described by the zero-1 hierarchy. This hierarchy accounts for the case of contact-induced language change (Hungarian). The second determinant of types of non-verbal predication is predication type. This factor determines the shape of the predicability hierarchy. This predicability hierarchy can be divided into a super-hierarchy (equative > ascriptive) and a sub-hierarchy (ascriptive, i.e. locative > adjective > nominal > possessive). The super-

274

Eva Η. van Lier

(34) Non-verbal predication and its typological and non-typological explanatory domains domain

determinants

non-verbal predicate predicatype tion predication type

Hierarchies

Data

zero-1 hierarchy

-

predicability hierarchy

superhierarchy / zero-2 hierarchy

-

sub-hierarchy

-

-

-

-

contact-induced change (Hungarian)* internal change: pathway 2 (Chinese) contact-induced change: pathway 2 (Sranan, Swahili)* synchronic variation (AAVE)* internal change: pathway 1 (IberoRomance languages) language acquisition (English, Spanish)*

hierarchy is relevant for expression formats as described by the zero-2 hierarchy. Furthermore, it relates to diachronic pathway 2, as manifested in internally motivated language change (Chinese), and contact-induced language change (Sranan and Swahili). The sub-hierarchy (ascriptive predication) accounts for intra-linguistic synchronic variation or language change in progress (AAVE), language-internally motivated change (pathway 1, as attested in Ibero-Romance languages) and first and second language acquisition (of English and Spanish respectively). Finally, the cases in the fourth column that are marked with an asterisk constitute the new explanatory domains that have surfaced in the present study. These cases are (i) the role of the zero-1 hierarchy in contactinduced change, (ii) the role of the predicability super-hierarchy in contactinduced change, (iii) the role of the predicability sub-hierarchy in lan-

The explanatory power of typological hierarchies

275

guage-internal synchronic variation, and (iv) in first and second language acquisition. Thus, not only can it be concluded that the developmental data assessed in this study can be accounted for in terms of the typological generalisations proposed by Hengeveld, but moreover some of these generalisations have even wider explanatory domains than suggested by his typological research. These conclusions give reason to consider the approach explored in this study a potentially fruitful one. Of course, more research is needed to gain further insight into the systematic explanatory power of typology by assessing other universals, hierarchies and cases than the ones discussed here. Another very important issue that is in need of further research, in my view, concerns the possible interactions between linguistic universals and social variables in determining the (dynamic) shape of human language.

Notes 1.

2.

3.

4. 5.

Some copulas, such as English become, do add some meaning to the construction in which they occur. This meaning is mostly of an aspectual nature (ingressive, in the case of become). This type of copula is called "semi-copula" by Hengeveld (1992:34-39) and will not be discussed here. In fact, ascriptive predications can also express existence. However, such predications belong to the subdomain of so-called ascriptive presentative predications. These play no significant role in the studies discussed in the remainder of this article. Furthermore, according to Hengeveld (1992), their behaviour in languages is considerably different from other types of non-verbal predications. For these reasons, I will not take presentative predications into account and will refer to non-presentative ascriptive predications simply as "ascriptive". In fact, Hengeveld (1992: 141-144) proposes a second sub-hierarchy for equative predications, stating that classifying predications are more easily predicable than identificational ones. However, this hierarchy is based on only one language in his sample (Abkhaz). Furthermore, he cites two different analyses for the relevant data. For these reasons, the sub-hierarchy of equative predications will not be taken into account in my further analyses. For details on the social characteristics of the groups studied and full references, see Rickford et al. (1991). A second argument supporting the idea that the distinction between ascriptive and equative predications based on a noun is not relevant for the analysis of

276

Eva Η. van Lier

the copula deletion pattern in AAVE is Hengeveld's observation that languages which use the same copula form in locative predications as in equative classifying predications must use that form in predications based on an adjective as well (Hengeveld 1992: 216). This strengthens the idea of the adjective occupying a position in between locatives and nouns, irrespective of whether the latter forms part of an ascriptive predications or an equative one. 6. This issue is discussed further in section 4.3.1. 7. Hungarian does not allow bare nominal or referential predicates in non-finite constructions. Therefore, they are expressed by means of a finite relative clause. 8. De Groot (2005) adds to this translation: [lit. 'the clever being girl']. Although the 'being' is not expressed, this is to make clear that (17c) is a nonfinite, embedded predication. The same applies to (18). 9. Norval Smith (personal communication), suggests that the origin of the locative copula de may also be its (currently still existing) function as a deictic adverbial meaning "there". In this scenario, the adverbial became reanalysed as a locative copula and developed from there its secondary function as a marker of durative aspect. Van den Berg (2000) cites examples from early Sranan, which support this claim in so far as the locative copula function of de is attested earlier than its aspect marking function: (i) 1757 eevi a ben die (=de) na tampatie if he PAST be in Tempati 'If he was in Tempati' 1762 Ο goedoe gado masra, koudemaas de foen mi misjie oh good God master, Koudemaas DUR strike my woman 'Oh good heavens, master, Koudemaas is raping my wife!' (van den Berg 2000: 92,94) Furthermore, in West Greenlandic the equative copula tassa also originated from a deictic adverbial, also expressed as massa: (ii) a. Hansi tassapisurtaq Hansi COP leader 'Hansi is the leader' b. massa arnat tikiraa -t ina-at here.is women.REL visitor.PL -REL room-POSS 'Here is the women's visitors room' (Fortescue 1984: 72, 77) 10. This type of construction is termed "attributive" by Arends, and is regarded as a subtype of equative predication, whereas in Hengeveld's terminology it would be an instance of ascriptive nominal predication (status-assigning). To avoid confusion, I make use of Hengeveld's classification in my analysis.

The explanatory power of typological hierarchies

277

11. The status of possessive predicates is unclear in this analysis; a problem that can possibly be related to the somewhat unclear status of possessives of the time stability scale, as mentioned in section 4.1.1. 12. For more discussion on the influence of substrate languages versus universal diachronic processes on the development of copula systems in Creole languages, see for example McWhorter (1997). 13. No reference is made to the predicability and/or the expression of possessive predications. 14. In McWhorter's terminology, these types of predication are all termed 'equative'. As before, in further analysis Hengeveld's categorisation is used, in which they are called 'ascriptive'. 15. As was the case with the AAVE study (section 4.1.2), Becker (2000) does not differentiate between ascriptive and equative predications based on a noun. However, for the same reasons as discussed in section 4.1.2, the distinction is not relevant for the proposed analysis. 16. In fact, Spanish uses estar in locative predications only when the first argument is a term that does not designate an event (but an animate being or an object). When the first argument is an event, ser must be used (Hengeveld 1986: 396-397). However, as will be shown below, VanPatten does not discuss the latter type of locative prediaction. 17. VanPatten (1987) includes the use of estar with the gerundio or progressive (Juan esta estudiando, 'Juan is studying') in his analysis. In my view, however, in those constructions estar is not a copula but rather an auxiliary. Therefore, I will not consider them any further. 18. As is shown by Herrero-Blanco and Salazar-Garcia (this volume), the explanatory power of the typological hierarchies for non-verbal predication also includes data on signed, rather than spoken language data.

References Arends, Jacques 1986 Genesis and development of the equative copula in Sranan. In Substrata versus universals in Creole genesis, Pieter Muysken and Norval Smith (eds.). Amsterdam: John Benjamins. 1989 Syntactic developments in Sranan. Ph.D. diss., University of Nijmegen. Becker, Misha 1999 The acquisition of copulas. Dissertation prospectus, University of California.

278

Eva Η. van Lier

2000

The development of the copula in child English: The lightness of be. Ph.D. diss., University of California. Berg, Marinus van den 2000 "Mi no sal tron tongo". Early Sranan in court records. 1667-1767. M.A. thesis, University of Nijmegen. Dik, Simon C. 1980 Studies in Functional Grammar. London: Academic Press. 1997 The theory offunctional grammar 1: The structure of the clause. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Fortescue, Michael 1984 West Greenlandic. Croom Helm descriptive grammars. London: Croom Helm. Giacalone Ramat, Anna (ed.) 2003 Typology and second language acquisition. Berlin: Mouton de Gruyter. Givon, Talmy 1984 Syntax: a functional-typological introduction. Vol. 1. Amsterdam: John Benjamins. Greenberg, Joseph 1978 Diachrony, synchrony and language universals. In Universals of human language. Volume 1: Method and theory, Joseph Greenberg (ed.), 61-91. Stanford: Stanford University Press. Groot, Casper de 2005 Typological observations in the grammars of Hungarian outside Hungary. In Hungarian language contact outside Hungary: studies on Hungarian as a minority language, Anna Fenyvesi (ed.). Amsterdam: Benjamins. Hengeveld, Kees 1986 Copular verbs in a functional grammar of Spanish. Linguistics 24: 393-420. 1991 Tipologia, diachronia, sincronia. Foro Hispanico 2. 1992 Non-verbal predication. Theory, typology, diachrony. Berlin: Mouton de Gruyter. Herrero-Blanco, Angel and Ventura Salazar-Gracia this vol. Non-verbal predicability and copula support rule in Spanish Sign Language.

The explanatory power of typological hierarchies

279

Li, Charles N. and Sandra Thompson 1977 A mechanism for the development of copula. In Mechanisms of syntactic change, Charles N. Li (ed.), 419-444. Austin: University of Texas Press. Lier, Eva van 2002 Typology and SLA: The acquisition of complement constructions in Dutch as a second language. M.A. thesis, University of Amsterdam. Mallison, Graham 1986 Rumanian. Croom Helm descriptive grammars. London: Croom Helm. McWhorter, John 1992 Ni and the copula system in Swahili. A diachronic approach. Diachronica XI: 15-46. 1997 Towards a new model of Creole genesis. New York: Peter Lang Publishing. Meyers Scotton, Carol 1985 Language universals and syntactic changes in Swahili as a second language. Studies in African Linguistics, supplement 9: 290-292. Pustet, Regina 2003 Copulas. Universals in the categorization of the lexicon. Oxford: Oxford University Press. Rickford, John R., Aretha Ball, Renee Blake, Raina Jackson and Nomi Martin 1991 Rappin' on the copula coffin: Theoretical and methodological issues in the analysis of copula variation in African-American Vernacular English. Language variation and change 3: 103-132. Short, David 1993 Slovak. In The Slavonic languages, Bernard Comrie and Greville Corbett (eds.), 533-592. London: Routledge. Stassen, Leon 1992 A hierarchy of main predicate encoding. In Meaning and grammar: cross-linguistic perspectives, Michel Kefer and Johan van der Auwera (eds.), 179-201. Berlin: Mouton de Gruyter. 1997 Intransitive predication. Oxford: Oxford University Press. Thomason, Sandra 2001 Language contact, an introduction. Edinburgh: Edinburgh University Press. VanPatten, Bill 1987 The acquisition of ser and estar: accounting for developmental patterns. In Foreign Language Learning: a research perspective, Bill

280

Eva Η. van Lier

VanPatten, Trisha R. Dvorak and James F. Lee (eds.)· Cambridge: Newbury House. Verhagen, Josje 2002 Typological universals and SLA: The acquisition of expletives in Dutch as a second language. Toegepaste Taalwetenschap in Artikelen 69:1, 39-55.

Non-verbal predicability and copula support rule in Spanish Sign Language r

Angel Herrero-Blanco Ventura Salazar-Garcia

1.

Introduction

Spanish Sign Language (lengua de signos espanola: LSE) is a visualgestural language used by the majority of the Spanish deaf community.1 Although we do not know exactly when it dates back to, we do know that it has been in existence for a long time. This is borne out by the fact that in the middle of the 19th century, Villabrille (1851) produced a dictionary in which approximately fifteen thousand signs were described.2 In recent times, particularly since Rodriguez (1992), an increasing interest has been shown in studying and researching this language (Munoz 1999; Herrero & Alfaro 1999; Morales et alii 2000; Chapa 2000; Herrero 2000, 2002, 2003; Herrero & Salazar 2002). However, as occurs with other sign languages (SLs), there is still a great deal to be done. Two important subjects bear mentioning: a) the general agreement on verbal morphology (Supalla 1986; Padden 1990; Liddell 1990; Valli & Lucas 1992; Neidle, Kegl, MacLaughlin, Bahan & Lee 2002), and b) the linguistic problem of topographical (descriptive) space. The point at issue is whether SLs make use of two codes, linguistic and visual, with different grammars (Cogill-Koez 2000), or there is one LS grammar in two modalities (Liddell 1994, 2000; Emmorey 1996; Liddell and Metzger 1989; Meier, Cormier & Quinto-Pozos 2002). This question, and the correlative one about the classifiers, is open (cf Herrero 2002: 13). In any case, the non-verbal predicability may be studied without considering either the specific topographical space or the non-manual copula formulated for ASL (Liddell 1980), which is irrelevant in LSE (cf Herrero & Salazar 2002). From our point of view, SLs, as symbolic instruments for human interaction, fulfil all the requirements for being considered natural languages in

282

Angel Herrero-Blanco and Ventura Salazar-Garcia

their own right. Apart from the special conditions of their output, their structural configuration reveals a clear linguistic nature. Therefore, it is not only possible, but also necessary, to approach research into SLs from a typological perspective. This paper aspires to offer a modest contribution to this effect. After this introduction, the second section is devoted to briefly reviewing the treatment of the copula and non-verbal predication in Dik's model (2.1), focusing our attention on Hengeveld's contribution. We will also explain the initial hypotheses that guide our paper (2.2). These hypotheses have in common the idea that typological regularities in oral languages (OLs) must be valid for LSE. The third section deals with nominal and adjectival predications in LSE. There, we will take into account the oppositions between equative and ascriptive relations (3.1), and those between characterization and specification (3.2). Our study reveals that ascriptive predication and equative one expressing characterization share the same structure. However, the equative predication expressing specification offers in LSE a marked construction, which may be considered as a kind of clefting. All these cases are non-verbal constructions without any copula. In a special subsection (3.3), we discuss which type of alternative for the copula is used. The fourth section analyses presentative and nonpresentative predications expressing possession. The presentative sentences are characterized by the use of a verbal lexical predicate, which we transcribe as HAVE (4.1). Such a predicate allows a transitive use as well as a pseudo-transitive, a circumstance that marks the grammatical difference between possessive and locative-existential predicates. We also present the reader with a brief subsection about the special use of HAVE with an absentive value (4.2). Finally, we study the possessive non-presentative constructions (4.3). The following section refers to non-presentative locative predications; it is convenient to mark on them the differences between sentences with deictic (direct and indirect) predicates (5.1) and sentences with non-deictic predicates (5.2). In the last ones, we have noticed the existence of a copula, which is the product of the grammaticalization of the demonstrative adverb, for indirect deixis, THERE(i). For this reason, we include the corresponding copula support rule (5.3). Lastly, we draw our conclusions in section number 6. There, we make explicit, in the frame of Dik's Functional Grammar, the mechanisms of formalization that regulate the different constructions under study, and we also evaluate the degree of accuracy of our initial hypotheses.

Non-verbalpredicability

and copula support rule

283

2. Non-verbal predicability in Functional Grammar 2.1. Types of non-verbal predicates The theoretical framework that we are going to adopt is the Functional Grammar (FG) designed by Simon C. Dik (1989, 1997). The high level of typological adequacy that this model has attained makes it a very promising instrument for undertaking the linguistic analysis of SLs. Moreover, its most recent developments, oriented towards a Functional Discourse Grammar (cf Hengeveld 2004) distinguish clearly between the grammatical organization of linguistic expressions (interpersonal, representational and expression levels) and their actual output (articulation). In our opinion, this allows us to see that the differences between OLs and SLs (except for those derived from topographical or descriptive space) occur in the final step: articulation, which is not, strictly speaking, a level within grammar. Nothing impedes complete congruence on the previous levels. Another important factor which has influenced the choice of theoretical framework is the fact that, within FG, important research has been carried out on non-verbal predications. Dik (1989: 161-182; 1997, vol. I: 193-216) offers the general guidelines for the analysis of this type of predication. He points out that non-verbal predications are neither exceptional nor anomalous in the languages of the world. Likewise, in the framework of FG he incorporates the 'copula support theory'. This theory states that copulative constructions are a particular type of non-verbal predications.3 From this point of view, the copula is a semantically empty supportive device which acts as a carrier for certain operators. Consequently, in FG the copula is not deleted in those conditions in which it does not occur, but is introduced, by means of an expression rule, in those conditions in which it appears. For his part, Hengeveld has studied this subject in depth, as is reflected in various publications (Hengeveld 1986, 1990a, 1990b, 1991, 1992a, and, especially, 1992b) in which he makes an important theoretical and typological contribution by presenting new evidence which substantiates the copula support theory. So, the study of non-verbal predication has in FG a very solid background. It facilitates considerably our approach to this aspect of LSE. Besides, we believe that this background makes such constructions a particularly promising field for evaluating the viability of a functional linguistic approach for researching into SLs. According to Dik (1980: 102-104) and Hengeveld (1992b: 74ff) the following types of non-verbal predicates may be distinguished:

284

Angel Herrero-Blanco and Ventura Salazar-Garcia

(1)

a. Bare predicates. Predicates whose distinguishing use is nonpredicative b. Referential predicates. Predicates based on terms or larger referential units c. Relational predicates. Predicates based on a referential expression that carries an indication of a semantic function

Each of these types is associated with the expression of determinate semantic relations. Thus, the application of bare predicates to terms gives rise to ascriptive predications which express property assignment (adjectival predicates, (2a)) or status assignment (nominal predicates, (2b)): (2)

a. My pullover is brown. b. John F. Kennedy was president.

The application of referential predicates to terms produces equative predications expressing characterization (3a) or specification (3b): (3)

a. Madrid is the capital of Spain. b. The capital of Spain is Madrid.

The application of relational predicates to terms leads to ascriptive predications with a specific semantic value: location (4a), existence (4b),4 possession (4c), etc: (4)

a. We are in Alicante. b. There is an important meeting. c. This book is mine.

Within these groups, we may discover other semantic distinctions which are relevant from the grammatical viewpoint. Thus, in referential predicates, their definite or indefinite character expresses the distinction between identification (5a) and classification (5b). Relational predicates expressing location-existence or possession may reveal a presentative value (6a, 6b) or a non-presentative one (6c, 6d):5 (5)

a. This is the book I was looking for. b. A peccary is a South American mammal.

Non-verbalpredicability

(6)

and copula support rule

285

a. There is a book on the table. b. Est mihi equ-us alb-us. C0P.3.sg.pres l.sg.dat horse-nom.m white-nom.m Ί have got a white horse.' (Latin; lit: 'a white horse is for me.') c. The book is on the table. d. The white horse is mine.

Nevertheless, there is no complete correspondence between the grammatical constructions and the contents expressed. This is due to the fact that the presence and productivity of the different types of non-verbal predicates are subject to typological variation, and they are not the same in all the languages of the world. This fact can be illustrated by some examples. Spanish possesses no non-verbal construction for expressing the presentative variety of existential or locative meanings. Instead, it uses a pseudotransitive structure with a verbal predicate (haber). In Yagaria, an IndoPacific language in which adjectives cannot be used predicatively, property assignment is expressed by means of an equative construction (Hengeveld 1992a: 77). Finally, we should remember that Mandarin Chinese does not allow the predicative use of non-verbal possessive predicates. Possession is expressed by means of an existential construction (cf. Hengeveld 1990b, 1992b: 127ff). For all these reasons, research into non-verbal predications must determine, for each language, the scope of non-verbal predicability, which Hengeveld (1992a: 83) defines as follows: (7)

Non-verbal predicability The possibility of grammatically acceptable application of a nonverbal predicate β to an argument a.

From among the typological regularities formulated by Hengeveld (1992b) regarding non-verbal predicability, we would like to highlight: (8)

a. Equative non-verbal predications are universal or almost universal b. Ascriptive non-verbal predications are ordered according to the Non-Verbal Predication Hierarchy (NPH):LOC > A > Ν > POSS c. Many languages use different expressions for presentative and non-presentative contents

286

Angel Herrero-Blanco and Ventura Salazar-Garcia

Hengeveld (1992b: 185ff) indicates that, in OLs, non-verbal predication has, basically, two alternative formats for the copula, which correspond to two different strategies. In the first of these, known as 'zero-Γ, the nonverbal predicate shows the same morphosyntactic behaviour as an (intransitive) verbal predicate, as in (9a). The second strategy, 'zero-2', does not permit such morphosyntactic correspondence, and the non-verbal predicate simply appears juxtaposed to the argument, as in (9b). Another significant difference is that the characteristic format of the zero-1 strategy is incompatible with the introduction of a copula (which would be ungrammatical). On the other hand, the zero-2 strategy frequently permits the introduction of a copula as an optional element: (9)

a. Abkhaz Da-ps 9-w-p'. 3. SG. SB J-dead-PRES-DECL 'He is dead.' 6 b. Latin Summ-um ius summ-a iniuri-a greatest-NOM.N law.NOM greatest-NOM.F injustice-NOM.F (est). (COP.3.SG.PRES) 'The greatest law is the greatest injustice.'

2.2. Hypotheses with respect to LSE The validity of the previous formulations has been amply confirmed in OLs. Our aim here is to ascertain to what extent they are fulfilled in the case of LSE. The idea that complete congruity exists between grammars of SLs and OLs leads us to suggest the following basic hypotheses: (10) a. LSE will have equative non-verbal predications, expressing semantic relations of characterization or specification b. If LSE has ascriptive non-verbal predications, their scope will be shown to be consistent with NPH c. We cannot determine beforehand the existence or non-existence of copula in LSE, since the non-verbal constructions can adopt alternative strategies. Now, assuming that a copula existed in as-

Non-verbalpredicability

and copula support rule

287

criptive predications, then, logically, its appearance would also be consistent with NPH. That is to say, the possibility of using the copula will decrease as we advance towards the right in NPH d. It is highly probable that significant grammatical differences exist in the expression of presentative and non-presentative contents In the following pages we will analyse how semantic relations typical of non-verbal predicates are expressed in LSE. This will allow us to determine the scope of non-verbal predicability in LSE and check to what extent our hypotheses are correct.

3. Nominal and adjectival predicates in Spanish Sign Language 3.1. Ascriptive and equative relationships In LSE we frequently detect the presence of verbless sentences, in which a noun or an adjective has a predicative function. In such cases, the nonverbal predicate appears after its argument, in accordance with an unmarked order 'argument + predicate' with no copula. Now, given that in LSE there are no articles, the question is how to determine whether these sentences establish an ascriptive or an equative relationship.7 At this phase of our research, we have not found any systematic procedure to distinguish between the two relationships, at least in the field of syntagmatic organization of the constituents in the utterance. Nevertheless, there are other factors that we have not studied in depth yet; for instance, prosodic contour (established by pauses and non-manual signs) in real communicative situations. We cannot refuse the idea that some of these factors might be grammatically relevant for the equative/ascriptive distinction. So, our considerations about this problem are necessarily provisory, and must be reopened in further studies. Taking into account what was said before, we may consider that, as far as we have been able to establish at this moment, in LSE there is no formal mechanism which allows us to distinguish, in nominal and adjectival predicates, between the ascriptive interpretation and the equative one. For this reason, the sentences included in (11) are ambiguous8 if we consider them in isolation:

288

Angel Herrero-Blanco and Ventura Salazar-Garcia

(11)

a. MY MOTHER PHYSICIAN 'My mother is a physician' 9 / 'My mother is the physician' b. MY JACKET WHITE 'My jacket is white' / 'My jacket is the white one'

In some cases, certain semantic or syntactic features provide keys for the correct interpretation of these sentences. For example, the presence of a demonstrative, a quantifier, or a possessive in the non-verbal predicate indicates that it is a term used predicatively, with identification value as in (12a). If the predicate is a hyperonym of the noun that appears as argument, it is probable (but not absolutely certain) that the closest interpretation corresponds to an equative relationship of classification as in (12b): (12)

a. 3.sg MY FRIEND ' {He/she} is my friend.' b. CAT ANIMAL Ά cat is an animal.'

In any case, these circumstances only arise in a relatively small number of non-verbal predicates. Therefore, we assume that, in general terms, the disambiguation of nominal and adjectival predicates is fundamentally determined by pragmatic factors. It is the linguistic or situational context which provides the keys necessary to decide whether a determinate predication has an ascriptive or an equative character. Whether a nominal predicate is or is not restricted (for example, by means of an adjective) is irrelevant when determining the ascriptive or equative character of predication. (13)

MY FRIEND PHYSICIAN FAMOUS 'My friend is a famous physician'/ 'My friend is the famous physician.' 1 0

We can relate this to the fact that ( l i b ) can have an equative interpretation. In principle, it is strange that an adjective should act as a head of an equative predication (i.e. a referential predicate). This fact can be explained in two possible ways. One of them would depend on admitting that LSE is a flexible language in which adjectives and nouns would not have a precise functional delimitation. The second explanation would indicate that, in such cases, the adjective is in reality adjacent to a noun which is

Non-verbalpredicability

and copula support rule

289

understood from the context. In accordance with this, it would really be a predication with a nominal head, although its external appearance coincides with that of adjectival predications. In our opinion, this second explanation is the one which better fits the facts detected in LSE. As far as we have been able to ascertain, LSE behaves like a specialized language,11 and not like a flexible one. The equative interpretation depends on determinate contextual conditions, among which there would be the possibility of recognizing the elided noun. In ( l i b ) the noun in question is easily retrievable, since it is present in the argument of predication: JACKET. This explanation is totally analogous to the one which Hengeveld (1992b: 61ff) puts forward for the Spanish case. In accordance with what has been stated up to now, we can point out that in LSE nominal predicates may appear in ascriptive predications and in equative predications. In the former case, it must be understood that the predicate is a bare nominal, and the predication has the semantic value of status assignment. In equative predications, the noun is really the head of a term used predicatively. As regards adjectives, they also appear as head of ascriptive predications, with a value of property assignment. When they appear in the predicate of an equative predication, they are understood to function as adjacent elements of an elliptical noun. The recognition of one type or the other depends on the pragmatic conditions, because, at least until this moment, we have not found any systematic difference in the formal appearance of that predication when the non-verbal predicate has no formally marked term operators. In any case, this fact is not the result of any characteristic of the non-verbal predicability in LSE. It is a logical consequence of the lack of articles in this language, and also affects verbal predication. In LSE, the nominal predicates of the lexicon can be directly integrated into discourse, in the form of terms, without it being necessary to introduce any formal determination marker. 12 But this also occurs in other languages, so it does not affect the congruity of LSE with OLs. Proof of this is the situation in Latin (cf Pinkster 1984: 2, 118-121), which in this respect is very similar to LSE. That is why the ambiguity that can be resolved from the context, which we pointed out in the examples (11) and (13), also exists in the case of (14): (14)

Alexander erat rex Macedonum. 'Alexander was {0/a/the} king of Macedonians.'

290

Angel Herrero-Blanco and Ventura Salazar-Garcia

3.2. Characterization vs. specification Up to now, when we have referred to equative predications, we have restricted our attention to those which expressed characterization (in its double aspect of classification and identification). Several authors, including Longobardi (1987), have warned about the necessity to bear in mind another type of equative predication, which Hengeveld (1992b: 82-89) calls specification.13 In many OLs clear grammatical differences can be seen between characterization predication and specification predication. 14 In LSE, we can also appreciate such differences. The way in which the semantic relation of specification can be expressed takes the form of a nonverbal predication without copula, but with certain characteristics which make it a marked construction. It consists of what we could call a 'question-answer construction'. The sentence is articulated in the form of a 'microdialogue': a question, a brief prosodic pause, and a reply. The question is marked by a non-manual sign: raised eyebrows. It is important to notice that, in true interrogative utterances, raised eyebrows are the prosodic marker for yes/no questions, not for wA-questions. Surprisingly, it is also used in this 'question-answer construction' (apparently, a wh- question). It is an interesting point, and we must pay more attention to this in further research. The non-manual sign may be accompanied by a wA-pronoun, which in LSE goes at the end of the question in (15a-b). But this interrogative pronoun is optional, as the interrogative is often marked only by the nonmanual sign in (16). Up to now, we have not been able to determine whether the presence or absence of the w/z-pronoun is conditioned by some dialectal, sociolinguistic, or pragmatic factors. From the grammatical point of view, both options are equally valid: 15 (15)

a. SAD WHOy/n, YOUR FATHER 'The sad one is your father' b. FUNNY THE-MOST WHAT y)n . EXCURSION 'The funniest is the excursion'

Non-verbal predicability and copula support rule (16)

291

a. FRANCE CAPITAL^. PARIS 'The capital of France is Paris.' b. MD y / n , YOUR FATHER 'The sad one is your father.' c. TEACHER y / n , 3.sg 'The teacher is {he/she}.'

A problem which could be considered is whether it is really a marked word order, or if, on the contrary, it is a different allocation of functional roles in the underlying semantic structure. In other words: which constituent is the predicate in this construction, the question or the reply? If the predicate is the first constituent (TEACHER, in (16c)), the difference with regard to the characterization constructions would be caused by a word order rule, the origin of which would surely have to be sought in the allocation of pragmatic functions (assignment of Focus, as we will remark). Now, it could be stated that the predicate is the second constituent (3.sg). In that case, the order of the constituents (argument + predicate) is maintained, and the difference in relation to the characterization constructions would come from the semantic structure. Jespersen (1924: 172-178) already pointed out that to identify the subject and the predicate in this type of coextensive predication is not really an important question. In any case, and at least with reference to English examples which he uses, he essentially opts to identify the subject with the constituent which precedes the copula, and the predicate with the one that follows it. Hengeveld (1992b: 87) also seems to prefer this option. But, in our opinion, the explanation based on constituent order fits in better with the intuition of LSE users. So, this will be the hypothesis we will adopt here. In any case, the facts are not conclusive, so it is still an open question. 16 It should be pointed out that the question-answer construction is also frequent in the verbal predications of LSE (at least the option with an explicit w/z-pronoun), as a means of highlighting one of the constituents: (17)

a. l.sg LIVE WHERE y/n , ALICANTE 'Where I live is in Alicante.' b. TODAY COME WHQy/„, MY BROTHER 'The person coming today is my brother.'

This fact shows that neither the w/z-pronoun nor the prosodic marker must be understood here as a copula. The most plausible hypothesis is based on

292

Angel Herrero-Blanco and Ventura Salazar-Garcia

considering the interrogative marker as a procedure for the identification of the constituent with the pragmatic function Focus. This could be related to the mechanisms of clefting in OLs. The introduction of the prosodic interrogative marker (raised eyebrows accompanying the wA-pronoun) is the result of an intonation rule intended to establish the prosodic contour. The w/i-pronoun, which can appear optionally in non-verbal predications, is introduced by means of a rule relating to the form of linguistic expression (cf Hengeveld 1992b: 13f). In our opinion, the marked word order (predicate + argument) and its appearance as a question-answer sequence are strategies in accordance with the principle of pragmatic highlighting. FG literature has explicitly pointed out the close relationship that exists among this principle, the expression of Focus, and such strategies as question-answer pairs (Dik 1997, I: 328ff) and 'special pragmatic positions' (Dik 1997,1: 403). A more detailed analysis of this question would require specific research into the conditions of use of these utterances. However, we believe that, as a working hypothesis, we can consider the idea that, in LSE, specification contents are necessarily associated with the assignment of the pragmatic function Focus to the argument of the predication. That is to say, a specified argument must have a specific informational status, which would essentially correspond to the function known as 'completive Focus'. So, for example, an LSE user would emit the sentence (15b) if, due to contextual conditions, it were necessary to specify that the predicate FUNNY THE-MOST ('the funniest') corresponds to the term EXCURSION, and not to other possible alternatives (for example: PARTY, SWIMMINGPOOL, etc.).

3.3. Copula formats: zero-1 and zero-2 strategy As we pointed out in 2.1, non-verbal predications have two alternative formats for the copula, which correspond to two different strategies: zero-1 and zero-2. Given that the most characteristic feature of LSE nominal and adjectival predications is the absence of copula, we must determine whether their expression format complies with the zero-1 or the zero-2 strategy. In the examples we have previously cited (11-13), what can be seen is juxtaposition between the argument term and the non-verbal predicate. This could make us believe that we are dealing with a zero-2 strategy. However, it must be taken into account that no predication operator figured

Non-verbalpredicability

and copula support rule

293

explicitly in these examples, and they are not, therefore, conclusive. In fact, in such conditions, the LSE verbal predication also shows a mere juxtaposition between the argument and the lexical root of the predicate, with no specific morphosyntactic marker. So, it is necessary to analyse the behaviour of predicates together with their operators. For this subject, the most relevant grammatical category in LSE is aspect.17 Tense and modality never produce variation in the morphological paradigms, since they are expressed by means of free constituents with clause-external pragmatic functions. 18 Taking these preliminaries into account, the facts show that the expression format of the predications studied here essentially corresponds to the zero-1 strategy: (18)

a. 3.sg WORK ' {He/she} works.' b. 3.sg WORK.cont ' {He/she} keeps working.' c. 3.sg SAD '{He/she} is sad.' d. 3sg SAD.cont ' {He/she} keeps being sad.'

These examples19 highlight the fact that the expression of the continuative aspect in verbal and non-verbal predication is carried out by means of the same procedure: a morphological change in the predicate, which consists of a prolongation and a slowing down of the movement of the hand. When the aspect is expressed by means of a free morpheme, verbal and nonverbal predications also show totally analogous behaviour. The marker of egressive and perfective aspects normally consists of a free sign, which can be translated as 'end'. This sign follows the predicate, independently of whether the said predicate is a verb (19a), an adjective (19b), or a noun (19c): (19) a. 3.sg WORK END ' {He/she} isn't working any more.' b. 3.sg SAD END ' {He/she} isn't sad any more.' c. 3.sg WAITER END ' {He/she} isn't a waiter any more.'

294

Angel Herrero-Blanco and Ventura Salazar-Garcia

Therefore, there seem to be reasonable motives to consider the nominal and adjectival predication of LSE to be examples of the zero-1 strategy.

4.

Presentative predications and non-presentative possessive contents

4.1. Presentative predications by means of HAVE As we pointed out in 2.1, we should pay attention to the distinction between presentative and non-presentative ascriptive predications, as in many languages they involve grammatical differences. Presentative predication is articulated around a referential predicate with a semantic value of locationexistence or possession. The study of LSE shows us that, in this language, presentative predication has a sign that we translate here as HAVE. This sign is used not only in expressions with a location-existence value, but also in those with a possessive meaning. Some authors, such as Rodriguez Gonzalez (1992: 329), have considered the possibility of viewing this sign as a copula. However, we believe that its analysis, in the light of FG, clearly shows that it is a lexical predicate of a verbal nature. As Hengeveld (1992a: 77) warns, the types of meaning (existence, location, possession, etc.) cannot be identified with the types of construction, and what is expressed in one language by means of a copula may be expressed in another language by an alternative construction. In the case of LSE, we can see that the presentative constructions stand outside the limits of the non-verbal predicability of LSE. This can be observed in the following examples. (20a) is an existential expression, in which location is to be understood as lexically void, (20b) is a locative expression, and (20c) is a possessive expression: (20)

a. WATER HAVE 'There is water.' b. MY HOUSE DOG HAVE 'There is a dog in my house.' c. MY FATHER DOG HAVE 'My father has got a dog.'

Apparently, both (20b) and (20c) have the same structure. It may be thought that they correspond to the same kind of construction. For example, we may formulate the hypothesis that in LSE location is encoded by

Non-verbal predicability and copula support rule

295

means of a possessive construction, in which the locative term (MY HOUSE) would act as the 'possessing entity' of the patient argument (DOG). However, this is not so, since an important grammatical difference exists between both sentences. (20c) permits the introduction of a satellite with a semantic function [location], whereas (20b) does not: (21)

a. *HERE MY HOUSE DOG HAVE *'My house has got a dog here.' b. HERE MY FATHER DOG HAVE 'My father has got a dog here.'

This is due to the fact that MY HOUSE, in (20b), receives the semantic function [location], something which does not occur with the first argument (MY FATHER) in (20c). Therefore, the two sentences correspond to different grammatical structures. In our opinion, the cause of this difference should be sought in the lexical alternatives that Hengeveld (1992b: 157-160) reports in OLs for the cases in which a determinate content is not predicable by means of non-verbal predication. In (20c), HAVE acts as a true lexical transitive verb (strategy: LEX1), whose first argument (possessor entity) lacks a semantic function [ 0 ] , and whose second argument (possessed entity) has a semantic function [patient]. In (20b), HAVE must be understood as a pseudo-transitive verb; that is, an originally transitive verb which has lost some of its lexical characteristics (strategy: LEX2). In this case, there is a single argument, as [patient], and a satellite term as [location]. Formally, the construction has no subject (in the traditional sense of this word), because no constituent really acts as a first argument. Hengeveld illustrates strategy LEX2 with the case of the Spanish verb haber, which was originally a transitive verb with possessive meaning. This verb has an auxiliary use, and also a pseudo-transitive one. Nowadays, it no longer retains its original transitive use, in which it has been replaced by the verb tener. But in the 15th and 16th centuries it still preserved its possessive meaning. Leaving aside the auxiliary function, it can be said that the behaviour of HAVE, in LSE, is very similar to that of haber in Spanish five centuries ago. We believe that this close relationship is not only typological, but also genetic. Specifically, we think that there are sufficient arguments to believe that HAVE comes etymologically from hay, which is the third person singular in the present tense of haber, in its pseudo-transitive use. 20 The transition from Spanish to LSE must have taken place by a dactylological incorporation. Dactylology (or fingerspel-

296

Angel Herrero-Blanco and Ventura Salazar-Garcia

ling) is a procedure which represents letters of the alphabet on the hands. It began in the middle of the 16th century, with the work of a Spanish Benedictine monk, Pedro Ponce de Leon. Its goal was to contribute towards teaching deaf people to read and write (Herrero 2000). Today, dactylology is used fundamentally for proper names that have no specific sign. The sign for the verb HAVE in LSE corresponds to the dactylological sequence formed by the letters a and i, which coincides whith the phonic form of hay (/ai/). 21 Moreover, it would not be exceptional, as there are other, more frequent signs in LSE (for example: AND, OR, etc.), for which we could also suggest a similar source (cf Rodriguez Gonzalez 1992: 320ff). In short, from a grammatical point of view, the typological congruity of LSE with OLs (in this case, Spanish) is once again confirmed. From the sociolinguistic point of view, the existence of loan words as a result of linguistic contact can be seen.

4.2. Absentive use of HAVE Albeit briefly, we wish to draw attention to another use of the verb HAVE which we consider to be extremely interesting. In its transitive use, this verb can also acquire an absentive value; it indicates that an entity is in a different place to that in which the communicative act is taking place (i.e. the place where the speaker and/or the addressee are). In this case, the sentence expresses locative content, but it is really encoded by means of a possessive construction: the located entity (prototypically, with the semantic feature [+human being]) appears as the first argument of the verb HAVE, and the location as the second argument with the semantic function [patient]: (22)

MY SISTER UNIVERSITY HAVE 'My sister is (off) at the university.' (lit: 'my sister has got university.') [Context: neither the speaker nor the addressee are at the university]

De Groot (2000) has shown that the absentive content is grammatically relevant in many OLs. Nevertheless, while in languages like Hungarian and Italian the said content is expressed by means of a non-verbal predication with a copula, in LSE, as we have seen, it is expressed by a verbal predication. Although this question would require a more detailed study, which is

Non-verbal predicability and copula support rule

297

beyond the scope of this paper, we think that the LSE data are a clear indication that the LEX1 strategy acts not only as an alternative for presentative non-verbal predication, but also for the absentive one.

4.3. Non-presentative possessive contents We have already seen how possessive contents of a presentative nature remain outside non-verbal predicability. Now, we will look at the situation of possessive ascription when it has a non-presentative character. Here we no longer find the verb HAVE, but another sign, which may be considered as a noun, and which corresponds to the English word owner. (23)

a. DOG OWNER,,/,,. MY FATHER b. DOG OWNER WHOy,„ MY FATHER

We may translate both sentences into English by means of a verbal predication (the dog belongs to my father) or a non-verbal ascriptive predication (the dog is my father's). But, in reality, (23a) and (23b) are neither verbal predications nor non-verbal ascriptive ones. They are both codified as a non-verbal equative predication, expressing specification (the dog's owner is my father). The predicate is a term with a nominal head (OWNER) and a restrictor (the possessed entity). The argument term refers to the possessor entity. We find the same expression formats that we saw in section 3.2: a question-answer construction, with (23b) or without a w/z-pronoun (23a). According to these data, we must conclude that, in LSE, there is no ascriptive possessive predication. Of course, LSE users can transmit this kind of meaning, but they do so in an alternative way: they construe an equative predication expressing specification. This fact permits us to reach various remarks. On the one hand, we can confirm that, for possessive contents, presentative and non-presentative predications maintain clear grammatical differences, which tallies with what has been confirmed in many OLs. On the other hand, it may be observed that possession is a content which remains outside the limits of nonverbal predicability in LSE. In other words, in LSE possessive terms cannot be applied predicatively. In a presentative sentence, the possessive term is codified as an argument of a predication with a verb head (HAVE); in a non-presentative sentence, it is codified as an argument of a predication with a noun head (OWNER). Finally, we may also conclude that the con-

298

Angel Herrero-Blanco and Ventura Salazar-Garcia

structions which we have analysed up to now fulfil entirely the predictions derived from NPH (8b). The possessive contents are those that appear furthest to the right within the hierarchy, and are, therefore, those which have the smallest possibility of being expressed by means of non-verbal predications. In LSE, the 'cut-off point' for NPH is situated to the right of nouns. Nouns are admitted as non-verbal predicates, as we saw in 3.1. But possessive constituents do not have this possibility; so they must, by necessity, be linked to a verbal predicate or to an alternative non-verbal predicate. Lastly, it should be noticed that at this point LSE diverges from the typological pattern of Spanish, which is the oral language with which it maintains a greater sociolinguistic contact. In presentative predications there is, in fact, a high degree of overlap, since both languages make use of the same strategies: LEX1 for possessive contents and LEX2 for locative and existential contents. Moreover, we have already seen the more than probable etymological link between the verbs haber (Spanish) and HAVE (LSE). In non-presentative possessive predication, this typological correspondence does not exist. Contrary to LSE, Spanish completes NPH, and admits the existence of possessive non-verbal predicates (cf Hengeveld 1986, 1991). All this proves that no 'typological dependence' can be established between LSE and Spanish. It shows that it would be a serious error to think that LSE is a merely 'Spanish on the hands'; i.e. a mechanism for articulating an oral language (in this case, Spanish) in a visual-gestural manner.

5.

Non-presentative locative predications

5.1. Direct vs. indirect deixis We have left the analysis of non-presentative locative predications until the end, since, due to their special characteristics, they merit detailed attention. For this purpose, three classes of locative predicates should be established. In the first place we must distinguish between deictic predicates, represented by demonstrative adverbs (here', 'there'), and non-deictic predicates, which are terms with a nominal head ('in Amsterdam'). Within the deictic predicates we should also make a distinction between direct deixis and indirect deixis. Direct deixis concerns a physical space which forms part of the setting where the communicative act takes place, and which can normally be perceived visually by the speaker and/or the addressee. In SLs,

Non-verbal predicability and copula support rule

299

this communicative setting also includes constructed actions: narrative procedures to introduce absent entities and play their roles, blending the situational space with a narrative one (cf Liddell & Metzger 1998: 665673). Indirect deixis refers to a space outside the communication setting. These distinctions imply important grammatical consequences for LSE. In order to identify a place by means of direct deixis, LSE opts to point it out directly: the index finger of the right hand points towards the physical space that we wish to identify. It is, therefore, a fixed manual sign in its configuration, but variable, due to pragmatic conditioning, in terms of its orientation. The translation of such sign into an oral language, such as English, would depend on the greater or lesser proximity of the location in question. We will use HERE(d) and THERE(d), where (d) indicates precisely the direct nature of the deixis. For indirect deixis, LSE uses a single sign: the index finger of the right hand moves diagonally, from right to left away from the body. This movement takes place approximately at head level. Here we will transcribe this sign as THERE(i), where (i) indicates that it is indirect deixis. This said, we can indicate that a locative predication with a deictic predicate behaves, in LSE, in a similar way to nominal and adjectival ascriptive predications. That is, the absence of a copula gives rise to the zero1 strategy. This applies to both direct and indirect deixis: (24)

a. MY BOOK HERE(d) 'My book is here.' b. MY MOTHER THERE(i) END 'My mother is not there any more.' [Context: the location alluded to by THERE(i) is outside the setting of the communicative act]

5.2. Non-deictic predicates When the predicate is a non-deictic term, then the construction changes in a very striking way. The typical sign of indirect deixis, THERE(i), in spite of not having here any referential function, must appear after the predicate. This is valid for any kind of argument: a first-order entity (individual; human or non-human), or a second-order entity (event):

300

Angel Herrero-Blanco and Ventura Salazar-Garcia

(25)

a. MY FRIEND ALICANTE THERE(i) 'My friend is in Alicante.' b. CAR GARAGE THERE(i) 'The car is in the garage.' c. MEETING OFFICE THERE(i) 'The meeting is in the office.'

The first observation that can be made as a result of this data is that the sign THERE(i) is seen in (25) to be semantically empty, or at least redundant. The same does not hold for (24b), where the deictic predicates have a clear referential value and, therefore, are semantically necessary. Otherwise the location would have remained unspecified. In (25) the situation is quite different. The location is specified by means of a lexical sign (ALICANTE, GARAGE, and OFFICE, respectively) which acts as a predicate head. The sign THERE(i) does not act here as a locative predicate, but as an element which permits the lexical sign to function as a relational predicate. By virtue of this fact, we believe that there are reasons to state that in locative predications with a non-deictic predicate, THERE(i) has undergone a process of grammaticalization by which it has stopped being a deictic element to become a non-verbal predicativizing copula. It is important to underline the fact that this process of grammaticalization only affects the typical sign of indirect deixis. The demonstrative adverbs used in direct deixis never lose their referential value. When a direct deictic adverb occurs in final position, preceded by a noun, this noun is necessarily interpreted as an argument, not a predicate. This occurs even if the noun has a spatial meaning (e.g. OFFICE). If preceded by another nominal constituent (e.g. BOSS), this should be interpreted as a restrictor of the argument. In any case, the deictic sign continues to act as the predicate of the sentence: (26)

a. BOSS OFFICE THERE(d) 'The boss's office is there.' [Not: 'the boss is in the office.'] b. BOSS OFFICE THERE(i) 'The boss is in the office.' [Not: 'the boss's office is there.']

There are other data which support the hypothesis that in locative predication with a non-deictic predicate, such as (25), the sign THERE(i) acts as a copula. For instance, it can be confirmed that this sign enters into paradig-

Non-verbal predicability and copula support rule

301

matic relationship with verbs that incorporate in their predication a term with semantic function [location]: (27)

a. MY FRIEND GARAGE THERE(i) 'My friend is in the garage.' b. MY FRIEND GARAGE STAY 'My friend stays in the garage.' c. MY FRIEND GARAGE WORK 'My friend works in the garage.'

This circumstance, moreover, invalidates a possible alternative hypothesis: to suggest that THERE(i) could be something like a 'pronominal copy' of the locative predicate. It is true that the pronominal copy is a grammatical phenomenon which appears quite frequently in many SLs. But in LSE the situation is very different. In this language we have barely been able to document the existence of this phenomenon. Furthermore, the appearance of the copy normally generates ungrammaticality. 22 The verbal predications exemplified in (27b) and (27c) do not admit such a copy, and there is therefore no justification to suggest a different situation for (27a): (28)

a. MY FRIEND GARAGE STAY *THERE(i) 'My friend stays in the garage (*there).' b. MY FRIEND GARAGE WORK *THERE(i) 'My friend works in the garage (*there).'

Another remark that we must incidentally make about these examples is related to the status of the sign that we have transcribed here as STAY. Some studies on LSE (e.g. Rodriguez Gonzalez 1992: 328) put forward the idea of considering this sign, in locative constructions, as a copula, with a similar behaviour to the Spanish verb estar. Nevertheless, we have observed that this sign is not semantically empty, since it offers, at least, an Aktionsart of marked continuative meaning. In fact, (27a) and (27b) are not synonymous. This makes us think that it is really a verbal predicate, not a copula. So it cannot be considered as equivalent to the English be (its translation as 'stay' seems to be much more accurate), and, obviously, it cannot be identified with the Spanish estar. A further argument in favour of the copulative character of THERE(i), in non-presentative locative predications, is provided by the markers which usually appear in post-verbal position: negative polarity, some aspectual

302

Angel Herrero-Blanco and Ventura Salazar-Garcia

indications, and indicators of temporal duration. Their behaviour in the presence of THERE(i) is completely analogous to that maintained in the presence of verbal and non-verbal predicates. That is, THERE(i) behaves, syntactically, like a verb, because it adopts the formal characteristics of a predicate. Nevertheless, it doesn't provide any semantic content to the sentence, because such a content is lexically expressed by a locative constituent. Therefore, we must conclude that, in this construction, THERE(i) functions as a copula. (29)

a. MY MOTHER ALICANTE THERE(i) NO 'My mother is not in Alicante.' b. MY MOTHER ALICANTE WORK NO 'My mother doesn't work in Alicante.'

(30)

a. MY MOTHER ALICANTE THERE(i) END 'My mother is not in Alicante anymore.' b. MY MOTHER ALICANTE WORK END 'My mother doesn't work in Alicante anymore.'

(31)

a. MY MOTHER ALICANTE THERE(i) FOR-TWO-YEARS 'My mother has been in Alicante for two years.' b. MY MOTHER ALICANTE WORK FOR-TWO-YEARS 'My mother has worked in Alicante for two years.'

Finally, it should be noted that THERE(i), in its non-deictic use, also admits the presence of an adjective as secondary predication, co-referential with the argument of the predication. Here, also, the syntactic behaviour is totally analogous to that of verbal predications: (32) a. MY FATHER MADRID THERE(i) SAD 'My father is sad in Madrid.' b. MY FATHER MADRID WORK SAD 'My father works sadly [lit: sad] in Madrid.' Example (32a) must be considered as a locative construction, not adjectival, since it is the locative term (MADRID) which acts as the main predicate. Nevertheless, it is not too much to think that, in the future development of LSE, the adjective may change to take on the function of main predicate, relegating the locative term to a satellite function. This is a proc-

Non-verbalpredicability and copula support rule

303

ess which seems to have taken place in some oral languages, such as Spanish (cf Hengeveld 1991; Salazar 2002), and has given rise to a gradual broadening, in the guidelines laid down by the NPH, of the copulative uses of estar. If the same were to occur in LSE (which is something we are unable to foresee at present), it could be a means of broadening the copulative uses of THERE(i) towards adjectival predication, in a way which is totally congruent with oral languages. Without denying that these last observations are quite speculative, we cannot fail to underline the fact that they are highly thought-provoking. Hengeveld (1992b: 208-211) found some OLs in which the deictic or non-deictic condition of the predicate is closely related to the presence of a copula. So this aspect does not question the typological congruity of LSE with OLs. In Tamil, Vietnamese, and West Greenlandic, the copula is optional in locative predications when the predicate is a deictic term; otherwise it is obligatory. In !Xu and Thai, an analogous phenomenon is observed in equative predications. The situation in LSE is not identical, but the differences can be explained on the basis of some typological factors, not directly related with the visual-gestural articulation. We must notice that the said OLs are all rigid, without adjectives (or just with a few of them). They normally use, as an alternative to copula, a zero-2 strategy. LSE, on the contrary, is a specialized language which uses the zero-1 strategy. In any case, the presence or absence of copula in LSE is consistent with the predictions derived from the notion of deictic centre (cf Comrie 1985; Dik 1997, vol. I: 40). 24 That is to say, the lack of a copula occurs in locative predications which are directly linked to the parameters of the communicative situation. In the light of all the facts stated throughout this article, we believe that everything points towards the idea that THERE(i), when it is grammaticalized and has lost its deictic nature, must be considered as a predicativizing copula with the appearance of a demonstrative adverb. Its use is confined to non-verbal ascriptive predications, with non-presentative locative meaning and without direct spatial deixis. One might criticize this conclusion on the basis of the fact that some of our informants accepted as grammatical the expression of these contents by means of a zero-1 strategy, devoid of this sign (Herrero & Salazar 2002). Nevertheless, this could be explained on the basis that the evolutionary process from deictic sign to predicativizing copula, despite having reached a very advanced stage, is as yet incomplete. And this would lead to a certain margin for sociolinguistic variation. After all, the presence of this copula was fully accepted by all the infor-

304

Angel Herrero-Blanco and Ventura Salazar-Garcia

mants, while the alternative without a copula was only endorsed by a minority that did not demonstrate an unequivocal preference for it.

5.3. A copula support rule for Spanish Sign Language In accordance with all we have stated in the previous section, we hold that the LSE sign which we transcribe as THERE(i), in non-presentative locative predications with a non-deictic predicate, functions as a predicativizing copula. For that reason, following the formalization conventions of FG, we are going to present in this section the expression rule which gives rise to the appearance of the said copula. The 'THERE(i) support rule' can be formulated as follows: (33)

INPUT:

[ττ2 e,: (ττ, {(ω χ,: φ, ( Xi ): ... : φ* ( x D W (xj)o σι) a 2 (e)] CONDITION: φ, d e i c t i c DmAdv OUTPUT: [τr2 e,: (ττ, THERE(i) DmAdv {(ω χ,: φι ( Xi ): ... Ά (xO)ioc} (xj)0 σι) ^2(e)]

The input of this rule consists of a predication formed by a relational predicate and a single argument. The predicate consists of a term, with semantic function [location], used predicatively. The argument (xj) is a term which lacks a specific semantic function. This input gives us information about some of the necessary requisites for the correct application of the rule. The non-verbal nature of the predicate has already been specified by its term structure. The non-presentative content of the predication is inferred from the fact that the term with semantic function [location] acts as a predicate. In presentative predication it acts as the argument of the verbal predicate HAVE. As a result, we consider that, in this format of the THERE(i) support rule, it is only necessary to specify one condition: the predicate head φι must not be recognized as a deictic demonstrative adverb. The resultant output maintains the previous structure, with the addition of the copula THERE(i) (originally a demonstrative adverb: DmAdv). This copula serves to carry certain operators to express aspect, polarity, etc., which, either as bound morphemes or free morphemes, must necessarily appear linked to a predicativizing element. The fact that this copula possesses an adverbial character is not an impediment in this respect, since, as Hengeveld (1992b: 188f) has shown, there are predicativizing copulas

Non-verbal predicability and copula support rule

305

other than verbal ones. In the cases where the copula is absent, the zero-1 strategy allows the non-verbal predicate to act in an analogous way to a verb. Similarly, a non-verbal predicativizing copula is capable of carrying (prototypically) verbal distinctions. We cannot determine exactly the place occupied by the THERE(i) support rule within the general sequence of application of the expression rules. To do so, it would be necessary to make a general study of the expression mechanisms used by LSE, which is beyond the scope of this paper. Nevertheless, the so-called 'dynamic model', proposed by Bakker (2001), now permits us to establish the first clarifications in this field, from which we can infer its relative position with regard to other rules. In the linear chain of expression, THERE(i) always occupies a position after the locative predicate. Consequently, and in order to respect the left-to-right principle (Bakker 2001: 33), the copula support rule must intervene after the rules which determine the final form of the predicate. Moreover, it should be recognized that the introduction of this support rule must necessarily precede the replacive rules intended to express the operators linked to the copula. Otherwise, such a copula would not be necessary. In the last place, the application of this rule also has to be situated in a phase prior to the expression of secondary predication, if this applies (cf. 32a), and the rules aimed at identifying certain pragmatic functions; in particular, the New Focus (questioning or completive), that appear after the copula. This is valid for the marked constructions which we have already studied (special word order and question-answer construction), and it can be extended to other analogous structures. This can be seen in the following example, which constitutes a true question: (34) VALENCIA THERE(i) WHO, 'Who is in Valencia?'

6.

Conclusions

In accordance with what has been stated throughout this paper, we can pinpoint the following conclusions. A) LSE has non-verbal predication with a nominal predicate. It is noncopulative predication, with the zero-1 strategy, appropriate to unmarked syntactic patterns. The nominal predicate can have an ascriptive interpreta-

306

Angel Herrero-Blanco and Ventura Salazar-Garcia

tion or an equative one (expressing characterization). In the former case, it would be a bare predicate; in the latter, a referential predicate. We have not found at the moment any formal procedure to differentiate exactly between the alternatives. So, nominal predication seems to be normally, in itself, ambiguous. It is the context which must provide the necessary clues for the correct interpretation. We may represent these predications as follows: (35)

a. Nominal ascriptivepredications (e: [(f,: PredN (f,)) (Xi)0](e)) b. Nominal equative predications, expressing characterization (e: [{(x,: PredN ( Xi )) 0 } (xj)0] (e))

B) LSE also has non-verbal predication with an adjectival predicate. Its characteristics are very similar to those of nominal predication, since it also responds to a zero-1 strategy, with an unmarked construction. The ascriptive interpretation depends on considering the adjective as a bare predicate. The equative interpretation occurs if the adjective acts as a restrictor or an elliptical noun, recognizable by the context. As in the case of nominal predication, we conclude in a provisory way that there is no formal procedure to distinguish the two possibilities (equative and ascriptive). The disambiguation is possible thanks to the context. (36)

a. Adjectival ascriptive predications (e: [(f,: PredA (f,)) (xO0](e)) b. 'Adjectival' equative predications (e: [{(Xi: Pred(0) N (Xi): PredA (xO)0} (xj)0] (e))

C) Equative predication that expresses specification also bears witness to the absence of a copula with zero-1 strategy. But, in this case, the expression adopts a clearly marked syntactic option: a question-answer construction, with or without a w/z-pronoun. This construction, which can be considered as a kind of clefting, is also possible in verbal predications, with a clear pragmatic meaning. For this reason, we defend the hypothesis that, in LSE, specification is by necessity associated in its conditions of use with the assignment of the pragmatic function Focus (prototypically, completive Focus) to the argument of the predication. (37) Nominal equative predications, expressing specification (e: [{(Xj: PredN (x;))0} (xi)0fOC] (e))

Non-verbalpredicability

and copula support rule

307

D) As regards ascriptive presentative contents, we observe that they remain outside the non-verbal predicability of LSE. Both existential-locative constructions and possessive ones have a verbal predicate (HAVE) which acts as the head of the respective predications. The difference lies in the fact that HAVE, in locative and existential sentences, behaves like a pseudo-transitive verb, whereas in possessive sentences it behaves as a transitive verb. The verb HAVE is never used in non-presentative predications. This confirms that, in LSE, the distinction between presentative and non-presentative contents is grammatically relevant. For ascriptive nonpresentative sentences with possessive meaning, LSE has a noun predicate which can be transcribed as OWNER. This predicate acts as a head of a real equative predication, expressing specification. The possessed entity acts as a restrictor of the noun head. Therefore, possessive contents remain outside non-verbal predicability in LSE. This is in keeping with the predictions derived from the NPH, where possession is the element situated furthest to the right. (38) a. Presentative locative predications (e: [(f,: HAVE v (fi)) (xOPat (Xj)ioc](e)) b. Presentative possessive predications (e: [(f,: HAVE v (f,)) ( Xi ) 0 (Xj)pat](e)) c. Non-presentative possessive predications (e: [{( XI : O W N E R n ( X j ): Xj-: Pred N ( X j )) 0 } ( x . W ] (e))

E) The expression format of non-presentative locative contents depends on the deictic or non-deictic nature of the predicate. Deictic predicates (which also occur in constructed actions) display a zero-1 strategy, essentially similar to that which appears in ascriptive nominal and adjectival predications. Non-deictic predicates are followed by the sign THERE(i), a demonstrative adverb which, in this context, has lost its deictic value. In these pages, we have shown the arguments that support the idea that this sign functions here as a non-verbal predicativizing copula, and we have presented the rule which governs its appearance. This is the only copula that we have documented in LSE, and it is in full accordance with the predictions coming from the NPH. Given that the locative contents are those which are situated furthest to the left of the said hierarchy, they are also those that can most easily adopt a copula as an element of support.

308

Angel Herrero-Blanco and Ventura

Salazar-Garcia

(39) a. Non-presentative locative predications, with a deictic predicate (e: [{(Xi: DmAdv (Xj))loc} ( Xj ) 0 ] (e)) b. Non-presentative locative predications, with a non-deictic predicate (e: [{(x;: PredN (Xi))ioc} (xj)0] (e)) These conclusions can be summarized by means of the following table: Table 1. Summary of conclusions Equative

Characterization

Specification

-VP - COP(Ol) - Marked

-VP - COP(Ol) +Marked

Ascriptive

Loc

Presentative

+ VP: HAVE (LEX2)

Nonpresentative

-VP -COP [+ deictic] + COP [- deictic]

A

Ν

Poss + VP: HAVE (LEX1)

-VP -COP(Ol) - Marked

-VP -COP(Ol) -Marked

0 [codified as equative (specification): -VP: OWNER -COP(Ol) +Marked]

From all the above-mentioned, we can say that the initial hypotheses included in (10) have been confirmed and extended in the light of the data provided by LSE. It is specially relevant, in our opinion, that those typological regularities found by Hengeveld for non-verbal predications in OLs - which we summarize in (8) - are also valid for LSE. We believe that this constitutes an important argument in favour of the typological congruity of LSE in relation to the OLs.

Non-verbal predicability and copula support rule

309

Notes 1.

This paper is part of the research project BFF2002-01016, funded by the Spanish Ministry of Science and Technology (MCYT). Our thanks to Mary Joan Dunmall for checking the English version of the text. We would also like to thank the anonymous reviewer whose insightful remarks so enriched the final text. Abbreviations used in glosses, translations, and representations, are as follows: 1. Constituent classes: A (adjective); COP (copula); DmAdv (demonstrative adverb); LOC (locative); Ν (noun); POSS (possessive); Pred (predicate); Pred(0) (understood or elliptic predicate); V (verb); VP (verbal predicate); 0 (lack of constituent). 2. Pragmatic functions: foe (Focus). 3. Semantic functions: 0 (zero); loc (location); pat (patient). 4. Layers: e (predication); f (predicate); χ (term); φ (term restrictor). 5. Satellites and operators: 7t1 (any predicate operator); 7t2 (any predication operator); σΐ (any predicate satellite) a l (any predication satellite); ω (any term operator); cont (continuative aspect); f (feminine); m (masculine); nt (neuter); pres (present tense); deel (declarative). Others: lit (literally); 1 (1st person); 3 (3rd person); sg (singular); dat (dative); gen (genitive); nom (nominative); sbj (subject); y/n (prosodic marker for yes/no questions); ? (prosodic marker for wh- questions). LSE signs are written in capital letters. We do not use other marks (nonmanual, index of agreement, etc) to the extent that they are not necessary for our purposes.

2. 3. 4.

It is available in Spanish and LSE at the Biblioteca de Signos website. The history of this theory dates back at least to Meillet (1903: 355-357). In accordance with Lyons (1977: 723) and Hengeveld (1992b: 96ff), we consider the existential content simply a special kind of locative one. Presentative constructions (re-)introduce a referent into the discourse. In nonpresentative constructions, the existence of this referent has first been established by the discourse or the communicative situation (cf Hannay 1985; Hengeveld 1992b: 118-121). In (6b) we select an example in Latin, because in English, as we may see in the translation, presentative possessive content is expressed by means of a verbal predication. So, it is not part of the non-verbal predicability of this language. Example used by Hengeveld (1992b: 186). The equative relationship that we take into consideration here is that which has a characterizing aim (with a value of classification or identification). The equative relationship of a specifying aim, as will be seen later, can be formally recognized.

5.

6. 7.

310 8. 9.

10.

11. 12. 13.

14. 15.

16.

17.

Angel Herrero-Blanco and Ventura Salazar-Garcia About abbreviations used in LSE glosses, see note 1. In the translation, we place the ascriptive interpretation first and the equative one second. In English, the presence of the indefinite article a is obligatory. This is due to the fact that, unlike other OLs, English does not permit an ascriptive construction here; therefore the corresponding content is expressed by means of an equative construction of classification. In languages where the opposition between classification and ascription exists, this is rather subtle but semantically relevant (Dik 1997, vol. I: 203). In some languages, such as Dutch, bare nominal predicates cannot be restricted or modified. But in many others, for example Spanish, that possibility is valid, at least under certain conditions. Spanish admits three possibilities for this case: without article, with indefinite article or with definite article. These possibilities correspond, respectively, to an ascriptive interpretation, an equative interpretation of classification and an equative interpretation of identification: Mi amigo es {0/un/el} medico famoso. For the distinction between specialized, flexible, and rigid languages, cf Hengeveld (1992b: 47-72). Cf. Coseriu (1956) for an analysis of determination from a structuralist perspective. Specification appears when the predicate and argument term are coextensive. Semantically, specificational contents answer the question who/what is x?, whereas characterizational contents answer the question what can you tell me about x? (Hengeveld 1992b: 83). Strictly speaking, also within specification, we might make a distinction between identification and classification. But, for the purposes of this paper, we do not consider it necessary to take such precision into account. For instance, in Thai different copulas are used for specificational and characterizational sentences (Hengeveld 1992b: 83). Normally, the transcription by glosses of LSE represents the non-manual question mark by a line above the word. For typographic reasons, we opt here for underlining. In English, a language with a rather strict word order, the facts point to a difference in the underlying structure. But in Spanish, with a more flexible word order and a larger presence of agreement phenomena, the explanation based on an expression rule is, in our opinion, more satisfactory. In LSE, the order of constituents is quite flexible, and depends to a great extent on pragmatic factors (cf Climent & Herrero 2000: 63). About aspect and aspectuality in FG, cf Dik (1997, vol. I: 221-225).

Non-verbal predicability and copula support rule

311

18. Tense appears as a Theme constituent at the beginning of the discourse, and from then it remains implied whilst no change occurs in it. Modal elements generally occupy the typical position of the pragmatic function Tail. 19. LSE doesn't mark noun-adjective agreement if the locus of adjective is the body (SAD) or the centre of the neutral space (WORK). In other cases, agreement is expressed by means of noun locus (Chapa 2000). Anyway, agreement does not affect our argumentation, and we won't transcribe it. The said agreement also occurs with HAVE, in examples below. 20. The form in the auxiliary use is ha. This is the only morphological difference between the auxiliary and pseudo-transitive uses of haber. 21. Nowadays the configuration of this sign has partially changed in its first segment, but its dactylological origin is still clear. 22. Cf. Herrero & Salazar (2002), for a more detailed explanation of this aspect. 23. Probably, the fact that MADRID acts as the main predicate is not obvious from the English translation. This is because in English the presence of the copula be is rather common in adjectival predications including a locative satellite. But this is not so in LSE. In this language, locative satellites in adjectival predications are normally placed in specific pragmatic positions (prototypically, a Theme position): THERE(i), MY FATHER SAD ('there, my father is sad'); MADRID, MY FATHER SAD ('in Madrid, my father is sad'). In these cases, the co-occurrence of deictic and non-deictic signs is ungrammatical: *MADRID THERE(i), MY FATHER SAD (""in Madrid there, my father is sad'). Therefore, the combined presence of a lexical locative constituent and the deictic sign THERE(i) in (32a) implies the locative nature of the predication, in which the adjective is an additional element. 24. This notion is closely related to some classical contributions by structural linguistics: the analysis of shifters (Jakobson 1957), and the distinction between temps du discours and temps de l'histoire (Benveniste 1950).

References Bakker, Dik 2001 The FG expression rules: a dynamic model. Revista Canaria de Estudios Ingles es 42: 15-54. Benveniste, Emile 1950 La phrase nominale. In Problemes de linguistique generale, vol. 1, Emile Benveniste, 151-167. Paris: Gallimard.

312

Angel Herrero-Blanco and Ventura Salazar-Garcia

Chapa, Carmen 2000 Los usos del espacio en la LSE. In Signolingiiistica. Introduction a la lingüistica de la LSE, 261-274. Valencia: Fesord CV. Climent, Jaime and Angel Herrero-Blanco 2000 La estructura sintäctica. In Signolingiiistica. Introduction a la lingüistica de la LSE, 57-65. Valencia: Fesord CV. Cogill-Koez, Dorothea 2000 Signed language classifier predicates. Linguistic structures or visual representation? Sign Language and Linguistics 3(2), 153-207. Comrie, Bernard 1985 Tense. Cambridge: Cambridge University Press. Coseriu, Eugenio 1956 Determinacion y entorno. In Teoria del lenguajey lingüistica general, 282-323. Madrid: Gredos. Dik, Simon C. 1980 Studies in Functional Grammar, London: Academic Press. 1989 The Theory of Functional Grammar. The Structure of the Clause. Dordrecht: Foris. 1997 The Theory of Functional Grammar. 2 vols. 2nd, revised edition, edited Kees Hengeveld. Berlin: Mouton de Gruyter. Emmorey, Karen 1996 The confluence of space and language in signed languages. In Language and Space, Paul Bloom, Mary A. Peterson, Lynn Nadel and Merrill F. Garret (eds.), 171-209. Cambridge, MA: MIT Press. Groot, Casper de 2000 The absentive. In Tense and aspect in the languages of Europe, Osten Dahl (ed.) 693-719. Berlin: Mouton de Gruyter. Hannay, Mike 1985 English Existentials in Functional Grammar. Dordrecht: Foris. Hengeveld, Kees 1986 Copular verbs in a Functional Grammar of Spanish, Linguistics 24 2: 393-420. 1990a Semantic relations in non-verbal predications. In Layers and levels of representation in language theory: A functional view, Jan Nuyts, A. Machtelt Bolkestein and Co Vet (eds.) 101-122. Amsterdam: John Benjamins. 1990b A functional analysis of copula constructions in Mandarin Chinese. Studies in Language 14 (2): 291-323. 1991 Tipologia, sincronia, diacronia. Foro Hispänico 2: 81-94.

Non-verbal predicability and copula support rule 1992a

313

Non-verbal predicability. In Meaning and Grammar: Crosslinguistic Perspectives, Michel Kefer and Johan van der Auwera (eds.), 77-94. Berlin: Mouton de Gruyter. 1992b Non-verbal predication. Theory, typology, diachrony. Berlin: Mouton de Gruyter. 2004 The architecture of a Functional Discourse Grammar. In A New Architecture for Functional Grammar, J. Lachlan Mackenzie and Maria A. Gomez-Gonzalez (eds.). Berlin: Mouton de Gruyter. Herrero-Blanco, Angel 2000 Historia del lenguaje de signos. In Signolingüistica. Introducciön a la lingiiistica de la LSE, 93-100. Valencia: Fesord CV. 2002 La investigation lingüistica de las lenguas de signos. LynX. Panorama de Estudios Lingüisticos 1: 9-47. 2003 Escritura alfabetica de la lengua de signos Espanola. Once lecciones. Alicante: Universidad de Alicante. Herrero-Blanco, Angel and Juan Jose Alfaro 1999 Fonologia y escritura de la Lengua de Signos Espanola. Estudios de Lingiiistica 13, 89-116. Herrero-Blanco, Angel and Ventura Salazar-Garcia 2002 Copula y predicaciones no verbales en la LSE. In Estudios lingüisticos sobre la Lengua de Signos Espanola, Montserrat Veyrat (ed.), 105-144. Valencia: Nau Libres. Jakobson, Roman 1985 Reprint. Los conmutadores, las categories verbales y el verbo ruso. In Ensayos de lingiiistica general, Josep M. Pugol and Jem Cabanes (eds.), 307-332. Barcelona: Planeta-Agostini. Originally published as Shifters, verbal categories, and the Russian verb. Harvard: dept. of Slavic languages and literatures, Harvard university, 1957. Jespersen, Otto 1924 La filosofia de la gramdtica. Barcelona: Anagrama, 1975. Liddell, Scott 1980 American Sign Language syntax. New York: Mouton. 1990 Four functions of a locus: Re-examining the structure of space in ASL. In Sign language research: Theoretical issues, Ceil Lucas (ed.), 176-198. Washington DC: Gallaudet University Press. 1994 A real, surrogate and token space: grammatical consequences in ASL. In Language, gesture and space, Karen Emmorey and Judy Reilly (eds.), 19-41. Hillsdale: Lawrence Erlbaum.

314

Angel Herrero-Blanco and Ventura Salazar-Garcia 2000

Blended spaces and deixis in Sign Language discourse. In Language and Gesture, David McNeill (ed.), 331-358. Cambridge: Cambridge University Press. Liddell, Scott and Melanie Metzger 1998 Gesture in Sign Language discourse. Journal of Pragmatics 30: 657697. Longobardi, Giuseppe 1987 Las oraciones copulativas en la teoria sintactica actual. In Sintaxis de las lenguas romanicas, Violeta Demonte and Marina Fernandez Lagunilla (eds.), 233-251. Madrid: El Arquero. Lyons, John 1977 Semantics. 2 vols. Cambridge: Cambridge University Press. Meier, Richard P., Kearsy Cormier, David Quinto-Pozos (eds.) 2002 Modality and Structure in Signed and Spoken Languages. Cambridge: Cambridge University Press. Meillet, Antoine 1964 Reprint. Introduction a I 'etude comparative des langues indoeuropeennes. Alabama: University of Alabama Press. Original edition, Paris: Hachette, 1903. Morales Lopez, Esperanza, Cristina Perez Casanova, Cesar Reigosa Varela, Emma Blanco Diaz, Nancy Bobillo Garcia, Cristina Freire Rodriguez, Begona Mallo Novas, Gabriela Prego Vazquez 2000 Aspectos gramaticales de la LSE. In Apuntes de lingüistica de la Lengua de Signos Espanola, Francisco Martinez Sanchez (ed.), 69131. Madrid: CNSE. Munoz Baell, Irma 1999 iComo se articula la lengua de signos espanola?. Madrid: CNSE. Neidle, Carol, Judy Kegl, Dawn MacLaughlin, Benjamin Bahan, Robert G. Lee 2002 The syntax of American Sign Language. Functional categories and hierarchical structure. Cambridge, MA: The MIT Press. Padden, Carol 1990 The relation between space and grammar in ASL morphology. In Sign Language research: Theoretical issues, Ceil Lucas (ed.), 118132. Washington DC: Gallaudet University Press. Pinkster, Harm 1995 Sintaxis y semäntica del latin. Madrid: Ediciones Cläsicas. Rodriguez Gonzalez, Maria Angeles 1992 Lenguaje de signos. Madrid: CNSE & Fundacion ONCE.

Non-verbal predicability and copula support rule

315

Salazar-Garcia, Ventura 2002 La estructura 'Estar de + Ν' en el marco general de las construcciones copulativas del espanol. Language Design 4: 69-93. Supalla, Ted 1986 The classifier system in ASL. In Noun classes and categorization, Colette Craig (ed.), 181-214. Typological Studies in Language 7. Amsterdam: John Benjamins. Valli, Clayton and Ceil Lucas 1992 Linguistics of American Sign Language. Washington DC: Gallaudet University Press. Villabrille, Francisco Fernandez 1851 Diccionario usual de mimica y dactilologia. Madrid: Imprenta del colegio de sordo-mudos y ciegos.

A new view on the semantics and pragmatics of operators of aspect, tense and quantification Annerieke Boland

1.

Introduction1

Operators of aspect, tense and quantification are recurring topics in Functional Grammar (FG). The standard view (Dik 1997) is that in the underlying representation of the clause, operators reflect grammatical expressions. Operators can be described by their semantics and the scope of their modification, that is the part of the utterance they modify. However, there is no consensus on what the scope of specific operators exactly is, where they should be located in the underlying structure of the clause. In my view this is due to the fact that the semantics of the specific categories of operators are not always properly described. Furthermore, it is not always taken into account that the different layers in the underlying clause structure serve distinct communicative functions. In this paper I propose an alternative view on the semantics of aspect and tense, based on Bohnemeyer (1998) and Klein (1994). The semantics determine the communicative function of operators: to which function of the utterance do the operators contribute? It is this communicative function that determines the location of the operators in the underlying representation of the clause. First I will discuss the communicative function of layers and how we should view the role of operators at these layers (section 2). In section 3, the semantics of aspectual operators are described. I will claim that they should all be considered predicate operators, except for quantificational aspect. In section 4, the semantics of tense operators are discussed. I propose a definition that is slightly different from the standard one: it is not the state of affairs that is located in time; it is the part of the state of affairs that is relevant to the discourse that is located in time. In section 5, I will discuss the distinction between property- and event-quantification, following Anstey (2002). I will also propose a change in the definition of the habitual.

318

A nnerieke Boland

This paper focuses on English. Examples have been collected from different sources: from Collins' Wordbanks Online English Corpus (CWO), from the worldwide web (WWW), and from Biber et al. (1999) (indicated by a Β and page number), who made use of the Longman Spoken and Written English Corpus. Examples that have no reference have been made up by the author.

2. Layers, operators and scope 2.1. Introduction One of the most crucial features of the FG-model is the notion of layers and scope. Although there is a strong relation between the two notions, there is no one-to-one correspondence. Layers are parts within the utterance that serve a certain communicative function: the function of description, of reference and of transferring content. Within each layer, different entities are designated: at the first layer, the predicate designates a property or relation and term structures designate entities. At the second layer, the predication designates a state of affairs (SoA) and at the third layer, the proposition designates the content of the utterance. Layers should be considered as pragmatic and semantic notions, not as structural or syntactic unities. Operators are grammatical modifiers that contribute to the communicative functions of description, reference and transferring content. The scope of an operator is the part of the utterance - or rather, what is designated by that part - that is modified by that operator. Scope is partly related to the layer at which an operator is located in that operators of lower layers fall within the scope of operators of higher layers. However, several modifications may apply at the same time to the same designated entity (property, SoA or proposition) so that operators of the same layer may have scope over each other. Scope relations between operators are thus not an a priori argument for locating those operators at a different layer. The only criterion that should be used for locating operators at a specific layer is their communicative function: what do they contribute to the utterance, semantically and communicatively? This paper is restricted to the discussion of operators of aspect, tense and quantification, located at the first and second layer of the representational level.

Anew view on the semantics and pragmatics of operators

319

My aim will thus be to describe layers and operators in terms of their pragmatics, more specifically their communicative function, and their semantics. The pragmatic and semantic definition will later form the basis for locating specific operators at a certain layer. Syntactic repercussions (order effects) are not considered a crucial criterion for the location of operators, although there may of course be syntactic reflections of underlying semantic or pragmatic relations (for example Bybee 1985, Hengeveld 1989, 1997). In this respect I agree with Anstey (2002:8): 'If one believes that syntax iconically and imperfectly reflects semantics then languages will be expected to reflect the order of operators to various degrees, but not to represent it.'

2.2. The descriptive layer The communicative function of the first layer is building up a proper description of the state of affairs. This is done by several elements: the predicate that designates a property or relation, the term structure(s) that designate entities and, potentially, additional participants or properties such as for instance beneficiary, company, direction or path2 (cf. Dik 1997: 106, based on Verkuyl 1972). Both the lexemes in the argument slots and the lexemes in the predicate slot can be modified by term operators and by predicate operators (7rl)3 and satellites, respectively. Now what exactly is the function of predicate operators and satellites? Dik's definition (1997: 63) is that predicate operators are grammatical operators that in one way or another specify the internal dynamics of the SoA. This may be true for most τ\ -operators, but is not the essential characteristic that distinguishes π1-operators from other elements, since these other elements can also influence the internal dynamics of the SoA. Satellites of direction or location may change the internal dynamics of a SoA as well: the SOAs walk in the park vs. walk to the park differ in telicity. Even the specificity or number of the arguments may change the internal dynamics of a SoA: the SOAs draw a circle and draw circles differ in telicity. The same holds for The demonstrators were passing the station and Demonstrators were passing the station. On the other hand, a predicate operator or satellite may specify the description of the SoA, without interfering with the internal dynamics. I know English vs. / must know English describes the same type of SoA (a state) but the relation between the argument(s) and the designated property is different. Satellites

320

Annerieke Boland

of Manner, Degree, Speed and Duration do not change the internal dynamics of the SoA either but they do specify the SoA. What does distinguish π1-operators and σΐ-satellites from other elements is that they modify the property or relation that is ascribed to the argument(s). The description of this property or relation is somehow made more specific. In Hengeveld's definition (1989:133) first-layer operators contribute to "building up a proper description of the situation the speaker wishes to refer to". This is correct, but it is only a specific part of the SoA description that is influenced by predicate operators and satellites. This can but need not interfere with the internal dynamics of the SoA. Thus on the one hand, the modification of a predicate by a τ\ -operator does not influence the term structure in an argument slot. On the other hand, a term structure is modified without influencing the predicate or the additional participants. The description of a SoA is the result of the combination of all these 'modified' elements. I adhere to Hengeveld's approach (1989) that this description of the SoA cannot be further modified as a whole, contrary to Dik (1997) and Cuvalay (1996). In short, the combination of elements at the first layer - modified arguments, modified additional participants and modified predicate - forms the proper description of the SoA. Each of these elements can have repercussions on the type of SoA, such as the dynamicity, the telicity, control, etc. A predicate operator is an operator that changes the description of the property or relation that is ascribed to the argument(s). It however does not modify the description of the arguments itself.

2.3. The referring layer The second layer designates the SoA. The communicative function of this layer is reference; the speaker links the description of the SoA to a SoA in the real or imaginary world. This constitutes the referring function within an utterance. Predication operators (ττ2) contribute to the referring function, they "relate the description of a SoA to the occurrence of that SoA in a real or imaginary world." (Hengeveld 1989: 134). In Dik's formulation (1997: 218) 7i2 and σ2 elements "leave the internal constitution of the SoA intact, but either quantify it or locate it with respect to spatial, temporal and 'objective' cognitive dimensions." Thus, predication operators do not change anything about what is predicated of the arguments, and do not change the internal dynamics of the SoA.

A new view on the semantics and pragmatics of operators

321

2.4. Summary In sum, the first layer fulfils the function of description within an utterance. Operators that contribute to the description of a SoA by specifying what is predicated of the argument(s) are considered ττΐ-operators and have scope over the predicate only. The second layer fulfils the function of reference within an utterance, of linking the SoA to the real or an imaginary world. Operators that contribute to the linking of a SoA to a real or imaginary world are 7t2-operators. They have scope over the full predication. The definitions of operators as given here are based on the communicative function, not on formal characteristics. The definition of the communicative function of specific operators is thus crucial in determining where they should be located in the underlying clause structure. In order to do so, the semantics need to be properly described.

3. Operators of aspect 3.1. Semantics Aspect is the first semantic domain to be discussed. Dik (1997: 221-222) defines different types of grammatical aspect: perfective/imperfective aspect, phasal aspect and perspectival aspect.4 A survey by Anstey (2002: 3)5 shows that there is no agreement on the location of aspectual categories and he himself adds another proposal. The most important question in settling this issue is which operators contribute to the description of the SoA by modifying the property ascribed to the argument(s) and which operators contribute to referring to the real or imaginary world. In order to answer this question it is important to reconsider the semantics of aspect. I propose an alternative view of the semantics of aspect, following Bohnemeyer (1998), who treats the three aspectual categories as one unified category of 'boundary' operators. According to this view aspectual operators select a specific part of an event or the pre-state or post-state of an event.6 It is only this part of the event that is ascribed to the arguments). Pre- and post-states differ from pre-and post-times: Pre- and post-times are time intervals that precede and follow the event, respectively. By contrast, pre- and post-states are related to the event in some non-temporal way in addition to temporal sequentiality. (...) Post-states reflect some kind of trace that the occurrence of the target event leaves behind

322

Annerieke Boland

in the world. (...) the target event may be the consequence or purpose of the pre-state, or an agent may be said to be in an intentional pre-state with respect to the target event. (...) I assume that pre- and post-states of an event Ε are events which do not form part of E, but encompass Ε in a natural 'causal chain'. (Bohnemeyer 1998: 76-77) An important point to note is that aspectual operators do not select part of the full SoA, designated by the predication, but of the temporal structure that inherently belongs to the property or relation designated by the predicate. A property or relation has extension in time, although this does not mean location in time. It is not always the complete temporal structure that is predicated of the argument(s), it is often only part of the temporal structure that the speaker wants to assert. The selection of part of the temporal structure modifies neither the meaning of the arguments nor that of the additional participants; it only modifies what is predicated of the arguments), what is asserted of them. Of course this has repercussions on the final SoA-description, since the selected part of the temporal structure of the property or relation is still the only part that helps build up the full SoA description. Aspectual operators should thus in my view be defined as follows: by using aspectual operators, a speaker selects different parts of the temporal structure of a property or relation - including the pre- or poststate - to focus on. It is only this selected part of the temporal structure that is the predicated property or relation of the argument(s). To understand the idea of selecting part of the temporal structure, it is helpful to consider a maximal temporal structure of a property or relation. Although different properties or relations should be represented by different structures a very general structure is presented in (1), for example for the relation described by the predicate write:1 (1)

Representation of temporal structure of the unmodified relation or property WRITE (x,) (x2)

The time span represented by the horizontal line could represent a homogenous or a heterogeneous interval, with or without a state change within the boundaries. The length of the time span is not exactly specified in the temporal structure of the unmodified property or relation.

A new view on the semantics and pragmatics of operators

323

The initial and terminal boundary, represented by the left and right vertical lines, can simply be the beginning and endpoint but they can also represent the point of beginning and completion where the predicate is being used in the description of a telic SoA. However, whether the temporal boundaries are points of ending or of completion is determined by the combination of the predicate, the arguments and the additional participants. In the case of states, there are no boundaries if the state is conceived of as permanent, but there are boundaries if the state is conceived of as temporary. States without boundaries cannot be subject to aspectual operations. When aspectual operators are used, the temporal structure is modified. Parts of the temporal structure may be selected, or even added and then selected. Only these selected parts of the temporal structure are ascribed to the argument(s), and nothing more. This will be illustrated below by representations of common aspectual distinctions.8 In these representations, the complete temporal structure is represented by the dotted line. The circle indicates the part of the temporal structure that is selected and ascribed to the argument(s). Besides the selected or focused part, some other parts of the temporal structure are entailed, i.e. they can be inferred from the linguistic expression. The complete part of the temporal structure that is implied is represented by the solid line. This line indicates the part of the temporal structure that is taken into account, which is not necessarily equal to the part of the structure that is selected and ascribed to the argument(s). It is also not necessarily any interval earlier in the time structure. For example, it is part of world knowledge that once an event is going on, it must have begun, but the initial boundary is not made explicit by the progressive, as in (2) below, whereas it is made explicit by the continuative, as in (3) below.

3.2. Phasal aspect The first category to be discussed is the category of phasal aspect. By using a progressive the speaker selects a subinterval of the time span to ascribe to the argument(s), without taking the initial or terminal boundary into account. Consider (2): (2)

Progressive: (WRITING) (x,) (x2)

324

Annerieke Boland

The temporal structure of a property is rather flexible. A predicate that by default describes a permanent state, naughty, may be used to describe a temporary state as well, hence with boundaries in the temporal structure. To indicate this non-default interpretation, a Progressive can be used: being naughty. A predicate that in general describes properties or relations of very short duration may also be used to describe this same property or relation as having a longer duration either because the property is repeated (jump, fire) or because the stage leading up to the state change is stretched (as in reach the top). To indicate this interpretation, a Progressive can be used. Another phasal aspect operator is the continuative. When a speaker uses a continuative operator, he concentrates on a subinterval of the temporal structure, as with a progressive or imperfective, but now the initial boundary is made explicit by the expression. In English, this is expressed by the construction keep -ing? Consider (3): (3)

Continuative: (KEEP WRITING) (x,) (x2)

It expresses that the argument(s) have started and still continue their activity. Examples are presented in (4-5): (4) (5)

She keeps smelling the washing powder. (B746) I don't want to keep living with my mum. (CWO)

Phasal aspect operators like the ingressive and egressive focus on the initial or terminal boundary respectively. In English, these can only be expressed lexically by stop, begin, end, finish, etc. The ingressive and egressive are represented in (6): (6)

Ingressive and egressive (START WRITING) (x,) (x2)

(STOP WRITING) ( Xl ) (x2) /O.

ν ^

A new view on the semantics and pragmatics of operators

325

3.3. (Im)perfective aspect Dik (1997: 221) considers (im)perfective aspect to be different from phasal aspect. I think, however, that (im)perfective aspect has the same communicative function as phasal aspect, although less specific. Operators of (im)perfectivity make a selection of the temporal structure, but this selection is either a subinterval (imperfective) or the entire (perfective) temporal structure. The representation of the selection of an imperfective operator is thus equal to the representation of a progressive or maybe a continuative. The representation of the selection of a perfective operator is presented in (7). A perfective selects the complete temporal structure, including the initial and terminal boundaries: (7)

Perfective (not grammatically marked in English)

Anstey (2002:7) argues that '(im)perfectivity is a prototypical characteristic of events, and not of properties. Only events (and not properties) can be viewed from either an external perspective as wholes (perfective) or an internal one as imperfective.' I adhere, however, to Klein (1994: 28-29), who states that the definition of 'viewing' an event is a metaphor that is not a clear explanation of the function of aspect. The description presented above, that aspectual operators make a selection of the temporal structure and that it is only the selected interval of the temporal structure that is ascribed to the argument(s) makes it clear what the communicative function of aspectual operators actually is. (Im)perfective operators have the same communicative function as phasal aspect operators, but the selections are less fine-grained. They either select a subinterval (imperfective) or the complete structure (perfective) to ascribe to the argument(s).

3.4. Perspectival aspect The communicative function of perspectival aspect - prospective and perfect - is not different in principle from phasal aspect operators or

326

Annerieke Boland

(im)perfective operators. If a speaker uses a prospective, he adds a prestate to the temporal structure.10 Only this 'pre-state' of the event is predicated of the argument(s). So, the argument is not in the state of writing, but only in the state of going to write. Consider (8): (8)

Prospective: (GOING TO WRITE) (x,) (x2)

If a speaker uses a perfect, then a post-state is added to the temporal structure. It is exactly this post-state that is predicated of the argument(s). The perfect is the counterpart of the prospective. Consider (9): (9)

Perfect: (HAVE WRITTEN) (x,) (x2)

The post-state of an event starts immediately after the terminal boundary of an event, but it may stretch in time for a long period. Note that a simple perfect makes the complete temporal structure explicit. It is therefore possible to get the so-called continuative perfect reading as in He has lived here for ten years or I have already run five miles. However, although the preceding event is implied, the focus is on the poststate, since the 'having lived somewhere for ten years' is only true in the post-state of this event (see also Klein (1994: 113) for a similar analysis). Smith (1991: 148) claims as well that we should consider the post-state selected by the perfect to be the modified property ascribed to the argument, by saying: "Present perfect sentences ascribe to their subjects a property that results from their participation in the prior situation. If at some time Henry has laughed, danced, built a sandcastle, the property of having done these things is asserted of Henry." (italics mine). In Bohnemeyer's words (1998: 77): 'Pre-and post-state operators select semantic components of the causal chain in which the event is embedded, in the same manner in which the other boundary operators select parts of the event itself.'

A new view on the semantics and pragmatics of operators

327

It may be clear that I disagree with Dik's classification of perspectival aspect as second-layer operators. He defines perspectival aspect distinctions as the way in which the SoA is viewed from an external point in time: 'What can be said on the basis of information available at some reference point tj about the occurrence of some SoA at some interval tj?' (Dik 1997: 239). Dik clearly treats the pre-state and post-state as a pre-time and posttime and not as part of the temporal structure. However, both the prospective and the perfect change the internal temporal structure of the predicate and it is only the pre- or post-state that is predicated of the argument(s). They thus fulfil exactly the function of πΐ-operators: i) they modify the property ascribed to the argument(s) since they only ascribe the pre-state of writing and not writing in its entirety or the post-state of writing and not writing in its entirety; ii) they do not change the selection restrictions of the predicate; iii) they do not in any way link the SoA to the real or imaginary world. Although there is a temporal relation between the pre- or post-state and the event (the event took place before the post-state or the event will take place after the pre-state), this temporal relation is only an implication and not part of the meaning. When this implication becomes more and more an essential part of the meaning (as happened to a more or lesser degree in German and Dutch, cf. Boogaart 1999: 156) then it grammaticalizes into a tense marker. The fact that a perfect or prospective can have scope over a progressive in English is no argument for positioning them at a different layer. The combination of a perfect with a progressive leads to a very sophisticated specification of what is predicated of the argument, that is: the argument is in a post-state of a certain activity, without the implication that this activity has ended. Consider (10): (10) Perfect progressive: (HAVE BEEN WRITING) (x,) (x2)

Note that the terminal boundary is not implied by this expression. So, John has been writing a letter does not imply that John has finished writing the letter, whereas John has written a letter does imply that the letter is finished. The choice of this complex aspectual expression does not change anything about the meaning of the arguments themselves and it does not link the arguments to the real world, neither in time nor in space.

328

Annerieke Boland

The opposite can also be expressed: John is going to be writing a letter. In that case, the pre-state of the activity of writing is ascribed to the arguments), without making explicit the boundaries of the temporal structure. Consider (11): (11)

Prospective progressive: (GOING TO BE WRITING) (x,) (x2)

It is also possible to combine a perfect or prospective with an ingressive or an egressive. Consider (12) for the representation of a prospective ingressive: (12)

Prospective ingressive: (GOING TO START WRITING) (x,) (x2)

So, perspectival aspect can have other aspectual operators in its scope, but this is not considered an argument for locating these operators at different layers. The communicative function is equal for all aspectual operators, that is, they select the part of the temporal structure that is ascribed to the argument(s). The combination of a perspectival operator and another aspectual operator results in a more complex selection of the temporal structure, with different parts of the structure taken into account. A final argument for treating perspectival aspect and phasal aspect as belonging to the same layer is presented by Hengeveld (1989: 134), based on Steedman (1977). He shows that both perspectival aspect and phasal aspect interfere with the internal dynamics of the SoA. And although this is not the crucial characteristic of a πΐ-operator, once an operator does change the internal dynamics, it must be part of the first layer. When a dynamic SoA, such as: (13)

(f: run (f)) (x:John (x))Ag

is combined with a progressive, a prospective or a perfect, it results in a derived stative reading. We could say that John is no longer a participant in a dynamic SoA, but in a derived stative SoA. This can be shown by the fact

A new view on the semantics and pragmatics of operators

329

that SOAs modified by a prospective, perfect or progressive cannot occur in pseudo-cleft constructions of the following type, which would be possible with dynamic SoAs (Steedman 1977: 221): (14)

What he did was run / *be running/ *be going to run/ *have run

This shows that a predicate that normally describes a dynamic SoA, once modified by a progressive, perfect or prospective, interferes with the internal dynamics of the SoA. This holds equally for phasal aspect and for perspectival aspect. The same conclusion follows from the combination with the modal auxiliary must. When must is combined with a dynamic SoA, it yields a deontic interpretation, whereas the combination with a Stative SoA usually yields an epistemic interpretation (or a general deontic interpretation that holds for a generic or non-specific argument, as in you must have a MA to enter the programme). Consider (15-16) (Steedman 1977: 228): (15) (16)

must + [+dyn]SoA :He must leave now. > deontic must + [-dynjSoA: He must be the chief plumber. > epistemic

When must is combined with a SoA in which the predicate is qualified by a progressive, prospective or perfect, then the only possible interpretation of must is epistemic, which again proves that the aspectual marker has interfered with the type of SoA: (17)

They must be running / be going to run / have run > epistemic

So the progressive, the prospective and the perfect behave in a similar way: they select a part of the temporal structure of the property or relation and only this part is ascribed to the argument(s). This semantic analysis is supported by the fact that they interfere with the internal dynamics of a SoA: the modified predicate is part of a derived stative SoA. Perspectival and phasal aspect should thus be treated as the same kind of operators, located at the same layer.

330

Annerieke Boland

3.5. Summary In sum, I define aspect as a grammatical means to select a specific part of the temporal structure of a property or relation to be ascribed to the argument(s). Selecting the pre-state or post-state of a property or relation does not differ in principle from selecting one of the boundaries, a subinterval or the complete temporal structure. Therefore, the categories of perspectival aspect, (im)perfective and phasal aspect all have the same communicative function: they all select part of the temporal structure of the property or relation that is designated by the predicate. All aspectual markers modify the property designated by the predicate, since the use of the operator modifies what is predicated of the participant. It restricts or defines the part of the temporal extension that is ascribed to the argument(s). The fact that aspectual operators may have scope over each other (perspectival aspect over other aspectual markers) is no reason for locating them at different layers. The combination of a Perfect and Progressive marker in English still functions as a selection of the time structure, albeit a more complex selection than a bare Perfect or a bare Progressive. The communicative function of a predicate operator as stated in section 2 is that it changes the description of the property or relation that is ascribed to the argument(s). It should however not modify the description of the argument itself, not change the selection restrictions on the arguments or on additional participants (since they fall outside the scope of the operator), and not refer to the real or imaginary world. If aspect is defined as the selection of part of the temporal structure, then it can only belong to the first layer and all aspectual markers should be considered 7rl -operators. It must be stressed that it is language-specific which aspectual categories are marked by grammatical means. In English, for instance, lexical verbs such as start or finish are needed to describe an initial or terminal boundary of an event; in fact, in English these verbs describe a SoA by themselves, the SoA of starting or finishing (something). Note that simple tenses may imply, but do not encode, a perfective or imperfective viewpoint. I return to this matter in the next section.

A new view on the semantics and pragmatics of operators

331

4. Operators of tense 4.1. Semantics The second semantic domain to be discussed is the domain of tense. The standard view on tense (Comrie, 1985) is that the time of the event Ε is located with respect to the time of speech S or another reference point in time (R). Ε can be either simultaneous with (present), before (past) or after (future) S (or R). The FG view on tense (Dik 1997: 237) closely resembles this idea: 'Tense operators serve to locate the SoA on the time axis in relation to some t r .' The moment tr can represent the moment of speech (absolute tense) or another moment in time (relative tense). In spite of the generally shared view on the semantics of tense, I want to propose a slight change, based on proposals by Klein. Klein (1994: 22-23) has shown that the standard view cannot account for some common phenomena. First of all, a past tense marker can be used for a state that still holds at the moment of speech, such as It was a beautiful book, but it was in Japanese. On the other hand, a future marker can be used for a state that already holds at the moment of speech. In answer to Will you be here at eight? one may answer Yes, I will be here. In both cases, it cannot be maintained that the (entire) SoA is located in time, since the SoA also holds in the present. Klein solves this problem by the notion of topic time, that is 'the time span to which the speaker's claim on this occasion is confined' (1994: 4). In his view, tense does not locate the SoA on the time line, but at the topic time. Aspectual operators serve to link the time of the SoA to the topic time interval11 and the topic time interval is located on the temporal axis by tense markers. I agree with Klein that tense does not necessarily locate the entire SoA in time. I also agree with Klein that there is something like a topic time, a time span about which the speaker is asserting something. However, I do not agree with his view that tense locates the topic time interval on the temporal axis. In my view, topic time is not something that is linguistically marked but there are several cues that language users can rely on to infer the topic time, such as tense markers, temporal adverbial phrases, world knowledge about the sequence of events, etc. The relation between the topic time interval and the temporal structure of a property or relation is crucial. I think that tense does locate the SoA in time, but only the part that is relevant to topic time.12 Speaker and addressee mutually understand what is the relevant time span about which the speaker asserts something.

332

Annerieke Boland

It is because of the maxims of Grice (1975) that tense marking only concerns the part of the SoA that is relevant to topic time and not the complete SoA. If a speaker selects a part of the temporal structure by means of aspectual operators, then this indicates that only that specific part of the temporal structure is relevant at topic time, and that it is only that specific part of the temporal structure that is ascribed to the argument(s) for the time interval about which the speaker wants to assert something. If no selection is made of the temporal structure, then the temporal structure of the property or relation determines what is the part relevant to topic time. In cases of a temporal structure with boundaries, it can be assumed that the complete SoA is relevant to topic time and that the Speaker wants to ascribe the entire temporal structure to the argument(s). In cases of a temporal structure without boundaries (properties conceived of as permanent states) only the part of the state that overlaps with topic time is located in time; it is a matter of pragmatic inference that not the complete state is located in time. No selection of the temporal structure is made through aspectual operators, but locating the entire SoA is virtually impossible (or at least pragmatically very inconvenient) since it holds anterior to, simultaneous with and posterior to the topic time. And continuously marking this fact (/ knew, know and will know her very well) is not very relevant in conversation. Language users know that once a permanent state holds for a certain topic time, it will also hold for other topic times. So, only the part of the State that is relevant to the topic time of the ongoing discourse needs to be located in time. In this view, the problematic examples of Klein can still be accounted for. With respect to states for example, this view still explains why it is possible to use a past tense for a Stative SoA that still holds at the moment of speaking. Consider (18): (18)

He didn 't come to the party. He was ill.

In the light of his not coming to the party in the past, it is only relevant to talk about the interval of the state of his being ill that held at the same time interval as his not coming to the party, although he might still be ill. So, if the standard view on tense is slightly modified, it could account for the problems that Klein has pointed out. I therefore propose to define the function of tense operators as follows: 'Tense operators locate the part of the SoA that is relevant to the discourse on the time axis in relation to the speech time interval.' 13

A new view on the semantics and pragmatics of operators

333

Whenever the present tense is used, the relevant part of the SoA overlaps with the speech time interval (S). Whenever the past tense is used, the relevant part of the SoA is anterior to the S. Whenever the future tense is used the relevant part of the SoA is posterior to S. Although the semantic definition of tense operators has been changed, the communicative function is equal to the standard view. Tense operators do not change the description of the SoA. They contribute to the referring function of the utterance, that is, they link the SoA to a real or imaginary world. Tense operators are therefore predication-operators (7i2), located at the second layer.

4.2. Tense and aspect What are the implications of the semantic definition of tense for the relation between tense and aspect operators? Aspectual operators mark the relevant part of the temporal structure of a property or relation. Only this part helps build up the proper SoA description. Aspectual operators thus explicitly select the part of the temporal structure (and therefore of the SoA) that is relevant to the topic time of the discourse. Tense markers, on the other hand, locate the part of the SoA that is relevant to the discourse on the time axis in relation to the speech time interval. For instance, in He has written a letter, the relevant part of the SoA is the post-state, selected by the perfect. This post-state is located as overlapping with the speech time interval (indicated by S) because of the present tense marking. Consider (19): (19) Representation of present tense and perfect: He has written a letter. S

Φ I

Time line

I topic

If a past tense marker is used, then the relevant part of the SoA, the poststate in case of perfect marking, is located anterior to S. Consider (20):

334

Annerieke Boland

(20) Representation of past tense and perfect: He had written a letter S

^ I

Time line

I topic

If a future tense marker is used, the relevant part of the SoA, the post-state in case of perfect marking, is located posterior to S. Consider (21) for two possible representations of He will have written a letter. (21) Representations of future tense and perfect: He will have written a letter

I1 I1

I V

N

1^

Time line

^

Time line

topic

/t v

-χ

\ y topic

When a prospective is used, it is the pre-state that is located on the time line with respect to the S, when an ingressive is used, it is the initial boundary that is located on the time line with respect to the S, etc.

4.3. Interpretation of Simple Past and Present It is claimed, among others by Comrie (1976: 66), that the present tense is essentially imperfective. But is this indeed the case? In the proposed account it is assumed that in the Simple Past and Present in English, no selec-

A new view on the semantics and pragmatics of operators

335

tion of the temporal structure is made. So, only the meaning of the tenses applies: for the Simple Present (the relevant part of) the SoA overlaps with the moment of speaking and for the Simple Past (the relevant part of) the SoA is anterior to the moment of speaking. Since no selection is explicitly made, it can only be assumed that the entire temporal structure is the relevant part of the SoA, which is a perfective view. Is this compatible with the possible interpretations of a Simple Present in English? The topic time determines which part of the SoA is relevant to the discourse; in the case of the Simple Present, this is the complete temporal structure. The interpretation of the Simple Present therefore depends on the relation between topic time, SoA-time and S. The first possibility is that the time for which the speaker wants to assert something (topic time) is more or less confined to S. This interpretation is possible if the interval of the SoA indeed more or less coincides with S and this is only the case for momentaneous SOAs. The effect is a reportative interpretation such as in there he comes/goes and in sports reports, such as spreads it wide (Mackenzie, to appear). Consider (22) for the representation of a reportative interpretation: (22) Representation of reportative interpretation of Simple Present S

I I I

Time line topic

A second possibility is that the time for which the speaker wants to assert something is much larger than S. In those cases, most SOAs do not overlap with the entire topic time and the interpretation is that the SoA is repeated, such that it fills the topic time. The interpretation may be either habitual or distributive, depending on the number and specificity of the argument(s). For example, he runs the marathon is interpreted as habitual, since the SoA in which a specific individual is involved is repeated. A tiger hunts for wild pigs (WWW) or Tigers fight for food (WWW) are interpreted as distributive, since the interpretation of the first argument is generic: the SoA is repeated because it holds for several individuals. Consider (23) for a repre-

336

Annerieke Boland

sentation of the habitual or distributive 14 interpretation of the present tense: (23)

Representation of habitual or distributive interpretation of Simple S

t#H

— J

•

Time line

topic

Present In this interpretation, the boundaries of the temporal structure of one SoA are included in topic time, but at the same time, the boundaries of the repeated SoA-sequence are not included in topic time. This yields the interpretation of a universal statement, a situation that holds at all times. The topic time determines that the sequence of SOAs that overlaps with S is relevant to discourse, hence the present tense. Note that it is not necessarily the case that one of the repeated SOAs indeed takes place at the same time interval as S. The interpretation that arises is that the situation of the repeated SOAs holds at an interval that overlaps with the speech time interval. A habitual or distributive interpretation therefore resembles a permanent state. As argued earlier (see 4.1), only the part of the state that is relevant to the discourse needs to be located in time and language users infer that it may hold at other time intervals as well. Consider (24): (24)

Representation of Simple Present in combination with a Stative SoA

\

Time line topic

In a habitual or distributive reading, the same inference takes place. Tense marking only concerns the sequence of repeated SOAs that is relevant to the discourse and language users infer that it may hold at other time inter-

A new view on the semantics and pragmatics of operators

337

vals as well, hence a sort of permanent state. This differs from a real state in that a habitual or distributive reading there are time intervals at which the SoA is not actually taking place. Finally, in English, there is a future interpretation of the Simple Present form possible for scheduled events, as in The train leaves at five tomorrow but also in temporal adverbial phrases such as When he comes here, I'll tell him (from Comrie 1976: 68). The combination cannot be interpreted as habitual because of the temporal adverbial phrase. How can we account for the fact that the Simple Present can be used in these cases? I follow Klein's (fc.) analysis that the time for which the speaker wants to assert something (topic time) stretches from the present into the future and includes the time of the complete temporal structure. Apparently, the Speaker wants to assert something about a time interval for which the SoA does not (yet)15 hold, up to the interval for which the SoA holds in its entirety. The complete temporal structure of the SoA including the preceding temporal span is relevant to discourse and located on the time axis. Since this interval overlaps with speech time, it is possible to use a present tense. The resulting reading has a sense of a scheduled or planned event. Consider (25) for a representation: (25) Representation of scheduled event interpretation of Simple Present S Time line

J topic

The counterpart of the scheduled event interpretation is that the topic time extends into the past. The speaker wants to assert something for a time span that stretches from some point in the past up till the moment of speech. This may account for the use of the Simple Present for narrating past events, the historic present. (Cf. also Klein fc.) It is often assumed that a present tense is imperfective. This claim should in my view be modified. A Simple Present does not select a specific part of the SoA to ascribe to the argument(s) but the complete temporal structure. In this sense it is only perfective, which is compatible with the interpretation in (22) and (25). However, the most common interpretation

338

Annerieke Boland

is that the topic time extends the speech time and that the SoA is repeated. In a habitual or distributive reading the Simple Present does have an imperfective sense, since the repeated temporal structure overlaps with topic time, but on the other hand it has a perfective viewpoint, since the temporal structure is in its entirety ascribed to the argument(s). Note that the Simple Present does not mark one of the above interpretations: it only marks present tense and the different interpretations are the result of implicatures or inferences. The interpretation of the Simple Past is less complex. The Simple Past in English indicates that (the relevant part of) the SoA occurred anterior to S. Since no selection of the temporal structure is made, the entire temporal structure is ascribed to the argument(s) and located anterior to S. This results in a perfective viewpoint, at least for SOAs that have terminal boundaries. For a state without boundaries, only the part relevant to the discourse is located with respect to S and it can be inferred that the state may still hold at the speech time interval. This fact is however not relevant to the discourse. The difference between the simple tenses in English and a real imperfective and perfective marker is that the latter explicitly selects a subinterval or the entirely of the temporal structure, whereas with simple tenses, no selection is explicitly encoded. And since there is not a more specific selection, the complete temporal structure must form part of the description of the SoA.

4.4. Summary In sum, tense operators contribute to linking the designated SoA to the real or imaginary world. They locate the part of the SoA that is relevant to the discourse on the temporal axis in relation to some tp Tense operators leave the description of the SoA intact and do not change the internal dynamics of the SoA nor any other aspect of the parts that build up the SoA (the arguments, the properly or relation, the additional participants). They may therefore be properly located at the second layer as predication operators (7i2), which is indeed the standard approach in FG. I have shown that the relation between topic time, the time span of the SoA and S may result in different interpretations of the Simple Present. In all of these interpretations, however, it is the complete temporal structure that is ascribed to the argument(s), so that the viewpoint is perfective. In

A new view on the semantics and pragmatics of operators

339

the habitual or distributive interpretation of a Simple Present, there is an imperfective sense as well, since it is not the entire sequence of repeated SOAs that is located in time, but only the part of the sequence that is relevant to the discourse. This is the same interpretation that applies to states without boundaries.

5.

Operators of quantification

5.1. Semantics The final semantic domain to be discussed is the domain of quantification (or frequency). Dik (1997) uses the term quantificational aspect for this domain, which includes grammatical expressions of semelfactive, iterative, frequentative, distributive and habitual. In the standard FG approach quantification operators are defined as markers of the frequency of the SoA designated by the predication. Since they do not influence the description of the SoA these operators are considered as 7i2-operators. However, both Klein (1994) and Anstey (2002) distinguish between operators that indicate the frequency of a property or relation and operators that indicate the frequency of the entire SoA. Anstey notes that in some languages there are operators that can mark the iteration/intensity of the predicate. In Biblical Hebrew for example the predicate 'break' has a derived predicate 'break into pieces'. Anstey (2002: 6) reserves the term predicate quantification for these operators, but I will use the term property quantification. These operators modify the temporal structure of the property or relation designated by the predicate and are therefore located at the first layer as predicate operators (πΙ). Operators that mark the frequency of the entire SoA are called eventquantification, following Anstey (2002: 6). In the case of eventquantification the temporal structure of the SoA is not changed, but there are several more or less 'identical' SOAs that are relevant to the discourse. The speaker does not refer to one specific event in the real or imaginary world, but to a sequence or group of events. In order to find out whether a marker of quantification is thus a 7rl - or a 7t2-operator, one has to know whether it modifies the property or relation ascribed to the argument(s) or whether it has no such effect and links the SoA to the real world. How do we know whether a frequency marker indicates property quantification or event quantification? I will discuss two

340

Annerieke Boland

expressions in English to illustrate this matter, the habitual marker used to and the expression keep -ing.

5.2. Habitual One of the habitual markers in English (alongside would and will) is used to. The crucial question is: is used to part of the SoA description, that is, part of the first layer, a predicate operator, or does it contribute to the referring function within an utterance? For example, the SoA described in He used to drink could in principle be analyzed as either: PAST ((HE, DRINK)frequently) or as PAST (HE, DRINKfrequently) A second point of discussion concerns the semantics of a habitual marker. According to Dik's definition, the habitual is characterized by two senses: "the SoA (potentially) recurs due to a habitual propensity of the participant involved." (Dik 1997: 236, italics mine). This definition seems to waver between two opinions: the SoA is quantified (I will call this the repetitive sense), but the property designated by the predicate is also described as a propensity of the participant (I will call this the habitual sense), which comes close to the function of inherent modality markers. Now, where the habitual sense is still an important sense of used to, we would expect selection restrictions on the first argument and the predicate, for a habit can only be ascribed to entities that can have propensities or habits (? Houses have the habit of being big / tend to be big) and habits are only properties or relations that are controlled (?He has the habit of being ill / tends to be ill). So, are there still selection restrictions of this kind on the expression of used to? As the examples in (26-30) show, there do not seem to be any restrictions on the arguments or predicates with which used to can combine: (26) I used to read a lot. (CWO) (27) Everyone used to call her smelly. (CWO) (28) We used to get grants you know for the (...) (CWO). (29) The flags used to be in the colors of the "Home " Club. (WWW) (30) There used to be a playground here. (WWW) We could therefore conclude that the 'habitual' sense of used to has been marginalized. It may be a matter of world knowledge or inference that if a specific entity is involved in a repeated [+control] SoA, the most likely

A new view on the semantics and pragmatics of operators

341

interpretation is that the repetition of the SoA is due to the habit or propensity of that entity. However, the repetitive sense of Dik's definition of the habitual does not seem to be completely compatible with the use of used to either, as illustrated most clearly by (30). There is no repeated SoA here (as far as we know, a playground is present for one extended interval in time, not with interruptions) and there is no participant with a propensity to this SoA. What does used to express in this utterance? How does it differ from There was a playground here? I assume that Dik's definition of the habitual has to be modified, following Klein's approach once more. In Klein's analysis (1994: 47) a speaker chooses to speak about a series of topic times when using a habitual marker, and for all of these topic times, the SoA holds. It is thus not necessarily the case that the SoA indeed recurs, but the time interval relevant to the discourse is a sequence of intervals in the past. The representation of a habitual marker with a SoA without boundaries is represented in (31). Note that the SoA holds at all times, but since the topic time intervals determine which parts of the state are relevant to the discourse, they create a sense of repetition. (31)

Representation of a habitual in combination with a state, without boundaries S J I

II

II

Ii

II

•

Time line

I topic

The representation of a habitual in combination with a SoA with boundaries is presented in (32). Here, the SoA is indeed repeated.

342

Annerieke Boland

(32) Representation of a habitual in combination with a SoA with boundaries S

HHHΗΗ

LJ l_l LJ LJ LJ marvellous —> nice —> beautiful b. incredible —» astonishing -» amazing -» surprising

2) In general, the less marked exclamative constructions are used to express the low degrees of exclamation whereas the more marked constructions are reserved to mediate the high ones. For example, in MSA, exclamation is mainly expressed by two predicate forms: ma ?aftal and Fafiil bi. The former is the more natural and more frequent way to carry the modality at issue; the latter is the marked one, which is consequently used to express a higher degree of exclamation. The difference between the sentences of the following pair is mainly a difference of evaluative degree: (39) Modern Standard Arabic a. ma ? akrama Zaydan! EXCL generous Zayd.ACC 'How generous Zayd is!' b. 7 akrim bi-Zaydin! generous.EXCL with-Zayd.GEN 'How very generous Zayd is!' In the same vein, one can assume that in the languages where Exclamation can be expressed either through declarative or interrogative constructions, these two types of exclamative constructions, despite having the same propositional content, seem to differ from each other with respect to the degree of exclamation which is intended to be carried. In general exclamatory interrogative constructions are more marked than exclamatory declarative constructions. This can be explained, it seems to me, by the fact that exclamation typically presupposes the truth of the fact which causes the exclamative reaction. On the basis of this observation, one can assume that sentence (40b) carries a higher degree of exclamation than sentence (40a):

Exclamation: Sentence type, illocution or modality?

(40)

365

a. She has GROWN! b. Hasn't she GROWN!

3) Some languages are provided with grammatical means which enable them to reinforce the exclamation value already basically expressed by a certain kind of exclamative construction. The reinforced exclamative construction permits to carry a higher degree of exclamation than the normal non-reinforced one. This falls under the ancient Arabic grammatical principle that "Adding in the form is adding in the meaning" or its modern version called the 'iconicity principle' by Givon (1983). If we generalize it in such a way that it can also apply to other notions than Topic, this principle may be reformulated as follows: (41) Iconicity principle The greater the semantic / pragmatic content of a linguistic expression, the more coding material is needed to express it. Let us consider some examples from French, SMA, Moroccan Arabic and Egyptian Arabic which support the applicability of principle (41) to exclamative constructions: (42)

French a. Elle est BELLE! b. Quelle est BELLE! c. Qu 'est-ce qu 'eile est BELLE! d. Dieu! Qu 'est-ce qu 'eile est BELLE!

(43)

Standard Modern Arabic a. ma ? ajmala Hindan! EXCL beautiful Hind.ACC 'How beautiful Hind.' b. ?ala ma ?ajmala Hindan! REINF EXCL beautiful Hind.ACC 'How very beautiful Hind is!'

366

Ahmed Moutaouakil

(44) Moroccan Arabic a. sa klina! Dinner have.eaten.we 'What a dinner we had!' c

b.

c

la sa klina! Over dinner have.eaten.we 'What a good dinner we had!' c

c

c. walayinni la sa klina! But over dinner have.we.eaten 'What a delicious dinner we had!' (45) Egyptian Arabic

Q

a. ?akalna asa! have.we.eaten dinner 'What a dinner we had!' b. ?akalna hittit asa! have.we.eaten some dinner 'What a good dinner we had!' c. ?akalna hittit °asa ?innama have.we.eaten some diner but 'What a delicious diner we had!'

?eh! what

The common feature of all these sentences is that they express increasing appreciation by adding or cumulating special morphemes (quantifiers, particles, etc.). That constructions such as (42)-(45) indeed carry increasing exclamation can be shown by the fact they differ with respect to the extent to which they tolerate restrictive comments. Compare: (46)

a. Elle est BELLE mais eile ne m 'impressionnepas beaucoup! 'She is BEAUTIFUL but she doesn't impress me very much!' b. ?Qu 'eile est BELLE mais elle ne m 'impressionne pas beaucoup! 'How BEAUTIFUL she is but she doesn't impress me very much!'

Exclamation: Sentence type, illocution or modality?

367

c. ??Qu 'est-ce qu 'eile est BELLE mais eile ne m 'impressionne 'How very BEAUTIFUL she is but she doesn't impress me pas beaucoup! very much!' d. ???Dieu, qu 'est-ce qu 'elle est BELLE mais eile ne 'Oh God! How very BEAUTIFUL she is but she doesn't m 'impressionnepas beaucoup! impress me very much!'

4.3. Exclamative modality and layering We have pointed out above that exclamative modality can occur not only at the propositional but also at the term layer. In the following two subsections, we will examine the different scope relationships that can be involved in exclamative clauses and their representation in the underlying clause and term structures. In subsection 4.3.3, we will claim that exclamative modality can also have in its scope a whole stretch of discourse, which allows us to speak of discourse exclamation in addition to proposition and term exclamations.

4.3.1. Propositional exclamation When it occurs in the propositional layer, exclamative modality can be represented by a) propositional satellites, b) a propositional operator or c) modal predicates. In the cases where it is expressed by lexical means, propositional exclamative modality takes the form of adverbial expressions such as surprisingly, wonderfully, amazingly, etc. These adverbs have the status of propositional satellites. To give an example, sentence (47) has (48) as its (simplified) underlying structure: (47) Surprisingly, John passed the exam. (48) DECL ASS E;: [X;: [Passv (surprisingly)]

(John)p oSu bjTop ( e x a m ) G o 0 b j ] F o c

368

Ahmed Moutaouakil

When exclamation is expressed by grammatical (morphological, syntactic and/or prosodic) means, it is to be represented as a propositional operator as becomes clear from (49): (49) Tp ILL E,: [appr/depr EXCL^^X,: [Predication]] (Tp= sentence type; ILL = illocution; appr=Appreciative; depr= depreciative; l...n= indices of degrees of exclamation) For example, the Arabic sentence (50), in which exclamative modality is coded in the form of the predicate ?akrim, has (51) as its underlying representation: (50)

?akrim bi-Zaydin! generous with-Zayd.GEN 'How generous Zayd is!'

(51) DECL ASS Ej: [Appr Excl2 Xf: [Pres e,: /k.r.m./

(Zayd)p oSu bjTop]Foc]

I have argued elsewhere (Moutaouakil 1993), elaborating on the analysis proposed in Vet (1990), that constructions like (52): (52) I am surprised that John passed the exam. involve what I called a "pseudo-embedding" phenomenon in the sense that, in spite of their surface syntactic configuration, they consist of a simple clause where the proposition modality is expressed by a modal formula. Since this modal formula plays the same role as any modal morpheme or particle, the most natural way to handle it (given the current FG model) is to represent it as an instantiation of the proposition operator. According to this analysis, the structure underlying (52) can be something like (53) where the expression I am surprised functions, as a whole, as an exclamative propositional operator: (53) DECL ASS Ej: [I am surprised X;: [Past e,: [Passv (John)p 0 subjTop (exam) G oObj]] Foe]

Exclamation: Sentence type, illocution or modality?

369

4.3.2. Term exclamation In Moutaouakil (1993), it is argued that exclamation (as well as other subtypes of subjective modality) can also occur at the term level. Examples of exclamative terms are the terms avec quel enthousiasme! and une femme merveilleuse in sentences (54) and (55): (54) Jean travaille avec quel enthousiasme! 'With what enthusiasm Jean works!' (55) J'ai rencontre une femme merveilleuse! Ί met a marvellous woman' It was suggested there to extend the layered model of term structure proposed in Rijkhoff (1992) by adding a fourth layer representing the different modality distinctions (including exclamation) which can occur in a term. More recently, starting from Rijkhoff s proposal, Dik (1997: I, 163) presents a modified version of underlying term structure whose general schema is (56): (56) co2 - loc co2 - quant x: [cc>i - qual pred [N] (args)] It is clear from (56) that three kinds of term operators (and consequently three layers) are distinguished: a) qualifying operators, b) quantifying operators and c) localizing operators. My suggestion here is to adapt the proposal made in Moutaouakil (1993) in such a way that it can be incorporated into the underlying term structure represented in (56). By adding a fourth kind of operator (and a fourth term layer) which can be generically labelled "modal operator", we get the general schema (57): (57) co3-mod [co2-loc [co2-quant χ: [ω,- qual pred [N] (args) θ,] θ 2 ] θ3]) As far as exclamation is concerned, it can be represented as a value of the modal operator and possibly also by a modal satellite, as becomes clear from (58):

370

Ahmed Moutaouakil

(58)

([appr/depr EXCLi ...„ [co2-loc [co2-quant x: [ωΐ-qual pred [N] (args) θ ^ θ 2 ] θ 3 ])

According to this proposal, the underlying structures of the second terms in sentences (54) and (55) are (59) and (60) respectively: (59)

([appr EXCL 2 [i [1 xf: [enthousiasme [Ν] ] θ] Θ]) ManFoc

(60)

([appr EXCL, [i [1 Xj·. [femme [Ν] ] θ] Θ] (merveilleuse)])

4.3.3. Discourse

GoFoc

exclamation

Exclamation is, as far as I know, commonly thought of as a clausal feature. Yet, it can also have in its scope a whole stretch of discourse. In this respect, one can invoke, as typical examples, the well-known varieties of panegyrical discourses. Here is a simple example: (61)

What a nice girl I saw yesterday in Amsterdam! What beautiful eyes she had! How lovely was her smile!

One of the insightful assumptions advocated in Dik (1997: Π, 419) and which opens new avenues for very promising research in the functional grammar of discourse is that discourse can be viewed as containing the same layers as those involved at the clause level. In the light of this assumption, Dik postulates the existence of "discourse illocution" which "takes a whole (section of) a discourse in its scope", as becomes clear from representation (62) (Dik 1997:1, 419): (62)

ILL (Discourse-episode)

In order to account for the exclamative character of stretches of discourse such as (61), it is also possible, as pointed out in Moutaouakil (1998), to postulate, elaborating on Dik's (1997: Π, 424-432) assumption that clause layering can be projected onto the discourse level, a "discourse modality" layer. According to this assumption, the structure of discourse is rather (63): (63)

ILL (mod (Discourse-episode))

Exclamation: Sentence type, illocution or modality?

371

If we consider exclamation to be one of the possible values of the discourse modal operator (Mod), we can represent the general structure of exclamative discourse as follows: (64)

ILL (appr/depr Excli...„ (Discourse-episode))

Accordingly, the (very simplified) underlying structure which we can assign to (61) can be something like (65): (65) DECL ASS [appr EXCL2 [saw (I) (a nice girl) (yesterday) (in Amsterdam). Had (she) (beautiful eyes).Was lovely (her smile ...] Notice that it is possible to re-interprete Benveniste's dichotomy "Recit vs. Discours" as values of the D-mod operator, as suggested in Moutaouakil (1998). In this case, subjective modal distinctions (including exclamation) would be subtypes of Discourse modality.

4.4. Implications and perspectives As is to be expected, the analysis of exclamation as a type of subjective modality (rather than an illocution) necessarily has some consequences for both the representation and the derivation of linguistic expressions in FG. In order to give a complete and precise picture of the implications of such an approach, further research is needed. In what follows, I will mention only two of these implications: a) the fact that the parallelism between term and predication advocated in Rijkhoff (1992) can be said also to hold between term, clause and discourse, and b) the fact that so-called sentence types are not determined only by (basic) illocutions but are rather results of an interplay of illocutionary and modal features. In the same connection, I will briefly examine the possibility of a modular approach to (some types of) subjective modality.

4.4.1. Extending the parallelism between predication and term The parallelism between term and predication has been discussed and argued for in Rijkhoff (1992) and Dik (1997). As shown above, this parallelism can be pushed a step forward to also hold between term, clause and

372

Ahmed Moutaouakil

discourse. The structural similarity between these three entities reveals itself at the representational and interpersonal level. Term, clause and discourse contain, in addition to the element which constitutes the representational level (i.e. a predication or a discourse-episode) modality and illocution layers which together define the interpersonal level. As regards the modality layer, the structural isomorphism between the three linguistic entities at issue is shown in representations (66)-(68): (66)

Term structure [co3 [Predication] (θ3) ]

(67) Proposition structure [π3 [Predication] (σ3) ] (68) Discourse structure a. [D-mod [D-episode] D-mod satellites] b. [π-mod [D-episode] σ-mod] Their complete structural isomorphism is represented in (69a, b), (70) and (71): (69) Discourse structure a. [D-ill [D-mod [D-episode] D-mod satellites] D-ill satellites] b. [π-ill [π-mod [D-episode] σ-mod] σ-ill] (70)

Clause structure [π4 [π3 [Predication] (σ3)] (σ4)]

(71)

Term structure [ω4 [ω3 [Predication] (θ3)] (θ4)]

Notice that the question whether terms can also have their own illocutionary values when they occur in a 'complete' sentence (cf. Mackenzie 1998 for the underlying representation of 'holophrastic' linguistic expressions) is left open, as indicated by the interrogative mark in (71).

Exclamation: Sentence type, illocution or modality?

373

4.4.2. Exclamative modality and sentence type It is generally agreed that what determines the sentence type of a given linguistic expression is the (basic) illocution it conveys. However, the sentence types to which linguistic expressions belong result, in fact, from the interplay of different semantic/pragmatic features such as illocution, pragmatic functions and, as I will claim, modality. So we can speak of a 'Modality Indicating Device' (MID) alongside the well-known 'Elocutionary Force Indicating Device' (IFID). As far as exclamative constructions are concerned, the way in which this interplay takes place can be summarized as follows: 1) As for the 'normal' exclamative constructions, exclamation values (Appreciative vs Depreciative) and degrees determine the prosodic contour, which is specific to this type of construction. These features can also determine the morphology of the exclamatory term phrase or the predicate of the exclamatory clause, according to the general rule (72) (Dik 1997: I, 349): (72)

OP [A] = Β

To give an example, the rules responsible for the form of the term quel enthousiasme in (54) and the predicate in (50) can be formulated as (73) and (74) respectively: (73)

Appr Excl, [Enthousiasme] = Quel

(74)

Appr Excl 2 [k.r.m v ] = ?akrim

enthousiasme

In some languages, word order (within the clause as well as within the term phrase) can also be sensitive to exclamative modal features. In Moroccan Arabic, the fronting of the adjectival predicate (i.e. its placement in clauseinitial position) expresses a relatively high degree of exclamation, as becomes clear from the comparison between (75a) and (75b): (75)

a. had d-dar zwina! this the-house nice 'This house is nice!'

374

Ahmed Moutaouakil

b. zwina had d-darl Nice this the-house 'How nice this house is!' In French, the adjectival restrictor is, as is well known, typically placed after the head noun. This order can, however, be reversed as is shown by the contrast between (76a) and (76b): (76) a. J'ai participS ä une soiree sacree, I have participated to an evening sacred Ί took part in a religious ceremony.' b. J'ai participe a une sacree soiree. I have participated to a memorable evening party Ί took part in a memorable evening party.' As assumed in Moutaouakil (1993), the adjective sacree in (76a) is a descriptive restrictor which thus belongs to the quality layer whereas, in (76b), it rather expresses an emotional evaluation of the entity referred to by the term phrase and thus serves to function as a satellite of the modal layer. The difference between (76a) and (76b) is accounted for in underlying term structures (77a) and (77b): (77) a. ([i [1 xf: [soireeN sacreeA] Θ] 0])GoFOC b. (EXCL [i [1 Xj: [soiree] θ] Θ] (sacreeA))GoFoc In order to describe the difference in word order between (76a) and (76b), a general placement rule can be formulated which will assign the appropriate position to the adjectival restrictor in French on the basis of its underlying (modal vs non-modal) status. 2) Exclamative constructions can undergo a grammaticalization process involving a conflictual interaction between IFID and MID. Through such a process, it can happen that exclamative modality progressively neutralizes the formal effect of illocution. The final stage of this neutralization process yields idiomatic exclamative constructions.

Exclamation: Sentence type, illocution or modality?

375

4.4.3. Towards a transmodular approach to (subjective) modality The basic claim advocated in Kroon (1997), Bolkestein (1998), Vet (1998), van den Berg (1998) and Liedtke (1998) is that a separate pragmatic / discourse module should be distinguished from the grammatical module. This module is intended to deal with the contextually determined properties of linguistic expressions, such as speech acts (Vet 1998) and pragmatic functions (Bolkestein 1998). In the same perspective, I would like to suggest that such a module1 can especially well host the representation of exclamation (and the subjective modality distinctions in general), given its contextual (speaker-oriented) nature. The transmodular derivation of exclamative constructions can be roughly conceived of as follows (cf. also Vet 1998 for a different conception). In the pragmatic module, a (pragmatic) underlying structure represents the illocutionary force, the exclamative modality and the pragmatic functions (Topic and Focus) associated with the linguistic expression; in the grammatical module, a (grammatical) underlying structure encodes the semantic and structural properties; these two underlying structures (which I suggest in formulating, for the sake of uniformity and simplicity, in the same FG standard style) are taken as inputs to expression rules which deliver the final syntactic form. This is visualized in Figure 1. To give an example, sentence (78) can be, in this approach, be thought of as deriving from underlying structures2 (79) and (80) located in the pragmatic and the grammatical modules respectively: (78) How beautiful she is! (79) ASS E;: [appr EXCL2 X,: [ ej : f (xDtoJfoc] (80) DECL Uj! [Xf: [pres e;: beautiful (she)PosUbj]] where U = utterance Let us notice that the other modules (epistemic, logic, social and perceptual) of the Model of Natural Language User (MNLU) probably intervene in this derivation. I will not go into this issue here.

376

Ahmed Moutaouakil

MNLU

Figure 1. The modular approach

5.

Conclusions

The claim I have tried to advocate in this study is that there are empirical as well as theoretical reasons to conceive of exclamation as a subtype of subjective modality rather than an illocution. Viewed in the light of this assumption, (non-idiomatic) exclamative constructions turn out to be subjectively modalized declarative, interrogative or imperative constructions typically conveying an assertion as a basic or (derived) illocution. Such an approach makes it possible to account in a principled way for the formal

Exclamation: Sentence type, illocution or modality?

377

properties of exclamative constructions that are determined by the different values and degrees of exclamation. A more satisfactory description and explanation of these properties can be arrived at if we consider sentence types as the result of a conflictual interplay between IFID and MID which may be progressively solved by a grammaticalization process yielding exclamative idioms. The re-interpretation of exclamation as a subjective modality allows us to postulate a modal layer at the term and discourse levels and then to account for the full structural parallelism between these three entities. Given that exclamation (as well as the other categories of subjective modality) is, like illocutionary force and pragmatic functions, a contextually determined notion, a transmodular approach can be proposed in which the derivation of exclamative constructions (and all linguistic expressions in general) is achieved through the interaction of the pragmatic and the grammatical modules. It goes without saying that further research within the FG framework will hopefully explore other interesting aspects of exclamation (i.e. lexical and "templatic" exclamative idioms, exclamative "holophrases" etc.) and reveal other possibly more satisfactory ways to come to grips with this highly rich and complex phenomenon.

Notes 1.

2.

Pragmatic module is not synonymous, in my opinion, with discourse module if we think of discourse, following Dik (1997b), as a linguistic entity larger than a sentence or a clause (i.e. a "Texte" in the proposal made in Moutaouakil 1998).. We can therefore deal with discourse partly in the grammatical module of the MNLU and partly in the pragmatic module (just like the clause). In this case, only the pure pragmatic features of discourse such as discourse illocution, subjective discourse modality and discourse pragmatic functions are to be accounted for in the pragmatic module. According to this conception of the modular approach to discourse, the underlying text structure advocated in Moutaouakil (1998) can be maintained. I intend to return to this very important issue in subsequent research. As to the way in wich the underlying (clause or discourse) structure is to be represented, I will adopt Dik's well-known hypothesis that all the modules of the MNLU, except for the perceptual module, "speak" the same language. Consequently, the pragmatic structure will be represented in the same FG

378

Ahmed Moutaouakil style as the grammatical structure with the difference, however, that it does not contain the same information.

References Berg, Marinus.van den 1998 An outline of a pragmatic functional grammar. In Bolkestein and Hannay (eds.), 77-103. Bolkestein, A. Machtelt 1998 What to do with Topic and Focus? Evaluating pragmatic information. In Bolkestein and Hannay (eds.), 193-214. Amsterdam: Benjamins. Bolkestein, A. Machtelt and Mike Hannay (eds.) 1998 Functional Grammar and Verbal Interaction. SLCS 44. Amsterdam: Benjamins. Dik, Simon C. 1997 The Theory of Functional Grammar. 2 vols. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Givon, Talmy 1983 Topic Continuity in Discourse: a Quantitative Cross-linguistic Study. Amsterdam: Benjamins. Kroon, Caroline 1997 Discourse markers, discourse structure and Functional Grammar. In Discourse and pragmatics in Functional Grammar, J.H. Connolly, R.M. Vismans & R.A. Gatward (eds.), 17-32. Berlin: Mouton de Gruyter. Lakoff, George 1987 Women, fire and dangerous things. What categories reveal about the mind. Chicago: University of Chicago Press. Liedtke, Frank 1998 Illocution and grammar: a double level approach. In Bolkestein & Hannay (eds.), 107-127. Mackenzie, J.L. 1998 The basis of syntax in the holophrase. In Bolkestein and Hannay (eds.), 267-296. Moutaouakil, A. 1993 Reflections on the layered underlying representation in Functional Grammar. Faculty of Letters, Rabat.

Exclamation: Sentence type, illocution or modality? 1995

379

qadaya l-luga l-arabiyya fi l-lisaniyyati l-wadifiyya. [Issues in Arabic functional linguistics], Rabat: Dar Al Aman. 1998 Benveniste's Recit and Discours as discourse operators in Functional Grammar. In Bolkestein & Hannay (eds.), 25-41. Rijkhoff, Jan N.M. 1992 The noun phrase. Dissertation, University of Amsterdam. Searle, J., R. 1979 Meaning and expression. Cambridge: Cambridge University Press. Vet, Co 1990 Asymmetries in the use of tense and modality. In Layers and levels of representation in language theory: a functional view, Jan Nuyts, Machtelt Bolkestein, Co Vet (eds.), 123-138. Amsterdam: Benjamins. 1998 The multilayered structure of the utterance. In Bolkestein and Hannay (eds.), 149-168.

Close appositions Eve lien Keizer

1.

Introduction1

This paper will deal with a number of very similar binominal expressions, often referred to in the literature as 'close appositions'. The more general category of appositions, which has been the subject of a considerable amount of research, has proved rather difficult to define and, in the course of the debate, has come to include so many different constructions that it is difficult to conceive of these constructions as constituting one category. Here, however, I will not be concerned with the question of which constructions should and which should not be called appositions; instead, the discussion to follow will be restricted to a small subset of appositional constructions - a subset which is generally referred to as close or restrictive appositions and which consists of a number of binominal constructions which, formally as well as functionally, behave in an apparently unified manner. Nevertheless, the aim of this chapter will be to show that there is considerable variation (in terms of syntactic, semantic and pragmatic behaviour) within this group, and that we are really dealing with a number of subtypes, given in (la-d). (1)

a. The actor Orson Welles The word recession b. Orson Welles the actor c. Actor Orson Welles d. My friend Orson Welles My friend the actor

Type Type Type Type Type Type

la: det + Ν + NP lb: det + Ν + Ν 2: NP + det + N 3: Ν + NP 4a: ProFoss + Ν + NP 4b: ProPoss+ Ν + det +N

In the present paper I will confine myself to Types la&b and Type 2; for a more comprehensive discussion of all of the types distinguished in (1), see Keizer (in prep.).

3 82

Evelien Keizer

I will begin by briefly discussing some of the criteria for close apposition put forward in previous treatments. As none of these accounts seem to be quite satisfactory, I will suggest that instead of regarding close appositions as one homogeneous group (with one basic underlying form), we should distinguish a number of different types. In order to decide how best to represent these types I will consider evidence from defmiteness, as well as further syntactic and semantic evidence concerning headedness. I will then propose different underlying representations for construction types 1 and 2. Finally, I will consider the various discourse functions of these two types of close apposition. Before I start, a brief remark on terminology and examples may be required. As will become apparent in the discussion of previous analyses, a considerable number of different terms have been used to refer to close appositions as a whole and the two component elements. In what follows the term close apposition will be used to refer to each of the constructions in (1). The two nominal elements will be referred to either as the first element (or N l ) and the second element (or N2), or, alternatively, as the proper name (or uniquely denoting element) and the descriptive element. As for the examples used: most authentic examples are from the ICE-GB corpus;2 most of these examples are taken from spoken language and have a clearly restrictive intonation.

2. Previous analyses 2.1. Some criteria Over the years, a number of different treatments have been proposed for the constructions in (la-b). Some of these will be reviewed in the present section. Rather than discussing these treatments separately and in full, however, I will present a list of criteria for close appositions extracted from these earlier works, ordered according to a number of important formal and semantic features. The list is not exhaustive; it simply serves to introduce the concepts and criteria employed in discussions of close apposition, in preparation for the analyses presented in Sections 2.5 and 3.3. 1.

Intonation a. the stress pattern is secondary-primary (Haugen 1953; cf. Francis 1958: 302);

Close appositions b.

383

the two elements belong to the same information unit, indicated in speech by inclusion in the same tone unit (Quirk et al. 1972/1985).

2.

Class and form of the elements a. the two parts must belong to the same major form class (Fries 1952: 187; Hockett 1955: 101; Francis 1958: 301; Sopher 1971; Quirk et al. 1972/1985); b. close appositions contain (i) a modifier consisting of the definite article followed by a nontitular class noun (i.e. which normally requires an article); (ii) a head, which is a substantive expression containing a nonclass noun, i.e. proper noun or mass/substance noun (one which can normally occur without an article) (Haugen 1953; see also Curme 1931: 92-93).

3.

Structure and syntactic headedness A number of different analyses have been proposed, four of which are given below: a. In constructions of the type the poet Burns 'the second element is restrictive and is necessary to limit, or restrict, or define the meaning of the first' (Lee 1952; cf. Francis 1958: 301); = NPjiead + NP Mod b. A close apposition is a modifier-head construction, containing (i) a modifier consisting of the definite article followed by a nontitular class noun; (ii) a head, which is ... proper name or mass/substance noun (Haugen 1953); = NP Mod + N P H e a d c. The construction must be endocentric, while there must be no more justification for taking the first part as attribute of the second as head than for the reverse (Hockett 1955); = N P H e a d / M o d + NP H e ad/Mod

d.

4.

Constructions of the type the poet Burns are modifier-head constructions; the definite article has scope of both elements (Burton-Roberts 1975; see also Burton-Roberts 1994, Acuna-Farina 1996); = Det [NMod +NHead]

Syntactic omissibility a. On a syntactic level the apposition [=N2] ... is structurally independent and can itself function as the subject of the sentence (Sopher 1971); b. Each of the appositives can be separately omitted without affecting the acceptability of the sentence (Quirk et al. 1972/1985).

384

Evelien Keizer

5.

Semantic reference a. The two parts must refer to the same entity (e.g. Hockett 1955); The two parts must be identical in reference or else the reference of the one must be included in the reference of the other (Quirk et al. 1972/1985); b. 'It is a logical impossibility ... for coreferential items to constitute between them a higher NP' (Burton-Roberts 1975: 396).

6.

Semantic omissibility a. [On a notional level ... ] the two nouns are interchangeable without changing the meaning of the sentence (Sopher 1971); b. There is no difference between the original sentence and either of the resulting sentences in extralinguistic reference (Burton-Roberts 1975).

7.

Positions of the elements The two parts are arbitrarily reversible (e.g. Sopher 1971).

There seems to be more or less general consensus on the criteria given in 1 (intonation) and 2 (form). The following sections will therefore concentrate on the more controversial properties mentioned in 3 to 7.

2.2. Headedness and syntactic omissibility For many linguists, headedness and syntactic omissibility have been important criteria for determining the internal structure of close appositions. One look at the list of criteria given in the previous section (items 3 and 4), however, shows that this has not led to one generally accepted analysis. This is largely due to the fact that the notions involved are often ill-defined and tend to be applied in an intuitive and inconsistent manner. This has resulted in four different basic underlying structures for close appositions: 1.

NPHead + N P M o d

2. NP Mod +

NPHead

3·

NP H e a d/Mod + NPHead/Mod

4.

D e t [ N M o d +NHead]

Let us briefly consider each of these four analyses. The first analysis is simply based on the (intuitive) assumption that in close appositions of the type the poet Burns 'the second element is restrictive and is necessary to limit, or restrict, or define the meaning of the first' (Lee 1952; cf. Francis

Close appositions

385

1958). Haugen (1953), however, rejects such an analysis. He does so on the basis of the 'replacement-by-zero test', arguing that the relative function of any two linguistic elements can be determined by omitting each one of them in turn and seeing which one of them can stand alone. Thus, in a construction such as the poet Burns it is not possible, Haugen (ibid.: 165166) claims, for the proper noun to be replaced by zero, as the element the poet cannot be used in the same context, e.g. to start a discourse by talking about a poet the hearer may be assumed never to have heard of; this being irreconcilable with the use of the definite article. If, on the other hand, we omit the poet, Haugen (ibid: 166) continues, we have "a perfectly satisfactory sentence". Haugen concludes that rather than the first, it must be the second noun which functions as the head of the construction. The question that arises, however, is what is meant by a "perfectly satisfactory sentence". From a syntactic point of view, omission of either element leads to an acceptable result. Semantically, too, the resulting constructions seem to be perfectly satisfactory, no matter which element is being omitted. Nevertheless, Haugen is right in recognizing that omission of the first element may lead to a difference in acceptability. What Haugen has in mind, therefore, is pragmatic acceptability, or felicitousness, of the resulting construction (see also sections 2.3 and 4). In other words, Haugen's use of the replacement-by-zero test to determine headedness within close appositions fails to prove his point because he does not specify the level at which the resulting constructions are to be judged as acceptable. Similarly inconclusive is the analysis of close appositions proposed by Hockett (1955), who deems it inappropriate to speak of modifier and head when dealing with these constructions. The alternative he offers, however, is not altogether convincing. He concedes that in most endocentric constructions the evidence points either to interpreting A as attribute and Β as head, or to interpreting A as head an Β as attribute. There are, however, cases, according to Hockett (1955: 101), where there "is cogent evidence for both of these attributive alternatives. When we find this to be the case, we speak of 'apposition.'" There is, however something distinctly odd about two NPs being mutually attributive; moreover, Hockett does not actually specify any evidence to support this claim. The final analysis listed above differs from the first three in that it does not regard close appositions as consisting of two separate NPs, but rather as one single NP containing two nominal elements. Burton-Roberts (1975: 393), for instance, proposes to analyse close appositions like the poet

386

Evelien Keizer

Burns as modifier-head constructions, in which a proper noun is being modified by an adjectival noun (see also Acuna-Farina 1996). Like the others, however, Burton-Roberts does not provide much syntactic evidence for determining headedness. His justification is entirely based on the assumption that appositions like the poet Burns are derived from constructions containing a relative clause (for more details, see Section 3.1.2). Acuna-Farina (1996: 28ff), adopting Burton-Roberts's analysis of constructions like the poet Burns, does try to substantiate the claim that these constructions are modifier-head constructions in which the definite article has scope over both modifier and head. Unfortunately, his argumentation is based on a number of incorrect assumptions. First of all, Acuna-Farina argues that the proper noun must be the head because it is this noun which imposes restrictions on the elements with which it occurs in a close apposition construction. Thus, he continues, proper nouns in English can only cooccur with a determiner when it is modified (*the Burns vs the excellent Burns). Since close appositions behave in a similar way (the poet Burns), they, too, can be assumed to be head-modifier constructions. Secondly, Acuna-Farina argues, the proper name in a close apposition can only combine with a definite determiner (*a poet Burns)·, this, too, shows these constructions follow the normal modifier-head pattern, as constructions with proper names can never be indefinite (whether unmodified, as in *a Burns, or modified, as in *an excellent Burns). However, the data used by Acuna-Farina turn out to be incorrect. It is, of course, true that on their prototypical, referring use, proper nouns denote entities that are unique (within the discourse context), and that, on this use, they may be seen as inherently definite. If, on the other hand, the proper noun cannot be assumed to refer uniquely, it is possible to add modifiers to enable the hearer to identify the entity referred to. If, on the basis of this extra information, the speaker assumes the hearer to be able to (uniquely) identify the referent, the definite article is used to communicate this assumption, resulting in such expressions as this Burns, the poet Burns, the excellent Burns, or the Burns who came here yesterday (Acuna-Farina 1996: 28). If, on the other hand, the additional information cannot be assumed to guide the hearer to the identity of the referent, the indefinite article is used, resulting in such constructions as a Burns, a poet Burns, an excellent Burns and a Burns who came here yesterday, all of which, contrary to AcunaFarina's claim, are quite acceptable, as shown by examples (2)-(4). In (2) examples are given of unmodified indefinite constructions with a proper

Close appositions

387

noun: in (2a) the indefmiteness is due to a lack of unique reference; in (2b) to the hearer's assumed unfamiliarity with both referent and the name and in (2c) to the non-specificity of the reference. Examples of constructions containing a modified proper noun can be found in (3). Some examples of indefinite close appositions can be found in (4). (2)

a. Every now and then a Forsyte would come up, sidle round, and take a good look at him. (John Galsworthy, The man of property) b. And uh after research by her solicitors, she consulted a Mr Patrick James, who apparently is a world authority on this type of surgery. c. Do you know an Alice.

(3)

a. By 1918, Croatia's political class yearned for a Yugoslavia where an extended Serbia would have been balanced by a consolidated Croatia. b. Mr Speaker does the Prime Minister realise that if she faced a Labour Party which was not equally divided equally muddled equally confused on this matter she would today face a motion of no confidence in her failure at this historic moment. c. An embarrassed Sir Patrick Mayhew, QC, the Attorney-General, who was to chair the meeting as head of the Bar, had to cancel it after only 48 of the 6,000 practising barristers in England and Wales arrived at the Camden Centre in north London.

(4)

a. The Prime Minister, Mr John Major the Senior Secretary of State for Scotland Ian Lang, Tom King the Secretary of Defence, military chiefs General Sir Peter de la Billiere, Sir David Craig.

b. I have a friend John who's in Linguistics with me. c. He also has to put up with a soppy elder brother Robert who is forever mooning about some girl or other and a sister Ethel who has all the brisk no-nonsense superiority of a true Wodehouse gel.

388

Evelien Keizer

What is equally interesting is that the modifiers in expressions such as an excellent Burns often turn out to have a non-restrictive reading. Of course, the modifier excellent may have a restrictive reading, in which case contrast is implied with some other entity (the excellent Burns vs. the unremarkable Burns), but this is, in fact, a rather implausible reading. It is much more likely that the adjective is used non-restrictively: there is only one Burns, to whom the property 'excellent' applies (compare example (3c)). On both readings, however, the element Burns in the construction the excellent Burns functions as a common noun and does not in itself refer. The same can be said about the poet Burns, which also allows for both readings (for more details, see Section 4); in either case, neither the element poet nor the proper noun Burns is used referentially - it is only the construction as a whole which can be said to refer. A quick search of the corpus further shows that there are, in fact, plenty of instances where a proper noun co-occurs with a non-restrictive modifying adjective only. Again, the proper noun itself would have allowed the addressee to pick out the intended referent, which, moreover, tends to be given, or at least inferrable, in the discourse context. Examples are given in (5): (5)

a. Shocked Tony Blair blasted the murder as "wicked in the extreme. " (The Sun, 15 Jan 2003, front page) b. And David O'Leary will float the free kick forward towards big Allan Smith. c. Clever Keith, finger right on the spot as usual. d. How are you keeping yourself back in sunny Paris where the food is good?

We may therefore conclude that the examples provided by Acuna-Farina fail to reflect actual language use and as such cannot be used as a basis for determining which element functions as the head of the construction. Finally, Acuna-Farina sets out to show that the definite determiner and the element poet in the construction the poet Burns do not form a constituent. As evidence Acuna-Farina uses the examples in (6), where the heaviness of the first element forces an interpretation in which the definite article has scope over the first element only. This, it appears, results in an ungrammatical construction:

Close appositions (6)

389

a. *the poet of the decade Burns. b. *thepoet distinguished by the critics Burns. c. *the poet who supported the revolution Burns.

These observations, however, prove to be incorrect. As it turns out, the first element in these constructions can be modified, by prepositional phrases as well as by relative clauses, as shown in examples (7a) and (7b), respectively. (7)

a. Uhm let me bring in the Conservative leader of the county council Ken Thompson. [det] Burns WH be poet (relativisation, obi) => the Burns who is a poet => the Burns poet (relative reduction, opt) => the poet Burns (by attribute preposing, obi)

A problem with this analysis, recognised by Burton-Roberts himself, is that at no stage in the derivation is the element poet specified as definite (note in particular that (18c) would be ungrammatical with the definite instead of the indefinite article). The presence of the definite article remains, therefore, unaccounted for. Burton-Roberts does, however, answer the question of why the definite article appears. Thus, he explains, since it is Burns that is being modified, it is Burns that is being determined; just like any other name, when modified, it acquires a determiner. In this respect the construction the poet Burns does not differ from the expression in (19b), the only difference

396

Evelien Keizer

being the form of the modifier (adjective versus adjectival noun). Moreover, Burton-Roberts argues, such an approach accounts for the fact that these constructions cannot be indefinite: the head, being a name and therefore inherently definite, requires the definite determiner, thus ruling out such expressions as *a poet Burns or *a friend John in example (20a). (19) a. the poetAN BurnsPN, b. the confidentA BurnsPN, (20) a. * a poet Burns. b. a confident Burns. There are, however, two problems with this explanation. First of all, it is obviously only valid if the proper name is analysed as the head of the construction. Secondly, on Burton-Roberts's line of reasoning, any cooccurrence of a proper name and an indefinite determiner should be impossible. As shown in example (20b)), however, this is not the case. It is true that proper names are inherently definite; it is, however, equally true that when a proper name does not uniquely denote an entity, it loses its inherent definiteness. Thus we can have a confident Burns, a triumphant Tony Blair, a true Forsyte or even simply a Forsyte (see also examples (2) and (3))· Nor, as we have seen, is there any reason to assume that close appositions cannot be indefinite (see example (4)). Indeed, on an introductory use of the proper name, indefinite close appositions seem to be perfectly acceptable. Examples with the indefinite article are given in (21a-c); in (2Id) we find an instance of discourse introductory this. (21)

a. / have a friend John who's in Linguistics with me. (= (4b)) b. He also has to put up with a soppy elder brother Robert who is forever mooning about some girl or other and a sister Ethel who has all the brisk no-nonsense superiority of a true Wodehouse gel. (= (4c)) c. And supposing I take a value K, and I want to know if FXaffects that value. d. Oh I remember I was talking to this bloke Mark some sort of this really old friend of mine.

Close appositions

397

We may therefore conclude that defmiteness is not a defining feature of close appositions. In the representation in (16b), the determiner performs the same function as in any other noun phrase, and can as such be either definite or indefinite.

3.1.3. Type 2: NP + det + Ν When we apply this line of reasoning to close appositions of Type 2, Burns the poet, we are, however, faced with a problem. So far, it has been assumed that the definite determiner should be seen as fulfilling the same function as in normal NPs; i.e. as indicating the identifiability or familiarity of a referent. If, however, the second element does not have a referent, then how can we account for the presence of the definite article? As I have argued elsewhere (Keizer 1992a), a problem with many theories of defmiteness is that they are defined strictly in terms of definite reference. This is one of the reasons why such theories cannot explain the use of the definite article in so-called 'non-referential noun phrases', like the predicative elements in example (22). (22)

a. You 're the expert at this. b. My Comic Relief T-shirt'd be the ideal garment. c. I wasn't exactly the most model student here anyway so.

If, however, defmiteness is viewed as a discourse function, as a pragmatic property reflecting the assumed identifiability or familiarity of an entity for the hearer, and if discourse entities are regarded as mental concepts rather than objects in the real world, there is no reason why defmiteness should apply only to referring expressions in argument position. The nonreferential noun phrases in (22) also introduce discourse entities or mental concepts, namely the properties denoted by the nominal predicates. Where these properties can be presented as 'identifiable' for the hearer (activating a unique mental representation), use of a definite description is appropriate (sse also Section 3.3). In the same way, the second element in a construction like Burns the poet, though not referring to a particular person, can nevertheless be presented as containing familiar or inferrable information. In fact, N2 in these

398

Evelien Keizer

constructions typically functions to provide the hearer with a unique link between the proper name and his/her knowledge base. Consider in this respect example (23). This sentence is part of a rather lengthy passage dealing with a change of occupier at Number 10 Downing Street, with Mrs Thatcher moving out and the new prime minister, John Major, moving in. (23) Even Humphrey the Cabinet cat seems affected by the imminent change of occupier. Now, in all likelihood, the reader is not expected to know who Humphrey is; nor will (s)he be expected to be aware of the existence of a Cabinet cat. Nevertheless, the descriptive information, part of which is activated by the context, provides an unambiguous link between the entity referred to and this context. This accounts for the definiteness of N2, despite the fact that this element does not independently refer to an identifiable entity.4 (See also discussion in Section 3.3.)

3.2. Headedness Before moving on to the discourse uses of close appositions, let us briefly return to the question of headedness. As we have seen, opinions vary as to which of the two elements functions as the head; moreover, apart from the omissibility test, little evidence for headedness is generally provided. I will therefore apply some additional tests: at the semantic level I will look at the selection restrictions of the verb; at the syntactic level I will consider pluralisation and subject-verb agreement.5

3.2.1. Semantic considerations Selection restrictions of the verb are generally taken to hold between a verb and the head of a noun phrase. In most cases both elements of a close apposition are compatible with the selection restrictions of the verb. In some cases, however, one of the elements, when used on its own, violates these restrictions. Two examples are given in (24). In both examples, it is the descriptive element which complies with the selection restrictions: one can have a friend, or earn a name, but not a person.

Close appositions (24)

399

a. I have a friend John who's in Linguistics with me. (= (4b), (21a)) b. ... thus earning the animal the name emperor tamarin. b'. *... thus earning the animal emperor tamarin the name.

Note that in (24b) we cannot reverse the order of the elements, as shown in example (24b'). This may be taken to indicate that the two orders differ with regard to which element functions as the head.

3.2.2. Syntactic considerations Pluralisation Generally speaking, when a close apposition has plural reference, N l , the descriptive element, occurs in the plural, while N2, the proper name or uniquely describing element, commonly consists of two or more coordinated elements. Examples can be found in (25). (25)

a. their regimental slow march written by Johann Valentine Hamm for the Milanova sisters Teresa and Maria.

b. Those who would see this kind of picture in a conventional linguistics text book would really imagine little Cs and Vs, or even the words consonant and vowels or little phoneme symbols written under here.

If, as has been suggested in discussions on headedness, the head is the element that gets inflected (Zwicky 1985, 1993; Hudson 1987, 1993), it is Nl which functions as the head of the construction. This is supported by the fact that in those cases where a plural descriptive element is followed by a singular proper name, as in (26), overall reference here is plural.6 (26)

the biologists Huxley. (= (15a))

As shown in example (27), appositions of Type 2 cannot appear in the plural. This seems to support an analysis in which the proper name functions as the head: since proper names cannot be pluralised (unless they function

400

Evelien Keizer

as normal nominal predicates, in which case they are preceded by the definite article), and since the plural marker must attach to the head of a construction, constructions of this type cannot appear in the plural. (27)

*Huxley(s) the biologists; *Teresa and Maria the Milanova sisters.

Subject-verb agreement In constructions of Type 1 number agreement obtains between N1 and the finite verb. In most cases, both elements of close appositions show agreement with the verb. Occasionally, however, N1 appears in the plural, while N2 does not. As shown in example (28), it is N1 which in such cases agrees with the verb. (28)

a. The biologists Huxley were both equally brilliant, (cf. (15a); (26)) b. I have two friends John, who work for the same company.

Equally interesting are instances of Type lb involving a multiple but noncoordinated uniquely denoting element. Examples are given in (29). As shown in the primed examples, the multiple elements occur with a singular verb when used by themselves. This once again suggests that N1 functions as the syntactic head of the construction, as it is this element which exhibits number agreement with the verb.7 (29)

a. I think the uhm words unconditional withdrawal have/*has some uneasy echoes to some people. a'. 'Unconditional withdrawal' has/*have some uneasy echoes to some people. b. The initials J.H. stand/* stands for John Hamish. b'. J.H. stands/*stand for John Hamish.

3.3. Underlying representations In the Functional Grammar framework developed over the 1990s, the two types of close apposition discussed in this section might be distinguished by assigning them the following underlying representations:

Close appositions (30)

(31)

Type 1: det + Ν + Ή head/first restrictor: a. the poet Burns b. the word recession

401

^-construction descriptive element (dlxj: (f;: poet N ): (ζ: Burns Np )) (dlxj: (£: word N ): (ζ: recession N ))

Type 2: NP + det + Ή-construction proper name head/first restrictor: (dlxji ( i : Burns Np ): (dfj: poet N )) Burns the poet

The underlying representations in (30) reflect the non-referentiality of the two elements and the fact that the determiner has scope over both. The descriptive element functions as the first restrictor, the proper name as modifier (second restrictor). The representation in (31) reflects the fact that in these constructions the proper name functions as the first restrictor and that the second element is specified for defmiteness. Since the proper name, despite being used non-referentially, is uniquely determining, it remains intrinsically definite; as such, it does not trigger the definite determiner for the construction as a whole. 8 In Hengeveld's recent proposal for a Functional Discourse Grammar (FDG), however, defmiteness is assumed to operate at the interpersonal, rather than the representational level (Hengeveld 2004). This is indeed a plausible assumption; as argued in Keizer (1992a), defmiteness functions to indicate the assumed presence of an activated or retrievable (unequivocal) link between the intended referent and the knowledge base of the addressee. In other words, use of a definite expression is felicitous when an entity can be assumed to be familiar to or identifiable for the addressee on the basis of the communicative context of the utterance, which includes not only textual and situational information, but also (assumptions about) the long-time knowledge of the speech participants. However, where the representation given in (30) and (31) differ from most existing proposals in Functional Grammar is that, at the interpersonal level, defmiteness is believed to apply not to referential acts (R-acts), but also to ascriptive acts (T-acts). This is in line with Keizer's (1992a) proposal for the representation of non-referential expressions of copular constructions (see example (22)). As in the case of copular constructions, close appositions contain non-referential elements which may look like terms but which fulfil a property-assigning rather than a referential function. The fact that in a close apposition like Burns the poet the second element contains a definite determiner does therefore not automatically mean that it is a refer-

402

Evelien Keizer

ential expression: although defmiteness is typically associated with R-acts, its function (indicating the presence of an activated or retrievable link between two concepts at the cognitive level) may just as well apply to T-acts. In the latter case, this cognitive status of an element as (in)definite is normally not formally expressed. Where the expression in question has a nominal first restrictor, however, defmiteness may be expressed in the way typical for referential expressions (terms), i.e. through the presence of a definite or indefinite determiner. This gives rise to a non-prototypical term: one which has the formal properties of a term, but which lacks its typical referring function. In other words, the representations in (30) and (31) are to be interpreted as follows. Close appositions as a whole (typically) have a referential function (although they may, of course, be used non-referentially), and as such constitute R-acts at the interpersonal level. At the representational level the intended referent of the construction as a whole will then be symbolized by means of an x-variable (or rather, any type of variable, depending on the order of the entity referred to). Each close apposition does, however, consist of two non-referential elements. These constitute T-acts at the interpersonal level, which, at the representational level, are given content by means of a description of the properties assigned to the overall referent. These properties are zero-order order entities and as such represented by means of an f-variable (in accordance with Keizer 1992a,b). For a proper interpretation of the communicative content to be conveyed (symbolized by the variable C in Hengeveld's FDG model), an indication of the cognitive defmiteness status (as (un)familiar, (un)identifiable) of the various elements will be required, not only in the case of the overall R-act, but also for these Tacts, which together provide the descriptive information needed for the addressee to pick out the intended referent. This, then, accounts for the presence of a defmiteness operator at the f-level in the representation in (31).

4. A classification of uses This section will consider the different discourse functions close appositions may perform. Four different uses will be distinguished: the functionally identifying use, the descriptionally identifying use, the introductory use and the contrastive use.

Close appositions

403

4.1. The functionally identifying use At first sight the constructions in (32) may seem redundant: since we all can be assumed to know that 'four' is a number, 'blue' a colour and 'Algernon' a name, there seems no reason to add an element explicitly stating this information. What seems to justify this apparent redundancy is the fact that the uniquely denoting element (four, blue, etc.) does not appear in its prototypical function. Thus four is not used as a quantifier, blue does not function as an adjectival modifier, while the proper name Algernon is not used to refer to an individual but to the name itself. (32) a. b. c. d. e.

The number four is my lucky number. I've always liked, the colour blue. The letter Έ ' stands for euro. I don't care much for the name Algernon. I don't like to use the word 'drop-out'.

Consider in this respect the examples in (33), where N1 clearly has a disambiguating function. In (33c), for instance, omission of N1 will lead the hearer to interpret the description background paintings as referring to a set of objects, not to the description. The function of N1 is, therefore, to indicate the type of referent of the construction as a whole, with N2 specifying the name this referent goes by. (33)

a. But now even here people are beginning to come to terms with the word unemployment and the signs are it's going to get a lot worse before it gets any better. b. Even if you have never been here you may recognise the name Avignon. c. I really hate the word background paintings ...

As shown in example (34), the functionally identifying use does not normally allow for the use of indefinite determiners. This can be explained by the fact that the relation between N1 and N2 is exhaustive, and, once established, renders the reference unique. (34)

a. * A number seven is my lucky number.9 b. */ hate a word background paintings.

404

Evelien Keizer

4.2. The descriptionally identifying use The descriptionally identifying use is the most common use of close appositions. The term descriptionally identifying, as is indeed the notion itself, is taken from Declerck's (1988) classification of copular sentences.10 On this use, the descriptive element provides information which allows the hearer to link the referent of the construction to his/her 'knowledge base', thus forestalling questions like 'Who is he/she?' or 'Tell me more about him/her'.

4.2.1. Type 1: det + Ν + N(P) Some examples of constructions of Type la are given in (35). In (35a), the hearer may be aware of the fact that Tom Walkenshaw is the Jaguar boss, but this is not assumed to be the case. Omission of the part the Jaguar boss may therefore leave the hearer wondering who Tom Walkenshaw is and why he is mentioned here; indeed, it is the description the Jaguar boss which renders the passage coherent. At the same time it will be clear that without the proper name the result is not quite satisfactory either: even if the hearer cannot be assumed to be familiar with the name, it makes sense to introduce this name, if only to facilitate future reference. A similar example is given in (35b). (35) a. This is a really big home win for Jaguar and indeed for Silverstone because the Jaguar boss Tom Walkenshaw is of course the uh managing director of Silverstone Circuits the BRDC. b. The uh background uh to the uh matter which has been argued a and uh relied on in the Registrar's judgement uh is a confusing state of cross claims between the debtor Mr Daniel on the one hand a and Mr Sigrani, uh with whom he had dealings on the other. Since the main purpose of the descriptively identifying use is to link the referent of the apposition to the hearer's knowledge base, it is not surprising that in this use appositions are not likely to be indefinite. Clearly, unmodified indefinites cannot have the required anchoring function (Prince 1981: 236). One might expect indefinites with an anchoring function, like

Close appositions

405

the ones in (36), to be acceptable. However, as these anchored indefinites are necessarily modified, the result tends to be rather clumsy. (36) a. ?A friend of mine Andy may be able to help you. b. ?A mate at work Tim had the same problem.

4.2.2. Type 2: NP + det + Ν In appositions of Type 2 the descriptive element likewise functions to identify the referent by linking the proper name to the hearer's knowledge base. Here the definiteness of the second element indicates that, within the domain of discourse, this link is a unique one. This was illustrated before by means of example (23), here repeated as (37a). A similar example can be found in (37b), where the description the Finance Director helps the hearer to identify Mr Hunt by providing an unequivocal link between proper name and discourse. (37) a. Even Humphrey the Cabinet cat seems affected by the imminent change of occupier. b. Uh, he said that uh his uh feeling was that uh Mr Hook and uh the other guy who was there who uh was Mr Hunt the Finance Director were uh talking about two different things to him at this meeting that Mr Hook was saying that he was interested in the further expansion of the Female bus Ferndale business. c. And uh the pace should be injected by Lord Chalmer the front runner. A special case is (37c), where the proper name may be assumed to be familiar, having been mentioned before. In such cases, the descriptive element may be used to justify a claim made by the speaker (compare (35a)). Thus, the name Lord Chalmer can be assumed to be familiar; the descriptive element front-runner, however, functions to explain why it is that the speaker thinks that this horse should inject the pace. In other cases, the property denoted by N2 can be assumed to be identifiable on the basis of general knowledge. In (38a), for instance, the proper name, Moro, has been mentioned before and may therefore be assumed to

406

Evelien Keizer

be known to the reader. The descriptive element {loner) functions to provide further information about the referent, by supplying a property which not only activates a unique concept (the stereotypical loner), but which in addition provides a link with the preceding discourse. The same is true for the descriptive elements in (38b). (38) a. There are charges that these culminated in the kidnapping and execution of former Premier Aldo Moro, whose insistence on defying an American veto on admitting Communists into the Cabinet infuriated Washington. Defying Andreotti and the Christian Democrat right, Moro was on the way to Parliament with the deal fixed when he was kidnapped. William Colby, CIA chief of the day, claimed recently that his agency had spent nearly $100 million 'supporting politicians'. But Moro the loner was not for sale. b. "Ha! You put me off do you? " said our new visitor, taking a step forward and shaking his hunting-crop. "I know you, you scoundrel! I have heard of you before. You are Holmes the meddler." My friend smiled. "Holmes the busybody!" His smile broadened. "Holmes the Scotland Yard Jack-in-office!" (Arthur Conan Doyle, The Speckled Band)u Observe finally that this analysis also explains why close appositions with a uniquely denoting element do not allow for a descriptively identifying use, as shown in (39). (39) a. * Written on the door was four the number. b. */ have always detested Algernon the name. c. * Έ ' the letter here stands for euro. First of all, there is little use in providing further descriptive information about such entities as a number, name or letter. Moreover, it is unlikely that the information in the second element will function to link the referent to the discourse context.

Close appositions

407

4.3. The introductory use The introductory use of close appositions is closely related to the descriptionally identifying use. On the introductory use, however, the descriptive information does not link the referent to the previous discourse; instead it provides new information about the referent. This use therefore differs from the descriptionally identifying use in the pragmatic status of the descriptive element: in the descriptionally identifying use this element contains activated or inferrable information; in the introductory use it provides new information.

4.3.1. Type 1: det + Ν + NP An example of the introductory use of close appositions is the poet Burns in (40a), where the description poet is not in any way related to the previous discourse. The fact that Burns was a poet is not presupposed; nor is it asserted. Instead, the construction provides the kind of background information that a hearer may need to appreciate the relevance of the comment in the context. Similarly, in (40b), the description author is obviously unrelated to the preceding discourse, its function being to provide background information enabling listeners to understand the relevance of the utterance. (40) a. The poet Robert Burns lived here from 1786 to 1788. (= (14a)) b. An enquiry's begun into the sinking of a trawler by a nuclear sub off the Scottish coast and the author Roald Dahl has died at the age of seventy-four. Not surprisingly, indefinite appositions typically have an introductory use. In (41) for instance, the referent of the apposition is assumed to be unknown to the hearer; moreover, the description supplied does not allow the hearer to relate this referent unambiguously to his/her knowledge base. (41)

a. I have a friend John who' s in Linguistics with me b. Oh I remember I was talking to this bloke Mark some sort of this really oldfriend of mine

408

Evelien

Keizer

4.3.2. Type 2: NP + det + Ν The introductory use is not available for constructions of Type 2, where the descriptive element will always in some way be activated in or inferrable from the context. If such a relation is lacking, use of this construction becomes infelicitous, as shown in example (42). (42) a. *?Robert Burns the poet lived here from 1786 to 1788. (= (14b)) b. *?An enquiry's begun into the sinking of a trawler by a nuclear sub off the Scottish coast and Roald Dahl the author has died at the age of seventy-four.

4.4. The contrastive use In each of the construction types discussed the descriptive element can be used contrastively. Some examples are given in example (43) (where contrastive stress is indicated by the use of small capitals). (43)

a. Who are you referring to? The CRITIC Paul Jones or the AUTHOR Paul Jones? b. It is Algernon the NAME I don 7 like, not Algernon the PERSON.

In each construction type, the contrast expressed may be of three kinds. In the first place, there is the contrast between two entities of the same kind (e.g. two persons, as in (43a)). Secondly, the contrast may be between entities of a different kind, as in example (43b), where a name is being distinguished from a person. In the third type of contrastive use, illustrated in (44), both constructions seem to refer to the same entity (person), contrasting different properties (roles) of this person. Note, however, that in terms of discourse referents (and mental representations), we are dealing with two separate entities which share the name of Welles. (44)

(In a conversation on Orson Welles:) The ACTOR Welles is fascinating·, the DIRECTOR Welles is absolutely brilliant.

Close appositions

409

The question that arises is whether contrastive constructions of this kind still qualify as appositions. According to the definition given in Section 2.5, one of the elements must be a proper noun or some other uniquely denoting element. It will be clear, however, that in the examples in (43) and (44) the proper names are no longer uniquely denoting, but act more or less like common nouns. The difficulty is, of course, that if they are not appositions, then what are they? Since on the whole they seem to have more in common with appositions than with any other category, we will for the moment continue to regard them as appositions, but of a less prototypical kind. In addition, it is, of course, equally possible for the proper noun or uniquely describing element to be used contrastively, in which case two entities are being contrasted which share the same property. (45)

a. Who are we going to discuss today? The poet BURNS or the poet POPE?

b. but their the word REGRET seems to be the most favourite diplomatic word knocking around rather than ANGER, ... c. He avoids the word PHONEME preferring

SOUND and this

explains

the title of the article. Finally, the contrastive use of appositions does allow for an indefinite determiner, although, again, such constructions may be assumed to be rare. Here, too, contrast may be expressed between the descriptive elements (example (46a)), or between the unique elements (example (46b)). (46)

a. I have a FRIEND John, but no COLLEAGUE John. b. What we have here is a constituent ON THE MAT, not a constituent THE MAT.

4.5. Underlying representations It will be clear that what distinguishes the three uses of close appositions is the pragmatic function of the two elements and/or the construction as a whole. The relevant distinctions are:

410 -

Evelien Keizer

Completive Focus (NewFoc) versus Given/Subtopic (descriptive element) Presentative Focus (NewTop) (typical function of Introductory / Descriptionally identifying uses) versus GivenTop (in Contrastive use) Contrastive versus Non-contrastive Focus (either element)

The various possibilities are listed in Table 1 below (note that where no pragmatic function is indicated, various functions are possible). As in Section 3.3, the representations have been given in the conventional FG-format for the sake of convenience, but comprise information specified at both the interpersonal and the representational levels. The procedure is, in fact, very similar to that described in Section 3.3. Thus, like definiteness, the pragmatic functions of Topic and Focus apply at the interpersonal level of an utterance. The pragmatic status of an element may, of course, be indicated in a variety of ways: by choosing a particular (Focus) construction, by means of word order (PI position, End Focus), by prosodic means (focal, emphatic or contrastive stress), or by formal means (e.g. use of a pronoun or a definite description), and often through a combination of two or more of these devices. More importantly, however, these functions may apply to the content of both a referential and an ascriptive act. 12 Just like information concerning the definiteness status of the elements in question, information concerning the pragmatic status of the close apposition as a whole, as well as of its composite elements, is determined at the interpersonal level, with the performance of the relevant R-acts and T-acts. The representations in Table 1 are to be seen as containing the information needed for the execution of these acts. Where relevant, information concerning the (assumed) familiarity or identifiability of the elements in question, as well as information pertaining to their role as new, inferrable or established discourse entities, is therefore given for each element, irrespective of whether this element fulfils a referring or property-assigning function.

412

5.

Evelien Keizer

Conclusion

This paper has been an attempt at developing a unified and consistent functional account of the form and use of two types of close apposition. One important conclusion is that, contrary to what has been assumed in most existing treatments, neither nominal element of a close apposition has independent reference, and that, as such, the frequently applied test of omissibility as a way of establishing which element functions as the (semantic or syntactic) head of a close apposition. Instead, it was suggested that close appositions be analysed as consisting of two predicating nominals both of which fall within the scope of a determiner (if present). In addition, it was established that, although superficially similar, a distinction has to be made between (at least) two main types of close apposition. The discussion of (in)definiteness in Section 3.1 and of headedness in Section 3.2 provided additional evidence to support this assumption. This eventually resulted in the representations in Section 3.3, which provides different analyses of the two types of close apposition, involving different head-modifier relations. Section 4 subsequently dealt with the discourse functions of the two types of close apposition distinguished. The result was a classification of four uses of close appositions, based on the communicative function of the two elements. In addition, the two types of close apposition were shown to differ with regard to the number of uses they allow. By examining relevant examples in their discourse context, it could be demonstrated that the form of a (particular type of) close apposition is not arbitrary: within a given discourse situation, it is the most efficient linguistic means available to the speaker to refer to, introduce and/or describe a particular discourse entity.

Notes 1.

This paper has been written as part of the AHRB research project 'The English Noun Phrase: an empirical study', carried out at the Survey of English Usage, University College London. I would like to thank the project supervisor, Bas Aarts, for useful comments on an earlier version of this paper. Thanks are also due to the editors of the present volume and one anonymous

Close appositions

2.

3.

4. 5.

6.

413

reviewer for further comments and suggestions for improvement. Any remaining errors are, of course, entirely my own. The British component of the International Corpus of English (ICE-GB) is a fully tagged, parsed and checked one-million word corpus of written and spoken English, compiled and grammatically analysed at the Survey of English Usage, University College London, between 1990 and 1998 (see e.g. Nelson et al. 2002). According to Acuna-Farina (1996: 37), 'we do not have, at least naturally, the tall poet Burns because we do not have the tall Burns'. Note, however, that there is nothing unnatural about the combination of the adjective tall and a proper noun: (i) Behind him his cousin, the tall George, the son of the fifth Forsyte, Roger, had a Quilpish look on his fleshy face ... (John Galsworthy, The man of property) This also enables us to account for the fact that close appositions of this type cannot be indefinite; see section 4.3.2. There is large body of psycholinguistic literature on the question of which factors influence the number of the finite verb, in particular in constructions with collective nouns (e.g. crowd, fleet, police) and constructions containing more than one noun, where at least two nouns differ in number (e.g. Bock and Miller 1991, Bock and Eberhard 1993, Nicol 1995, Vigliocco et al. 1996, Eberhard 1997, Berg 1998, Vigliocco and Nicol 1998, Bock et al. 1999, Franck et al. 2002). Findings, however, are far from unequivocal and various determining factors have been suggested (including the semantic number of the subject NP, the syntactic distance between the verb and the controlling noun and the linear distance between verb and noun). Although this clearly shows that subject-noun agreement should not be relied on blindly in determining headedness, it may nevertheless be a useful test for headedness, provided that is used in combination with other evidence and is applied with the necessary caution. For a more detailed discussion, see Keizer (in prep.). For linguistic discussions of subject-verb agreement, see e.g. Zandvoort (1961), Morgan (1972), Moravcsik (1978), Corbett (1979, 1994), Quirk et al. (1985), Dowty and Jacobson (1988), Fries (1988) and Reid (1991). Where the uniquely defining element allows for pluralisation, both elements appear in the plural (example (ii)), which suggest that the lack of a plural inflection in (25a) and (26) is simply due to the fact proper names, as a rule, cannot be pluralised. (ii) He's an in a very inadequate and dangerous source, to use if we 're trying to understand, what the ancient Gauls, and their cousins the Britons and so on were really like.

414

Evelien Keizer

Moreover, as pointed out by Burton-Roberts (1975: 403), plural reference can also be achieved by means of a plural proper name and a singular N1 (cf. iiia): (iiia) the biologist Huxleys (iiib) the lady presidents Here, it would seem, it is the second element which qualifies for headedness. Observe, however, that this construction seems to behave more like a compound (cf. iiib). 7. Note that in other, related, aspects, too, N1 seems to function as the (syntactic) head of the apposition. In (iva), for instance, it is the descriptive element which agrees in number with the numeral or quantifier, while in (ivb), it is again the plurality of N1 which licenses use of the expressions together and each other. (iva) I know two/all biologists Huxley. (ivb) The biologists Huxley work together / despise each other. 8. Compare in this respect the examples in (8), which lack a definite determiner for the same reason. 9. As soon as an indefinite article is used, the construction can only be interpreted as referring, not to the number itself, but to some object described as (number) seven: (v) #A (number) seven is missing. 10. The fact that we borrow Declerck's (1988) notion of descriptionally identifying does not imply any direct relationship between close appositions and copular constructions. Unlike some existing theories, I do not regard appositions as derived from constructions containing a copular relative clause. The similarity between close appositions and copular constructions is due to the fact that both contain two nominal predicates applying to the same entity. 11. In some cases the second element loses its discourse connection to become more of a general characterisation, serving as a nickname or as part of the proper name, as in Edward the Confessor or William the Conqueror. 12. This is clearest in the case of Focus, which function can obviously be assigned to referring and predicating elements alike. For the function Topic, it depends on the exact definition given of Topic. For a discussion see Mackenzie and Keizer (1991).

References Acuna-Farina, Juan Carlos 1996 The puzzle of apposition: on so-called appositive structures in English. Universidade de Santiago de Compostela.

Close appositions

415

1999 On apposition. English Language and Linguistics 3 - 1: 59-81. Berg, Thomas 1998 The resolution of number conflicts in English and German agreement patterns. Linguistics 36: 41-70. Bock, J. Kathryn and Kathleen M. Eberhard 1993 Meaning, sound and syntax in English number agreement. Language and Cognitive Processes 8: 57-99. Bock, J. Kathryn and Carol A. Miller 1991 Broken Agreement. Cognitive Psychology 23: 45-93. Bock, J. Kathryn, Nicol, Janet and J. Cooper Cutting 1999 The ties that bind: creating number agreement in speech. Journal of Memory and Language 40: 330-346. Burton-Roberts, Noel 1975 Nominal apposition. Foundations of Language 13: 391-419. 1994 Apposition. In The Encyclopaedia of Language and Linguistics, Vol. I., R.E. Asher and J.M.Y. Simpson (eds.), 184-187. Oxford: Pergamon Press. Corbett, Greville G. 1979 The agreement hierarchy. Journal of Linguistics 15: 203-224. 1994 Agreement. In The Encyclopaedia of Language and Linguistics, Vol. I., R.E. Asher and J.M.Y. Simpson (eds.), 54-60. Oxford: Pergamon Press. Corbett, Greville G., Norman M. Fraser and Scott McGlashan (eds.) 1993 Heads in grammatical theory. Cambridge: Cambridge University Press. Curme, George O. 1931 Syntax. Boston: Heath and Co. Declerck, Renaat 1988 Studies on copular sentences, clefts and pseudoclefts. Leuven: Leuven University Press / Fortis Publications. Dowty, David and Pauline Jacobson 1988 Agreement as a semantic phenomenon. In Proceedings of the Fifth Eastern States Conference on Linguistics, J. Powers, & K. de Jong (eds.), 95-108. Columbus: Ohio State University. Eberhard, Kathleen M. 1997 The marked effect of number on subject-verb agreement. Journal of Memory and Language 36: 147-164.

416

Evelien Keizer

Fortescue, Michael, Peter Harder and Lars Kristoffersen (eds.) 1992 Layered structure and reference in a functional perspective, Amsterdam: Benjamins. Francis, W. Nelson 1958 The structure of American English. New York: Ronald. Franck, Julie, Gabriella Vigliocco and Janet Nicol 2002 Subject-verb agreement errors in French and English: the role of syntactic hierarchy. Language and Cognitive Processes 17: 371-404. Fries, Udo 1988 The crew have abandoned the ship. Concord with collective nouns revisited. Arbeiten aus Anglistik and Amerikanistik 13, 98-104. Greenberg, Joseph Η. (ed.) 1978 Universals of human language, Vol. 4.: Syntax. Stanford: Stanford University Press. Haugen, Einar 1953 On resolving the close apposition. American Speech 28, 165-170. Hawkins, John A. 1978 Definiteness and indefiniteness. London: CroomHelm. Hengeveld, Kees 2004 The architecture of a Functional Discourse Grammar. In A new architecture for Functional Grammar, J.Lachlan Mackenzie and Maria Ä. Gomez-Gonzälez (eds.). Berlin: Mouton de Grayter. Hockett, Charles F. 1955 Attribution and apposition in English. American Speech 30: 99-102. Hudson, Richard A. 1987 Zwicky on heads. Journal of Linguistics 23: 109-132. 1993 Do we have heads in our minds? In Corbett et. al., 266-291. Keizer, M. Evelien 1992a Reference, predication and (in)definiteness in Functional Grammar. A functional approach to English copular sentences. Ph. D. diss., Vrije Universiteit Amsterdam. 1992b Predicates as referring expressions. In Fortescue et al. (eds), 1-27. in prep. Aspects of the English Noun Phrase: structure, cognition and communication. Lee, Donald W. 1952 Close apposition: an unresolved problem. American Speech 27: 268275. Mackenzie J. Lachlan and M. Evelien Keizer 1991 On assigning pragmatic functions in FG. Pragmatics 1 - 2: 169-215.

Close appositions

417

Meyer, Charles F. 1992 Apposition in Contemporary English. Cambridge: Cambridge University Press. Moravcsik, Edith A. 1978 Agreement. In: J.H. Greenberg (ed.), 331-374. Morgan, Jerry L. 1972 Verb agreement as a rule of English. Papers from the Regional Meeting of the Chicago Linguistic Society 8: 278-286. Nelson, Gerald, Sean Wallis and Bas Aarts 2002 Exploring Natural Language: Working with the British Component of the International Corpus of English. Amsterdam: John Benjamins. Nicol, Janet 1995 Effect of clausal structure on subject-verb agreement errors. Journal of Psycholinguistic Research 24: 507-516. Reid, Wallis 1991 Verb and number agreement in English. London: Longman. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik 1985 A Comprehensive Grammar of the English Language. London: Longman. Sopher, H. 1971 Apposition. English Studies 52: 401-412. Vigliocco, Gabriella, Brian Butterworth and Merrill F. Garrett 1996 Subject-verb agreement in Spanish and English: the role of conceptual factors. Cognition 61: 261-298. Vigliocco, Gabriella and Janet Nicol 1998 Separating hierarchical relations and word oder in language production: is proximity concord syntactic or linear? Cognition 68: Β13B29. Zandvoort, Reinard W. 1961 Varia Syntactica. In Language and Society. Essays presented to Arthur M. Jensen on his seventieth birthday, Frank Behre and C.A. Bodeisen (eds.), 193-203. Copenhagen: Det Berlingske Bogtrykkeri. Zwicky, Arnold 1985 Heads. Journal of Linguistics 21:1 -29. 1993 Heads, bases and functors. In Corbett et al. (eds.), 192-315.

Inversion and the absence of grammatical relations in Plains Cree Arok Wolvengrey

1.

Introduction

The analysis of Plains Cree data in this paper is meant as an illustration of a theoretically possible, but descriptively under-represented language type wherein the syntactic functions of subject and object are neither necessary nor delineated within the grammar. Instead, the two levels of pragmatic and semantic functions, as represented in the Algonquian - or even Creespecific - combination of person and direction marking, suffice to disambiguate participants without recourse to a third level of grammatical relations. The key to this system is the way in which person-marking is accomplished separate from but linked to the interaction of pragmatic and semantic role hierarchies. The formal instantiation of this link is known as the Direct-Inverse system, which has not generally been identified with "case-marking", due mostly to a formalist bias towards the traditional definition of case-marking, but which functionally serves the equivalent purpose of role-indexing. As preamble to the data, then, a brief introduction of some of the pertinent details of the Cree verbal person-marking system is necessary.

1.1. Person-marking Analyses of Cree and the Algonquian languages have generally not included any discussion of a case-marking system. Given the traditional definition of "case-marking", this is perhaps justified, since there is no specific marking on nouns that can be used to identify the difference between nominative vs. accusative or ergative vs. absolutive, etc. Neither full nominals nor independent pronominals nor person-marking affixes bear any marking specific to any core semantic function of the verb. Nevertheless,

420

Arok Wolvengrey

the person marking affixes do play an important part in establishing these roles. In order to illustrate the person-marking affixes, another important aspect of the Cree verbal system requires explication. Cree (and Algonquian) verbs have traditionally been divided into four subtypes, according to the criteria of transitivity and the animacy of one of the key participants. This division is illustrated in table 1. Table 1. Cree verb subclasses Transitivity

Inanimate

VII Inanimate Intransitive

VTI Transitive Inanimate

Animate

Animacy

Transitive Intransitive (animacy describes the (animacy describes the goal / 'object') single participant)

VAI Animate Intransitive

VTA Transitive Animate

Thus, intransitive verbs are of two types, depending on whether the lone participant is inanimate (VII) or animate (VAI), while transitive verbs are similarly divided into two subtypes, depending on whether the goal is inanimate (VTI) or animate (VTA). These four verbal subclasses are illustrated in (l)-(4) below: (1) (2)

VE VAI

pikopayin. pikopayiw.

(3) (4)

VTI VTA

pikonam. pikonew.

'it is broken.' 's/he is broke (i.e. short of cash); it (animate) is broken.' 's/he broke it' 's/he broke him/her or it (animate)'

It has been noted that the animacy agreement with intransitive 'subjects' and transitive 'objects' is suggestive of an ergative patterning (cf. Hewson 1987; Campana 1989), but this falls naturally from the criteria of classifi-

Inversion and the absence of grammatical relations in Plains Cree

421

cation: transitive verbs must necessarily and logically take an animate actor.2 Thus it is only the animacy of the goal which can fluctuate. In each of these four verb types, person-marking is obligatory and occurs in the form of bound pronominals (sometimes misidentified as mere verb agreement), which renders independent pronouns redundant and optional (which is to say that their occurrence is free to be used for pragmatic emphasis). Even in such instances cross-linguistically, in which participants are coded merely by person-marking affixes, it is usually the case that these affixes will specify both the person involved as well as the syntactic role of the participant, and thus participate in the case-marking system of the language. In Cree, however, the person-marking affixes are merely that, markers of person, and indicate neither semantic nor syntactic role. This can be demonstrated by the following examples (in which VII verbs will be ignored, since their person-marking possibilities are much reduced due to the absence of all but inanimate third person arguments). Examples will be given to show the agreement patterns of first person plural exclusive (lp) and third person plural (3p) participants. The first examples are from the VAI paradigms.3 (5)

a.

ninestosinän. ni- nestosi-nan 1 VAI:be tired lp 'we are tired.'

b.

nipimipahtänän. ni- pimipahtä- -nan 1 VALrun lp 'we run.'

c. nestosiwak. nestosi-(w)ak VAI:be tired 3p 'they are tired.' d. pimipahtäwak. pimipahtä-(w)ak VAI:run 3p 'they run.'

422

Arok Wolvengrey

The lp forms in (5a) and (5b) show that the discontinuous morpheme ni-nän is used whether or not the verb is active (e.g. pimipahtä- 'run') or stative (e.g. nestosi- 'be tired'). Similarly, the 3p inflection -(w)ak occurs in both (5c) and (5d) ([w] is given in brackets as it can be separated as the third person singular marker). This immediately eliminates the possibility of an Active-Stative (or Split Intransitive) case-marking system.4 In such systems, intransitive participants or "subjects" receive differential marking (hence the name "split intransitive") dependent on the semantic role of the participant. The most common roles differentiated are agents or "active" participants and patients (or undergoers, themes, experiencers, processed, zero, etc.) or "stative" participants (hence the name "active-stative"). In the next examples, the same pronominal morphology occurs in the VTI paradigms: (6)

a. nipikonenän. ni- pikon1 TI:break s.t. 'we break it.'

(e)nan lp

b. nikisihtänän. ni- kisihtä1 TLfmish s.t. 'we finish it.'

nan lp

c.

pikonamwak. pikonTI:break s.t. 'they break it.'

d. kisihtäwak. kisihtäTI:fmish s.t. 'they finish it.'

(amw)ak 3p

(w)ak 3p

The verbs in (6) represent two subclasses of VTI verbs, both transitive with inanimate objects. The difference occurs in the morphology rather than the syntax. The verb in (a) and (c), pikon-, is considered a VTI class 1, or the normal VTI type, as it exhibits some slight differences from the VAI type morphologically (to be attributed to the historical derivation of these

Inversion and the absence of grammatical relations in Plains Cree

423

forms). In contrast, the verb in (b) and (d), kisihtä-, is a transitive verb which has surface morphology completely identical to VAI verbs (cf. the verbs given in (5) above). 5 Regardless of the very slight differences in these paradigms, the same pronominal morphology is present as the "subject" of these VTI verbs as was found for the "subject" for the VAI verbs given in (5) above. Thus far, then, it might appear that we have an accusative system, but more examples are required. The examples in (7) illustrate the agreement of both l p and 3p participants interacting in a transitive animate or VTA verb. (7)

a.

b.

niwicihänänak. niwicihä1 VTA:help s.o. DIR 'we help them.'

nänlp

ak 3p

niwicihikonänak. niwicihiko1 VTA:help s.o. INV 'they help us.'

nänlp

ak 3p

Since the verb can now fully index two animate participants, the same verb can be used to illustrate both l p and 3p participants. What is interesting and diagnostic of the Cree person-marking system is that, regardless of the role of the participant, the same person indexors found elsewhere continue to be in use in the TA paradigm. In other words, the familiar discontinuous l p morpheme ni- -nän is found in the same position in both (7a) (where it indicates the "subject" or actor) and (7b) (where it indicates the "object" or goal), while the third person pluralizer -ak similarly occurs in the same form and position in both examples despite a change from goal (7a) to actor (7b) semantically. There is, then, absolutely no formal difference between the way l p and 3p participants are marked in any of the paradigms surveyed thus far - VAI, VTI, and VTA - and so, no formal differences based on semantic (e.g. agent vs. goal/theme) or syntactic (i.e. subject vs. object) role.

424

Arok Wolvengrey

2. Hierarchical Relations in the Cree Direct-Inverse What does enable us to assign a particular semantic role to each participant is a "direction-marking" morpheme, completely separate from the person markers, which specifies which of the two participants is the actor/agent and which is the goal/theme. This involves the interaction or "alignment" of two hierarchies, a Person (or Pragmatic/Topicality) Hierarchy and a Semantic Function Hierarchy. These are similar, though not identical, to the person and semantic function hierarchies most commonly cited in Functional Grammar (cf. Dik 1997), requiring some Algonquian-specific modifications. Following the cross-linguistic Person Hierarchy, cited in (8), Cree speech-act participants outrank third persons as expected, but the Algonquian Person Hierarchy (ΑΡΗ), given in (9), requires that second persons outrank first persons. (8)

(9)

The Person Hierarchy Speech Act Participant (Dik 1997: 36) The Algonquian Person Hierarchy 2 -> 1 - ^ 3

Non-Participant

3'

The details of the Algonquian ranking of second above first persons will not be discussed here. It is important to note, however, that while by no means can a universally preferred ranking of first and second persons be determined, this particular ranking is reminiscent of that given by Myhill (1988, cited in Dik 1997:38) in his discussion of Spanish "clitic climbing". Similarly, the separation of non-participants or third persons into "proximate" (3) and "obviative" (3') is another Algonquian-specific sub-division which functions in disjoint reference and is determined largely by pragmatic factors. Obviation is vital to Cree and Algonquian grammar in general, and plays an especially important role in conjunction with DirectInverse morphology. Therefore, a brief introduction to this phenomenon will be necessary here, facilitating a better understanding of further examples cited below in section 3.1. In any situation in which two or more third person participants are present only one can be proximate (typically the most discourse-central or topical participant) while all others must be marked as obviative. Changing either the assignment of obviative status or the direction of the verb (as

Inversion and the absence of grammatical relations in Plains Cree

425

discussed further below) will result in a shift in semantic role interpretation, while shifting both obviation and direction will leave the semantic interpretation identical, thus illustrating the pragmatic uses of these Algonquian grammatical devices. The following examples illustrate the interaction of obviation with direction marking. (10)

a. cäniy ki-wicihew meriwa. J klwicih -e 3 TNS VTA:help s.o. DIR 3-3' 'Johnny helped Mary.'

-w 3

b. cäniwa ki-wicihew meriy. J-wa kiwicih-e 3' TNS VTA:help s.o. DIR 3-3' 'Mary helped Johnny.' c. cäniy ki-wicihik meriwa. J kiwicih-ikw 3 TNS VTA:help s.o. INV 3'-3 'Mary helped Johnny.'

M-wa 3'

-w 3

Μ 3

(-w) M-wa 3 3'

d. cäniwa ki-wicihik meriy. J-wa kiwicih-ikw 3' TNS VTA:help s.o. INV 3'-3 'Johnny helped Mary.'

(-w) Μ 3 3

In (10), the shift of obviation from meriwa in (a) to cäniwa in (b) results in a reversal of semantic role interpretation. The difference between (a) and (c) is a shift in direction, but this similarly reverses the semantic role interpretation. Only where both obviation and direction have been changed (as between (a) and (d)), does the semantic interpretation remain unchanged. The difference between (a) and (d) (or for that matter (b) and (c)) is purely a pragmatic one, based on the speaker's choice of proximate / topic / viewpoint assignment.

426

Arok Wolvengrey

The examples in (10) not only illustrate Cree obviation, but also show the necessary interaction of the Algonquian Person Hierarchy with semantic functions as ranked in the Algonquian Semantic Function Hierarchy, given in (12). Again, this is not precisely equivalent to the cross-linguistic hierarchy cited in Dik (1997; see (11a) below) since Algonquian languages, and certainly Cree, consistently rank the animate dative (recipient, benefactive, etc.) above the not-necessarily-animate theme.6 This is closer to Givon's (1984) semantic case-role hierarchy, cited in (1 lb). (11)

(12)

The Semantic Function Hierarchy a. Agent Goal (Theme) Recipient Dik (1997: 37) b. Agent -> Dative Patient Givon (1984: 134)

Beneficiary

...

The Algonquian Semantic Function Hierarchy Agent -> Recipient/Benefactive -> Theme

Note, however, that Givon (1984: 139) refers to his case-role hierarchy as a "topic hierarchy of the major semantic case-roles", belying an underlying interaction with or modification by some other, pragmatically-based hierarchy. This must clearly be an Animacy Hierarchy of the type cited in Dik (1997; see (13) below), for the Algonquian animacy-based gender system is perhaps the most important aspect of Cree grammar and the simplified animacy hierarchy given in (14) functions in virtually ever Cree utterance. (13)

(14)

The Animacy Hierarchy human -> other animate Dik(1997: 37)

inanimate force

inanimate

Cree Animacy Animate Inanimate

For the purposes of illustrating the interaction of the Algonquian Person and Semantic Function Hierarchies, the primacy of theme/patient versus recipient/benefactive will be ignored. All interactions will simply be represented as between agent and goal. The interaction of a typical direct construction is given in (15a), where the higher-ranking person fulfills the

Inversion and the absence of grammatical relations in Plains Cree

427

higher-ranking semantic role (i.e. agent), while the lower-ranking person fulfils the lower-ranking semantic role (i.e. goal). (15)

Interaction of Algonquian Person (Topicality) and Semantic Hierarchies a. Prototypical Topicality SAPs: lp 3p Direct=

Agent -» Prototypical Agency b.

Goal

Prototypical Topicality SAPs: lp 3p Inverse Agent ->· Goal Prototypical Agency

In contrast, when the higher-ranking person fulfils the lower-ranking semantic role, and the lower-ranking person fulfils the higher-ranking semantic role, inverse marking is required (15b). The term "alignment" has been used to describe this system (cf. Blain 1998), with the situation calling for direct marking being referred to as "aligned", and the inverse being invoked when the hierarchies are not properly aligned or mis-aligned. Here, the traditional terms of direct and inverse will be preferred. The conclusion that we can reach from these observations is that the direct-inverse system, paired with Cree person-marking, is the functional equivalent of a nominal case-marking system. However, other conclusions have been reached in previous analyses.

3.

(Mis)Identification of Grammatical Relations/Syntactic Functions

The inverse has been variously and contradictorily analyzed as an active (e.g. Dahlstrom 1991, Wolfart 1991) and a passive (e.g. Jolley 1982, Rho-

428

Arok Wolvengrey

des 1976). As one example, Dahlstrom (1991) provided evidence in support of her analysis of the inverse as an active construction with tests for recognizing both actor as subject and goal/patient as object. While "subject" tests involve both inverse and unspecified actor data (see further below), the important test for "object" is specific to the VTA inverse. For the purposes of this section of the paper, then, only the status of a grammatical object will be investigated. And though the following data was originally used by Dahlstrom to argue for an analysis of the inverse as an active construction, it can, in fact, be shown that the data does no such thing. Dahlstrom's (1991) "quantifier float" test for objecthood rests on the observation that a quantifier, such as kahkiyaw, 'all' in (16), must be construed with the object (i.e. patient or goal) of a transitive verb. (16) piyisk kahkiyaw mestinam otayäna;... piyisk kahkiyaw mestin-am o(t)IPC QNT VTI 3 3P Finally all use s.t. up (3-0') 'At last he had spent all his belongings.'

-ayänNDI possessions

-a O'p

In (16), the association of the quantifier with the "object", otayäna, is obviously unambiguous since only the "object" is plural. Therefore, only the "object" could potentially be modified by such a quantifier.

3.1. maskwa(k) examples However, Dahlstrom includes in her argument two examples in which ambiguity does exist (i.e. both participants are potentially plural). The first is reproduced here as (17),7 while the second, example (18), will be repeated and discussed more fully as (27) in section 3.2 below. (17)

nisto nipahewak maskwa näpewak. nisto nipah- -e -wak maskw-a näpewNUM VTA DIR 3p NA 3' NA three kill s.o. 3p-3' bear(s) men 'the men killed three bears.' •'three men killed a bear/bears.' (Dahlstrom 1991:83)

-ak 3p

Inversion and the absence of grammatical relations in Plains Cree (18)

kahkiyaw säkihikwak kahkiyaw säkihQNT VTA all love

429

otänisiwäwa iskwewak. -iwäw -tänis-ikw -(w)ak oNDA 3pP INV 3p 3P daughter(s) 3'-3p

iskwew- -ak NA 3p women 'their daughters love all women.' (i.e. 'all women are loved by their daughters.') '*all their daughters love the women.'(Dahlstrom 1991:87)

-a 3'

In (17), following the grammaticality judgements of Plains Cree speakers, the quantifier (nisto, 'three') can only be construed with the goal (maskwa, 'bear(s)') despite the fact that the actor is also plural. Furthermore, nipahewak is a direct VTA unambiguously identified as an active verb in which the goal is equated with "object". It is this identification of the goal with "object", and the fact that the quantifier only construes with the goal, that allows Dahlstrom to extend the analysis of goal as object to the inverse. The example in (18) shows that the quantifier (kahkiyaw, 'all') can only be construed with the inverse goal (iskwewak, 'women'), again following native speaker grammaticality judgements. Thus, it appears from these two examples that the quantifier can only be associated with the goal of a VTA verb, regardless of its status as obviative (maskwa) or proximate (iskwewak) or of the status of the verb as direct or inverse. For Dahlstrom, then, this illustrates that the goal is the object in both direct and inverse VTA verb constructions. However, this can be disproven. A simple test exists which can illustrate that the interpretation of a floated quantifier has nothing whatsoever to do with a participant's status as goal (or actor). This entails modifying the examples in (17) and (18) by reversing the direction markers on the TA verbs. Thus, in (19), for instance, the direct theme marker present in (17) has been changed to an inverse theme, with no other changes occurring between the two constructions. This has the effect of reversing the semantic roles of the participants, i.e. changing the status of actor and goal. If Dahlstrom's analysis is correct, the quantifier must now construe with the new "object" or goal. This is not the case.

430

Arok Wolvengrey

(19)

nisto nipahikwak maskwa näpewak. nisto nipah-ikw -wak maskw-a NUM VTA INV 3p NA 3' three kill s.o. 3'-3p bear(s) 'three bears killed the men.'(i.e. 'the men were bears.') '*bears killed three men.'

näpew-ak NA 3p men killed by three

In (19), then, we have the exact opposite situation to that found by Dahlstrom with examples like (17). Similarly, if we modify the example in (18) by changing the direction marking, the result is as given in (20) where the quantifier must still construe with iskwewak even though it is now the actor of the direct-marked verb. (20)

kahkiyaw säkihewak otänisiwäwa iskwewak. kahkiyaw säkih-e -(w)ak o- -tänis-iwäw -a QNT VTA DIR 3p 3P NDA 3pP 3' All love s.o. 3p-3' daughters iskwew- -ak NA 3p women 'all women love their daughters.' ('*women love all their daughters.')

With data such as the examples in (17) and (18) alone, Dahlstrom concluded floating quantifiers associate only with goals and that the inverse is therefore active with goal as object. However, if only examples such as (19) and (20) had been investigated, the conclusion would have been the opposite; that floating quantifiers provide a test for subjecthood rather than objecthood (and that, as a consequence, perhaps, the inverse would again have been analyzed as a passive of sorts). The fact that all of these examples are grammatical in Cree suggests that the status of a participant as actor or goal has no necessary bearing on the interpretation of a floating quantifier. Data such as this cannot be used as a test for subject or object at all. Floating quantifiers are construed with goals in (17) and (18), but with actors in (19) and (20). Floating quantifiers are construed with proximate participants in (18) and (20), but with obviative participants in (17) and (19). Additionally, word order also fails to

Inversion and the absence of grammatical relations in Plains Cree

431

have any bearing on interpretation. In (17) and (19), it is the participant immediately following the verb (and therefore closer to the floating quantifier), while in (18) and (20), it is the participant in sentence-final position (further away from the quantifier) that associates with the quantifier. Further modifications of these important examples serve to reinforce the observation that word order appears irrelevant. Reversing the order of the two NPs in sentences (17) and (19) yields (21) and (22) respectively. (21)

(22)

nisto nipahewak näpewak maskwa. nisto nipah-e -wak näpewNUM VTA DIR 3p ΝΑ Three kill s.o. 3p-3' men bear(s) 'the men killed three bears.' ('*three men killed a bear/bears.')

-ak maskw- -a 3p NA 3'

nisto nipahikwak näpewak maskwa. nisto nipah-ikw -wak näpewNUM VTA INV 3p NA Three kill s.o. 3'-3p men 'three bears killed the men.' (i.e. 'the men were killed by three bears.') ""bears killed three men.'

-ak 3p

maskw- -a NA 3' bear(s)

Changing the word order has no effect on the interpretation. (17) and (21) are interpreted identically, while (19) and (22) are similarly unchanged in meaning. In all four of these examples, regardless of semantic role and word order, the quantifier associates with maskwa. If neither word order nor semantic role can be used to predict this phenomenon, we must look elsewhere, particularly at pragmatic roles or pragmatic information status. In each of the maskwa examples (17, 19, 21, 22), the only common denominator is the obviative status of the noun. As has already been observed, though, (18) shows that it is not always the obviative with which the quantifier is associated. However, I would like to suggest that there is a fundamental difference in the information status of the obviative participants in (17) versus (18) and that this difference underlies the interpretation of floating quantifiers in the data. In examples (17) and (19) we have already noted that the quantifier must be construed with the obviative participant. Why should this be? This is a difficult question to answer and one which might lead to many

432

Arok Wolvengrey

random hypotheses. Is it something specific to bears? Is there something in the difference between bears and men semantically (i.e. only men are +human)? The first thought was never to be taken seriously and is easily discarded. Dahlstrom's original sentence included möswa rather than maskwa, such that whatever holds for bears also holds for moose. This still leaves the possibility that näpewak, 'men', as a human referent, somehow outranks a non-human referent such as maskwa, 'bear(s)'. This possibility can also be rejected by observing the result of simply reversing the assignment of proximate and obviative in (17) and (19) above, yielding (23) and (24) respectively. (23) nisto nipahewak maskwak näpewa. nisto nipah- -e -wak maskw- -ak NUM VTA DIR 3p NA 3p Three kill s.o. 3p-3' bears 'the bears killed three men.' ('*three bears killed a man/men.') (24) nisto nipahikwak maskwak näpewa. nisto nipah-ikw -wak maskw NUM VTA INV 3p NA Three kill s.o. 3'-3p bears 'three men killed the bears.' (i.e. 'the bears were killed by three men.') ('*a man/men killed three bears.')

näpewNA men

-ak 3p

näpewNA men

-a 3'

-a 3'

Once maskwak is marked as proximate, and näpewa is marked obviative, the quantifier continues to construe with the obviative participant. Thus, the notion that the semantic feature [+human] plays any role can be discounted. It is, for these examples, simply the obviative status of the noun which appears determinative. Hence, an explanation for this is still required. I believe the answer is to be found in the pragmatic information status indicated in these examples by the choice of obviation. In each of the examples including maskwa(k), the choice of obviative is not obligatory. The contrast between (25) and (26), for instance, shows that the same basic semantic proposition (that of men killing bears) can be imparted by two different structures, involving a shift of obviative assignment in conjunc-

Inversion and the absence of grammatical relations in Plains Cree

433

tion with a change in direction marking. These examples are identical to (17) and (24) respectively, with the removal of the complicating quantifier. (25) nipahewak maskwa näpewak. nipah-e -wak maskw-a VTA DIR 3p NA 3' kill s.o 3p-3' bear(s) 'the men killed a bear/some bears.' (26)

näpewNA men

-ak 3p

nipahikwak maskwak näpewa. nipah- -ikw -wak maskw- -ak näpew-a VTA INV 3p NA 3p NA 3' kill s.o. 3'-3p bears men 'a man/some men killed the bears.' (i.e. 'the bears were killed by a man/some men.')

Thus, the contrast between (17) and (25) is only that a quantifier has been included in (17), and must be associated with the obviative referent. Similarly, the contrast between (24) and (26) is only that (24) includes a quantifier which also must be construed with the referent assigned obviative status. Without the quantifier, the underlying proposition is unaffected. However, the information status is certainly affected by choice of obviative. This is illustrated by the definite/indefinite contrast in the English glosses in (25) and (26). The obviative referent is given an indefinite reading, while the proximate is definite. This correlates with the information status of the obviative as less topical, more likely to be new information and quite possibly indefinite. In contrast, the proximate referent is most likely the sentence topic and, therefore, represents given, definite information. Even in contextless, "out-of-the-blue" sentences, informants will associate any possibly ambiguous descriptive information (such as quantifiers) as modifying the referent most in need of further specification. Salient, topical, proximate referents do not require further specification; they are known. Less topical, new and indefinite information coded as obviative participants, on the other hand, may well require further information to establish proper reference. The result, as illustrated in all of the the maskwa(k) examples, is that the quantifier is always interpreted as providing further specification of the obviative referent. Pragmatic factors dictate where neither syntactic word order nor semantic role plays a part.

434

Arok Wolvengrey

3.2. iskwewak examples Returning to example (18), repeated here as (27), we observe that it is the proximate goal, rather than the obviative actor with which the quantifier must construe. (27) kahkiyaw säkihikwak otänisiwäwa iskwewak. kahkiyaw säkih- ikw- (w)ak- o- tänis- iwäw- a QNT VTA INV 3p 3P NDA 3pP 3' All love 3'-3p daughter(s) iskwew- -ak NA 3p women 'their daughters love all women.' (i.e. 'all women are loved by their daughters.') '*all their daughters love the women.' This would seem to parallel the example cited earlier as (17) in which the quantifier could only construe with the 'object', and contradict the analysis offered immediately above. Similarly, however, we have already seen how this example can be altered (as (20) above, repeated here as (28)) to reveal that it is not the goal which must be construed with the quantifier at all, but rather the proximate noun phrase iskwewak, regardless of its semantic role. (28) kahkiyaw säkihewak otänisiwäwa iskwewak. Kahkiyaw säkihe(w)ak o- tänisiwäw- a QNT VTA DIR 3p 3P NDA 3pP 3' All love s.o. 3p-3' daughters iskwew- ak NA 3p women 'all women love their daughters.' '*women love all their daughters.' Though iskwewak 'women' is now undisputedly the actor of the Direct VTA verb in (28), the quantifier must remain associated with this participant. While these examples together give further evidence against Dahlstrom's analysis of the goal as object, they also present an obvious problem

Inversion and the absence of grammatical relations in Plains Cree

435

for the current analysis for, as already noted, the quantifier is always construed with the proximate participant, iskwewak, not the obviative. In these examples, however, the information status of proximate and obviative referents is not identical to that in the maskwa(k) examples. Here, because of the presence of a possessive relationship between the two referents, the assignment of obviation to the possessum is obligatory. It follows from the necessity of establishing the possessum's reference via the possessor that the possessum is less salient in the discourse than the possessor. It therefore has a similar status to the obviative referents in the maskwa(k) examples; they represent new information. The real difference lies in the status of the proximate referents. Whereas the proximate referents in the maskwa(k) examples are interpreted as highly topical, the proximate participant in (27) and (28) need not be interpreted as being highly topical or given information at all. Instead, it may also be new (or re-introduced) information, and only marginally more topical than the possessum. Faced with two referents low in topicality, the quantifier must be construed with one of them. The evidence from the iskwewak examples indicates that it is the proximate referent, the referent through which the possessum must take its own reference. Thus, the quantifier is construed with the proximate in these examples in order to further establish the reference of the possessor, which is itself necessary to properly establishing the reference of the possessum. This does not, however, explain why the quantifier cannot be construed with the obviative possessum, and this is a question which I leave for further investigation. In conclusion, the test for objecthood suggested by Dahlstrom (1991) is not valid. Neither syntactic nor semantic roles play a part in establishing the association of floating quantifiers with nominals. In the additional absence of strict word order constraints on interpretation, the only available explanation for this phenomenon is that the pragmatic information status of the arguments contributes to or even dictates the disambiguation of reference of floating quantifiers.

3.3. Unspecified Actor vs. Passive (kakwe test) We will now turn to the formal characteristics of the "unspecified actor" paradigms in Cree. These have, under the guise of "indefinite actor" or "passive" constructions, been amply described in the past (e.g. Wolfart 1973, Jolley 1982, Ahenakew 1987, Dahlstrom 1991, Dryer 1996). Syntac-

436

Arok Wolvengrey

tically, we can note the obligatory absence of a lexical actor, as shown in (29). (29)

a.

niki-asamikawinän. ni- kiasam-ikawi -nan 1 TNS VTA XAct lp past feed s.o. (X-)lp 'We were fed. / (Someone) fed us.'

b. *awiyak niki-asamikawinän. awiyak ni- kiasam-ikawi -nan PR (3) 1 TNS VTA XAct lp Someone past feed s.o. (X-)lp 'Someone fed us / We were fed by someone.' c. awiyak niki-asamikonän. awiyak ni- kiasam-iko -nan PR (3) 1 TNS VTA INV lp someone past feed s.o. 3-lp 'Someone fed us.' In (29a), the unspecified actor form stands alone as a grammatical sentence, while (29b) illustrates the ungrammaticality of trying to include even an indefinite pronominal (awiyak) as actor. (29c) is included to show that if an indefinite pronoun is used, it is interpreted as any other third person participant, and the verb must occur in the VTA inverse rather than the unspecified actor. Morphologically, there is a split among the Cree VTA unspecified actors between speech act ("local", "you-me set", first and second person) participants on the one hand, and third person arguments on the other. This formal difference has lead to the following suggested modification to the ΑΡΗ, in which X represents an unspecified actor (cf. Jolley 1982, Dechaine & Reinholtz 1998): (30) Modified Algonquian Person Hierarchy SAPs XAct Third Persons 2 1 ^ X 3 3'

Inversion and the absence of grammatical relations in Plains Cree

437

Interestingly, though they both represent the ΑΡΗ as in (30), Jolley (1982) argues that "indefinite actors" are passives with promotion to subject, while Dechaine & Reinholtz (1998) argue that promotion to subject does not occur. Ultimately, both positions hinge on one or the other possibility. Theoretically, both possibilities should not be substantiated in one and the same construction and we must therefore search for tests which illustrate which is the correct choice. Dahlstrom (1991:76) discusses unspecified actor data, including the example in (31), as part of her copying-to-object test for subject. (31)

nikiskiyimäwak e-ki-sikihihcik. nikiskeyim- -ä -wak eki1 VTA DIR 3p CMPL TNS know s.o. l-3p past Ί know they were scared.'

sekih-iht -ik VTA XAct 3p scare s.o. (X-)3p

The sole participant of the unspecified actor VTA in the subordinate clause is the goal. This goal is coreferential with the goal of the main clause. Following Dahlstrom's arguments that only a subordinate clause subject can copy to object, she concludes that the goal of an unspecified actor form must be the subject. Hence, unspecified actors are passives. However, another interpretation of the data is again available. In direct and inverse alike, the actor (i.e. highest ranking semantic function present) must be the argument copied to object. With an unspecified actor form, the actor has been obligatorily demoted and cannot participate in the syntax (though it does remain semantically and pragmatically), leaving the goal as the highest ranking participant on the Semantic Function Hierarchy (SFH). It is the only possible option, just as when an unquestionably intransitive verb occurs in the subordinate clause (cf. Dahlstrom 1991:67): (32)

nikiskeyimäw e-nöhte-sipwehtet. nikiskeyim-ä -w enöhtesipwehte- -t 1 VTA DIR 3 CMPL IPV VAI 3 know s.o. 1-3 want to leave Ί know he wants to leave.'

Unless another test can confirm Dahlstrom's hypothesis, control by highest ranking role on the SFH is at least as plausible an analysis. In fact, addi-

438

Arok Wolvengrey

tional data does exist which will confirm that the goal of an unspecified actor VTA verb cannot be equated with a "subject". The data in the following examples, (33)-(38), illustrate that pragmatic context dictates whether the actor or goal of an unspecified actor can exert control over a higher predicate (represented by the preverb kakwe-). In one and the same construction, both are possible. (33)

kakwe-wäpamikawi! kakwe wäpam- -ikawi -0 IPV VTA XAct 2(IMP) try see s.o. (X-)2 'Try to be seen!' *?'try for someone to see you'

In (33), the verb is inflected as a second person singular imperative. The context for this example is that it is uttered as advice given to a dancer before entering the circle at a pow-wow. As such, the second person is expected to take volitional control of the event (i.e. to bring about the event of his/her being seen). The English translation indicates this with a passive construction. The examples in (34) and (35) stem from similar contexts in which the goal can be expected to take volitional control. Again, a passive translation is called for, consistent with a promotion of the unspecified actor verb goal to "subject" status. (34) niki-kakwe-wäpamikawin e-mekwä-nimihitoyän. nikikakwe- wäpam-ikawi -n e1 TNS IPV VTA XAct SAP CMPL past try see s.o. (X-)l mmihito- -yän VAI 1 dance Ί tried to be seen while dancing.' (35) ohcitaw ta-kakwe-wäpamikawiyan (nöhte-otahoweyani). ohcitaw takakwe- wäpam- -ikawi -yan IPC CMPL IPV VTA XAct 2 necessary try see s.o. (X-)2 'You have to try to be seen (if you want to win).'

mekwäIPV while

Inversion and the absence of grammatical relations in Plains Cree

439

In (34), the unspecified actor verb is in the Independent mode (typically used for main clauses), while in (35) the verb is in the Conjunct mode (frequently but not exclusively used for subordinate clauses, and therefore introduced by a complementizer (CMPL)), but this difference does not appear to matter. In (34) and (35), the goal takes volitional control over the action meant to bring about the event. If this was always the case, the analyses of Dahlstrom (1991) and Jolley (1982), among others, would hold. However, the data in (36)-(38) show that pragmatic contexts also exist in which the unspecified actor retains volitional control, and the goal thus remains merely a goal. (36)

tahto-kisikäw mäna e-ki-kakwe-wäpamikawiyän. tahto-kisikäw mäna ekikakwe- wäpam- -ikawi -yän IPC IPC CMPL TNS IPV VTA XAct 1 everyday always past try see s.o. (X-)l 'Someone's always trying to see me every day.'

(37)

mäka mina kape-kisik niki-kakwe-wäpamikawin nawin.). mäka mina kape-kisik ni- kikakwe- wäpamIPH IPC 1 TNS IPV VTA as usual all day past try see s.o. 'As usual, someone tries to see me throughout the busy).'

(38)

(mistahi

nica-

-ikawi -n XAct SAP (X-)l day. (I'm very

otäkosihk esa öma e-ki-kakwe-wäpamikawiyän, ... otäkosihk esa öma ekikakwe- wäpam- -ikawi -yän IPC IPC IPC CMPL TNS IPV VTA XAct 1 Yesterday past try see s.o. (X-)l Ί understand someone tried to see me yesterday, ...' or Ί understand there was an attempt to see me yesterday, ...' (..., mäka anima mekwäc e-mämawapiyähk δta kä- pe- takohtecik.) (..., but they arrived here while we were in a meeting.')

We do not expect to find imperatives among these examples, but both Independent and Conjunct mode inflections are represented. The context for all three examples is similar, with some indication that an unspecified actor (whether definite or indefinite, specific or non-specific) has made an attempt or is constantly making attempts to see the first person goal. In (36)

440

Arok Wolvengrey

and (38), the verb occurs in the Conjunct mode, while in (37) the verb is Independent. Hence, neither interpretation is specifically associated with the Independent or the Conjunct. It is merely pragmatic context which determines the interpretation of control. Finally, it can be noted that the exact same verb form occurs in both (34) and (37), repeated here as (39). Devoid of context, either interpretation is open to Plains Cree speakers, and either English translation is possible. (39)

niki-kakwe-wäpamikawin.

nikikakwewäpam- -ikawi -n 1 TNS IPV VTA XAct SAP past try see s.o. (X-)l 'Someone tried to see me / 1 tried to be seen.' What these examples have illustrated is that the VTA unspecified actor paradigm does not have a grammaticalized subject or object. The actor is unspecified, but not pragmatically absent, and may act as controller given an appropriate context (as in (36)-(38)). This could be interpreted as an example of the demotion of the agent without apparent promotion of an object. In contrast, appropriate contexts can also be found which dictate that the goal will serve as a controller (as in (33)-(35)), in which case, passivization including promotion to "subject" appears to occur. In this construction in Plains Cree, though, both options are possible, and the choice is not grammaticalized but remains open to determination by pragmatic context. The debate over whether unspecifed actor forms are full passives or not has hinged entirely on the identification of the goal with either "subject" or "object" status. The debate has now been unhinged. There is no necessarily grammaticalized subject or object at all. The choice of controller is left entirely to pragmatics as one and the same construction serves the purpose for two distinct constructions, with two different grammaticalized subjects, in languages like English.

4.

Conclusion

Ultimately, searching for tests for "subjecthood" and "objecthood" in Cree is futile, since the grammatical relations of "subject" and "object" do not themselves exist. In Cree, the interaction of pragmatic and semantic func-

Inversion and the absence of grammatical relations in Plains Cree

441

tions is enough to disambiguate all necessary interactions without recourse to a third level of grammatical functions. The Algonquian Semantic Function Hierarchy is rigidly defined and assigned in Cree verbal structure, and this Algonquian-specific SFH differs from Dik's (1997: 37) SFH only due to the overriding importance of animacy in Cree grammar. Animate participants must be treated as more salient/important than inanimates, and thus prototypically animate dative/recipient/benefactive participants outrank patient/theme participants for the status of "goal". This need not entail a recasting of the Functional Grammar Semantic Function Hierarchy, but should indicate that other factors, including other hierarchies, may take language-specific precedence, requiring language-specific modifications. The extent to which one hierarchy can impact on another, and even the observation of which hierarchies can impact on others, is left for future investigation. Despite this Algonquian-specific modification, the SFH plays an important role in Cree grammar, especially in the coding of V T A verbs. The only way in which a lower-ranking semantic function can take precedence over a higher-ranking one (e.g. goal over actor) is for the higher-ranking role to be completely removed syntactically (e.g. via unspecified actor inflection). As we have seen, though, there is no necessary concomitant re-assignment of perspective (i.e. grammaticalization of syntactic functions) involved. Goals remain semantic goals and are only potentially raised in status through a particular pragmatic interpretation of the clause, not through a syntactic shift of perspective. Most importantly, the Cree and Algonquian Direct-Inverse system works through the interaction of the rigid Algonquian Semantic Function Hierarchy and the equally rigid pragmaticallybased Algonquian Person Hierarchy. Thus pragmatic factors alone are utilized to assign the appropriate semantic roles to the participants involved. It is only the grammaticalized perspective of languages such as English that force the choice of subject and/or object in translation. The Direct-Inverse system in Cree makes this third level of perspectivizing functions or grammatical roles unnecessary.

Notes 1.

I am grateful to numerous colleagues and students at the First Nations University of Canada (formerly the Saskatchewan Indian Federated College) who have helped to shape the present discussion, with particular thanks to my part-

442

2.

3.

Arok Wolvengrey ner Jean Okimasis. I would also like to express my gratitude to the participants in the 10th Functional Grammar Conference and to an anonymous reviewer whose helpful comments have served to guide me (I hope) in improving the text. Where I may have failed to incorporate suggestions, from colleague and reviewer alike, and where fault remains, I alone am responsible. It is, in fact, possible to have an inanimate actor, but such exceptional forms exist as a subtype of the Transitive Animate (VTA) paradigms and, as such, do not constitute a separate subclass on the same level as VII, VAI, VTI, and VTA verbs. Abbreviations used in examples and throughout the text include the following: 1 first person (singular) lp first person plural exclusive 2 second person (singular) 21 first person plural inclusive 2p second person plural 3 third person (proximate) singular 3P third person (singular)possessor 3p third person (proximate) plural 3pP third person (proximate) plural possessor 3' third person obviative 0 inanimate third person (proximate) singular Op inanimate third person (proximate) plural 0' inanimate third person obviative singular 0'p inanimate third person obviative plural 3-3' third person proximate singular acting on third person obviative 3-0' third person proximate singular acting on inanimate third person obviative 3p-3' third person proximate plural acting on third person obviative 3 '-3 third person obviative acting on third person proximate singular 3 '-3p third person obviative acting on third person proximate plural ΑΡΗ CMPL DIR IMP INV IPC IPV ΝΑ NDA NDI

Algonquian Person Hierarchy complementizer direct imperative inverse indeclinable particle indeclinable preverb animate noun dependent (inalienable) animate noun dependent (inalienable) in-

PR QNT SAP SFH s.o. s.t. TNS VAI VII VTA VTI

pronoun quantifier speech act participant Semantic Function Hierarchy someone; animate object something; inanimate object tense animate intransitive verb inanimate intransitive verb transitive animate verb transitive inanimate verb

Inversion and the absence of grammatical relations in Plains Cree

NI NUM 4.

5.

6.

7.

animate noun inanimate noun numeral

443

X unspecified actor XAct unspecified actor inflection

At least in terms of the verbal morphology. Thanks are extended to an anonymous reviewer for pointing out that it is still theoretically possible for an Active-Stative system to manifest itself in nominal marking in the absence of such marking verbally. This is, however, not an issue for Cree and Algonquian in general where, as stated earlier, there is no more differential marking on nouns or independent pronouns than there is on bound pronominals. This substantial group of verbs have received varying classification by Algonquianists. Bloomfield (e.g. 1946) referred to these syntactically transitive but morphologically intransitive verbs as "pseudo-transitive" verbs, while still treating them primarily as part of the intransitive type. Many simply take the morphological criterion of paradigm similarity as primary and group them with the VAI class (cf. Wolfart 1973, Ahenakew 1987). Others have taken the syntactic criterion of transitivity as primary and grouped them as a subclass of VTI verbs (cf. Okimasis and Ratt 1999). Still others, seeking to recognize both morphological and syntactic features, have recently utilized such abbreviations as VAI-T (Ellis 1995) and VAIt (Wolvengrey 2001). Here we have the potential for a conflict between traditional Algonquianist terminology (following Bloomfield, for instance) and the terminology of Functional Grammar (following Dik). In FG's Semantic Function Hierarchy, only the "agent" ranks above a "goal", which can be equated with the semantic role of "theme" or "patient". In Algonquian terms, the agent (or simply the highest ranking semantic role present) is named the "actor", while the term "goal" is most commonly associated with the second highest ranking semantic role. This can, in monotransitives, be a theme or patient, but in the case of ditransitives, the Algonquian "goal" will always be the recipient or benefactive, not a theme. In essence, then, the terminologies are virtually equivalent, the only difference being in what semantic function is treated as primary following the agent. The current example is identical to that cited by Dahlstrom (1991: 83) with the substitution of the third person obviative noun maskwa "bear(s)" for the third person obviative noun moswa "moose". This has been done only to make some of the subsequent examples more pragmatically acceptable to the judgements of speakers who see the killing of men by bears as more likely than the killing of men by moose. The substitution has no other bearing on the arguments put forth here.

444

Arok Wolvengrey

References Ahenakew, Freda 1987 Cree Language Structures: A Cree Approach. Winnipeg: Pemmican Publications. Β lain, Eleanor 1998 The role of hierarchies and alignment in direct/inverse. In Papers of the 29lh Algonquian Conference, David H. Pentland (ed.), 43-56. Winnipeg: University of Manitoba. Bloomfield, Leonard 1946 Algonquian. In Linguistic Structures of Native America, Harry Hoijer (ed.), 85-129. (Viking Fund Publications in Anthropology 6.) New York: Wenner-Gren Foundation for Anthropological Research. Campana, Mark 1989 Algonquian and the absolutive case hypothesis. In Actes du vingtieme congres des algonquinistes, William Cowan (ed.), 70-78. Ottawa: Carleton University. Dahlstrom, Amy 1991 Plains Cree Morphosyntax. New York: Garland Publishing, Inc. Dechaine, Rose-Marie and Charlotte Reinholtz 1998 Direct/Inverse as case-licensing. Paper read at the Third Workshop on Structure and Constituency in the Languages of the Americas, March 27, Saskatchewan Indian Federated College, Regina. Dik, Simon C. 1997 The Theory of Functional Grammar 1: The Structure of the Clause. Second, revised edition, edited by Kees Hengeveld. New York: Mouton de Gruyter. Dryer, Matthew 1996 Passive vs. indefinite actor construction in Plains Cree. In Papers of the 27th Algonquian Conference, David H. Pentland (ed.), 54-79. Winnipeg: University of Manitoba. Ellis, C. Douglas 1995 ätalöhkäna nesta tipäcimowina / Cree Legends and Narratives. Publications of the Algonquian Text Society. Winnipeg: The University of Manitoba Press. Givon, Talmy 1984 Syntax: A Functional-Typological Introduction. Amsterdam: John Benjamins.

Inversion and the absence of grammatical relations in Plains Cree

445

Hewson, John 1987 Are Algonquian languages ergative? In Papers of the 18th Algonquian Conference, William Cowan (ed.), 147-153. Ottawa: Carleton University. Jolley, Catherine A. 1982 On the Plains Cree passive: an analysis of syntactic and lexical rules. Ohio State University Working Papers In Linguistics, 26: 1-33. Myhill, John 1988 Variation in Spanish clitic climbing. Georgetown University Round Table on Languages and Linguistics 1988. Washington: Georgetown University Press. Okimasis, Jean L., and Solomon Ratt 1999 Cree, Language of the Plains. Regina: Canadian Plains Research Center. Rhodes, Richard 1976 The Morphosyntax of the Central Ojibwa Verb. Ph. D. diss., University of Michigan. Wolfart, H. Christoph 1973 Plains Cree: A Grammatical Study. Transactions of the American Philosophical Society, n.s. vol. 63, pt. 5. Philadelphia. 1991 Passives with and without agents. In Linguistic Studies Presented to John L. Finlay, H. Christoph Wolfart (ed.), 171-190. (Algonquian and Iroquoian Linguistics, Memoir 8.) Winnipeg. Wolvengrey, Arok 2001 nehiyawewin: itwewina / Cree: Words. Regina: Canadian Plains Research Center.

Direction diathesis and obviation in Functional Grammar: The case of the inverse in Mapudungun, an indigenous language of south central Chile

Ole Nedergaard Thomsen

1.

Introduction

Standard Functional Grammar (Dik 1997) conceives of voice as a formal expression of alternative perspectivizations of a given State of Affairs designated by a clausal predication: an active transitive predication is seen from the perspective of the Agent argument, a passive predication from the perspective of the Patient, and a dative-shifted ditransitive predication (say, John gave Mary the book) has its secondary perspective with the Recipient.1 These perspectivizations are conceived of as active 'assignments', on the part of the speaker, of Perspectival Functions to predicational terms. The model recognizes at most two such functions, viz. Primary Vantage Point ('Subject') and Secondary Vantage Point ('Object'). The present version of FG, additionally, recognizes the existence of predicational operators of Diathesis (Nedergaard Thomsen 1994). These operators are expressed by the category of voice (e.g. inflection, participles, periphrasis). Prototypical passive Diathesis is the inverse perspective on the causal connection between an activity and its resultant state as designated by the verbal predicate, i.e. from result to cause (e.g., The stick was broken by the child). The antipassive is the opposite, viz. from causal activity to resultant state (Nedergaard Thomsen 1994). However, there are other kinds of Diathesis in the languages of the world than active-passive and ergative-antipassive, one being the category of Direction, distinguishing between Direct and Inverse. Roughly, this kind of Diathesis does not involve reperspectivization of a given state of affairs, the perspective being 'fixed' according to a Discourse Participant Empathy Hierarchy (Kuno 1976), perhaps better known as the Animacy

448

Ole Nedergaard Thorns en

Hierarchy (Silverstein 1976; Comrie 1989), a unified scale for nominal denotations, ranking from Speech Act Participant-denoting personal pronouns over 3rd person pronouns and (personal) proper names, to human, animate, and inanimate entity denoting nouns. The Direction system functions in such a way that the Primary Vantage Point is always with the higher referent on the Animacy Hierarchy, irrespective of its semantic role as Agent or Patient. If this referent performs the role of Agent in the state of affairs, the Direction is Direct, whereas if it performs the role of Patient, the Direction is Inverse, taking for granted that it is more 'natural' (optimal) for a higher-ranked referent to be in control of the situation than the reverse. The ranking of the Speech Act Participants (SAPs), i.e. speaker, hearer, and third person, is prototypically according to the speaker's empathy - it is easiest for the speaker to empathize with him- or herself, etc. This may be cast in two ways: either 'objectively', according to some parameter of closeness to ego ("me first!" - "I am closest to myself' - the speaker is inside the scope) whereby first person is ranked above second person, or 'subjectively', in accordance with a parameter of politeness or modesty ("you first!" - "you are closest to me" - the speaker is outside the scope) whereby second person is ranked above first person. A Direction system normally also distinguishes two kinds of third person, according to degree of closeness to ego (proximity), namely a Proximate third person and an Obviative third person (cf. Hockett 1966). The Proximate is always ranked above the Obviative in an Obviation system. The Direction system conforms to a conception of an optimal ('expected') role-referent configuration of a state of affairs, such that a deviation from this ideal is coded by a specific category, the Inverse, for the non-optimal ('unexpected') state of affairs. What "Vantage Point" exactly means is not obvious, or may even differ from language to language and/or type of diathesis. Nedergaard Thomsen (1994; cf. Haberland and Nedergaard Thomsen 1994) proposes that Vantage Point be understood as grammaticized cognitive-pragmatic 'aboutness' (topicality) of a predication or proposition, and that there are (at most?) three 'degrees' of aboutness, namely Topic 1, Topic 2, and Topic 3. These textual functions are thus either assigned to or stay fixed with the different referents of the State of Affairs designated by the predication. The assignment is according to the 'topicworthiness' of the syntacticosemantic (ideational) roles, arranged in a Topicworthiness (Semantic Role) Hierarchy, for instance, Agent > Patient > Recipient >... . The Topics (1-3), as mentioned above, could also be fixed with respect to the referential-

Direction diathesis and obviation in Functional Grammar

449

semantic (interpersonal) Animacy Hierarchy, so that the highest-ranked of the two or three central referents of a given State of Affairs will be Topic 1, etc. With respect to Obviation, Proximate is the higher-ranked referential role of the two third persons. What qualifies a referent as Proximate third person is the discoursal status of given information (topically more prominent; Arnold 1994, 1998), or being the focus of the speaker's interests (Hockett 1966), that is being pragmatically 'near' the speaker. Obviative is the opposite ('distant'). The basic function of Obviation is to track third person referents in discourse. Direct Diathesis is the congruent alignment of topicworthiness (semantic role) and topicality value (Vantage Point): the more topicworthy semantic role, Agent, is coupled with the more topical of the referents of the states of affairs (for instance, a first person, indexing the Speaker), whereas the less topicworthy semantic role, Patient, is coupled with the less topical referent (for instance, a third person). Inverse Diathesis is a non-congruent alignment, signalled by the inverse voice: a highly topical role (e.g. Agent) is of actual low topicality (Topic 2), and vice versa. Now, it has been claimed, both within Functional Grammar proper (TFG1) and cognitive-functional grammar (Tomlin 1995), that Vantage Points are not textual/discoursal 'topicality', or 'aboutness' functions. In the latter model they are the speaker's actual focus of attention at the moment s/he utters the predication. The most obvious case of attentional focus is the focusing occurring during visual perception of a state of affairs and the on-line description of it. If attentional focus is grammaticized in a given language, the choice of subject (Primary Vantage Point), and thereby the selection of verbal voice, is causally determined by the speaker's attentional focusing in the intentional-conceptual input system to his or her grammar. Even though both TFG and cognitive-functional grammar explicitly deny that syntactic functions (and, derivatively, verbal voice) are grammaticizations of textual notions of topicality, or claim that topicality is 'really' cognitive focusing (Tomlin 1995), there is nevertheless empirical evidence (Forrest 1996) that 'topicality', understood as textual aboutness, may be a dominant factor in subject selection, but that visual focusing may delay utterance production, if the visual focus is not the same as the topic chosen (according to the goal of the discourse): the re-orientation takes time. The above description - in terms of topicality, not focal attention - of the two third persons in the Obviation system, the Proximate and the Obviative, goes against the hypothesis of (cognitive-) functional grammar

450

Ole Nedergaard Thomsen

mentioned above. Therefore, it was proposed to apply Tomlin's Fish Film experiment (Tomlin 1995) to a language with Direction Diathesis and Obviation to see whether voice and subject selection in utterances correlate with visual attention focusing (rather than the speaker's focus of interest, which may be quite another thing) in this kind of language, in situations where the speaker's focal attention is or can be controlled experimentally in on-line speech production. The South American Indian language Mapudungun (Araucanian) of south central Chile (and, marginally, Argentina, cf. Croese 1985, Smeets 1989, Fernandez Garay 2001) was selected as a language with Direction Diathesis and Obviation (Fontanella 1967; Grimes 1985; Arnold 1994, 1996). The hypothesis was accordingly that Proximate would be chosen to code a visually primed participant in Tomlin's experiment, and that a primed Agent (Proximate) would select direct voice whereas a primed Patient would select inverse voice. The paper reports preliminary results of a pilot application of the Fish Film experiment in a fieldwork setting with 4 Mapudungun speakers of Padre Las Casas / the city of Temuco, south central Chile (region called La Frontera/La Araucania). Implications will be drawn for the adequacy of FG as a grammatical model (typological, pragmatic, and psychological adequacy) as well as for the FG conception of the Natural Language User (concerning the existence of an intentional-conceptual input to underlying semantic structure). The paper will conclude that Mapudungun is better accounted for in terms of Hengeveld's (2004a, b) 'top-down' Functional Discourse Grammar than in terms of Dik's (1997) bottom-up TFG.

2.

Mapudungun

Mapudungun is a suffixing, head-marking, polysynthetic language (Lenz 1935), coding on the verb the 'case', person, and number of the Subject (Tl) and the Object (T2) - there is no nominal case, i.e. dependent marking to distinguish the external arguments of a predication. The endings are basically (with important exceptions, see below) agglutinative, separating the expression of the grammatical categories. The 'person' endings are the direct manifestation of the terms of the predication: Mapudungun is a "prodrop", or Pronominal Argument, language, having cross-reference, not agreement, so that the explicit mentioning of a participant by way of a per-

Direction diathesis and obviation in Functional Grammar

451

sonal pronoun is emphatic or contrastive. The word order of the language is fairly free, with declarative SVO, SOV, and VOS. In the analysis of Mapudungun several grammatical notions are required. First, the arguments of a clausal predicate are ideational semanticosyntactic macro-roles, which are the single argument of an intransitive predication (S), the transitive subject (A, actor, proto-agent), the transitive object (P, undergoer, proto-patient), and the indirect object (I, which may be a receptor, or a possessor-source in actions of taking away and the like); the I of a ditransitive clause is coded the same way as the Ρ of a monotransitive clause, thus constituting a Primary Object (PO), different from the Ρ of the ditransitive clause, the Secondary Object (SO), which does not trigger cross-reference (Mapudungun is thus primarily a Pronominal Argument language, secondarily a Lexical Argument language). Furthermore, an Oblique (OBL) function is recognized (coded by postpositions). Additionally, there are predicative complements (PC) in complex predicates, such as copula constructions. Second, the three textual Perspectival Functions ('topics') mentioned above are relevant, namely Primary Vantage Point (Tl), Secondary Vantage Point (T2), and Tertiary Vantage Point (T3); additionally, an Adjunct function is needed. The ideational and textual functions behave in a hierarchical way: Macro-role Empathy Hierarchy A,S > I,Ρ (PO) > Ρ (SO) > OBL Topic/Empathy Hierarchy Tl > T2 > T3 > Adjunct The default topicality, or topicworthiness, of the macro-roles is according to the Topic/Empathy Hierarchy: A,S(T1), etc. (cf. Kuno 1976's Surface Structure Empathy Hierarchy). The terms of the predication are distinguished as to the interpersonal, deictic functions of Discourse Participant Status: Interlocutor Status (± Speech Act Participant, SAP; Speaker vs. Addressee) and Obviation Status ([-SAP]) only: central vs. peripheral Discourse Participant: Proximate vs. Obviative; cf. Kuno's 1976 Topic Empathy Hierarchy). They also behave in a hierarchical way: Interlocutor Empathy Hierarchy [+SAP] > [-SAP]

452

Ole Nedergaard Thorns en

Obviation Empathy Hierarchy [-SAP]: Proximate > Obviative Speaker (1), Addressee (2) > [-SAP] (3) 1 > 3; 2 > 3; 2 >1 > 2 (N.B. Questioned referents are Obviative) The Interlocutor Hierarchy and the Obviation Hierarchy are integrated into the Discourse Participant Empathy Hierarchy (termed 'focalization hierarchy' by Adalberte) Salas, cf. e.g. Salas 1992b): Discourse Participant Empathy Hierarchy 2 > 1 > 2 > 3 Proximate > 3 Obviative Notice that the first person may be ranked below the second person (speaker-modest, 'you first!'), and the second person below the first (speaker-assertive, 'me first!'). We shall see below that an innovation has occurred from the 'speaker-modest' to the 'speaker-assertive' version of the person hierarchy, retaining, however, a degree of politeness by including the second person P/PO in the sphere of the first person A, in accordance with Heath's (1998) 'pragmatic-skewing model', which observes in the languages of the world an avoidance of transparency of coding with first and/or second person affixes. The integrated saliency hierarchy given above also functions as a 'topicworthiness' hierarchy, to the effect that, of two primary participants in a state of affairs (A or P/PO) - irrespective of their macro-role status - , the more salient will be (chosen as) Tl, the other T2, etc. Thus, the hierarchy is intimately correlated with discourse status. In another perspective the hierarchy is also an 'agent potentiality' hierarchy. (Notice that animacy per se is not a factor in Mapudungun: inanimates may be Proximate.) Before embarking on an analysis of the Mapudungun head-marking system in detail (in Sections 2.1 and 2.2), it will be convenient to have a look at the morphological template of a Mapudungun verb. The order of morphemes in the Mapudungun verbal desinences is the following (where |T1| is a primary topic independent of voice; T2 is a secondary topic in the direct diathesis, T2' is a secondary topic in the inverse; "X 2.1 and the inverse -e+ in 2.1 -> 1.1, and of the deponent reflexive/reciprocal -w+ in 1.1 -n 2.2-n and the inverse morph -mu+ in 2.2-n 1.1 as well as in 2.7-« 1.2-n. It is evident that these morphs symbolize Direction and index ('incorporate') number (sing, vs. non-sing.). The paradigm is rather transparent: the Direct Obviative third person Ρ T2, -fi-, originates in the stem of the third person pronoun, fei-, the Inverse Obviative third person A T2' -(m)ew stems from the instrumental/causal postposition mew, and is partly integrated into the verb form, thereby losing its initial m. Notice that the Inverse Proximate third person Ρ T l ' has 'periphrastic' number conjugation, applying the number part of the third person pronoun, -eng-u for the dual and -eng-ün for the plural, as a 'floating' and emphatic number indicator (cf. also intransitive (Of) and (Oi')). The transitive paradigm of Table 3 will be exemplified in the following examples (1): (1).

a. pe-fi-{n}. see-3.T2[P]-{Ind: 1 .sg.Tl [A]} Ί saw 3 rd pers.' b. pe-e-{n}-ew. see-Inv- {Ind: 1 .sg.T 1 [P]} -3 ,T2 [A] '3rd pers. saw me'

460

Ole Nedergaard Thomsen

b'.pe-ye-{n}-o inche. 'S/he saw me.' [Argentine Mapudungun (AM), Fernandez Garay 2001] c.

pe-fl-i-0. see-3.obv.T2[P]-Ind-3.prox.Tl[A] '3 rd pers. (prox.) saw 3rd pers.'

d.

pe-e-i-0-ew. see-Inv-Ind-3 .prox.T 1 [P]-3 .obv.T2 [A] '3 rd pers. (obv.) saw 3rd pers.'

e. pe-fi-i-ng-u eng-ün. see-3,obv.T2[P]-Ind-3.prox-du.T 1 [A] 3.-pi. 'They 2 (prox.) saw them.' f.

pe-w-i-0. see-R-Ind-3 .prox. .Τ 1 [S] '3rd pers. saw 3rd pers.self; they saw each other.'

g. pe-nge-{n}. see-Pass-{Ind:l.sg .T1[S]} Ί was seen.' h.

pe-{e+-i-i-u). see-{I+-Ind-l.-du.Tl[l.sg.Tl[A]:2.sg.T2[P]]} Ί see you 1.' (modern central form: number incorporated)

h'. pe-e-i-m-i-0 see-In v-Ind-2. -sg.T 1 [Ρ] -1. sg.T2 [A] Ί see you 1.' (older, peripheral form) i.

pe-{w+-i-i-n}. see- {R+-Ind-1 .-plur.T 1 [1.T1 [A]:2.T2 [P]]} Ί (or we, any number) saw you ( > 3 , number incorporated)', or w-i-i-n 'R-Ind-l.-pl.Tl[S]: 'We saw ourselves/each other'. Modern, central form)

Direction diathesis and obviation in Functional Grammar

461

j.

pe-{e+-{n}-+0}. see-{Inv/2.=sg-{Ind:l.sg.Tl [P]}-2.T2[A]} 'You 1 saw me.' (only two individuals, speaker and addressee)

k.

pe-{mu+-{n}-+0} see- {Inv/2 ,>sg- {Ind: 1 .sg.T 1 [P]} -2 ,T2 [A]} 'You (2-n) saw me (us).'

1.

pe-{mu+-i-i-n-+0}. see-{Inv/2.>sg-Ind-l ,-plur.Tl [P]-T2[A]} 'You (any number) saw us (me, such that

m. kawelloelu-vi-i-i-n miliko. [AM] horse.T3[SO] give-3.T2[PO]-Ind-l.-pl.Tl[A] soldier[PO] 'We gave the horse to the soldier.' n. elu-e-{n}-o epu kuchillo. [AM] give-Inv-{Ind: 1 .sg.Tl[PO]}-3.T2[A] two knives.T3[SO] 'He gave me two knives.' o. munt-e-{n}-o chanikollii. [AM] take:away-Inv-{Ind: 1 .sg.Tl[PO]}-3.T2[A] matra.T3[SO] '3rd pers. took away the matra from me.' p. epe munti-ma-te-v-e-{n}-o almost take.away-Appl-It-Evid-Inv-Ind. 1 .sg.Tl [PO]-3.sg.T2[A] in (- ni kure [AM] my wife.T3[SO] 'He almost took away my wife from me.' q. inche ngiidiiv_el_a_{e+-y-u} I [A] sew-Appl-Fut-{I+-Ind. 1 -du [l.sg.Tl[A]:2.sg.T2[PO]]} mi wün. [AM] your mouth.T3[SO] 'I'll sew your mouth.'

462

Ole Nedergaard Thomsen

2.2. Mapudungun head-marking in diachrony: grammaticalization The Mapudungun person endings can be seen as coding not only Discourse Participant Status but also semantico-syntactic role, i.e. they also contain (a kind of) case markers. In the Direct diathesis A [Tl] is nominative and Ρ [T2] is accusative, in that A is identical to S, and A is different from P. As the Inverse (outside third person constellations) is not a result of active perspectivization, but rather an automatic consequence of the constraints of the Discourse Participant Empathy Hierarchy, it is in fact just as "direct" (i.e. active) as the real Direct diathesis. Therefore, P/PO [ T l ' ] is absolutive (nominative), in that Ρ is identical to S, and Ρ is different from A, and A [T2'] consequently ergative. This seems consonant with the fact that the partially grammaticalized, enclitic -(/# m)ew seems to stem from the homonymous, general postposition mew which among other functions has the instrumental/causative one. The implication of this analysis is that Mapudungun is a split nominative/accusative-ergative/absolutive language, with overlap in the third person transition (3 is 3rd person Proximate, 3' 3rd person Obviative): Voice forms (A -> P) Nominative/ accusative 1 3 2 3 3 3' -0-

1 -> 2 -e+-w+-

1 p.3.T2[P]) Direct P> (f,: 'see' [V]) (x,: )A (d, x2: 'animal' [N] )P b. 'The animal saw me' (p.3.T2[A] < p.l.Tl[P]) Inverse *gAccsing *You.ACC.SG

der. ^NomPiur der. there. You.NOM.PL there. Jet"Acc piur der. der. *You.ACC.PL there there

How can we interpret this mixed case usage, and what is the semiotic gain of expanding the use of the accusative at the expense of the nominative, which is perfectly functional as it stands? It is not easy to say, apart from the truism that a marked form is taking over from an unmarked one, to the point that it is becoming the default case, the case you resort to when you do not know what else to choose. But why not abandon the nominative all together? It may still be a question of distinctiveness.9 The personal pronouns carry very few semantic features, apart from person, number, natural gender han and hun, or grammatical gender den and det. Α trivalent predicate with three nominal arguments, as in (4), more often than not has +Hum or +Anim for two of the three arguments, the first and the third. The semantic fillings of the arguments may help to determine the role played by these arguments. But in the absence of good semantic distinctions, as with the personal pronouns, case distinctions may help to distinguish between the various sentence constituents, and thus between the roles. Note that it is only on the left side of the word order template that a case distinction between argument 1 on the one hand and arguments 2 and 3 on the other hand is useful: as when the subject stands in field II and some other pronominal argument constituent in PI:

Pronominal expression rule ordering in Danish

517

(29) HamobjAcc kender hanSubjNom ikke. He.ACC knows he.NOM not 'Him he doesn't know.' (30)

(a review of the play 'Macbeth':) Allext kender 'Macbeth'x2, men ingenxl All know 'Macbeth', but nobody Macbethx2 Macbeth -

ikke engang not even

hanNomxi he.NOM

kender knows

selv. self

'Everybody knows 'Macbeth', but nobody knows Macbeth - not even he himself.' (Pol.16.6.2003 11:6) Heltoft (1997: 234) interprets the fact that pure nominative pronouns are not replaced by accusatives, but that pronouns with attributes or in paratactic constructions may be, as a difference between a bare anaphoric subject on the one hand and all other possibilities (non-bare anaphoric subjects as well as non-subjects with the accusative) on the other hand. A bare anaphoric pronoun has no other function but to maintain the reference to a former discourse referent. An example of bare anaphoric subjects contrasting with other grammatical functions can be seen in (6). (20)-(25) are examples with semantic features added to the constituent, which therefore displays greater distinctiveness of its own. Thus one can afford to use the accusative as a non-distinctive form in itself. Moreover, the added semantic features often introduce new entities into the discourse world, as can be seen in (20) and (24), hence they have no pure anaphoric function. As regards (21) - (25) the interpretation is, according to Jorgensen, (2000: 172), that the nominative lives on when a person is directly addressed. In this case it is not a real anaphora, but a deictic item, and the reference is derived from another - pragmatic - source.

6.

Expression rules and mixed case usage?

As for the relevant expression rules which might be formulated within the layered model, it is rather unclear what to do. First, there is the question of referring. In the second volume of Dik (1997) some indexing devices in the underlying structure are presented, which should secure the right relation

518

Lisbeth Falster Jakobsen

between the anaphor and its antecedents, the referents. But, as Cornish rightly says in his criticism of Dik's ideas (Cornish 2000), these indices only work out seen from the point of view of production: the speaker is supposed to know whom or what he is referring to with his anaphors. But the hearer cannot experience these indices; he has to rely on a smaller or larger part of the context to find the correct referent. Consider example (6). Three male persons are involved here: DC Hansen, the missing Mr. Ankermand, and the drunk head clerk Andersen. The pronoun to be used for all of them is a case form of han, 'he'. The anaphora is only a trigger which induces the hearer to find a suitable antecedent. And this is a result not of underlying meaning, but of processing the relevant textual segment containing the clues as to the trigger (Cornish 2000: 15). Thus, the real question is whether anaphorical reference is a question of textual meaning, contained in the underlying structure, or the result of an act within the communicative frame. The latter demands "a much greater 'division of labour' in the FG account of discourse anaphora, as between the concerns of underlying clause structure (the representation of anaphoric terms as well as of the relevant properties of the anaphoric predication as a whole), and the broader discourse context in which the anaphoric predication is set. This is the dialectical relation between bottom-up and top-down processes which is instrumental in resolving an anaphor's reference..." (Cornish 2000: 16). It is the question of the addressee, A, trying to reconstruct the intention of the speaker, S, according to Bolkestein's model of verbal interaction (1998: 209), now within the frame of the whole context of utterances. This speaks in favour of an FDG. Second, there is the problem of the mixed case usage. It is called mixed here because it is not a result of some uniform underlying function. A uniform underlying function should produce one expression, but here, as it stands, we have two expressions as resulting from one underlying representation. The variation in usage can be put down to a difference between written and oral language, but not to the effect that in one of the two modes one is obliged to use one variable, and in the other mode the other variable. Some speakers do not use the oral expressions at all, neither in written nor in oral production, some use some of them, some many of them. We are in the middle of a diachronic development which renders the orally marked usage still less unmarked, to the extent that it is sometimes carried over into the written language. (21) is presumably conceived in the written mode, (22), (23), and (25) are written interviews, (24) is a play, and (20) a dialogue in a novel.

Pronominal expression rule ordering in Danish

519

Now, how can this be modelled in an underlying representation? "We are witnessing a [...] period where forms are multifunctional, and functions are expressed simultaneously by several forms" (Bakker 1998: 16). No essential underlying functions have been altered, but are there some marginal, different functions at play? For some speakers, the accusative instead of the nominative is clearly understood as adhering to the colloquial, nonstandard mode, not to be used in written texts. In this case, it may be interpreted as having some sort of stylistic function; thus the difference in usage will carry some content difference, which may be exploited consciously as in (20) and (24). For other people, the accusative case may be the only personal and correct choice in both modes. Again, there might be a difference between the understanding of the speaker and the addressee in the same verbal interaction. The speaker might produce an accusative as a clear mark of the colloquial mode, whereas the addressee understands it as the perfectly normal form - or vice versa. This may create difficulties for FDG-models, e.g. the Bolkestein model, in the assessment of the intention of S towards A and the interpretation of this intention by A, respectively. Should different modes of language manifestation then be modelled in the rules somehow? If you appoint the user of correct standard Danish as the ideal speaker and model him or her in the rules, e.g. in the FG layered model, what you model is an educated elderly gentleman or lady from the better middle classes, but you do not model actual Danish. If you try to model actual usage, you will meet a problem with the various stages of language development of speakers according to the diachronic change we are in the middle of. Bakker says (1999: 43) that his dynamic model: "top down - left to right - depth first" is geared to language production rather than reception, and that there might be other constraints on comprehension than the ones he has put forward as regards production. To work out these constraints as well may be a realistic project - and a necessary one, if you really want to take language seriously. This can be seen from the mixed case usage, and from (19), an example of child language, where what is totally faulty, or maybe impossible according to the ideal rules seen from an adult production view, is nevertheless perfectly understood by everybody. In any case, a top-down model is called for if style or mode has to be represented by rules. Hengeveld (2000) proposes a quite detailed FDG model, where an extended hierachical layering is combined with a modular approach. It is a top-down model. Presumably mode and style will go into the interpersonal level, as referential acts do. This is then a functional level

520

Lisbeth Falster Jakobsen

which somehow has to get an interface with the expression level. However, as presented by Hengeveld (2000: 6), the expression level has the paragraph as its highest level; this would not be high enough for mode or style. Bakker's model (1999) is top-down as well, but starts at the main clause level, without pragmatic factors, which would leave us with no means to trigger the mode which again triggers a possible choice of pronominal case. A third possibility would be to face the fact that the pragmatic intentions of the speaker and the hearer are past assessment and modelling; you cannot know for sure about the intentions of the speaker or about what the hearer understands by the utterance, thus you cannot model all the semantic and pragmatic factors. The highest level of assessment will be a normative assessment of what is normally meant by the produced utterance, i.e. what the normal intention and hence the normal interpretation is. Not the actual intention and interpretation. But even this would sometimes produce very inaccurate assessments, as in cases of ongoing changes in the expression. Normative usage would again only remove the dynamic aspect of the model, the production part - as well as the reception part which is still missing from discourse grammar, which is essentially a dynamic grammar catering for specific instances. It would leave it as a model of possible or potential standard combinations of items in a language and the standard understanding of the manifestations of these combinations. But maybe this is the only feasible grammar we can produce, be it an FG grammar or an FDG grammar. Anything more specific than that will leave the field of modelling applicable to many instances behind, to become a mere description of specific instances. And this is not really the idea of a grammar.10

Notes 1. 2.

Especially Connolly et al. (1997) and Hannay and Bolkestein (1998) focus on discourse and pragmatics. That the lexical filling is decisive for the interpretation of the argument role value can be seen from examples which are erroneously constructed: (i) a. Men de seneste ar har TV2Goai, subj overgäet den sk£ebneAG, Arg2 at blive vraget af sinepolitiske stotter... (Pol. 5.6.02:11,1) 'But in the last years TV2 has befallen the fate of being rejected by its political supporters...' instead of

Pronominal expression rule ordering in Danish

521

(i) b. Men i de seneste är har den skcebneAg uhj overgäet TV2Cwai, Ar .2'But in the last years the fate has befallen TV2.' I use the terms Topic and Focus in a rather loose way, cf. Bolkestein's just comments in Bolkestein (1998). There is no exclusive formal marking of Topic and Focus in Danish. A minor matter is the question of the comparative subordinating elements som 'as' and end 'than', which are not considered real prepositions, as they do not govern a fixed case. Rather, the case assignment is regulated by agreement with the sentence constituent case value of the other part of the comparison: (ii) a. Nogle af demAgSubj. tjener mere end jegNom. Some of them make more (money) than I (do). b. HunAgSubjNom er asldre end jeg Nom. She is older than I (am). c. DuAgSubjNom kender hendeGoAcc bedre end jegNom. You know her better than I (do). d. DuAgSubjNom kender hendeGoAcc bedre end migAcc. You know her better than me. This is a rather simplified version which depends on the idea that there is only one Focus in a sentence and that this Focus does not depend on specific features of the constituent such as determination or Given or New. In fact there might be two Foci in a sentence. One is obligatorily assigned to the peak of the new information in the sentence, normally near the end of the sentence. The other is a contrastive focus which points a constituent out among a set of presented or presupposed possible items (Dik 1997a: 332). When this type of Focus is assigned to a whole constituent this will normally be [+Def], as can be seen in (4c) and (lOc-d). Exceptions exist, as in (4d). The constituent with contrastive Focus may keep its unmarked position, but is otherwise marked by specific stress or an additional lexical marker as in (lOc-d). In (9b) the two Focus types are combined in one constituent. On the whole the two types of Focus have different stress and intonation patterns. A distinction between the two Foci might be added to the rules in (13) in order to make (lOc-d) possible as well. The displaced constituent need not originate from the immediate subordinate clause; it may come from further down the subordination line, viz. (17). However, as discussed in section 6, indexing does not guide the addressee at all to the correct understanding. In the case of sentence intertwining the absence of the idea of dependence as a feature of lexical items and constructions involved is apparent. Such a feature might guide the addressee to understanding the sentence constituent value of the displaced constituent in the subordinate clause. S

g

522 8.

Lisbeth Falster Jakobsen It seems that the plural third person pronoun de in the nominative lends itself a little more to attributes: (iii) (about smoking hash:) [DeNom, der er til den slags pä for eksempel en Roskilde They-nom who are to this sort on for example a Roskilde Festival,]Sub deSubCorNom ved, hvad Festival, they-nom know what

9.

der kan ske derude. there may happen out there

'Those who like that sort of things at a Roskilde Festival for instance, they know what might happen there.' (Pol. 12.7.2002 1,6) In a construction governed by a preposition, the nominative case is becoming still more frequent, perhaps due to the growing lack of certainty of usage. In prepositional phrases the case as such has no grammatically distinctive value. This lies in the preposition, thus the case form is not crucial: (iv) Men for denom der ikke husker Informations But for they-acc who not remember Information-gen begyndelse, har avisen et image som venstreorienteret. start, has newspaper-def an image as leftish

'For those who do not remember the start of "Information", the newspaper has the image of being leftish.' (Pol. 25.1.2003 11,6) 10. Professor Michael Fortescue has had the kindness to correct my English language. Remaining errors and inadvertencies are of course my own responsibility.

References Bakker, Dik 1998 Language change: some implications for the theory of Functional Grammar. Working Papers in Functional Grammar 66, University of Amsterdam. 1999 FG Expression Rules: from templates to constituent structure. Working Papers in Functional Grammar 67, University of Amsterdam. Bolkestein, A. Machtelt 1998 What to do with Topic and Focus? Evaluating pragmatic information. In Hannay and Bolkestein (eds.), 193-214.

Pronominal expression rule ordering in Danish

523

Connolly, John H., Roel M. Vismans, Christopher S. Butler and Richard A. Gatward (eds.) 1997 Discourse and Pragmatics in Functional Grammar. Berlin: Mouton de Gruyter. Dik, Simon C. 1997 The Theory of Functional Grammar, 2 vols. 2nd, revised edition, edited by Kees Hengeveld. Berlin: Mouton de Gruyter. Hannay, Mike and A. Machtelt Bolkestein (eds.) 1998 Functional Grammar and Verbal Interaction. Amsterdam: Benjamins. Heltoft, Lars 1997 Hvem opslugte hvo - Et bidrag til det moderne kasussystems udvikling. In Ord, Sprog og artige Diet. Festskrift til Poul Lindegärd Hjort, Flemming Lundgreen-Nielsen, Marita Akhoj Nielsen, and Jens Kousgaard Sorensen, 227-256. Copenhagen: C.A.Reitzel. Hengeveld, Kees 1997 Cohesion in Functional Grammar. In Connolly, Vismans, Butler, and Gatward (eds.), 1-16. Kroon, Caroline 1998 Discourse markers, discourse structure and Functional Grammar. In Connolly, Vismans, Butler, and Gatward (eds.), 17-32. Liedtke, Frank 1998 Illocution and grammar: a double level approach. In Hannay and Bolkestein (eds.), 107-128. Jorgensen, Henrik 1991 Om de danske personlige pronominer [About the Danish personal pronouns]. Danske Studier 1991: 5-28. Copenhagen: C.A.Reitzels Forlag. 2000 Studien zur Morphologie und Syntax der festlandskandinavischen Personalpronomina, mit besonderer Berücksichtigung des Dänischen. Ärhus: Aarhus University Press. Moutaouakil, Ahmed 1998 Benveniste's recit and discours as discourse operators in Functional Grammar. In Hannay and Bolkestein (eds.), 25-42. Vet, Co 1998 The multilayered structure of the utterance: About illocution, modality and discourse moves. In Hannay and Bolkestein (eds.), 1-24.

524

Lisbeth Falster Jakobsen

Primary data Nergaard, Lise 2001 Julen er hjerternes fest. Copenhagen: Gyldendal. (N0rgaard) Thorup, Kirsten 1979 Den lange sommer. Copenhagen. (Thorup) Copies of the newspaper Politiken (Pol.) Copies of the weekly Weekendavisen (Weekend.)

Index of names

Aarts, Bas 412 Acuna-Farina, Juan Carlos 383, 386, 388,389, 390,413 Aissen, Judith 464, 465 Alfaro, Juan Jose 281 Anagnostopoulos, Elena 213 Anderson, Gregory D.S. 213,214 Anderson, John M. 499 Anderson, Stephen R. 97, 100, 129, 499 Anstey, Matthew 83, 317, 319, 321, 325, 339, 344, 347 Aoun, Joseph E. 27, 29, 36, 37 Arends, Jacques 264, 265, 266, 277 Ariel, Mira 36,209,221,242 Arnold, Jennifer 449, 450, 466, 479 Augusta, Felix Jose 455, 456, 458 Bahloul, Mäher 17 Baker, Mark C. 87, 97, 98, 99, 115, 118, 119, 128, 129 Bakker, Dik 1, 4, 5, 30, 35, 36, 41, 45, 46, 47, 48, 49, 53, 70, 72, 73, 83, 206, 207, 221, 222, 234, 235, 242, 243,305,519, 520 Barlow, Michael 205, 206 Bates, Elisabeth 475 Becker, Misha 269, 270, 277 Belnap, Robert K. 29 Benveniste, Emile 311 Berg, Marinus van den 276,375, 413 Bhatt, D.N.S. 36 Biber, Douglas 318 Birner, Betty J. 169, 170, 171, 172, 184, 189, 195 Blinkenberg, Andreas 194 Bloomfield, Leonard 443

Bohnemeyer, Juergen 317,321, 322, 326, 347, 348 Bolkestein, A. Machtelt 375, 503, 518,519,520 Boogaart, Ronny 327, 344 Borillo, Andree. 165, 166, 168, 184 Bresnan, Joan 196,204 Broadbent, Sylvia Μ. 217 Bybee, Joan 319 Carminati, Maria Nella 173,176, 193, 195, 197, 198 Chapa, Carmen 281,311 Chisarik, Erika Z. 149,160 Chomsky, Noam 34, 37, 92, 204 Climent, Jaime 310 Cogill-Koez, Dorothea 281 Comrie, Bernard 69,233,303,331, 334, 336, 448 Connolly, John H. 42, 43, 44, 45, 48, 195, 196, 198, 520 Cook, Eung-do 95 Cooke, Joseph R. 242 Corbett, Greville G. 18, 19, 20, 25, 29, 33, 106, 182,203,413 Cornish, Francis 168, 184, 185, 518 Coseriu, Eugenio 310 Costa, Joäo 172, 175, 176, 177, 195, 196, 197 Coulthard, Richard M. 57 Croese, Robert A. 450 Croft, William C. 33, 34, 205 Crowley, Terry 243 Cuvalay, Martine 107, 108, 129, 130, 320, 347 Daglish, Gerard M. 214 Dahl, Osten 407, 408

526

Index of names

Dahlgren, Sven Olof 179 Davies, H. John 233, 234 Dechaine, Rose Mary 436, 437 Dik, Simon C. 3, 5, 8, 12, 33, 34, 37,41,42, 43,48, 53,83,84, 90, 107, 111, 127, 129, 130, 137, 163, 165,204, 205,206, 207, 215,241,242, 250,251,282, 283,292, 303,310,317,319, 320,321,325,327,331,339, 340, 341,345,347,348, 351, 353, 355, 357, 359, 360, 369, 370,371,373,377, 424, 426, 441,443,447, 450, 462, 497, 503, 506,509,512,517,518,521 Dixon, R.M.W. 94, 123, 124, 129 Dryer, Matthew S. 435,489 Ebert, Karen 210 Emmorey, Karen 281 Erteschik-Shir, Nomi 174 Evans, Nicholas 127 Everett, Daniel 154,156,160 Fernandez Garay, Ana 450, 460 Foley, William A. 11,55,56 Fontanella, Maria Beatriz 450, 459 Forrest, Linda B. 449,470 Fortescue, Michael 277, 522 Fournier, Ntalie 165, 166, 167, 170, 178 Frajzyngier, Zygmunt 213, 220 Franck, Julie 413 Friberg, Barbara 10 Fries, Udo 383,413 Fuchs, Catherine 166,178 Garcia Velasco, Daniel 68,91,171, 195,498 Gamal-Eldin, Saad 13, 26 Gelderen, Elly van 35 Gerdts, Donna Β. 66

Givon, Talmy 186,209,241,257, 426 Goossens, Louis 343 Greenberg, Joseph 249 Gregores, Emma 236 Grice, H.Paul 332 Grimes, Joseph E. 450 Groot, Casper de 84, 127, 137, 159, 207,211,215,261,262, 263, 276, 296, 347 Gundel, Jeanet Ν. 221 Haberland, Hartmut 448 Haegeman, Liliaene Μ. 9 Haiman, John 20, 34, 210, 233, 243 Hannay, Michael 57, 182, 185, 186, 187, 188, 189, 309, 520 Harbert, Wayne 17 Harder, Peter 501 Haugen, Einar 382, 383, 385, 389 Hawkins, John A. 395 Heath, Jeffrey 452 Heine, Bernd 211 Hekking, Ewald 235 Heltoft, Lars 517 Hengeveld, Kees 5, 34, 53, 64, 68, 83,90,91,92, 107, 108, 127, 129, 135, 138, 163, 164, 182, 184, 185, 190, 220, 249, 250, 251,252, 253, 254, 256, 259, 260, 262, 271,275,276, 277, 282, 283, 285, 286, 289, 290, 291,292, 294, 295,298,303, 304, 308, 309,310,319, 320, 328,346, 347, 401,402, 450, 453, 469, 476, 479, 483, 488, 498, 503,519, 520 Herrero-Blanco, Angel 277 Herring, Susan 190 Hieda, Osamu 2 Hobaek Haff, Marianne 196 Hockett, Charles F. 383, 384, 385, 389, 448, 449, 465

Index of names Hopper, Paul 101,128 Hovdhaugen, Even 58 Hudson, Richard A. 399 Huffman, Alan 164, 184 Jakobson, Roman 59,311 Jespersen, Otto 291 Jorgensen, Henrik 515, 516, 517 Kanerva, Jonni M. 196 Keizer, M.Evelien 107, 108, 129, 130, 347,381,397, 401,402, 413,414 Kern, Barbara 154,156,160 Kiss, E. Katalin 148 Klein, Wolfgang 317,325,326, 331,332, 336,337,339, 341, 347, 348 Korzen, Hanna 196, 197 Kroon, Caroline 57, 90, 220, 375 Kuno, Susumo 447, 451 Laczko, Tibor 149, 160 Lakoff, George 357, 358 Lambrecht, Knut 167, 186, 195 Langacker, Ronald W. 95 Lazard, Gilbert 191, 192 Lee, Donald W. 383,384 Lee, Kee-dong 102, 104, 105, 129 Lehmann, Christian 205, 206 Lenz, Rodolfo 450 Levelt, Willem J.M. 54, 469, 476 Levin, Beth 170, 196 Li, Charles N. 260,261 Liddell, Scott 281, 299 Liedtke, Frank 375 Lier, Eva van 147,272 Limburg, Machiel J. 207, 211,215 Longobardi, Giuseppe 290 Lord, Carol 243, 405 Lucas, Ceil 281 Lyons, John 309

527

MacDonald, Lorna 208, 218 Mackenzie, J.Lachlan 53, 56, 64, 68, 83, 103,335,347, 372,414, 489 MacWhinney, Brian 475 Mallison, Graham 264 Mchombo, Sam A. 204 McWhorter, John 266, 267, 268, 277 Meijs, Willem 159 Meillet, Antoine 309 Metzger, Melanie 281,299 Meyer, Charles F. 381 Mithun, Marianne 87, 88, 93, 94, 95,96, 97, 100, 101, 102, 103, 104, 106, 111, 112, 116, 118, 119, 120, 122, 124, 125, 126, 128, 129, 130 Mohammad, Mohammad A. 12,18, 35 Moravcsik, Edith A. 205, 413 Morgan, Jerry L. 413 Mosel, Ulrike 58 Moutaouakil, Ahmed 179,180,183, 195, 197, 198,351,357,361, 368, 369, 370, 371, 374, 377 Munro, Pamela 243 Nichols, Joanna 204 Nicol, Janet 413 N0lke, Henning 166, 168 Oates, Lynette F. 116,129 Olmsted Gary, Judit 13,26 Olson, Clifford 204 Padden, Carol 281 Payne, Doris L. 3,70,214 Perlmutter, David M. 192,193,194 Pinkster, Harm 289 Pinto, Manuela 172, 173, 174, 175, 193, 197

528

Index of names

Polinsky, Maria 473 Prince, Ellen 221,404 Pustejovsky, James 173 Pustet, Regina 258 Quirk, Randolph 383, 384, 389, 413 Reid, Wallis 413 Rickford, John R 255, 256, 257, 258, 276 Rijkhoff, Jan N.M. 22,369,371 Rivano, Emilio 463, 473, 474, 479 Rosen, Sara T. 97,99, 101, 115, 124, 128, 129 Sadock, Jerrold M. 87, 92, 94, 97, 98, 110, 127, 128 Salazar-Garcia, Ventura 277 Saltarelli, Mario 2 Sasse, Hans-Jürgen 127 Searle, John R. 354,355,468 Seiler, Hansjakob. 205 Shay, Erin 213 Siewierska, Anna 4, 5, 30, 35, 36, 165, 183,206, 207,219, 241, 242, 479 Silverstein, Michael 448 Sinclair, John M. 57 Smeets, Ineke 450, 466, 467, 468, 479 Smith, Carlota S. 264, 276, 326, 343, 347, 388 Sopher, H. 383,384,389

Speas, Margaret 204 Staden, Miriam van 55 Stanchev, SvillenB. 165,183 Stassen, Leon 257 Steedman, Mark J. 328, 329 Steen, Gert van der 36 Supalla, Ted 281 Thomason, Sarah 255 Thompson, Sandra A. 101,128, 260, 261, 389 Thomsen, Ole Nedergaard 70, 71, 447, 448, 459, 462, 464, 473 Tomlin, Russell S. 449, 450, 468, 469, 470, 471,479 Valli, Clayton 281 Van Valin, Robert D. 9, 204, 241 VanPatten, Bill 271,277 Velazquez-Castillo, Maura 87, 96 Verhagen, Josje 272 Verkuyl, Henk J. 319 Vet, Co 107, 129, 347, 368, 375 Vigliocco, Gabriella 413 Villabrille, Francisco Fernandez 281 Vries, Lourens de 83 Ward, Gregory 172, 184, 189, 195 Wilhelm, Andrea 95 Wurm, Stephen 483 Zandvoort, Reinard W. 413 Züniga, Fernando 458, 479 Zwicky, Arnold 399

Index of languages

African-American Vernacular English 255, 256, 257, 258, 269, 273, 274, 275, 276, 277 Ani 211,241 Arabic 3, 5, 12, 13, 16, 17, 18, 19, 20,21,24, 25,26, 27, 28,29, 30, 31,35,36, 37, 163, 164, 179, 180, 181, 182, 183, 191, 192, 195, 205, 221, 260, 357, 362, 364, 365, 366 Bartangi 214 Basque 1, 2 Boumaa Fijian 94, 124, 129 Caddo 95 Camling 210,211 Chichewa 196,205 Chinese 260, 261, 273, 274, 275, 285 Danish 70, 71, 474, 503, 504, 506, 507, 508,511,512,513,514, 519, 521 Dutch 33, 34, 68, 103, 135, 136, 138, 142, 143, 144, 145, 146, 154, 158, 159,214,215,223, 228,231,238,243,310,327 English 6, 36, 42, 43, 44, 66, 76, 81, 99, 103, 109, 114, 159, 163, 164, 165, 169, 170, 171, 172, 173, 174, 180, 182, 185, 187, 188, 193, 194, 195, 196, 197, 199, 208,216, 231,234, 235,243, 251,255,257, 269, 270, 271, 272, 273,274, 275,291,297,

299, 301,309,310,311,318, 319, 324, 325,327,330, 334, 336, 337, 338, 339, 340, 342, 343, 345, 347, 352, 357, 386, 412,413,433,438, 440, 441, 474, 500, 507, 522 Fore 11 French 69, 163, 164, 165, 166, 168, 169, 170, 171, 172, 173, 175, 178, 180, 182, 192, 193, 194, 195, 196, 197, 199, 241,356, 357, 363, 365, 374, 391 German 241, 263, 264, 327, 391, 392 Greek 213 Guarani 96, 236 Gumawana 204 Gunwinggu 116,117,129 Hdi 212,213 Hua 20, 209, 210 Hungarian 101, 135, 146, 147, 150, 152, 154, 158, 159, 160, 179, 216, 261, 262, 263, 273, 274, 276, 296 Italian 163, 164, 172, 173, 174, 175, 180, 183, 192, 193, 195, 296 Ket 213,214,239 Kobon 233,234 Koegu 1, 2 Konjo 10 Kusaiean 102, 104, 105, 106, 107, 109, 110

530

Index of languages

Latin 42, 204, 260, 285, 286, 289, 309 Mapudungun 450,451,452,453, 454, 456, 457, 460, 462, 463, 464, 470, 471,472, 473,474, 475, 476, 477, 479 Mohawk 93, 112, 115, 124, 125, 204 Mupun 220 Nakanai 211,241 Ono 165,232,233 Otomi 235 Paolillo 190 Plains Cree 419,429,440 Portuguese 163, 164, 172, 175, 176, 183, 185, 191, 192, 197, 243,260 Rumanian 263, 264 Samoan 58, 93, 94 Sandawe 214 Saramaccan 265 Selknam 1,2,241 Slovak 264

Southern Tiwa 66, 76, 79, 81, 84, 96, 97, 99, 115 Spanish 58, 59, 76, 78, 81, 204, 212,213,216, 230, 231,232, 234, 235, 236, 237, 260, 271, 274, 275, 277, 285, 287, 289, 295, 298, 301, 303, 304, 309, 310, 424, 500 Sranan 264, 265, 266, 268, 273, 274, 275, 276 Swahili 266, 267, 268, 273, 274, 275 Tanggu 483, 484, 486, 487, 488, 489, 490, 491, 496, 497, 498, 499 Tauya 208 Tidore 55 Turkish 101, 102, 163, 164, 177, 178, 179, 180, 183, 184, 190, 195 Tuscarora 118 Wari(OroNao) 135,154,157,158, 160 West-Greenlandic 93,99, 115, 128 Yagua 1,3,70 Yucatec Mayan 93, 102, 110, 111, 122

Index of subjects

Aboutness 186,448 Absentive 296 Absolutive 419,483 Accessibility 21, 221 Active 427,447 Agreement 1, 203 Alignment 427 Anaphoric operator 215 Anaphorical chain 55 Animacy hierarchy 426 Apposition 210, 381 Approximation 68 Articulation 57 Aspect 317 Continuative - 293,324 Egressive - 324 Frequentative 342 Habitual - 340 Imperfective - 321 Ingressive - 324 Nominal - 22,369 Perfect - 326 Perfective - 321 Perspectival - 321,325 Phasal - 68,321,323 Progressive - 323 Prospective - 326 Assertion 358 Autolexical syntax 98 Bare predicate 251,284 Boundary operator 321 Case marking 484, 503 - system 419 Case-manipulative noun incorporation 111

Centripical expansion 3 Classifier noun incorporation 118 Clitic doubling 213 Close apposition 381 Conceptual component 57 Constituency 24 Constituent ordering 41 Constructed action 298 Contextual component 57 Continuative aspect 293, 324 Coordination 25 Copula 283 - deletion 256 -support 250,281,304 Copying 211 Cross-reference 203, 456 Dactylology 295 Defmiteness 19, 68, 388, 394 Degrees of exclamation 363 Deontic modality 329 Depth-first principle 4, 46, 73, 519 Descriptive layer 319 Diachronic change 515 Diathesis 447 Discourse act 56 Discourse discontinuity 490 Discourse function 402 Discourse grammar 55 Discourse Linking Theory 205 Discourse particle 55 Discourse-manipulative noun incorporation 116 Dynamic Expression Model 1,41, 53 Egressive aspect 324

532

Index of subjects

Elastic template 44 Embedded non-verbal predication 261 Empathy hierarchy 451,473 Epistemic modality 329 Ergativeverb 169 Exclamation 351 Expression of clauses 154 Expression of terms 146 Expression of words 135 Expression rules 54 Familiarity scale 221 Fingerspelling 295 Form-order 10 Formulation 54, 67 Frame 46, 67, 145 Act- 78,79,81 Communicative - 518 Declarative - 78 Event - 73,78 Illocutionary - 64, 90 Imperative - 73 Predicate- 102,465,476 Predication- 76, 107, 108, 118, 477 Frequentative aspect 342 Functional Discourse Grammar 53, 60, 87, 163, 476, 503 Functions 72 Gender 14 Givenness hierarchy 221 Government and Binding 9, 98 Grammatical relations 419 Grammaticalization 462 Ground 182,473 Habitual aspect 340 Headedness 384,398 Hierarchy of predicability 252 Holophrases 56

Iconicity principle 365 Illocution 351 Imperfective aspect 321 Inalienable possessed 124 Information structure 167,183,187 Ingressive aspect 324 Interpersonal level 63, 90, 105, 113, 189 Intransitive verb 169 Inversion 419,447 Language acquisition 269, 271 Language change 259 Language contact 234, 261 Levels of representation 59 Lexemes 67, 319 Lexical Functional Grammar 204 Lexicon 498 Linearization 42 Locative inversion 163 Mapping rules 54 Mitigation 68 Modality 351 Deontic- 329 Epistemic - 329 Objective - 68 Subjective - 68,360,375 Modes of message management 187 Morphosyntactic encoding 68 Morphosyntactic templates 135 Move 57 Natural language user 475 Neutral mode 187 New subject 182 Nominal aspect 22, 369 Non-deictic predicate 299 Non-presentative predication 298 Non-verbal predication 249, 281 Embedded - 261 Noun-incorporation 87, 92 case-manipulative - 111

Index of subjects classifier- 118 discourse-manipulative- 116 unitary - 104 Number 14 Object 428 Objective modality 68 Obviation 447 - span 465 - tier 465 Obviative 424,431 One-word utterance 76 Operator 321 Order of constituents 504 Output component 57 Overgeneration 4 Owner 297 Parallel processing 23 Passive 427,447 Past participle 135 Perfect 326 Perfective aspect 321 Person 14 Person hierarchy 424, 436 Person marking 203, 217, 419 Personal pronoun 503 Perspectival aspect 321, 325 Phasal aspect 68,321,323 Phonological encoding 70 Phonological level 66, 151 Placement rule 43 Portmanteau 217 Possessive construction 295, 297 Possessor raising 124 Pragmatic function 12 Pre-affixal syntax 98 Predicability 252,281 Predicate, bare 251,284 Predicate, relation 351 Predicate-frame 102, 465, 476 Predicate worthiness 258

533

Predication-frame 76, 107, 108, 118,477 Present 334 Presentative mode 187 Presentative predication 294 Primary Object 474 Primary operator 68, 143 Primitives 67 Pro-drop 204 Progressive aspect 323 Prospective 326 Proximate 424,432 Quantification 317,339 Quantifier float 428 Radical Construction Grammar 205 Reaction mode 186 Redundancy rules 12 Reference 389 Referential predicate 251,284 Referring layer 320 Relation predicate 251 Relational Grammar 193 Relational predicate 284 Reportative 68 Representation of person 222 Representational level 64, 90, 106, 114, 150, 190, 497 Resolution 18 Role and Reference Grammar 9 Sandwich application 509 Secondary Object 474 Secondary operator 152 Semantic function hierarchy 426 Semantic omissibility 389 Sign language 281 Simple past 334 Speech act 354 Speech errors 234 Stager 182 State of affairs 319,331,447

534

Index of subjects

Structural level 65, 91, 109, 115, 150, 190 Subject 428, 438, 473 Subjective modality 68, 360, 375 Synchronic variation 255 Syntactic template 43 Synthetic compound 138 Synthetic omissibility 384 Tag-question 358 Tail-head linkage 55 Template 6, 43, 74, 119, 135, 146, 165,231,504 Temporal structure 331 Tense 14,68,317 Term structure 369 Thetic 185 Time-stability scale 270 Top-down process 6, 54, 163, 519 Topic mode 186 Topic time interval 331

Topicworthiness 448 Trace 47 Typological hierarchies 249 Typology of exclamatives 3 61 Typology of noun incorporation 120 Unaccusative 169 Undergeneration 1 Unitary noun incorporation 104 Vantage point 448 Variation 513 Verb agreement 12,400,421 Wh-question 290 Yes-no question 290 Zero-1 strategy 261, 286, 292 Zero-2 strategy 266, 286, 292