TITEL$1863 20-12-07 08:13:04 Rev 18.02 META Systems

The Noun Phrase in Functional Discourse Grammar

≥

Trends in LinguisticsStudies and Monographs 195

Editors

Walter Bisang(main editor for this volume)

Hans Henrich HockWerner Winter

Mouton de GruyterBerlin · New York

The Noun Phrase inFunctional Discourse Grammar

edited by

Daniel Garcı́a VelascoJan Rijkhoff

Mouton de GruyterBerlin · New York

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

�� Printed on acid-free paper which falls within the guidelinesof the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication Data

The noun phrase in functional discourse grammar / edited by DanielGarcı́a Velasco, Jan Rijkhoff.

p. cm. � (Trends in linguistics. Studies and monographs ; 195)Includes bibliographical references and indexes.ISBN 978-3-11-019867-6 (hardcover : alk. paper)1. Grammar, Comparative and general � Noun phrase. 2. Func-

tionalism (Linguistics) I. Garcı́a Velasco, Daniel. II. Rijkhoff,Jan.

P271.N679 20084151.5�dc22

2007050816

ISBN 978-3-11-019867-6ISSN 1861-4302

Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie;detailed bibliographic data are available in the Internet at http://dnb.d-nb.de.

” Copyright 2008 by Walter de Gruyter GmbH & Co. KG, D-10785 BerlinAll rights reserved, including those of translation into foreign languages. No part of thisbook may be reproduced or transmitted in any form or by any means, electronic or mechan-ical, including photocopy, recording or any information storage and retrieval system, with-out permission in writing from the publisher.Cover design: Christopher Schneider, Berlin.Printed in Germany.

Abbreviations vii

Introduction1

Prototypical and non-prototypical noun phrasesin Functional Discourse Grammar

43

Layers, levels and contexts in Functional Discourse Grammar63

On noun phrase structure in Functional (Discourse) Grammar:Some conceptual issues

117

Reference and ascription in Functional Discourse Grammar:An inventory of problems and some possible solutions

181

Interpersonal meaning in the noun phrase221

Freestanding noun phrases within documents:A pragmatic approach based on Functional Discourse Grammar

263

Agreement in the noun phrase:The dynamic expression of terms and what can go wrong

287

Functional Discourse Grammar and extraction from (complex)noun phrases

321

vi

Index of authors 365Index of subjects 369Index of languages 375

1 first person2 second person3 third personA addressee / adjectiveABL ablativeABS absolutiveACC accusative caseAdv adverb(ial)AdvP adverbial phraseALL allativeART articleAP adjective phraseASP aspectAUG augmentativeBNC British National CorpusC common genderCLF classifierCNPC Complex Noun Phrase

ConstraintCOND conditionalCOP copulaCREA Corpus de Referencia del

EspaD discourseDAT dative caseINF infinitiveDEF definiteDEM demonstrativeDET determinerDI Domain IntegrityDIM diminutiveDS different subjectECC Extra Clausal ConstituentEIC Early Immediate

Constituents hypothesis

EXH exhortativeF feminineF(D)G Functional Grammar and

Functional DiscourseGrammar

FDG Functional DiscourseGrammar

FG Functional GrammarFUT futureGEN genitive caseHP Head ProximityHPSG Head-Driven Phrase

Structure GrammarIND indicativeINDEF indefiniteIRR irrealisLCA Linear Co-occurrence

AxiomLFG Lexical Functional

GrammarM masculineN noun / neuterNEG negativeNF non-futureNH non-humanNOM nominative caseNP noun phraseNUM numeralOBJ objectPRF perfectivePL pluralPOSS possessivePP prepositional phrasePPT past participlePRES present

viii

PROG progressivePROH prohibitiveQ question markerR realisREFL reflexiveREL relative markerRSM resumptive markerS speaker / subject /

sentenceSG singularSIM simultaneousSTC Stepwise Term

ConstructionTNR time nominalizerTR transitional soundV verbVf finite verbVi nonfinite verbVP verb phrase

(see fn. 9 in Introduction)VOC vocative

Zero (semantic function)1 singularA Discourse Act /

addressee /anaphorical operator

Ag Agent(semantic function)

CL claused/def definiteExp Experiencer

(semantic function)i indefiniteIL Interpersonal LevelILL illocutionIMP imperative

L layerLex lexemem pluralML Morphosyntactic LevelP1 the (special)

clause-initial positionP speech act participantPat Patient

(semantic function)PL Phonological LevelPredP predicative phraseprox proximityR Referential Subact /

relative operatorRL Representational LevelRP referential phraseSo Source

(semantic function)T Ascriptive SubactUR underlying representationVP verb phrase in FDG

C Communicated contentE Speech act (FG)e State-of-Affairsep Episodef Property / relationF Speech occurrencel LocationM Movep Propositional content

(FDG)t Timex IndividualX Propositional content

(FG)

The articles in this volume analyse the noun phrase within the frameworkof (FDG), the successor to Simon C. Dik

(FG). was the main conferencetheme of the ,which took place at the University of Oviedo (Gij2004 and the present anthology consists of a selection of the manuscriptsthat were submitted after a call for contributions following the conference.1

There were several reasons to devote the theme session of the confer-ence to the noun phrase (NP). First, the most recent treatment of NPs byDik in terms of umously) in 1997, inthe first volume of . Given the fact thatFDG presents a strongly revised version of Dikkian FG with respect torules, variables, representations and overall design, it seemed appropriateto investigate how the new theory deals with one of the most basic gram-matical constructions, the noun phrase.Secondly, in a recent, cross-linguistic investigation on the structure of

the NP, Rijkhoff (2002) presented an alternative analysis of the NP withinthe FG framework. This study contains several new facts and ideas, whichmade it an interesting challenge to investigate to what extent the proposalshe put forward could or should be integrated into the new FDG model (seealso Rijkhoff this volume). All in all, the time seemed ripe for a detailedinvestigation of the way NPs are handled in FDG. Moreover, to analyse amajor linguistic construction from various perspectives (textual, typologi-cal, logical, semantic, morphosyntactic, etc.) is an excellent way to test anew model of grammar with regard to some of the standards of adequacyfor linguistic theories (see also section 1).In order to contextualize the papers in the present volume, we will first

lay out the main differences between FG and FDG and explain why someFG scholars felt that a general reorganization of the model was necessary.Section 2 gives a brief overview of the history of the way NPs have beenanalysed in F(D)G,2 paying special attention to variables for the variousentities (i.e. the

2

(i.e. the ntational frames, layers, levels, modules).Finally, in section 3, we will summarize the most relevant aspects of thearticles included in the volume. The reader is invited to consult Dik(1997a; 1997b), Anstey and Mackenzie eds. (2005), Hengeveld (2004a),Hengeveld and Mackenzie (2006) and Hengeveld and Mackenzie (forth-coming) for more detailed expositions of F(D)G.

FDG may be seen as the natural descendant of FG inasmuch as it sharesmany if not most of the central assumptions and goals formulated by Dik(see Butler 2003 and Anstey 2004 for excellent presentations of the evolu-tion of FG). In Dik -13) view the aim of FG was

to provide the means and principles by which functional grammars of par-ticular languages can be developed. And the highest aim of a functionalgrammar of a particular language is to give a complete and adequate accountof the grammatical organization of connected discourse in that language( -quacy (in particular descriptive adequacy) such as have been formulated byChomsky

But since FG does not share Chomsky s syntactocentric perspective ongrammar, Dik (1989: 12; 1997a: 13) added that r-ences with respect to what has been called atory adequacypoint he introduced three additional standards of adequacy for grammaticaltheories, which subsequently have also been accepted in other functionalapproaches to grammar (e.g. Van Valin and LaPolla 1997: 8): pragmatic,psychological and typological adequacy. Undoubtedly, the standard oftypological adequacy has been satisfied to a greater extent than the othertwo, as many FG publications contain references to studies on a wide vari-ety of linguistic phenomena in languages from many different families,which have had a considerable influence on the general architecture of themodel. This strong commitment to typology continues to be an importanthallmark of FDG.There is little doubt that FG has been less successful in its aim to be-

come a theory of grammar that is also pragmatically and psychologicallyadequate (Butler 1991; 1999), and to some extent this has motivated thebirth of FDG. As far as pragmatic adequacy is concerned, some FG practi-

3

tioners felt that the 1997 model (in spite of Dikin the quotation above) contained two major obstacles to progress towardsa truly functional theory of grammar: (i) the fact that research in FG con-centrated almost exclusively on the analysis of the sentence and its con-stituents, and (ii) the lack of integration of FG into a theory of verbal inter-action. As for psychological adequacy, which roughly requires the theoryto be compatible with well-established findings in the field of psycholin-guistics, let us simply say that the number of studies in FG that make refer-ence to the results of psycholinguistic research is extremely limited, whichmeans that the theory has remained untested from a psycholinguistic pointof view.Thus, on the one hand FDG can be seen as a continuation of FG, on the

other hand the original model as conceived by Dik has been restructured insuch a way that it is better equipped to meet the standards of adequacymentioned above. Another major difference between FG and FDG is thatFDG is explicitly designed to model the speakers(Hengeveld and Mackenzie 2006: 669). So even though FDG shares certainfeatures with its predecessor FG, the differences are substantial enough tosay that it offers a new research programme in the linguistic landscape.Next we will examine the differences between the two models under theheadings ,, and

1.1. Sentence grammar vs. discourse grammar

In various places Dik expressed his commitment to the study of connecteddiscourse in language and it is true that over the years some FG linguistshave concerned themselves with discourse phenomena. Nevertheless, inreality FG has mostly concentrated on the analysis of the internal organiza-tion of single, isolated sentences. In fact, this development was alreadyanticipated in the first presentation of the theory, when Dik (1978: 15)stated:

FG is meant to cover any type of linguistic expression ( e-stricted to the internal structure of sentences, inasmuch as there are combi-nations of sentences related by syntactic and semantic rules. (

[emphasis ours].

4

Perhaps the focus on the internal structure of sentences was simply a mat-ter of priorities in the early stages of the theory, but Butler (1991) suggeststhat it may also be a consequence of the formal (in the sense of explicit)orientation of the theory, as discourse pragmatic aspects may be ratherdifficult to capture in a rigid notational system such as the one employed inFG. This is arguably an important drawback, which, according to Siewier-ska (1991: 2) makes FG amenable to the attacks of both formalists andfunctionalists alike. It seems, then, that over the years FG has re-interpreted its commitment to the study of language-in-use towards de-veloping a grammatical theory that is merely with a model ofverbal interaction.The need for a more serious treatment of discourse phenomena in FG

and a greater integration of grammar (the rules) and pragmatics (when toapply these rules) into the theory was especially strongly felt in the latenineties (see the papers in Hannay and Bolkestein eds. 1998). In this periodseveral authors proposed to enrich the architecture of the model either byadding a pragmatic module, the so-called (Kroon 1997,Bolkestein 1998, Vet 1998), or by extending Hengeveldanalysis of the sentence to account for the hierarchical organization ofdiscourse ( G 1996, Hengeveld1997). Essentially, this set the pillars for the new architecture of the model,since traits of both approaches can be found in FDG.Significantly, in his presentation of FDG, Hengeveld (2004a) notes that

a considerable number of grammatical phenomena relate to units that areeither larger or smaller than the clause and for that reason cannot be ade-quately described by a sentence grammar. In particular, he argues that FDGmust account for the fact that a quite a few verbal exchanges are not real-ized in the form of fully-fledged sentences, but rather in the form of frag-ments, or, generally speaking, non-clausal linguistic units (Mackenzie1998a). Thus, FDG replaces the sentence with the (Hannayand Kroon 2005) as a basic unit of grammatical analysis. The followingutterances are examples of non-sentential discourse acts:

(1) a.b.c.

The expression in (1a) may act as a vocative or address in order to initiatea verbal interaction, whereas (1b) and (1c) can serve as natural replies to aquestion. Hence, apart from their non-sentential nature, what these expres-

5

sions have in common is that they serve as complete contributions to dis-course interaction, and must obviously be studied within the context inwhich they occur.At the same time, there are linguistic phenomena that relate to the

organization of connected discourse and extend their influence over severalsentences. In his chapters on the representation of discourse, Dik (1997b)cites a number of phenomena that serve to establish coherence relations.These include iconic sequencing, topic continuity, focus assignment and

, which is here illustrated with an example from Kombai(adapted from De Vries 2005: 364):

(2) a.go.3SG.NF-until.DS year one finished.3SG.NF-TR-DS die.3SG.NF

b.die.3SG.NF-TR-DS bury.3PL.NF

Tail-head linkage usually involves the repetition of the verb at the begin-ning of the next clause (as in the case of in the example above),but sometimes it involves the verb of the penultimate clause or even theverb of the clause before that. Such cases show even better that tail-headlinkage can only be properly handled by a grammatical theory that goesbeyond the boundaries of a traditional sentence grammar.

1.2. Top-down vs. bottom-up

Unlike FG, FDG is a in which the generation of a linguis-tic expression is assumed to start from a communicative intention whichultimately leads to the grammatical coding of a piece of information and itsfinal articulation or execution ( n to articulationby Anstey (2004: 45), this adds to the pragmatic-centricity of the model asopposed to the predicate-centricity of FG. In this respect FDG differs frommost contemporary grammatical theories, which see the lexicon as thepoint of departure in the generation of linguistic structures. The top-downorganization of FDG is a reflection of its commitment to the standard ofpsychological adequacy mentioned earlier and was strongly inspired by thepsycholinguistic research of Levelt (1989).

6

Levelt, a and an . Conceptualization in-

volves the creation of a communicative intention and the construction ofpreverbal message, i.e. a conceptual structure that will serve as input to theFormulator. The process of Formulation translates this preverbal concep-tual structure into a linguistic structure (Levelt 1989: 11) and Articulationinvolves the execution of an acoustic plan by means of the appropriatephysiological organs. All three components have a place in the FDGmodel, as can be seen in (cf.

and ).Figure 1 shows that there are four major modules in the organization of

FDG: the conceptual, the contextual, the grammatical and the output com-ponents. Hengeveld (2004b: 369) emphasizes that the conceptual compo-nent is the trigger for the grammatical component to operate. In his view,the conceptual component is responsible for two types of processes: thedevelopment of a communicative intention, which has a direct link with theinterpersonal level in the grammar (see 1.3 below), and the creation of aconceptualization of that communicative intention, which directly connectsto the representational level (Figure 2).3

The contextual component represents the speech situation and includesboth linguistic and non-linguistic perceptual information. As the dynamicsof discourse unfold, the contextual component receives information fromthe grammatical component and provides the conceptual component withdata that are potentially relevant for the creation of new communicativeintentions and conceptualizations. One could say that it is of the majorfunctions of the contextual component to connect the grammatical compo-nent with the other modules.The output component corresponds to Levelt ator. However,

given the fact that FDG attempts to understand the structure of discourseacts as reflections of different kinds of knowledge deployed by the

(J. Lachlan Mackenzie, personal communication), ratherthan just the , the output may take different forms of expression(written, signed or spoken signs).It is important to emphasize that the conceptual, the contextual and the

output components are not part of the grammatical component (Hengeveldand Mackenzie 2006: 669). Although some scholars have made proposalsas to the contents or internal structure of the conceptual component (An-stey 2002, Nuyts 2004) and the contextual component (Connolly 2004),4

FDG, as a linguistic theory, is centrally concerned with the grammatical

7

component, whose internal structure will be laid out in greater detail in thefollowing section.

General layout of FDG (Hengeveld and Mackenzie 2006: 669)

1.3. Levels and layers

The grammatical component in FDG is shown in more detail in Figure 2(Hengeveld and Mackenzie 2006: 670).

Primitives

Pragmatics, Semantics

Morphosyntax, Phonology

Primitives

Formulation

Encoding

Articulation

Expression

8

The grammatical component in FDG

Figure 2 shows that FDG posits four levels of representation in the gram-matical component: the Representational level, the Interpersonal level(both inherited from classical FG), as well as the Morphosyntactic level(also referred to as the or ) and the Phonological

FramesLexemesPrimary operators

TemplatesAuxiliariesSecondary operators

Prosodic patternsMorphemesSecondary operators

Interpersonal Level

Representational Level

Morphosyntactic level

Phonological level

Formulation

Morphosyntactic Encoding

Phonological Encoding

9

level. As noted by Anstey (2004), there is a clear correspondence betweenthe levels and the main areas of linguistic analysis:

(3)

The ovals in Figures 1 and 2 indicate stages in the procedure at which par-ticular operations take place in the grammatical component.means that at this point pragmatic and semantic representations are beingproduced; the two stages indicate where morphosyntactic andphonological representations (in that order) are being generated. Eachoperation has of its own set of primitives in the form of frames, templatesand operators (among others). Although primitives are assumed to be lan-guage-specific (the lexemes of a language are perhaps the most obviousexamples of language-dependent primitives), FDG aims at discoveringsignificant cross-linguistic generalizations and hierarchies, which canpredict the number and type of frames and templates that a languageemploys on the basis of a limited set of parameters.Unlike FG, the four levels of representation are independently organ-

ized and relate to one another through , signified by arrows.One of the features of FG that is preserved in FDG concerns the use ofhierarchical (layered) representations to account for differences in scope.In the next section we will discuss the internal organization of each level inthe grammatical component in more detail. The discussion of layering,however, has been postponed to section 2.2, which deals with the layeredanalysis of both NPs and clauses (or ).

The interpersonal level captures the relevant details of the linguistic ex-pression that ad-dressee and Mackenzie 2006: 671). The structure of the inter-personal level is shown in (4) (Hengeveld 2005: 63):

(4) (M1: [(A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))

10

The central unit of analysis at this level is the Move (M), which is definedafter Kroon (1997: 20) as of discourse that is able toenter into an exchange structurecommunicative intentions (invitations, proposals, requests, etc.) and aremade up of one or more Discourse Acts (A). Mackenzie (2004: 183) arguesthat the expression in (5) contains one move, but three discourse acts, andcan thus be represented as in (6).

(5)(6) (M1: [(A1, A2, A3)] (M1))

Each discourse act is characterized on the basis of its illocution, repre-sented by means of an illocutionary frame in which variables are intro-duced for the speaker (P1)S and the addressee (P2)A

5 The general inventoryof illocutionary primitives is given in the following table (Hengeveld andMackenzie 2006: 672):

. Illocutionary primitives

DECLarativeINTERrogativeIMPERativePROHibitiveOPTATiveHORTativeIMPRectaiveADMOnitiveCAUTionaryCOMMissive

The third argument of the illocutionary frame is the communicated content(C), constructed on the basis of (R) and (T) .These variables, which were absent in FG, make it possible to distinguishbetween semantic entities and the pragmatic functions of reference andpredication or ascription. This is shown in the following example, in whicha term designating a spatial object (act in (7a) and in an ascriptive act in (7b):

11

(7) a.b.

The two NPs are now formalized as in (8):

(8) a. (R1: [a carpenter] (R1))b. (T1: [a carpenter] (T1))

Since all pragmatic aspects that determine the actual form of the linguisticexpression must be specified at the interpersonal level, pragmatic functionssuch as Topic and Focus are also added to the schema at this level.

The representational level s-tic unit that reflect its role in establishing a relationship with the real orimagined world it describes and Mackenzie 2006: 673), thatis, it reflects the use of language as a representational system rather than asa socializing system as at the interpersonal level. The following representa-tion shows the hierarchical organization of this level (adapted fromHengeveld 2005: 64):

(9) (ep1: [(p1: [(e1: [(f1: 1))] (e1))] (p1)) n) ] (ep1))

The variable , defined as a set of .There is an important difference between FG and FDG in that the proposi-tion (represented by the p-variable in FDG) is now assumed to be part ofthe representational level rather than the interpersonal level. As in FG, therepresentational level is constructed on the basis of predicates (symbolizedby the f-variable6) which designate a property or a relation. All lexemes ofa language are analysed as predicates and therefore represented as in (10):

(10) (f1: (f1))

When the slot is filled by a first-order noun (i.e. a nominal predi-cate designating a property of a spatial entity), the schema in (10) is usedto give a semantic description of one or more individuals (represented bythe x-variable; variables for other entity types are discussed below):

12

(11) (x1: (f1: N (f1)) (x1))

Terms and predicates may be combined to create predications, which des-ignate States of Affairs and are represented by the e-variable. As an exam-ple, consider the representation of the predicationdoor (see section 2.3 for a more complete representation of NPs in FDG):

(12) (Past e1: (f1: openV (f1)) (1 x1: (f2: manN (f2)) (x1))Ag(1 x2: (f3: doorN (f3)) (x2))Pat (e1))

The verbal predicate designates a relation between two individuals,and , which are characterized as the Agent (Ag) and the Patient

(Pat) of the action, respectively. The combination of these three units isused to describe a State of Affairs (symbolized by the e-variable), which, inturn, is located in the past time by the e representationin (12) also contains the term operator and , whichstands for singular number.

In classical FG, underlying representations like (12) are converted intoactual linguistic expressions after the application of a number of expres-sion rules that take care of the form and order of the constituents in sen-tences. Since these expression rules involve language specific features, it ishere that the differences between individual languages are made explicit.This indicates that in FG syntax is merely regarded as the actualization ofan underlying semantic representation. Indeed, syntactic constituents andword order are not considered primitive notions in FG, but the result ofcomplex interactions between general ordering principles.In FDG, by contrast, linear order and constituent structure have been

given a more prominent status with the introduction of a separate morpho-syntactic level. In accordance with basic functional methodology, FDGassumes that syntactic order can generally be explained on the basis of themeaning and use of linguistic expressions. However, the theory also admitsthat certain syntactic facts may be governed by independent principles,which warrant arate morphosyntactic level withinthe grammar, rather than as the output of the grammar as in FG(Hengeveld and Mackenzie 2006: 675).

13

Like the other levels of representation, the morphosyntactic level is fedby primitives in the form of , which provide slots for theinsertion of lexical units. Syntactic templates are reminiscent of

7 in FG, i.e. rather simple ordering schemas which define basicclausal syntactic configurations. Here are the functional patterns proposedby Dik (1980: 218, 220) for some clause types in English (Vf = finite verb,Vi = infinitival verb):

8

(13) a. P1 S Vf Vi O Xb. P1 Vf S Vi O Xc. P1 Vf Vi S X

Constituents at the representational level are assigned a position in a func-tional pattern through the application of placement rules, which are in turnsensitive to information specified in the underlying representation. How-ever, in this approach a ational leveldoes not always map directly onto a syntactic position in the functionalpattern. This problem does not occur in FDG, where the separation be-tween the representational and the morphosyntactic level allows for inde-pendent semantic and morphosyntactic representations of linguistic expres-sions. In view of the fact that F(D)G does not permit movement operations,this is especially useful in the case of syntactic discontinuity, extraposition,raising, etc. (see Hengeveld and Mackenzie 2005). Having separate levelsof representation in the grammatical component also makes it possible toaccount for cases where different semantic representations are mappedonto the same syntactic template (De Groot 2005: 150). Moreover, sincefunctional patterns as used in FG cannot handle any restrictions on theordering of elements within a construction, FDG has extended its inventoryof syntactic templates, which now also includes templates for the constitu-ents of the clause.The organization of the morphosyntactic level has been slightly mod-

ified since the first presentation of the FDG model. Initially the followinghierarchical structure was proposed (Hengeveld 2004a: 6):

(14) (Para1: [(S1: [(Cl1: [(PrP1: [(Lex1)] (PrP1)) (RP1: [(Lex2)] (RP1))](Cl1))] (S1))] (Para1))

The units distinguished in (14) are: (Para), (S),(Cl), (PrP), (RP) and (Lex).It was understood at the time that this representation is only a first ap-

14

proximation, given that languages may be more specific in the syntacticunits they distinguish. In subsequent publications, though, the terms Predi-cative Phrase and Referential Phrase have been replaced by traditionalconstituency labels such as Verb Phrase or Noun Phrase, for reasons to bediscussed in section 2.9

Moreover, the structure has been simplified by eliminating the variableson the right of the representation, whose role in the representation was notentirely clear. Thus, the morphosyntactic level in current FDG may also berepresented as follows:

(15) Para [ S [ Cl [ VP [ Lex1 n] NP [ Lex1 n ] ] ] ]

Although the morphosyntactic (and other) levels in FDG are supposed tobe fed by primitives, and can thus be considered static, some authors haveargued for a dynamic interpretation of FG expression rules (Bakker 2001,2005; Bakker and Siewierska 2002). Hengeveld (2005) shows that this ideais compatible with the general architecture of FDG, thus adding to the psy-chological adequacy of the model.

Before presenting a brief overview of the way Noun Phrases (Terms, Ref-erential Acts/Phrases) have been represented in and

, we must first say a few words about thelabel . There has always been a fair degree of reluctance touse the label Noun Phrase in FG and its successor FDG. The reason whyNoun Phrase is generally avoided in these theories is that it is a formallabel which only informs us about the intrinsic properties of a constituent,rather than a functional name that specifies the relation of a constituent tothe construction in which it occurs (cf. Dik 1997a: 126-127). If we were touse only functional names (such as , or ), however, wewould know very little about the intrinsic properties of the constituent inquestion. Ideally there should be straightforward names for linguistic cate-gories that tell us something about the formal the functional propertiesof a linguistic form or construction, but apparently such category labels arenot always easy to find (see below).So as to avoid the use of purely formal category labels in F(D)G, sev-

eral alternatives for the name Noun Phrase have been proposed, but so far

15

it has been difficult to come up with an acceptable label that avoids theform/function bias in its name. Thus, Hengeveld(Hengeveld 2004a: 7; on , see also Hengeveld this volume) metwith some resistance, because it did not quite cover all the forms and con-structions it was supposed to include (for example, semanticists claim thatphrases can only be referential if they have definite or specific reference;cf. Saeed 2003: 25).10 Earlier Dik (1997a: 55, 127) had proposed the word

, which was defined as ich can be used torefer to an entity or entities in some world f-erential Phrase is perhaps too narrow, the definition of Term is rather wide,since it basically includes any kind of expression that can fill an argumentor adjunct (satellite) position (e.g. pronouns, complement clauses, headlessrelative constructions).11 In sum, when we use a formal label such as NounPhrase, we exclude forms and constructions that can occur in the samefunction as a noun phrase (such as pronouns or complement clauses). Onthe other hand, names such as Term or Referential Phrase are either toogeneral (covering a wide variety of forms and constructions) or too narrowin that the definition only includes certain terms or noun phrases.The basic problem with names for linguistic categories seems to be that

they tend to be based either on formal or on functional properties and thatthere is usually no direct relationship between them. The same constituentmay occur in different functions, and the same function may apply to dif-ferent forms or constructions (cf. Dik 1997a: 26). This is shown in the fol-lowing examples from Dutch, each of which involves an adnominal modi-fier introduced by the preposition mples demonstratethat the same kind of form or rather construction ( ositional phrasecan be employed as a classifying, a qualifying or a localizing modifier.12

(16)a man of God

(i.e. a priest, a prophet, a religious leader)

(17)a woman of medium age

-aged woman

16

(18)the car of my sister

r

Conversely, a qualifying modifier (a functionally defined category) cantake the form of an adjective, a noun phrase introduced byclause:

(19)

(20)

(21)

These examples also show that we can characterize a linguistic category byusing a combination of a formal and a functional name. Thus, a name suchas informs us about the function (qualifying, ratherthan e.g. quantifying or localizing) as well as the form of the modifier inquestion (namely that the constituent belongs to the word class Adjective).One could do the same with Noun Phrase, i.e. combine this formal categorylabel with functional labels such as Subject ( ) or Qualifying( ), but it seems that in this case the problem outlined aboveis more difficult to circumvent (perhaps this is typical for higher-levelcategories like Noun Phrase or Term).13

In spite of this labelling problem, and for all practical purposes, wehave decided to use the name Noun Phrase, even though this volume is notrestricted to prototypical noun phrases (also devoting discussion to propernames, pronominal terms and other terms not headed by a proper noun) andin spite of the fact that a wide variety of both formal and functional issuesconcerningsame chapter (see e.g. Garc ution; for more discussionon the terminology issue see Bakker and Pfau this volume).

17

2.1. The Noun Phrase in Functional (Discourse) Grammar

It is possible to recognize three stages in the representation of NPs inF(D)G:

1. FG I: -up model;2. FG II: multi-layered (hierarchical) representations in a bottom-up model;3. FDG: multi-layered representations in a multiple-level, top-down model.

The first two stages are separated by the introduction of a layered model ofthe NP in Dik ; the introduction of

(section 1) marks the beginning of the third and presentstage.Initially NPs ( erms -hierarchical or

structures (Dik 1978: 57; Dik 1989: 55, 115):14

(22) ( xi: 1(xi): 2(xi): ...: n(xi)) [n 1]

or (Dik 1997a: 132):

(23) ( xi: 1(xi): 2(xi): ...: n(xi)) [n 1]

In this NP schema / stands for various kinds of grammatical categoriesin the NP (called term operators), such as

m-ber). xi is the variable ranging over the set of potential referents (see belowon variables for semantic categories).15 Each / signifies a predicate (typi-cally a nominal, adjectival or verbal predicate), and each (xi)/ (xi) is anopen predication in (xi), i.e. a frame in which all the argument positions butthat of xi have been filled. Open predications are also called restrictors,because they successively restrict the range of possible referents of the NP.The first restrictor ( 1/ 1) is normally the head noun; the others restrictorscan take a variety of forms, such as an adjective, a verb (often heading arelative clause), or a possessor NP (in which case we would be dealingwith a so-called term predicate; see (25) below). The colon between therestrictors indicates that the information to the right gives a specificationof, or a restriction on, the possible values of xi as it has been specified atthat point. To give an example, the underlying structure of the simple NP

is as follows:

18

(24) (d3xi: dogN (xi): bigA (xi))three big dogs

This is to be paraphrased as i such that the propertyi, such that the property i 1989:

115). A complex NP, which contains one or more modifiers,may look like this:

(25) (d3xi: dogN (xi): bigA (xi): {d1xj: gardenN (xj)Location}(xi))

In this example the third restrictor {d1xj: gardenN (xj)Location}(xi) is ex-pressed as a prepositional phrase ( ), which is analysed as aterm predicate with the semantic function Location (on term predicates, seeDik 1997a: 204-206).16

The next example has as the second restrictor a relative clause headed averb, which refers to an event or situation (i.e. a temporal entity), which issymbolized by the e-variable, introduced in Vet (1986), which is also em-ployed in FDG (see also section 1.3.2; m = plural, R = relative operator):

(26) (dmxi: dogN (xi): [Past ei: attackV (Rxi)Agent (dmxj: childN (xj))Goal])

Several changes were proposed to modify the original schema; this sectiononly highlights the use of different for semantic categories in themodelling of linguistic expressions. The employment of , anothermajor change in the representation of NPs, is discussed in section 2.2.Apart from variables for term operators ( ) and predicates ( ), the first

book-length publication on FG used only variables1978: xi). With the introduction of layering into FG

(Hengeveld 1989), however, the number of variables forexpanded considerably. Dik (1989: 50; 1997a: 93) lists variables for

five categories (notice that the y-variable has disappeared):

(27) ORDER STRUCTURE TYPE VARIABLE0 predicate Property/Relation f1 1st order term Spatial entity17 x2 predication State of Affairs e3 proposition Possible fact X4 clause Speech Act E

19

Dik (1989: 113; 1997a: 129) explicitly stated that entities are mental con-structs. Presumably many mental constructs of spatio-temporal entities donot exist independently of entities in the external physical world, but therelation between a mental construct of an entity and its counterpart in thereal world is a difficult issue with a long philosophical history that we willnot discuss here. Suffice to say that the problematic relation between men-tal and physical entities in the representation of linguistic expressions hasalso been given due attention in FG and FDG (e.g. Vet 1998), and is alsotouched upon in this book (see, for example, the contributions by Escri-bano, Keizer and Rijkhoff).Having different variables for the various kinds of entities offers several

advantages for the representation of linguistic expressions. To mention onethat is relevant for the current volume, they make it possible to distinguishbetween nouns that are used to talk about different kinds of entities (Dik1989: 180-181). Thus, an NP headed by a noun denoting a concrete objectlike (a first-order noun) contains an x-variable, as in (d1 xi: tableN(xi))

18 , whereas an NP headed by an event noun such as(a second-order noun) will contain an e-variable as in (d1 ei: meetingN (ei))

(Dik 1997a: 214-216).The use of different variables for different kinds of entities is also moti-

vated by the fact that different kinds of entities are specified for differentkinds of properties. Thus, a first-order noun like can be specified forspatial properties (such as weight, size, or colour

), whereas a second-order noun like can be characterized interms of temporal properties (e.g. duration, as in ; cf.Rijkhoff 2001).19 There are also other grammatical phenomena that can beaccounted for by using separate variables for distinct categories, such asthe fact that languages may employ different anaphoric pronouns for dif-ferent kinds of entities (Dik 1997b: 223-228). For example, English usesanaphoric for spatial objects (symbolized by the x-variable), whereasanaphoric is used for possible facts (here symbolized by the X-variable;notice that we only use skeleton representations to bring out the contrast):

(28) a.b. drop (Cherie) (xi: the briefcase), butpick_up (Tony) (Axi)

(29) a.b. think (John) (Xi: [Bill would win]) andthink (Peter) (AXi) too

20

In addition to the categories listed in (27), variables have been proposedfor places, times, manners and quantities and it seems that -logy of entities and Mackenzie eds. 2005: 166) will continue tobe expanded.20 The current list of variables for semantic categories in FDGat the Representational Level is as follows (Hengeveld and Mackenzie2006: 673):21

(30) SEMANTIC CATEGORY VARIABLE EXAMPLES (NOUNS)Individual xState-of-Affairs ePropositional content pProperty/relation fLocation lTime t

Notice furthermore that linguistic expressions such as NPs and clauses canbe regarded as constituting a separate ontological category with their ownset of variables. This makes it possible to distinguish between linguisticexpressions and the referents of those expressions (Rijkhoff 2002: 228).22

For example, one could argue that in (31) the pronoun refers to thereferent of , but that the pronoun in boldface refers to the name

rather than the referent of that name:

(31) A:B:(i.e. did you stop calling her ?

In FDG anaphoric reference to linguistic forms or constructions is ac-counted for at the morphosyntactic level, where they are provided with anindex (Hengeveld and Mackenzie 2006: 675; Hengeveld this volume).

2.2. Layering in the noun phrase

The most significant development in FG was Hengeveld a-lyse the clause as a hierarchically organized layered structure, in whicheach layer defines a different kind of entity. Furthermore, each individuallayer comes with its own set of operators and satellites, symbolized byindexed variables and (representing various kinds of grammatical andlexical modifiers respectively; Hengeveld 1989; Dik 1989: 50).

21

Hengeveld(Hengeveld 1990: 3-6, 12):

(32) (Ei: 4 ILL: 4 (S) (A) ( 3 Xi: (Xi): 3 (Xi))] (Ei): 5 (Ei)) IL

( 2 ei: [ 1 Predicate : 1 ( xi) n)] (ei): 2 (ei)) RL

xi: term : term operatorspred: predicate frame 1: predicate operators 1: predicate satellitesei: predication 2: predication operators 2: predication satellitesXi: proposition 3: proposition operators 3: proposition satellitesILL: illocution frame 4: illocution operators 4: illocution satellitesEi: clause 5: clause satellites

To illustrate this proposal with a concrete example, let us consider the rep-resentation of . Atthe Representational Level (RL), the speaker describes one event in whichthree big dogs engage in the activity of barking. This is constructed on thebasis of the predicate frame (roughly, argument structure) of the lexeme

, which offers one slot for the agent of the action. It is in that positionthat the NP representation in (23) is inserted. The verbal predicate is fur-thermore modified by the 1 operator Prog (progressive aspect), and thewhole predication is located in time by the 2 operator Past and in space bythe 2 satellite :

(33) (Past ei: [Prog barkV (d3xi: dogN (xi): bigA (xi))Agent ] (ei):(d1xj: gardenN (xj))Location (ei))

At the Interpersonal Level (IL), the speaker evaluates the actual occurrenceof the State of Affairs described in (33) as probable through the 3 proposi-tion satellite . Finally, the whole linguistic expression is character-ized as a declarative speech act, which is represented through the 4 illocu-tion operator DECL:

22

(34) (Ei: DECL ILL (S) (A) (Xi: (Xi): (probableA)Manner (Xi))] (Ei))

(Past ei: [Prog barkV (d3xi: dogN (xi): bigA (xi))Agent] (ei):(d1xj: gardenN (xj))Location (ei))

Hengeveld (1989,1997a), but not in its entirety. One major difference concerned the wayspeech acts are represented in FG: as an E-variable in Dik (1989: 50;1997a: 66) and as a predicate in Hengeveld r-rent FDG. The schema in (32) shows that Hengeveld i-cate ILL takes the speaker (S), the addressee (A) and the proposition (X) asits arguments and in this schema the E-variable stands for the utterancerather than the speech act. The illocutionary predicate frame is maintainedin FDG, as we saw in section 1.1.1, although the variable for speech par-ticipants is nowthe addressee.Hengeveld r-

personal Level and the Representational Level, a distinction due to B([1934] 1999) and Halliday ([1985] 2004). Even though there are obvioussimilarities between the schema in (32) and the schemas that are used atthe interpersonal and representational levels in FDG (section 1), there arealso certain differences, in particular with respect to the distribution oflayers and entities. In current FDG, for example, the proposition (nowsymbolized by the p-variable) is an entity that is specified at the Represen-tational level, as shown in (9), rather than the Interpersonal Level, as in FG(shown in (32) above).Inspired by Hengeveld -b, 1988, 1989) first proposals for a lay-

ered analysis of clausal structures and by some of Aristotle( V. II), Rijkhoff (1988) subsequently proposed a layered model ofthe noun phrase. Initially only three layers were recognized: the innermostlayer for qualifying modifiers ( ), the outermost layer for lo-calizing modifiers ( ) and in between a toaccommodate quantifying modifiers (it is important to point out that in thisapproachcategories, i.e. operators and restrictors or satellites). It was argued fromthe very beginning that the three-layered analysis of NPs also applied toclauses, indicating that NPs and clauses (or ratheranalysed in a similar fashion. To emphasize parallels between the underly-

23

ing structure of the NP and the clause, non-first restrictors were analysedas satellites.23 In Figure 3 the hierarchical organization of the layers is rep-resented in a concentric fashion, showing more clearly that modifiers at anouter layer have semantic scope over material specified at the inner layeror layers ( = clause satellite, = NP satellite). Recall that there is no one-to-one relationship between form and function and this is especially truefor satellites (as was shown in examples (16-21) above). Consequently,semantic modifier categories such as or , which are ex-pressed by affixes, function words or other grammatical markers, can stillbe more or less profitably connected with a certain layer, but in the case ofsatellites the relation between form and function is so indirect that we canonly list the various forms or constructions that are used as modifiers in theclause or NP (e.g. adverb(ial), adjective, relative clause).The three-layered NP structure also contained a slot for nominal aspect

markers, a new grammatical category whose members further specify theof a noun (i.e. lexicalized nominal aspect or

indicating that the entity it denotes is, for example, a singular object ratherthan a collective entity (Rijkhoff 1991; 2002: 100-121; this volume).24

Nominal aspect is, of course, the counterpart of verbal aspect: perfective orimperfective aspect markers further specify the of a verb (lexi-calized verbal aspect orAs in the case of Hengeveld -

layered analysis of NPs (and clauses or rather icationsadopted by Dik (1997a), in particular the distinction between qualifying,quantifying, and localizing operators (Dik 1997a: 163, 218).

(35) a.2-Loc 2-Quant ei: [ 1-Qual pred [V/A] (args)]

b.2-Loc 2-Quant xi: [ 1-Qual pred [N] (args)]

Notice, however, that both localizing and quantifying operators are treatedas (1997a: 219) believed e-search should be done on the interaction between Quantification and Lo-calization of SoAs

24

Location

3 3

Quantity

2 2

Quality

1 1

, Clause satellites: e.g.

V E R B

N O U NNP satellites: e.g

1 1

Quality

2 2

Quantity

3 3

Location

. Symmetry in the underlying structure of the clause and the NP: 1992version

Since then Rijkhoff has added two layers of modification to his version ofthe layered NP/clause model, one for in theouter periphery (layer number 4), the other for closeto the core (layer number zero). Figure 4 shows how the five layers distin-guished in Rijkhoff proposal (plus two additional clausal layers) are dis-tributed over the Interpersonal and the Representational Levels in FDG(notice that in this model all four entities have an intersubjective dimen-sion).

25

At the , modifiers are concerned with theInterpersonal Status of kinds of entities in the :[i] clauses (or rather the messages contained in the clauses),[ii] propositions, [iii] events and [iv] things.

( 6, 6)inform ddressee about the illocution-ary status of the clause (Decl, Int, ).

( 5, 5)inform aboutassessment of / attitude towards aproposition Xi as regards theprobability, possibility or desirabilityof the actual occurrence of event ei.

( 4, 4) ( 4, 4)specify the existential status of thing xi or event ei in the

At the , modifiers specify properties of spatio-temporalentities (things, events) in the in terms of the notionsKind (Class), Quality, Quantity, and Location.

3. ( 3, 3) 3. ( 3, 3)2. ( 2, 2) 2. ( 2, 2)1. ( 1) 1. ( 1)0. ( 0, 0) 0. ( 0, 0)

. Parallels in the layered model of the noun phrase and the clause(Rijkhoff to appear, this volume).25

The 5-layered NP model can be formally represented as the structure in(36), in which each operator ( , , rtain layer(L) as its argument. Notice that in this representation the Interpersonal andthe Representational levels do not clearly co-exist as separate entities (as incurrent FDG) and that the f-variable has been omitted (see note 17).

26

(36) NPi: 4[ 3[ 2[ 0[ NOUN(xi) ]L0 1(L0) ]L1 2(L1) ]L2 3(L2) ]L3 4(L3)

A somewhat simplified representation of the NPlooks like this:

(37) NPi: Def [ DemRemote [ 3 [ 0 [ dog(xi) ]L0 sniffer(L0) ]L1black(L1) ]L2 3(L2) ]L3 4(L3)

In this schema, xi symbolizes the referent of the NP and L stands for3 is the layer accommodating localizing modifiers 3 and 3).

Operators and satellites of layer N take the same argument (i.e. informationspecified in LayerN-1). The indexed variable NPi allows for reference to thenoun phrase as such.Rijkhoff argued that certain localizing modifiers, such as definite arti-

cles and noun modifiers like or , are better analysed as dis-course-referential operators ( 4) or satellites ( 4), in that they are directlyconcerned with the status of an entity in the shared world of discourse(Rijkhoff 2002: 229-238). Instances of discourse-referential modifiers inthe clause are realis/irrealis markers ( 4) and adverbials such as or

( 4). The claim that articles and (ir)realis markers should occupy thesame kind of modifier slot in a layered representation of the NP and theclause is substantiated by the fact that some languages use the same markerfor spatial and temporal entities to indicate that an entity is grounded in theworld of discourse (Rijkhoff 1990; Rijkhoff and Seibt 2005). In these ex-amples from Fongbe the element in question is glossed as DET ( r-miner 26

(Lefebvre 1998: 94, 99; see also Lefebvre and Brousseau 2002)(38)

I eat crab DET

(in question/that we know of)

(39)John arrive DET

Further research into the layered NP structure and similarities between theunderlying structure of the NPs and the clause has resulted in a schema thatalso has a special layer for classifying modifiers (Rijkhoff to appear, forth-

27

coming a-b; this volume). Classifying modifiers ( 0/ 0, 0/ 0) indicate whatof entity is being denoted, specifying features that (sub)-categorize

entities into a system of smaller sets, as in (40a-b):

(40) a.b 27

There is no longer a complete symmetry between operators and satellites inthe 5-layered NP/clause model (Figure 4). It has emerged that the elementsthat were originally categorized as operators are actually

operators and that ably do not occur(this is explained in Rijkhoff this volume).

2.3. The noun phrase in Functional Discourse Grammar.

We have seen two major developments in the representation of the NP inF(D)G. One was Rijkhoff y-sis of clausal structures to the NP, first suggested in 1988 and later partlyadopted by Dik (1997a). The other major change occurred with the intro-duction of FDG (Hengeveld 2004a-b, 2005; Hengeveld and Mackenzie2006). As mentioned in section 1, in certain respects FDG is a continuationof FG, in particular with regard to the basic assumptions, methods andgoals that characterize structural-functional approaches to grammar (Butler2003). But we can also see a clear break between FDG and its predecessor:the top-down organization, the focus on discourse acts rather than sen-tences as the basic unit of analysis, and the strictly modular approach withrespect both to the four main components in FDG (grammatical, concep-tual, contextual, output) and to the four levels inside the grammatical com-ponent (pragmatic, semantic, morphosyntactic, phonological).As to the NP, some features of the way NPs were analysed in FG have

been adopted in FDG, albeit perhaps in some modified form, but there arealso some important differences. First of all, we find a strict separationbetween the four levels in the grammatical component mentioned earlier.This is shown in (41) and (42) for the interpersonal and the representa-tional levels (Hengeveld this volume):

(41)( R RI: ............ (RI):

R (RI))

28

(42)( x xi: (

f fi: LexN (fi):f (fi)) (xi):

x (xi))

R Identifiability, Specificity R Subjective attitudex Location, Number x Referent modificationf Shape, Measure f Reference modification

Since discourse units are now analysed at four distinct levels, we also getfour representations of the same NP:

at the Interpersonal Level (IL)at the Representational Level (RL)at the Morphosyntactic Level (ML)at the Phonological Level (PL).

If we temporarily ignore the Phonological Level, the NPis represented as follows in FDG (Hengeveld this volume; R = referentialsubact, T = ascriptive subact):

(43) ( id RI: TI TJ (RI)) (IL)

(1 xi: (fi: girlN (fi)) (xi) : (fj: intelligentA (fj))(xi) ) (RL)

[[theart] [[intelligentA]AP girlN- NP1]NP2 (ML)

The same unit would be represented in classical FG as follows:

(44) (d1xi: girlN (xi): intelligentA (xi))

The most conspicuous difference between the two models resides in thegreater degree of exhaustiveness of the FDG representation thanks to theseparation of levels. The FG representation in (44) comes close to the Rep-resentational Level in (43), but again, there are some significant differ-ences. Definiteness, symbolized by r-sonal level in FDG and represented through the combination of theIdentifiability operator ( ntial subact variable

29

opens the way to a search for more interpersonal modifiers in the NP, atopic which is touched upon by several authors, especially Butler, in theircontributions to this volume.Another obvious difference results from the addition of the morphosyn-

tactic level in FDG. FG representations show semantic rather than syntac-tic properties and relations and, consequently, there is no underlying syn-tactic structure. Differences in constituent order across languages areaccounted for through general principles, which restrict the range of possi-ble syntactic configurations. FDG, however, assumes that general princi-ples of word order impose restrictions not on semantic representations, buton syntactic configurations, and thus restrict the potential set of templateswhich are available as primitives in the grammar of languages.The representations in (43) also show that, strictly speaking, we cannot

use labels such as sonal and the Representa-tional Levels in FDG. At the Interpersonal Level we only know that thelinguistic expression to be produced in the output component is the resultof a definite referential subact (RI), which contains the two ascriptivesubacts TI (girlN) and TJ (intelligentA). The Representational Level justgives us a semantic schema or frame in which thefilled with the nominal predicate .Another important contrast, this time with Rijkhoff , concerns

the way layers are organized in FDG, which uses the same basic format atall levels (Hengeveld and Mackenzie 2006: 671):

(45) ( 1: [head] ( 1): 1))

In this structure, 1 symbolizes some variable (e.g. Rx

by operator (and function ), andby the head and optional modifiers

By contrast, Rijkhoff instead of non-first restrictors)and both operators and satellites basically serve as predicates that take acertain layer in its scope (rather than a variable). Nevertheless, one couldstill argue that at some abstract level of representation NPs and clauses areassigned the same underlying structure both in Rijkhoffand in FDG (even though this is perhaps not always evident from thenames used for same-level operator and modifier categories in FDG, asshown in (42)).

30

Notice, finally, that the superscripts on the variables for operators andmodifiers in (40) and (42) indicate that FDG layers are not so much estab-lished on the basis of scopal differences between the various kind of opera-tors and satellites, as in Rijkhoff s model, but is rather motivated by theavailability of a variable for a particular kind of entity (R, x, f).

The previous sections have shown that there are basically three proposalsconcerning the way NPs should be analysed in F(D)G:

1. the analysis according to 1997);2. the analysis proposed in FDG (Hengeveld this volume; Hengeveld andMackenzie 2006);

3. the combined NP/clause analysis put forward by Rijkhoff (this volume;to appear).

FDG, as the natural successor to FG, has adopted some aspects of the wayNPs have been analysed in FG, but as we saw above, the new architecturehas also made it possible, and sometimes even necessary, to come up withsome new features (notably the strict separation between levels). To whatextent FDG can or will accommodate Rijkhoff sals depends onseveral factors. One of the major obstacles seems to be that in FDG layer-ing is intricately connected with variables for entities at the representa-tional level, whereas in Rijkhoff f-ferences in the semantic scope of operators and satellites.As may be expected, some of the papers in the present volume (those by

Hengeveld, Escribano, Rijkhoff and Keizer) primarily deal with matters ofrepresentation. Others, however, are more concerned with the practicalapplication of the model with regard to discourse-interpersonal matters(Butler, Connolly), whereas the contributions by Bakker and Pfau and byGarc mainly deal with morphosyntactic issues. It is also true,however, that one cannot make a very strict thematic division between thechapters in this volume, as one of the advantages of the FDG model is pre-cisely the fact that grammatical phenomena can be treated from differentperspectives (pragmatic, morphosyntactic, etc.) in a coherent fashion.In the opening article, Kees Hengeveld (University of Amsterdam, the

Netherlands) lays bare the analysis of the NP within the general structureof FDG. He argues that the separation between the interpersonal, the repre-

31

sentational, and the morphosyntactic levels of analysis in FDG allows for amore transparent and systematic treatment of noun phrases. Hengeveldtakes the prototypical noun phrase as his point of departure, which he de-fines as an NP with a nominal head that denotes a concrete, first-orderentity by lexical means and is used referentially rather than ascriptively. Hethen discusses examples of non-prototypical NPs from various languagesand shows how they can be analysed in FDG.Jan Rijkhoff (University of Aarhus, Denmark) puts forward an alterna-

tive, 5-layered model of the NP (with parallels in the clause) within thegeneral framework of FDG, which has separate layers for classifying,qualifying, quantifying, localizing and discourse-referential modifiers (inthis approach the termmodifier categories). In his view, there should be no special slot for amodifier that specifies a subjective attitudinal meaning (Hengeveld -modifiers), since such meanings can be expressed in many different ways(e.g. lexically, grammatically, morphosyntactically, prosodically or acombination of these). Rijkhoff then suggests that all components of theFDG model represent some kind of context and argues that a separate

( -context o-date elements from the extra-linguistic context (notably the speech situa-tion, including the speech participants), which in current FDG arerepresented in the and at the Interpersonal level inthe .Jos

provides a critical assessment of the ways the NP has been analysed in FGand FDG. He observes a number of inconsistencies in the way FDG usesvariables, operators and scope, which, he argues, can be eliminated if ahierarchical, binary-branching NP structure is assumed. He suggests FDGshould reintroduce Dik idea of and extendthe current ontology of entities. The result would be a more cogent treat-ment of scope and NP syntax, which does not force the theory to abandonany of its fundamental methodological principles.Evelien Keizer (University of Amsterdam, the Netherlands) also deals

with matters of representation in her analysis of the notions of referenceand ascription in FDG. She argues that variables at the representationallevel represent the t ntity described, ratherthan the intended referent. As a consequence, a clear separation is estab-lished between the grammatical component and the contextual componentin FDG. She tests the validity of her proposal against a wide variety of

32

forms and constructions, such as copular sentences, proper names, pro-nouns and appositive structures.In his article on interpersonal meaning, Christopher S. Butler (Univer-

sity of Wales, Swansea, UK) deals with several issues. The first part fo-cuses on the way speaker attitude is handled in the two NP models pre-sented in this volume: as a modifier by Hengeveld and as a globalqualification of an entity by Rijkhoff. In the second part Butler argues thatthere are two types of interpersonal meaning: one is concerned with thediscourse context (orienting the addressee towards the actuality or non-actuality of an event), the other with speech acts and certain types of mo-dality (having to do with social and personal context). The main body ofthe article investigates in considerable detail the kind of interpersonalmeaning that is oriented towards the social and personal context (i.e. thesecond type), which is illustrated with many examples from English andSpanish. The last section offers a detailed corpus-based investigation ofEnglish as an interpersonal modifier, when it is used to express sym-pathy towards the referent named in the NP.Pragmatic issues also take centre stage in the article by John Connolly

(Loughborough University, England, UK) on freestanding NPs in writtendocuments: NPs that are more or less peripheral to the main body of dis-course and have been characterized as instances of nguage(Quirk et al. 1985: 845 847). The article investigates how freestandingNPs may be treated in FDG and in particular examines the implications forthe integration of grammar and pragmatics. Connolly discusses severalkinds of pragmatic functions that a freestanding NP can fulfil, showing thatthese extra-clausal NPs are used in relation with linguistic as well as non-linguistic material (such as images). He argues that FDG should incorpo-rate freestanding NPs, proposes a preliminary list of additional pragmaticfunctions and suggests that the formulation ofrules acilitate the integration of FDG into a wider theory of verbalinteraction.Dik Bakker and Roland Pfau (both University of Amsterdam) deal with

agreement in the NP against the background of theproposed by Bakker in several earlier studies. The authors assume a

process interpretation of the F(D)G model in which matters of form andorder are interlaced. Their contribution is particularly concerned withagreement phenomena in the German noun phrase, which is noted for itsmorphological complexity. They show that their model can account forNP-internal agreement phenomena, including speech errors, as attested in alarge corpus of spoken German.

33

Daniel Garc (University of Oviedo) examines the so-calledComplex Noun Phrase Constraint within the context of FDG. The exis-tence of restrictions on the displacement of syntactic units is arguably achallenge to functional theories of grammar, which do not make use ofmovement transformations, because they seem to provide a strong argu-ment in favour of the need for an autonomous syntactic component ingrammar. However, the author argues that the constraint cannot be ex-plained on the basis of configurational restrictions alone and that the acti-vation status of referents in discourse is crucial to a proper understandingof the phenomenon. He then shows how the cognitive status of referentscan be represented in the contextual component of FDG.We hope this introduction gives the reader a good idea of the way NPs

are handled in FDG and that it shows that this new theory offers an inter-esting and valuable framework in which to describe and explain grammati-cal phenomena. The relation between linguistic theories and linguistic re-search should be bidirectional: theories open up new paths of research, andresearch, in its turn, provides results that eventually lead to changes in thetheory. Most of the papers in the present volume illustrate this dual rela-tionship: they test the theory by analysing the NP from different angles andon the basis of facts from various languages, ultimately suggesting certainmodifications of the theory. This, we believe, is clear evidence of thestrength and the flexibility of the new theory of

. Given the youth of the model there are of course many aspectsthat have remained undiscussed or need a more detailed treatment, but welook forward to the start of a lively debate on the basis of the data and pro-posals presented in this volume. Finally, we wish to express our sinceregratitude to several people who offered their generous help in the editingprocess. Unfortunately, the names of about a dozen colleagues cannot berevealed as they have acted as anonymous referees in the selection proce-dure, but we would like to express here our warmest thanks for their valu-able advice. We are also grateful to Lachlan Mackenzie who carefully readthe entire manuscript and offered many very useful comments, with regardto both form and content. Daniel Garc also wishes to acknow-ledge the financial support of the Dept. of Anglogermanic and FrenchStudies and the Research Vice-Rectorate of the University of Oviedo.

GijOctober 2007

34

1. The name is still used in the conference announcements,even though nowadays most of the contributions are concerned with

.2. In what follows, we will employ the acronym F(D)G as a way of conflating the

theory of Functional Grammar (FG) and Functional Discourse Grammar(FDG).

3. The need for a conceptual component in FG was noted as far back as Nuyts(1992). Anstey (2002) is a proposal for the internal structure of the conceptualcomponent within FDG.

4. See the articles by Butler, Connolly, Rijkhoff and Garc in this vol-ume.

5. In Hengeveld and Mackenzie (2006) the Illocutionary frame restricts a vari-able gnate a Speech Occurrence.

6. However, as will be shown in 2.1, FDG introduces two more variablesctively.

7. As a matter of fact, Connolly (1991) renamed functional patterns as syntactictemplates

8. Dik (1989: 348) defines P1 as initial position(...), used for special purposes, including the placement of constituents withTopic or Focus function

9. In FDG or only includes the predicate and itsmodifiers, so it should not be confused with the VP in Chomskyan-type gen-erative grammars (this is the way it is used in Escribano . Thus,the predicate phrase in only consists of one element, .Section 2 discusses the way is used in this book.

10. For this reason Dik (1997a: 130) distinguished between two kinds of r-ring and . Notice that in F(D)Gwe cannot really speak of n-tactic level.

11. In its logical sense, a (Latin ) is an atomic element without astructure of its own (Vendler 1971: 116; Lyons 1977: 148). Even though inFG the notion n-serted into an argument slot of a predicate or into a adjunct (satellite) position,it was added that -order spatialentities 1997b: 223). In a similar vein, Hengeveld (this volume) charac-terizes i-nal head that denotes a concrete object and is used referentially.

12. It is shown in Rijkhoff (forthcoming b) that the formal properties of this pos-sessive construction in Dutch depend on the function it has as a modifier andvary systematically along certain parameters (Attribution, Predication, Refer-ence).

35

13. See also Dik (1997b: 92) who uses the nameare used to talk about spatial entities.

14. Mackenzie (1987) suggested omitting the x-variable after the colon in an at-tempt to get away from the logical-semantic nature of the representation andmake it more pragmatically adequate (see also Dik 1989: 262; Dik 1997a: 63;Butler 2003: 278-279).

15. The x-variable gets a similar interpretation in FDG: the set of possible first-order referents.

16. Mackenzie (1992b) has argued that certain prepositions are better categorizedas predicates (rather than the formal expression of some function).

17. For the distinction between first-order entities (spatial objects), second-orderentities (events, states, and other temporal objects or states of affairs), third-order entities (i.e. propositions), etc. see Lyons (1977).

18. Note that this representation does not contain an Dik (1997a: 63)justifies its absence in representations for the sake of simplicity.

19. These are simple examples. In practice matters can be rather more interesting.For example, when a second order modifier such as is used in combina-tion with a first order entity such as , as in , the adjectiveforces us to interpret the book as a temporal entity, as it refers to the momentthe book was published (i.e. an event).

20. See Mackenzie (1992a, 1998b, 2001) on variables for places, times and man-ners, Olbertz (1998) on times and Hengeveld and Wanders (2007) on quanti-ties. Some important contributions concerning variables for entities in FG canbe found in a collection of and Mackenzie(2005), who also put the articles in a wider historical perspective.

21. Note that the ysis ofthe Interpersonal level in FDG with the variables for the Discourse Act (A)and Speech Occurrence (F), as shown in 1.1.1.

22. For a critical discussion of this proposal, see Keizer (this volume). On vari-ables in the NP schemas, see also Rijkhoff (1992: 190-191). Escribano (thisvolume) is a critical treatment of variables in F(D)G from a strictly logicalperspective.

23. The name was first used in Mackenzie (1983). Interestingly, inFDG a development in the reverse direction has taken place: the clausal ad-juncts that were called

24. On see Rijkhoff (2002: ch. 2).25. In Rijkhoff's version, modifiers at the Interpersonal Level specify

rather than properties (as in FDG).26. The relationship between discourse-referential modifiers in the clause and the

NP is discussed in more detail in Rijkhoff and Seibt (2005); see also Rijkhoff(this volume).

36

27. The phrase is potentially ambiguous, meaning eitheran unidentified boy -classifying) or of shirt y-ing); cf. Taylor (1996: 665) and Willemse (2005).

Anstey, Matthew P.2002 Layers and operators revisited.

77.2004 Functional Grammar from its inception. In: J. Lachlan Mackenzie

and Mar -Gonz, 23 71. (Functional Grammar Series

24.) Berlin/New York: Mouton de Gruyter.Anstey, Matthew P. and J. Lachlan Mackenzie (eds.)2005 (Functional Grammar

Series 26). Berlin/New York: Mouton de Gruyter.Bolkestein, A. Machtelt1998 What to do with Topic and Focus? Evaluating pragmatic information.

In: Mike Hannay and A. Machtelt Bolkestein (eds.),, 193 214. Amsterdam: Benjamins.

B1999 Reprint. .

Stuttgart: UTB. Jena: Fischer, 1934.Butler, Christopher S.1991 Standards of Adequacy in Functional Grammar.

27: 499-515.1999 Nuevas perspectivas de la Gram ncional: los est

adecuaciJavier Mart

219-256. Barcelona: Ariel.2003

. Amsterdam: Benja-mins.

this vol. Interpersonal meaning in the noun phrase.Connolly, John H.1991 . Berlin: Foris.2004 The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and Marmez-Gonz eds.), ,

37

211 242. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

this vol. Freestanding noun phrases within documents: A pragmatic approachbased on FDG.

Connolly, John H., Roel M. Vismans, Christopher S. Butler and Richard A. Gat-ward (eds.)

1997 Berlin and NewYork: Mouton de Gruyter.

Dik, Simon C.1978 (Publications in Language Sciences 7).

Dordrecht: Foris.1980 . London: Academic Press.1989

(Functional Grammar Series 9). Dordrecht: Foris.1997a

(Functional Grammar Series 20). Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

1997b(Functional Grammar Series 21). Second, revised edi-

tion. Edited by Kees Hengeveld. Berlin/New York: Mouton deGruyter.

Escribano, Josthis vol. On noun phrase structure in F(D)G: Some conceptual issues.

Garcthis vol. Functional Discourse Grammar and extraction from (complex) noun

phrases.G1996 La organizaci n-

cional. In: M. Caneda and J. P, 42-68. Vigo: Universidade

de Vigo.Groot, Casper de2005 Morphosyntactic templates. In: Casper de Groot and Kees Hengeveld

(eds.), , 135-161. (Functional Grammar Series 27.) Berlin/New York: Mouton deGruyter.

Hannay, Mike and A. Machtelt Bolkestein (eds.)1998 . Amsterdam: Benja-

mins.Hannay, Mike and Caroline Kroon2005 Acts and the relationship between discourse and grammar.

12-1: 87 124.

38

Halliday, Michael Alexander Kirkwood2004 (third edition, revised by

Christian M.I.M. Matthiessen). London: Arnold. First published1985.

Hengeveld, Kees1987a The Spanish mood system.

22.1987b Clause structure and modality. In: Johan van der Auwera and Louis

Goossens (eds.), (Functional Gram-mar Series 6), 53 66. Dordrecht: Foris.

1988 Illocution, mood and modality in a functional grammar of Spanish.6: 227 269.

1989 Layers and operators. 25: 127-157. [also in:Anstey and Mackenzie (eds.), 1 46]

1990 The hierarchical structure of utterances. In: Jan Nuyts, A. MachteltBolkestein and Co Vet (eds.),

, 1 23. Amsterdam: Benjamins.1992 . Berlin/New

York: Mouton de Gruyter.1997 Cohesion in Functional Grammar. In: John H. Connolly, Roel M.

Vismans, Christopher S. Butler and Richard A. Gatward (eds.),1 16. (Functional

Grammar Series 18.) Berlin/New York: Mouton de Gruyter.2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzie and Mar G -Gonz lez (eds.),, 1 21. (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzie and Mar

mez-Gonz ,365 378. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

2005 Dynamic expression in Functional Discourse Grammar. In: Casper deGroot and Kees Hengeveld (eds.),

, 53-86. (Functional Grammar Series 27.) Ber-lin/New York: Mouton de Gruyter.

this vol. Prototypical and non-prototypical noun phrases in FDG.Hengeveld, Kees and J. Lachlan Mackenzie2005 Interpersonal functions, representational categories, and syntactic

templates in Functional Discourse Grammar. In: Mar -Gonz

, 9 27. Bern: Peter Lang.

39

2006 Functional Discourse Grammar. In: Keith Brown (ed.),2nd Edition, Vol. 4, 668 676. Oxford:

Elsevier.fc. . Oxford: Oxford University Press.

Hengeveld, Kees and Gerry Wanders2007 Adverbial conjunctions in Functional Discourse Grammar. In: Mike

Hannay and Gerard Steen (eds.),, 211-227. Am-

sterdam/Philadelphia PA: Benjamins.Keizer, Evelienthis vol. Reference and ascription in FDG: an inventory of problems and some

possible solutions.Kroon, Caroline1997 Discourse markers, discourse structure and Functional Grammar. In:

John H. Connolly, Roel M. Vismans, Christopher S. Butler andRichard A. Gatward (eds.),

17 32. (Functional Grammar Series 18.) Berlin/NewYork: Mouton de Gruyter.

Lefebvre, Claire1998 Multifunctionality and variation among grammars: the case of the

determiner in Haitian and in Fongbe.13-1: 93 150.

Lefebvre, Claire and Anne-Marie Brousseau2002 . (Mouton Grammar Library 25.) Berlin/New

York: Mouton de Gruyter.Levelt, Willem J.M.1989 . Cambridge, Mass.: The

MIT Press.Lyons, John1977 (II vols.). Cambridge: Cambridge University Press.

Mackenzie, J. Lachlan1983 Nominal predicates in a Functional Grammar of English. In: Simon

C. Dik (ed.), (Publications in theLanguage Sciences 11), 31-51. Dordrecht: Foris.

1987 The representation of nominal predicates in the fund.25.

1992a Places and things. In: Michael Fortescue, Peter Harder and LarsKristoffersen (eds.),

, 253 276. Amsterdam: Benjamins.1992b English spatial prepositions in Functional Grammar.

46.

40

1998a The basis of syntax in the holophrase. In: Mike Hannay and A.Machtelt Bolkestein (eds.),

, 267 296. Amsterdam: Benjamins.1998b On referring to manners. In: Johan van der Auwera, Frank Durieux

and Ludo Lejeune (eds.),, 245-251. Munich: Lincom Europa.

2001 Adverbs and adpositions: The Cinderella categories of FunctionalGrammar. 41: 119 135.

2004 Functional Discourse Grammar and language production. In: J. Lach-lan Mackenzie and Mar -Gonz

, 179 195. (FunctionalGrammar Series 24.) Berlin/New York: Mouton de Gruyter.

Mackenzie, J. Lachlan and G -Gonz2004 (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2005 (Linguistic Insights 26.)

Bern: Peter Lang.Nuyts, Jan1992 . Amsterdam:

Benjamins.2004 Remarks on layering in a cognitive-functional language production

model. In: J. Lachlan Mackenzie and Mar eles G -Gonz , 275298. (Functional Grammar Series 24.) Berlin/New York: Mouton deGruyter.

Olbertz, Hella1998 (Functional

Grammar Series 22). Berlin/New York: Mouton de Gruyter.Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik1985 . London:

Longman.Rijkhoff, Jan1988 A typology of operators.

29.1990 Toward a unified analysis of terms and predications. In: Jan Nuyts,

A. Machtelt Bolkestein and Co Vet (eds.),, 165 191. Amster-

dam: Benjamins. [also in: Anstey and Mackenzie (eds.), 47 74].1991 Nominal aspect. 8: 291 309.1992 The noun phrase: a typological study of its form and structure. Ph.D.

diss., University of Amsterdam.2001 Dimensions of adnominal modification. In: Martin Haspelmath et al.

(eds.)

41

, 522-533. Berlin/New York: Walter deGruyter.

2002 . Oxford: Oxford University Press [a revised andexpanded paperback edition was published in 2004].

this vol. Layers, levels and contexts in Functional Discourse Grammar.to appear Descriptive and discourse-referential modifiers in a layered model of

the noun phrase. 46-4, 2008.fc. a Synchronic and diachronic evidence for parallels between the layered

structure of NPs and sentences. In: Folke Josephson and IngmarS

. Amsterdam/Philadelphia: John Benjamins.fc. b On the co-variation between form and function of adnominal posses-

sive modifiers in Dutch and English. In: William B. McGregor (ed.),(The Expression of Cognitive Catego-

ries, Volume 2). Berlin/New York: Mouton de Gruyter.Rijkhoff, Jan and Johanna Seibt2005 Mood, definiteness and specificity: A linguistic and a philosophical

account of their similarities and differences.3 2: 85-132.

[http://ojs.statsbiblioteket.dk/index.php/tfs/index]Saeed, John I.2003 , 2nd Edition. Oxford: Blackwell.

Siewierska, Anna1991 . London: Routledge.

Taylor, John R.1996 . Oxford: Oxford University Press.

Van Valin, Robert D., Jr. and Randy J. LaPolla1997 . Cambridge: Cambridge

University Press.Vendler, Zeno1971 Singular Terms. In: D. D. Steinberg, L. and A. Jakobovits (eds.),

, 115 133. Cambridge: Cambridge University Press.Vet, Co1986 A pragmatic approach to Tense in Functional Grammar.

16.1998 The multilayered structure of the utterance. In: Mike Hannay and A.

Machtelt Bolkestein (eds.),, 1 13. Amsterdam: Benjamins. [also in: Anstey and

Mackenzie (eds.), 299 324]

42

Vries, Lourens de2005 Towards a typology of tail-head linkage in Papuan languages.

29-2: 363 384.Willemse, Peter2005 Nominal reference-point constructions: possessives and esphoric NPs

in English. Ph.D. diss., Catholic University of Leuven.

1

The aim of this paper is to show how the various functions and forms of anoun phrase can be handled in Functional Discourse Grammar (FDG). Inorder to do so, I will take what may be called the standard, prototypicalnoun phrase as my point of departure. The standard noun phrase (i) has anominal head; (ii) denotes a concrete, first-order entity; (iii) denotes bylexical means; and (iv) is used referentially.2 This type of noun phrase isdiscussed in section 3, after a brief introduction to FDG in section 2. Sec-tions 4 7 then study noun phrases which lack one of the properties of stan-dard noun phrases, in the order in which these properties are listed above.The conclusion will be that the separation between the interpersonal, therepresentational, and the morphosyntactic levels of analysis in FDG allowsfor a systematic treatment of standard and non-standard noun phrases.

3

2.1. Introduction

Figure 1 gives a general overview of the FDG model. A summary of thevarious properties of this model may be found in Hengeveld (2005) andHengeveld and Mackenzie (2006); a full presentation of the model is givenin Hengeveld and Mackenzie (fc.).At the interpersonal level the hierarchical structure given in (1) applies:

(1) (M1: [(A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))

The hierarchically highest unit of interpersonal analysis given here is themove (M), which may contain one or more discourse acts (A). A discourseact is organized on the basis of an illocutionary frame (ILL), which has two

44

speech act participants (P, the speaker S and the addressee A) and thecommunicated content C evoked by the speaker as its arguments. Thecommunicated content, in turn, contains a varying number of ascriptive (T)and referential (R) Subacts. Note that the latter two units are operative atthe same layer, i.e. there is no hierarchical relation between them. In gen-eral, then, at the interpersonal level units are analysed in terms of theircommunicative function.

Outline of FDG

FramesLexemesPrimary Operators

(M1: [(A1: [ILL (P1)S (P2)A (C1: [(T1) (R1)](C1))] (A1))] (M1))(Interpersonal Level)

/ xxx#XXX#xxx#XXX \ /(Phonological Level)

[[[lexemeAdj]AdjP lexemeN]NP [lexemeV[lexemeAdv]AdvP]VP]CL

(Morphosyntactic Level)

(ep1: [(p1: [(e1: [(f1) (x1)] (e1))] (p1))] (ep1)(Representational Level)

TemplatesAuxiliariesSecondary Operators

Formulation

Morphosyntactic Encoding

Phonological EncodingProsodic PatternsMorphemesSecondary Operators

45

At the representational level the layers presented in (2) are relevant:

(2) (ep1: [(p1: [(e1: [(f1) (x1)] (e1))] (p1))] (ep1))

At this level of analysis linguistic units are described in terms of the entitytype they denote. These entity types are of different orders: third-orderentities or propositional contents (p); second-order entities or states ofaffairs (e); first-order entities or individuals (x); and zero-order entities orproperties (f). Propositions may furthermore be joined into episodes (ep).Note that first-order and zero-order entities belong to the same layer, i.e.there is no hierarchical relation between them.At the structural level, constituent structure representations of clauses,

phrases and words are given, such as for instance in (3):

(3) [[[lexemeA]AP lexemeN]NP [lexemeV [lexemeAdv]AdvP]VP]CLe.g. A manN dancedV badlyAdv .

At this level underlying units become more language-specific, but the as-sumption is that differences between languages can be described system-atically along typological parameters.An important property of the model is that the interpersonal, representa-

tional, and morphosyntactic levels of linguistic organization are built upusing different sets of primitives. The interpersonal and representationallevels of organization are structured on the basis of pragmatic and semanticframes, into which lexemes and primary operators (i.e. operators that aredefined in terms of their meaning) are inserted. The morphosyntactic levelis organized in terms of structural templates, into which, apart from lexicalmaterial from the preceding levels, grammatical words and morphosyntac-tic secondary operators (i.e. operators anticipating bound grammatical ex-pressions) are inserted.Finally, it is important to note that levels are related to each other

through operations, represented in ovals Figure 1. There is a fundamentaldistinction between on the one hand, and on theother. The process of formulation is concerned with specifying those prag-matic and semantic configurations that are encoded within the language. Interms of formulation, languages may differ in e.g. the kind of pragmaticand semantic functions that are relevant for a description of their gram-matical system, irrespective of whether these functions are encodedthrough syntax, morphology, etc. The process of encoding is concernedwith the morphosyntactic and phonological form pragmatic/semantic con-

46

figurations take in the language. In terms of encoding, languages may dif-fer in e.g. their word order, morphological types, phoneme inventory, etc.

3.1. Introduction

An example of the standard noun phrase is given in (4):

(4)

The noun phrase is (i) headed by a noun ( ); (ii) de-notes a first-order entity (the concrete object llexical means, i.e. uses lexical items ( , ) to build up a pictureof the concept transmitted; and (iv) is used by the speaker to refer to thefirst order-entity denoted.

3.2. The standard case in FDG

In FDG a noun phrase such as the one in (4) may be represented as in (5):

(5) ( id RI: TI TJ (RI)) (IL)

(1 xi: (fi: girlN (fi)) (xi) : (fj: intelligentA (fj))(xi) ) (RL)

[[theart] [[intelligentA]AP girlN- NP1]NP2 (ML)

The referential use of the noun phrase is represented at the InterpersonalLevel (IL), where RI indicates that the noun phrase instantiates a referentialsubact. This referential subact contains two instantiations of ascriptivesubacts (TI and TJ). The denotation of the noun phrase is dealt with at theRepresentational Level (RL). Here xi indicates that the noun phrase de-notes a first-order entity. This first-order entity has the lexically expressedproperties fi and fj, which shows that denotation is achieved by lexicalmeans. The nominal nature of the noun phrase is indicated at the represen-

47

tational level, too, where the subscript of the lexical item functioning as thehead is r-sonal and representational levels, the morphosyntactic encoder produces anoun phrase at the morphosyntactic level (ML).

3.3. Operators and modifiers

As indicated in section 2, and illustrated in (5), there are positions at theinterpersonal and representational levels for various kinds of primary op-erators, i.e. operators that capture grammatical distinctions in terms of theirmeaning. For every layer within these levels, there is a separate category ofoperators, represented by en-tational level. Similarly, every layer may be modified by a separate cate-gory of modifiers, captured by p-resentational level. Within the maximal structure for noun phrases used asreferential subacts, the operator and modifier positions given in (6) areavailable:

(6) ( R RI: .......................................... (RI):R (RI))

( x xi: (f fi: LexN (fi):

f (fi)) (xi):x (xi))

The various operator positions in (6) capture the following types of mean-ing:

(7)R Identifiability, Specificityx Location, Numberf Shape, Measure

Identifiability and specificity are properties of referential subacts (R), sincethey have to do with the speaker ssessment of the knowledge of thehearer concerning the referent of the noun phrase. Location and numberconcern properties of the entity denoted by the noun phrase as a whole inthe external world, and therefore operate at the highest layer of the repre-

48

sentational level. Shape and measure specify properties of the property (f)expressed by the head noun, rather than of the entity (x) denoted as awhole, and therefore apply at the lowest layer of the representational level.An example of a noun phrase containing expressions of all three operatortypes is given in (8), represented in (9):4

(8) R x fHEAD

DEF THREE PAIR shoe-PL

(9) (+Id RI: ............................. (RI))(3 xi: (paar fi: schoenN (fi)) (xi) )

This example shows how the surface order of the various operator expres-sions within the noun phrase iconically reflects the underlying scope rela-tions, similar to what one tends to find at the clausal level (see Foley andVan Valin 1984; Hengeveld 1989). This relation between noun phrases andclauses has been stressed in work by Rijkhoff, most recently in Rijkhoff(this volume).

The various modifier positions in (6) capture the following types of mean-ing:

(10)R Subjective attitudex Referent modificationf Reference modification

Modifiers of R apply at the interpersonal level and are therefore speaker-bound. They express the attitude of the speaker with respect to the referentof the term. Modifiers at the highest layer of the representational level (x)specify properties of the entity denoted as a whole. And those at the lowestlevel (f) specify subproperties of the property expressed by the head noun,rather than of the entity denoted as a whole. The latter two types of modifi-cation have been called a-

49

tion (1967). The differences between examples(11) (13), all involving the adjective , illustrate these three classes ofmodifier. The examples were all encountered in a free internet search. Fora detailed corpus-based study of these constructions, and a discussion oftheir theoretical implications, see Butler (this volume).

(11) ( R)

(12)( x)

(13) ( f)

In (11) the speaker expresses his or her sympathy for the doctor referred toby means of the use of . In (12) the adjective indicates a property ofthe entity referred to: this entity is a doctor and is poor. In (13), on theother hand, the adjective has a more restricted scope: the entity referred tois poor as a doctor, i.e. it is the doctorhood that is modified by the adjec-tive. The structure in (6) offers three different positions for these threedifferent readings of , as illustrated in (14) (16), where in each casethe layer modified is different in the sense that the scope of the adjectivedecreases from the R-level in (14), through the x-level in (15) to the f-levelin (16):

(14) (RI: ........................ (RI): (fj: poorA (fj)) (RI))(xi: (fi: doctorN (fi)) (xi))

(15) (RI: ........................................................ (RI))(xi: (fi: doctorN (fi)) (xi): (fj: poorA (fj)) (xi))

(16) (RI: ...................................................... (RI))(xi: (fi: doctorN (fi): (fj: poorA (fj)) (fi)) (xi))

The differences between the various uses of in (11) (13) are reflectedin its behaviour in certain grammatical contexts. Examples (17) (19) showhow behaves differently from , which can only be used for refer-ent-modification:

50

(17) a.b.

(18) a.b.

(19) a.b.

Example (17) shows that only an interpersonal adjective can be used inexclamations of the type illustrated; (18) shows that both adjectives may beused at the x-level specifying opposite values, excluding theand the ; and (19) shows that in the context of

... , which forces a low-scope reading upon the adjective, onlyreference modification is possible.5

I will now turn to a type of noun phrase that is different from the standardnoun phrase only in the fact that its head is not a noun. Consider the fol-lowing example from Hupa:

(Na-Dene; Golla 1985: 59)(20)

3SG.POSS-horn-3SG.OBJ-plenty

At first sight it seems that the expression in (20) is not a noun phrase but aclause. However, as shown in (21), the same expression may take a posses-sive prefix, which a clause could never take, thus clearly showing thephrasal nature of the expression:

(Na-Dene; Golla 1985: 59)(21)

1SG.POSS-3SG.POSS-horn-3SG.OBJ-plentyy cow plenty on it

One way of interpreting this construction is that a concrete entity is charac-terized in terms of a state-of-affairs in which it is typically involved. In

51

other words: a first-order entity is characterized in terms of a second-orderentity. Example (20) may be represented as in (22):

(22) (xi: (ei: [(fi: le:n (fi)) (xj: de j): (xi: mi (xi))Poss (xj)) (xi: xo (xi))Ref](ei)) (xi) )

This analysis is similar to the one given by Dik (1997b: 92) for internallyheaded relative clauses, and indeed, as (22) shows, the variable (xi) of theterm as a whole is filled in with another coreferential term (xi: xo (xi))within the predication (ei) that occupies the head position of the term as awhole. This predication is thus a closed predication in FG terms.The possibility of a term containing an open predication as its head may

now also be considered. As proposed by Van der Auwera (1990: 151ff.),cases like (23) are instantiations of this situation:

(23)

The headless relative (in italics) in (23) can be represented as in (24):

(24) (xi: (Pres ei: [(fi: read (fi)) (xj: you (xj))Ag (xi)Pat] (ei)))

Note that here the variable (xi) within the embedded predication is notfilled with any lexical material, but bound by the variable of the term as awhole. In this sense it is different from the Hupa noun phrase in (21),which contains a closed rather than an open predication.Taking this analysis one step further, productive nominalizations like

the one in (25) may receive a similar treatment:

(25)INDEF teach-AG.NR

This nominalization might be represented as in (26):

(26) (1 xi: (ei: [(fi: teach-V (fi)) (xi)Ag] (ei)))

The fact that this configuration is expressed as a noun would then be takencare of by the morphosyntactic encoder. An advantage of this approach,over the one generally advocated in FG, is that no predicate formation ruleis needed, and derivational expression is dealt with in the same way asinflectional expression. Of course, such a syntactic approach to deriva-

52

tional morphology is only possible to the extent that the derivation in-volved is productive and regular. As soon as a derivation becomes lexicallyspecialized it would have to be entered into the lexicon as a separate lex-eme.

Another way in which noun phrases may deviate from the standard case isby denoting a non-first order entity. Cases in point are listed in (27):

(27)x individualf propertye state-of-affairsp propositional contentl locationt time

As demonstrated in Hengeveld and Mackenzie (fc.), semantic classes suchas the ones listed in (27) are needed to account for differences in thegrammatical behaviour of classes of nouns and noun phrases crosslinguis-tically. A case in point is nominalization in English, where different deriva-tional processes produce lexemes denoting properties of different kinds ofentities. Consider the nouns in (28):

(28)x , ,fe ,p , ,lt

Note that the resulting meanings of the various processes can be classifiedin terms of the semantic categories in (27). English does not have a produc-tive derivational process for time expressions, but some other languagesdo, and this then produces meanings that are expressed in English by com-pounds such as or . Consider the following exam-ple from Supyire:

53

(Carlson 1994: 113)(29)

his TNR-die-DEF.CL NEG PRF be.long.time NEG

FDG uses different types of variables at the representational level to ac-count for grammatically relevant semantic classes denoted by nounphrases. Thus, the examples listed in (27) would be represented as in (30):

(30) (xi: (fi: chairN (fi)) (xi))(fi: (fi: colourN (fi)) (fi))(ei: (fi: meetingN (fi)) (ei))(pi: (fi: ideaN (fi)) (pi))(li: (fi: gardenN (fi)) (li))(ti: (fi: weekN (fi)) (ti))

6.1. Two cases of lexical non-denotation

A third way in which noun phrases may differ from the standard case iswhen they denote, but not by lexical means, or when they do not denote atall. These cases are discussed in turn in the following sections.

6.2. Proper names and pronouns

Proper nouns and pronouns can be said to have no meaning of their own, inthe sense that they do not designate properties of entities in the externalworld, but have a conventionalized referential use6 only. In FDG thesewords are therefore interpreted as direct and unique instantiations of refer-ential subacts. For this reason, they are represented at the interpersonallevel, as restrictors of referential subacts, rather than at the representationallevel, as restrictors of entity descriptions.7 The sentence in (31) can there-fore be represented as in (32):

(31)

54

(32) (TI ) (RI: (RI)) (RJ: (RJ))(ei: [ (fi: seeV (fi)) (xi)Exp (xj)Pat ] (ei))

Given their interpersonal status, proper names and pronouns may be ex-pected to be modifiable by R operators and R modifiers only, but not byoperators from the representational level. Consider examples (33) and (34)and their representations in (35) and (36):

(33)(34)

(35) (RI: you (RI): (fj: poorA (fj)) (RI))(36) (RJ: John (RJ): (fj: poorA (fj)) (RJ))

in (33) and (34) can only be interpreted as an expression of a subjec-tive attitude of the speaker with respect to the referent of the term, andnever as an instance of referent modification or reference modification (seesection 3.3.3).

6.3. Vocatives

In (35) (36) the pronoun and proper name have a representational counter-part, even though it is one that is not lexically filled. When these sameelements are used as vocatives, they have no representational counterpart atall. They do not denote an entity in the external world, but have an inter-personal function only, bound to the speech situation itself. The expres-sions in (37) (38) can be formalized as in (39) (40) (see Hengeveld andMackenzie fc.: ch.4):

(37)(38)

(39) (M1: [(A1: [(FI: VOC (FI)) (P1)S (P2: John (P2))A] (AI))] (MI))(40) (M1: [(A1: [(FI: hey (FI)) (P1)S (P2: you (P2))A] (AI))] (MI))

In some languages the different uses of proper names are reflected in theirgrammatical behaviour. Thus, in Portuguese, proper names carry a definite

55

article when used referentially, but appear without an article in vocativefunction, as in the following examples:

(41)see.PAST.PRF.1SG DEF.SG Jo

(42)VOC Jo DEF.SG what COP-2SG.PRES do-PROGR

The fact that the vocative use of pronouns and proper names is differentfrom their referential use is furthermore reflected in the fact that subjectivemodifiers such as in (33) and (34) cannot be used in the vocative con-struction.

(43)see.PAST.PRF.1SG DEF.SG poor Jo

(44) *VOC poor Jo DEF.SG what COP-2SG.PRES do-PROGR

are you doing?

6.4. (Non-)denotation: summary

In sum, Table 1 represents the situations with regard to the denotation ornon-denotation of noun phrases. Each situation may be interpreted as acombination of features obtaining at the interpersonal and representationallevels of representation.

. (Non-)denotation

Interpersonal RepresentationalStandard case (R: (x: Lexeme (x))Proper names (R: Lexeme (R)) (x:Vocatives (P: Lexeme (P))

56

7.1. Two cases of non-referentiality

A fourth way in which noun phrases may differ from the standard case iswhen they are non-referential. Here again there are two different cases thatmay be distinguished (cf. Rijkhoff 2002: 56 57): One in which the nounphrase has no interpersonal function at all, and one in which it has an as-criptive rather than a referential interpersonal function. These cases arediscussed one by one in the following sections.

7.2. Noun incorporation

Smit (2005) distinguishes three cases of noun incorporation, differentiatingthe three types according to the layered underlying structure of the incor-porated unit. These three cases are listed in (45):

(45)f-incorporation: (fi: LexN (fi))x-incorporation: (xi: (fi: LexN (fi)) (xi))R-incorporation: (RI: .................... (RI))

(xi: (fi: LexN (fi)) (xi))

In the first case what is incorporated is a noun, not a noun phrase. In theother two cases what is incorporated is a noun phrase. The phrasal natureof these two cases can be demonstrated by the fact that the incorporatednoun allows an (external) modifier. The difference between the two phrasalcases has to do with the referentiality of the incorporated unit: if it is a caseof x-incorporation it is impossible to refer back to the incorporated unit; ifit is a case of R-incorporation anaphoric reference is possible. These vari-ous properties are listed in Table 2.

. Noun incorporation

Modification Referencef-incorporation - -x-incorporation + -R-incorporation + +

57

R-incorporation is simply a matter of a different type of morphosyntacticexpression of a standard noun phrase. The case of x-incorporation, how-ever, cannot be interpreted as a standard case, since the noun phrase is notused referentially. Consider the following example from Caddo, discussedin Mithun (1984: 864 866):

(Caddoan; Mithun 1984: 866)(46) .

a.lot PROGR-grass-grow-PROGR

The incorporated noun ifieroration. At

the same time, it is not a referential phrase. As Mithun (1984: 866) notes:orated nouns, KH] in

these [...] constructions is often deducible from context, the IN m-selves are not, strictly speaking, referential. An extensive examination oftexts shows that they are not used to establish discourse referents as inde-pendent N atively rare cases where enti-ties first appear in discourse as IN equent mention of them regu-larly includes a restatement of the N, either incorporated or independent.In view of this non-referential nature of the noun phrase, the head of

which is incorporated, it may be represented in FDG as in (47):

(TI)(47) (ei: [ (fi: - - (fi)) (xi: k'uht (xi): wayah (xi))] (ei))

This representation indicates that at the representational level there is a fullphrasal description of a first order entity, but that this semantic unit has nointerpersonal counterpart.

7.3. Ascriptive noun phrases

A second case in which noun phrases are used non-referentially is whenthey have an ascriptive interpersonal function. A case in point is (48):

(48)

This sentence may be represented as in (49):

58

(TI) (RI)(49) (ei: [ (1 xi: (fi: criminal (fi)) (xi)) (prox xj: (fj: man (fj)) (xj)) ] (ei))

The formalization in (49) indicates that the first-order entity description xidoes not instantiate a referential subact (R) but an ascriptive subact (T).8

Once the distinction between the ascriptive and referential use of nounphrases is made and can be formalized, some differences in their behaviourcan be accounted for. One example concerns anaphoric reference. Considerthe following examples:

(50)

(51)

Anaphoric reference to a noun phrase used ascriptively requires the use ofor as in (50), while the use of a personal pronoun is required in the

case of referential use.Once the above analysis is accepted, identificational constructions have

to receive an analysis different from the one advocated in Dik (1980: chap-ter 4) and Hengeveld (1992). Keizer (1992, this volume) already noticedproblems with this analysis. Consider the following example:

(52)

If in (52) is taken as the predicate, as it is in Dik (1980) andHengeveld (1992), then one would expect (53) to be grammatical, which itisn

(53)

Rather than being an ascriptive subact, should thus be taken as theinstantiation of a referential subact, and (52) should be represented as in(54), in consonance with Keizer (1992):

(RI: [he] (RI)) (RJ: Peter (RJ))(54) (Pres ei: [ (xi) (xi) ] (ei))

59

The copula is then introduced at the morphosyntactic level, to accommo-date the expression of the tense operator in the absence of a verbal element.

7.4. (Non-)referentiality: summary

The cases of (non-)referentiality discussed in this section may now besummarized as in Table 3.

. (Non-)referentiality

Interpersonal RepresentationalStandard case R x etc.Predicate nominals T x etc.Incorporation x etc.

In the previous sections, the various types of noun phrases have been dis-tinguished from one another in the way summarized in Table 4.

. Summary

Head Order Representational Interpersonal

Standard noun phrase ( N 1 (x: lex (x)) (R:

Non-nominal noun phrase ( non-N 1 (x: ([..lex..]) (x)) (R:

Non-first order noun phrases ( N non-1 (e: lex (e)) etc. (R:

Proper names and pronouns ( N 1 (x: (R: lex (R))

Vocatives ( N 1 (P: lex (P))

x-incorporated noun phrases ( N 1 (x: lex (x))

Ascriptive noun phrases ( N 1 (x: lex (x)) (T:

Table 4 shows, among other things, that standard noun phrases ( -nominal noun phrases ( -first order noun phrases ( nnot bedistinguished from one another at the interpersonal level, while they aredistinct at the representational level. Conversely, standard noun phrases( -incorporated noun phrases (( e another at the representationallevel, while they are distinct at the interpersonal level. Thus, by combining

60

the interpersonal and representational levels, unique configurations arisethat are sufficient to trigger the encoding devices that are needed to ac-count for the morphosyntactic differences between these constructiontypes.

1. This article is to a large extent inspired by joint work with Lachlan Mackenzieon Functional Discourse Grammar (Hengeveld and Mackenzie fc.). I am grate-ful to the editors of this volume and to an anonymous referee for comments onan earlier version of this paper.

2. I use the term ssionand the entities external to the language system to which that expression ap-plies (Lyons 1977: 207f). The termfor the use a speaker makes of a linguistic expression to identify an entity foran addressee. As Lyons (1977: 177) states, it is the speaker who refers (byusing some appropriate expression): he invests the expression with referenceby the act of referring. (2002: 228)when he talks about the descriptive versus referential function of nounphrases.

3. This section is largely based on Hengeveld and Mackenzie (2005).4. The fact that does not take a plural ending shows that it is a grammatical

expression rather than a head noun.5. Note that this subdivision of adjectives solves the problem mentioned by Es-

cribano (this volume) that not all modifiers are intersective. In the classifica-tion used here only x-modifiers, as illustrated in (15), are intersective.

6. This is probably the reason why they were called1997a: 61).

7. See also Coates (2006), who provides a series of arguments for the pragmaticnature of referring by means of proper names. See also Keizer (this volume)for an alternative view, in which proper names are treated as restrictors at therepresentational level.

8. Keizer (this volume) provides the noun phrase used ascriptively with an addi-tional f-variable at the representational level. I see no need for this additionalvariable, as the property-assigning nature of such noun phrases is sufficientlycaptured by the fact they are used as ascriptive subacts.

61

Bolinger, Dwight1967 Adjectives in English: Attribution and predication. 18: 1 34.

Butler, Christopher S.this vol. Interpersonal meaning in the noun phrase.

Carlson, Robert1994 (Mouton Grammar Library 14). Berlin/New

York: Mouton de Gruyter.Coates, Richard2006 Properhood. 82-2: 356 382.

Dik, Simon C.1980 . London: Academic Press.1997a

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

1997b(Functional Grammar Series 21.) Edited by Kees

Hengeveld. Berlin/New York: Mouton de Gruyter.Escribano, Josthis vol. On noun phrase structure in F(D)G: Some conceptual issues.

Foley, William A. and Robert D. Van Valin, Jr.1984 . Cambridge: Cambridge

University Press.Golla, Victor1985 . Hoopa Vally: Hupa Language

Program.Hengeveld, Kees1989 Layers and operators in Functional Grammar.

25-1: 127 157.1992 (Functional

Grammar Series 15). Berlin/New York: Mouton de Gruyter.2005 Dynamic expression in Functional Discourse Grammar. In: Casper de

Groot and Kees Hengeveld (eds.)53 86. (Functional Grammar Series 27.) Ber-

lin/New York: Mouton de Gruyter.Hengeveld, Kees and J. Lachlan Mackenzie2005 Interpersonal functions, representational categories, and syntactic

templates in Functional Discourse Grammar. In: Mar de losG -Gonz

, 9 27. (Linguistic Insights 26.)Berne: Peter Lang.

62

2006 Functional Discourse Grammar. In: Keith Brown (ed.),2nd Edition, Vol. 4, 668 676. Oxford:

Elsevier.fc. . Oxford: Oxford University Press.

Keizer, M. Evelien1992 Reference, predication and (in)definiteness in Functional Grammar. A

functional approach to English copular sentences. Ph.D. diss., VrijeUniversiteit Amsterdam.

this vol. Reference and ascription in FDG: an inventory of problems and somepossible solutions.

Lyons, John1977 (II vols.). Cambridge: Cambridge University Press.

Mithun, Marianne1984 The evolution of noun incorporation. 60: 847 894.

Rijkhoff, Jan2002 (Oxford studies in typology and linguistic theory).

Oxford: Oxford University Press.this vol. Layers, levels, and contexts in Functional Discourse Grammar.

Smit, Niels2005 Noun incorporation in Functional Discourse Grammar. In: Casper de

Groot and Kees Hengeveld (eds.),Grammar, 87 134. (Functional Grammar Series 27.) Ber-

lin/New York: Mouton de Gruyter.Van der Auwera, Johan1990 . Thesis, Universitaire Instelling Antwerpen.

1

This contribution has three main goals. Firstly, so as to be able to accom-modate classifying modifiers, I will propose a revised model of the nounphrase, which consists of five concentric layers of modification rather thanfour, as it did in the previous model (section 2).2 It will also be claimed thatdiscourse-referential modifiers, which are specified at the InterpersonalLevel, only relate to the status of the referent in the world of discourse(Rijkhoff fc. a; Rijkhoff and Seibt 2005). That is to say, attitudinal modifi-ers of the kind proposed in Hengeveld (2004b) (see also Hengeveld thisvolume; Butler this volume) are not deemed to have their own slot in thelayered model of the noun phrase defended here.Secondly, I will propose some modifications regarding the contextual

component with special attention to noun phrases and argue that externalreality (the aratecomponent and that each component of the FDG model constitutes a dif-ferent kind of context (section 3). A single rule will be proposed to capturethe influence of any contextual factor on the form, function or meaning ofa linguistic expression.Section 4, finally, argues that in the current FDG model the Interper-

sonal Level in the grammatical component contains elements that actuallybelong to the external component (or -context oposes alternativeschemas for the NP and the clause without variables for Speaker and Ad-dressee. In this proposal the descriptive modifiers (i.e. classifying, qualify-ing, quantifying, and localizing modifiers) are specified at the Representa-tional Level ( ntentregarded as that part of the grammatical component that is concerned with

ation,one for things or events, one for propositions, and one for clauses.

64

In Rijkhoff (2002) the noun phrase (NP) is analysed as having four layersof modification, accommodating (i) qualifying, (ii) quantifying, (iii) local-izing, and (iv) discourse modifiers (or discourse-referential modifiers). Itappears, however, that a complete analysis of the NP requires an addi-tional, fifth layer of modification for what might be called

of the noun (as announced in Rijkhoff 2004, fc. a). Moreover, itwill be argued that discourse-referential modifiers, which are specified atthe Interpersonal Level in the grammatical component, are only concernedwith the status of the referent (thing, event) as a discourse entity and not,as Hengeveld (2004a, 2004b, this volume) has claimed, also with emo-tional or attitudinal phenomena.This section is structured as follows. After a brief presentation of the

-layered model of the noun phrase (section 2.1), I will argue forthe existence of an additional, fifth category of so-called

, which (further) specify is being referred to by thespeaker (sections 2.2 and 2.3).

2.1. The four-layered analysis of the NP

In Rijkhoff (2002) modifiers in the NP are divided into two main catego-ries: descriptive and discourse-referential modifiers (or representationaland interpersonal modifiers respectively). This division reflects the doublefunction of NPs: they provide a physical description of an entity and at thesame time they are the constructions used to refer to entities in the world ofdiscourse. Descriptive modifiers specify properties of the referent of theNP or clause in terms of the notions Quality (how it is), Quantity (howmuch/many it is) and Location (where it is). The fourth and outermostlayer of modification accommodates discourse-referential modifiers, whichpertain to IT IS

events) in the shared world of discourse (e.g. Definite d-nominal adjectives, possessives and relative clauses are instances of lexicalNP modifiers or lized by in Figure 1), because theyinvolve members of lexical categories (verbs, nouns, adjectives, adverbs).Operator slots (symbolized by in Figure 1) are reserved for grammaticalmodifier categories in the NP such as articles or demonstratives, which canbe expressed as free forms, particles or affixes (for a discussion of

, see Rijkhoff 2002: 100 121). Figure 1 also shows that clauses can

65

be analysed in a similar fashion ( symbolizes operators, stands for satel-lites; see Rijkhoff 2002: chapter 7); discourse-referential modifiers, whichare symbolized by the variables - - - elow(Figure 3).

Location

3 3

Quantity

2 2

Quality

1 1

,

1 1

Quality

2 2

Quantity

3 3

Location

Symmetry in the underlying structure of the clause and the NP as inRijkhoff 2002: descriptive modifiers (but cf. Figure 10 below).

Languages use only a subset of the modifier categories mentioned in Fig-ure 1 and there is no one-to-one relationship between the form and thefunction of a modifier (only some typical forms or constructions are speci-fied in Figure 1). For example, in many languages the adnominal distaldemonstrative (a localizing operator) is also used as a discourse-referential

66

operator to mark definiteness, and relative clauses can be employed asqualifying, quantifying or localizing satellites (Rijkhoff 2002 and fc. a).A simplified version of the layered NP model is represented in (1), in

which each operator ( , , as itsargument (cf. Escribano this volume). Note that I am here only concernedwith the analysis of NPs used for first-order or spatial entities (symbolizedby the x-variable) and that the schema in (1) does not contain the predicatevariable f (see Figure 3 and further below on discourse-referential modifi-ers 3

(1) NPi: 4[ 3[ 2[ 1[ NOUN(xi) ]L0 1(L0) ]L1 2(L1) ]L2 3(L2) ]L3 4(L3)

x referent variable (symbolizes the referent of the NP);term (NP) operator: 1 = qualifying operator, 2 = quantifying op-erator, 3 = localizing operator, 4 = discourse-referential operator;

/ term (NP) satellite: 1 = qualifying satellite, 2 = quantifying satel-lite, 3 = localizing satellite, 4 = discourse-referential satellite.

L1 constitutes the , which contains the head of the constructionand accommodates ( 1 in the NP, 1 in the clause) and

( 1 in the NP, 1 in the clause). In the 2002 model these modifiersonly relate to the property that is designated by the head, such as nominalor verbal aspect markers, (typically) adjectives if the language underanalysis has them and adverb(ial)s of manner or speed (but cf. section2.2). The (L2) contains the quality layer and accommodates

( 2, 2; 2, 2), which have to do with number dis-tinctions (singular, plural) or cardinality. The (L3) containsboth the quality layer and the quantity layer and accommodates

( 3, 3; 3, 3), which specify properties concerning the locationof the thing or event, such as demonstratives, tense markers, (localizing)relative clauses or adverb(ial)s such as e-call there is no direct correlation between form and function of modifiers,especially in the case of satellites).

67

THOSE TWO DOG BLACK IN THE GARDEN

1 1

Quality

2 2

Quantity

3 3

Location

. Simplified representation of

Thus, in an NP like only modi-fies the head noun and the quantifying modifier specifies the numberof black dog entities. Both and specify the location ofdog entities with all their qualitative and quantitative properties.In addition to the three descriptive layers shown in Figure 1, the under-

lying NP/clause structure contains a layer to accommodate modifiers thatare concerned with the status of the thing or event as a discourse entity(Figure 3). Here we find values for operators such as Definite or Specific(in the clause: Realis or Actual) and satellites such asthe clause e.g. 4 Discourse-referential (DR) modifiersappear in the outermost layer, as they have the descriptive modifiers intheir scope (see also Figures 14 and 15 on other interpersonal modifiercategories).Parallels between discourse-referential operators in the clause and in the

NP are particularly interesting because of the various relationships thathold between realis/irrealis and definiteness/indefiniteness. I will showbelow that, in order to understand these relationships, we must make cru-cial reference to the grammatical category of specificity (see Rijkhoff andSeibt 2005 for a more detailed discussion).

68

4

( 3-2-0) ( 0-1-2-3)( )

( e.g. ally----------------------------------------------------------------------------------------------------

e.g.

( )( 3-2-0) ( 0-1-2-3)

4 4

. Symmetry in the underlying structure of the clause and the NP:Discourse-Referential modifiers - - -

The grammatical categories Realis and Definite are similar in that theyboth indicate that the entity, i.e. the event or thing referred to by thespeaker, (already) has a certain location (isdiscourse. By contrast, Irrealis and Indefinite signal that the entity referredto by the speaker does have a proper location (is notworld of discourse at least not yet. There are thus two symmetries, onelinking the grammatical categories Realis and Definite, the other linkingIrrealis and Indefinite.

69

Noun Phrase(thing)

Clause(event)

Definite THING OR EVENT (ALREADY) HAS ALOCATION IN THE DISCOURSE WORLD

Realis

Indefinite THING OR EVENT DOES NOT HAVE A

LOCATION IN THE DISCOURSE WORLD (YET)Irrealis

. Symmetry between Definite/Realis and Indefinite/Irrealis

There is also an interesting anti-symmetry between (in)definiteness and(ir)realis (Rijkhoff 1988), which has to do with the number of ways anentity can be definite/indefinite (thing) or actual/non-actual (event). HereDefinite aligns with Irrealis in that both definite things (referents of defi-nite noun phrases) and non-actual events (referents of irrealis clauses) canoccur in the world of discourse for many different reasons. For example,the referent of an NP can be definite because

it has been mentioned before (anaphoric use)

(2)

it is available in the physical context (situational or deictic use)

(3)

it is deemed identifiable (or accessible) by the speaker because of so-called -reference

(4)

it is relatively unique in the given contextual setting, so that the hearercan identify it on the basis of his or her general knowledge.

(5)

70

There are also many reasons why reference is made to a non-actual event;for example, because the speaker expresses a wish or a request:

(6) [interlocutors are not in the same place](7) [the door is not closed]

In both cases the speaker refers to an event which has not occurred (yet)and which therefore cannot be located in a particular spatio-temporal re-gion in the world of discourse, i.e. the referent of a non-actual event is not

5 All languages have linguistic de-vices to mark the various kinds of moods (e.g. epistemic, epistemological,deontic) and speech acts (e.g. indicative, imperative, optative) that signal ina more or less indirect fashion to the addressee whether or not reference isbeing made to an actual event. There are, however, quite a few languagesthat (also) employ a special realis or irrealis marker for this purpose. Onesuch language is the Papuan language Amele.

(IRR = irrealis; Roberts 1990: 371 372)(8)

pig SIM-run.out-3SG.DS. 3PL hit-3PL-

(9)pig SIM-run.out-3SG.DS. 3PL hit-3PL-

(10)pig SIM-run.out-3SG.DS. hit-NEG.F.3PL

By contrast, there is basically one reason why an NP is indefinite: becausethe entity (thing) designated by the indefinite NP has not been properlyintroduced into the world of discourse hence the speaker assumes that theaddressee does not know (yet) what particular thing is being referred to.

(11)

There is also only one reason why a sentence is in the realis mood: becausethe entity (event) designated by the sentence is real, i.e. it has happened (oris happening).

71

(R = realis; Roberts 1990: 371 372)(12)

pig SIM-run.out-3SG.DS. 3PL hit-3PL-

Thus there are many reasons why the speaker may assume the referent of adefinite NP to be recoverable for the hearer (anaphoric reference, deicticreference, etc.), but there is basically only way for an event to be actual:because it (has) occurred or is occurring. And vice versa: while there isbasically one reason why the speaker assumes the referent of an indefiniteNP to be irretrievable (because its referent is unidentifiable / unfamiliar /inaccessible when it is being referred to for the first time), there are nu-merous reasons why the speaker chooses to refer to a non-actual event. Forexample, because the speaker expresses a desire, a wish, a hope, a fear, anintention, a possibility, a probability, a request, or a command.6

Noun Phrase(thing)

Clause(event)

Definite MANY ONE RealisIndefinite ONE MANY Irrealis

. Anti-symmetry between Definite/Realis and Indefinite/Irrealis.

In order to explain symmetrical and anti-symmetrical relations between(ir)realis and (in)definiteness we need to distinguish between specific andnon-specific indefinite reference. Consider the following examples (Kart-tunen 1976):

:(13)

:(14)

72

These examples show that there are two ways to continuedogspecific dog (whereas in (14) the speaker does not refer to any particular dog (Englishhas no special article for specific or non-specific reference unlike, for ex-ample, many Polynesian or sub-Saharan African languages).It will be recalled that realis events and definite things are located in a

certain spatio-temporal region in that world of discourse; in other words,these events and things can be said to be grounded in that world of dis-course. There is a difference, however, in that realis events, which are typi-cally tensed (if the language uses tense marking), ground themselves uponbeing referred to by the speaker, whereas definite things often first need tobe introduced by an indefinite NP before they can be grounded.Referents of irrealis clauses (see examples 8 10) and nonspecific-

indefinite NPs (as in 14) are events and things that may also be said toexist in some fashion once they have been mentioned (since it is possible torefer to them anaphorically), but they are not grounded in the world ofdiscourse: it is impossible to say where or when they are actually occurringor have occurred.By contrast, referents of -indefinite NPs (such as

) are grounded immediately in the world of discourseonce they have been referred to by the speaker (see also 13).7 In this sensereferents of specific-indefinite NPs are like realis events in that theyground themselves (anti-symmetry).8

Notice that non-actuality of an event does not necessarily mean thattense is always absent (in languages that have tense marking). When tenseis used for a non-actual event, it specifies an event was .The following example from Nyigina (Australia) illustrates:

(Stokes 1982: 24)(15)

NEG 3SG-NF.IRR-go-PAST Derby-ALL

Summing up, we can now say that the symmetry between (in)definitenessand (ir)realis is due to the fact that(a) and indicate that the thing or event being referred to bythe speaker is grounded (occupies a certain spatio-temporal region) inthe world of discourse, and

73

(b) and indicate that the thing or event beingreferred to by the speaker is not grounded (does occupy a certainspatio-temporal region) in the world of discourse.

Noun Phrase(thing)

Clause (event)

Definite GROUNDED IN DISCOURSE WORLD RealisNonspecific-indefinite

NOT GROUNDED IN DISCOURSE WORLD Irrealis

. Symmetry between Definite/Realis and Nonspecific-indefinite/Irrealis.

Fongbe (a Kwa language mainly spoken in Benin) is an example of a lan-guage that employs the same marker for realis and definite reference:

(Lefebvre 1998: 94, 99; see also Lefebvre and Brousseau 2002)(16)

I eat crab DET

(in question/that we know of)

John arrive DET

Jacaltec (Mayan) uses the same marker for irrealis and nonspecific-indefinite reference (the variation is due to vowel harmony):

(Craig 1977: 93, see also Martin 1998)(18)

sleep-OJ EXH CLF/he

(19)ASP-ABS.3-start PL woman look_for-FUT

INDEF-OJ poted looking for a pot.

[nonspecific-indefinite reference]

74

Summing up, anti-symmetry between (in)definiteness and (ir)realis is dueto the fact that(a) there is basically one way for an entity to be indefinite (thing) or actual(event); also, referents of and

for the first time.(b) there are numerous ways an entity can be definite (thing) or less thancompletely actual (event).

Noun Phrase(thing)

Clause(event)

Definite MANY ONE

(GROUNDS ITSELF)Realis

-indefinite

ONE

(GROUNDS ITSELF)MANY Irrealis

Anti-symmetry between Definite/Realis and Specific indefinite/Irrealis.

I would like to make clear at the outset that this section is not concernedwith attitudinal modifiers as described in Dik (1997: 295f.), which specifythe speaker -order entities inthe discourse world, such as the propositional contents of his or her ownutterance, as in:

(20)(21)

Here I will mostly be concerned with manifestations of the speaker o-tional attitude such as sympathy or scorn towards animate entities. Forexample, in many languages the diminutive is used to express affection(Jurafsky 1996):

75

(22)I have just with my colleague-DIM talked

Adjectives likespeaker typically human or animateentity:

(23)

In a recent article Hengeveld (2004b) has argued that in (23) is aninstance of a special category of modifiers ( -modifiers e-sented at the Interpersonal Level in the FDG model (see also Butler thisvolume; Keizer 2004: 15).9 There are several reasons why I am reluctant toposit a special category of attitudinal modifiers at the Interpersonal Levelof the NP model presented above (Figure 3). First of all, a layered model ofany linguistic structure is supposed to reflect differences in semantic scopethat exist between the various modifier categories. There is, however, nogood evidence to show that the scope of in the sense ofin 23) differs from the scope of with another sense (moneyattributive modifier , it only has the noun in its scope. This does notimply that all adjectives, or rather all qualifying satellites, are equal. Prop-erties denoted by qualifying satellites range from objective/permanent (e.g.

A table A chair A ap-ple A book A idea A performanceiconically reflected in the way qualifying adjectives are ordered relative tothe head noun: across languages adjectives denoting more inherent (per-manent, objective) properties have a strong tendency to occur closer to thehead noun than adjectives denoting less inherent (temporary, subjective)properties:10

(24)(25) ?

Thus, the observation that a modifier such as in (23)subjective evaluation by the speaker 2004b: 373) is no reasonto introduce a special category of attitudinal satellites at the Interpersonal

76

Level of the NP ( not the same as ctive 11 Wenoted above that many, perhaps most adjectives reflect some kind of sub-jective evaluation by the speaker ( , , , , ,

). In fact, a wide variety of modifier categories can be said toexpress subjective evaluations: verbal aspect markers (Comrie 1976: 17),expressions such as vs. , adverbial modifiers such as

or , forms of address (expressing different degrees of polite-ness) they all involve the speakerspeaker might have said something else under similar circumstances.12

Nevertheless it is true that the NP in (23) has anadditional communicative value in that it displays sympathy for the refer-ent of the phrase on the part of the speaker. However, treating as aninterpersonal modifier in this construction would imply that the surplusattitudinal value is only situated in the adjective, whereas in fact speakerattitudes such as sympathy or scorn tend to manifest themselves in a vari-ety of ways (morpho-syntactically, prosodically, in voice quality) as aproperty of a construction rather than through a distinct modifier categoryrepresented at one particular layer in the underlying structure (Halliday2004: 61, 238, 318 9; Butler this volume). In the following example scornis expressed morpho-syntactically through the use of a distal demonstrative

-je al possessive construction13

(26)I find that new book-DIM of you totally nothing

Leaving out the diminutive and using the proximal form of the demonstra-tive or the prenominal possessive turns the NP into a neutral expression.

(27)I find your new book absolutely nothing

Thus it seems that attitudes of the kind discussed here are global propertiesat the level of a construction rather than a local property that resides in asingle modifier.

77

How can we capture such global expressions of speaker attitude inFDG? Obviously, when the input from the pre-linguistic conceptual mod-ule enters the grammatical component, it must already be clear which partof the structure will be affected by speaker attitude. In FDG one could use,for example, a dotted line to indicate which elements in the underlyingrepresentation are within the Attitude Domain, which will effect the form,content and expression of material in that attitude domain. This kind ofrepresentation is in fact rather similar to the way discourse pragmatics, inparticular Focus Structure, is handled in(Van Valin and LaPolla 1997: 210 218). In RRG a dotted line is used tomark the potential focus domain (that part of the sentence in which focuscan occur) and a triangle indicates what part of the structure is the actualfocus domain (the part of the sentence which is in focus).

2.2. Evidence for the layered NP/clause model

The fact that in languages such as Jacaltec or Fongbe the same morphemeis used as a discourse-referential operator in the NP and in the clause (sec-tion 2.1.3) provides strong support for the NP/clause model defended here.Evidence for the descriptive layers in the NP/clause model in Figure 1comes from parts-of-speech systems, historical linguistics, cognitive lin-guistics and morpho-syntax.As to syntactic evidence, there are 24 ways to linearize a demonstrative,

a numeral, an adjective and a noun in a single integral noun phrase, butmost logically possible patterns are not attested in the world es.Interestingly there are no gaps in the eight t-terns that reflect the layered organization of the NP presented above (withthe adjective typically a qualifying modifier always next to the nounand the demonstrative typically a localizing modifier always in theperiphery (Rijkhoff 2004):

78

(28)dem num A N Alamblak, Dutch, Georgian, Hungarian,

Kayardild, Ket, Nama Hottentot, ImbaburaQuechua, Pipil, Tamil, Turkish

dem num N A Burushaski, GuaranRomance languages)

dem A N num Zandedem N A num Bambaranum A N dem Berbice Dutch Creole, Bislama, Sranannum N A dem Basque, Hmong NjuaA N num dem SangoN A num dem Oromo, Fa d'Ambu, Nubi

By contrast, languages that employ one of the remaining 16 -iconicpatterns as the basic order seem to be extremely rare. Greenberg (1966:86 87) had one such language in his sample, Kikuyu, which has the order[N dem num A], but Seiler (1978: 322) claims that this is avariant A num dem]. Hawkins(1983), whose sample contained over 300 languages, mentioned two otherBantu languages with a non-iconic basic pattern in the NP: Aghem [N Adem num] and Noni (which has [N dem num A] as well as [N dem Anum]). It remains to be seen, however, if the Aghem and Noni sequencesare really integral NPs (see Rijkhoff 2002: 272 6 for some arguments toshow that we may be dealing with appositional or complex structuresrather than simple integral NPs in the case of non-iconic patterns).There is also evidence from morphology and clause-internal syntax to

support the layered model outlined in Figure 1. In English, for example,different kinds of temporal satellites tend to occur in the order time dura-tion ( ), time frequency (or so ) and time position (

), as in Quirk et al. (1985: 551):

(29)

Regarding morphological evidence, Bybeeof inflectional morphemes relative to the verb in a representative sample offifty languages revealed that l-lowed by tense, and then by mood. The only exception to this orderingfound in the 50-language sample is in Ojibwa, where the Dubitative suffixprecedes the Preterite suffix

79

Historical evidence to support parallel treatment in the layered analysisof the clause and the noun phrase comes from various sources. For exam-ple, Gildea (1993) has shown how demonstratives have developed intotense markers in Panare. Several authors have described phonological simi-larities between markers of nominal and verbal plurality (see for exampleFrajzyngier 1977, 1997; Mithun 1988; Newman 1990) and historical con-nections between markers of collectivity and perfectivity (both: Quality)were already discussed by Von Garnier (1909).As to lexical word classes, it appears that languages across the globe

employ one of the seven major parts-of-speech systems listed in Figure 8(for a more detailed typology of parts-of-speech systems that also includesintermediate types, see Hengeveld et al. 2004).

Type 1Type 2 verbType 3 verb nounType 4 verb noun adjective adverbType 5 verb noun adjectiveType 6 verb nounType 7 verb

Parts of speech system (adverb = manner adverb; based on Hengeveld1992: 69).

Apart from quantitative differences (number of distinct word classes, rang-ing from one to four), languages may also differ with respect to the degreeof flexibility displayed by the members of a certain word class. For exam-ple, in English the adjective beautiful needs to be provided with an ad-verbializing suffix before it can modify a verb:

(30)(31)

But there are also languages such as Ngiti (Type 3), which do not distin-guish between adjectives and (manner) adverbs (Kutsch Lojenga 1994:336):

80

(Kutsch Lojenga 1992: 338)(32)

child RSM-carry:PRF.PRES light load

(33)light child RSM-carry:PRF.PRES load

load easily

In other words, Ngiti and other languages with a Type 3 parts-of-speechsystem the same element is used as a qualifying modifier ( l-lite n Figure 1) in the NP and in the clause.

It appears that the four-layered model presented above needs to be ex-panded with another layer of modification, both for the NP and the clause.This layer concerns , which further specify the ofentity denoted by the noun (or verb) and forms the innermost layer ofmodification, i.e. that between the head constituent and the layer that ac-commodates qualifying modifiers.Classifying satellites are lexical items that further specify what kind of

entity is being denoted by the head noun, for example: inreport in inor in ginal numbering of the layersintact, this additional descriptive layer will be referred as the -layer(Figure 9).14

Classifying modifiers typically appear adjacent to the head noun anddiffer from qualifying modifiers in that they do not specify an objectiveproperty of the entity ( , ) or the speakerattitude towards the entity ( , ), butrather a particular subclass of the entity in question. Here are some moreexamples of classifying satellites from English and other languages:

(34) a.bc.d. 15

81

(Koptjevskaja-Tamm 2003: 539 40)(35) a.

a:C people-DEF.C-GEN theatre

b.a:C school-DEF.C.-GEN task

(Koptjevskaja-Tamm 2002: 155)(36) a.

bread GEN knifead knife

b.coffee-GEN cup

Clause operators ( Clause satellites (4. Discourse-Referential3. Location2. Quantity1. Quality0. Kind

0. Kind1. Quality2. Quantity3. Location4. Discourse-Referential

NP operators ( NP satellites (

A five-layered NP/clause model: preliminary version (cf. Figure 10).

82

Classifying satellites also occur in the kind of classifier construction that isattested in Arrernte and other Australian languages.

(Wilkins 2000: 179 200)(37) a.

ant meat-antb.traditional medicine meat-ant

c.socially relevant place meat-ant

Even though all the constructions in (37) could be translated asWilkins (2000: 192) states that in each case the semantic effect of the clas-sifying noun is different. For example, the combination in(37b) could be paraphrased as follows:

In using the classifier constructionmeat-ant ddressee to think aboutthe referent of the noun phrase from the point of view of its having the prop-erties of an -ant r-rent point in the discourse are its properties as an i-cine

In other words, the classification is based on the way gures ina particular context: as food in (37a), as a medicine in (37b), or as an entityin aIn English classifying adjectives differ from qualifying adjectives in

that they usually do not admit intensifiers, comparison, or predicative posi-tion (Quirk et al. 1985: 1339; on non-predicative adjectives, see also e.g.Farsi 1968, Levi 1973):

(38) vs. *[intensifier]

(39) vs. *[comparison]

(40) vs. *[predicative position]

83

The semantic range of classifying satellites is rather broad and includessuch categories as material, purpose and function, status and rank, origin,and mode of operation (Halliday 2004: 320). Essentially they relate to anyfeature that may serve to classify entities into a system of smaller sets. Theexamples show that various forms and constructions may serve as an ad-nominal classifying satellite (e.g. adjective, noun, genitive NP), and that itmay be difficult to draw the line between a noun + classifying satellitecombination and compounds or quasi-compounds (cf. also Giegerich2005).16 Furthermore, the same adjective can function either as a classify-ing or a qualifying satellite (Farsi 1968; Halliday 2004: 319). For example,an adjective such as serves as a qualifying satellite in(civil rightsExamples of classifying satellites at the level of the clause are more dif-

ficult to find, since it seems they tend to occur as incorporated forms, as in(41b):

(Merlan 1976: 185)(41) a.

where 3.SG:is knife

:(41) b.

3.SG 3.SG:it-knife-cut bread

There is, however, also the phenomenon called1986, 1989; Gerds 1998) or1984). Whereas an incorporated element is part of another word, a strippednoun is a separate word (according to phonological criteria such as stressplacement), which must appear next to the verb. Thus, in Kusaiean adverbscan appear between verb and object (42a) but not between verb andstripped noun (42b).

(Gerds 1998: 94; original example in K. Lee 1975)(42) a.

Sah he sharpen diligently knife the

84

(42) b.Sah he sharpen knife diligently

-sharpening

As in the case of incorporation we see that in (42) a distinction is madebetween sharpening in general and a certain kind of sharpening, namelyknife-sharpening. The crucial difference is that in the case of a strippednoun we are dealing a more or less independent element that serves as aclassifying satellite at the level of the clause.17

Classifying operators are members of grammatical (rather than lexical)modifier categories which (further) specify what kind of spatial or tempo-ral entity (thing, event) is being referred to. As a matter of fact classifyingoperators were already represented as a distinct modifier category in the

-layered NP/clause model, where they were erroneously catego-rized as d 2).In earlier studies I argued that, just as verbs are coded for a particular

((

noun un(Dik 1997: 224f.;

note that is also known as lexicalized aspectuality), nominalaspect markers can change the of a noun (Rijkhoff 2002: 100121). For example, nouns in Oromo are lexically coded for a that Ihave calleddirect construction with a numeral), but when they are provided with a(what I call) collective or an individual aspect marker, they designate aspecial kind of set, viz. a or a (with just onemember) respectively:

(Stroomer 1987: 76 77, 84 85)(43) horse/horses set) vs.

(collective set)

(44)(singleton set)

85

It turns out now that the nominal and verbal aspect markers I erroneouslyregarded as qualifying operators in the -layered model have to dowith icated in theway they were characterized (Rijkhoff 2002: 101): It is important to makeclear at the outset that .. nominal aspect markers .. specify what kind of ..entity is being referred to ..rather than qualifying operators in that they relate to the kind of thing (e.g.individual, collective) or event (e.g. process, action) that is being referredto by the speaker. But that leaves us with the question: what about qualify-ing operators? What is the grammatical equivalent of the qualifying satel-lite in the NP and in the clause? This is discussed in the next section.

Operators are members of modifier categories that are expressed gram-matically rather than lexically. They belong to a closed paradigm andmanifest themselves morphologically as inflectional modifications of lexi-cal items or as tical or functionwords). For example, if the operator value o-logical expression it typically takes the form of an article or an affix, butnote that Definiteness may also be expressed through phonological or syn-tactic means (Rijkhoff 2002: ch. 6). Since the members of a grammaticalword class by definition constitute a smallish, closed set of items, the dis-tinctions expressed by operators are absolute (Dik 1997: 160). For exam-ple, if Definiteness is a grammatical category in a language, the only twoavailable choices are +Definite or Definite (indefinite). More examplesare given in Table 1, which shows that operator values in the various se-mantic domains typically capture a limited number of crucial distinctions(no attempt has been made to give an exhaustive listing of domains or val-ues).

86

. Distinctions in operator categories (based on Dik 1997: 160)

NUMBERTENSEPOLARITYASPECTMOOD

ILLOCUTION

Let us now return to the question posed in the title of this section:qualifying operators exist?inherent objective or subjective properties of an entity, typically expressedby adjectives if a language has them. Interestingly all major semantictypes of adjective distinguished in Dixon (1982: 16) include gradableproperties: dimension ( x-tremely big lour(value (feature of qualifying modifiers (as is suggested by these facts), this wouldexplain why there are no qualifying operators: operators do not lend them-selves to specifying degrees of some variable property: an object can be

Satellite fairly SingularOperator18

2.3. Formal representation

A formal representation of the underlying structure of the NP may appearas in (45). Note that here the interpersonal and the representational levelsare conflated, whereas in Hengeveld and Mackenzieco-exist as separate entities (cf. Hengeveld this volume):

(45) NPi: 4[ 3[ 2[ - [ 0[ (NOUN (fi))(xi) ]Lc 0(Lc) ]L0 1(L0) ]L12(L1) ]L2 3(L2) ]L3 4(L3)

As in (1), each operator ( , ,argument. For example, quantity layer L2 is the argument of operator 3

and satellite 3.

87

= term (NP) operator; the argument of classifying operator 0 is theinnermost layer (the core layer, hence labeled LC), whereas the ar-gument of discourse-referential operator 4 is layer L3 (which con-tains all other layers);

f = predicate variable (here symbolizing the head noun);x = referent variable (symbolizes the referent of the NP);/ term (NP) satellite; the argument of classifying satellite 0 is the

core layer (Lc), whereas the argument of discourse-referential satel-lite 4 is the material contained in localizing layer L3.

CLAUSE OPERATORS CLAUSE SATELLITES

4 4. Discourse-Referential 4

3 3. Location 2

2 2. Quantity 2

1. Quality 1

0 0. Kind 0

0 0. Kind 0

1. Quality 1

2 2. Quantity 2

3 3. Location 3

4 4. Discourse-Referential 4

NP OPERATORS NP SATELLITES

A five-layered NP/clause model.

88

This section investigates how (elements in) the revised NP structure pro-posed above interact with contextual factors.19 The three main claims ofthis section are:(a) the contextual component in the current FDG model must be split

up into two separate contextual components (see also Keizer thisvolume): one for discourse (co-text) and another for the discourseevent orexternal world);

(b) all major components in the FDG model provide a context of somesort: a grammatical context (G-context), a conceptual/cognitive/mental context (C-context), a discourse context (D-context), and asituational/external context (E-context);

(c) a single rule can account for any kind of contextual influence on alinguistic expression.

3.1. Components as contexts

The current FDG model (Figure 11) consists of four interacting modules:one central grammatical unit surrounded by three marginal modules (theconceptual, contextual, and output components).

The contextual component is characterized as follows (Hengeveld 2005:58):

The CONTEXTUAL COMPONENT contains a description of the discourse do-main as it has been built up during the current discourse to the extent thatthis is relevant to the form that subsequent utterances may take. It does notonly contain a description of the content and form of the preceding dis-course, but also of the actual perceivable setting in which the speech eventtakes place.

89

General layout of FDG (Hengeveld 2005: 61).

This description clearly states that the contextual component actually con-sists of two distinct parts: a textual component (co-text) and a situational

FramesLexemesPrimaryoperators

Interpersonal Level

Phonological Level

Structural Level

Representational Level

Morpho-syntactic Encoding

Phonological Encoding

ProsodicpatternsMorphemesSecondaryoperators

Articulation

Utterance

Formulation

TemplatesAuxiliariesSecondaryoperators

90

component. Since FDG is a theory of grammar (rather than a theory of theworld), I believe there are several reasons to argue for a strict division be-tween the purely linguistic context and alinowski1923). Firstly, it is important not to confuse different kinds of entities:

in the sense of co-text is a linguistic entity, a isnot.20 Secondly, as co-text is a finite entity (basically consistingof the linguistic material preceding and following an utterance in somediscourse), whereas the external context in which a discourse event takesplace is essentially an open-ended phenomenon.21 Consequently, we haveto be very selective in our representation of the situational context if wewant to avoid the impossible (and for linguists: unnecessary) task of havingto give a complete description of the physical context, which could meanthe entire universe (where does1978: 210f.). The discriminative and subjective nature of event representa-tion is also reflected in statements by e.g. Dik (1997: 13) and, more re-cently, Connolly (2004: 105), who wrote: p-tion will have to be selective. There is no possibility of including every factabout the situation. The basic principle is to include what one feels to berelevant, insofar as one ledge and the evidence permits.The problem is complicated further by the fact that features of the com-

ponents of the speech situation (such as sex, age, class, caste, country oforigin, generation, region, schooling, ethnicity, kinship relation and degreeof intimacy with other speech participants, occupational status; the physi-cal and psychological setting of the speech act; topic and purpose of com-munication) are often intricately connected, and that their relative impor-tance varies from language to language (Rijkhoff 1998). For example,whereas the social significance of sex is primary in one language, socialrank may override all other features in another language (see e.g. Ervin-Tripp 1972: 224 225 on such differences in the systems of address inBisaya and Korean).In my view grammarians must construct and test theories of (discourse)

grammar; it is not their task to give a principled account of all the elementsin the extralinguistic context that may have an influence on linguisticforms, meanings or functions. Therefore I suggest that one FDG compo-nent be strictly reserved for discourse in the sense of co-text (D-context)and another component for those elements in the speech event (E-context)that are considered to have an impact on the form, function, or meaning of(part of) an utterance (see also Butler this volume).

91

Hengeveld (2005: 57) gives the following characterization of the concep-tual component:

The CONCEPTUAL COMPONENT is not part of the grammar but is the drivingforce behind the grammatical component as a whole. [...] it represents theconversion of a prelinguistic conceptual representation into the linguisticallyrelevant semantic and pragmatic representations that are allowed by thegrammar of the language concerned. The conceptual component is responsi-ble for the development of both a communicative intention relevant for thecurrent speech event and the associated conceptualizations with respect tothe relevant events in the external real or imaginary world.

This description more or less covers the notion of context as employed inthe cognitive sciences, where context is first and foremost a mental phe-nomenon. The FDG interpretation of conceptual knowledge also clearlyrelates to the way context is defined by language philosophers and pragma-linguists: the set of background assumptions that are necessary for an ut-terance to be intelligible (Ungerer and Schmid 1996: 45 46). I will refer tothis component in the FDG model as C-context (see also fn.19).

The grammatical component constitutes yet another kind of context in theFDG model (G-context). This is the module where the discourse act isanalysed pragmatically, semantically, morphosyntactically and phonologi-cally (but see section 4.1). Essentially each level provides another envi-ronment containing elements that may affect the form, function or contentof (part of) the linguistic expression. For example, at the RepresentationalLevel the plural operator is nothing but a contextual element in the under-lying representation of the NP that affects the form and meaning of an Eng-lish nominal predicate in the head slot (Dik 1997: 349).

(46) Pl [ N] =

92

The actual linguistic expression, which manifests itself as a series of spo-ken, signed or written symbols, is an entity that is produced and perceivedin the external world and which is, therefore, located in space and time. Forthis reason I regard the output component, whichsigned, or orthographic expressions 2004a: 6), as part of theE-context. Obviously, the same goes for the collection of linguistic expres-sions that together constitute the co-text. However, theof these two entities (the co-text and its constituent expressions) are

to be modelled in the grammatical and the discourse compo-nents and as such they are only theoretical constructs that exist in linguistsminds.

3.2. FDG components: some modifications

Having motivated the need for a separate component in FDG to account forthe influence of elements in the external/situational context (section 3.1.1),and having established that all major components in the FDG model repre-sent some kind of context in the cognitive, anthropological, philosophical,grammatical or textual sense (sections 3.1.2 4), we can now draw the fol-lowing picture of the components in a revised FDG model.

Contextual components in FDG.

Linguists working in the framework of FDG are mostly concerned with theG-context and the D-context. The most conspicuous difference between themodel outlined in Figure 12 and the current FDG model (as represented in

(External/Situational Component)

(ConceptualComponent)

(Grammatical Component)

a. Interpersonal Level (pragmatic analysis)b. Representational Level (semantic analysis)c. Structural Level (morpho-syntactic analysis)d. Phonological Level (phonological analysis)

(DiscourseComponent)

Co-text

93

Figure 11) concerns the organization of the components, but there are alsodifferences regarding the contents of the components. They will be dis-cussed in section 4.

3.3. A context rule

In section 3.1 I argued that the plural operator is an example of a contex-tual element in the underlying structure of the NP in the grammatical com-ponent of the FDG model. The rules that deal with the effect of an operatoron some linguistic expression are called xpression rules k1997: 349 358) and take the following general format:

(47) Operator [Operandum] = Value

This rule is easily transformed into a generalmodel which produces linguistic output for the grammatical component(which in turn can serve as the input for another context rule):22

(48) Context [input] output

Thus, the plural formation of English nouns may be accounted for as fol-lows in FDG:

(49)G-context: ([.. [.. [PL [ - [.. [ N ] ..]L0 ..]L1 ..]L2 ..]L3 ..]L4)input: bookNoutput: bookN-PL

This could be paraphrased as follows: -context contains an opera-tor with the valuefor an inflectional suffix to mark plural number: -PLInflform of this inflectional plural suffix is specified at the phonological leveland depends on the properties of the input predicate (Dik 1997: 351). Thismeans that, in order to produce the appropriate form, the grammar needs tocheck the following set of rules:

94

(50)1. Pl = /Iz/ if last phoneme of input predicate is sibilant; otherwise2. Pl = /s/ if last phoneme of input predicate is voiceless;

otherwise,3. Pl = /z/

(51)G-context: N-PLInflinput: /b k/[Condition: if the last phoneme of the predicate is voiceless,

then PL = /s/]output: / /

This could be paraphrased as: -context (at the phonological level)contains a marker for inflectional suffix PLInflfor possible conditions, then substitute PLInfl e-ment .Let us now look at some cases where the form or meaning of a linguis-

tic structure in the grammatical component is sensitive to some contextualelement in another component (C-context, D-context, E-context).

The interaction between NP structure in the grammatical component andinformation in the conceptual component (C-context) is manifested, amongother ways, in the value taken by certain operators after the formulator hasproduced an underlying structure for the grammatical component. For ex-ample, the reason why we normally refer toto our long-term knowledge of the world (or rather universe), which iscontained in the conceptual component. It is part of our conceptual know-ledge that there is only one such entity in our solar system. Since there isno danger that this entity will be confused with another entity of the sametype in normal conversation, the speaker may assume that the addresseewill be able to identify the intended referent ofentity may not have been mentioned before or is invisible at the time ofspeaking. In other words, the fixed value for the discourse-referential op-erator in the NP -context:

95

(52)C-context: the sun is a unique entity in our solar systeminput: ([value 4 = ? [.. [.. [ - [.. [ sunN ] ..]L0 ..]L1 ..]L2 ..]L3 ..]L4)output: ([value 4 = Def. [.. [.. [ - [.. [ sunN ] ..]L0 ..]L1 ..]L2 ..]L3 ..]L4)

The discourse component (D-context) contains a description of the co-textof the utterance. Obviously, FDG needs this component to account for allsorts of reflections of discourse phenomena in a clause, such as tail-headlinkage (Hengeveld 2005: 55).

(Van Staden 2000: 275): tail-head linkage(53) ...

... then Jafa carry.on.the.back basket 3.NH=there then

3.NH=there ascend upwards. upwards 3.SG.M pick banana...

picked the bananas ..

Discourse phenomena like tail-head linkage, clause-chaining or switch-referencing (cf. de Vries 2005) often require multiple interactions betweenvarious components, but a simplified account of tail-head linkage couldlook like this:

(54)D-context: In episode E of discourse D, the speaker wants to ex-

press coherence between Sentence B (whose underlyingstructure is being prepared for the grammatical compo-nent) and preceding sentence A (which is part of the D-context).

Input: prelinguistic version of sentence B in conceptual com-ponent.

Output: underlying structure of sentence B in grammatical com-ponent (sentence A.

: at the Structural Level the last word of Sentence A mustbe the first word in Sentence B.

96

Since linguistic expressions are always used by at a particularand , each language has devices to integrate information from theexternal, situational context. In the FDG model this means that the E-context needs to be checked for all instances of person, time, and placedeixis (notice that here the world of discourse overlaps with the externalreality; cf. fn. 19). For example, in Spanish a predicate adjective is markedfor the gender of the addressee:

(Hengeveld 2005: 58)(55)

what pale-F.SG COP-IND.PRES.2.SG

(56)E-context: Addressee is singular female personInput: predicative adjective ( )Output: predicative adjective + - suffix ( )

In the last couple of decades two more deictic categories have been intro-duced:1. discourse deixis, which concerns references to portions of the sur-

rounding discourse;2. social deixis, which is concerned with the grammaticalization, or

encoding in language structure, of social information (Levinson1987: 93).23

Whereas discourse deixis is obviously concerned with the D-context, socialdeixis must also be explained on the basis of information specified in theE-context. For example, many languages have a set of pronouns for specialsituations, as when the speaker wants to show respect or deference for theaddressee. The impact of the social status of the addressee in the externalreality (to be measured relative to the speaker c-ture in the grammatical component can be captured in the following way:

97

(57)E-context: Addressee is a person who in the current situation needs

to be addressed in a respectful fashion by the Speaker.Input: A=2SGOutput: U [= polite form of 2SG pronoun]

FDG is a theory of grammar in which the discourse act is the basic unit ofanalysis. The termit may suggest that the basic unit of analysis is a verbal activity, rather thana linguistic form or construction, i.e. the product of that activity.24 This isespecially true in studies in which FDG is presented as a dynamic modelreflecting the various phases in the process of language production (intention to articulationsection that in my view a theory of grammar must provide a model for theanalysis of linguistic expressions, not a model of verbal activities or theproduction process that results in these linguistic expressions (for a similarpoint see Hengeveld 2004b). In other words, what is being modelled in thegrammatical component is not an activity or a process but a linguistic con-struction (or 2004).So far we have been concerned with the layered structure of the NP and

the various modules (a closer look at the contents of the grammatical component or G-context. Iwill argue that the grammatical component is not the place in FDG to de-scribe non-linguistic entities such as (features of) Speaker and Addresseeor other aspects of the physical and psychological setting of the speechsituation. The G-context must only be used to describe and analyse theform, function and meaning of linguistic constructions.In section 4.2 I will elaborate on the observed parallels between NPs

and clauses (Figure 10, section 2.2.3). It will be argued that descriptivemodifiers (concerning Kind, Quality, Quantity, and Location) are specifiedat the semantic or Representational Level ( uage as carrier of con-tent -referential and other interpersonal modifiers arespecified at the pragmatic level (communicational process 2003: 111; Halliday 2004: 61). Theother interpersonal modifiers in the clause inform the Addressee about:

98

a. the of a spatial or temporal entity (thing xi andevent ei) in the world of discourse (some relevant modifier catego-ries: Definite, Specific, Realis, expressions of objective modality,discourse-referential satellites; section 2.1 3);

b. the of proposition Xi, in particular the speaker n-tal or cognitive attitude towards the proposition (some relevantmodifier categories: expressions of subjective modality and eviden-tiality, propositional satellites; Dik 1997: 295 299);

c. the of the clause (some relevant modifier cate-gories: Declarative - Interrogative - Imperative - Exclamative, illo-cutionary satellites; Dik 1997: 300 307).

This means, among other things, that propositions, which are part of theRepresentational Level in current FDG, appear at the Interpersonal Levelin the version proposed here (section 4.2).

4.1. The place of speaker and addressee in FDG

If the grammatical component is concerned with the grammatical analysisof linguistic entities (NP clause co-text), this component should not in-clude descriptions of psychological or physical entities that are part of theextra-linguistic context. For example, both Hengeveld (2004b) and Con-nolly (2004: 108, 112) have proposed that the Interpersonal Level can alsobe used as a platform for the specification of (features of) Speaker P1 andAddressee P2. In current FDG, the linguistic expression (symbolized by C =communicated content, i.e. the information transmitted in the discourse act)is an argument of an abstract three-place illocutionary predicate (the othertwo arguments being the Speaker and the Addressee).25

(M1: [(A1: [ILL (P1)S (P2)A (C1: [...(T1) (R1)...] (C1))] (A1))] (M1))

The Interpersonal Level (Hengeveld 2004a: 5).

In the FDG model defended here, the grammatical component is strictlyseparated from the external/situational component, which contains (fea-tures of) entities that are part of the external, physical world (section3.1.1). A modified version of the discourse act (i.e. the product) may be

99

shown as (58). For the sake of simplicity, the move in (58) consists of onediscourse act, which in turn consists of a single, full clause. In (58) theinterpersonal and the representational level are connected, but recall thatthey co-exist as separate entities in Hengeveld and Mackenzie rsion(see section 3.2.; the distribution of layers over the Interpersonal and theRepresentational Level is discussed in section 4.2).

(58) THE UNDERLYING STRUCTURE OF THE CLAUSE(Lc = core layer; see also (45)):

Interpersonal Level (LANGUAGE AS EXCHANGE)Movei: ([ 6 [ 5[ 4[ ]L3 4(L3)]L4 5(L4)]L5 6(L5)]L6)

[ 3[ 2[-[ 0[( (fi))(ei)]Lc 0(Lc)]L0 1(L0) + ]L1 2(L1)]L2 3(L2)]L3Representational Level (LGE. AS CARRIER OF CONTENT; see Fig. 10)

Operators and satellites at the Interpersonal Level:26

4 / 4 = discourse-referential operator / satellite (L3: Event orAffairs i)

5 / 5 = proposition operator / satellite (L4: proposition Xi)6 / 6 = illocutionary operator / satellite (L5: discourse act Ai)

In (58) the discourse act (a linguistic entity; see note 24) is still the basicunit of analysis within the move (Kroon 1995), but note that the schemadoes not contain variables for (features of) Speaker or Addressee, and thatlocution 6) rather than an abstract frame (cf.

Dik 1997: 66 7, 300f.).

4.2. Parallels between the NP and the clause revisited: the place of modalcategories in FDG

In section 2.2.3 I proposed a five-layered model of the NP: four layers toaccommodate descriptive modifier categories (classifying, qualifying,quantifying, localizing modifiers) and one layer for modifiers that are con-cerned with the status of entities in the world of discourse (hence s-course-referential modifiers component of FDG thedescriptive modifiers are specified at the Representational (semantic)

100

Level, whereas discourse-referential modifiers are specified at the Interper-sonal (pragmatic) Level; recall that here we are only concerned with NPsthat are used for spatial entities (symbolized by the x-variable):

INTERPERSONAL LEVEL ( LANGUAGE AS EXCHANGEInterpersonal modifiers in the NP are concerned with the statusof things in the .

Discourse-Referential 4 4

----------------------------------------------------------------------------------REPRESENTATIONAL LEVEL ( LGE. AS CARRIER OF CONTENTDescriptive modifiers specify properties of an entity in the

in terms of Kind, Quality, Quantity, and Location

Location 3 3

Quantity 2 2

Quality - 1

Kind 0 0

NP layers at the Interpersonal and the Representational Level.

If we now also include the other layers in the underlying structure of theclause, we see that two layers must be added:

101

At the , modifiers are concerned with theInterpersonal Status of kinds of entities in the :[i] clauses (or rather the messages contained in the clauses),[ii] propositions, [iii] events and [iv] things.

( 6, 6)inform ddressee about the illocution-ary status of the clause (Decl, Int, ).

( 5, 5)inform aboutassessment of / attitude towards aproposition Xi as regards theprobability, possibility or desirabilityof the actual occurrence of event ei.

( 4, 4) ( 4, 4)specify the existential status of thing xi or event ei in the

At the , modifiers specify properties of spatio-temporalentities (things, events) in the in terms of the notionsKind (Class), Quality, Quantity, and Location.

3. ( 3, 3) 3. ( 3, 3)2. ( 2, 2) 2. ( 2, 2)1. ( 1) 1. ( 1)0. ( 0, 0) 0. ( 0, 0)

NP and clause layers at the Interpersonal and theRepresentational Level (Rijkhoff fc. a).

Figure 15 shows that we find three layers of modification at the Interper-sonal Level, each of which is concerned with language as an instrument ofsocial interaction (Dik 1997: 5):27

(i) discourse-referential modifiers, which inform the Addressee aboutthe status of first and second order entities in the world of discourse (e.g.Definite, Realis). These modifiers specify whether or not an entity occu-pies a certain spatio-temporal region in the discourse world. Or they may

102

indicate the chance of occurrence of an event as measured against the back-ground of the speaker of the world and systems of moral,legal, or social norms (i.e. expressions of objective modality; Dik 1997:241 3).(ii) proposition modifiers, which inform the Addressee about the status

of a proposition (a third-order entity) in the discourse. They specify thespeakerproposition Xi with regard to the probability, possibility or desirability ofevent ei actually taking place. Interestingly, propositions (symbolized bythe p-variable in Figure 16 below) appear at the Representational Level inFDG, despite the strong interpersonal nature of propositional modifiers(Halliday 2004: 146 150).28

(pi: [(ei: [(fi) (xi)] (ei))] (pi))

The representational level in FDG (Hengeveld 2004a: 6).

I consider propositional modifiers to be part of the Interpersonal Level,because they clearly contain an instruction on the part of the speaker indi-cating to what extent statement Xi about event ei can be taken to be true orfactual by the addressee. The relevant modifier categories are attitudinalsatellites and expressions of evidentiality or subjective modality (Dik1997: 295 6).(iii) illocution modifiers, which inform the Addressee about the status

of the message (a fourth-order entity). These modifiers specify how themessage that is being communicated should be interpreted: as a statement(declarative), a question (interrogative), etc. (Sadock and Zwicky 1985).

This article argues that the Interpersonal Level in the grammatical compo-nent of the FDG model accommodates three layers of modification, each ofwhich is concerned with u-nicational process 2003: 111; Halliday 2004: 61): one for clauses(fourth-order entities), one for propositions (third-order entities) and onefor things or events (first and second order entities, respectively). All threelayers are relevant for the analysis of clauses at the Interpersonal Level,

103

whereas only one is relevant for the analysis of linguistic expressions thatare used to talk about concrete things, i.e. noun phrases.According to the model of FDG defended here, the Representational

Level in the grammatical component accommodates four layers ofmodification for both clauses and NPs. These descriptive layers areconcerned withrepresentation tler 2003: 111) and operators and satellites at this levelare distributed over classifying, qualifying, quantifying, and localizinglayers of modification (but recall that there are no qualifying operators).Since the speaker otional attitude towards persons or things is a globalproperty rather than a lexical feature of a particular modifier category, thisis accounted for in terms of anthat the speaker uses to express his or her emotional attitude (this could bemarked by a dotted underscore or other typographical means). This will bean indication that a special (non-lexical) meaning has been added, whichwill then also trigger the appropriate morpho-syntactic and phonologicalrules.In this version of the FDG model, the grammatical component does notcontain descriptions of elements from the external world (such as Speakeror Addressee), which is represented by a separate component, and all com-ponents are treated as different kinds of contexts:

Finally, it was shown that a single u-ence that any contextual factor in these components may have on the form,meaning or function of (part of) a clause.

1. I am grateful to Daniel Garc o and J. Lachlan Mackenzie for theirvaluable comments; the usual disclaimers apply.

2. In this article,grammatical modifier categories (satellites and operators).

3. Since the NP as well as its referent (x) can be the referred to anaphorically,both need to be indexed in some fashion in order to be able to distinguish be-tween the two, but I am no longer convinced that the structure proposed in

104

Rijkhoff (2002) is the best way to do so. Hence an alternative schema is pro-posed here, where xi symbolizes the referent and the indexed labelstands for the construction that is used to refer to that entity.

4. Although genericity is often discussed together with definiteness and specific-ity, it is perhaps better regarded as a property of propositions rather than NPs.Whereas one can find many languages with articles for definiteness or speci-ficity, I am not aware of any language that employs a special article for ge-neric reference (Behrens 2005: 276, Rijkhoff fc. a: fn. 7).

5. It is safe to say that there is no consensus on the status of referents, i.e.whether they are entities in the external, physical world or mental representa-tions of entities that exist in a discourse world (cf. Keizer this volume; see alsofn. 19). For the sake of the argument I will simply assume that in a normalconversation speaker and hearer set up a shared world of discourse in whichrealis clauses are used to refer to events that are deemed to have a correlate inthe external, physical world and in which irrealis clauses are used to refer toevents that may or may not have a counterpart in the external world.

6. Chung and Timberlake (1985: 241) wrote:way for an event to be actual, there are numerous ways that an event can beless than completely actual.whereas there is basically one way for a thing to be indefinite, there are nu-merous ways for a thing to be definite (Rijkhoff 1988).

7. Cf. Von Heusinger (2002a: 2):property of an NP being referentially anchoredfor a recent discussion of the notion

8. The difference between referents of specific and nonspecific-indefinite NPsmay also have to do with the fact that in the case of nonspecific reference weneed to set up separate discourse worlds for the various interlocutors (ratherthan having just one shared world of discourse for both speaker and hearer; cf.Kamp 2004).

9. As to the other meanings of d ortrained) one could hypothesize that each sense of rentset of meaning components of the nominal predicate Huang andAhrens 2003 on classifier coercion of nouns). Thus, -1 s to fea-tures concerning personal, human qualities (poor as a person), whereas -2qua doctor). Essentially the same procedure is used to account for the seman-tics of flexible parts-of-speech (see Hengeveld et al. 2004, Hengeveld andRijkhoff 2005, Rijkhoff fc. b).

10. See e.g. Farsi (1968); Clark and Clark (1977: 474); Hetzron (1978); Ney(1982, 1983); Fries (1985); Sichelschmidt (1986).

11. See Traugott and Dasher (2001) and Smith (2004) on diachronic develop-ments from subjective to intersubjective.

105

12. In Tongan a speakerselecting a special hward 1953, Hendrick 2005:908). Tomasello (2003: 12) notes that damentallyperspectival in the sense that a person may refer to one and the same entity as

, , or , or the same event as , , , ordepending on the communicative goals with respect to the lis-

tener nal states [13. The Interpersonal Level in FDG was inspired by the interpersonal metafunc-

tion in Hallidaythis level is concerned with discourse and attitudinal phenomena, only the lat-ter are part of the interpersonal metafunction in SFG (discourse phenomenafall under the textual component). It may also be useful to point out that inSFG no distinction is made between operators and satellites, grammar andlexicon, semantics and pragmatics, indicating that it can be dangerous to sim-ply copy an element of one theory into another. Note furthermore thatHengeveld ifier at the Interpersonal Levelmay have been motivated by his idea that NPs should be compared withpropositions rather than predications (Hengeveld 1998). For this comparisonto work, one would need to find a counterpart of subjective modality in theNP. In current FDG, however, subjective mood is represented at the Represen-tational Level (see also section 4.2).

14. See Chappell and McGregor (1989: 28); Quirk et al. (1987: 1339 40); Halli-day (2004: 319-320).

15. Apparently is ambiguous, meaning either tail of an unidenti-fied canine , personal com-munication); see also below on the difficulty of distinguishing between classi-fying and qualifying modifiers without a proper context. Notice furthermorethere is a dependency relation between the value of the discourse operator andthe function of a satellite (Quirk et al. 1987: 1276). For example, in a definiteNP such as or as a local-izing or identifying satellite, whereas in an indefinite NP such as

or16. In addition to the observed relation between compounding and classification

there also a connection between inalienability and classification (as inmountain topand McGregor 1989).

17. Consider also examples such ashandwashed the laundry

18. On gradable adjectives see e.g. Cruse (1986: 202f.); for a recent discussion ofgradability, see Wechsler (2005).

106

19. Some recent studies on the notion ofAkman and Bazzanella (2003) and Giv (2005).

20. Theopen-ended and could either be represented in the conceptual component(since this component is responsible for the development of iatedconceptualizations with respect to the relevant events in the external real orimaginary world see section 3.1.2) or in a separate module in the FDGmodel.

21. On this problem, see also e.g. Giv (1989: 76):the existence of an irreducible residue, a recalcitrant escape clause concerningthe open-endedness of ptured,however exhaustive and refined one ay be.

22. Ultimately this rule is probably due to Frege (1891): [] .

23. Bystander deixis is usually regarded as a subcategory of social deixis andsubsumes a variety of linguistic phenomena such as haplology, avoidancestyles and secret languages (Rijkhoff 1998).

24. The fact that , as well as the acts that they contain (and ) are said to be 2004a: 5-6)

also suggests that what is being analysed in FDG are activities rather than theof these activities.

25. M = move, A = discourse act, Ill = illocution, C = communicated content, T =ascriptive subact, R = referential subact. Since all linguistic constructions areindexed, they can still be connected to Speakers, Addressee and other entitiesin the non-grammatical modules (

26. In theory one could also include operators and satellites at the level of theMove, as in ( 7[ 6 [---]L5 6(L5)]L6 7(L6)), where 7 is a move operator and7 a move satellite.

27. There is of course also the pragmatic status of an entity in terms of newswor-thiness (topic, focus), but this is realized as a function rather than a modifier.

28. Before the introduction of FDG, the proposition was still part of the Interper-sonal Level (cf. Hengeveld 1990: 1, 6; see also fn. 12).

107

Akman, Varol and Carla Bazzanella (eds.)2003 The complexity of context. 35-3 (special issue

onAnstey, Matthew P.2004 Functional Grammar from its inception. In: J. Lachlan Mackenzie

and Mar -Gonz, 22-71. Berlin/New York: Mouton de Gruyter.

Bateson, Gregory1979 . New York: Dutton.

Behrens, Leila2005 Genericity from a cross-linguistic perspective. 43-2:

275-344.Butler, Christopher S.2003

. Amsterdam: Benja-mins.

this vol. Interpersonal meaning in the noun phrase.Bybee, Joan L.1985 .

(Typological Studies in Language 9.) Amsterdam: Benjamins.Chappell, Hillary and William McGregor1989 Alienability, inalienability and nominal classification. In: Kiria Hall,

Michael Meacham and Richard Shapiro (eds.),

, 24 36. Berkeley: Berkeley Lin-guistics Society.

Chung, Sandra and Alan Timberlake1985 Tense, aspect, and mood. In: Timothy Shopen (ed.),

, 202 258. Cambridge: Cambridge UniversityPress.

Churchward, C. Maxwell1953 . London: Oxford University Press.

108

Clark, Herbert H. and Eve V. Clark1977 .

New York: Harcourt Brace Jovanovich.Comrie, Bernard1976 . Cambridge: Cambridge University Press.

Connolly, John H.2004 The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and Mar(eds.), , 89 116 (Func-tional Grammar Series 24.) Berlin/New York: Mouton de Gruyter.

Craig, Colette Grinevald1977 . Austin: University of Texas Press.

Cruse, David Alan1986 . Cambridge: Cambridge University Press.

Denning, Keith and Suzanne Kemmer (eds.)1990 Stanford:

Stanford University Press.Dik, Simon C.1997

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

Dixon, Robert M.W.1982

. Berlin/New York: Mouton de Gruyter.Ervin-Tripp, Susan1972 On sociolinguistic rules: Alternation and co-occurrence. In: John J.

Gumperz and Dell Hymes (eds.),, 213 250. New York: Holt,

Rinehart and Winston.Escribano, Josthis vol. On noun phrase structure in F(D)G: Some conceptual issues.

Farsi, A.A.1968 Classification of adjectives. 18: 45 60.

Fortescue, Michael2004 The complementarity of the process and pattern interpretations of

Functional Grammar. In: J. Lachlan Mackenzieand MarGonz , 151178 (Functional Grammar Series 24.) Berlin/New York: Mouton deGruyter.

Frajzyngier, Zygmunt1977 The plural in Chadic. In: Paul Newman and Roxana Ma Newman

(eds.), , 37 56. Leiden: Afrika-Studiecentrum.

109

1997 Grammaticalization of number: from demonstratives to nominal andverbal plural. 1-2: 193 242.

Frege, Gottlob1983 . Function und Begriff (Nachdruck). In: Karel Berka and

Lothar Kreiser,, 63 82. (dritte, erweiterte Auflage unter

Mitarbeit von S. Gottwald und W. Stelzner). Darmstadt: Wissen-schaftliche Buchgesellschaft. First published 1891.

Fries, Peter H.1985 Towards a discussion of the ordering of adjectives in the English

noun phrase. In: Benjamin F. Elson (ed.),

, 123 133. Dallas: Summer Instituteof Linguistics.

Garnier, K. von1909 COM als perfektierendes Praefix bei Plautus, SAM im Rigveda,

CYN bei Homer. 25: 86 109.Gerds, Donna B.1998 Incorporation. In: Andrew Spencer and Arnold M. Zwicky (eds.),

, 84 100. Oxford: Blackwell.Giegerich, Heinz J.2005 Associative adjectives in English and the lexicon-syntax interface.

41-3: 571 591.Gildea, Spike1993 The development of tense markers from demonstrative pronouns in

Panare (Cariban). 17-1: 53 73.Giv1989 . Hillsdale NJ: Law-

rence Erlbaum.2005

. Amsterdam: Benjamins.Greenberg, Joseph H.1966 Some universals of grammar with particular reference to the order of

meaningful elements. In: Joseph H. Greenberg (ed.),(2nd edition), 73 113. [Reprinted in Denning and Kemmer

(eds.), 40 70.] Cambridge: The MIT Press.Halliday, Michael Alexander Kirkwood2004 (third edition, revised by

Christian M.I.M. Matthiessen). London: Arnold.

110

Hawkins, John A.1983 . (Quantitative Analyses of Linguistic Struc-

ture.) New York: Academic Press.Hendrick, Randall2005 Tongan determiners and semantic composition. 81-4: 907

926.Hengeveld, Kees1990 The hierarchical structure of utterances. In: Jan Nuyts, A. Machtelt

Bolkestein and Co Vet (eds.),, 1 23. Amsterdam: Benjamins.

1992 . Berlin/NewYork: Mouton de Gruyter.

1998 Terms and propositions. Paper read at the(ICFG8), Free University Amster-

dam, 8 July 1998.2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzie and Mar, 1 21. (Functional Grammar Series

24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzie and Mar

(eds.), , 365 378.(Functional Grammar Series 24.) Berlin/New York: Mouton deGruyter.

2005 Dynamic expression in Functional Discourse Grammar. In: Casper deGroot and Kees Hengeveld (eds.),

, 53 86. Berlin/New York: Mouton de Gruyter.this vol. Prototypical and non-prototypical noun phrases in FDG.

Hengeveld, Kees and J. Lachlan Mackenzie2005 Interpersonal functions, representational categories, and syntactic

templates in Functional Discourse Grammar. In: Mar -Gonz

, 9 27. Bern: Peter Lang.2006 Functional Discourse Grammar. In: Keith Brown (ed.),

2nd Edition, Vol. 4, 668 676. Oxford:Elsevier.

Hengeveld, Kees and Jan Rijkhoff2005 Mundari as a flexible language. 9-3: 406 431.

Hengeveld Kees, Jan Rijkhoff and Anna Siewierska2004 Parts-of-speech systems and word order. 40-3:

527 570.Hetzron, Robert1978 On the relative order of adjectives. In: Hansjakob Seiler (ed.),

111

, 165 184. T ngen:Narr.

Heusinger, Klaus von2002a Specificity and definiteness in sentence and discourse structure.

19: 209 211.2002b The cross-linguistic implementations of specificity. In: K. Jaszczolt

and Ken Turner (eds.) Vol-ume 2, 405 421. (Pragmatics and Beyond NS 100.) Amsterdam:Benjamins.

Huang, Chu-Ren and Kathleen Ahrens2003 Individuals, kinds and events: Classifier coercion of nouns.

25: 353 373.Jurafsky, Daniel1996 Universal tendencies in the semantics of the diminutive.

72-3: 533 578.Kamp, Hans2004 Specific-indefinites and anchored logical form. Paper presented at

Copenhagen Determination Symposium, Copenhagen BusinessSchool (Denmark), 26 August 2004.

Karttunen, Lauri1976 Discourse referents. In: James D. McCawley (ed.),

, 363 387. (Syntax and Semantics 7.) NewYork: Academic Press.

Keizer, Evelien2004 Term structure in FG: A modest proposal.

78.this vol. Reference and ascription in FDG: An inventory of problems and

some possible solutions.Koptjevskaja-Tamm, Maria2002 Adnominal possession in the European languages: form and function.

(STUF) 55 2: 141 172.2003 A woman of sin, a man of duty, and a hell of a mess: non-determiner

genitives in Swedish. In: Frans Plank (ed.),, 515 558. Berlin/New York: Mouton de

Gruyter.Kroon, Caroline1995 nam, enim, autem, vero

at. Amsterdam: Gieben.Kutsch Lojenga, Constance1994 . K

112

Lee, K.1975 . Honolulu: University Press of Hawaii.

Lefebvre, Claire1998 Multifunctionality and variation among grammars: The case of the

determiner in Haitian and in Fongbe.13-1: 93 150.

Lefebvre, Claire and Anne-Marie Brousseau2002 . (Mouton Grammar Library 25.) Berlin/New

York: Mouton de Gruyter.Levi, Judith N.1973 Where do all those other adjectives come from? Papers from the

ninth regional meeting of the Chicago Linguistic Society, April 1315. In: Claudia Corum, T. Cedric Smith-Stark and Ann Weiser (eds.),

, 332 345. Chicago: Chicago Linguistic Society.Levinson, Stephen C.1987 . Cambridge: Cambridge University Press. First published

1983.Malinowski, Bronislaw1923 The problem of meaning in primitive language. In: C. Ogden and I.

Richards (eds.),, 296

336. New York: Harcourt, Brace and World.Martin Laura1998 Irrealis constructions in Mocho (Mayan).

40 2: 198 213.Merlan, Francisca1976 Noun incorporation and discourse reference in Modern Nahuatl.

42-3: 177 191.Miner, Kenneth L.1986 Noun stripping and loose incorporation in Zuni.

52-3: 242 254.1989 A note on noun stripping.

55: 476 7.Mithun, Marianne1984 The evolution of noun incorporation. 60-4: 847 894.1988 Lexical categories and the evolution of number marking. In: Michael

Hammond and Michael Noonan (eds.),, 211 234. New York: Academic

Press.

113

Newman Paul1990 . (Publications in African

Languages and Linguistics 12.) Dordrecht: Foris.Ney, J.W.1982 The order of adjectives and adverbs in English.

6: 217 257.1983 Optionality and choice in the selection of order of adjectives in Eng-

lish. 23: 94 128.Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik1985 . London:

Longman.Roberts, John R.1990 Modality in Amele and other Papuan languages.

26: 363 401.Rijkhoff, Jan1988 A typology of operators.

29.1998 Bystander deixis. In: Yaron Matras (ed.),

, 51 67. (Sondersprachenforschung 3.) Wies-baden: Harrassowitz.

2002 . (Oxford Studies in Typology and TheoreticalLinguistics.) Oxford: Oxford University Press.

2004 Iconic and non-iconic word order patterns: on symmetry in the NPand counter examples to Universal 20. In: Henk Aertsen, Mike Han-nay and Rod Lyall (eds

, 169 180. Amsterdam: Free University, Facultyof Arts.

fc. a Descriptive and discourse-referential modifiers in a layered model ofthe noun phrase. 46-4 (2008).

fc. b On flexible and rigid nouns (and verbs). To appear in.

Rijkhoff, Jan and Johanna Seibt2005 Mood, definiteness and specificity: A linguistic and a philosophical

account of their similarities and differences.3 2: 85 132.

[http://ojs.statsbiblioteket.dk/index.php/tfs/index]Sadock, Jerrold M. and Arnold M. Zwicky1985 Speech act distinctions in syntax. In: Timothy Shopen (ed.),

155 196. Cambridge: Cambridge University Press.Searle, John1978 Literal meaning. 13: 207 224.

114

Sichelschmidt, Lorenz1986 Optionality and choice in the analysis of adjective order: comments

on Ney. 40-2: 135 148.Seiler, Hansjakob1978 Determination: A functional dimension for interlanguage compari-

son. In: Hansjakob Seiler (ed.),

, 301 328. TSmith, John Charles2004 An analysis of Romance th Penn

Linguistics Colloquium, 27-29 February 2004, Department of Lin-guistics, University of Pennsylvania PA, USA.

Staden, Miriam van2000 Tidore: A linguistic description of a language of the North Moluccas.

Ph.D. dissertation, University of Leiden.Stokes, Bronwyn1982 A description of Nyigina, a language of the West Kimberley, West-

ern Australia. Ph.D. dissertation, Australian National University.Stroomer Harry1987

. (Cushitic Language Stud-ies 6.) Hamburg: Buske.

Tomasello, Michael2003

. Cambridge, M.A.: Harvard University Press.Traugott, Elizabeth Closs and Richard B. Dasher2001 . (Cambridge Studies in Linguistics

97.) Cambridge: Cambridge University Press.Ungerer, Friedrich and Hans-J1997 London/New York: Long-

man.Vries, Lourens de2005 Towards a typology of tail-head linkage in Papuan languages.

29-2: 363 384.Van Valin, Robert D., Jr. and Randy J. LaPolla1997 . Cambridge: Cambridge

University Press.Wechsler, Stephen2005 Weighing in on scales: A reply to Goldberg and Jackendoff.

81-2: 465 473.

115

Wilkins, David P.2000 Ants, ancestors and medicine: A semantic and pragmatic account of

classifier constructions in Arrernte (Central Australia). In: GunterSenft (ed.), , 147 216. (Language,Culture and Cognition 4.) Cambridge: Cambridge University Press.

1

The F(D)G account of noun phrase (NP) senses rests crucially on the con-cepts of , , and . Yet, the status of F(D)G variablesand operators is problematic, and extant F(D)G pronouncements on theirscope as well as that of other constituents of the underlying structure ofNPs are contradictory. To a great extent this is because the F(D)G theoryof construction is vague. Correspondingly, surface NP syntax largely restson (DI), (HP), and , which, in the FGversion, yield only weak predictions.The purpose of this paper is to draw attention to such shortcomings, di-

agnose their causes, relate them to wider issues of ontology, semantics, andsyntactic theory, and extract the relevant consequences in the area of NPstructure and interpretation. Although inevitably critical, the paper is meantas a sympathetic outsider F(D)G programme, and itsultimate aim is therapeutic. Hence, along with careful critical discussion ofterm structures and their surface consequences, some suggestions will beoffered that might yield a more coherent and elegant account without aban-doning any of the constitutive assumptions of F(D)G.It must be emphasized, though, that the article remains basically criti-

cal-therapeutic: it is just a well-meaning outsiderF(D)G scholars to clarify certain aspects of their account of NPs. It is notan alternative full F(D)G theory of NPs (e.g., no attempt is made to decidewhich operators, core types, or types of satellites are appropriate, etc.), nora plea for any competing version of F(D)G. The suggestions that accom-pany critical discussion below are not incompatible with any deep F(D)Gprinciple, but they do diverge from current F(D)G doctrine and, to the ex-tent that they amount to a significant strengthening of its key principles,make the theory more vulnerable to empirical refutation. On methodologi-cal grounds, this is as it should be, and, with respect to Indo-European lan-guages like English, the empirical consequences are favourable, but

118

whether such a strengthening is desirable in view of the facts F(D)G hascollected about Language in general is a different matter well beyond thescope of this article or its author s-tions below, in sum, are offered in good faith and for what they may beworth. It remains for F(D)G scholars themselves to decide whether build-ing an alternative account of terms/NPs along such lines is wise and suitstheir overall research strategy or not.

Current F(D)G term representations have become very heterogeneous as tothe nature of the information they represent (which results from mergingdifferent analytical levels), and they are packed with variables whose im-port is not always clear, cf. Escribano (2004b). Of course, some arestraightforward: for example, every semanticist will understand why therepresentation of a quantified NP contains a variable, and, to the extent thatany NP can pick out different extralinguistic entities, why it carries a vari-able ranging over possible referents, although whether sense and referenceshould be represented in a single formula is, of course, a moot point (cf.Keizer this volume). Yet, what some other F(D)G variables are doing at therepresentational level is more difficult to understand. For example, whyshould all lexical predicates carry variables? Apparently, the idea of en-dowing lexical predicates with variables first appeared in Dik (1989: 50)and was developed in Keizer (1991, 1992) and Hengeveld (1992a), cf. Dik(1997a: 55 footnote). However, although considering every predicate as thespecific value of a predicate variable may be appropriate in a representa-tion of the speech act as a set of action choices available to the speaker,such variables can hardly play any role in a formula that intends to capturethe of an NP, where lexical items, by definition, must have replacedsuch variables. The same applies to the variables that lexical items receivein current FDG for a different reason, i.e., categories with semanticimport. As Hengeveld and Mackenzie (2006: 673) put it,that ontological categories are reflected in the language system, they havethe status of semantic categories, each of which is provided with its ownvariable 2 Hence, every lexical item (and every NP, etc.) is entitled tocarry both a Dikkian variable and one of Hengeveld and Mackenzieof course, those two among many others, since there are many other ana-lytical perspectives (e.g., to the extent that it consists of a phonologicalform, it might carry a phonological variable; if it plays a discourse role, it

119

may be endowed with a discourse-functional variable; to the degree that itstands for an entity of the psychological world, it may carry a Psych-Wvariable, etc.). However, it would be absurd to try to integrate all thoseheterogeneous variables along with predicates, arguments, andvariables3 (if any are needed) into, say, a representation of the sense ofterms as reflected at the representational level so the basic point is: whichvariables, if any, do play a role in the computation of the NP se and itsintrinsic referential potential?4

It seems that this essential question has been somewhat lost sight of inrecent F(D)G work, and the result is very heterogeneous e-sentations in which most variables are doing no work at all but, on the con-trary, creating confusion in the formalism and its interpretation. It may benecessary, therefore,a. to recall the nature and role of variables in generalb. to clarify the variables that figure in current F(D)G representations,

andc. to distinguish those that are appropriate to sense representations

from those that belong to other analytical levels.

Since this article is meant as a small contribution towards a better F(D)Gaccount of NP semantics and syntax, only variables involved in the compu-tation of the sense of terms will be discussed here in some detail, but theissues that arise apply to F(D)G variables in general, and the Ockhamianreasoning deployed here has multiple consequences at clause and otherlevels which will be tacitly invoked but not discussed in depth.

2.1. Why variables?

To place the issues in perspective, it is convenient to start by askingwhether, and why, term representations should contain variables at all. Fornatural language semanticists, the source of variables is the standard logi-cal treatment of elementary quantification, in which first-order variablesare introduced, bound by quantifiers, to replace awkward conjunctive state-ments with appropriate generalizations. In predicate logic formulae like( x) [P1(x) n(x)] Q(x), for example, the (presumed) dualvariable iversal quantifier and saturates the openposition of the first-order predicates P1 n, and Q. That argument positionis supposed to be exactly the same as is alternatively satisfied by thereferent of an ordinary name like in , represented by the

120

constant rmula like S(j). Hence, the variable-occur.

Yet, on a little reflection, j ntent,for whereasdetermined partly by linguistic convention (in English, is the name of amale, etc.) and partly through use in an appropriate discourse context,stands only for a strange entity whose ontological status metaphysicianshave been discussing, inconclusively, for over two thousand years, andwhich suffices to saturate a predicate only to the extent that it isby a quantifier.Predicate Logic quantifiers, in their turn, are analyzed, after Frege, as

second-order predicates that apply to structures like P(x) and express exis-tential statements likex 5 Under such an analysis, whereas a sentence like containstwo semantically relevant constituents coextensive with the linguist a-nonical NP and VP, one like has no semantic con-stituent to match the NP . This consequence, which has al-ways seemed counterintuitive to grammarians, is justified to the extent thatit accounts for the strange properties of under negation, i.e.,the puzzling non-contradictoriness of

or . An ordinaryproper name like on the contrary being well-behaved under negation,cf. , can be directly represented as anindividual constant ari-able at the representational level.Admittedly, most NPs are not like proper names. When terms are ex-

pressed by phrases with a common noun as head and perhaps one or moremodifiers, as in , if the right truthconditions and entailments are to follow, it is necessary to lay bare theintensional contribution of their various constituents. In the simplest cases,such phrases are constituted by monadic predicates (

, etc.)6 and, of course, predicates need subjects, sothe question arises what the subject of such predicates is, an issue that en-tails determining what is a possible subject or predicate, i.e., choosing anontology7 and struggling with rather complex metaphysical problems, cf.Strawson (1959).Playing on the safe side metaphysics-wise, early philosophical seman-

tics assumed an austere ontology containing only individuals, according towhich only extensional individuals could be subjects and the only predi-cates were first-order predicates. Russell e-

121

scriptions, for example, would represent the meaning of lexically complexNPs like as a conjunction offirst-order predications upon a ividual i-ablea variable does intervene in the logical form of definite NPs. Obviously,

y by empiricist standards, butconvenient in that it left the smallest possible unanalysed residue and al-lowed for a straightforward representation of the truth conditions intro-duced by the internal constituents of complex NPs.Yet, the ontological status of the e-

mains, far from clear. Under some metaphysical interpretations, e.g.,Locke 1997: 268-276) or Wittgenstein [1918] 1922: 2.021,2.024, 2.0131), if is representedby a formula including the string (...x) (x) & (x) & (x)..., thevariable r-ties icist standards on precisely thosegrounds. On the other hand, interpretingcluster of necessarily concurrent properties of an entity) is unwarranted,since in such a formula the (x) is one of the predi-cates of t be a constituent ofalternative, apparently (cf. Loux 2002: 96-138), is to interpret d-ing to the eory of individuals. Yet, under amust stand either for the bundle or for a selected subset of the bundle, and,in either case, insurmountable logical problems arise: Ifbundle in ...P(x) & Q(x) etc., where properties like P and Q are themselvesconstituents of the bundle, P(x), Q(x) etc., become tautological or down-right inconsistent; if, on the contrary,each of P(x), Q(x) etc., may well be informative, but, then, by definition,the content of ious clauses P(x), Q(x), etc. that are supposedto characterize an individual cannot be unified, and the bundle-theoreticdefinition of individual identity disintegrates. To put it simply, P, Q, etc.,are in that case being predicated of different particulars and the formula isno longer a representation of any consistent bundle.The status of such

any of the major metaphysical theories of individuals on the market, butone thing seems clear: if we grant, for the sake of argument, thatout -specified individuals n-der-specified substratum theorists) cannot be synonymous withproperty-less, forthat has long been far richer. Hence, to be, must

122

minimally consist of the ontological properties distinguishing individualsfrom other entities (events, properties, etc.), except existence, since that iswhat an existential quantifier contributes to such formulae.Although, somewhat surprisingly, the uncertain status of

perceived as a flaw (to my knowledge), that early approach to the logicalform of NPs soon proved to have other important limitations. For example,modifiers do not always receive intersective interpretations (e.g.,

is not an individual who is both good and a driver). Various attemptswere made in the literature to handle such cases without abandoning anaustere first-order ontology. A popular one consisted in increasing the adic-ity of to introduce a hidden argument, i.e., saying that reallymeanspatch would not work for cases like or modal NPs like

, and eventually semanticists had to assume higher-typedpredicates, subjects that were themselves properties, and a much richerontology. Correspondingly, it became technically possible to quantify overproperties (cf. natural language expressions like

, etc ) via formulae like P[P(i) P(j)], (where P = a property and i, j individuals), and property vari-ables became appropriate in such cases of quantification.Otherwise, the usual computational machinery suffices: second-order

predicates combine with first-order predicate subjects in exactly the sameway first-order predicates like combine with first-order arguments like

to yield i.e., both cases reduce to.8 Thus, , itself a first-order predicate (type <e,

t>), is the result of a second-order predicate (type <<e, t>, <e, t >>)being functionally composed with a first-order predicate (type <e,t>) (cf. Partee 1995, Heim and Kratzer 1998: 68-73). Informally, predicateslike just need bjects in such cases, and the resultingpredicate will in its turn saturate an appropriate determiner oftype <<e, t> e> to yield a name of type e (= individual), but neither theindividual variable u-tation of such senses. In other words, higher-order properties and their

esented, but nothing else need be.A linguist like Dik (1997a: 133-136, 149-150) might surely object that

such logical representations of complex NPs, even if they capture theirbare truth-conditional aspects, are just too poorly structured to do justice tothe way semantic composition operates to yield senses in natural language.For example, there is plenty of empirical evidence (e.g., from standardconstituency tests like -anaphora, Ellipsis, Coordination, etc.) that, in

123

cases of multiple modification, the modifiers of natural languages are notjust added one after another, as suggested by the u-lae of predicate logic, but stacked, or even attached to different hierarchi-cally organized invisible heads within the NP (cf. Cinque 1994 and work inCinque ed. 2002). Thus, the four modifiers in

modify not the head noun, but different expansions thereof, i.e., mini-mally (more structure may be involved), if the lowest modifier modifies thenoun , the next higher one modifies , etc., and, depending onwhich modifiers and operators (functional categories, etc.) intervene, thehigher heads may, but need not, be of the same semantic type. This is abso-lutely correct, and further types are surely involved, but that need not entailnew variables in term representations.The semantic ontology did not long remain austerely limited to indi-

viduals and properties. Another early influential enrichment was David-son orated on in Parsons (1990) andsubsequent work, to capture entailments among predicates of differentadicity sharing homonymous heads. Of course, predication and quantifica-tion over events9 did require a new event variable 10 and new semantictypes for events, predicates of events, and events as modified in variousways, although Davidsonignore the hierarchical organization of natural language VPs, where argu-ments and modifiers appear at different layers of structure and affect dif-ferent heads and different types of entities.11 Thus, constructing LogicalForms to mirror the way expressions are compositionally built in naturallanguages surely entails allowing for recursion of entities within entities,properties within properties, events within events,12 etc., and accepting arich type system, a fact acknowledged in all major linguistic approaches,including F(D)G,13 but representing stacking and recursion does not requireany new variables in the internal structure of sense representations, which,as argued below, reduces to functional application/composition of th-order predicates and their arguments.In sum: as semanticists have attempted to capture more subtle aspects of

the sense of NPs in natural languages, the ontology has expanded to in-clude properties/sets, events, etc., with their own properties, recursion, etc.,but variables remain necessary to compute senses only to the extent quanti-fication is involved, if at all, and play no role otherwise. The rule of thumbis just this: symbols can only be compositionally computed if they standfor entities belonging to homogeneous ontological domains (sounds, in-scriptions, actions, speech acts, senses, etc.), and, by definition, if the valueof a variable symbol is computed in a formula, the variable cannot be so

124

computed, which makes it redundant at whatever analytical level is in-volved.

2.2. Variables in F(D)G term structures

The original term formulae in Dik (1978: 16, 57) and subsequent FG workdown to Dik (1997a: 160 footnote 1) are obviously inspired by the standardFregean-Russellian treatment of quantification, i.e., an FG term is a struc-ture x: (x): (x): (x), where ) by aset of grammatical operators and onadic first-order predicates : : .14 Although FG terms do not quite have theproperties of predicate logic ones (cf. Dik 1989: 117-118), term variablesraise the same issues discussed above with reference to the status of i-vidual -specific problems thathave remained unsolved, if not unnoticed.15 In this section, an attempt willbe made to identify them and diagnose their causes.First, the variable by Dik at the core of FG terms is

problematic, and for the same reasons the logicianswere. Dik (1989: 55, 115-118; 1997a: 136) presents it as a first-order vari-able picking out individuals, but such a statement immediately clashes withhis other assumptions. Observe that ifexpressions denoting individuals should be freely substitutable for it, but,of course, that is incorrect (no expression denoting an individual may re-place erm formula), as well as being incompatible with the FGview (e.g., Keizer 1991; Dik 1997a: 131; 1997b: 11) that only terms referto individuals. The term-variablein the FG sense any more than it can pick out individuals in the logiciansrepresentations. Dik himself (1989: 126; 1997a: 148) observes that l-lows from the definition of ividuals cannot be re-stricted itstands for an individual.16 Dikthe effect that in the case of terms structures, ntityas nstructionthat it is not ividuals :figure as saturating their open position). But ifindividual, what do FG operators bind, and what saturates the open posi-tion of restrictors? Dik (1989: 122, 171 footnote 8; 1997a: 140, footnote10) alternatively analyses individuals as singleton sets, but that does notrescue -standing theoretical entity: if -

125

member set (intensionally: a property), then, type-wise, n-guishable in such cases from the property variablesets. One of them at least is otiose, or, rather, both are (cf. ).The problem does not seem to have been clearly perceived, but the early

status of FG scholars, for,since then, they have tended to deprive it of its original referential import,to add to the NP new discourse-functional, referential, or pragmatic vari-ables ( 1992, 2002,and to keep nverting it into a g-matic -229;2004; this volume) and other recent F(D)G writings distinguish within theirunderlying NP structures the2002 i.e., Quality, Quantity, and Location, four in Rijkhoff this volume, byimplication more in Keizer 2004, Mackenzie 2004, and Hengeveld thisvolume, cf. ) from a higher rsonallevel in Hengeveld this volume). Correspond-ingly, aside from the term variable d-ditional variable, khoffor rsonal level.17

Of course, something equivalent to a referential variable is still neces-sary if NPs are to be referentially usable, but not in the representations ofthe sense of NPs (cf. Escribano 2004b and Keizer this volume). Dikof course, figured in such representations because it was meant to serve adouble (representational and referential) purpose, although for the reasonsdescribed it cannot play the former role well. If a new c-counts for the NP ntial potential, thenjustified, perhaps as a variable standing for linguistic entities (= senses).However, ifas with the logicianssame holds ifexpressions bind variables, to capture the fact that the NP corresponds tothe speakerones) to refer to a certain referent (cf. Rijkhoff 2002: 229). Anyway, treat-ing NP-choice variable issimply incoherent. In sum, Dikand inconsistent, and to the extent it remains in formulae down to the latestF(D)G approaches to term structure (e.g. Hengeveld this volume; Rijkhoffthis volume), the original inconsistencies remain unsolved.An early proposal to sever

Rijkhoff ferent

126

In the mature form of that proposal, cf. Rijkhoff (2002: 227-229), the termvariable ferential, a function now entrustedto a-tions: in some passages (e.g., Rijkhoff 2002: 229) it ranges over the set ofNPs available to the speaker as descriptions of the referent, whereas else-where (Rijkhoff 2002: 228)

via lexical orgrammatical restrictors.18 However, the term variable l-taneously a -choice variable, unless byunderstood just the sense of an NP. On the contrary, what a speaker uses torefer to a referent is not the sense of an NP, but the NP , i.e., alinguistic NP and its sense are categorially different things; if NPs NPs(i.e., linguistic expressions in their full individuality) are particulars usedby speakers to refer to entities of the world of discourse (or their mentalcorrelates), the entities in the range of i-ties from a different ontological domain.However, even if x-

actly the same problems the logicians(and similarly, Hengeveld terally stand either forthe sense of an entire NP or for a zero-sense entity. It cannot pick out fullNP senses, which obviously cannot replace icatesexpressed by nouns and their modifiers (i.e., the sense of an NP like

cannot become the , ,etc.), and it cannot stand for zero sense, either, for, if it does, by definitionit cannot denote anything at all, and in that case nothing is bound by theoperators or saturates the open position of the restrictors, and the wholeterm structure collapses. Observe that, in virtue of the correlation [zerointension] <> {universal extension}, a strictly property-lessout the whole Universe of Discourse (call it W for y-thing in W. However, in Rijkhoff (2002), as in F(D)G generally, the worldcontains much more than individuals, e.g., properties, events, predications,possible facts, speech acts, locations, times, manners, and what not (cf.Mackenzie 2004; Hengeveld this volume; Hengeveld and Mackenzie fc.),so, clearly, nnot range over the set W, either. Under present F(D)Gontology,have some content, but by no means that of a full NP: as stated above withreference to the metaphysician imallyentities with the ontological features that distinguish individuals from therest of the ontology,19 whatever they are, and no matter what its content is,as the underlying structure of an NP is gradually built, its value must be

127

). The result is that, finally, thevarious tokens of the term variable pick out sets that cannot be unified,which makes any conjunctive representation incoherent.20

Hengeveld esentational structure ofNPs still keep Dik -internal ativelyevent-denoting nouns, -denoting ones, etc., cf. Hengeveld2004: 372; Hengeveld and Mackenzie 2006: 674, Hengeveld this volume,Hengeveld and Mackenzie, fc.) as thecomplex) nominal head plus its optional modifiers. In other words, stillessentially Dik cture proposed for the referential subactin Hengeveld (2004: 372), in particular, is ( 1: [ 1 [ 1: LexemeN (f1) :

1))] (x1): 1))] (R1): 1)).21 However, on careful inspection, certain

differences emerge: whereas in Dikstraightforwardly saturated the open predicates of the restrictor and all thesatellites, and thus contributed in a transparent way to the determination ofthe NP cave-ats expressed above), Hengeveld ratethe head nominal in that formula, it is not an argument of the Lexeme, andit is unclear how it can be an argument of the whole head structure [ 1:Lexeme N (f1): 1))], which is already internally quantified and fully satu-rated. A parallel problem arises at the higher level, where the loosely addedreferential variable R1 is bound by a referential operator 1 (irrelevanthere) and restricted by a saturated NP [ 1 [ 1: Lexeme N (f1) : 1))](x1): 1))] and a modifier 1), but, again, although R1 is an argumentof nclear how it can fit into the argument-structure of the saturatednominal. On the other hand, it seems implausible to claim that both N andthe NP each take two arguments of different ontological types or even do-mains, i.e., in the case of N, a property ( d-ual (the value of small in the case of the NP, an individual (buthardly the same individual) and a speech act R. The problem has not beensolved even in Hengeveldvolume), where a head noun like or an adjective like in

each carry two arguments, an i

level operator and the term variable ix xi opera-

tor.Such difficulties are ignored in Hengeveld pproach, and variables are

treated as if they could be freely added to previously saturated structures,but the price to pay is high:compositional derivation of sense and reference potential, and, correspond-ingly, bel to the NP, but nothing in the sense or

128

reference of the NP constrains its range. Thus, the correspondence amongsense, reference, and use becomes essentially arbitrary.Perhaps as a reaction to criticism in Escribano (2004b), Rijkhoff (this

volume) has now removed some of the variables in Rijkhoff (2002), in-cluding all (tokens of)but one. If I do not misunderstand his proposal, in his new term formula

esents the (full) term variable, and stands for the refer-ent of the NP (a discourse entity/its mental representation), but, contrary totraditional FG practice, unexpectedly occurs only once per term, at the verycore, next to the noun. Obviously, this avoids one of my criticisms of Dikand Hengeveldused, its value, whatever it is, will be unique, and the overall structure ofthe term will not become inconsistent on account of different tokens ofrepresenting different sets or sense entities. However, ifterm variable and stands for the referent, its placement in the core of theterm structure, as an argument of the bare noun, is inexplicable, for it is notthe noun that denotes the referent. In that case,at the top of the representational formula, i.e., whereRijkhoff ccount.22

As shown, the tension provoked by the Janus-like status of Dik i-nal term variable has not yet been satisfactorily resolved within F(D)G.The fluctuation between Dik

concept of f-erencechoice, etc. That variables should proliferate in term representations, thus,is unsurprising, for in recent F(D)G work several non-isomorphic ontologi-cal spheres are invoked whose mutual relations need to be made more pre-cise, i.e.:1. The extra-mental world(s) endlessly shifting along the dimension

of Time (Phys-W e-vant in F(D)G (cf. Dik 1997a: 129; Mackenzie 2004; Hengeveldand Mackenzie fc.).

2. The different psychological world(s) of Speaker and Addressee(Psych-Wreference (cf. Dik 1989: 113, 140; 1997a: 129; Keizer 1991: 1;Rijkhoff 2002: 27, 336; Mackenzie 2004: 973, etc.).

3. The transitory discourse-generated world(s) successively negoti-ated by S and A at particular stages of discourse and, again, differ-ent from each other, and obviously distinct from the Phys-WPsych-W -W

129

4. The speech-act world (Sp-Act-W, hereafter), constituted by S, A,their referential and ascriptive acts, time, place, etc. And, finally,

5. The intralinguistic ( -Whereafter), which also differ across speakers (and are bound to dif-fer for each individual speaker along the time dimension, althoughlinguists tend to dismiss changes of sense occurring within shortspans of time as irrelevant).23

Thus, clarifying the nature of F(D)G denotation/reference and the onto-logical domain over which each of the various F(D)G variables rangesshould contribute to understanding which should be kept and which re-moved from the term formulae of the representational level. Letthis briefly.Starting with the variables attached at the top of NPs, Hengeveld

apparently ranges over events in Sp-Act-W (acts of reference). As to theclassical term variable -W, and individuals from Psych-W (or Psych-W as linguistically construed)is what Dik pposed to be bound by,cf. Dik (1997a: 129),24 Rijkhoff (2002: 336), and Keizer (this volume, on

ntal extension setsout individuals from Disc-W,25 in principle distinct from Psych-W(s), al-though, obviously, Disc-W and Psych-W may overlap (e.g., in internalmonologue), and Disc-W(s) will be internally represented for both S andA. Hengeveld2002) cannot range overis no reason to take it to range over s-sion does not reduce to its sense. If it is interpreted as a sense variablecomparable to -senses), itis a Ling-W entity; otherwise it could range over events of Phys-W (physi-cal NP occurrences, in whatever medium, i.e., sounds, inscriptions) or enti-ties of some other r-ther down, in therange over Ling-W entities (cf. Hengeveld this volume; Keizer this vol-ume). Rijkhoff -W(s) or Disc-W(s), and is said to belong to the descriptive dimension ofthe term, which suggests that it stands for aLing-W, like the th-order properties over which hisYet, Rijkhoff also says that egions withminimal properties (2002: 228, fn. 19, 229), and in that case it must be areferent in Phys-W, an ontological domain that FG traditionally considers

130

irrelevant, and theccount cannot be reconciled.

Significantly, in Rijkhoff (this volume), which contains a revised versionof term structure, ferent of the NP, presumablya Disc-W entity as psychologically represented in virtue of the nature oflinguistic construction. However, under that interpretation, why s-sociated with the core layer, a purelysaid.In sum: it is necessary to clarify the ontological domain of

other variables in F(D)G term representations, which means no less thandeciding what kind of semantics F(D)G wants, perhaps a long-term enter-prise, although, in my view, only immanentist theories will eventually becapable of accounting for what happens in natural languages. However, theimmediate problem with all extant F(D)G proposals concerning the under-lying structure of NPs is not just that the variables range over heterogene-ous domains, relevant at different analytical levels, and clutter representa-tions (cf. Escribano 2004b; Keizer this volume), but the fact that the logicalarchitecture of the formulae makes even those variables that should play arole in the computation of senses ornamental (cf. ). As a consequence,the NP sense that must help the speaker assign a value tothe NP suitable to play the R role in a speech act does not result from theintensional value of the various tokens of via simpleinheritance or any explicit compositional operation. In other words, even

relevant sense variables play no role in the computation of theintension of the NP, nor in the determination of its referential value orrange of use in speech situations, which must be stipulated by extra-compositional means.

2.3. Multiple variables/variable-tokens and the unification/update issue

Let us now turn to a different aspect of the problem of FG variables. Thestructure of a term like (1), represented in (2), contains several variables orvariable tokens whose mutual relations also need clarification. I illustratethe problem with Dikarise with alternativeat the head N, its expansions, and its satellites, that remain in RijkhoffKeizer and Mackenzie

131

(1)(2) dprox3xi: studentN (xi) : newA (xi) : {(d1xj: AmsterdamN)Source (xi) }

According to Dik (1997b: 30) the relation between the various instances ofaphoric

and has the same index value throughout), and he even introduces an ana-phoric operator icuous. That assump-tion faithfully echoes the logicians ndividual i-abledynamically constructed terms (cf. Dik 1997b: 31). Indeed, Dik (1989:116-119) had long before clearly stated that the structure of a term like (1)does not resolve into the simple conjunction of clauses that constitutes itspredicate logic representation. The difference is that natural language re-strictors are stacked, the construction of terms being a step-wise process(Dik speaks of 26

whose effect is that restrictors previously attached constrain possible sub-jects and the predicability of subsequent restrictors, cf. Dik (1989: 116-119, 127-128; 1997a: 150; 1997b: 31-33, 37). In fact, that is what the orderof the predicates separated by r-mulae.27 This, of course, has immediate consequences for the value of thepossibly various occurrence(s) of the term variableis very clear in this respect:be understood asand, discussing the predicability of new restrictors, Dik (1997a: 150) adds:

e-ries of restrictors placed on a term as construed so far which determines thepredicability of the next restrictor in the seriesconsistent with his idea of dynamic term construction, but it has a conse-quence that Dik apparently did not notice: each successive occurrence ofthei.e., it is a different variable.Which sets are represented by each variable is not entirely clear,

though. Interpreting Dik ements on the structure ofterms and the relative scope of operators and restrictors (cf. Dik 1989: 127-128, 138, 308; Dik 1997a: 132ff.; Dik 1997b: 23; cf. also Keizer 1991: 6,fn. 4) is tricky, for they are contradictory and the FG theory of scope is toovague (cf. ), but, if operators are themselves stacked (Dik 1989: 308)28

and restrictors (including the head noun, cf. Dik 1997a: 151) are attachedin the order implied by the ation, what Dik ulae claim aboutthe structure of a term like (2) might be more perspicuously represented as

132

in our preliminary tree (3), left-to-right order irrelevant, and paraphrased asin (4) (parallel to Dik

, cf. Dik 1997a: 161).

(3) Term [d prox 3xi: studentN (xi) : newA (xi) :{(d1xj: Amsterdam N) Source (xi) }]

xi {(d1xj: Amsterdam N) Source(xi) }

xi newA (xi)

xi studentN (xi)

d ?

prox ?

3 xi

(4)

Observe, though, that in that case an inconsistency arises, i.e., the variablexi can hardly remain identical to itself throughout the stepwise derivation,or the effects of stacking will be vacuous. At the bottom of tree (3), xiwould presumably stand for the universe of discourse (Psych-W, in Dikwork), or the subset of ualsthe operator applies, trios of such entities (whatever they are) aredenoted; after the operator applies, trios of such entities near thespeaker are denoted; after the definite operator applies, discourse/context-relevant proximate trios of such entities are denoted, etc. The first restric-tor, the head noun will then restrict that set of contextually rele-vant proximate trios of individual entities in Psych-W to relevant proxi-mate trios of students, then the second restrictor, will restrict that setto that of trios of , and in its turn, will re-strict the resulting set, etc., until no further modifiers remain. In otherwords, the range of referents of the presumed term variable xi must be p-dated nstruction of the term proceeds. If the variable at the toprepresents the set to which the full term is intended to refer, the lower

133

variables will stand for different supersets thereof which cannot be unifiedwith it or among themselves. Thus, Dik, the relation between occur-rences of ction.The same difficulty arises if esumably

is in Rijkhoff (2002): assuming it is not a iable, after all,but has some content, the lowest e-notations in Ling-W, i.e., type specified individualsopposition to properties, events and other kinds of entities), but asx ument of N and of Quantityand Location satellites (cf. Rijkhoff 1992: 191; 2002: 228), it, too, must bepdatedYet, obviously, if ations,

the effect is devastating: the formula ceases to be a coherent description ofan individual or set (for the same reason a metaphysicianto be a bundle if n-dle). Of course, what holds of smallvariables, and ultimately of all variables that recur in stacked modificationstructures within NPs, VPthe earliest FG theory of terms, and, at bottom, a consequence of the onto-logical incoherence ofindividuals, remains unsolved in all later F(D)G analyses (e.g., Keizer2004; Keizer this volume; Hengeveld 2004; Hengeveld this volume;Hengeveld and Mackenzie 2006: 674) except Rijkhoff (this volume),where occurrences of the term variableif that single occurrence is attached to the core noun, it will not pick outthe referent of the full term, but a superset thereof, i.e., all individuals ofthe type specified by the head noun.

2.4. The variable-binding issue

As argued above, in formulae of the representational component there is noreason to add variables unless they have to be quantified over, and they arenot quantified over when the predicates that carry them are saturated byappropriate arguments, so it is unclear why variables should figure, alongwith items that express their values, in F(D)G representations. However, tostick to FG n-bach iables, as Dik (1997a: 54 footnote) says, variables mustbe bound by operators term-internally,29 and this is not always the case in

134

current F(D)G term formulae. On the contrary, multiple variable-bindingviolations arise.To start at the bottom, the f

satellites ( 1-n) of Rijkhoff -229) formulae raise aproblem, for they are not bound by any operator (assuming that the 1, 2a,and 2b operators cannot simultaneously bind x i-cally disparate categories, a reasonable assumption). Although Rijkhoff(2002: 227-228, and fn. 18) treats Quality satellites as being of a highertype than the rest (i.e., second-order, in the basic case) and has them takethe value of the i nbound variablei 0: if the nominal predicate

is to project its properties in a compositional way to its higher expansions,the result at L1 must be the fusion of the two predicates (N, Quality satel-lite). Even if the variable of the satellite is schargedby the sense of the head N, thebound by an operator, and there is no other operator. This part of the prob-lem is not solved in Rijkhoff (this volume), where thebeen dispensed with in the higher layers, but thenoun remains, and remains unbound, cf. RijkhoffFurther binding violations arise with Rijkhoff

occurrences of ring the implications ofSTC, cf. ) as tokens of the same variables, respectively, as the repre-sentation suggests, biuniqueness issues arise, i.e.,by more than one operator each, an impossible configuration under stan-dard views on operator-variable relations; if, on the contrary,are cover symbols for sets of different variables with a-tions, as STC entails, then either each operator can bind several differentvariables, an offending configuration, or some variables remain unbound.The problem persists in recent work like Hengeveld (2004: 372),Hengeveld and Mackenzie (2006: 673-674), Hengeveld and Mackenzie(fc.). In Hengeveld (2004), for example, the Quality operator 1 must bindthe 1

1 iable at the modifier remains unbound, and the same applies tothe higher operators 1 and 1 with respect to the pairs of variables oftype 1

levels. This problem is not solved in Rijkhoff -tier model of terms, where operators no longer bind the term variable thatsurvives at the core noun. Rijkhoff proposes, instead, to analyze operatorsas predicates taking the remainder of their respective layers as their argu-ment, a solution I approve of (cf. Escribano 2004b, 2006), but still: that

135

leaves the surviving iable of the noun as unbound as the survivingrmula (45).

Needless to say, the variable-binding problem becomes more acute themore articulate the internal structure of terms grows. Keizer (2004: 15-16)reasonably recalls that, since nouns may denote entities of different order(zero, 1st, 2nd, 3rd, 4th), we must expect arguments and modifiers of different

term, I suggest, is just different r-ders(at least) the levels Rijkhoff (2002, this volume) postulates (in fact, rathermore), no term can have more than one type of variable at its core, i.e., aterm may denote an individual, or an event, or a speech act, etc., but notseveral of such entities at various structural levels; the parallelism withclauses breaks down here. However, the point is that Keizer pproachobviously multiplies thepresent F(D)G approach, the number and types of the variables awaitingsuitable operators to bind them. I leave to the interested reader the task ofspecifying how many variables will remain unbound under a more articu-lated NP structure like Keizer

2.5. Summary on variables

In short, leaving aside (as heterogeneous and intensionally irrelevant, inprinciple) big X iate) and

e top of NPs, the variables that figure in the lower, onto-logically homogeneous, layers of FG esentational formulae(theneither an ndividual ble, and each complex termmay well contain several

l-ues, some, including Rijkhoff riable, in-evitably unbound, as are some of thevolume) and Hengeveld (this volume) introduce and Keizervolume) implied higher-order sense variables. Furthermore, in principle,the results of stacking under the stepwise construction of the term cannotbe computed if variables are not pdatedother hand, the formulae become inconsistent if they are, as they no longerrepresent the intended entity. As a consequence, current F(D)G term repre-sentations are ultimately incoherent and fail to capture the compositionalrelations between different elements of term-structure or their respective

136

contribution to the determination of sense, reference and pragmatic ade-quacy. This is a serious flaw, for once the status of variables is shown to bequestionable, everything else in the F(D)G representation of terms is, i.e., itis unclear what operators bind, what restrictors restrict, or how senses areassembled and reference compositionally determined.

Another major aspect of F(D)G term representations is the set of a-tors i-fiers, cf. Dik (1989: 137, fn. 1; 1997a: 160 fn. 1), but, under close inspec-tion, they are rather unlike them, after all. As stated above, standardpredicate logic quantifiers are formula-level constituents, explicitly takescope over restrictors containing unique unbound variables, and (vacuousquantification ignored) bind them. The syntactic and semantic function andscope of FG operators, on the contrary, is rather less clear, which castsfurther doubt on the soundness of the FG formalism.The first difference between logical quantifiers and FG term operators

is that, as Dik himself observes (1989: 149; 1997a: 168), term operators arenot sentence-level constituents, but NP-level ones. Strictly speaking, there-fore, Frege FG, and thesemantic quirks of etc. must ultimately be captured under a modifieranalysis (e.g., that of Montague 1974 or Generalized Quantifier Theory; cf.Barwise and Cooper 1981).However, what is specially disturbing about the nature of FG operators

is that, as shown directly, it is often unclear how they fit into term struc-tures, whether they must have restrictors in their scope, whether they mustbind variables (for variables are often unbound, as shown), and, if theymust, whether they bind only one variable or are expected to bind more,even of different ontological types and domains, as implicit in currentF(D)G formulae (cf. ). Therefore, the conceptual problems surround-ing variables and operators intertwine and boost each other.Dik presents term operators primarily as elements of abstract term struc-

ture, but does not clearly define either their semantic function or their syn-tactic structure with respect to their operands in the structure of terms/NPs,respectively. In general, an FG operator is a grammatical function applyingto an -300), but it isclear that Dik did not consider operators predicates, since theyare not assumed to take arguments and are not listed in the Fund, cf. Dik

137

(1997a: 160-161). Apparently, in Dik eory, they are i-mately: predicates) applying to the term variable i-ously operated on by lower operators (cf. Dik 1989: 308; Keizer 1991: 6),as suggested in our structure (3) above, or ofby both lower operators and lower restrictors, under Diklayered account of terms.Of course, if operators are indeed abstract features, we need not expect

them to surface as free forms with identifiable syntactic functions (i.e.,they can as well be realized as morphological or lexical features on thehead noun), but in English and similar languages most do have overt expo-nents, and archetypal ones are discrete determiners or quantifiers in surfacesyntax, i.e., under the F(D)G analysis of NPs, noun dependents of somesort in the syntax. Rijkhoff (2002: 99) calls them mmatical modifiersin opposition to iers andMackenzie (2006: 671) further add that what makes them different fromother functional features is that they are non-relational (i.e., monadic)predicates, an analysis shared by Rijkhoff (this volume), where each opera-tor takes the next lower layer of structure as its argument.To my knowledge, Rijkhoff

far, but also a major departure from traditional FG doctrine on term opera-tors. It definitely distinguishes them from logical ones, makes them modi-fiers, and of course entails that they cannot bind variables. As a conse-quence, either the variables are removed (as claimed in Escribano 2004band here), or they will remain unbound, and Rijkhoff removes most ofthem, although not all, cf. .30

However, that is not the standard FG view of term operators. If we con-sider what has been written on term operators since Dik (1989), their broadrole asonat least two different views co-exist:31

a. In his formulae, and even more explicitly in his occasional para-phrases thereof, e.g., Dik (1989: 315-316; 1997a: 62, 161; 1997b:23, etc.), operators operate on the term-variable before any restric-tors (not even the head!) apply, which strictly corresponds to thederivation (3) above. However, Dik

b. Operators are distributed in hierarchical layers, with those at anyone layer taking scope over operators and satellites at lower layers,cf. Dik (1989: 138), Dik (1997a: 161).

138

The problem, of course, is that no single standard derivation, as repre-sented by a graph structure, can satisfy both (a) and (b), so it is necessaryto make a choice and adapt formulae to their paraphrases and graphic rep-resentations. Structure (3) above, in particular, satisfies clause (a) but vio-lates (b). To satisfy clause (b), the structure of a term like (2) should belike (5).The layered representation (5), in the spirit of Dik r-

lier parallel analysis of layered clause structure, corresponds to what istaken to be the intended view in Dik (1997a: 161), and is approximately theanalysis of terms assumed in current F(D)G work like Rijkhoff (2002: 218-228, 238, 314, 337, 341-343), Hengeveld (2004: 372), Hengeveld andMackenzie (2006: 671), Keizer (2004), Hengeveld and Mackenzie (fc.),and Rijkhoff (this volume), except in what concerns the number of layers,which varies somewhat (five, in Rijkhoff zer2004).

(5) Term [dprox3 xi: studentN (xi) : newA (xi) :{(d1xj: AmsterdamN) Source (xi) }]32

d (xi) ? {(d1xj: Amsterdam N) Source (xi) }(who came earlier, above mentioned)

prox (xi) ? {(d1xj: Amsterdam N) Source (xi) }

3 xi

newA(xi) xi ? {(d1xj: Amsterdam N) Source (xi) }

studentN (xi) xi

The conflict between (a) and (b) arises because, unfortunately, Dikformulae are vague as to the structure of operators and were not revised inDik (1997a, 1997b) to match his new layered account of terms. The vague-ness is due to the lack of explicit indications as to how operators fit intoterms. Observe that whereas the hierarchical structure of restrictors isclearly defined by c-tors contribute to the semantics of the term and their scope, cf. Dik (1997a:62), operators are not separated by colons in term formulae, but juxta-posed. In Dik (1997a: 181), the term formula for is

139

still (6), but all (6) suggests is that all the operators apply toof the restrictors does, as in derivation (3) above. Since Dik also says else-where (Dik 1989: 308) that operators are stacked and take scope over eachother, in derivation (3) they have been hierarchically ordered, but as a mat-ter of fact no hierarchical ordering follows from term formulae.33 The sameambiguity remains in the schematic formula of underlying term structurethat Dik (1997a: 163) offers, i.e., exactly (7), which, again, suggests thatthe higher operators (all except Quality) apply to the variablecomplex predicate?; observe that in (7) only Quant has a variable) beforethe restrictors do, i.e., essentially still clause (a) and derivation (3) above,rather than clause (b) and derivation (5).

(6) (d prox 3 xi: [N]: [A])(7) 2 -Loc 2 - Quant x : [ 1 - Qual pred [N] (args)]

Thus, even if Dik ight as to operatorsover the whole term is ignored as inconsistent with his pronouncementselsewhere, FG term formulae, his own paraphrases thereof, and the layeredview of terms do not quite match in what concerns what operators operateon or modify, and substantial ambiguity remains.Adopting clause (b) raises another problem: The y-

ers in all current F(D)G term structures is inconsistent with STC and en-tails that operators and satellites of the same layer have each other in theirscope, which, in its turn, causes an argument-binding violation.34 The firstpoint should be obvious: if STC holds, layers containing more than twoconstituents, say an operator, a head and a satellite, just cannot beand nothing prevents multiple satellites (or even multiple operators) withinany given layer. As to the second point, Rijkhoff (p.c.) rejects it and pre-fers to say that operators and satellites of the same level Ln take the sameargument (Layer Ln-1), as he claims in Rijkhoff (this volume). Yet, it isunclear how two constituents of the same structural layer that contractpredicate-argument relations with a third constituent can fail to take scopeover each other. If A is an argument of P1, where P1 is, say, an operator,P1 and A must be constructed with each other, and must be in each otherscope (i.e., sisters); if a second predicate P2 say, a satellite also takes A(and not [P1+A]) as its argument, then P2 must also be constructed with A,and a sister to A. Hence A is eventually constructed with both P1 and P2,and, since s-ters, and therefore in each other (D)Gnot mean what ic and all other kinds of linguis-

140

tics,35 or operators and satellites of the same layer are indeed in eachother (D)G itive relation, orjust not be based on 36

an asymmetric relation, but in that case F(D)G scope should be carefullystated and regulated, or constituent ordering predictions based on it will beuntestable, cf. section 4 . However, nothing in the F(D)G literaturesuggests that ething different in F(D)G from what itmeans in all other major theoretical approaches in the field.As stated, the problem also has an argument-binding side: A constituent

cannot, in general, be simultaneously an argument of two predicates P1 andP2. This is a standard assumption in most linguistic approaches that takesemantics seriously. The relevant constraint has different names in varioustheories ( , in Bresnan m-mar, the in Chomskyan Principles and Parameters The-ory/Minimalist Grammar, etc.) but the basic idea is very simple and can bestated in theory-neutral terms: argumenthood is a functional concept, andfunctions must yield a unique value; hence, as soon as an argument re-ceives a context-dependent semantic role (Zero, Experiencer, etc.) from apredicate P1, it cannot receive a second semantic role from a second predi-cate P2. It follows that if a layer Li is an argument of the satellite, it cannotsimultaneously be an argument of the operator, and vice-versa.37

The other worrying respect in which F(D)G term operators differ fromstandard logical ones is that, under either Dik(2002) account, whenever a term contains a different number of operatorsand variables, either the variables must be multiply bound, a non-standardproperty of variables, or operators need not bind variables, a non-standardproperty of operators. This is the case in both (3), and (5) above, but thetechnical problem is different depending on how variable occurrences areinterpreted. If different occurrences oftokens of the same variable, the variable is usually multiply bound (assum-ing operators are operators and must bind a variable) and the resultingstructure is incoherent; on the contrary, if they are taken to stand for differ-ent variables, some of them may remain unbound, along with all variables(e.g., (D)G term formulae simply do not pro-vide operators, cf. . Under Rijkhoff's (2002: 228) analysis, for exam-ple, the lower operators i

but the variable, like Dikand ends up multiply bound. On the contrary, as stated, theRijkhoff's (2002) formulae are all unbound unless his operators can eachbind two variables of different ontological types (i.e. c-

141

tively), an uncanonical assumption. Even in more recent work likeHengeveld (2004), the operators ch bind two variablescorresponding to non-identical sets (i.e.,modifier,head vs. 38 As to Rijkhoffmodel (Rijkhoff this volume), only two variables remain (core N), and they are both unbound.Such problems result from a) the heterogeneous content of term formu-

lae and consequent proliferation of extraneous variables in them, and b)their layered, but weakly integrated, architecture. Within current F(D)Gassumptions, the obvious way out of the vacuous quantification pitfall is tofurther refine the hierarchical structure of terms, distinguish more layerswith variables denoting intermediate entities (sets, ensembles, restrictedsets, quantified restricted sets, etc.), and allow each such layer to containits own operator binding its own variable, i.e., to expand current analyseslike Keizer (2004) or Rijkhoff (2002, this volume) by adding new binarybranching layers as determined by STC. Since, as argued here, differentoccurrences of(i.e., they are technically different variables), adjusting layers to STC willsurely leave variables unbound.To get them bound may seem trivially easy at first sight: FDG might

simply introduce, beside variables, exactly (ad hoc) operators to bi-uniquely bind them. However, even if the additional operators can be justi-fied within a richer type system, enforcing biuniqueness in operator-variable relations will not of itself yield an integrated term structure thatcan be compositionally computed. The problem, in a nutshell, is this: to theextent it could be maintained, Dik iablethat held the various semantic elements of the term together, but onceis shown to stand for different sets, the glue that held terms together van-ishes, and it is necessary to replace it with something. Piling up perfectlycanonical layers of operators, correctly bound variables, etc. on top of eachother does not help, for such layers contain predications with respect todifferent objects (sets), and no integrated structure results whose sense canbe compositionally calculated. If the current strategy is pursued, FDG willhave to develop some other device to hold the various constituents of termstructure together as parts of a unified semantic representation of someentity or set of entities.In sum, the real issue, at bottom, is that the computation of sense (and

reference potential) cannot be compositionally done as Dik assumed, sosome other means is needed, but piling up parallel and carefully symmetri-

142

cal layers of structure on top of one another or even endowing each ofthem with related variables and operators, predicates and arguments, satel-lites and heads, etc. will not make the slightest difference. What is needed,instead, is a more closely integrated representation of terms and a composi-tional mechanism that monotonically calculates the intensional contribu-tion of the various heads and modifiers involved. One way to achieve that,of course, is to treat current F(D)G term operators as one-place functions(i.e., predicates) directly taking distinct layers of the term as their argu-ment, with no variables involved, as proposed in Escribano (2004b) andRijkhoff (this volume), although, obviously, such a strategy represents asignificant departure from FG tradition, with wide-ranging implications atother levels (clauses, etc.).

4.1. CurrentConstruction

To the extent each layer Ln of term structure contains one (or more) opera-tors ( n1 n-1), and possibly one or more satellites( n1,..), an additional conceptual objection to current F(D)G doctrine issimply that such multiple branching abstract structures cannot even be builtunless Dik ant idea of ction1997a: 134-136; 1997b: 31ff, 37) is abandoned. According to STC, notonly operators, but also satellites, cannot just be flatly added as sisters toprevious structure, but must be stacked bottom-up in the order in whicheach makes its semantic contribution to the term.39 Dikproperly crystallized into an explicit FG principle, although it plays animportant role in all his discussions and paraphrases of how the sense ofterms is built and interpreted, but in my view he underestimated its impor-tance, for, if STC is observed, drastic consequences follow, e.g., only bi-nary-branching structures can be derived, functional uniqueness follows,new operators and restrictors automatically acquire asymmetric scope40

over previously attached ones, and, by definition, the structural depth (i.e.,the number of construction stages) of a term representation built out ofconstituents is = .41

In the case at hand, of course, STC leaves -ary branching layers out ofthe question. Observe that, if only two term constituents intervene (i.e.,operator and core,42 or core and satellite), no problem arises, but if the

143

elements to be constructed into a given layer are a core, an operator, and asatellite (the prototypical case under current F(D)G analyses of intra-layerconstituency), STC requires two of them to be attached to each other first,i.e., either [Op+Core], or [Core+Sat] (left-to-right order irrelevant) mustfirst become a new constituent, and the third element (Sat, Op, respec-tively) must wait until the next constructive operation allows it to get at-tached to the new core, i.e., under STC *[Op+Core+Sat] is an unreachableoutput; possible outcomes are either [[Op+Core]+Sat] or [Op+[Core+Sat]].To my knowledge, the STC idea has never been explicitly abandoned in

FG, but, in practice, Dik himself ignored it in his analysis of clauses andNPs, and all subsequent work on term and NP structure (e.g., the currentlayered theory) has been flagrantly inconsistent with it. I take it, then, thatSTC is no longer, , a principle of F(D)G, but this is an unfortunateaccident, for if only the implications of STC are systematically extracted,there follows a much more explicit and semantically well motivated ac-count of construction, constituency, domains, layers, scope, and surfaceconstituent order. Also, given STC, current F(D)G principles like

(DI, hereafter) and (HP, hereafter) can bestrengthened, particularly as to their hierarchical importance,43 many ap-parent violations of DI and HP are avoided, their systematic exceptionsexplained, and better empirical predictions follow. Even more, not only DIand HP, but also other descriptively useful, but non-categorical and onlyapproximately accurate FG

, and the core cases of, cf. Dik (1997a: 399-401) become redundant, since they follow as con-

sequences of more general principles, systematic exceptions to them can beexplained, and the resulting structures make the conceptual problemsabove easier to solve.On the contrary, if STC is ignored, as has unfortunately occurred in FG,

given constituents, possible outcomes of construction multiply, constitu-ents that are necessary at both the semantic and the syntactic level are notconstructed, only loosely integrated domains like NP and Clause tend to beconsidered, constituent ordering possibilities and DI and/or HP violationsproliferate, scope between constituents is symmetric where, under asym-metric scope, predictions would improve, etc. The negative consequencesaffect all aspects of semantics and syntax, including constituent order atsurface structure, and cast unnecessary doubts on the standard (and, in-deed, optimal) view that expression rules must operate subject to Iconicityon underlying semantic representations, cf. Rijkhoff (2002: 226, 253, 337-338, etc.).

144

4.2. Nice consequences of strengthening key F(D)G syntactic concepts

Unfortunately, although the concepts and constructive operationsused in F(D)G do not differ significantly from those in other major andexplicitly formalized linguistic theories, what exactly happens when twoelements A and B are constructed into a bigger element C has never beencarefully discussed in the F(D)G literature, to my knowledge, and the resultis an informal view of and loose concepts ofand which negatively affect the nature of the principles that F(D)Gcan formulate (preference rules) and unnecessarily weaken their empiricalpredictions. Such formal details, simply, have not attracted much attentionin F(D)G, but it is necessary to work them out, or rather it is a pity not to,so I will here, in a more constructive vein than in the preceding sections,try to spell out, in terms as theory-neutral as possible, what Dikreally entails given F(D)Gand must be reinstated to the place of honour it deserves among F(D)Gprinciples.We must start with the concept of . Needless to say, when

A and B are here said to yield [A+B] the symbolconvenient indication of the existence of a construction (dependence, etc.;a theory-neutral term intended) between A and B. As a matter of fact, theconstruction of A and B can hardly reduce to a set-theoretic operation likethe sum (or the product) of A and B ( Chomsky 2005; cf. Escribano2005b). Rather, all linguistic theories explicitly or tacitly agree that A mustbe functionally integrated into the structure of B (or ), and it is sointegrated to the extent it -neutral terms intended; readfeatures nder various names, what everybody sees occurring incases of Construction corresponds to the intuitive idea of ctionfunctional properties of various kinds (adicity, semantic role, inflection,etc.). Roughly, A may be constructed with B mainly for one of three rea-sons: 1) to satisfy Bof a certain type and assigns to it a certain semantic role, but that argumentremains unsatisfied, i.e., valueless, until A is attached to B); 2) to satisfyone of its own functional attributes, e.g., if A is a referring term, it will beinterpretable only if it receives a semantic role, so a valued semantic rolefeature is inherently expected, but the value of that feature44 (Agent, Bene-ficiary, Zero, emains unspecified until A is constructed with some headB; and, depending on how morphology is handled, another reason may be3) to license A atures (e.g., a NP in the Nominative,

145

Accusative, etc., or a verb form carrying certain inflections, fit only in con-struction with certain other heads or dependents, etc.), cf. Escribano(2005b).When appropriate items come to be in construction with each other,

their respective requirements are satisfied, and the resulting structuretogether and becomes fully interpretable in exactly the sense in which analgebraic formula can be fully evaluated only when all its variable symbolshave been valued. Again, terminology varies greatly, some theories willsay that construction rests on

ementshared by everybody, as far as I know. I try to use the terms that seem moretheory-neutral and either in (tacit) use in F(D)G or not likely to be resistedby F(D)G scholars. See Escribano (2004a, 2005b) for details on what Ireally think happens.It is that allows/drives the constructive process. Notice that

if, say, B is fully satisfied, that is, functionally saturated for all its attrib-utes (adicity, roles, etc.), there will be no functional slot available for (afeature of) A to fill into B, and A will simply not be constructible with Band will not be constructed with it. Observe, furthermore, that Satisfactionis a two-place relation, i.e., A satisfies B, not a 3- or -place relation (cf.*A satisfies B C), which entails that Satisfaction-based construction isinherently a binary process (like Dik ction, evenmultiple mutual Satisfaction, is allowed,45 but Satisfaction must be unidi-rectional or bi-directional, and involve just two participant structures A andB each time it occurs, although each participant may consist of a rich set ofattribute value pairs. Crucially, it is not possible for A to satisfy both B andC, or vice-versa, for B and C to both satisfy A. The former case wouldviolate functional uniqueness (in whatever respect, semantic, inflectional,etc., is relevant); as to the latter, Satisfaction will trigger two distinct con-structive operations, and one will necessarily occur first. As a conse-quence, two of the three participants (say, B and A) will be constructedinto a new object O, and C will eventually satisfy a feature of A or B onlyin the sense that all features of A and B will have become features of O,but, strictly speaking, C will be satisfying a feature of O at the point it isconstructed with O.In the case of term structures, assuming participants to be Op(erator),

Core, and Sat(ellite),46 Satisfaction (> STC) yields two possible outcomes:a) Op+[Core], and b) [Core]+Sat, where [Core] may correspond to either a)or b), i.e., successive Cores (1 omatic adjustmentof their senses and referent potential (in the direction of increasing inten-

146

sion and decreasing extension). *Op+Sat, *Op+Op, *Sat+Sat, *Core+Coreare all directly forbidden provided Ops and Sats are basically predicates( s, in the first case) and core segments are nominalsfunctioning as their arguments, as assumed in F(D)G analyses likeRijkhoff vial: informally, apredicate needs an argument, and will not contract a Satisfaction relationwith another predicate, so it will not be constructible/constructed with it;47

correspondingly, a Core nominal inside an NP functions as an argument (ofSats or Ops, cf. Escribano 2004a, 2005a, 2005b, 2006, and Rijkhoff thisvolume), and will not satisfy, or be satisfied by, another nominal.48

Furthermore, a natural, and stronger, concept of immediatelyresults: a s-faction relation, or, informally, a step of STC in which a successful Satis-faction relation is established, i.e., the noun with its first complement, orCorei and its Sat (e.g., AP, PP, Cl, etc.), form a , and so doOp and its Corej (e.g., Determiner and Corek, Quantifier and Corel, etc.).On the other hand, a full NP structure is not a strong domain, since manyitems in a complex NP may be completely unrelated from a semantic andsyntactic viewpoint (i.e., not in a Satisfaction relation). The advantage ofsuch stronger domains is their strict cohesiveness, i.e., by definition, noth-ing may intervene between Op and its Core, or Core and its Sat. If a con-stituent X occurs between Core and its Sat, X must be constructed witheither Core or Sat; if X is constructed with Core, [Core+X] will becomeaccessible to Sat, but Core itself will not, and since [Core+X] will not be ofthe same as [Core] (if it is a possible type at all), even though it be-came accessible, it would not trigger construction with Sat under Satisfac-tion; correspondingly, if X is constructed with Sat, [Sat+X] will be acces-sible to Core, but Sat itself will not, and [Sat+X] will anyway not be theright type of predicate to combine with Core. Thus, ( )

follows automatically, and since eak (= FGrecursive construction of strong domains, DI is redundant and need not beadded to the theory at all.Similarly, a new concept of relativized to different stages of con-

struction (= Core) can be defined, with further consequences in the area ofsurface constituent order. In fact, STC already entails that the noun is thehead of the NP only at the first stage of construction; as N is constructedwith its own arguments, satellites, or operators, the Head (= Core) becomesa gradually expanding nominal phrase. It follows that at any stage S in aSTC, the participants in construction are invariably a Core (acting as Head,in the revised sense) and just another element, Arg, Op or Sat. That re-

147

moves the anomaly that Keizer (2004: 8) observes concerning the fact thatonly terms contain multiple restrictors, a problem not solved by just re-labelling N as the only restrictor and the rest as satellites, as she proposes.If only Core and Op are involved, assuming that Ops are monadic

predicates taking the Core as their only argument, Op cannot but be adja-cent to Core, since arguments must be satisfied and, by assumption, Op hasonly one argument, cf. Rijkhoff (this volume). Observe, furthermore, that,since both Op and its Core now are head(s), HP automatically follows fromSatisfaction in this case.If it is only Sat that is constructed with Core, as the ultimate heads of

satellites may not be monadic, two cases must be distinguished. IfHead(Sat)49 is monadic (e.g., an adjective like ), it will only re-quire itsrequires. Of course a monadic Sat may carry its own satellite (e.g., an ad-verb, as in ), but in that case the internal satellite of Satwill never intervene between Core and its Sat. The reason is, again, Satis-faction as applied subject to , another common assumption acrosscontemporary linguistic theories, i.e., in the broadest theory-neutral ver-sion, that attributes are satisfied in a certain order as a consequence of their

. Of course, Priority is itself a corollary of the binary character of Satis-faction: if two features are to be satisfied, one musthand, Core is an argument of Sat, and arguments of heads must generallybe satisfied before the resulting phrase, in its turn, satisfies a satellite, or itwill be too late to satisfy them at all, as the types of the head as modifiedby the satellite and its still unsatisfied argument would no longer match.50

Hence, Head(Sat)+Core will form a continuous constituent, and both DIand HP, again, follow directly. Now, if Head(Sat) is polyadic, in principleit would seem possible for a non-subject argument of Sat to be satisfiedfirst and linearly intervene between Head(Sat) and Core, in violation of DIand HP, and yet, interestingly (in languages like English, Spanish, French,Italian, etc., at least) such cases never occur, cf.

(cf. Escri-bano 2004a, 2005b).The reason is that an independently necessary additional principle with

an tervenes to control the correspondence betweenhierarchical position and linear precedence (i.e., a possible FGprincipleformulation of that principle is Kayne 4)

(LCA, hereafter), which, in an informal, theory-neutral, version that

148

FDG may easily adopt, says simply that if A has asymmetric scope over B,A precedes B.In conjunction with the type of construction defended here, LCA plays

a crucial role in preserving HP in this case for reasons, again, related toSatisfaction. When a phrase (e.g., a Sat) is constructed with another (e.g., aCore), their respective internal heads may both previously have been con-structed with arguments (typically) or satellites. As a consequence, if boththe noun and its Sat initially had arguments to satisfy, their respective ar-guments must be adjacent to their heads and may separate Head(Core)from Head(Sat), in violation of HP. Yet, as stated, the Satisfaction of ar-guments is subject to priority, cf. Escribano (2004a, 2005a), and, in par-ticular, subjects are always satisfied last, hence the open slot of Sat willalways be the last argument of Head(Sat) to be satisfied. SinceSTC ( ttom upCore, being the subject of Sat, will always be hierarchically higher thanother Sat arguments. It follows that Core will invariably take asymmetricscope over Head(Sat) and all Sat er arguments, and that Core willprecede Sat and all SatThat reasoning predicts two possible orderings, i.e., Core+[Head(Sat)

+Comp(Sat)], or Core+[Comp(Sat)+Head(Sat)]. In the second case,Comp(Sat) will intervene between Core and Head(Sat) and HP will still beviolated. However, LCA applies systematically, to all phrases, and this hasparticularly nice empirical consequences in SVO languages like English:51

since Head(Sat) is itself hierarchically higher than, and takes asymmetricscope over, the constituents of Comp(Sat) (by definition, since it is a sisterof Comp(Sat)), Head(Sat) itself precedes its own complement. In sum:Core precedes Head(Sat), and Head(Sat) precedes complement of Sat. Theonly possible configuration, then, is Core+[Head(Sat)+Comp(Sat)], i.e.,Core precedes any Sat containing complements (PPs, Clauses, complexAPs, etc.), and Head(Sat) is adjacent to Core, which, again, saves HP (un-der the definition of head = Core).This captures the true side of HP, but HP has systematic exceptions, and

it is important to understand why sometimes HP just cannot hold. With

bound to enter into conflict with the higher-ranking principle of Satisfac-tion. For example, the head of the Core nominalcan never be adjacent to the head of the Sat in

, cf.,

,

149

, etc. The reason is that all such hypothetical possibilities violateSatisfaction under Priority and/or LCA once or more. For example, in thefirst case, the argument of and the first argument ofhave not been constructed with their proper heads, and, of course, the at-tempt to satisfy HP between and causes two HP viola-tions elsewhere. In the second example, LCA is violated, since hasasymmetric scope over and should precede it, which willcause the systematic HP violation cited. Finally, in the third example, it isthe complement of that has not been properly attached to itshead, in violation of Priority and Satisfaction (plus HP elsewhere). GrantedSatisfaction under Priority, the complement of should have beensatisfied before its subject, not the converse, which yields the idiomaticallycorrect solution (at the expense of a systematic violation of HP).However, HP is just what Satisfaction predicts when the participants are

lexical heads, and when they are not, HP holds or not depending on otherprinciples. In short, HP is not a t some-times holds and sometimes systematically does not, but the crucial point isthat, in either case, the right facts follow from the way thegeneral principles of Satisfaction, Priority, and the LCA interact. Thus, it isnot necessary to add to the grammar such ordering preferences. As a matterof fact, much the same can be said of nearly all the other FG general order-ing principles in e.g., Dik (1997a: 399-404). Why it is so should be obviousto the reader at this point, but let us briefly recall the trivial reasoning in-volved in the case of SVO languages; for SOV and VSO languages thereasoning is more complex, and depends on assumptions about displace-ment that F(D)G does not share, so they will be ignored here.

depends on semantic function, which depends on Con-struction under Satisfaction, i.e., a constituent C will have function F withrespect to another constituent C a-tor) to the extent it satisfies (arguments), or is satisfied (satellites), or bothsatisfies and is satisfied by (features of) Csubject to Priority, which fixes stage of attachment to the construct(roughly: first arguments, if any, according to their prominence, then satel-lites, according to the types they select, then operators; same sequenceapplied to all participants in construction), and stage of attachment deter-mines Scope, which, under LCA, in turn determines surface order.Similarly, has two sides, one vertical, the other

horizontal, or, informally speaking, nion-like -imageConstruction under Satisfaction, which fixes the relative scope ofdependents; the latter is an automatic consequence of the internal structure

150

the latter is an automatic consequence of the internal structure of the de-pendents themselves, which follows from the same principles and the LCAas applied inside the dependents. Briefly, if a Sat (of whatever scope) ismonadic, it will precede its Core and Head(Sat) will be at the right edge ofSat and closest to its subject (a case of HP above); if Head(Sat) is not mo-nadic, Sat will follow its Core, which is its subject, has asymmetric scopeover Head(Sat) and Comp(Sat), and, granted LCA, must precede them (i.e.,in the case at hand, complex APs, PPs, and Clauses will follow the Core;monadic adjectives and participles will precede the Core and be adjacent toit). Finally, on the vertical axis, Op will be where its scope dictates (i.e.,where Satisfaction licenses its construction with the Core), and, under thepresent analysis (cf. also Rijkhoff this volume), being a monadic predicate,it will always precede and be adjacent to the Core (HP) unless it is an affixand special morphological rules apply.As to ( , another preference

rule in FG, cf. Dik 1997a: 404), of course, insofar as it is a consequence ofright-branching, it follows neatly: non-branching operators and satellites(monadic predicates) will precede the Core and all right-branching satel-lites will follow it. However, increasing complexity correlates with seman-tic weight and the need to add information, and to that extent it does notentirely follow from sheer geometry: if two or more APs, PPs, or Clauses,etc. concur, our principles predict only that all will be postnominal;whether one or the other linearly precedes depends on its relative scope. Ofcourse scope is predictable, but there is no obvious correlation betweenscope and llabic word mayhave widest scope (e.g., an article), and a longish clause may belong nextto the head (a complement clause like in

). Thus, it is not reasonable toexpect LCA, etc. to predict Increasing Complexity in full.Finally, the orders due to also follow from Sat-

isfaction and LCA if it is assumed that a constituent may have additionalattributes to satisfy which may force it to participate in

elsewhere. E.g., if a constituent C (with its semantic role, inflec-tion, etc.) is additionally specified as Focus, Topic, (Subject), Relative,etc., the Satisfaction of such additional features may lead to processes offronting, inversion, topicalization, extraposition, etc. of C (theory-neutralterminology intended). Incidentally, that constitutes the only violation ofDI the present theory tolerates, i.e., systematic violations of DI due to s-placement eory-neutral term intended) triggered by Satisfaction. Thus,why ntirely explicable: a constituent C marked

151

as [D-F: Foc/Top], [Cl-F: Subj], [Cl-F: Rel],52 etc., is a Focus, Topic, Sub-ject or Relator constituent C e-quently, such features just may not be satisfiable until C beenbuilt (recall that construction works bottom-up according to STC) and be-comes available to C for the purposes of Satisfaction.Of course, the positions assigned to Focus, Subject, etc. may vary, and

displacement may not be involved in all cases (or the features may be real-ized by morphological means subject to other rules), but in some languagestopics, foci, etc. are syntactic constituents and displacement occurs. InEnglish, for example, Topic, Subject, and certain types of Focus (WH-Interrogative, Neg-Constituent Preposing, Adverb Preposing, etc.; theory-neutral versions intended) do involve fronting, whereas unmarked Focus isXP-final and does not. Can the present theory predict the position of Topic,Focus, Subject, etc. in English-like languages? Indeed: a constituent Cmarked as Topic, Focus, or Subject must satisfy its own attribute of topi-cality, focality, or subjecthood and also supply a value to the topic, focusor subject attribute of its clause, C n-valued [A: v] pair [Top(Ccan be constructed with C -branching structure, so C willimmediately acquire asymmetric scope over everything inside Cgranted LCA, will precede it. Thus, in English, topics, subjects, and suchfoci will occur at the left edge of their clauses. Since Rel is also a clausefeature, but one with wider scope than Subject, Focus, and Topic, a con-stituent marked [Cl-F: Rel] must attach to Cconsequence: XP[Rel] will immediately take asymmetric scope over every-thing inside C n-tirely correct, and the reasoning can be extended to account for comparablefacts in many other languages.Other theories invoke metaphors like Copying, or Movement in such

cases, and various other devices (Slash, Unification, etc.) are available inHPSG, LFG, Categorial Grammar, etc.). FDG must also provide some de-vice to account for displacement phenomena, be it at the abstract level or inthe expression module. I suggest that displacement must be a) a deep phe-nomenon, since it has semantic consequences, but b) with surface reflexes.As to how it should be conceived of, FDG will have its aesthetic prefer-ences, but in my view the appropriate concept is just (secondary) Construc-tion, i.e., ultimately Satisfaction. Finally, as to the technical details, mybroad suggestion is that phonological material be muted and realizedin the Topic, Focus, or Subject slot, where it can satisfy the clause-level

152

feature that supports the relevant semantic interpretation of theconstituent.Needless to say, this is a very quick and schematic account of how se-

mantic and syntactic computation work under Satisfaction, and empiricaldiscussion is necessarily scarce and limited to SVO languages, but theessential point is this: Satisfaction, as specified, entails Diktheories, including F(D)G, assume that arguments are hierarchically or-dered (= Priority, basic case), so all that it has been necessary to add towhat F(D)G tacitly assumes is LCA, or an expression rule to that effect. Inother words, an extremely simple and general machinery that comes for(almost) free to F(D)G will elegantly account for binary branching struc-ture, the integrity of domains, predicability constraints, functional unique-ness, head update, correct scope relations, and correct surface order, in-cluding FG general ordering preferences, to the extent they are correct, aswell as explanations for systematic exceptions to them. As to the semanticfront, appropriate semantic types must be defined for heads, arguments,successively richer cores, satellites, and operators/modifiers, but F(D)G,like all theories, has to do that anyway. The advantages of this approach,on the other hand, are substantial: no variables are involved in sense repre-sentations, no variable-binding problems arise, no vacuous quantificationresults, different intensions are automatically assigned for each core nomi-nal as well as the topmost NP, correct scope follows, and, crucially, thevarious levels of term construction will be functionally integrated in thestrongest possible way, predicate-argument relations, and the sense andreference potential of the term will be compositionally determinable.All that follows from STC (<Satisfaction) once it is properly spelled out

as to what it presupposes and entails and its formal details are worked out;what has been done here is just to fill in the gaps and systematically extractthe consequences of a particular interpretation of STC that is perfectlycompatible with F(D)G doctrine in all essential respects, in my view. Onthe contrary, abandoning Dikallowed F(D)G doctrines that cause, severe semantic and syntactic prob-lems. I have already extensively referred to the former (cf. discussion ofthe inconsistency of variables, the variable-update issue, variable-bindingproblems, functional uniqueness, and general semantic disintegrationabove).As to the syntactic problems, of course, in languages like English, loose

multiply branching structures are blatantly incompatible with the results ofall standard syntactic tests (e.g., coordination, zeugma, substitution by pro-forms, anaphora, mobility, etc.), and adopting them simply amounts to

153

giving up on a precise formulation and explanation of processes like coor-dination, substitution, anaphora, ellipsis, processes resting on asymmetricscope (binding of reflexives, interpretation of pronouns, etc.), predicability,the fine details of constituent order and its structural motivations, and ex-plicit accounts of syntactic and semantic composition, in general. In thecase at hand, the flat layers predicted inside F(D)G NPs do not correspondto the structure that standard tests indicate exists, and make rules like NP-internal Coordination, One-Pronominalization, or NP-internal Ellipsis(theory-neutral interpretations intended) unstatable, apart from obscuringthe transparent iconic relation that exists between the adicity and selectionproperties of lexical predicates and their patterns of occurrence in NPstructures.Arguing and illustrating all that in detail would entail repeating well

known textbook level facts and require much more space than is availablehere, and, on the other hand, sadly, F(D)G is not overly concerned withsurface constituency, so I will in closing refer only to an aspect of NPstructure with which the fullest F(D)G account of NPs to date, Rijkhoff(2002), is deeply concerned, i.e., the surface order inside NPs. My mainclaim here will be that adopting DI, HP, and other standard FG principlesunnecessarily diminishes the accuracy of the predictions that accounts likeRijkhoff glish NPs.53

4.3. Unnecessary weakness of current F(D)G predictions on NP-internalsurface order

One of the crucial concepts in Rijkhoff -internal order is, which in turn supports (DI). Rijkhoff

248-253) domains coincide with NP and clause, but NPs are too big andloose structures. A single NP may contain multiple elements that bear nosemantic or syntactic relation to each other, and as a consequence DI isviolated more often than it would be if domain were equivalent to the set ofconstituents in a given unique relation of Satisfaction. Observe that evenelements of the nucleus of an NP, like two arguments of the same N, do notcontract any semantic or syntactic relations with each other. Hence eventhe strictto constitute a useful domain for the purposes of, say, effectively constrain-ing surface order. Of course, beyond the nucleus, the n-ner: a determiner or a modifier and an argument of the noun are unlikely tohave any connection whatsoever, and the same holds of quantifiers and

154

relative clauses, modifiers of different levels, operators, etc. As a conse-quence of this loose concept of Domain, Rijkhoff, like Dik (1997a: 402),must state DI in a weak form, as another tendency, cf. Rijkhoff (2002: 250)and add stipulations such as that embedded domains h-eral and not intervene between A and N, for example. On the contrary, ifour concept of domain above (= construction, based on Satisfaction) isadopted, DI is a categorical fact, although it need not be stated as a specificprinciple, since it follows from Satisfaction, and, in fact, theof embedded domains needs no special provision, either, as, to the extentthey are descriptively correct, both follow from the way Satisfaction andLCA work, as shown above, cf. Escribano (2004a, 2005a) for details(within a different theoretical framework).As a consequence of his flat view of layers, Rijkhoff

principle, another key factor in his account of NP-internal order, isalso much too weak, unsurprisingly, since it is a structure-dependent rela-tion and must rest on F(D)Gthe version that applies to satellites (a particular case, presumably),Rijkhoff states it as follows:the expression that they have in their scopemoderate content54 when only one modifier is present, but the fact thatmodifier and modified will be adjacent follows from Satisfaction, anyway(recall that the modified is an argument of the modifier), and, if there aremore modifiers, Rijkhoff ecomes vacuous.Rijkhoff (2002) does not discuss adjective order, but, just as an exercise

in theory-checking, think of alike Sat1, Sat2, Sat3, and Core, e.g., . All four are ineach other ), so Rijkhoffnothing useful about their respective order. It is compatible with resultslike, say,

* , etc., as well as thedesired .55 On the contrary, a process of constructionunder Satisfaction, as above, would enforce a binary branching structure,Sat1, Sat2, Sat3 and Core would contract asymmetric scope relations, andsuch ordering ambiguities would disappear, i.e., if, assuming delicate typ-ing of adjectives, a structure [ +[ +[ + ]]]] results fromlexical selection specifications and Satisfaction, clearly or willnever occur between e.g. and , and, granted LCA, outputs like* or * will just never be gener-ated. Observe that has asymmetric scope over [ ], and mustprecede them, just as has asymmetric scope over [ +[ ]]

155

and must precede them. In sum, there is much to be gained from adoptingSatisfaction, strong Domain, and asymmetric Scope.The third crucial factor of Rijkhoff -internal order is

, which, again, Rijkhoff (2002: 263-264) formulatesweakly, as just another tendency (cf. also Dik 1997b: 402; Rijkhoff 2002:264-290). This may be descriptively useful, but is theoretically inadequatein two respects: First, it does not give HP its due, predicting many more HPviolations than actually occur, and second, it does not explain why HPsystematically does not hold in certain cases. The first point is obvious: iflayers are flat, HP is bound to suffer multiple violations; for instance, if anoun is simply preceded (or followed) by two modifiers (two As, two Ops,an A and an Op, two PPs, two Rel Clauses, etc.) in the same layer, and thestructure isone of the modifiers and N will always violate HP. On the contrary, if sucha nominal is built under Satisfaction, the relativized concept of Head,strong Domains, asymmetric scope, etc., the structures will be binary-branching, e.g., [Mod1+[COREMod2+[COREN]]], [[COREMod2+[COREN]]+Mod1], or [[[COREN]+Mod2]+Mod1], etc., order subject to LCA, asabove, the relevant heads Mod2 and Core and Mod1 and Core will alwaysbe adjacent, and no HP violation will arise.56

Other statements by Rijkhoff in relation to HP, i.e., that the head N is asclose as possible to the head V of the clause, or his first HP-derived (nega-tive) prediction that the preferred position of any adnominal modifier57 isnot between the head noun and the head of the clause (Rijkhoff 2002: 263,290), are also inaccurate/incomplete and inferior to what Satisfaction, Pri-ority and LCA predict. Observe that, if correct, in SVO languages likeEnglish, NP objects carrying pre-modifiers (cf.

) and NP subjects containing post-modifiers (cf.) would both be non-preferred options, whereas, in

fact, they are not only perfectly normal, but the only options available,even if the post-modifiers are long branching APs, PPs or Rel Clauses. Ingeneral, in SVO languages like English we would expect modifiers to sys-tematically occur on opposite sides of the core depending on the NPfunction and distribution with respect to the higher head, i.e., if the corenoun is the head of a subject, its accompanying modifiers should all be pre-nominal, whereas if it is the head of an object, they should all be post-nominal, etc. Obviously, that is not the case, either in English or in relatedSVO languages, so a stronger statement is justified: there are perfectlysystematic circumstances in which an adnominal modifier of a subject(e.g., a PP or Clause) must be exactly in Rijkhoff pposedly non-

156

preferred position, between the head of the subject and the verb, and thereare circumstances in which a satellite cannot be there, etc.Rijkhoff -derived PHP1, i.e., that attributive adjectives are

expected next to the noun, sounds sensible, and yet, under scrutiny, some-times it does not hold, and when it does it follows automatically from moregeneral principles. The latter needs no comment: if the noun is an argumentof the adjective, it must be constructed with it under Satisfaction. Moreinteresting is why PHP1 is systematically false sometimes, i.e., preciselywhenever the noun ctiveexample, granted Satisfaction and Priority one or more arguments of thenoun may have to be satisfied (cf. ) before any adjectivalmodifier of the NP can in its turn be satisfied by taking the resulting NP asits subject. If the adjective itself is monadic, it will be constructed with thatNP and will not only be adjacent to it, but, in English, granted LCA, it willnecessarily precede it (cf. , vs.

);58 however, if the adjective is not monadic (cf.), adjacency between the two heads is impossible, in violation

of HP (which does not predict why this is so), i.e., the only possible outputis , cf. *

, *, etc. Now, if PHP1 were an option open in all cases, it might make

sense to include it in the grammar, but it is not, and with the right sort oftheory we know why, so mentioning incomplete/inaccurate generalizationslike PHP1 (or indeed HP itself, for the same reasons) is misleading andunnecessary.Actually, there is more evidence that PHP1 is just a false/incomplete

generalization and that more abstract and subtle principles than HP areinvolved. For example, HP/PHP1, like Satisfaction LCA, predict

, but the unacceptableshould in Rijkhoff

violations. Observe that the two heads are adjacent, as HP demands, andthat the embedded NP domain is at the periphery, as DI and IncreasingComplexity/Peripherality recommend. Of course, under the F(D)G analy-sis, the head of is not yet adjacent to , against HP, sothere is no other way to satisfy both HP demands but the configuration

, which is exactly what Satisfaction, Priority andLCA also predict will happen. However, isnot just mildly deviant, but unacceptable, so the violation is not a tolerableone in that case. Now, since what causes intolerable unacceptability in

is the same kind of

157

occurs in , the latter should be as marginal as the for-mer, but as a matter of fact it is not: the isquite acceptable in spite of its obvious HP violation. Of course, whatcauses the contrast between

and has little to do with HP/PHP1, andmuch to do with the adjectivesmust be satisfied, cf. Escribano (2005a).Under the layered account of NPs, in short, HP counterexamples are

bound to proliferate, and Rijkhoff (2002: 290, 308) must add thatones (e.g., those caused by the piling up of non-branching APs) are toler-ated better than large ones (those arising when embedded domains such asNPs or clauses intervene). Well, other reasons (LCA, etc.) prevent the em-bedding of XPs, as we saw, but, leaving that aside, the fact is that somesuchamong adjectives of size, colour, etc. which Rijkhoff does not discuss havealready been mentioned in this respect, so I will cite just a more interestingcase that suggests a necessary revision of the status of prepositions in FG.PPs like in orin , being structurally as simple as adjec-tives in underlying FG representations59 are under Rijkhoffpredicted to cause only the same ations as multiple adjec-tives cause. However, at least in English and similar languages, this is,interestingly, wrong when the violation occurs to the left of the noun: Inpre-nominal position, an intervening surface PP is always much worse thanan intervening noun or adjective on the left hand side of an NP. In fact,aside from idioms like in

, PPs are completely impossible in pre-nominal position, cf.

; see Escribano (2004a). Unless a bare noun with a role featurequalifies as an such examples can somehow beexcluded by DI, a dubious possibility, it follows that expressionprinciples would have to be added to HP, DI, etc., to allow someHP violations (those involving bare adjectives and nouns, cf. in

) and block others.60

Observe, though, that this problem arises only as a specific consequenceof the FG view of prepositions. Under broadly similar approaches to FGlike Hawkins n terms of an EarlyImmediate Constituents principle (EIC; omain2001), it does not, as prepositions are there all along and count for theevaluation of internal complexity. The same can be said of our approach

158

here: if , , etc., are the result of combining adyadic predicate , with its first argument to yield a monadic restric-tor, both the semantics and the syntax of such satellites follow nicely fromSatisfaction and LCA, i.e., still needs a subject (e.g.,

), and the subject will have asymmetric scope over and its inter-nal argument , hence, granted LCA, will also have toprecede and such deviant examples will be neatly ex-cluded. What this solution entails is an analytical distinction betweenfreely selectable prepositions like , on the one hand, and governedones (e.g., in , etc.) or bare functionalin , etc. The latter may arise from late shallow

features and expression rules, but since the former clearly add semanticcontent and their own arguments, they must be treated as dyadic predicatesto all purposes (i.e., they must be available in the Fund). This is a smalland well motivated modification which immediately brings a surprisingnumber of facts under the maximally general and elegant principles of Sat-isfaction and LCA, cf. Escribano (2004a).

In sum, the relatively poor performance of current F(D)G principles atpredicting and explaining the fine details of NP-internal order is but an-other consequence of the general looseness of F(D)G concepts that ourprevious discussion of variables, operators, scope, and NP semantics hadalready revealed. That looseness not only endangers the status of operators,variables, and other aspects of abstract structure, as shown above, but alsolargely invalidates F(D)G(2002: 313-335, 346) otherwise correct claim that important surface order-ing facts follow from Scope. They certainly do, as we have tried to prove,if Scope is explicitly formulated. Unfortunately, F(D)Ghas not been one of tightening its formalization and exploiting the interac-tion of semantic and syntactic principles to simplify the grammar. On thecontrary, Hengeveld has now apparently abandoned the idea of derivingorder from meaning through structure and Scope. In Hengeveld andMackenzie (2006: 675) order is said to follow, not from meaning, but from

rpho-syntactic principles(2004: 45-46) the order of head and modifier is said to be fixed either bysome templatic constraint, or by special indicators of the modification rela-tion. Such an approach to surface order amounts to saying that it is not

159

subject to any more general principles than the low-level idiosyncrasies ofindividual constructions, but, unless what has been said above about Satis-faction, Priority and LCA is completely off the mark, that is an unnecessar-ily weak assumption. On the contrary, a precise theory of construction,structure, and scope, coupled with the optimal assumption of iconicitybetween abstract and visible structures, as regulated by LCA, yields mostof surface order (cf. Kayne 1994; Ernst 2002; Escribano 2004a; 2005a).Obviously, this comparison of the Satisfaction-based account and the

standard F(D)G one in the area of NP-internal order is not conclusive evi-dence that the former is neatly superior . Before accepting such aconclusion, it would be necessary to compare the performance of the twoapproaches with respect to the rich cross-linguistic body of order facts ine.g. Rijkhoff (2002). However, an important difference should be under-lined: in the Satisfaction-based solution, all surface order predictions aredirectly obtained as a result of elegant interactions between principles likeSatisfaction, Priority, or the iconicity constraint LCA which also control allother aspects of construction. Granted such powerful principles, phenom-ena like DI or HP, to the extent they hold, clearly appear as mere conse-quences, or incomplete generalizations, which may be useful from a de-scriptive point of view, but fail to identify the real forces at play and lackexplanatory power. Once the stronger principles are revealed, there is nopoint, in my view, in proposing intuitive principles that are known to besubject to categorical and completely predictable exceptions.Most of the problems the F(D)G account of NPs faces ultimately result

from a vague theory of Construction. If Construction is made to depend onstrong (and, at bottom, cross-theoretically assumed) principles like Satis-faction subject to Priority and rigid (iconic) correspondence between struc-ture and linear order (Kayne a-tion between different elements of term structure are enforced, shallowdifferences like that between complementation and modification dissolve,cf. Escribano (2004a), correct constituency based on unambiguous binary-branching structures and rigid scope arises directly, FG principles likeIconic Ordering, Centripetal Orientation, Domain Integrity, Head Prox-imity and Increasing Complexity or even Pragmatic Highlighting (cf. Dik1997a: 399-404, and Rijkhoff 2002) all follow as consequences of interac-tions among such more powerful forces, and their limits and systematicexceptions are elegantly explained, at least in English-like languages.In the case of NP structure, the functional layers assumed in current

F(D)G can only arise if Dik principles areignored, and are not likely to be correct. Although they constitute a good

160

first approximation to the most important functions of various tracts of NPconstruction and allow a moderately successful account of the surface or-der of determiners, quantifiers, heads, complements, and modifiers, thefine details do not quite fit. The higher referential and discourse layersprobably do not exist as such, cf. footnote 16, (although NPs surely havesuch functions), whereas, on the contrary, many other layers (= construc-tions = domains = cores) are very likely to exist, for rather more delicatedistinctions are needed both inside the nucleus of NPs and also above it, inthe broad area of modification, cf. Keizer (2004) and Rijkhoff (this vol-ume) for recent F(D)G work in that direction, and e.g., Crisma (1993),Cinque (1994), Bernstein (1997), Longobardi (2001), Giusti (2002), Scott(2002), for relevant fine details from several well-studied languages. Infact, aside from conceptual considerations (only binary-branching struc-tures may result from Satisfaction), there is substantial empirical evidence,derived from constituency tests, binding facts, surface order, and semanticfacts, that points towards that kind of structure in NPs as well as clauses.For reasons of space, such evidence has not been cited here againstF(D)G and theo-retical discussions of constituency (cf. e.g. Pesetsky 1995).Certain minor syntactic details may also need rethinking in F(D)G, e.g.,

the status of prepositions and relative operators. In my view, very niceconsequences follow from assuming that prepositions with semantic importare dyadic predicates, with their complement and their(here, a nominal; in the case of adverbial PPs, a verbal phrase), cf. Escri-bano (2004a). As to relative operators, most of their properties can be co-herently derived from the assumption that Rel is also a dyadic predicatewith a special capacity to attract a clause-internal constituent, and that Rel(not the verb!) is the head of the Rel Clause. For one advantage, under suchan analysis,core nominal follows, and, particularly, the order of relative clauses withrespect to other NP constituents in English-like languages follows nicelyfrom scope and LCA, cf. Escribano (2003; 2004a).The major focus of attention in this article, though, has been placed on

semantic issues and problems derived from the representation of NP mean-ings in current F(D)G, mainly two, a) the ontological status of what ismeant, including the question of what symbols belong/do not belong inunderlying NP representations, and b) the overall architecture of term for-mulae. As to the former, F(D)G seems to operate with a broad concept of

hacts, which makes F(D)G term representations very heterogeneous. Of

161

course, it is perfectly legitimate to design models of speech situations withvarious slots for participants and relations in them and treat linguistic ex-pressions as values of appropriate variables in such models, or to connectNP senses with referents in e.g., models of Psych-W or Disc-W, but thetheory must first show how the senses of NPs are compositionally derived,and how they constrain referential and use potential. Thus, in my view, it isnecessary to be less ambitious and start from a more constrained concept of

61 NPs are used by speakers to refer to extralinguisticentities, but what they can name or describe is just entities of our Ling-W,i.e., senses.62 Therefore, only (symbols standing for) senses should figurein the representations on which the meaning = sense of NPs must be calcu-lated (cf. Escribano 2004b; Keizer this volume). That excludes

X icate variablethat Dik (1989: 50) envisaged for lexical items. What remains is a) thesenses of nouns of various types (individuals, substances, events, etc.) andadicities (properties, relations), b) the senses of their arguments, also ofvarious types (individuals, events, propositions) and adicities, and c) thesenses of their modifiers (properties, also of different adicities and th-order). The senses of predicates and arguments are composed by

, those of modifiers and their modifieds, i.e., higher-order predicates and the predicates that saturate them as their subjects, by

, both essentially the same operation based onSatisfaction of sense-related properties (Types).Operators are nth-order predicates, and no variables need be involved in

the representation of senses,63 since, by definition, a variable appears in-stead of a constant. Thus, the relation and composition of Ops and Cores isessentially the same as that of Sats and Cores, cf. Escribano (2004b, 2006),but a Sat and an Op never apply to the same core ( Rijkhoff thisvolume and Hengeveld this volume). Under Satisfaction, the richness ofthe hierarchical structure inside NPs directly entails a highly differentiatedstructure in what concerns semantic Types, which has not yet been elabo-rated. As a real compositional account of NP semantics and syntax isworked out, new, more delicate semantic Types will emerge to be added toexisting ones like ualsproperties of the latter two), cf. Pustejovsky (1995) for one such system,but that does not entail, and is not likely to determine, enrichments of themachinery in charge of construction (i.e., Satisfaction) and the monotoniccomputation of senses (Functional Application, Functional Composition),which is powerful enough, especially if semantic Co-Composition, cf.Pustejovsky (1995), is allowed along with (multiple) mutual satisfaction.

162

Of course, what specific types should be assigned to N and all the nominalcores above N is a matter that we cannot settle here, as it entails decisionson complex issues connected with morphological derivation and other mat-ters, but, whatever semantic types are chosen, no variables need be in-volved in sense representations.Needless to say, this quick overview just summarizes our previous di-

agnosis of conceptual problems in the area of term structure and adds bareguidelines, not detailed solutions to the substantive empirical issues thatarise, e.g., Which Types (possibly different from metaphysical types of

-Shifting operations are relevant in natural lan-guage semantics? However, it hopefully will have shown that a clearerview of the implications of Satisfaction, rigid concepts of Construction,Domain and Scope, and a universal? expression principle like KayneLCA that enforces a tight correspondence between structure and surfaceorder are a prerequisite to any adequate account of NP form and sense. Infact, Dikcorrect, programmatic statement anticipating what is formally proposedhere, and ideas like Satisfaction and Priority are already assumedin F(D)G. Also, since F(D)G has also already developed a rich linguisticontology (individuals, masses, ensembles, properties, events, etc.), furtherenrichment of its Type system, as proposed here, is a matter of degree, nota deep modification, and should not raise special difficulties. On the otherhand, it is true that the changes defended here would make F(D)G verydifferent from what it is now as to the details, more rigid, and more vulner-able to empirical refutation, which is as should be from a methodologicalviewpoint.As to the critical tone that has prevailed in this paper, in conclusion, al-

though in its present form the F(D)G treatment of underlying terms andsurface NPs is seriously inconsistent in various respects, the problems areunderstandable, largely accidental (inherited from certain types of philoso-phical semantics) and by no means inherent to the theory, whose basicprinciples seem to me correct and fertile. F(D)G should be further devel-oped into a proper account of the way natural languages work, semanticallyand syntactically. For this reason, the ultimate purpose of this contributionhas been critical, but in a well-meaning, constructive sense, i.e., to helpF(D)G scholars identify and perhaps solve some of their more pressingtheoretical problems, acting from the vantage point of a sympathetic out-sider with a broad view of linguistic theory, absolute academic freedom,and no axe to grind.

163

1. This work owes much to Daniel Garc , who facilitated my access tosome crucial sources, read several versions of the text, and gave me wise ad-vice on how to write a delicate piece like this, to Jan Rijkhoff, who, as an au-thor, reacted very elegantly to my objections, and, as editor of this volume, of-fered very detailed criticism and suggestions, and to three other anonymousreferees who discovered various obscurities and other infelicities in earlierversions that I hope have been avoided in this. I also thank Lachlan Mackenziefor quickly sending to me an offprint of an article of his that I needed. The un-derlying semantic research has been sponsored by the Spanish Ministry ofEducation under contract I+D MEC-04-HUM2004-018.

2. Interestingly, operators do not have them, which implies that they are notsemantic categories under the authorsfrom the fact that FG operators are considered grammatical modifiers. Yet,operators are in the language systemconsequences. This view of Hengeveld and Mackenzie e-lian concept of bjects and predicates), but leaves operatorsin a semantic limbo.

3. That is, those biuniquely bound by operators, cf. .4. Assuming, in the spirit of Hengeveld and Mackenzie above (cf. se-

mantic categories = ontological categories as reflected in the language sys-tem), that at the representational level F(D)G semantics is, like all linguisticsemantics, concerned with the sense of expressions, and that reference (to enti-ties of psychological worlds in the case of FG, cf. Dik 1997a: 129) is, alongstrictly Fregean lines, a function of linguistic sense plus additional parameters.

5. Actually, a logical quantifier like ( x) is conventionally assumed to vacuouslyapply also to formulae like ( z) P(z) that contain no unbound variables, andeven to formulae like P(j) containing no variables at all. Thus, in PredicateLogic, ( x) [( z) P(z)] and ( x) P(j) are not syntactically ill-formed, althoughthe quantifier is semantically inert in them, but the motivation has to do witheconomy of statement of the syntax of Predicate Calculus. Natural languages,of course, radically disallow vacuous quantification (cf.

).6. Properly, predicates of higher adicity are often involved, but the various modi-

fiers around a noun all behave as one-place predicates at the stage they are at-tached to their respective heads. Obviously, the head noun may itself be dy-adic or even triadic (e.g., , respectively) and modifiers may beheaded by dyadic adjectives (e.g., , as in ), dyadicprepositions, participles of any adicity, etc.

7. The term n-tory (or closed system) of types of entities (e.g., individuals, substances,events, etc.) that can be linguistically named or described.

164

8. Such terminology is standard in logic textbooks and in all Fregean approachesto linguistic semantics, e.g., Montague Grammar, Categorial Grammar(s),LFG, GPSG-HPSG, and broadlyor Jackendoff (1990). McGee Wood (1993), Steedman (1993) and Heim andKratzer (1998) contain clear and easy accounts for linguists.

9. E.g., representing as ( e) [Meeting (e) & Agent (e,b) & Theme (e, q) & At (e, h) & Time (e, < now) ], or

as ( e1) [Seeing (e1) & Experiencer (e1, t) & Theme (e1, b) &Time (e1, t) ] ( e2) [Reading (e2) & Agent (e2, b) & Time (e2, t)], etc.

10. Incidentally, a variable which is subject to exactly the same metaphysicaldifficulties posed byalso affected).

11. Actually, on the negative side, Davidson o induced generativegrammarians to counter-intuitively treat all constituents of the clause (i.e., thehead predicate, the subject, objects, adjuncts, tense affixes, etc.) as predicatesof events, suppressing the important differences between heads, arguments,and modifiers. Although any clause constituent adds its own truth conditionsand to that extent can be thought of as a fferent status ofheads, complements and adjuncts has wide-ranging syntactic and semanticconsequences in natural languages and cannot be ignored in any structure-dependent account of senses. The problem is particularly acute in currentChomskyan linguistics as understood by Kayne and Cinque, cf. Escribano(2006).

12. In fact, apart from stacking, recursion is also involved, for events recur em-bedded inside more encompassing events. Davidsondid not recognize the fact that coexisting verb modifiers often do not evenmake sense if predicated of a unique event. For example, it has been wellknown since at least Dowty (1979) that whereastelic VP ), dverbialsmodify atelic ones (e.g., ), and, insofar as no verbsense can be both telic and atelic, we do not expect the two types of adverbialsto co-occur. Yet, they do, and in either order (cf.

), whichproves that even seemingly simple VPevent structure, and therefore we need not only stacked modifiers, but modifi-ers applying to different event structures embedded within a single VP, i.e.,complex event structures and a hierarchy of types of events and properties ofevents.

13. Particularly since Dik (1989) and Hengeveld (1989) (cf., e.g., Keizer 1992,2004; Rijkhoff 1992; Mackenzie 2004), Yet, the structure attributed to eventsin current F(D)G is still pretty much Davidsonian. The formulae offered incurrent work as representations of expressions like

are absolutely flat. There is no attempt to capture the fact that accom-

165

plishment verbs like involve three different sub-events (state, change ofstate, causation thereof, plus an ation, Theme andAgent are constituents of different sub-events.

14. The ontology assumed in F(D)G has, of course, expanded considerably sinceDik (1978), and therefore neither predicates nor their arguments need anylonger be first-order, as explicitly acknowledged in Dik (1989) and later work.Keizer (2004), Hengeveld and Mackenzie (fc.), and Mackenzie (2004) containrecent discussions of the various types of entities (properties, individuals,places, times, manners, states of affairs, propositions, speech acts), and theirmodifiers. In fact, the test for entity-hood is: whenever it is possible to referanaphorically to or focus a WH-word on a constituent, its denotatum is an en-tity (cf. Hengeveld and Mackenzie fc.). Whether the ontological enrichmentaffects only the intralinguistic world of senses or also theobvious. Current pronouncements (e.g., in Hengeveld and Mackenzie 2006:673) to the effect that properties have no independent existence and perhapsDik between ggest thatthe extralinguistic ontology assumed in F(D)G is still austere, but, be that as itmay, its intralinguistic ontology is certainly rather rich (cf. ). As the on-tology that terms must name has expanded, parallel variables standing forproperties (etc., occur alternating with F(D)G terms (cf. Dik 1989:180-182; Keizer 1992 on properties; Hengeveld 1992a: 7, on speech acts; Dik1997a: 131-132, 136-138; Dik 1997b: 93; Hengeveld 2004; Keizer 2004: 15-16; Hengeveld 2005, and Mackenzie 2004 for a good short overview). Al-though I discuss matters here mostly with reference to the l-lel reasoning applies to

15. To be fair, at least early FG seems to have simply inherited the problem fromthe logicians of the time. Russellian-Davidsonian accounts were just about allthe semantic doctrine available on NPs in the early seventies, and it was natu-ral for Dik to adopt their views. Although FG did not share the realist theoryof meaning, he probably saw the mental world that FG terms refer to as suffi-ciently isomorphic to the t-ment.

16. Hengeveld (this volume) shares the view that proper nouns are not modifiablein the representational level, although they can be modified by predicates ofthe interpersonal level (cf. Cf. also Butler this volume).

17. Whether the higher variable and its function can indeed be associated with aspecific level of structure, rather than with a function of the full NP, is dubi-ous, though. Rijkhoff -231) discoursecorrespond to any well-defined continuous stretch of structure at the top ofNPs, since discourse-oriented modifiers like

, etc. must also often be structurally attached under the scope of Locationoperators and modifiers. And, as a matter of fact, the same applies to

166

Hengeveld i-cal features that account for the referential vs. ascriptive use of NPs, but alsoother elements (e.g., Hengeveld this volume). Since articles crucially help theaddressee construct or identify the referent, it seems reasonable toassociate them with that level, but articles cannot be purely -representationalpredicative uses, and, to that extent, their association with the rsonallevel is questionable (cf. Keizer this volume). On the other hand, other inter-personally relevant items may occur fairly deeply embedded within NP struc-ture (e.g. in , is visible at the interpersonal level but not atthe representational level (cf. Hengeveld this volume). Thus, whether thegrammar is top-down, as in current F(D)G, or bottom-up, as in classical FG,the interpersonal and discourse velsthe representationally relevant ones. Hence, calling reference/discourse-oriented properties of NPs a further s-leading: such a structural level does not seem to exist (cf. Butler this volume).A hint in this respect is that, according to Hengeveld and Mackenzie (fc.),Hengeveld (this volume) the difference between referential and ascriptive NPsis just functional, not structural. If so, labelling referential tokens of NPs

s-take. I suspect, though, that structural differences between referential andpredicative NPs may well be indirectly detectable, as they are in languageslike Spanish, and, anyway, Hengeveld and Mackenzieview that nothing at the representational level distinguishes ascriptive fromreferential uses of NPs is implausible: after all, ascriptive NPs denote proper-ties, whereas referential ones must denote individuals. Such a radical differ-ence in sense is unlikely to be irrelevant precisely at the representational level(cf. also Keizer this volume).

18. Properly, in that passage, Rijkhoff fits well with themetaphysical analysis of individuals in t-tempt to separate a substratum from the remaining intensional properties thatmay constitute an NP

19. Correspondingly, , etc., as appropriate in NPsdescribing events, properties, propositional attitudes, etc. that Rijkhoff (2002)does not discuss.

20. Apart from icate variables (fi m)ranging over senses, at N and the heads of all satellites. For reasons alreadystated, they play no role in the computation of senses or referents and will beignored here, although their status will re-emerge briefly below in our discus-sion of operators and binding issues surrounding them.

21. Terms denoting other types of entities follow the same pattern, with differentvariables

167

22. A further issue is what binds -bindingproblems .

23. Admittedly, this multiplicity holds only if individual competence matters.What view current F(D)G adopts with respect to the internalism/externalismissue and the private language debates is unclear to me. If individual compe-tence does not matter in F(D)G, there is a single, idealized stic worldthe world of senses generated by eachthat as it may, Ling-W has a place of honour in linguistic semantics. For struc-tural/internalist semanticists like Coseriu (1992) and Chomsky (2000), whatwe here call -W -wise. As toF(D)G, to the extent the equation [semantically relevant entity] = [entity asconstrued by the language system] is assumed (cf. Dik 1997a: 129; Hengeveldand Mackenzie 2006), it is Ling-W, not really Psych-W, that is the core do-main for semantics in F(D)G too.

24. Dik (1997a: 131) restricts the concept of ities named byterms functioning as arguments or satellites in clause structure, but that mustbe an oversight, or an NP like inside another NP like

will not be a referential expression. Why predicates, predications,propositions and clauses do not o entities of type f, e, X and E, respec-tively, but merely i-cian, Dik was a realist with an austere Phys-W ontology (just extensional indi-viduals) and considered predicates, predications, etc. as names of Ling-Wentities (mere senses), which would explain why he adds that designation issimilar to constructive reference. However, the rest of Dikmake room for such an asymmetry between terms naming first-order entitiesand those naming higher order ones.

25. Correspondingly, -order variables bound bydiscourse-constructed events, facts, speech acts, etc.

26. Although Dik uses both terms,what Dik has in mind and will be preferred here.

27. Of course, the truth conditions (but just the truth conditions) of a term likecan, in principle, be adequately represented by a

flat unordered conjunction of clauses like ... [girl(x) & Indian(x) & beautiful(x) & dark (x) & tall (x)], etc., but such a formula does not capture the factthat in natural languages head and modifiers are attached in a certain order thatis subject to strict, near-categorical, constraints, cf.

, , etc. Obviously, the grammariancan simply stipulate such constraints, but that is hardly an elegant strategy if itis possible to derive the order of modifiers from deeper principles, and Dikwas clearly taking the latter course when he talked about ynamic construc-tion

on predicability (hence, on the extension of the respective sets in-volved). Spelling out that idea a bit, the result is that predicates like

168

, etc., are predicable of different sets, but in that case thesets cannot be represented by the same variableof conjunctive formula above will have a different range; ,the various instances of iable.

28. Observe that when several operators occur, they are not separated bythey are not attached in the order of their left-to-right position in the formula.If they are stacked, as Dik (1989) claims, the determiner is predicted to havescope over the numerator, and the same result is obtained when, followingRijkhoff ciates each operator with a differentlayer of term structure. Of course, that would not yield a tree like (3), cf. infra.

29. See Reichenbach (1947: 87, footnote and ff.). Term-internal binding is theonly formal difference that Dik (1997a: 168) acknowledges between FG quan-tifying operators and logical ones. Recall that Frege iers are functionsapplying to propositional functions.

30. It also entails revising the status of all other usalones: a very significant rethinking of F(D)G, all in all.

31. Arguably, there is a third. Number, definiteness, and case, in particular, arealso said to extend their influence over the whole term on which they operate,cf. Dik (1989: 138-139, 317-319). A similar statement, generalized to all termoperators, remains in Dik (1997a: 161), but must be a mistake there, for a fewlines down on the same page Dik approvingly introduces Rijkhoff c-count of term operators as associated with successive onion-like layers of termstructure. That organization, of course, is strictly parallel to Diktreatment of clause-level operators as stacked around their operand, with inneroperators falling within the scope of outer ones, cf. Dik (1989: 308), which iskept in the revised version of Dik (1997a). However, Diknumber, etc. is far from dead in F(D)G. Bakker and Pfau (this volume) still in-sist on the idea that the number, definiteness, and case operators areoperators with scope over the whole term, although this may be due to a mis-reading of Dik yway, there are very good reasons why e.g. case mustbe a very (2005c) and section 4 .

32. In (5), will be higher or lower (hencewhether it is taken to be an identifying satellite (cf. Keizer 2004) or a qualify-ing one, but the point does not affect the argumentation here. I stick to thesame example just for the purposes of comparing alternative structures. If, in-stead of , the last restrictor were a relative clause like

or a discourse-level identifier like they woulddefinitely be attached as sisters to the operator d, as suggested.

33. Observe that although the left-to-right sequence of restrictors is meant to ex-press increasing scope, if the same criterion is applied to the sequence d prox3 we obtain exactly the converse of what Dik elsewhere claims to be the rela-tive scope of d, prox, and 3: the numeral will have scope over prox, and proxover d. Assuming the contrary, the operators are themselves stacked onion-like

169

with respect to the variable lds structures like (3), not (5).From Dik e-spect to restrictors.

34. Just as variables must be uniquely bound by appropriate operators, argumentsmust be uniquely rminology varies, but the dif-ferences are immaterial here) appropriate predicates, or they will not receive asemantic function and will be uninterpretable (just like uninterpreted variablesin a calculus).

35. The standard view of scope, used in mathematics, formal logic, and formallinguistics, entails that if A and B are constructed into a new entity [A*B] (or-der irrelevant; * = whatever combinatory operation applies), then A and Bhave scope over each other, and that if [A*B], in its turn, is constructed with anew object C into a structure C*[A*B] or [A*B]*C, then C and [A*B] are ineach otherB has scope over C.

36. Linear precedence is a non-starter, though, if the structure of terms is asclaimed in current layered models and the aim is to say that e.g., satellites atLn have scope over constituents of layer Ln-1, since when the onion-like termstructure is projected onto a linear sequence, the head noun precedes (andtherefore would have scope over) all post-nominal satellites.

37. Functional uniqueness entails that full NPs will not be further restricted: sincethe full NP must be able to receive a semantic function from the head it is con-structed with (e.g., a verb), it cannot receive a second semantic function assubject of a restrictor. Hence, restrictors must operate at intermediate levels ofterm structure, and on categories different from the NP (i.e., not on the termvariable ctional uniqueness that the same headnominal cannot be restricted by more than one restrictor, or by one restrictorand one operator, i.e., the structure cannot bebranching, as STC predicts. The head nominal is constantlyconstruction of the term proceeds (cf. our discussion of the variable update is-sue above).

38. In what concerns theHengeveld (this volume), where each of the restrictors (e.g., , , in

) carries its own operator and its own variable. However, thehigher nnot stand forthe same set (i.e., two different variables).

39. Note that, stepwise construction, according to Dik, is a deep property thataffects the sense of terms, not just a surface realization detail. Actually, Dikeven attributes psycholinguistic significance to it, for he claims that it reflects

logy), cf. Dik (1997a: 62, 135).40. A has asymmetric scope over B if A has scope over B but B does not have

scope over A, as in the structure A[B, C]. Sisters, on the contrary, have sym-metric or mutual scope, as in structures like [A, B, C].

170

41. Obviously, this is consistent with grouping subsets of such c-tion into as many functionalnature and semantic contribution of operators and satellites, e.g., a term mayhave a discourse status segment (for (in)definiteness), a spatial location seg-ment (for deictics), an ordinator segment (for etc.), a quantificationsegment, a modal segment (for modifiers like or ), and asmany aspectual, qualification, classification, and complementation segmentsas need to be established. Current F(D)G ur or five layers have that func-tional character, but the functional distinctions so far established, especially inwhat concerns the intermediate and lower levels, are still too broad. Work likeScott (2002) reported in Cinque ed. (2002) suggests that a much more delicatelayering of modifiers is justified, and recent proposals by F(D)G scholars likeKeizer (2004) and Rijkhoff (this volume) point in the same direction.

42. I will useits successive expansions as a consequence of stepwise construction with satel-lites and operators, i.e., in the sense of lexical or phrasal head of the NP. Ifnecessary, sub-indices can be used to unambiguously identify successivelybigger nominal cores, as X-bar theories do with bar levels, following an ideadue to Zellig S. Harris.

43. This is an important respect in which F(D)G differs from other approaches thatalso explain facts, particularly constituent order facts, in terms of a compro-mise among competingpapers in Barbosa et al. eds. (1998). The difference is that in OT, the respec-tive eting principles is explicitly defined in a hi-erarchy, the import of each type of violation is strictly quantified, and to thatextent it is possible to predict which of the competing forces will prevail incases of conflict. On the contrary, F(D)G does not say anything on the relativestrength of its ordering preferences, and it is impossible to predict what willhappen, except ex post facto.

44. It is assumed here that theoretical linguistic objects are sets or matrices of[attribute: value] pairs. An [A: v] pair is a function that applies to its carrierand yields a unique value over a restricted domain. Informally, an [A: v] pairsimply expresses a proposition about its carrier. All sorts of properties can beformalized with precision in that format, which is extensively used in gram-mars like LFG, GPSG, HPSG, Construction Grammar, and all related compu-tational approaches. Traditional features like +F, used in FG and many othertheories, correspond to a particular subset of [A: v] pairs whose value range isBoolean (+,-). Since many attributes necessary to characterize linguistic itemsare not Boolean, the [A: v] format is preferred, though. An attribute is l-uedleave their carriers as incalculable factors in computations, causing incom-pleteness or anomaly.

171

45. E.g., the construction of a verb, adjective or noun with its (PP) complement, oran adjective with its N(P), satisfies both an adicity requirement on the verb,noun or adjective, and a role value, and often a case value requirement, on thecomplement, and semanticists like Pustejovsky (1995) claim that co-composition (= bi-directional functional application) must be allowed.

46. Also Core+Argument, obviously; I ignore that case here, since the discussionis here focused on the structure of the upper FG layers at which Sats and Opsintervene. Arguments are also attached in a stepwise fashion, but I will ignorethe details here, since I suspect in matters like event structure inside nominalsthere is little common ground across different linguistic schools and I want tokeep this discussion as close to theory-neutral as possible.

47. In cases of predicate composition, one of the predicates functions as an argu-ment of the other, i.e., one is type <<e,t>,<e,t>>, a property of a property, (anth order property), whereas the other is just <e,t> a property of order n-1.

48. This does not entail that a noun cannot in the earlier stages of term construc-tion function as a predicate and take its own arguments, cf. Keizer (2004) foran F(D)G view of this. What is claimed here is that even relational nouns,once saturated, form cores that function as arguments of satellites at higherlevels, but cannot satisfy or be satisfied by another core nominal.

49.present context, head refers to thepredicate, usually an adjective, a semantically non-trivial preposition (cf. in-fra), a participle, etc.

50. There is a systematic but innocuous apparent exception: Since the subject isthe last (sometimes called xternalthat it is also the , the head of that predicate mayhave to satisfy a satellite of its own before satisfying its subject, since the se-mantic type of the object resulting after the head is completely satisfied wouldno longer be constructible with the satellite. This may happen only when thesatellite does not alter the type of the head in a significant way, e.g., in caseslike , vs.

. In this example, is a second-order degree property. It willtherefore satisfy its argument if it combines with , a gradablemonadic predicate, but will not match the type of the combination resultingwhen satisfies its subject, i.e., , a non-gradableproperty. Hence, either the adverb is satisfied first or it remains unsatisfied,dangling, and the whole structure fails. Thus, Satisfaction forces a violation ofPriority in this case, unsurprisingly, as Satisfaction is the real principle; Prior-ity just defines how it is to be applied.

51. The facts of SOV languages are different, and must be accounted for by addi-tional principles, either Kayneso easily deduced and saved from counterexamples in all cases. In this pro-

172

grammatic conceptual discussion, I will limit attention to how much followsfrom Satisfaction in English-like languages.

52. It is assumed here that Focus and Topic areconstituents that discharge them carry attributes like [D-F(unction): Top], etc.,and that the Subject is a -internal argu-ment), hence, a subject will be marked as e.g. [Cl-F(unction): Subj]. In othertheories the subject is treated as a special or l-ogy is immaterial. What matters is that the subject is not just an argument ofthe verb, although it is also an argument of the verb.

53. Only a few suggestive examples can be discussed here, though. Further cover-age of data is clearly impossible in the space available in this already long ar-ticle, and well beyond my competence in what concerns non-IE languages. Itfollows that current F(D)G concepts might still prove descriptively preferableto the present version in view of the facts of a large sample of languages likethat used by Rijkhoff (2002), so I am well aware that the adequacy of eitherthe proposal sketched here or F(D)G as currently developed cannot be judgedwith reference to just English and similar languages. However, those few lan-guages, especially English, happen to be those about which we know incompa-rably more, and to that extent it is understandable that the adequacy of theoriesbe first of all judged with respect to how well they handle really well investi-gated problems known to a great majority of linguists. Also, under the assump-tion that Human Language is based on a unique universal computationalmechanism (N.B.: such a mechanism need not be anot an tegy to adopt as a zero hy-pothesis that if an elegant set of principles makes really strong and delicatepredictions in one or a few well investigated languages, such principles maywell be appropriate to a wider domain. In this respect, the versions of Satisfac-tion, Priority, etc., presented above have a simplicity, coherence, and explana-tory power (if only, in English and similar languages) that make them strongcandidates worth comparing with alternative concepts. Obviously, it will bemuch easier to produce counterexamples to them than to HP, DI,etc. as currently understood in F(D)G, since the theory argued for here is morerigid, but the real issue is to determine whether the apparent counterexamplesare genuine and really invalidate the principles, or just follow from low levelfacts, or further principles of morphology, etc., without need to abandon theprinciples. Of course, that cannot be done here, so all I will do is compare thepredictions of the theory above with those that follow from the standardF(D)G concepts as developed mainly in Rijkhoff (2002) and Keizer (2004).

54. Even ignoring the hedge ifiedB will usually be adjacent, but does not say in which order (A+B, B+A?) theywill occur. As shown above, depending on the internal structure of A and B, itis possible to do rather better, cf. Escribano (2004a) for details.

173

55. Refining the layer structure, as in Keizer (2004) and Rijkhoff (this volume),may avoid piling up modifiers at a single layer and exclude some of the badpredictions, e.g., if one of the modifiers of is (

etc.), in Rijkhoff -layer model it might belong to the classifierlayer, which would force it to at least be adjacent to the head, i.e., only order-ings containing the sub-string or the deviant *will be predicted.

56. Of course, the FG analysis of PPs and Rel Clauses automatically induces aviolation of HP as soon as a PP or Rel Clause is attached to a noun, even if noother dependent intervenes, whereas under analyses that assume that P and Relare heads no HP violations arise in such cases. There is strong evidence insupport of the view that both unselected P and Rel are heads with their ownarguments, cf. Escribano (2004a) on Ps and Escribano (2003) on Rel Clauses.

57. Rijkhoff (2002: 290, 308) directly excludes demonstratives, numerals, andadjectives from the scope of his HP constraint. The reason adduced is thatsuch elements cause only minor violations that speakers can tolerate, whereasphrases cause intolerable processing difficulties. However, in the case of arti-cles, demonstratives and numerals, there is no need to weaken HP: since theynever take any other complements but the core nominal, appropriate layeringand the relativized concept of Head above will immediately eliminate the falseminor HP violation anyway. On the contrary, in the case of adjectives, whichcan take complements, it is a mistake to exempt them from HP, and the imme-diate consequence is that constructions like * will be li-censed (cf. ). Of course, Rijkhoff (2002) does not really discuss complexphrases, which disguises this type of problem, but it is a problem.

58. There are a few exceptions, of course, cf. ,, etc., but they are not significant at this level of abstraction. See Es-

cribano (2004a) and (2005a) for a possible explanation in terms of Focus.59. Underlyingly, is just a noun with a role feature; recall that prepo-

sitions are not predicates in FG and do not exist until the expression rules in-troduce them later.

60. One of the editors questions this analysis of aswithin FG. Yet, as far as I know, there is no other FG analysis of such modifi-ers: if prepositions are inserted r-lying structure that PP is just a noun with an Origin/Source? role feature. If thePP contained a full NP (e.g., ), it could ar-guably be treated as an embedded domain and excluded as a o-lation, but, as it stands, it is no different, in my view, from having adjectives orquantifiers intervening between determiners and nouns (thethat Rijkhoff rates by excluding them from the scope of HP).

61. It need not be an impoverished concept of sense, though. Thus, the affective-interactional aspects of meaning that Butler (this volume) discusses need notbe extraneous to Ling-W as here understood. This is obvious: to the extent

174

such features are expressible through linguistic means (i.e., not by means ofgestures, musical resources, etc.), they belong to the language, are subject toits conventions, etc.

62. Keizer (this volume) defends the same view. Rijkhoff (this volume), andHengeveld (this volume) emphasize the yers, but it remainsunclear to me whether their NPs ultimately denote Ling-W entities, or ratherextralinguistic ones. As to Butlerrather seems to correspond to what we here call C-World, i.e., his meaningsseem to be conceptual in a broad sense compatible with the existence of dif-ferent modules to handle affective-interactional concepts and narrowly con-ceptual-representational structures. The approach defended here, on the con-trary, is not conceptualist, but internalist, or language-immanentist, i.e., oursemantics is the kind of semantics that Saussure, Hjelmslev, Coseriu, (and,when he is consistently internalist, Chomsky) had in mind. The mind with allits content is surely relevant to any explanation of performance, and to the in-tegrated account of linguistic behaviour that F(D)G aims at (cf. Dik 1989, But-ler this volume), but linguistic expressions name what they can name, i.e., enti-ties from Ling-W. The relation between Ling-W entities and C-World (andPhys-W) ones is altogether a different story, and a fascinating one.

63. In Escribano (2004b), I proposed eliminating variable-binding issues and thevariable-update problem by limiting term operators and variables to a higher-order one at the top of the term structure. Of course, that was my compromisewith the traditional FG assumption that all terms and clauses have an ([ (this volume) now proposes to treat op-erators as modifiers and nearly dispenses with variables altogether, I feel freeto propose here in an F(D)G framework what I proposed in Escribano (thisvolume [written 2004]), in a broadly Chomskyan one. This proposal, there-fore, differs from, and is incompatible with all those that maintain sense/termvariables in the formulae of the representational component, e.g., Rijkhoff(this volume), Hengeveld (this volume) and Keizer (this volume).

Bakker, Dik and Pfau, Rolandthis vol. Agreement in the noun phrase: The dynamic expression of terms and

what can go wrong.Barbosa, Pilar, Danny Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky,

(eds.)1998 .

Cambridge, M.A.: The MIT Press.

175

Barwise, Jon and Robin Cooper1981 Generalized quantifiers in natural language.

4: 159 219.Bernstein, Judy1997 Demonstratives and reinforcers in Romance and Germanic lan-

guages. 102: 87-113.Butler, Christopher S.this vol. Interpersonal meaning in the noun phrase.

Chomsky, Noam2000 . Cambridge:

Cambridge University Press.2005 Three factors in language design. 36: 1 22.

Cinque, Guiglielmo1994 On the evidence for partial N-movement in the Romance noun

phrase. In: Guiglielmo Cinque (ed.),, 85 110. Washington D.C.: Georgetown University Press.

Cinque, Guiglielmo (ed.)2002 .

Oxford: Oxford University Press.Coseriu, Eugenio1992 Sem sco

Mars -282. Barcelona: Universidad deBarcelona.

Crisma, Paola1993 On adjective placement in Romance and Germanic event nominals.

18: 61 100.Davidson, Donald1967 The logical form of action sentences. In:

Oxford: Oxford University Press, 1980.Dik, Simon C.1989 Dordrecht: Foris Publi-

cations.1997a

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

1997b(Functional Grammar Series 21.) Edited by Kees

Hengeveld. Berlin/New York: Mouton de Gruyter.Dowty, David1979 . Dordrecht: Reidel.

Ernst, Thomas2002 Cambridge: Cambridge University Press.

176

Escribano, J. Luis G.2003 On relative clauses. 47: 157

182.2004a Head-final effects and the nature of modification.

40: 1 43.2004b On the structure of terms in FG: Some conceptual issues (A sympa-

thetic outsider Paper presented atUniversity of Oviedo, September,

2004.2005a Discontinuous APs. 43: 563 610.2005b Semantocentric Minimalist Grammar. 27-2: 57 74.2005c English compounds and the theory of abstract case. In: Luis Quereda

(ed.),, 107 131. Granada: Uni-

versidad de Granada.2006 NPs as just NPs. 28-6: 529-579.

Frege, Gottlob1952 Ox-

ford: Basil Blackwell.Giusti, Giuliana2002 The functional structure of noun phrases: A bare phrase structure

approach. In: Guiglielmo Cinque (ed.), 54-90. Oxford: Oxford University Press

Hawkins, John A.1994 Cambridge:

Cambridge University Press.2001 Why are categories adjacent? 37: 1 34.

Heim, Irene and Angelika Kratzer1998 . Oxford: Blackwell.

Hengeveld, Kees1989 Layers and operators in Functional Grammar.

25: 127 157.1992a . (Functional

Grammar Series 15.) Berlin/New York: Mouton-de Gruyter.1992b Parts of speech. In: Michael Fortescue, Peter Harder and Lars

Kristoffersen (eds.),, 29 55. Amsterdam/ Philadelphia: John Benja-

mins.2004 Epilogue. In: J. Lachlan Mackenzie and Mar

G -Gonz, 365 378. (Functional Grammar Series 24.) Berlin/New York:

Mouton de Gruyter.

177

2005 Dynamic expression in Functional Discourse Grammar. In: Casper deGroot and Kees Hengeveld (eds.)

53 86. (Functional Grammar Series 27.) Ber-lin/New York: Mouton de Gruyter.

this vol. Prototypical and non-prototypical noun phrases in FDG.Hengeveld, Kees and J. Lachlan Mackenzie2006 Functional Discourse Grammar. In: Keith Brown (ed.),

Volume 4, 668 676. Oxford: Elsevier.fc. . Oxford: Oxford University Press

Hengeveld, Kees, Jan Rijkhoff and Anna Siewierska2004 Parts of speech systems and word order. 40:

527 570.Jackendoff, Ray1990 . Cambridge, M.A.: The MIT Press.

Kayne, Richard S.1994 . Cambridge, M.A.: The MIT Press.

Keizer, Evelien1991 Referring in FG: How to define reference and referring expressions

43.1992 Predicates as referring expressions. In: Michael Fortescue, Peter

Harder and Lars Kristoffersen (eds.),, 1 27. Amsterdam/ Philadelphia:

John Benjamins.2004 Term structure in FG: A modest proposal.

78.this vol. Reference and ascription in FDG: An inventory of problems and

some possible solutions.Locke, John1997 Reprint. . London:

Penguin Books, 1690.Longobardi, Giuseppe2001 The structure of DPs: some principles, parameters, and problems. In:

Mark Baltin and Chris Collins (eds.),, 562 603. Oxford: Blackwell.

Loux, Michael J.2002 (2nd. ed.). London and

New York: Routledge.Mackenzie, J. Lachlan1992 English spatial prepositions in Functional Grammar.

46.1996 English nominalizations in the layered model of the sentence. In:

Betty Devriendt, Louis Goossens and Johan van der Auwera (eds.),

178

, 325 355. (Func-tional Grammar Series 17.) Berlin/New York: Mouton de Gruyter.

2004 Entity concepts. In: Geert Booi, Christian Lehmann, Joachim Mug-dam and Stavros Skopeteas (eds.),

, vol. 2. 973 982. Berlin/New York:Mouton de Gruyter.

McGee Wood, Mary1993 . London/New York: Routledge.

Montague, Richard1974 The proper treatment of quantification in ordinary English. In:

. Edited byRichard Thomason), 247 270. New Haven/London: Yale UniversityPress.

Parsons, Terence1990 .

Cambridge, M.A.: The MIT Press.Partee, Barbara H.1995 Lexical semantics and compositionality. In: John Gleitman and Mark

Liberman (eds.), , 311360. Cambridge, M.A.: The MIT Press.

Pesetsky, David1995. Cambridge, MA: The MIT Press.

Pustejovsky, James1995 . Cambridge, M.A.: The MIT Press.

Reichenbach, Hans1966 Reprint. . New York: The Free Press,

1947.Rijkhoff, Jan1992 The noun phrase: a typological study of its form and structure. Ph.D.

dissertation, University of Amsterdam.2002 . (Oxford Studies in Typology and Theoretical

Linguistics.) Oxford: Oxford University Press.this vol. Layers, levels and contexts in Functional Discourse Grammar.

Russell, Bertrand1905 On denoting. 14: 479 493.

Scott, Gary J.2002 Stacked adjectival modification and the structure of nominal phrases.

In: Guiglielmo Cinque (ed.) , 91120. Oxford: Oxford University Press.

Steedman, Mark1993 Categorial Grammar. 90: 221 258.

179

Strawson, Peter F.1959 . London/New

York: Routledge.Wittgenstein, Ludwig1922 Reprint. . London: Routledge and

Kegan Paul. First published 1918.

In the standard theory of Functional Grammar (henceforth FG), the distinc-tion between reference and predication has always proved to be somewhatproblematic, both before and after the introduction of the layered model(Hengeveld 1989; Dik 1989; see Keizer 2004). As such, this topic providesa genuine challenge for Functional Discourse Grammar (henceforth FDG):to design, within the broad outline of the model developed so far, a coher-ent, psychologically plausible and intuitively appealing way of dealingwith reference and ascription.2 Traditionally, the terms reference and as-cription (or predication) have been taken to belong to the realm of seman-tics. In FDG, however, they are used to describe acts performed at the in-terpersonal level. It will, therefore, be worthwhile to see whether thedistinction made within the FDG model between an interpersonal and arepresentational level may pave the way for a more adequate treatment ofreference and ascription. At this moment, however, it is difficult to evalu-ate the possible contribution the new model can make, since the internalstructure of referential and ascriptive subacts (at the interpersonal level)and the semantic units (at the representational level) have not yet receivedsystematic treatment; nor has the relation between the two levels been fullyexplored.If, in the development of the FDG model, we want to avoid lapsing into

the same old mistakes, now is the time to reconsider the way FG dealt withreference and ascription. The aim of this article will, therefore, be to offeran inventory of the problems with regard to reference and ascription thatcan be, and have been, identified in existing treatments within FG (Dik1978, 1989, 1997a, 1997b; Rijkhoff 2002; Anstey 2002), and to indicateways in which (some of) these problems can be avoided in FDG. Morespecifically, this article will address the following questions:

182

What is the nature of representation at the representational level?What does the x-variable at this level stand for?3

Is it justified to represent/symbolize the (extra-linguistic) referentof an expression within a grammar model?What is the relation between the variables at the representationallevel and any extra-linguistic entities in the contextual compo-nent? Is this relation part of the grammar model or not?How can we best exploit the possibilities offered by differentcombinations between, on the one hand, the subacts of referenceand ascription at the interpersonal level and, on the other, the dif-ferent semantic categories of the corresponding expressions at therepresentational level?

Throughout the article, proposals will be tested by applying them to copu-lar constructions in English, as it is in these constructions that the distinc-tion between reference and ascription is far from clear-cut. Detailed analy-ses will therefore be provided of classificational constructions (

), identity statements ( ) and identificationalsentences ( ), both at the interpersonal and at the representationallevel, first of all to demonstrate the need for a new approach, and secondlyto show the advantages of the new analysis to be presented.This article is organized as follows. Section 2 briefly discusses various

proposals concerning reference in FG. Next, Section 3 describes the rele-vant new features of FDG, in particular those relating to the treatment ofreference and ascription. It will be shown that, although in some respectsthe FDG treatment of reference and ascription is an improvement on for-mer approaches, some problems still persist. Section 4 addresses morefundamental questions about the nature of the variables at the representa-tional level and their relationship to the contextual component. It will beargued that if we conceive of these variables in a different way, this willallow for a treatment of reference and ascription that is both more consis-tent and more psychologically adequate. Section 5 demonstrates the advan-tages of the new proposal by applying it to proper names and pronouns, aswell as to restrictive and non-restrictive appositive constructions. Finally,Section 6 offers a brief summary of the main points.

183

2.1. Standard FG: Dik (1978, 1980, 1989, 1997a)

In standard FG, terms are defined in a seemingly straightforward manner asreferring expressions (Dik 1997a: 127; see also Dik 1978: 55; 1989: 111):

By a term we understand any expression which can be used to refer to an en-tity or entities in some world. The entities that a term can be used to refer toare the potential referents of that term. The entities that a term is meant torefer to are the intended referents of that term in that use.

To account for their formal properties, Dik (1978: 16, 57) offers the fol-lowing general schema for representing terms:

(1) ( i: i (xi): j (xi): ... n (xi))

where i is a variable ranging overa set of potential referents and each i) is an open predication in xi. Therepresentation is to be read as follows: the domain of the potential referentsxi is first restricted to the set of entities of which i (xi) is true; this latterset is restricted to the subset of which j (xi) is true, and so on until theopen predication n (xi) gives the last restriction on the set of potentialreferents.In subsequent versions of the theory of Functional Grammar, this basic

schema for terms remained virtually unchanged (Dik 1989: 115; 1997a:132). With the introduction of the layered model (Hengeveld 1989; Dik1989), the number of variables was, of course, extended, which meant thatthe term variable xi was from then on used only in terms referring to first-order entities (objects, individuals), while higher order entities (states-of-affairs, propositions, speech acts) were all provided with their own vari-able. These variables (e, X and E, respectively) were also used in termsreferring to these entities. At a later stage, the f-variable was introduced tosymbolize properties (Keizer 1992a, 1992b; Hengeveld 1992a, 1992b).None of these changes, however, affected the basic function and structureof the term as defined in Dik (1997a): terms still consisted of a term vari-able, one or more term operators and one or more restrictors. Thus, theimportant distinction between the actions of referring and predicating wasmaintained, both within the predication and within the term (cf. Strawson[1971] 1950; Searle 1969). In Dik (1997a: 127), for instance, we read that

184

The predication may be understood in terms of the two basic acts of refer-ring and predicatingwhich something is going to be predicated; predicating means assigningproperties to, and establishing relations between, such entities.

Within the predication, the difference was reflected in the distinction be-tween the predicate (verbal or non-verbal) on the one hand and its argu-ments (typically filled by terms) on the other; within the term, between theterm variable, representing a referent, and its restrictors, designating theproperties assigned to this referent. Since the actions of referring andpredicating were generally assumed to be mutually exclusive (Strawson1959: 142), this meant that terms could not be used to fulfil a predicatingfunction.Nevertheless, it was felt that in Dutch copular constructions of the kind

illustrated in (1) and (2), the post-copular expressions do serve to fulfil apredicating function (Dik 1980: 99-101). The examples in (1) could beaccounted for by arguing that the expression is not a term, butsimply a nominal predicate. Confirmation for such an analysis was foundin the fact that in Dutch such bare nominals do not contain a determiner,numeral or quantifier, may appear in the singular even with a plural sub-ject, and do not allow for (pre- or post-) modification. In other words, theybehave more like adjectives than like the first restrictor of a term; conse-quently, these nominal expressions were analysed in the same way as ad-jectives in this position, i.e. as non-verbal predicates (with the copula beinginserted at a later stage through the rule of copula support; Dik 1980: 100).

(1) a.Jan is painter

b.Jan and Piet are painter

c. * .Jan is famous painter

In (2a), on the other hand, the expression cannot be analysedthis way. Here the post-copular expression displays several of the formalproperties of a term (see also examples (2b and c)); as such, it was ana-lysed as a term.

185

(2) a.Jan is a painter

b.Jan and Piet are painters

c.Jan is a famous painter

Since, at the same time, it was still regarded as serving the same predicat-ing function as the bare nominal in (1a and 1b), a predicate formation rulewas introduced to convert terms into non-verbal predicates. This TermPredicate Formation rule, first introduced in (Dik 1980: 103), has remainedpart of the standard theory (Dik 1997a: 205):

(3)input: any term (t)output: {(t)} (x1)

A final important, and by now generally accepted, modification made inDik (1989) concerns the nature of the entities referred to. Whereas origi-nally referents were thought of as entities in the real world, Dik (1989)describes them as entities in the mind (Dik 1989: 113; see also Dik 1997a:129). As such they are part of a mental world, which need not correspondto the real world but may just as well be a representation of some mythical,fictional or hypothetical world.

2.2. Hengeveld (1992a, 1992b)

In Hengeveld (1989, 1990) we find the layered model presented for thefirst time. Inspired by Foley and Van Valin (1984), this model offered ahierarchical representation of the clause which distinguished between aninterpersonal level and a representational level. The interpersonal levelconsisted of two layers: the clause, represented by the variable E, and theproposition (X). At the lower, representational level the proposition was inturn restricted by a State-of-Affairs (e) consisting of (at least) a predicateand one or more terms (x).At a later stage the f-variable was introduced to symbolize the proper-

ties and relations designated by verbal, nominal and adjectival predicates(Hengeveld 1992a: 31; 1992b: 52-55). The need for this f-variable is illus-trated in the following examples, which show that anaphoric reference can

186

be made to properties (such as in example (4a)) and that these proper-ties can take their own modifiers (as shown for the adjectival predicate

in (4b)) (Hengeveld: 1992a: 33):

(4) a.b.

The f-variable also proved to be useful for the analysis of non-verbalpredications (Hengeveld 1992b: 77-80), where the non-verbal predicate(whether derived or non-derived) could now be provided with its own f-variable. Hengeveld also differs from Dik n-stead of a category of term-predicates, he distinguishes a larger class ofreferential predicates, which icates based on terms, i.e. refer-ring expressions with a nominal head, and predicates based on larger refer-ential units, i.e. predications, propositions, and clauses77). Apart from these differences, however, Hengeveldnominal non-verbal predicates was similar to that of Dik (1980, 1997a).Thus bare nominals like in (5a) were analysed as non-referentialpredicates, while post-copular terms like in (5b) were ana-lysed as referential predicates (Hengeveld 1992a: 32; 37; 1992b: 80), i.e. aspredicates (f) with a referential expression (x) in their scope. The differ-ence between the two analyses was reflected in their respective underlyingrepresentations (given here in simplified form):4

(5) a.a {(f1: president)} (x1: John)b.b {(f1: (d1x1: (best friend): my))} (x1: John)

Now, copular constructions with referential predicates come in differentkinds. Although various typologies have been proposed (e.g. Gundel 1977;Higgins 1979; Declerck 1988; see also Keizer 1992b, chapter 2), most ofthem acknowledge a basic distinction between classifying (or predica-tional) and identificational copular constructions. Hengeveld (1992b: 91),too, makes this distinction, relating it to a difference in the definiteness ofthe post-copular expression: whereas in classifying constructions the post-copular expression is indefinite, in identificational constructions the post-copular expression is definite.5 Both categories of copular construction aresubsequently further subdivided into (reversible) specificational sentencesand (non-reversible) characterizing sentences. Specificational and charac-

187

terizing sentences (whether classifying or identificational) further differwith regard to the number of referents they involve: whereas in specifica-tional sentences subject and referential predicate refer to the same entity, incharacterizing sentences the two expressions refer to different entities. Thisresults in the following typology (Hengeveld 1992b: 89):

. Referential predicates

(reversible) (non-reversible)

(definite referentialpredicate)

{(fi: (dxi)} (xi)6

{(fi: (dxi)} (xj)

(indefinite referentialpredicate)

{(fi: (ixi)} (xi) {(fi: (ixi)} (xj)

Note finally that so-called identity statements (e.g. Declerck 1988: 110-113) are described by Hengeveld (1992b: 87) as specificational construc-tions . Thus, an example like (6) is seen as serving no otherpurpose than to state that two names, and , maybe used for the same referent set (xi).

(6) a. (simplified):a (f1: (d1x1: Evening Star (x1) ) (f1)) (d1x1: Morning Star (x1) )

2.3. Keizer (1992a, 1992b)

In Keizer (1992a, 1992b) I drew attention to what I still consider to be amajor problem in both Dik Hengeveld -verbalpredications (see also Section 4). Whether we refer to the post-copularexpression as a term-predicate or a referential predicate, in both cases theimplication is that such expressions fulfil two functions at the same time:they refer to some entity while at the same time they predicate a propertyof an entity (either the same entity or some other entity). This means, firstof all, that referring and predicating are no longer two separate acts. Sec-ondly, from a logical point of view it does not seem to make sense to as-sume that if in a copular construction an expression like is

188

regarded as fulfilling a predicating (property-assigning) function, it shouldnevertheless be analysed as referring to an entity.In Keizer (1992a, 1992b) I therefore proposed putting the f-variable in

non-verbal predications to a slightly different use. As in Hengeveldanalyses of such constructions, it was used to symbolize the property as-signed to the expression in argument position. However, in those caseswhere the copular construction is a classifying one (whether specifica-tional, descriptional or predicational; see Keizer 1992b: 90-96), the f-variable was no longer restricted by a term; instead it merely representedthe property denoted by the (combined) predicate(s) contained within thisexpression. As the post-copular expression was no longer regarded as ful-filling a referential function, it no longer contained an x-variable. Thus,expressions like those in (7a) were seen as involving only one referent andone referential expression ( ); the expression has a property-assigning function and, as such, does not refer to an entity. Unlike in (5btherefore, the post-copular expression in (7) contains only an f-variable(representing the property) and no x-variable:

(7) a. John is a fool.a {(fi)} (xi)

The two approaches can thus be said to differ in terms of what is givenmore emphasis: the form of an element or its function in the predication.Thus, in the Dik-Hengeveld approach the post-copular expression in classi-fying copular constructions is analysed as a referential expression becauseit has (some of) the formal properties of a term. In my own approach it isthe function of this expression, as a non-verbal predicate, which leads to itsanalysis as a non-referring expression, despite the fact that it shares a num-ber of formal properties with terms.Evidence supporting this latter stance can be found in the fact there are

also differences in formal behaviour between referring expressions (inargument position) and the post-copular expressions in question. One ofthese concerns the set of pronouns used to indicate anaphoric relations withthe two types of expression: whereas expressions like or

, when used to refer to an individual, will be anaphorically referredto by means of the definite masculine or feminine personal pronouns or, the same expressions, when used predicatively, trigger the neuter defi-

nite pronouns , or , or the indefinite pronoun (cf. Keizer1992b: 124):

189

(8) a.

b.

In other words, there is both functional and formal evidence for the non-referring status of post-copular nominal expressions.Note furthermore that the approach suggested is also more efficient in

that it allows for a simplification of the theory: if the post-copular expres-sions in question are not terms but property-assigning expressions, they canbe used as predicates straight away, which means that there is no longerany need for a term-predicate formation rule.Finally, problems also arise with the analysis of identity statements

proposed by Hengeveld (1992b). First of all, it needs to be mentioned thataccording to Declerck (1988), on whose work the classification proposed

by Hengeveld is based identity statements like (6) are not specificational(see Declerck 1988: 113). In true identity statements, where the emphasisis on the copular verb, neither of the two expressions serves as a specifica-tion of the other, nor does either of the two expressions predicate over theother. After all, given that the order between the elements is reversible, andthe focus is on the copular verb, how can one determine which part predi-cates over the other? A more plausible analysis is one in which both ele-ments in an identity statement are regarded as taking argument position(for more details, see Keizer 1992b: 167).A second problem concerns the fact that on Hengeveld

specificational constructions, as well as identity statements, take the formof xi = xi; that is as expressing a relation between two identical entities. Aspointed out by Frege ([1892] (1977: 56), this would mean that a sentencelike does not differ from

, which holds , and can hardly bethought of as informative. Clearly, however, the two statements do notconvey the same message; moreover, identity statements can be highlyinformative. This suggests that an analysis involving equation of one andthe same entity is not adequate. Fregea relation between two (identical objects), but as a relation between thenames, or signs, used to designate these entities (Frege 1977: 56):

What is intended to be said by seems to be that the signs or namesandunder discussion; a relation between them would be asserted.

190

Effecting such ns to our knowledge 1977: 56) iscertainly one function of identificational constructions (see also the discus-sion of non-restrictive apposition in Section 5). Identity statements, how-ever, seem to serve a different function: here two previously introduceddiscourse entities are asserted to be identical (at least in the speaker n-tal representation of some world). It would therefore seem more appropri-ate to regard these constructions not as merely asserting a relation betweentwo names, nor as equating two identical objects, but as equating two sepa-rate (mental) entities. (For further discussion, see Section 4.)This leaves us, however, with the question of why it is that, in the case

of a (non-referring) noun phrase in post-copular expression, a predicatingelement displays so many of the formal properties of referring nounphrases? Moreover, when faced with such a discrepancy between the (typi-cal) form and function of a noun phrase, which criteria do we use to deter-mine whether we are dealing with a referring or a non-referring expres-sion? And is it plausible to assume that the distinction between referringand non-referring expression is a clear-cut one? In Keizer (1992b: 139) Itried to resolve this issue by distinguishing between prototypical and non-prototypical referring expressions.7 A prototypical referring expression (or

8

(9) : first-order entity: nominal head

no valency (is non-relational)all term operators may (in principle) apply

: argument9

The underlying assumption is, of course, that the category of referring ex-pressions is a graded category: some of its members are better examples ofthe category than others. This in turn raises the question of whether it isnevertheless a bounded category, i.e. whether we are to assume that allexpressions are to some extent (no matter how small) referring expressions,or whether there are also expressions which do not belong to the category.In view of earlier remarks concerning the distinction between referring andpredicating, the latter option seems the more likely one. In Keizer (1992b) Itherefore suggested using the third criterion (that of function) to determinecategory membership; that is to say, for an expression to be a (prototypicalor non-prototypical) referring expression, it must fill argument position.10 Itwill be clear that this approach not only makes it possible to use the termnon-referring meaningfully, but also leaves intact the important distinction

191

between the basic acts of referring and predicating (see also Keizer 1992b:360).

2.4. Rijkhoff (2002)

This section will be concerned with one of the several additions and modi-fications to the theory of FG introduced by Rijkhoff (2002): the introduc-tion of an extra variable to refer to the linguistic expression (in this casethe noun phrase) itself. In his proposal for a layered structure of nounphrases, Rijkhoff (2002: 227-229) distinguishes between X and x: the for-mer, as the referent variable, symbolizes the referent of the noun phrase,whereas the latter, the NP or term variable, symbolizes the noun phraseused by the speaker to refer to a discourse entity (Rijkhoff 2002: 228).This, Rijkhoff (2002: 229) explains, brings out the dual function of nounphrases: they have a referential side (concerned with the entity that istalked about and figures in the world of discourse) and a descriptive side(concerned with properties of this referent as a spatial entity).Rijkhoff uses this distinction to tackle the problem of identity state-

ments. Thus, he argues (following Frege 1977), in sentence like (10), wehave two different noun phrases (xi and xj) referring to the same referent(Xi).

(10)

A further advantage of such an analysis, Rijkhoff points out, is that theextra variable enables us to specify in the formal representation whetheranaphoric reference is made to a noun phrase (x) or to the referent of anoun phrase (X). Consider in this respect examples (11a and 11b) (whereA = anaphoric operator):

(11) A:he

B: him

In this example the pronouns and refer to the discourse entity (Xi),whereas refers to one of the speakernoun phrase indexed as (xj).

192

Both examples, however, turn out to be problematic. First of all, themost likely interpretation of (10) is one in which the hearer is assumed tobe unaware of the fact that both descriptions can be used to refer to oneand the same entity. Or rather, since entities are mental constructs, it maybe more correct to say that the speaker realizes that there is a discrepancybetween the mental representation of the hearer and his/her own, and, ap-parently convinced that his/her own representation is c-cordance with the speakerhis/her mental representation accordingly. In other words, in the hearermental representation, the two descriptions correspond to two differententities. The speaker, to achieve his/her goal, takes the hearer c-tive, thus using the two expressions to refer to two different mental enti-ties. This then also accounts for the fact that the sentence is informative:rather than Xi = Xi, the sentence states that Xi = Xj, the intended resultbeing a mental world in which the two descriptions apply to one and thesame entity (see also Keizer 1992b: 97).In (11B) we are faced with a different problem. According to Rijkhoff,

the pronoun refers back to the noun phrase. Closer examination, how-ever, shows that does not refer back to the noun phrase, but only to therestrictors within the noun phrase, i.e. the string . Thisis clear from the fact that the anaphoric reference does not include anyterm operators (note that replacing by the relevant descriptive elementwould yield an indefinite noun phrase). In other words, the anaphoric rela-tion does not hold at the level of the noun phrase, but at the level of thepredicates within this noun phrase. In FG, this can be captured by analys-ing the anaphoric relation at the f-level, which means that (for the represen-tation of this particular example) there is no need to introduce the extraNP-variable.

As the name suggests, the newly developed grammatical model of Func-tional Discourse Grammar (e.g. Hengeveld 2004a, 2004b, 2005; Hengeveldand Mackenzie 2005, 2006) is to be regarded as an expansion of FunctionalGrammar (FG) from a sentence grammar into a grammar that is also sensi-tive to discourse features. It has a number of novel features, some of whichturn out to be relevant to the present discussion. One of these concerns thetop-down nature of the model, which takes as its point of departure thecommunicative intention of the speaker. This intention, however, is not

193

part of the grammatical component itself; instead it is conceived at aprelinguistic conceptual level. Information from this conceptual level func-tions as input to the grammatical component, where it triggers the formula-tion of linguistic activity. In addition to this conceptual component, thegrammatical component is also linked to a contextual component, withwhich it interacts at various stages during the production process, and withan output component, which takes as its input the fully specified underly-ing representations produced by the grammatical component and convertsthese into the perceivable (spoken or written) expressions.A final important feature is the model r-

ticular the distinction between a pragmatic module, which converts theprelinguistic, conceptual information into representations at the interper-sonal level, and a semantic module, resulting in representations at the rep-resentational level. Thus, the interpersonal level accounts for all the for-mal aspects of a linguistic unit that reflect its role in the interactionbetween speaker and addressee Hengeveld and Mackenzie 2006: 671),while the representational level for all the formal aspects of alinguistic unit that reflect its role in establishing a relationship with the realor imagined world it describes, i.e. it concerns designation rather than evo-cation, the latter being the job of the interpersonal level Hengeveld andMackenzie 2006: 673).For referring expressions this means that they are no longer thought of

as being built up in bottom-up fashion from the various predicates (nowlexemes), but instead as originating as (sub)acts of reference at the inter-personal level. In other words, what we see is a separation between thepragmatic act of referring (representing the speakerand the semantic means selected by the speaker to transmit this intention(in this case, the grammatical and lexical clues provided by the speaker toenable identification of the intended referent by the hearer).At both levels the relevant units are represented by different variables

(see e.g. Hengeveld 2004a, 2005; Hengeveld and Mackenzie 2006). At theinterpersonal level, each discourse act (A1) consists of an illocution (F1),the two speech participants (P1 and P2), and a communicated content (C1).The communicated content, which contains all the information the speakerwishes to transmit in his/her communication with the addressee, containsone or more subacts. These subacts come in two kinds: a subact of ascrip-tion (T1), which reflects an attempt by the speaker to evoke a property, anda subact of reference (R1), reflecting an attempt by the speaker to evoke areferent. At the representational level we find the following (partly famil-iar) variables:

194

(12)Individual xState-of-Affairs ePropositional Content pProperty/relation fLocation lTime t

The general schemas for the representation of referring expressions at bothlevels are given in (13):

(13) IL: ( R1 [ 1): (R1))RL: ( i: i): ( i))

where and stand for the relevant operators, and for the relevantmodifiers, and for the functions of the respective units, and, finally, R1for the subact of reference and i for any of the variables listed in (12).Now, it will, of course, be interesting to see how these innovations af-

fect the analysis of copular constructions. First of all, the distinction be-tween a pragmatic level and a semantic level makes it possible to accountfor the fact that one and the same description at the representational levelcan be used to perform different kinds of subact at the interpersonal level(Hengeveld 2005: 13; see also Hengeveld 2004a: 12; Hengeveld andMackenzie 2005). Consider in this respect the sentences in (14) and (15)(from Hengeveld and Mackenzie 2006: 674):

(14) a.(Ascription of zero-order entity: T/f)

b.(Reference to zero-order entity: R/f)

(15) a.(Ascription of first-order entity: T/x)

b.(Reference to first-order entity: R/x)

In his analysis of these examples, Hengeveld distinguishes between thetype of subact at the interpersonal level (T or R) and the type of entity des-ignated at the representational level (f or x). Thus, in examples (14a) and(14b), the same property (

195

) in (14a), but referred to in (14b). Similarly, in (15), one andthe same the nominal expression, , fulfils an ascriptive func-tion in (15a), but a referring function in (15b). In both instances,Hengeveld continues, the type of entity is the same: in (15a) a first-orderentity is ascribed, while in (15b) a first-order entity is referred to(Hengeveld 2005: 13).11

This, however, leaves us with the same problem noted in Keizer(1992b) (see Section 2.3). According to the analysis offered here, the ex-pression in (15a) is used to ascribe a first-order entity to thereferent of the subject noun phrase ( ). First-order entities, however,cannot be ascribed; only properties can be ascribed. First-order entities areconcrete objects, with their own physical properties (colour, size etc.); onecan ascribe these properties to a concrete object, but one cannot ascribeone concrete object to another. It is, therefore, much more plausible toanalyse sentence (15a) at the representational level as involving only onefirst-order entity (the individual Sheila) and one property assigning element( ). The difference between in (15a) and (15b)thus not only concerns the type of subact performed at the interpersonallevel, but also the type of entity involved at the representational level (first-order versus zero-order). This leads to the following representations:

(16) a.(Ascription of property: T/f)

b.(Reference to first-order entity: R/x)

What causes the confusion is the fact that in both cases the same form ofexpression is used: a noun phrase. Now, as pointed out by Hengeveld (thisvolume), the prototypical noun phrase has a nominal head, is lexically des-ignating,12 denotes a first order entity, and is referential. An example is thenoun phrase in (15b)/(16b). The noun phrase in (15a)/(16a),on the other hand, is not prototypical: here is not used refer-entially (in terms of the definition of referring expressions in (9)) and doesnot denote a first-order entity. What we are dealing with, therefore, is anon-referential, non-prototypical noun phrase, which functions to assignthe complex property (or combination of properties) to afirst-order entity. This non-prototypical status is also reflected in the factthat, despite first appearances, the expression does not haveall the formal properties of noun phrases either (witness the use of differentanaphoric pronouns; see example (8) above). At the same time, it will be

196

clear that the subact of ascription is carried out by non-prototypical meansat the representational level: whereas an ascriptive subact is typically per-formed by means of a verb, adjective or bare nominal, here it takes theform of a noun phrase.Finally, let us consider the effect of the new features of FDG on the

analysis of identity statements. Originally, the FDG treatment of identitystatements was similar to the FG treatment proposed in Hengeveld(1992b). Consider once again example (6), here repeated as (17):

(17)

In Hengeveld (2004a: 15), we read thatprosodically prominent copula, serve the purpose of stating that the act ofreferring to an object by using a certain name is equivalent to the act ofreferring to that same object by another name; hence they are statementsabout the validity of acts of reference. rpreted asmeaning that (17) consists of two identical subacts of reference, i.e. Ri =Ri. Obviously, however, this is logically impossible: every linguistic act isuniquely situated in time; even the exact repetition of a linguistic act re-sults in a new act. Hengeveld and Mackenzie (2005: 16) therefore offer amodified analysis, according to which identificational sentences like (18)(i.e. Declerck ifying sentences) are regarded asconsisting of two different subacts of reference at the interpersonal level,but as involving only one entity at the representational level (see alsoHengeveld this volume):

(18)IL: (R1: [+S, -A] (R1)) (R2: Peter (R2))RL: (x1) (x1)

The distinction of two separate R-subacts at the interpersonal level isclearly an improvement on the original treatment of identificational sen-tences. Nevertheless, the proposed treatment is still problematic. First ofall, we still have an equation of two identical entities at the representa-tional level. As pointed out before, however, for such a construction to becommunicatively informative, the two descriptions, though possibly apply-ing to one and the same entity in the speakereven the e-sentation of the hearer. Secondly, sentence (18) is not an identity statement(unless the copula is given prosodic prominence). This is important, since

197

the two types of construction seem to differ exactly in the status of thepost-copular expression.To try and solve (part of) the puzzle, let us begin by distinguishing vari-

ous kinds of ictional ormythical world something thatsituation or any particular speech participants. Secondly, there are thespeech participantsthese will show considerable overlap, but there will always be differencesbetween them. Thirdly, there is the discourse world (or discourse domain,Vet 1986), consisting (among other things) of entities introduced into thediscourse by textual means, as well as entities present in the immediatediscourse situation. In the model of FDG this discourse world is repre-sented by the contextual component (see Connolly 2004: 103; Hengeveld2004b: 369). In what follows, I hope to show that a further distinction hasto be made between the discourse world (consisting of entities introducedinto the discourse by linguistic means) and the immediate situation (per-haps better regarded as part of the real or some fictional or mythicalworld).13

Now, let us see what happens in the case of an identity statement. Ac-cording to Dik (e.g. 1997a: 3), language is c-tion between human beings, used with the intention of establishing com-municative relationships ommunicative relationships, in turn, areestablished with the aim of bringing about a change in the knowledge (orpragmatic information) of the addressee: c-tion both S and A possess a huge amount of pragmatic information, PS andPA, respectively. In saying something to A, Smodification in PAhave a more or less detailed picture of the pragmatic information of A.This estimate of PA is thus part of PS, but clearly not of the discourseworld.We can apply this to identity statements by placing them in a context.

Consider in this respect the dialogues in (19) and (20).

(19) (Two people staring at the sky:)A:B:A:

198

(20) A:B:A:

In both cases, we have (at least) three different worlds or domains of refer-ence. Consider (19). Here we have an immediate situation shared by A andB, which includes a clearly visible, and presumably uniquely identifiableobject x1, referred to by A as and by B as . In addi-tion, there is the pragmatic information of the two participants, PA and PB.This pragmatic information can best be seen as mental representations ofthe real world (as well as of any number of fictional or mythical worlds).Since for both A and B the object referred to forms part of the real (visible)world, it can also be assumed to form part of their respective mental repre-sentations of this world.However, despite the fact that the immediate situation includes only one

element x1, the expressions used to refer to this object, andenter two entities into the discourse world. This situation corre-

sponds to PB, which, judging from Bdiscourse includes two entities: one labelled (an objectalready familiar to B) and now also one labelled (an object B is notfamiliar with, and which he or she assumes to be a separate entity). Thediscourse world does not, however, correspond to A a-tion, in which the labels and apply to one and thesame entity. Thus A, believing his/her representation to be correct, sets outto bring about a change in the mental representation of B. To achieve this,A utters an identity statement with the intention of bringing Binformation (i.e. Bhis/her own. In this identity statement reference is made to two differentdiscourse entities, the intended result being a modified PB in which the twodiscourse entities are represented as one entity.The example in (20) can be accounted for in much the same way, the

only difference being that here reference is made not to something in theimmediate situation, but to a (largely shared) fictional world. Once again,what are equated are not two identical entities, nor the descriptions used torefer to these entities, but two different discourse entities which, in someother world or domain, correspond to one and the same entity.Let us finally return to example (18). As pointed out before, this is not

an identity statement. The intention of the speaker is entirely different.Instead of equating two discourse entities, the speaker simply enters a newlexeme (a new label) into the discourse, which applies to the (already iden-

199

tifiable) referent of the subject noun phrase. This means that the expressionis not used referentially here: this sentence consists of one subact of

reference (performed by means of the pronoun ) and one subact of ascrip-tion (assigning the name to the individual referred to by ). I willreturn to the use of proper names in constructions of this kind in Section 5.

In the previous sections I have indicated a number of weaknesses in FGand FDG treatments of copular constructions and have tried, within thenew framework provided, to offer some suggestions for a more consistenttreatment of these constructions. The treatment eventually proposed differsfrom existing analyses mainly in the way in which the post-copular term isrepresented at the representational level: as a property rather than an entityin classificational (or predicational) sentences, and as a separate discourseentity in identity statements. The present section will address the morefundamental question of what exactly the variables at the representationallevel stand for, and, crucially, whether these variables should, in fact, bepart of the grammatical component. This will be the subject of Section 4.1,which will present a new proposal for the interpretation of variables at therepresentational level. In Section 4.2 this proposal will be applied oncemore to classificational sentences and identity statements.

4.1. Variables at the representational level

The grammatical component includes information relating to the formalaspects of linguistic expressions. These expressions are regarded as beingbuilt up, in a modular fashion, from linguistic primitives of various kinds(frames, lexemes, operators etc.). If, however, the grammatical componentis meant to contain linguistic information only, it follows that the variablesused within this component can only be interpreted as representing linguis-tic entities, i.e. either the linguistic primitives selected during the produc-tion of an utterance or the larger linguistic units built up from these primi-tives.At the interpersonal level, this requirement seems to be met. At this

level representations contain all (and only all) the linguistic informationneeded to reflect an expression ction. Thus, the dis-course act, represented by the variable A, is a linguistic unit, brought into

200

existence through the process of language production. The same can besaid of all the other units distinguished at the interpersonal level.14 The factthat these entities, once created, enter the discourse world (feed into thecontextual component) is irrelevant in this respect: they do not exist priorto the act of communicating and are, as such, linguistic entities. Nor doesthe morphosyntactic level form a problem in this respect. Here, too, theunits represented (phrases, clauses) are clearly linguistic in nature, theoutcome of the linguistic actions performed at the preceding stages of lan-guage production. Note that it is at this level that we now find the actualexpressions used to refer to some entity (cf. Rijkhoff -variable).But what about the variables at the representational level? In Hengeveld

(2004a: 6) we read that ative intention thespeaker will in most cases have to fill his utterance with basic semanticcontent, i.e. with descriptions of entities as they occur in the non-linguisticworld.representational level. Thus, what we should find here are the semanticunits used to describe extra-linguistic entities, not the extra-linguistic enti-ties themselves. In other words, the variables at the representational levelcannot be interpreted as representing extra-linguistic entities (of whatevertype): the extra-linguistic entities described at this level are part of the con-ceptual or contextual component, not of the grammatical component.It is, in fact, not difficult to see why extra-linguistic entities cannot be

part of the grammatical component. After all, as pointed out before, therelation between the description used and some (intended or selected) ref-erent is one of intention (in the case of the speaker) or interpretation (in thecase of the hearer). The relation itself is not established through language:the linguistic expression merely serves to enable the hearer to identify theintended referent (or restrict the set of intended referents). To achieve this,speakers will choose descriptions which are likely to lead the hearer to theintended referent, but this does not mean that the description actually ap-plies to this referent (in the real world or in the discourse world); nor doesit mean that reference will be successful. Neither the intended referent northe relation between the linguistic description and this referent are part ofthe grammar; as such, neither ought to be represented at the representa-tional level.This, of course, is a major departure from standard FG, where variables

symbolized the intended referent(s) of an expression. Recently, however,the problem of what exactly the variables at the representation level standfor has been acknowledged, the result being a different conception of therole of these variables in FDG, namely as indicating the type of entity de-

201

scribed by an expression (Hengeveld, p.c.). Now, although this is an inter-esting idea, it is not unproblematic. First of all, it would mean that the vari-ables at this level no longer actually represent an entity, but merely serveas additional instructions to the hearer. If so, it becomes difficult to defendthe presence of an index. After all, if the variable merely reflects the typeof entity referred to, it will suffice to choose the right type of variable:since the types (orders of entities) themselves are unique, indexing be-comes superfluous. Finally, indicating the type of entity described is notwhat variables are for: within FDG this is typically the kind of (non-lexical) information that is provided by operators (but operators on what?).I would therefore like to suggest a different use of the variables at the

representational level; a use which combines the useful ingredients of ear-lier approaches while respecting the function of variables and the nature ofrepresentation at the representational level. What I would like to propose isto regard the variables at the representational level as symbolizing what Iwill call a speakerextension set is, of course, not new going back at least as far as JohnLocke (1690) and is generallydefined as the set of entities to which the property described by a lexemeapplies (e.g. McCawley 1981: 154; Hanna and Harrison 2004: 232; Allen2001: 72).15 However, the exact nature of the relationship between lan-guage and reality has been an important subject of discussion in the phi-losophy of language from antiquity to the present. In their very detaileddiscussion of this debate, Hanna and Harrison (2004) distinguish two broadcamps: the Russellians and the Wittgensteinians.16 The former are advo-cates of referential realism, which insists that there is a direct relation be-tween a word and (some element in) the world, and that meaning is derivedfrom this relationship. Wittgensteinians deny the need for such a link, be-lieving instead that meaning exists independently of the external world. Ona radical relativistic view, people have no access to reality at all, only tosome mental representation of reality. In what follows I will adopt a mod-erate relativist view, according to which the relation between a linguisticexpression and some extra-linguistic element is indirect, i.e. mediated by alevel of mental (or conceptual) representation.17 Instead of the term exten-sion set I will, therefore, use a similar but nevertheless crucially differentterm, that of speaker 18

202

(21)The entire set of mental entities to which, in the view of speaker, theproperty (or properties) described by the lexeme(s) used in an ex-pression applies (apply).

As such, this mental extension set is not a set of external, extra-linguisticentities first of all because a speaker cannot possibly know all the mem-bers of this set (which would mean that the speaker him/herself does notknow what he/she is referring to); and secondly because, even if this werepossible, no two speakers would ever select the same set (in which casecommunication would become impossible). Instead, this mental extensionset can be seen as a linguistic entity, in that it forms part of the long-termlinguistic knowledge of a speech participant, being determined by themeaning definitions of the linguistic elements in question in the speakermental lexicon.19 This does not, of course, mean that this information iscompletely invariable. First of all, the meanings of lexemes can change,lexemes can obtain additional meanings, and new lexemes are introducedinto the lexicon. As part of the linguistic knowledge of a speech partici-pant, however, the mental extension set of a lexeme forms a relatively con-stant factor.Regarding the variables at the representational level as symbolizing the

mental extension set of the lexical elements (restrictors) used in an expres-sion means that it no longer represents the intended referent (set), but theentire class of mental entities from which the intended referent (set) is tobe selected. Thus, whereas the selection of the referent set of an expressionmay require further (context-dependent) information, including the kind ofinformation provided by operators, the mental extension set exists inde-pendently of these operators (i.e. the operators have scope over the exten-sion set) and is determined by lexical restriction only20 (Lyons (1977) usesthe term tterance-independentthe lexemes includes information about the type of entity described. It istherefore possible to use the existing range of variables to reflect this in-formation; it is, however, not necessary to do so.Consider the simple noun phrase in (22) as an example:

(22) a.b. (1 prox x1: [f1: dog])

Here the term variable x1 symbolizes the mental extension set of the de-scription, i.e. the entire set of mental (first-order) entities to which the lex-

203

eme dog applies. The term operators r-mation about the intended referent; information which the speaker assumeswill be required for the hearer to pick out this referent. The intended refer-ent itself, however, is not represented at the representational level but ei-ther forms part of (or is introduced into) the contextual component or isassumed to be retrievable from long-term pragmatic information of thehearer. The relation between the description and the referent remains amatter of intention (speaker) or interpretation (hearer) and is therefore notcoded in the grammar model.In this respect the approach suggested here differs crucially from the

representation proposed by Connolly (2004). Connolly, too, assumes thatthe contextual component includes, among other things, the referents of theterms used in the discourse (Connolly 2004: 103). Where the present pro-posal differs from Connolly ation approach is that inhis representations the variables at the interpersonal, representational andmorphosyntactic levels are all co-indexed with (i.e. uniquely linked to) thenon-linguistic entities at the contextual level. Thus we read that (Connolly2004: 108)

The line 251, x252)individuals x251, x252 (which are primarily constituents of the representa-tional level, and given shape as JayN and KayN at the expression level) arereferring to certain phenomena in the context. These referents are enumer-ated in the contextual description.

In the present proposal, by contrast, it will be assumed that the actual rela-tion between the linguistic expressions and their intended or selected refer-ents is not coded in the language, and should as such not be represented atany level in underlying representation. The subact of reference at the inter-personal level merely reflects the speaker n-tity, but is not in any way linguistically linked to that referent. Similarly,the lexical and formal features provided at the representational and mor-phosyntactic levels merely serve to help the hearer in his/her attempt topick out the intended referent(s). The reference, may, however, not be suc-cessful, in which case the wrong entity is selected, or no entity at all. Thisis possible exactly because the identity of the extra-linguistic entity is notpart of the linguistic expression.The links between expressions and extralinguistic entities will, of

course, be part of the speech participants ntal representations of thediscourse world. As argued earlier, however, these need to be distinguishedfrom both the grammatical and the contextual components: the grammati-

204

cal component contains only (largely shared) linguistic information, andthe contextual component consists of shared knowledge only (identity ofthe speech participants, time and place of the discourse, (perceivable) enti-ties in the immediate situation), while a speech participant e-sentation of the discourse world may include much more information andmay differ considerably from that of other speech participants.Next, I would like to argue that for the proposed use of variables at the

representational level it is still useful to provide these variables with anindex. Different (combinations of) lexemes introduce different mental ex-tension sets and there is reason to assume that these need to be distin-guished (uniquely) within the grammar model. Consider in this respectexamples (23) and (24).

(23) A:B: they

(24) A:B: They

So far the use of the italicized anaphoric pronouns in these examples hasalways been somewhat puzzling. Since they obviously do not refer back tothe discourse referents explicitly introduced in the preceding sentences (

and ), their use could only be explained interms of inference, with the pronouns referring to all the members of aclass inferred from the specific referent. We can now explain this use ofthe definite pronoun by assuming that it is meant to refer to all themembers of the mental extension set introduced by the noun phrases

and . For this to be possible, these exten-sion sets must have been introduced as discourse entities into the contex-tual component, thus becoming available for anaphoric reference; as such itmay be argued that, like all other linguistic elements, these extension setsneed to be distinguishable from each other.Note finally that such an analysis confirms the idea that mental exten-

sion sets are determined by lexical restrictors only: what is included in theanaphoric reference in (23) and (24) is the information supplied by thenominal and adjectival lexemes ( and / , respec-tively), not the information provided by the term operators.

205

4.2. Copular constructions revisited

In this section I will describe the implications of the new proposal for theanalysis of copular constructions. We saw that, as far as classificationalcopular constructions are concerned, the main problem consists in the factthat it is difficult to reconcile the function of the post-copular expressionwith its linguistic form. Since these expressions are used in an ascriptivefunction, they may be assumed to describe a property: after all, it is proper-ties that are ascribed. Thus, in a sentence like , the nounphrase is felt to describe a property to the individual referred to asCharlie. Such an analysis is supported by the fact that a separate set ofpronouns is used to indicate anaphoricity with these expressions (the in-definite pronouns , as well as the neuter pronouns , and incases of an antecedent describing a person; again see example (8) above).At the same time, the form (a noun phrase, even if non-prototypical) sug-gests the presence of a (first-order) referent.Now let us look at expressions of this kind in the light of what has been

proposed in this section. Consider the following example:

(25) A:B:

As in examples (23) and (24), the definite pronoun here seems to takeas its referent a mental extension set; in this case the extension set intro-duced by the post-copular expression . This clearly suggeststhat post-copular expressions do have an extension set, even if the descrip-tion as a whole does not refer to a specific entity from that set. We can nowexploit the f-variables and the (newly defined) x-variables at the represen-tational level to represent this situation:

(26)IL: (T1)RL: (f1: [mx1: [f2: car]: [f3: old]])

At the interpersonal level, the post-copular expression is analysed as ansubact of ascription (T1). At the representational level, we find a complexproperty (f1), restricted by the mental extension set x1 ( ), which inturn is determined by two properties, (f2) and (f3). The extension setis specified by the plural operator, thus giving the noun phrase .Now the use of different anaphoric pronouns can be accounted for. The

206

pronouns , , and , in example (8) take as their antecedent thecomplex property ascribed to the referent of the subject noun phrase (i.e.f1). The pronoun in (25) takes as its antecedent the extension set of thedescription in post-copular position (i.e. x1). Finally, the indefinite pronoun

can be used to replace only the property ascribed by the nominal firstrestrictor (i.e. f2, as in ).The approach advocated also allows for a more insightful analysis of

identity statements. Consider once more example (27) (cf. example (19);CC = contextual component):

(27)IL: (R1) (R2)RL: (x1: [f1 = (x2: [f2CC: entity-1 = entity-2

The representations in (27) no longer involve any contradiction. Sentencesof this type do not involve the equation of two subacts (there is no equationat the interpersonal level), nor of two identical entities. Instead, what isasserted is that the mental extension sets of the two descriptions (at therepresentational level) as well as their respective discourse referents (in thecontextual component), taken to be different by the hearer, are (accordingto the speaker) identical. A sentence like (27) is thus informative in thesense that it is meant to change the pragmatic information of the hearer,both linguistic (equation of two long-term mental extension sets) and non-linguistic (equation of two discourse entities). This shows that, althoughprimitives (in this case lexemes) are part of long-term knowledge, they arenot invariable. Different people may attach different meaning definitions tocertain lexical elements.More generally, this means that, for the FDG-model to have some kind

of psychological adequacy, it must be assumed that the primitives used forcreating linguistic expressions (not only lexemes, but also frames, tem-plates, operators etc.) are those available to the speaker, i.e. those primi-tives that are part of the linguistic pragmatic information of the speaker.Obviously, most of these linguistic primitives are shared by all adult nativespeakers of a language. Nevertheless, there are always borderline cases,where native speaker judgements vary. This can only be accounted for byassuming speakers to have access to slightly different sets of primitives.Linguistic knowledge, in other words, does not differ from any other kindof long-term knowledge: it varies from person to person and can be modi-fied and extended.

207

Note, however, that the change in pragmatic information brought aboutby an identity statement need not be linguistic in nature. In sentence (27)above, the modification intended concerned a change both in long-termlinguistic and in long-term extra-linguistic (contextual) information: thenew information being that the mental extension of the lexeme isidentical to that of the lexeme , and that the two correspond-ing discourse entities are, in fact, one and the same. In other cases, how-ever, the change is only a short-term and contextual one. Consider (28):

(28)IL: (R1) (R2)RL: (x1: [f1 = (x2: [f2CC: entity-1 = entity-2

Here, again, the speaker wishes to convey the information that two dis-course entities, which correspond to two separate entities in the hearermental representation of the real word, are one and the same entity. Thespeaker thus asserts that in the given context (note the use of context-dependent elements such as and ), the two expressions are co-extensive. However, since the descriptive elements involved clearly havedifferent intensions (or meanings), they can be used to refer to differententities in other contexts. As such, the descriptions in (28) are merely usedas an instruction to the hearer to find the two discourse referents in ques-tion; their (long-term) mental extension sets are not equated.

The analysis of descriptions at the representational level proposed in theprevious section can also be fruitfully applied to various other types ofconstruction. In what follows we will look at some of these constructions:proper names and pronouns in Section 5.1 and appositive constructions(restrictive and non-restrictive) in Section 5.2.

5.1. Proper names and pronouns

Let us first of all consider the analysis of proper names. According to re-cent proposals (Hengeveld this volume), referentially used proper namesare analysed as referential subacts restricted by the proper name (a lexeme)

208

at the interpersonal level. At the representational level, these constructionsare represented by means of the term variable only:

(29)IL (R1: Peter)RL (x1)

This analysis reflects the idea that proper names have a pragmatic function(identification of a referent) but are semantically empty and should there-fore not be represented at the representational level. Application of theanalysis proposed in Section 4, however, leads to the following representa-tion in (30):

(30)IL (R1)RL (x1: Peter)

Here, the proper name is represented at the representational level, as it isregarded as fulfilling a semantic (restrictive) function. It is true that propernames fail to assign a property, but they do have an extension set (all theentities known by the name Peter).21 Moreover, they do provide the hearerwith semantic information about the intended referent (such as gender).22

Their lack of descriptive meaning is reflected in the fact that they are notprovided with an f-variable. In this way we can account for the fact thatanaphoric reference can take the form of a definite pronoun, but not of theindefinite pronoun .In addition, we are now in a position to provide a more plausible analy-

sis of example (18), which, using a somewhat simplified form, was repre-sented as follows:

(31)IL: (R1) (R2: Peter)RL: (x1) (x1)

As the intention of the speaker is not to equate two first-order entities, weare not dealing with an identity statement here; nor is the post-copular ex-pression used to assign a property to the referent of the subject. Instead,

is used as a special type of ascription: rather than instructing thehearer to assign a property, it instructs him/her to attach the label toan existing discourse entity; or, to put it differently, to expand the mental

209

extension set of the lexeme to include the referent of . If variables atthe representational level stand for the extension set of an expression, thismeans that the variables of the two nominal expressions in (31) should notbe co-indexed, as the extension sets of the two expressions are not identi-cal. I would therefore like to suggest the following representation instead:

(32)IL: (R1) (T1)RL: (x1) (x2: Peter)

Unlike proper names, the extension sets of first person singular and secondperson pronouns can be thought of as being fully determined at the inter-personal level, where will automatically be identified as the speaker and

as the hearer(s). It seems therefore justified to analyse these pronounsat the interpersonal level as restrictors on the R-subact. The status of thefirst person plural pronoun is a bit more problematic, since referenceincludes individuals other than the speaker referents whose identity is notdetermined by the reference act itself. The same is true for the third personpronouns, which provide the minimum of lexical information needed tolead the hearer to some intended referent (set). For these pronouns, repre-sentation at the representational level may seem a more plausible option.The fact that the pronouns and provide semantic information (ongender) seems to support such an analysis.23

5.2. Restrictive and non-restrictive apposition

Keizer (2005, 2007) gives an analysis of the form and function of restric-tive (or close) appositions of the kind ,

, , etc. Although the different subtypes callfor slightly different analyses, it is argued that all close appositions oughtto be regarded as involving one referring expression (the appositional con-struction as a whole), which consists of two non-referential nominal con-structions. Furthermore, it is argued that, irrespective of the type of closeapposition, the first element functions as the head and the second elementas a modifier.Translated into FDG terminology, this means that an expression like

is the result of a single R-subact at the interper-sonal level, while at the representational level we find two elements: aproper name and a description. Since the extensions sets of the two ele-

210

ments are clearly not identical, they will be represented by different vari-ables (for arguments see Keizer 2005, 2007). Once again, the proper nameis analysed as directly restricting the x-variable: since it does not assign aproperty, it is not provided with an f-variable. Observe also that, in accor-dance with proposals so far, the modifier is analysed as a T-subact withinthe R-subact.

(33)IL: (R1: (T1) )RL: (x1: (x2: [Orson Welles]) (x3: [f1: actor]) )

(34)IL: (R1: (T1) )RL: (x1: (x2: [f1: actor]) (x3: [Orson Welles]) )

The present proposal also proves useful in the representation of non-restrictive apposition constructions. Hannay and Keizer (2005) offer a dis-course-based classification of non-restrictive nominal appositions in Eng-lish, along with a (provisional) FDG-analysis of the various types distin-guished. In the analysis proposed, the appositive element is in all casesrepresented as a separate discourse act, consisting of either a subact ofreference or a subact of ascription. In one of the types distinguished, thespeaker uses the apposition to reformulate the description provided in thehost element. An example can be found in (35) (from Hannay and Keizer2005: 173):

(35)

<ICE-GB:S2B-010 #131:1:A>24

Reformulation can serve a number of purposes. In (35) the non-restrictiveapposition is added as a simplification of the description usedthe host element ( ). In examples of this kind, the host is typically anR-act at the interpersonal level, while the apposition will be interpreted asa separate discourse act consisting of a T-act only. At the representationallevel, however, both expressions will be analysed as having the same men-tal extension set: both descriptions apply to the same set of entities,25 theonly difference being that the description provided in the appositive ele-ment is simpler or more accessible for the hearer than that provided in the

211

host element. The actual lexemes used in the two descriptions are, ofcourse, not identical. This leads to the following representation at the in-terpersonal and representational levels:

(36)IL: (A1: (R1) )Nucl (A2: (T1))SimplRL: (x1: [f1: jihad]) (x1: [f2: war] [f3 holy])

This article has addressed a number of issues related to reference and as-cription in FDG, all of which have turned out to centre around two mainquestions: (1) how to deal with reference and ascription in FDG, especiallyin copular constructions? and (2) what is the nature of representation, morespecifically of the x-variable, at the representational level? In order to an-swer these questions a number of basic assumptions of the FDG modelneeded to be made explicit. One of these is the assumption that in thegrammatical component only linguistic entities and relations are repre-sented. It was argued that this requirement is fulfilled at the interpersonallevel, where variables represent units that are created in the process oflanguage production. At the representational level, however, variables aretraditionally regarded as representing (discourse) entities; i.e. extra-linguistic entities. In other words, whereas the referents of the units distin-guished at the interpersonal level are part of the linguistic system, those atthe representational level are not.This does not mean, however, that the information provided at the rep-

resentational level is extra-linguistic. After all, it is at this level that thesemantic content (or meaning) of an expression is specified, which is, ofcourse, very much linguistic in nature. All that needs to be done to solvethe problem is, therefore, to have the variables at the representational levelrepresent the semantic content of expressions, rather than the entities re-ferred to. It was therefore suggested that rather than the intended referent,the variables at this level be regarded as symbolizing the mental extensionset of an expression. This mental extension set the entire set of mentalentities to which, in the view of speaker, the property (or properties) de-scribed by the lexeme(s) used in an expression applies (apply) is part ofthe long-term linguistic knowledge of the speaker and is not directly re-lated to any entity or entities in some (discourse) world; as such it fulfilsthe requirement of being linguistic in nature. Such an approach was shown

212

to have the additional advantages of being psychologically adequate and ofretaining representational and notational consistency over the various lev-els and modules of the model.In addition, the proposal calls for a strict separation of the grammatical

and contextual components. Whereas the former contains only linguisticentities, the latter includes the extra-linguistic entities referred to by aspeaker performing a referential act. Although the linguistic entities repre-sented at the representational level may themselves become (extra-linguistic) discourse entities (i.e., be fed into the contextual component),neither these extra-linguistic entities themselves, nor the relation betweenthe linguistic expressions used and these entities form part of the gram-matical component. Instead, the grammatical component is seen as consist-ing only of the (long-term) linguistic knowledge of the speaker, a consider-able part of which (in most cases) overlaps with (long-term) linguisticknowledge of the hearer.Finally, it has been shown that the approach suggested not only leads to

a more consistent treatment of copular constructions, but also proves usefulin dealing with proper names and pronouns, as well as with restrictive andnon-restrictive appositive constructions in English.

1. Thanks are due to Lachlan Mackenzie for useful comments on an earlier ver-sion of this article and to Kees Hengeveld for fruitful discussions on the sub-ject (which, however, have not yet led to consensus on all points; seeHengeveld this volume). I am also much indebted to the detailed comments ofthree anonymous referees, which, I hope, have resulted in a clearer presenta-tion of my ideas. Any remaining errors or inconsistencies are, of course, en-tirely my own.

2. In FDG, the term predication is reserved for the representational level, whereit is used to describe the unit designating a state-of-affairs. At the interpersonallevel, the term ascription is used in the sense of pertyof/assigning a property to some entitysubacts of reference and ascription that are being performed; the linguisticmeans to perform these acts are specified at the representational level.

3. This question is also addressed by Escribano (this volume), who describesfrom the logician w the problems involved in the F(D)G use ofvariables, and their relation to sense, reference and the choice of NP.

4. Following standard FG theory, the possessive pronoun is here a restrictor onthe x-variable. As argued by Mackenzie (1987), it might be more appropriate

213

to analyse possessive constructions as arguments (in the case of inherent rela-tions) or satellites (in the case of possession proper) of the nominal predicate,with the semantic functions Reference and Possessive, respectively. Such anapproach, the expression would be represented (again simpli-fied) as (d1x1: (best friend) (d1xj: I)Ref). See also Keizer (1992b, 2004).

5. Note, however, that in most classifications of copular constructions (includingDeclerck 1988), there is no direct link between the type of copular construc-tion and the (in)definiteness of the post-copular expression: generally speak-ing, it is assumed that both identificational and classificational constructionscan contain either a definite or an indefinite post-copular expression (see alsoe.g. Keizer 1992b, chapter 6).

6. The elements in this sentence can, of course, be reversed. In that case, how-ever, the sentence is no longer regarded as a characterizational sentence, but asa specificational sentence. As pointed out by Higgins (1979) and Declerck(1988), the two types of construction are used in different contexts and exhibitdifferent syntactic behaviour. Specificational sentences contain two referentialelements, one of which is weakly referential (a kind of variable: we know theentity referred to exists, but cannot identify it), the other strongly referential (avalue) (Declerck 1988: 5). The function of a specificational sentence is tospecify the value of the variable, i.e. to fully establish the identity of theweakly referential element. It is typically used in answer to some (explicit orimplicit) -question of the type a-phrased as The value (Y) therefore forms the focus ofthe sentence; as such it is given prosodic prominence and may appear as thefocus of an -cleft ( ). Finally, specifica-tional sentences have an ey imply that the list ofvalues satisfying the variable is exhaustive (Paris is the only capital of France).Hengeveld ntences seem to cover Declerck a-tional and descriptionally identifying sentences (Declerck 1988, chapter 1).They are used to assign a property (i.e. provide information, in this case

) to an entity ( ) and typically answer the questionis X (like)?/Tell me more about X ei-ther element is given focal stress (the sentence has a neutral intonation pat-tern), and -clefts are inappropriate. Finally, they are not exhaustive: Paris canbe characterized in many ways.

7. Note that Hengeveld (this volume) uses the notion of prototypicality in histreatment of noun phrases. According to Hengeveld o-typical noun phrase (a) has a nominal head; (b) is lexically designating; (c)denotes a first order entity; and (d) is referential. It is important to realize that,despite the overlap in defining features, we are dealing with two separate typesof category: where referring expressions form a functional category (partly de-fined in terms of form), noun phrases form a formal category (partly defined interms of function). See also the discussion in Section 3.

214

8. This is a simplified version of the definition given in Keizer (1992b: 139).9. It might be argued that fully referential NPs can also appear in satellite posi-

tion, as in , where the (fully referential) NPfunctions as a satellite term (with the semantic function of beneficiary).

Note, however, that the basic distinction between referring and predicatingapplies to the nuclear predication, and that within this nuclear predication, fullterms must indeed function as arguments. Moreover, if we assume that mostprepositions (including ) are predicates (e.g. Mackenzie 1992, Keizer toappear), it is not the term that functions as a satellite, but the preposi-tional phrase , where functions as the argument of .

10. This means that non-specific and generic expressions do belong to the cate-gory of referring expressions. They are taken to refer to some non-specific en-tity or a class of entities, respectively. As such they are not prototypical refer-ring expressions; they do, however, fill an argument slot (and, therefore, arebeing predicated about).

11. Note that on this approach, there is no longer a need for term-predicate forma-tion (see Garc and Hengeveld 2002).

12. As pointed out by one of the reviewers, this term is tautological, in the sensethat only (and all) lexical elements designate.

13. It goes without saying that the contextual component contains more than rep-resentations of discourse entities and entities present in the immediate context.As, in this article, I am concerned with the availability and identifiability ofdiscourse referents, I will restrict myself to that part of the contextual compo-nent that is relevant to the discussion. For further discussion of the contextualcomponent, see Rijkhoff (this volume); compare also Escribano l-ume) distinction between Physical World, Psychological Worlds, DiscourseWorlds, Speech-Act Worlds and Linguistic Worlds. For a discussion of the in-teraction between the contextual and the conceptual components, see Butler(this volume).

14. Note that including the speech participants (PS and PA) at the interpersonallevel is justified on account on the fact that what is relevant here are their rolesin the communicative situation, not the (extra-linguistic) individuals in ques-tion. Since these roles are fixed elements of any communicative setting, andsince their specific (discourse-dependent) features may be reflected in theformal aspects of a linguistic expression, their representation at the interper-sonal level seems to be justified (for an alternative view, see Rijkhoff this vol-ume).

15. The term extension is closely related to the notion ofto other related terms like intension, designation, connotation, meaning andsense; e.g. Frege 1892 [1977]; Mill 1856; Carnap 1956; Lyons 1977; for adiscussion, see e.g. Harder 1996: 16-29). Unfortunately, these terms have beenused in many different, but often very similar and partially overlapping ways,leading to what Geach (1970: 55) describes as

215

16. Apart from Bertrand Russell, the realist camp is represented by such philoso-phers as Willard V.O. Quine, Saul Kripke, Hilary Putnam, Donald Davidson,Michael Dummett and John McDowell; among the followers of Ludwig Witt-genstein we find Maurice Merleau-Ponty and Jacques Derrida.

17. For a detailed discussion of the cognitive dimension in meaning, see Harder(1996), Chapter 3.

18. Note that, despite its conceptual nature, thissame as the intension of a expression. The intension of an expression is gener-ally defined as and Har-rison 2004: 232), or, somewhat more specifically, as i-ates to each world the EXTENSION of that expression in that world.(McCawley 1981: 401). The intension, in other words, can be seen as themeaning definition of a lexeme, that which determines its (mental) extensionset.

19. One of the reviewers raised the interesting question of whether the concept ofextension can be applied to temporal (t) and locative (l) terms as well. He orshe correctly pointed out that in that case the difference between the expres-sions and , represented (in standard F(D)G) as (1x1: (f1:table)) and (1l1: (f1: table))Loc, respectively, cannot be derived from the mean-ing definition of the lexeme , but that the correct interpretation can be ar-rived at by analysing prepositions as lexical elements, with their own exten-sion set. I do, indeed, believe that this is the right approach (see Keizer toappear).

20. I deliberately avoid use of the term referent(s), since it is not clear tome whether the information provided by term operators plays a role in deter-mining this set.

21. This has, in fact, been the generally accepted view among logicians since JohnStuart Mill (1856): proper names lack sense but do have denotation; or, asCoates (2006: 371) puts it, properhood can be defined as tingNote that Coates himself prefers the term p-erhood as iscounts the sense of any lexical itemsin the expression that is used to do the referringdescriptive expressions like or canbecome proper if they are used merely to refer to an object, without ascribingany properties. However, since (synchronically) senseless proper names (suchas ) need not be used to refer (see example (32)), I still prefer the termsenseless denoting.

22. For a similar view, see Allan (2001: 87-99). Here proper names are seen asdesignators conveying a certain amount of lexical information (e.g. referent ishuman, female, etc.). It is true that they are typically used as a-tors 1972: 269), i.e. as designating the same object in any possibleworld (Allan 2001: 89). In that case, the proper name will function as a nounphrase (Allan 2001: 97). As a linguistic element, however, a proper name is a

216

noun, which means that it can be (and indeed often is) used as a regular firstrestrictor (see also Keizer 1992b: 97).

23. In logical semantics this may be an unusual point of view: here pronouns aretypically seen as mere pointers, devoid of any meaning. Within the pragmaticdiscourse-based tradition, as well as in cognitive approaches, on the otherhand, pronouns are seen as having a certain degree of lexical content. Allan(2001: 69), in an attempt to bring together the various views, describes pro-nouns as having none for the zero anaphor;

.24. This example is taken from the British component of the International Corpus

of English (ICE-GB), a fully tagged, parsed and checked one-million wordcorpus of written and spoken English, compiled and grammatically analysed atthe Survey of English Usage, University College London, between 1990 and1998.

25. Note that has, by now, two senses: apart from designating a ggleging war,

it has come to denote any conflict waged for principle or belief and is oftentranslated to mean -Webster Online Dictionary). Al-though in example (35) the expression can have either interpretation,the analysis provided here is based on the second sense of .

Allan, Keith2001 . Oxford: Blackwell.

Anstey, Matthew P.2002 Layers and operators revisited.

77.Butler, Christopher S.this vol. Interpersonal meaning in the noun phrase.

Carnap, Rudolph1956 . Chicago: University of Chicago Press.

Coates, Richard2006 Properhood. 82: 356 382.

Connolly, John H.2004 The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and Marmez-Gonz ,211 242. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

217

Dik, Simon C.1978 . (Publication in Language Sciences 7). Dor-

drecht: Foris.1980 . London/New York: Academic

Press.1989 . Dordrecht: Foris.1997a

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

1997b(Functional Grammar Series 21.) Edited by Kees

Hengeveld. Berlin/New York: Mouton de Gruyter.Declerck, Renaat1988 . Leuven:

Leuven University Press/Foris Publications.Escribano, Josthis vol. On noun phrase structure in F(D)G: Some conceptual issues.

Foley, William A. and Robert D. Van Valin1984 . Cambridge: Cambridge

University Press.Frege, Gottlob1977 Reprint. On sense and reference. In: P.T. Geach and M. Black (eds.),

1977,, 56 78. Oxford: Blackwell. First published 1892.

Garc2002 Do we need predicate fames? In: Ricardo Mairal Us

Jes, 95 123. (Functional Grammar Series

25.) Berlin/New York: Mouton de Gruyter.Geach, Peter T.1970 . Ithaca: Cornell University Press.

Gundel, Jeanette K.1977 Where do cleft-sentences come from? 53: 543 559.

Hanna, Patricia and Bernhard Harrison2004 . Cambridge: Cambridge University Press.

Hannay, Mike and M. Evelien Keizer2005 A discourse treatment of English non-restrictive nominal appositions

in functional discourse grammar. In: Mar -GonzJ. Lachlan Mackenzie (eds.),

, 151 185. (Linguistic Insights 26.) Bern: Peter Lang.Harder, Peter1996

Berlin/New York: Mouton de Gruyter.

218

Hengeveld, Kees1989 Layers and operators in Functional Grammar. 25: 127

157.1990 The hierarchical structure of utterances. In: Jan Nuyts, A. Machtelt

Bolkestein and Co Vet (eds.),, 1 23. Amsterdam/ Philadel-

phia: John Benjamins.1992a Parts of speech. In: Michael Fortescue, Peter Harder and Lars

Kristoffersen (eds.),, 29 55. Amsterdam/ Philadelphia: John Benja-

mins.1992b . Berlin/New

York: Mouton de Gruyter.2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzieand Mar -Gonz, 1 21. (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzie and Mar

mez-Gonz ,365 378. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

2005 Dynamic expression in Functional Discourse Grammar. In: Casper deGroot and Kees Hengeveld (eds.)

53 86. (Functional Grammar Series 27.) Ber-lin/New York: Mouton de Gruyter.

this vol. Prototypical and non-prototypical noun phrases in FDG.Hengeveld, Kees and J. Lachlan Mackenzie2005 Interpersonal functions, representational categories, and syntactic

templates in Functional Discourse Grammar. In: Mar de losG -Gonz

, 9 27. (Linguistic Insights 26.)Berne: Peter Lang.

2006 Functional Discourse Grammar. In: Keith Brown (ed.),2nd Edition, Vol. 4, 668-676. Oxford:

Elsevier.Higgins, F. Roger1979 . New York: Garland Pub-

lishing.Keizer, M. Evelien1992a Predicates as referring expressions. In: Michael Fortescue, Peter

Harder and Lars Kristoffersen (eds.),, 1 27. Amsterdam/ Philadelphia:

John Benjamins.

219

1992b Reference, predication and (in)definiteness in Functional Grammar. Afunctional approach to English copular sentences. Ph.D. diss., VrijeUniversiteit Amsterdam.

2004 Term structure in FG: a modest proposal.78.

2005 Close appositions. In: Casper de Groot and Kees Hengeveld (eds.),, 381 417.

(Functional Grammar Series 27.) Berlin/New York: Mouton deGruyter.

2007 .Cambridge: Cambridge University Press.

to appear English prepositions in Functional Discourse Grammar.15-2, 2008.

Kripke, Saul1972 Naming and necessity. In: Donald Davidson and Gilbert Harman

(eds.), , 353 355. Dordrecht: Reidel.Lyons, John1977 . Cambridge: Cambridge University Press.

Mackenzie, J. Lachlan1987 The representation of nominal predicates in the fund.

25.1992 English spatial prepositions in Functional Grammar.

46.McCawley, James D.1982

Oxford: Blackwell.Mill, John Stuart1856 . 4th ed. London: John W. Parker and Son.

Rijkhoff, Jan2002 . Oxford: Oxford University Press.this vol. Layers, levels and contexts in FDG.

Searle, John1969 . Cambridge: Cambridge University Press.

Strawson, Peter F.1971 Reprint. On referring. In: Peter F. Strawson (1971),

, 1 27. London: Methuen. First published 1950.Strawson, Peter F.1959 . London: Methuen.

Vet, Co1986 A pragmatic approach to tense in Functional Grammar.

16.

This article has two interrelated aims: firstly, to develop the account ofinterpersonal meaning in the noun phrase given, within the framework ofFunctional (Discourse) Grammar, by Hengeveld and Rijkhoff; and sec-ondly to illustrate some of the complexities which we must face up to if weare to develop an account which genuinely attempts to model the interper-sonal dimension of the language actually used by speakers and writers.The main body of the article is arranged in four major sections, with

progressively greater degrees of n-terpersonal meaning. and Rijkhoffwith respect to those properties of units realized morphosyntactically asnoun phrases which Hengeveld considers to belong to the interpersonallevel of the grammar. This comparison will lead to the recognition, inof two types of broadly interpersonal meaning (in units realized as clausesas well as in those realized as noun phrases), one oriented towards the so-cial and personal context of interaction, the other towards the discoursecontext (co-text).The remainder of the article focuses on interpersonal meaning oriented

towards the social and personal context. Indifferent types are discussed, from both English and Spanish. Focusingdown even further, -based study of just onetype of interpersonal modifier: the use of in English as an indicator ofthe speaker or writerthe noun phrase. This study is used to illustrate the richness of detailedpatterning which must be accommodated in a model which aims to showhow we exploit the resources of our language in order to carry out impor-tant functional tasks such as the expression of attitudes. Some conse-quences of such data for the FDG model are outlined. The article ends witha brief section (

222

The current FDG model (see Hengeveld 2005, this volume; Hengeveld andMackenzie 2006, forthcoming) consists of four components. The central

component interacts with a component whichresponsible for the development of both a communicative intention rele-vant for the current speech event and the associated conceptualization withrespect to the relevant events in the external real or imaginary world(Hengeveld 2005: 57), and with a component whichonly contain a description of the content and form of the preceding dis-course, but also of the actual perceivable setting in which the speech eventtakes place componentwhich ographic expressions(Hengeveld 2005: 57).The grammatical component consists of four levels: the

(pragmatic) level, which represents a linguistic unit in terms of the com-municative function it fulfils; the (semantic) level, whichis concerned with the ontological status of a linguistic unit; the

level (called the level in earlier work); and thelevel, housing phonological representations (Hengeveld 2005: 59 60).At the interpersonal level, the utterance is represented as a discourse

move, consisting of one of more acts, each specified for its illocutionaryforce and its communicated content, this latter consisting of ascriptive andreferential subacts.2 The schematic structure at the interpersonal level is asin (1) below (Hengeveld 2004a: 5, 2005: 63, this volume):

(1) (M1: [(A1: [ILL (P1)S (P2)A (C1: [ 1) (R1) 1))] (A1))] (M1))

Here, M represents a move, A an act, ILL the illocutionary force, P a par-ticipant in the discourse, C the communicated content, T a subact of ascrip-tion and R an subact of reference, and the subscripts S and A representspeaker and addressee, respectively.At the representational level, the schematic structure is as in (2)

(Hengeveld 2005: 64, this volume):

(2) (ep1: [(p1: [(e1: [(f1) (x1)] (e1))] (p1))] (ep1))

223

where represents an episode (a set of propositional contents joined to-gether in narrative texts), the propositional content, the state of affairs,a property or relation, and an individual.Hengeveld (2004b: 372) notes that the structures proposed in FDG

make available new slots for operators and modifiers, with respect to clas-sical FG, and that one of the areas in which this would appear to be usefulis the layered structure of the term phrase. He proposes a layered structurefor the referential subact in which there are three sets of operators andthree corresponding sets of modifiers. Hengeveld points out that the struc-ture he sets out has the same number of operator and modifier slots as thatproposed by Rijkhoff (1992), and claims that these slots can now be under-pinned with their functions at various layers.Hengeveld (this volume) proposes the maximal structure shown in (3)

for the standard or prototypical noun phrase, which is referential, has alexical head, is used to designate a concrete first order entity, and desig-nates by lexical means.

(3) ( R RI: 1):R (RI))

( x xi: (f f1: LexN (f1):

f (f1)) (x1):x (x1))

R operators/modifiers are concerned with the situating of the referentialsubact in the communicative situation, while x operators/modifiers relate tothe properties of the set of entities designated, and f operators/modifiersare concerned with the nature of the property f itself. At the R level wehave operators for identifiability and specificity, concerned with thespeaker arer knows about the referent,and also modifiers indicating subjective attitude; at the x level we haveoperators for location and number, and modifiers indicating referent modi-fication; and at the f level the operators are concerned with shape andmeasure, and the modifiers with reference modification.3

Hengeveld (2004b, this volume) points out that such a schema allows usto account for the different values of in , where the ad-jective modifies the property of being a driver (i.e. ( f) in the most recentrepresentation), , in which it selects a subclass of the entitiesreferred to by the head noun ( x), and , where the adjectiveencodes a subjective attitude on the part of the speaker ( R).In Rijkhoff (this volume), his earlier 4-layer structure for the noun

phrase (see Rijkhoff 2002: 227) is expanded into a 5-layer structure asshown in (4):

224

(4) NPi: 4[ 3[ 2[ - [ 0[ (NOUN (fi))(xi) ]Lc 0(Lc) ]L0 1(L0) ]L12(L1) ]L2 3(L2) ]L3 4(L3)

where subscripts 0 to 4 indicate the five layers, / represents a term (NP)operator and /T a term (NP) satellite (parallel to Hengeveldbeing a predicate variable symbolizing the property represented by thehead noun, and x a variable symbolizing the referent of the noun phrase.Operators at layer 0 in this new model correspond to what in the earlier(2002) model were labelled as

, i.e. the kind of thing (e.g. individual, collective) being referred toby the head noun; level 0 satellites relate to the of the type ofentity represented by the head noun, as in or .Both operators and satellites at level 0 take the core layer (Lc) as their ar-gument. There are now no operators at level 1, as indicated by the emptyslot at this point in 4; level 1 satellites are concerned with more or lessinherent properties of the referent normally expressed by adjectives in lan-guages which have them. Level 2 houses operators and satellites concernedwith quantity, i.e. number distinctions and cardinality. Level 3 accommo-dates operators and satellites which locate the referent in space and time.While levels 0 to 3 are descriptive, level 4 is concerned with the status ofthe referent in terms of the discourse itself, and so deals with definitenessand specificity and with notions of sameness, difference, etc.If we compare the latest schemes of Hengeveld and Rijkhoff, we see

that Hengeveld f operators correspond to Rijkhoff 0 operators, in thatboth are concerned with specifying the nature of the property f itself.Hengeveld x operators, however, cover two of Rijkhoffnamely 2 operators concerned with quantity and 3 operators relating tolocation. The R operators of Hengeveld i-ability and specificity, clearly have the same role as the 4 discourse op-erators of Rijkhoff.Although there are clear, if incomplete, parallels between the two

schemes in terms of operators, there is much less similarity between thetwo accounts as far as modifiers/satellites are concerned. Hengeveld f

modifiers would presumably include Rijkhoff 0 satellites.Rijkhoff (this volume), however, rejects Hengeveldinterpretations of in ( f, x and R), partly on the groundsthat no differences of scope appear to exist for the three meanings, andpartly because subjectivity is a property shared by many adjectives, andindeed also by several other modifying categories. For Rijkhoff, then, allthree readings would be analysed in terms of 1 (qualifying) satellites. Fur-

225

thermore, Hengeveld x category presumably covers lexical expressionsof number (Rijkhoff 1) and location ( 3). It also remains an open ques-tion as to where Rijkhoff 4), such as

or , would fit into HengeveldCentral to the discussion in the present article is the differentiation be-

tween, on the one hand, the expression of speaker attitude in the nounphrase, which Hengeveld deals with in terms of R modifiers and Rijkhoffin terms of the global qualification of a whole construction, and on theother hand the discourse properties of the noun phrase, which Hengeveldhandles as interpersonal R operators and Rijkhoff as 4 discourse opera-tors. It is to a more detailed discussion of this issue that we now turn.

Attitudinal meanings and meanings related to discourse properties are simi-lar in some respects but different in others. They are similar in that both areconcerned with how the speaker or writer chooses to present the content ofwhat s/he wants to say: in that sense, both are broadly interpersonal in na-ture. They differ, however, in that attitudinal markers are speaker-oriented:they represent the speakerparticular point of view, while discourse properties such as definiteness arehearer-oriented, in that they represent the speakerhearer e of the entities referred to in the discourse.These two types of meaning are sharply distinguished in Systemic

Functional Linguistics (see Halliday and Matthiessen 1999, 2004), whereattitudinal meaning falls under the interpersonal component, while dis-course properties such as definiteness/indefiniteness and specificity formpart of a textual component of meaning which has not been separately rec-ognized in any version of FG.4 A number of types of evidence are cited bysystemic linguists in favour of the recognition of these two components inaddition to ideational meaning (consisting of an experiential subcomponentwhich is parallel to the representational level of FG, plus a logical sub-component dealing with relations such as coordination, subordination,etc.). Although, as discussed in detail in Butler (2003a: 241-246), this evi-dence is in many ways problematic, it is at least possible to see a cleardifference between the kinds of contextual parameter towards which thevarious types of meaning are oriented. The attitudinal content of a message(which I see as deriving from the affective/interactional part of acomponent onent

226

is related to the context of interaction in personal and social terms, in thesense that the appropriate expression of attitudes depends on, for example,those social relationships between speaker and addressee which are re-flected in the degree of formality of the interaction; on the other hand, dis-course meanings such as (in)definiteness are related to the context pro-vided by the unfolding discourse itself and its relationship with the worlddescribed. I therefore propose to distinguish, within FDG, between twotypes of broadly interpersonal phenomenon: those which are oriented to-wards the social and personal context and those oriented towards the dis-course context.The distinction made here thus relies on a parallel distinction between

two kinds of context. The ial and personal context oposed in thepresent article forms part of what Connolly (2004a, 2007; see also Con-nolly 2004b: 103-104) refers to as ediate contextpart of scourse context-text scourse context

similar vein, Rijkhoff (this volume) proposes that the contextual compo-nent of FDG should be split into two separate components, one relating tosocial context and the other to the co-text. The former is seen as a situ-ational/external E-context, while the latter is part of the D-context, relatingto the discourse. Thus there is now a considerable measure of agreement onthe need to recognize the two types of context which underlie the distinc-tions I am proposing here. Note, however, that as we saw instill appears to hold a very restrictive view of the contextual component,according to which it comprises the discourse context andaspects of the immediate situation, such as the participants and the entitiesavailable to interactants, a view which is perpetuated by Keizer (this vol-ume). If we are to account for the kinds of phenomenon studied in the pre-sent article, we need to go beyond this, to include relevant aspects of thesocial situation as specified by Connolly and Rijkhoff.The split between two kinds of broadly interpersonal meaning proposed

above clearly has implications which go beyond the noun phrase. Rijkhoff(1992, 2002, this volume, forthcoming) has shown that there are clear par-allels between the layered structures he proposes for the noun phrase andthose for the clause. Particularly interesting in terms of the topic of thepresent article is his suggestion that the clause should be provided withdiscourse operators and satellites, concerned with the actuality or non-actuality of the event designated by the clause. Rijkhoff points out thatthere is a parallel between these discourse operators and satellites andthose for the noun phrase, in that both are concerned with whether the re-

227

spective referents do or do not have an identifiable place within the dis-course world. Reinterpreting this in terms of our two types of interpersonalmeaning, operators and satellites concerned with orienting the addresseetowards the actuality or non-actuality of an event belong to that part of theinterpersonal meaning which is oriented towards the discourse context,while those concerned with speech acts and certain types of modality be-long to interpersonal meaning oriented towards the social and personalcontext.

A number of types of phenomenon can be considered to fall under the so-cially and personally oriented type of interpersonal meaning within thenoun phrase. Firstly, there is the kind of item which served as an exampleof an R-modifier in Hengeveld xamples from Englishand Spanish are given in (5) and (6) below:5

(5) (BNC CK0 3608)

(6)Usually say-3PL that in the house the poor

man not speakt home the poor man doesn (CREA

sample XXI, text 2M/2H-197-d [Venezuela])

In English, more than one interpersonal modifier is often used to empha-size the meaning, as in (7), in which neither nor carries its refer-ential meaning, since the referent need be neither impecunious nor aged forthe utterance to be felicitous:

(7) (BNC G12 2473)

Swear words are also good examples of interpersonal modifiers, which canbe used directly as a modifier of the head noun, as in (8) and (9), or at-tached to another modifier, as in (10):

228

(8) . (BNC KD5 5835)

(9) . (BNC HH9 1051)

(10) . (BNC GUN 2153)

Examples of R-operators are the diminutive and augmentative in Spanish,in some though not all of their uses. Compare (11), in which the diminutiveencodes a representational meaning, that of small size ( == rsonal meaning, in this casethe affectionate attitude of the speaker towards the grandfather:

(11)I I in house my have-1SG a kitten

(CREA Centro de ense

(12)And my grandfather-DIM was English

(CREA CSHC-87 text 115, [Venezuela])

In (13), which forms part of a conversation on football, this operator iscombined with the choice of a modifier which itself carries interpersonalmeaning. Note also the augmentative ending on the word foragain carrying interpersonal meaning, but this time of a negative nature,contrasting with the speaker

(13)poor-DIM ball all to-it give-3PL kick-AUG

(CREA, , 06/06/1986, TVE1 [Spain]

The interpersonal use of the diminutive can carry other positive attitudinalmeanings besides affection (as in 12) and sympathy (as in 13), as is evidentin the following example:

229

(14)left-3PL all clean-DIM the vegetables peeled-DIM and

the cooking-pots clean-DIM

cooking pots cleaned up -87 Entrevista 158)

So far, my illustration of interpersonal meaning which is realized in thenoun phrase has been by means of examples of morphosyntactic modifica-tion through operators, and also lexical modifiers which have been pre-sented as having an attitudinal function. There are, however, some compli-cations here which need to be attended to.Firstly, note that even modifiers which convey a primarily attitudinal

meaning often do so via their representational import. Consider, in thisrespect, example (15) below:

(15)(BNC CG1 444)

Although the modifier clearly carries on the positive attitudinal stanceindicated by , it is highly unlikely that one would refer to one ofthe North American Great Lakes as : rather, thesmallness of the lake is seen as one of its charming features.6

Secondly, speakers have available to them another important mecha-nism for representing their attitude, namely the choice between attitudi-nally loaded lexical items and more neutral ones, or between positively andnegatively loaded items, and this applies to head nouns as well as to modi-fiers. In (16), the use of the word carries strong negative over-tones, compared with the neutral :

(16)(BNC CRE 2234)

Consider also (17) and (18):

(17)

(BNC G07 126)

230

(18) (BNC AE0 1984)

In (17) the woman is presented negatively as being , this being rein-forced by other modifiers, , , and and by referenceto and . In (18), on the other hand, the girl is presented positivelyas , the favourable impression being strengthened by the further use of

. Similarly, in (19) we have an example where the property ofthere being many people living and working in a city is presented nega-tively by means of the modifier , while in (20) the same basicproperty is presented from a positive viewpoint by means of :

(19).

(BNC AB6 1384)

(20)(BNC AM0 432)

The examples discussed above demonstrate that interpersonal meaningscan be realized at various places in a single nominal phrase, through theselection of attitudinally loaded nouns as heads, and of modifiers with aninterpersonal function overlaid on their representational import, togetherwith grammatical devices such as diminutivization in languages such asSpanish. In spoken language a further important contribution is made bythe intonation, and sometimes also voice quality, chosen by the speaker.This tendency for interpersonal meanings to be strung out through nominalphrases has been commented on by Halliday and Matthiessen (2004: 238):indeed, it is claimed (Halliday and Matthiessen 2004: 61) that such o-sodic n-ing not only in noun phrases but also in the clause. This tendency has alsobeen noted by Rijkhoff (this volume).It is not clear from the current literature how choices between lexical

items with roughly the same representational content, but different inter-personal overlay, should be handled within Functional Discourse Gram-mar. Although the lexicon has become much more prominent in FG in re-cent years, largely through the work of proponents of the FunctionalLexematic Model (see e.g. Mart Mingorance 1998, Faber and MairalUs 1999), the factors involved in selecting from a range of possible lexi-cal items, in order to refer to some entity or quality, have not received theattention they deserve. This is no doubt due largely to the concern of main-

231

stream FG practitioners to concentrate their efforts within the bounds ofthe grammar itself. Even in FDG, with its conceptual and contextual com-ponents, it has been made clear that the linguist still withthe grammatical component of the model. On the other hand, data such asthose presented in the present article suggest that we need to go beyondthis, to look at the role of the interpersonal component in mapping concep-tual material on to linguistic categories, a process which is strongly condi-tioned by contextual factors. In this respect, Systemic Functional Linguis-tics, with its strong interest in the relationship between linguistic choiceand context, has advanced rather further than F(D)G.

5.1. Introduction to the study

In rsonalmeaning realized in the noun phrase is concerned with the use of adjectivalmodifiers with attitudinal function. This section reports the results of adetailed study of the modifier in the British National Corpus (originalversion), using the concordancing program WordSmith Tools.7 The con-cordance for (with more than 15,000 occurrences in the corpus) wassorted firstly on the position 1 to the right of the headword (hence-forth position 1R), and within this ordering a further sort was made onposition 2 to the right (2R), so facilitating the study of as a modifierof following nouns. By means of the collocation tool provided by the soft-ware, the most frequent collocates at positions 1 to 5 to right and left of theheadword were obtained. The concordance and collocation list were thenused to study the properties of as a modifier.Three main senses of were found in the corpus:

rable/expectedthe interpersonal use, to express sympathy.

In the first of these senses, collocates with the word , words forcategories of people (e.g. , , ) and for geographicalentities (e.g. , , ). In the second sense, it is associatedwith a wide range of items such as , , , ,

, , etc. It is, of course, the third sense which we are par-ticularly interested in here.

232

5.2. Types of nominal head modified by interpersonal poor

In its interpersonal usage, collocates at position 1R with several typesof nominal head, and I shall concentrate on these collocates here, though itshould be realized that some of these nouns will also appear at position 2R,with a further adjective between and the noun (see also the discussionof these other adjectives inFirstly, there is a very large group of examples where this modifier is

followed by a proper name, which may or may not be preceded by a title(e.g. , , + surname). We may also include here usesof names for relatives (e.g. , ) when used as forms of address.Secondly, is followed by the personal pronouns (12 times) and(11 times). The use of with proper nouns and pronouns is com-

mented on by Huddleston and Pullum (2002: 430, 520), and Biber(1999: 509) note its use with pronouns. Hengeveld (this volume) commentsthat given their interpersonal nature, proper names and personal pronounswould indeed be expected to take only interpersonal operators and modifi-ers, and that this seems to be the case ( ).Thirdly, there is a further large group of examples in which is fol-

lowed by a common noun. In Table 1 are shown all the common nouns at1R which occur at least twice with in its interpersonal usage.8 Foreach of these nouns, the total frequencies and the frequencies of clearlyinterpersonal usages are given, together with an indication of whether thecombination appears in each of the possible syntactic environments.

233

Interpersonal + common noun at position R1, in the BNC(original edition)

16 100 * * * *9 100 * * * * *16 100 * * *9 78 * * *6 100 * * *12 100 * * *6 100 *61 87 * * * * * *14 93 * *21 100 * * * *3 100 * * *45 100 * * *117 76 * * * * *10 90 * * * * *44 100 * * * * * *28 100 * * * * *29 100 * * * *52 100 * * * * *22 100 * * * * *8 100 * *20 90 * * *44 100 * *9 100 * * *4 100 *6 33 * *156 94 * * * * * *21 100 * * * * *3 100 *4 75 *4 100 *10 70 * * *2 100 *41 100 * * * *24 100 * * * *

234

22 100 * * * *10 100 * *2 100 *10 100 * * * *3 100 *311 57 * * * * * *2 100 *56 86 * *2 100 *237 9 * *31 10 *13 100 * * *8 100 *47 100 * * *63 100 * * * * * *142 93 * * * *18 100 *139 71 * * * * *18 100 * * *

Table 1 allows us to draw some quite clear and interesting conclusionsabout the use of as an interpersonal modifier with common nouns inthe corpus. Firstly, note that 37/53 or 70% of the items in Table 1 co-occurwith only in its interpersonal meaning, and 42/53 or 79% of the com-binations have in its interpersonal sense in 90% or more of the cases.Examination of the nouns themselves reveals that they may be classifiedinto five groups:

(i) Generic nouns for male and/or female human beings, , ,, , , , , , which are neutral as to register and

also normally used literally, though they can occur with interpersonal over-lay, as in example (21) below, where is clearly being applied to a man,quite probably advanced in years, and in example (22), where

is addressed to a grown woman:

(21)

(BNC HTG 0288)

235

(22)

(BNC CCM 2662 2669)

(ii) A group of words which also refer generically to a male or femalehuman being, but which are informal in register: e.g. , , , ,

, , . The data in the first column of Table 1 revealan interesting difference between this group and the more neutral group:while many of the neutral items occur quite frequently with in itsliteral sense of used with the informalitems is almost always the interpersonal one. The choice of an informalrather than a neutral lexeme serves to reinforce the sympathetic attitudeencoded in . This is even clearer with the nouns , , ,

, , , , , , , , , , ,, , , all of which have a clear interpersonal overlay.

Again, most of these nouns take only in the interpersonal use, theexceptions being , , and : significantly, when

is used with poor in its literal sense, it does not refer to a person buthas its more neutral meaning of(iii) A set of words for relatives (

), some of which, as we have already seen, can also beused as proper nouns.(iv) Words which refer not to human beings as wholes, but to their

bodies ( ) and body parts, especially at the extremities (), though not necessarily so ( ). In these cases, the syntac-

tic environment in the corpus is always that of a noun phrase with a defi-nite, possessive determiner.(v) Items ( ) referring to domestic animals which are often

treated as quasi-human (e.g. by the use of the pronouns ).

5.3. Syntactic environments

Now let us turn to the syntactic environments in which interpersonal+ noun appears, as detailed in Table 1. We may recognize two broadenvironments (excluding theinstances which did not fit into any of the other groups), both of which canbe divided into sub-types:

236

(i) Occurrence as an independent unit, which may be appositional orleft- or right-dislocated:

(a) simply + noun:

(23) , (BNC G3B 1038)

(24) unhappy, , and perhaps(BNC ARJ 0569)

(b) + noun, preceded by a definite determiner (article, possessive,demonstrative):

(25) , I(BNC CBF 12768)

(26)(BNC AC5 0976)

(27) . I(BNC K8V 3031-3032)

(c) + noun, preceded by a pronoun ( )

(28) . .(BNC CM7 1287-1288)

(ii) In a noun phrase which is fully integrated into the clause:

(a) indefinite, in a singular noun phrase with , , , , ,or a plural noun phrase with zero determiner, numeral,

, , etc.

(29)(BNC G1M 1560)

(30)

(BNC HGS 0535)

237

(31)(BNC CB5 2955)

(32)(BNC JSY 062)

(b) with definite determiner, in a singular or plural noun phrase

(33) gave hope (BNC JY6 4179)

(34)

(BNC G02 0899)

As expected from the anaphoric function of definite noun phrases, thesestructures are frequently used in place of pronominal reference, where asympathetic attitude is being conveyed.

(c) bare, with ellipsis of article in a singular noun phrase acting as Sub-ject

(35) (BNC KBD 1631)

(36) (BNC KCY 1899)

(d) preceded by / :

(37)

(BNC J1E 0700)

(38)(BNC CD8 1141)

Of these various syntactic patterns, the most widespread are that with adefinite determiner in a fully integrated clause NP (52/53 or 98% of itemsinclude this pattern) and the independent form with no determiner (40/53or 75%).In addition to combinations in which is immediately followed by a

noun, there are those in which a further adjective separates from its

238

head. By far the most frequent of these additional adjectives are(546 occurrences), (281) and (34), all of which, significantly,refer to a property which confers vulnerability, through either age (low orhigh) or small size. The conditions for the use of these adjectives are morecomplex than those for itself: while, as noted in referent char-acterized as does have to be adult (we do not find , forexample), advanced age is not a prerequisite: (BNC KBL0877) does not imply that Arthur is literally old. On the other hand,and are not used for referents other than those which relate to smallor young animate entities: for instance the speaker who says

(BNC GV6 2448) is unlikely to be referring to a seven foot giant.The types of noun which occur in these combinations are basically thosewhich occur with by itself.

5.4. Interpersonal meaning within a conceptual and contextual framework:Integrating the properties of interpersonal poor into FDG

We have seen that , used as an interpersonal modifier, has a complexset of properties, involving strong collocational tendencies and preferredsyntactic environments. What, then, are the implications of this rich pat-terning for Functional Discourse Grammar? The answer to this questiondepends on what we see as included in the FDG model, and this is some-thing about which there is considerable ambiguity. For instance, althoughHengeveld (2004b) and Hengeveld and Mackenzie (2006) present the con-ceptual and contextual levels as the FDG model, in the latter ac-count (2006: 669) we also find the statement thatcomponent of a wider theory of verbal interactionextragrammatical components are themselves part of FDG, but adjunctsto it, within the kind of wider theory of verbal interaction proposed for FGby Dik (1997). Thus there is still considerable reluctance to engage fullywith the description of either the concepts which the grammar encodes orthe context, social and personal as well as discourse-oriented, in whichlanguage is embedded and which shapes the formation of appropriate ut-terances, as well as being progressively modified by those utterances. If wetake the more conservative stance, then all we need to do is whatHengeveld has already done, i.e. show that can act as an interpersonalmodifier, and set up appropriate structures within the grammar to deal with

239

this. But if we are committed to a truly functional explanation of why weuse particular forms of utterances (and this, after all, is surely the ultimateaim of functional linguistics), then we must boldly go where few FG lin-guists have been before, and tackle the thorny issues of concepts and con-text, as well as the area of lexical choice, all of which have so far been theCinderellas of FDG. What follows is no more than an initial step in thisdirection, using interpersonal as an example.In attempting to account, within an FDG framework, for the complex

set of properties displayed by interpersonal in English, there are anumber of issues we need to address:(i) What are the relevant concepts which the speaker needs to activate

in referring to the entity which is viewed in sympathetic terms?(ii) Where do attitudes themselves belong in the global FDG model?(iii) What are the choices the speaker has available when s/he wishes to

express a sympathetic attitude towards a referent using ?(iv) Finally, how can the answers to the above three questions be inte-

grated into an FDG account?I shall examine each of these issues in turn.

The entity itself will have a corresponding concept in the conceptual ontol-ogy on which the speaker draws, and this concept will preferentially be inthat part of the ontology which is concerned with animate beings, eitherhuman or treated as quasi-human, such as entities with the conceptual fea-tures [ANIMAL], [DOMESTIC].9 Even more restrictedly, constructions withinterpersonal relate to particular subparts of the category

(i) concepts which represent specific human beings;(ii) concepts which represent generic categories of human being,

characterizable by means of conceptual features such as[MALE/FEMALE], [ADULT/NOT ADULT];

(iii) concepts which represent relationships between human beings,characterizable by means of conceptual features such as [PARENTOF], [OFFSPRING OF], [SPOUSE OF], as well as gender features;

(iv) concepts characterized as [BODY PART], and usually also as[PERIPHERAL].

240

Clearly, expressing some attitude towards a referent (or a situation: seelater) has some extralinguistic basis, just as the expressions we use to referto entities have a basis in the conceptual ontology. The question then arisesas to whether attitudes and their expression are also part of the conceptualcomponent. If they are not, then we need to decide just where in the overallmodel they are to be placed.A search for definitions of the term

search engine yielded two broad classes. Firstly, there is the everyday useof the term to mean a general idea. Secondly, we have the more technicaldefinitions of the term which are of interest in the present context, and inwhich concepts are seen as mental entities which classify objects, events,properties and relations according to their shared characteristics. Clearly,then, accepted usage links concepts to entities (in a broad sense of theterm) and their extensional properties, whose linguistic correlates enterinto propositional structures. Attitudes such as sympathy, on the otherhand, are not reflected in propositional structures, but rather in the situa-tion-bound, non-extensional and non-propositional structures of the inter-personal level of the grammar. It therefore seems unwise to treat them asconcepts. This view is confirmed by work on conceptual ontologies de-signed in association with computational applications, such as that embed-ded in the Mikrokosmos knowledge-based machine translation system (seee.g. Beale, Nirenburg and Mahesh 1995). In this system, the ontology itselfis purely representational in that it consists only of concepts related to ob-jects, events and properties in the world under description, while non-propositional aspects of text structure are dealt with by means of i-crotheories l-ogy.An alternative might be to see attitudes, illocutionary and perlocution-

ary intentions, and the like, as aspects of the mental context for the produc-tion of an utterance. This, however, seems to me to be incorrect. Context,as I see it, is a complex of factors which, in the production of utterances,condition both the choice of concepts and the selection of ways of repre-senting those concepts linguistically; it need not, however, be part of whatthe speaker wishes to express. For this reason I part company here with theaccount given by Connolly (2004b), who holds that we do not need a sepa-rate conceptual component, but that if such a component is indeed pro-posed, it should be seen as a proper subset of the contextual component.

241

Since attitudes, emotions and the like are not concepts, and yet are stillpart of what we express using language,10 I propose that instead of just theconceptual component of the current FDG model, we should recognize a

, composed of two sub-components: theproper and the ,11 the latter being the

home not only of the attitudes and emotions we express through languageand also other means such as gesture, but also of the speech act forces wewish to convey: for instance, the speaker may choose directive speech actforce and a particular, high level of politeness from the interactional con-tent component (the second of these being conditioned by specifications inthe contextual component), and this may then be realized linguistically inEnglish as, say, a modalized interrogative (e.g. ). Furthermore,just as concepts can relate to fferent kinds and sizes in theworld under representation, so can the interactional content which accom-panies them. For instance, just as a conceptual ontology can contain nodesfor events (or more generally, what we might call situations) as well asobjects (entities) and properties, so the interactional content may relate towhole situations as well as to the entities and properties which form part ofthese situations. For example, the modal concept of possibility may beattached to a situation, as in (39) below, or to (a property associated with)an entity, as in (40).

(39) . (BNC AAV 1126)

(40)

. (BNC AKY 0007)

Similarly, a sympathetic attitude can be attached either to a whole situation(example 41), to an entity (42), or indeed to both (43):

(41)(BNC CEP 10522)

(42) (BNC GW8 0150)

(43) (BNC KD2 0840)

The conceptual and affective/interactional content sub-components clearlylink up with the representational and interpersonal levels of the FDG

242

model, respectively. It also seems possible that they are related in interest-ing ways with aspects of human neurophysiology.12 Springer and Deutsch(1998: 181) report thatsome qualities of humor seem to depend on right-hemisphere abilitiesSchumann (1997: 216-221) and Wray (2002: 241 242) survey a range ofstudies which link damage in the right hemisphere of the brain to abilitiesconcerned with pragmatics and discourse structure, and with processing ofemotional/affective aspects of language. The literature survey by Joanette,Goulet and Hannequin (1990), cited by Schumann (1997: 216-217), showsthat the pragmatic effects of right hemisphere lesions can include organiza-tional difficulties (e.g. with the combining of utterances into coherent dis-course or the elaboration of stories), problems of interpretation (e.g. relat-ing discourse to its context, dealing with metaphor, or understandingindirect speech acts), and difficulties with both production and comprehen-sion of types of prosody which convey emotional reaction or speech actforce. Both Wray and Schumann also cite the work of Locke (see e.g.Locke 1995), who postulates that first language acquisition is controlled bytwo different neural systems, one which has responsibility for what Locketerms l cognition nder the control of the right hemisphere,the other concerned with grammatical analysis, and under left hemispherecontrol.Particularly interesting in the present context is the account of Solms

and Turnbull (2002), who distinguish between brain mechanisms for deal-ing with the external world of things and situations, and with the internalworld of emotions, claiming (p. 19) thatto mediate this divide of the brain which deal with in-formation from the outer world are rily -dependent func-tions -fenced, using directed path-ways in the brain. On the other hand, communication of informationconcerned with the internal world

that reflect changes in the of the organismemphasis in original). This distinction correlates with a difference in thetypes of neurotransmitters involved: the channel-dependent functions aremediated primarily by the so-called a-tory, such as glutamate and aspartate, or inhibitory, such as gamma-aminobutyric acid), while what Solms and Turnbull call the c-tions with the inner world, are also mediated through the ac-tion of other transmitters such as serotonin and dopamine (p. 36). In termsof language, what I have called conceptual content lines up with our per-ception and representation of the world, whereas affective or

243

content rather reflects ption of , not of theobject world s-tributed nature of emotional responses in the brain may be iconically re-lated to the linguistic realization of affective meanings in distributed

-contained propositions.

Given the choice of a particular combination of conceptual and interac-tional content, a number of decisions need to be made about how thespeaker gure 1 are shown thechoices which are revealed by the corpus-based study described earlier,expressed as the kind of system network used to model paradigmatic op-tions in Systemic Functional Grammar. The notations used are shown inFigure 2, and the realizations of choices are given in Table 2. The symbol- t-work and its realization rules account for the ways in which sympatheticattitude can be expressed lexically in the whole noun phrase, through thecombination of choices in heads and modifiers. In this way, we can main-tain Hengeveld cular items,while at the same time accommodating the observation of Halliday andRijkhoff that such meaning is often distributedlinguistic unit. Clearly, we could also add to the network choices which arerealized phonologically, in intonation contours and stress.The systematization of speaker choice in this way raises the question of

how such choices might be represented within a Functional DiscourseGrammar. One possibility would be the incorporation into the model of anexplicitly paradigmatic schema within the conceptual and affec-tive/interactional components, on the lines of the networks used in Sys-temic Functional Grammar and co-opted for use here. Space does not per-mit the further discussion of this possibility here: for further detail of whata paradigmatic component might offer to FG and Role and ReferenceGrammar, see Butler (2003a: 235 239, 439 448; 2003b: 493 494).

independent

bare

neutral

definite

possessive

demonstrative

assituation

1stperson

indefinite

personal

_integrated

2ndperson

attachedtoreferent

_

addedexpressionofsympathyinmodifier

__ informalitem

forhead

_

Choicesintheexpressionofsympathy

244

245

x x

either x or y y x or y or z

y z

x

y

both (x or y) and (a or b)

a

b

x a

b if x or y, then a or b or c

y c

Notational conventions for system networks

Realizations of choices in the sympathy network

as situation X, X,X

attached to referent noun phrase with as modifierbareindependent isolated noun phraseintegrated noun phrase as constituent of clauseindefinite , , , , , + (singular)

zero det., numeral, , , etc. (plural)2nd person before1st person / beforeneutral beforepossessive possessive determiner ( , , , etc.) beforedemonstrative demonstrative det. ( , ) beforeadded expression ofsympathy in modifier

additional modifier ( , ) after

informal item for head choice of head lexeme with feature [infml.] (e.g. , )

246

The structure in (44) for , where the modifier is used inter-personally, is modelled on that given by Hengeveld (this volume):

(44) (Def RI: I): (fj: poorA (fj)) (RI))(xi: (fi: doctorN (fi)) (xi))

We may now see this as the first stage in the linguistic representation of aptual) entry which maps on to the

English word is accompanied by an (affective/interactional) entrycorresponding to sympathetic attitude. This entails a lexicon in which lex-emes are indexed for the conceptual and affective/interactional featureswhich they realize. But this is clearly not the whole story. The choice ofappropriate lexical items to encode the relevant content depends on selec-tions from the contextual component, especially in terms of degree of for-mality. Adopting a highly simplified three-point scale of formality (infor-mal, neutral, formal), we may say that the use of interpersonal modifierssuch as is characteristic of interactions with informal or neutral for-mality rather than those at the top end of the scale. This claim is stronglysupported by the information in Table 3, which lists, for various combina-tions involving , the sections of the BNC (World Edition) which haveat least three occurrences, together with the frequency per million words ineach section13.

247

Distribution of combinations with in the BNC (World Edition)

written, fiction, prose 58 3.6spoken, conversation 15 3.6written, biography 3 0.9written, popular lore 5 0.7

poor thing

written, miscellaneous 4 0.4written, fiction, prose 12 0.8written, non-academic,humanities, arts

3 0.8

spoken, conversation 3 0.7

poor things

written, miscellaneous 4 0.4written, fiction, prose 19 1.2poor sodspoken, conversation 4 1.0spoken, conversation 3 0.7poor sodswritten, fiction, prose 6 0.4

poor bugger written, fiction, prose 11 0.7poor devil written, fiction, prose 18 1.1

spoken, conversation 3 0.7poor devilswritten, fiction, prose 10 0.6written, fiction, drama 6 131.1spoken, interview,oral history

7 8.6

written, biography 33 6.2written, fiction, prose 97 6.1written, religion 6 5.3written, non-academic,humanities, arts

9 2.4

spoken, conversation 6 1.4written, popular lore 4 0.5written, miscellaneous 4 0.4

my poor

written, non-academic,social science

3 0.7

written, fiction, prose 44 2.8you poorspoken, conversation 8 1.9written, fiction, prose 174 10.9spoken, conversation 33 7.8written, biography 13 3.7written, popular lore 7 0.9

poor little

written, miscellaneous 8 0.9

248

spoken, live sportscommentary

5 150.1

spoken, consultation 6 43.5spoken, lecture,social science

4 25.0

spoken, conversation 78 18.5written, fiction, prose 261 16.4spoken, interview,oral history

12 14.7

written, newspaper,tabloid

6 8.2

written, newspaper,other, social

9 7.9

spoken, meeting 6 4.4spoken, broadcastdiscussion

3 4.0

written, popular lore 29 3.9written, religion 4 3.6written, biography 12 3.4written, non-academic,social science

14 3.4

written, newspaper,other, sports

3 2.9

written, newspaper,broadsheet, national,miscellaneous

3 2.9

written, miscellaneous 18 2.0written, non-academic,political, law, education

6 1.3

written, non-academic,natural science

3 1.2

poor old

written, non-academic,humanities, arts

3 0.8

written, academic,social science

6 1.4poor young

written, fiction, prose 14 0.9

It can be seen that most of the combinations listed are most prevalent inconversation and in those registers of written English which might be ex-pected to incorporate quite large amounts of dialogue (dramatic and prosefiction, biography): indeed, it is interesting that more use seems to be madeof these expressions of sympathy in the simulated conversation of fiction

249

than in real conversation. This corroborates the findings of Biber(1999: 509), who report that in fiction, premodifying is commonlyemotive. Furthermore, in informal contexts only, there is the further choiceof whether to use an item of neutral formality such as or , or aninformal term such as or . This second choice, as wehave seen, reinforces the sympathetic attitude, even though items such as

or do not themselves have any inherent interpersonal meaning. Athird choice is to use as a head noun one of the items which does have in-terpersonal overlay, such as . Thus there aresome items which must be marked in the lexicon for their conceptual andaffective features and for their register specificity. Consider the example in(45):

(45)

(BNC CDE 1938 1939)

In the production of , the conceptual features [HUMAN,NOT ADULT, FEMALE] are activated indeed, they are already activated byprevious mentions of the same referent by means of other expressions (inthe extract given in (45), these are and ). If all these featureswere realized in the linguistic output, the corresponding lexemes would be

or one of the more informal equivalents such as . But of course thespeaker is not obliged to realize all the conceptual features every time thereferent is mentioned. Instead, s/he may choose a superordinate term, suchas , , corresponding to just the features [HUMAN, NOT ADULT]. Thechoice of rather than , or rather than , is motivated not byconceptual factors but by a complex of other considerations: sympatheticattitude, the contextual factor of informality, and also, in the case of ,the socio-geographical characteristics of the speaker. Informality (or neu-trality) in the contextual component also licenses the use of the interper-sonal modifier , and also, together with the conceptual feature [NOTADULT], the use of the further interpersonal modifier .Also at the interpersonal level, we must represent the distinction be-

tween the uses which I have labelled as ependentthe noun phrase containing is integrated as a constituent of the clause.The first type can be treated as a separate discourse act, while in the secondcase we have a referential phrase which, in the most recent manifestationsof FDG, would be seen as constituting just one of a number of subactswithin an act.

250

By way of summary, let us look at the complete derivation of a particu-lar noun phrase containing , in its literal and non-literal uses.

(46)

(BNC FUB 0776 0777)

(47)(BNC HNY 2693)

The contexts of these examples in the corpus show that in the first, bothand are being used literally, to refer to an impecunious mendi-

cant, while in the second we have the interpersonal use expressing sympa-thy for the person referred to.The noun phrase in (46) reflects the choice of a con-

ceptual configuration which specifies that we are dealing with a male per-son who lives by asking people for money or food, and that a property ofthis person is that he has little money. This configuration is encoded in theinterpersonal and representational levels of the grammatical component asin (48), modelled on a slightly simplified version of the parallel structuregiven in Hengeveld (this volume):

(48) (Def RI)(1 xi: (fi beggarN (fi)) (xi): (fj: poorA (fj)) (xi))Rec

We have here a structure in which the referent is specified as a beggar andthen as poor. Both and will need to be specified in the lexiconwith reference to the conceptual configurations they realize (seeThe structure in (48) is passed to the morphosyntactic level, where it isreinterpreted as in (49):

(49) [[theArt] [[poorA]AP beggarN- ]NP1]NP2

Finally, at the phonological level we have the configuration in (50):

(50) / /

Turning now to the same surface noun phrase in (47), we need a very dif-ferent configuration in the content component, in both conceptual and af-fective/interactional subcomponents. The only strictly necessary concep-

251

tual feature for the use of in its interpersonally-orientedsense is [+HUMAN], but in this case (and indeed in the very few other casesof this usage in the corpus examined, though the sample is too small toallow firm conclusions) the referent is also characterized as [+MALE]. Thisuse of is specialized for co-occurrence with an evaluative, interper-sonally-coloured adjective, which in this case brings with it from the affec-tive/interactional subcomponent the feature of sympathetic attitude on thepart of the speaker, although with adjectives other than we may finddifferent attitudes (e.g. envy in , mild criticism in). We also have the specification of informality in the contextual com-

ponent. All these features lead to the encoding of the content in the inter-personal and representational levels shown in (51):

(51) (Def RI: (fi: beggarN (fi)) I): (fj: poorA (fj)) (RI))(1 xi)Ag

The structures at the morphosyntactic and phonological levels are the sameas before.

The specification of lexical entries is not yet highly developed within FDG,so I shall make only a few brief and informal suggestions for how we mighthandle the kind of variation documented here. As shown in the previoussection, a lexical entry may need to contain information not only about theconceptual structure represented by a lexeme and any affective / interac-tional components of the meaning, but also about contextual variables suchas register and dialect. In Table 4 I present an informal specification forvarious lexemes discussed in 14. Items in small capitals are assumedto be taken from the conceptual ontology. It must be emphasized that theentries sketched in Table 4 are not intended as anything more than prelimi-nary informal notes: in particular, I do not wish to claim any inherent supe-riority for the notation I have used, and it is obvious that most of the con-ceptual labels used are themselves complex and in need of furtherexplication. I hope, however, to have given some idea of the kinds of in-formation which need to be present if we are to deal adequately with thecomplexities of interpersonal as well as representational meaning, and withsocial contextual variation in the lexicon.

Lexicalentries

child

Na

HUMAN(x)&NOT(ADULT

(x))

(RI)

(xi:(fi:child N(fi))

(xi))

kid 1

NHUMAN(x)&NOT(ADULT

(x))

Informal

(RI)

(xi:(fi:kid N(fi))

(xi))

lass

NHUMAN(x)&FEMALE(x)&

YOUNG(x)

Informal

Northern

UK,inc.

Scotland

(RI)

(xi:(fi:lassN(fi))

(xi))

beggar1

NHUMAN(x)&HUMAN(y)&

((LIVE(x)(REQUEST(x,

(GIVE(y,M

ONEY

FOOD,

x)))) Means)

(RI)

(xi:(fi:beggarN(fi))

(xi))

beggar2

NHUMAN(x)

SYMP(S(x))

Informal

(RI:(fi:beggarN

(fi))(R

I):(f j:

(LexInterpers-adj)(fj))

(RI))

(xi)

252

poor1

AHAVE(x,LITTLE(MONEY&

POSSESSIONS))

(RI)

(xi:(fi:LexN(fi))

(xi):(fj:poorA(fj))

(xi))

poor2

ADO(x,Activity x)

( BAD) Manner

(RI)

(xi:(fi:LexN(fi):

(fj:poorA(fj))(fi))

(xi))

poor3

A___

SYMP(S(x))

Informal/

neutral

(RI:

I):

(fj:poorA(fj))(R

I))

(xi:(fi:LexN(fi))

(xi))

253

254

The entry for specifies simply that the referent must be characterizedas HUMAN and as NOT ADULT. The entry for (so labelled because thereis at least one other meaning for ) adds to that of the fact that theregister must be informal for the appropriate use of the lexeme. For ,we specify not only that the referent must be HUMAN, FEMALE and YOUNG(though not necessarily NOT ADULT), but also that that the register is in-formal and that the lexeme is used in Northern and Scottish dialects ofBritish English. The entry for is that for the literal use, and speci-fies that a beggar is HUMAN and LIVEs by REQUESTing some other HUMANentity to GIVE him or her MONEY or FOOD. In all these cases, thepattern e-ferring phrase. The entry for , on the other hand, simply indicatesthat the referent must be HUMAN, and the structural pattern shows that thisusage of co-occurs with an evaluative adjective such as and/or

and involves an instance of speaker-bound, interpersonal modificationexpressing the speaker The entry for

indicates the literal usage, in which the referent is said to HAVE LIT-TLE MONEY and POSSESSIONS, and the structural pattern is for referentmodification. That for is for the meaning in which what is beingdescribed is the bad manner in which the referent carries out the activityimplicit in the designation (e.g. in the sense of one who doesthe job of doctoring badly), and so its structural pattern indicates referencemodification; furthermore, the activity mentioned in the conceptual struc-ture is indexed to show that it is the activity in which the entity character-ized as is assumed to engage by virtue of being described by a par-ticular head noun. The entry for , on the other hand, is the subjective,interpersonal use, specifying only the sympathetic attitude of the speakertowards the referent and the fact that the register must be informal or neu-tral, together with a structural pattern showing modification at the interper-sonal level.During the process of language production, the speaker needs to draw

on lexical entries such as these, together with information from the concep-tual, affective/interactional and social contextual and discourse contextualcomponents of the revised FDG model. For instance, consider the follow-ing input from the various components:

255

[HUMAN, MALE, ADULT]Sympathetic attitude of speaker towardsreferentInformal registerReferent already activated

The combination of conceptual and social contextual features allows15 theselection from the lexicon of an item with matching features, such asor ,16 while the social contextual information leads to the choice of anentry for in its interpersonal function, and that from the discoursecontext results in the choice of a definite noun phrase, signalled by thedefinite article . The resulting noun phrase is thus .Similarly, in language comprehension, if the speaker produces a phrase

such as , this gives the hearer access to the information that(i) the speaker is referring to a male, adult human being, (ii) s/he is ex-pressing a sympathetic attitude towards this referent,17 (iii) s/he regards thecontext of interaction as informal, (iv) s/he regards the referent as alreadyactivated in the discourse and so recoverable by the hearer.

In this article, it was first demonstrated that the parallels betweenHengeveld s structure for the noun phrase within FDG and the structuresproposed by Rijkhoff are not as simple or as complete as they might firstappear to be. Detailed consideration of these proposals led to the postula-tion of two types of broadly interpersonal meaning, one oriented towardsthe social and personal context of interaction, the other towards the dis-course context. A more detailed examination of the first of these types ofinterpersonal meaning, as expressed within the noun phrase, highlightedthe need to study the factors which lead to the choice of one lexical itemrather than another, an endeavour which inevitably takes us beyond thegrammatical component of FDG, requiring an examination of the interac-tion between the grammar and the conceptual and contextual components.The way in which this might work was demonstrated by a detailed cor-

pus-based study of the use of as an interpersonal modifier in the Eng-lish noun phrase. This study led to the postulation that the conceptual com-ponent of current FDG should be replaced by a content component, dividedinto two parts, conceptual and affective/interactional, a proposal whichaccords with what is known about the partitioning of brain functions and

256

their mediation. The choices available to the speaker in the expression ofsympathy using were then systematized using a network notation. Thestages in the conversion of a complex of conceptual and affec-tive/interactional content into an FDG structure were then discussed, andfinally some very preliminary and sketchy suggestions were made for thecontent of lexical entries relating to some of the lexical items investigatedin this work.

1. The work reported here was carried out within the research project DiscourseAnalysis in English: Aspects of cognition, typology and L2 acquisition,awarded to the SCIMITAR research group (http://ietsil.usc.es/scimitar) andsponsored by the Spanish Ministry of Education, FEDER funds and the Xuntade Galicia (XUGA) (grant numbers BFF2002-02441, PGIDIT03PXIC20403PN). I am grateful to Lachlan Mackenzie and to three anonymous re-viewers for their valuable comments on earlier drafts of this article, which ledto substantial improvements. The usual disclaimers apply.

2. In Mackenzie s (1998, 2000) Incremental Functional Grammar model, a vari-ant of FDG, the referential and ascriptivere-labelled asMackenzie (2006). I shall adopt the term

3. The distinction betweenis taken from Bolinger (1967).

4. A broad interpretation of interpersonal meaning within FG is also found in thework of Hannay (1991) on the ementsuggests govern the distribution of Topic and Focus pragmatic functions, andwhich he treats as subtypes of declarative illocutionary force. He thus treats asa kind of interpersonal meaning what, in SFG terms, would be treated as partof the textual metafunction. For discussion see Butler (2003b: 96-97).

5. English examples are from the original version of the British National Corpus(see http://info.ox.ac.uk/bnc for details). Spanish examples are from the Cor-pus de Referencia del Espahttp://www.rae.es.

6. I am grateful to Gordon Tucker for first pointing out this interaction to me.7. For details of WordSmith Tools see http://www.lexically.net/wordsmith.8. Usually, the surrounding context indicated clearly those cases in which the

meaning was representational (i.e.cases, especially with the noun , categorization proved more difficult.The frequencies given for interpersonal usages in Table 1 are for those exam-ples which were clearly non-representational.

257

9. I leave open here the important and somewhat vexed question of whetherconceptual features should be regarded as universal. Work on a Natural Se-mantic Metalanguage by Wierzbicka and her colleagues (see e.g. Wierzbicka1992, 1996; Goddard and Wierzbicka 1994, 2002) suggests that we may in-deed be able to express conceptual meanings in terms of a set of universalprimitives, while Levinson (1997) claims that we normally package sets ofatomic conceptual features into what he callsthat while some concepts are truly universal because of their importance in allhuman cultures, others are specific to particular sets of languages/cultures, andperhaps yet others to a particular language. What is clear, however, theapproach in Wierzbicka n-ceptual and semantic representations (for discussion see Levinson 1997, Pe-derson and Nuyts 1997, Nuyts 1992, 2001).

10. In other words, they are still part ofsense, as pointed out by Nuyts (2001: 6).

11. The substance of this proposal is adumbrated by Hengeveld (2004b: 369),who states that ptual component there is an important dis-tinction between communicative intentions on the one hand, and conceptuali-zations on the other r-able not to regard communicative intentions, attitudes, etc. as conceptual innature.

12. I am grateful to Lachlan Mackenzie for alerting me to this possibility andmaking me aware of the work of Solms and Turnbull. I am also indebted toAlison Wray for information on Schumann

13. The information in this table was derived from the Variation in English Wordsand Phrases (VIEW) online search facility for the BNC World Edition devel-oped by Mark Davies (http://view.byu.edu). Note that the frequencies givenare for all occurrences of the word pairs in question, irrespective of their syn-tactic contexts.

14. The word definitions are based on information in the Collins Cobuild Diction-ary.

15. It should be noted that the choice of a lexeme marked as informal is probablybest regarded as an option in informal contexts, rather than obligatory, since aphrase such as would not be out of place in such situations.What is blocked, however, except for humorous effect, is the choice of an itemmarked as being formal.

16. If the social and/or geographical dialect parameters in the contextual compo-nent were suitable, other more restricted choices such as would also bepossible.

17. The question arises here of how the hearer knows that the speaker is express-ing a sympathetic attitude rather than attributing an impecunious status to thereferent. My intuition is that if the attribute of poorness has not been men-tioned in the recent discourse context, and if is not stressed, then the in-

258

terpersonal reading is the most likely, especially if other elements of the dis-course context are conducive to such an interpretation. However, it wouldneed a further study to corroborate or refute these hunches.

Beale, Stephen, Sergei Nirenburg and Kavi Mahesh1995 Semantic analysis in the Mikrokosmos Machine Translation Project.

In:, August 2 4. Kaser Sart University, Bangkok,

Thailand.Biber, Douglas, S. Johansson, G. Leech, S. Conrad and E. Finegan1999 . Harlow: Pearson

Education Limited.Bolinger, Dwight1967 Adjectives in English: attribution and predication. 18: 134.

Butler, Christopher S.2003a

.Amsterdam/Philadelphia: John Benjamins.

2003b.

Amsterdam/Philadelphia: John Benjamins.Connolly, John H.2004a Context in Functional Discourse Grammar. Unpublished ms.2004b The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and Marmez-Gonz ,89 116. (Functional Grammar Series 24.) Berlin/New York: Moutonde Gruyter.

2007 Mental context and the expression of terms within the English clause:An approach based on Functional Discourse Grammar. In: MikeHannay and Gerard J. Steen (eds.)

, 193-208. Am-sterdam/Philadelphia: John Benjamins.

Dik, Simon C.1997

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

Faber, Pamela and Ricardo Mairal Us1999 (Functional Grammar

Series 23.) Berlin/New York: Mouton de Gruyter.

259

Goddard, Cliff and Anna Wierzbicka (eds.)1994

Amsterdam/Philadelphia: John Benjamins.2002

Amsterdam/Philadelphia: John Benjamins.Halliday, Michael A. K. and Christian M. I. M. Matthiessen1999

London/New York: Cassell.2004 , 3rd edn., revised by Chris-

tian M. I. M. Matthiessen. London: Arnold.Hannay, Mike1991 Pragmatic function assignment and word order variation in a func-

tional grammar of English. 16: 131 55.Hengeveld, Kees2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzie and Mar -Gonz, 1 21. (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzieand Mar

G -Gonz, 365 378. (Functional Grammar Series 24.) Berlin/New York:

Mouton de Gruyter.2005 Dynamic expression in Functional Discourse Grammar. In: Casper de

Groot and Kees Hengeveld (eds.)53 86. (Functional Grammar Series 27.) Ber-

lin/New York: Mouton de Gruyter.this vol. Prototypical and non-prototypical noun phrases in FDG.

Hengeveld, Kees and J. Lachlan Mackenzie2006 Functional Discourse Grammar. In: Keith Brown (ed.),

2nd Edition, Vol. 4, 668-676. Oxford:Elsevier.

fc. . Oxford: Oxford University Press.Huddleston, Rodney and Geoffrey Pullum2002 . Cambridge:

Cambridge University Press.Joanette, Yves, Pierre Goulet and Didier Hannequin1990 . New York: Verlag.

Keizer, Evelienthis vol. Reference and ascription in F(D)G: an inventory of problems and

some possible solutions.

260

Levinson, Stephen C.1997 From outer to inner space: linguistic categories and non-linguistic

thinking. In: Jan Nuyts and Eric Pederson (eds.),, 13 45. Cambridge: Cambridge University Press.

Locke, John L.1995 Development of the capacity for spoken language. In: Paul F.

Fletcher and Brain MacWhinney (eds.), 278 302. Oxford: Blackwell.

Mackenzie, J. Lachlan1998 The basis of syntax in the holophrase. In: Mike Hannay and A.

Machtelt Bolkestein (eds.),, 267-295. (Studies in Language Companion Series 44.)

Amsterdam/Philadelphia: John Benjamins.2000 First things first: towards an Incremental Functional Grammar.

32: 23-44.Mart , Mingorance, Leocadio1998

Edited by Amalia Mar ales. Granada:University of Granada.

Nuyts, Jan1992 Amster-

dam/Philadelphia: John Benjamins.2001

. Amsterdam/Philadelphia: John Benjamins.Pederson, Eric and Jan Nuyts1997 Overview: on the relationship between language and conceptualiza-

tion. In: Jan Nuyts and Eric Pederson (eds.), 1 12. Cambridge: Cambridge University Press.

Rijkhoff, Jan1992 The noun phrase: A typological study of its form and structure. Ph.D.

diss., University of Amsterdam.2002 Oxford: Oxford University Press.this vol. Layers, levels and contexts in FDG.fc. Layering and iconicity in the noun phrase: Descriptive and interper-

sonal modifiers. In: Christopher S. Butler and Miriam Taverniers(eds.), special issue of devoted to layering in functionalgrammars.

Schumann, John H.1997 The neurobiology of affect in language. Supplement 1 to

48.Solms, Mark and Oliver Turnbull2002

. New York: Other Press/Karnac Books.

261

Springer, Sally P. and Georg Deutsch1998 .

5th edn. New York: Freeman.Wierzbicka, Anna1992

. New York/Oxford: Oxford Univer-sity Press.

1996 Oxford: Oxford University Press.Wray, Alison2002 . Cambridge: Cambridge Uni-

versity Press.

In English and many other languages, the noun phrase (NP) is capable offulfilling a variety of functions. An NP may typically appear as any of thefollowing:

(1) a. A constituent of a clause, for example subject or object.b. A constituent of another phrase, for example complement of theadposition within an adpositional phrase.

c. An extra-clausal constituent (ECC), as defined and described inDik (1997b: 379-407).

The category of ECC is subdivided by Dik (1997b: 383) into four posi-tional subcategories:

(2) a. Freestanding or absolute (unattached to any clause).Sam!

b. Pre-clausal.Sam, you have been most helpful!

c. Clause-internal or parenthetical.You, Sam, have been most helpful!

d. Post-clausal.You have been most helpful, Sam!

Among these subcategories, the present paper is concerned specificallywith freestanding NPs, as found in (written English-language) documentssuch as books, articles, web-pages, and so forth.We shall take it for granted that any of these documents may contain

not only language-text but also non-linguistic material, particularly images,

264

such as pictures or diagrams. As Kress and van Leeuwen (1996: 39) re-mind us, discourse is in general multimodal.Attention will be focused here especially upon those freestanding NPs

which can be characterized as being:

(3) a. Relatively peripheral to the main body of discourse.b. Characterized by the sort of syntactic structure for which Quirk etal. (1985: 845-847) employ the term nguage .

The term et al. to the following:

(4) a. Headings.b. Titles.c. Labels.d. Notices.e. Newspaper headlines.f. Advertisements.

Quirk et al. (1985: 845) state that simple block-language expressions tendto take the form not of sentences but of NPs or nominal clauses. Othertypes of structure are also found. However, in the present paper, of course,we are concerned only with those of the NP category.Before going any further, let us make clear how the term

used in the present paper. It will be assumed that (in English syntax) an NPconsists of the following:

(5) a. A head-word, which is either a noun or a pronoun, and which isobligatory, except in instances of ellipsis

b. Optionally, one or more modifiers. Structurally, a modifier maycomprise (i) a single word, (ii) a group of two or more wordsforming an embedded phrase, or (iii) an embedded clause. Sub-ject to various constraints, modifiers may appear either before orafter the head-word. However, because modifiers are generallyoptional, it is possible for an NP to contain just its head-word.

We need also to consider the representation of freestanding NPs withinFDG. As an example, let us suppose that the NP (6a) occursas the heading of a chapter in a document. This NP would be representedalong the lines of (6b) at the interpersonal level and (6c) at the representa-tional level:2

265

(6) a.b. (+id R1: T1 (R1))c. (1p1: (f1: problemN (f1)) (p1))

No syntactic or semantic function has been assigned to this NP, as the termplays no role within any larger grammatical structure. (If one were to insistupon at least a semantic function, then this would presumably have to beon grounds of consistency with possible expansions such as

or in which theassigned to the NP in question.) Nor is any proposition-related representa-tion suggested, since (6a) does not embody a claim of either truth or falsity.Rather, the salient function of the chapter-heading (6) consists in its

discourse-pragmatic role of introducing (or ushering in), and at the sametime encapsulating, the ensuingus, therefore, turn to the subject of pragmatics, for it is with this domain,rather than with grammatical details, that the present paper is fundamen-tally concerned.3 In the next section we shall consider how freestandingNPs may be treated from the point of view of pragmatics and discourse,within the framework of Functional Discourse Grammar (FDG) outlined byHengeveld (2004a, 2004b). This will lead on to a discussion of the implica-tions for the broader theory of verbal interaction envisaged by Dik (1997a:4). Finally, some general conclusions will be drawn.

Let us, then, consider the pragmatic treatment of freestanding NPs. Weshall begin with functions that reflect the structure and subject-matter ofdiscourse, such as the titles of books or articles and the headings of chap-ters, sections and sub-sections within individual chapters.A freestanding constituent is, as noted above, a type of ECC. Dik

(1997b: 384) states that an ECC may have one of the following functions:

266

(7) a. Interaction managementKim! (attracting the addressee

to engagement in conversation).b. Attitude specification

Wow! (expressing amazement).c. Discourse organization

Introduction (as a chapter heading).d. Discourse execution

Yes (in response to a question during the conduct of aconversation).

Clearly, the kind of headings with which we are currently concerned be-long in principle to DikHowever, there is a slight problem here. Dik (1997b: 386-403) divides thecategory in question into the following subcategories:

(8) a. Boundary markingAnyway... (indicating the transition of one part of a

conversation to the next part).b. Orientation

As for you, ...c. Tail

[It

All of these subcategories apply to constituents that are attached to clauses,rather than to freestanding items like headings or titles, and are thereforenot suitable for our present purpose. Instead, we shall employ a subset ofthe functions identified by Quirk et al. (listed in (4) above), plus some oth-ers:

(9) a. Headingb. Titlingc. Prelimd. Finishe. Labelf. Caption

These terms will be explicated more fully as we proceed. (It is possible thatfuture research will indicate the need for additional functions, but for pre-sent purposes we shall confine ourselves to those in (9).)

267

2.1. Heading, Titling, Prelim and Finish

First of all, we shall consider headings and titlings. Ausher in a particular unit within the discourse hierarchy. Headings can belayered, with the topmost (layer-1) being the most general, while subordi-nate headings occupy layer-2, and so on downwards. For purposes of dis-course representation, a layer-1 heading may be denoted as in (10a), alayer-2 subheading as in (10b), and so on.

(10) a. heading[1]b. heading[2]

Asome self-contained item (such as an image) that appears within a docu-ment. The term mpass not only the title itself,but also any directly-associated material, particularly the name of the au-thor, if this is juxtaposed to the title.How may we accommodate headings and titlings within the framework

of FDG? Let us start with an example. Suppose that we have a book calledby J. Smith, and that we are going to treat this book

as constituting a single, coherent discourse. Suppose also that the firstchapter is headed , and that we are going to regard that chapteras one of the transactions within the discourse concerned (though not nec-essarily the first, as there may, for example, be a preface). Suppose, fur-ther, that the opening sentence of the first chapter reads:

In sum:

(11) a.Functional LinguisticsJ. Smith

b.Introduction

c.Functional Linguistics is the study of human language as a sys-tem of human communication.

In the present example, the titling of the book can, in fact, be considered tobe the opening move in the entire discourse. This move contains two asser-tive acts, communicating respectively the title and the author.

268

Note that if the title had beenthen this would, in itself, have consisted of two acts: an as-

sertion and a question.4 The second of these acts would conventionally becalled the -title t the latter term is not really necessary for our pur-poses.The titling in (11a) also doubles up as the topmost heading of the dis-

course hierarchy. However, a titling can generally be expected to do morethan just usher in the text of the document. It can also be expected to pro-vide what we might (figuratively) call the e-sents to the outside world. That is to say, the titling generally appears onthe cover of the document, and furthermore it serves to identify the docu-ment distinctively (and perhaps uniquely) and thus to provide a handle forstorage and retrieval, as for instance in a situation where books or reportsare catalogued and requested by their authors and titles. Consequently, weshall treat the function of distinct from that ofMoreover, we shall encounter examples below in which titlings do notsimultaneously serve as headings.Accordingly, employing (in essence) the style of discourse representa-

tion developed in Connolly (2004) and omitting dispensable detail, we mayrepresent the material in (11) in the following manner:5

(12) ENTER DISCOURSE D1

ENTER MOVE M1

M1 = titling = heading[1]

ENTER ACT A1 /* */6

A1: [ASSERT (Author) (Message m1) (Audience)] (A1)ILLS(A1) = ILLA(A1) = representativeLEAVE ACT A1ENTER ACT A2 /* */A2: [ASSERT (Author) (Message m2) (Audience)] (A2)ILLS(A2) = ILLA(A2) = representativeLEAVE ACT A2LEAVE MOVE M1

ENTER TRANSACTION T2

ENTER MOVE M11

269

M11 = heading[2]

ENTER ACT A51 /* */A51: [ASSERT (Author) (Message m51) (Audience)] (A51)ILLS(A51) = ILLA(A51) = representativeLEAVE ACT A51LEAVE MOVE M11

ENTER MOVE M12 /* */

ENTER ACT A52 /* */A52: [ASSERT (Author) (Message m52) (Audience)] (A52)ILLS(A52) = ILLA(A52) = representativeLEAVE ACT A52LEAVE MOVE M12

LEAVE TRANSACTION T2

LEAVE DISCOURSE D1

The fact that this representation makes explicit the assignment of thepragmatic functions r-tant, as it makes clear that although these NPs may lack a syntactic or se-mantic-relational function, nevertheless they do have a specific discourse-related function.The term aterial which

precedes the body text, but does not act as a heading or titling, while aerial that occurs at the end of a body

text. Examples of prelim material can be seen when the senderand the date are included at the top of a letter, or when the date and placeof publication are displayed at the beginning of a book. Such informationmay help to contextualize the discourse. However, if information referringto the date and place of the origin of a document is placed at the end of abody text, such as the preface of a book, then it constitutes a finish.Note that the salutation ( ) at the beginning of a letter and the

sign-off at the end ( ) belong under the heading of interaction man-

270

agement rather than discourse organization; see (7) above. Dik (1997b:384-385) uses the terms -takingrespective functions.

2.2. Other matters of block language

At this juncture it is convenient to consider briefly those items listed in (4)above that we do propose to adopt here as functions of freestandingNPs, namely notices, headlines and advertisements. An example of ablock-language notice would be a road-sign containing just the freestand-ing NP:

(13) Low Bridge

This has very little structure, and from the discourse point of view there isno titling or heading, but simply the message itself. The latter comprises asingle move containing only one assertive act. The latter serves as a warn-ing to the drivers of tall vehicles, and thus has the illocutionary force of adirective,7 since it seeks to influence the overt behaviour of drivers whoread it. An appropriate FDG-style representation of this notice wouldtherefore be as follows:

(14) ENTER DISCOURSE D1

ENTER MOVE M1

ENTER ACT A1A1: [ASSERT (Author) (Message m1) (Audience)] (A1)ILLS(A1) = ILLA(A1) = directiveLEAVE ACT A1LEAVE MOVE M1

LEAVE DISCOURSE D1

Simple as this may be, it appears to be all that is required, and there seemsto be no reason to include a discourse-pragmatic function such asOf course, it is possible for notices to be more elaborate than this. For

instance, we could imagine a notice with a heading (such as) and a body (such as and so forth). In

sum:

271

(15) a.Rules of the House

b.No smoking!No denims!

Such a notice would be represented along the following lines:

(16) ENTER DISCOURSE D1

ENTER MOVE M1

M1 = heading[1]

ENTER ACT A1 /* */A1: [ASSERT (Author) (Message m1) (Audience)] (A1)ILLS(A1) = ILLA(A1) = representativeLEAVE ACT A1LEAVE MOVE M1

ENTER MOVE M2 /* */

ENTER ACT A2 /* */A2: [ASSERT (Author) (Message m2) (Audience)] (A2)ILLS(A2) = ILLA(A2) = directiveLEAVE ACT A2ENTER ACT A3 /* */A3: [ASSERT (Author) (Message m3) (Audience)] (A3)ILLS(A3) = ILLA(A3) = directiveLEAVE ACT A3

LEAVE MOVE M2

LEAVE DISCOURSE D1

As for headlines in newspapers, these can be considered to constitute atype of heading. However, accommodating newspaper headlines and arti-cles within discourse analysis is not necessarily straightforward. Let ustake an example. Suppose that there exists a newspaper entitled

, aimed at UK-nationals living overseas who wish to be kept up-to-date with current events in the UK. This newspaper is likely to consist of

272

articles on a set of topics which bear little relationship to one another, otherthan the fact that they happen to be of current interest. Consequently, wemay be reluctant to regard the entire document as a single discourse, butmay prefer, instead, to treat each article (together with its associated head-line) as a separate discourse-strand.8

To continue with our example, suppose that an issue of this newspapercontains an article with the headline andanother with the headline . The first article openswith the sentence: Thesecond begins: Insum:

(17) a.Home Territory

b.Improvement in Train Punctuality

c.Rail travellers received some good news yesterday.

d.Increase in Wine Imports

e.The British appetite for wine shows no sign of abating.

Note that because the titling NP stands outside of all the various discourse-strands, it does not function as a discourse heading. Thus, the material in(17) may be represented along the following lines:

(18) ENTER DOCUMENT

ENTER MOVE M1

M1 = titling

ENTER ACT A1 /* */A1: [ASSERT (Author) (Message m1) (Audience)] (A1)ILLS(A1) = ILLA(A1) = representativeLEAVE ACT A1LEAVE MOVE M1

ENTER DISCOURSE D1

ENTER MOVE M2 /* */

273

M2 = heading[1]

ENTER ACT A2A2: [ASSERT (Author) (Message m2) (Audience)] (A2)ILLS(A2) = ILLA(A2) = representativeLEAVE ACT A2LEAVE MOVE M2

ENTER MOVE M3 /* */

ENTER ACT A3 /* */A3: [ASSERT (Author) (Message m3) (Audience)] (A3)ILLS(A3) = ILLA(A3) = representativeLEAVE ACT A3LEAVE MOVE M3

LEAVE DISCOURSE D1ENTER DISCOURSE D2

ENTER MOVE M11 /* */M11 = heading[1]

ENTER ACT A61A61: [ASSERT (Author) (Message m61) (Audience)] (A61)ILLS(A61) = ILLA(A61) = representativeLEAVE ACT A61LEAVE MOVE M11

ENTER MOVE M12 /* */

ENTER ACT A62 /* */A62: [ASSERT (Author) (Message m62) (Audience)] (A62)ILLS(A62) = ILLA(A62) = representativeLEAVE ACT A62LEAVE MOVE M12

LEAVE DISCOURSE D2

LEAVE DOCUMENT

274

There are two further points to note. Firstly, if an article contains subhead-ings, then these can be treated as layer-2 headings, as before. Secondly, thearticle headings will almost certainly appear subordinated to the newspapertitling through being printed in smaller-sized lettering. However, as alreadystated, this does not mean that the newspaper titling constitutes, in terms ofdiscourse-level analysis, the top-level heading for the individual articles.This state-of-affairs is reflected in the representation (18).As for advertisements, a comprehensive treatment of this large subject

is not feasible here. However, see, for instance, Cook (2001) who providesa useful overview, and Rush (1998) whose study of the NP in this type ofdiscourse reveals further syntactic details, especially in relation to thestructure of the modification preceding the head-word.Nevertheless, we are in a position to make some brief remarks. To begin

with, certain advertisements consist of language-text only, while others aremultimodal in character. For the moment, we shall confine ourselves to thepurely linguistic type.Some advertisements are very simple, for instance

while others are more complex. A newspaper or magazine may have col-umns of classified advertisements of items for sale, organized under head-ings such as , , and so on. In the body text of thecolumn are listed a series of entries, such as

Block advertisements, too, may involve hierar-chical organization. They may appear under general headings, such as

, and they may also manifest an internal hierarchical structure.For example, an advertisement might bear the titling

and contain one or more vacancies, each of which comprises a head-ing followed by some body text, for example

andIn

sum:

275

(19) a.University of Cheesebury

b.Lecturer in Linguistics.

c.An interest in FDG would be an advantage.

d.Lecturer in Phonetics

e.Experience of Experimental Phonetics essential.

It is apparent that this example is capable of being assigned a representa-tion along much the same lines as (18) above.

Let us now proceed to a consideration of freestanding NPs that relate not todiscourse units but to the visual illustrations that often occur in documentsof various kinds. Again it will be helpful to take an example. See Figure 1.

1994 1996 1998 2000

Year

0

20

10

1990 1992

Currency Crisis

Exports of lollipops from Slurpland during the 1990s

This chart contains several freestanding NPs, representing three differentfunctions:

276

(20) a. Figure 1Exports of lollipops from Slurpland during the 1990s

b. Exports (millions)Year9

c Currency Crisis

The term rally, it applies to theimage as a whole. A otationwhich indicates the intended meaning of some element within the image. Inthe present example the horizontal axis is labelled and the verticalaxis is labelled Athe image or to some element within it, for the purpose of aiding interpreta-tion; cf. Kosslyn (1994: 16). Here, the NP helps us to un-derstand the sharp downturn in the number of exports during 1995, andtherefore functions as a caption. It is not a label, since the element of thegraph to which it applies (namely the emand a label suchas and thelowest point on the curve would demand a label such as .The chart is not, of itself, intended to represent a graph of the occurrenceof the causal factors that might influence the level of exports, and this is, ofcourse, precisely why additional information supplied by the explanatorycaption is helpful.We are thus put in mind of the fact pointed out by Barthes (1977: 37-

41) that images typically allow for more than one possible interpretation.For instance, a chart portraying a curve within the framework of two or-thogonal axes (as in Figure 1), but without any textual annotation or ac-companiment, would be capable of being assigned an indefinitely largenumber of possible meanings. What titlings, labels and captions do, how-ever, is to point the viewer of the image towards one particular interpreta-tion among the different possibilities. Barthes terms this effect r-agerather than drifting from one possible understanding of the image to an-other. (It would also be possible for the suggested interpretation to be givenan attitudinal slant, for instance by replacing in Figure 1with a more pejorative caption such as .)Titlings, labels and captions attached to images can all be analysed in

terms of their pragmatics. Each of the five examples in (20) constitutes anindividual communicative act that is not directly a part of a larger lan-guage-text (though the image to which it is attached could be regarded as aconstituent of the overall, multimodal discourse out of which the document

277

is constructed; see Connolly 2000 for further details). These communica-tive acts may be represented along the following lines:

(21) ENTER MOVE M40

M40 = titlingENTER ACT A101 /* */A101: [ASSERT (Author) (Message m101) (Audience)] (A101)ILLS(A101) = ILLA(A101) = representativeLEAVE ACT A101ENTER ACT A102 /* */A102: [ASSERT (Author) (Message m102) (Audience)] (A102)ILLS(A102) = ILLA(A102) = representativeLEAVE ACT A102LEAVE MOVE M40

ENTER ACT A103 /* */A103 = labelA103: [ASSERT (Author) (Message m103) (Audience)] (A103)ILLS(A103) = ILLA(A103) = representativeLEAVE ACT A103

ENTER ACT A105 /* */A105 = labelA105: [ASSERT (Author) (Message m105) (Audience)] (A105)ILLS(A105) = ILLA(A105) = representativeLEAVE ACT A105

The pragmatic functions of titling, label and caption, respectively, are dulymade explicit in this representation. It may be noted that the titling in thisexample is not a heading within a discourse hierarchy. Within the titlingthere are two acts the title itself being preceded by a descriptor () whose purpose is to identify the image as a graphical element within thediscourse and to number it within a sequence of such elements within thetext as a whole.Another issue calling for comment is the question of whether the prag-

matic functions with which we are concerned in the present paper apply tomoves, acts or both. The examples adduced above suggest that if the dis-course analysis of a document reveals the grouping of acts into moves, thenfunctions such as heading or titling apply in general to entire moves

278

(though it is, of course, possible for a move to contain just one act). How-ever, it appears that pragmatic functions such as labels on images are notgenerally grouped into moves; and in cases where the layer of the move issuperfluous, the pragmatic functions concerned have to be regarded asapplying to individual acts.Further examples of NPs with the functions of titling or label can be

found abundantly in web pages. A web page regularly contains a title anddisplay of body material which is in general multimodal in character. Thismaterial often includes clickable items that facilitate links to other parts ofthe document or to other documents. These items are frequently (though byno means always) picked out in some special way, for instance throughunderlining or through mounting on a button-icon. Such items serve aslabels for the relevant clickable areas in the display, for the purpose ofindicating the information to be accessed, or the action to be performed, byclicking on the screen-areas concerned.At this point we may return briefly to advertisements, but this time to

consider those in which language-text is combined with one or more im-ages. It turns out that the pragmatic functions of titling, label and captionare all applicable to freestanding NPs attached to such advertisements. Forinstance, a page of advertisements for cars may include a series of entries,each of which consists of a picture of a particular type of car, together witha titling giving the make and model. Possibly some of these pictures mightcontain labels, such as , attached to the appropri-ate region of the image. Alternatively, an advertisement for car-interiorupholstery might consist of picture of a car interior, devoid of any titling orlabel, but accompanied simply by the caption (Whether other prag-matic functions would also be needed to deal adequately with the use offreestanding NPs in advertisements will be left as an open question.)

Dik (1997a: 1-4) envisages that a functional account of grammar shouldform part of a broader theory of verbal interaction, to be described (at leastin part) in terms of (2007). Let usnow give some consideration to this issue.As stated earlier, the syntactic structure of the freestanding NPs with

which this paper is concerned is relatively simple. Consequently, the inter-pretation of these NPs is not assisted very substantially by the grammaticalinformation embodied in them. Therefore, the reader has to place consider-

279

able reliance also on other factors in order to interpret the NPs appropri-ately.How, then, do we recognize an item such as a titling or heading for

what it is? In fact, there are at least four sources of cues to aid the interpre-tation of freestanding NPs:

(22) a. Information in the linguistic expression, describable in terms ofcontent and structure

b. Information at the graphetic level

The graphetic level is the counterpart, in the written language, of the pho-netic level in the spoken language. It is concerned with linguistically non-significant attributes of text, such as the size of alphanumeric charactersand the physical layout of text. (To describe a phenomenon as

non-significant is to state that it is not part of the language system. Itdoes not, however, in any way imply that the phenomenon in question is

irrelevant.)

c. Discourse-pragmatic factorsd. Contextual factors

In relation to information within the linguistic expression, we have to takeaccount of basically two considerations. Firstly, there is the internal com-position of the NP concerned. If the latter takes the form of block language,then this is itself a cue as to the expression atus. Secondly, there is thepossibility that the NP will be accompanied by an explicit descriptor, suchas , or a heading number such as .Graphetic-level cues often serve as an aid to the identification of titlings

and headings of discourse units, which are typically distinguished frombody text through the use of bold or italic styles of typeface. Sometimesthey are also set in larger-sized characters. In addition, titlings, labels andcaptions relating to images may differ from the body text in terms of type-face (font, style and/or size). Furthermore, the physical layout of text onthe page can provide interpretive cues. Titlings, headings and prelims pre-cede the body text to which they relate, while finishes occur at the end ofit. Moreover, titlings and headings of discourse units may be centred,whereas certain prelim items, such as the date at the top of a letter, may beright-justified. Titlings of images are normally placed immediately aboveor below the images concerned, while labels and captions are placed eitherwithin or adjacent to the images to which they relate.

280

Under the rubric of discourse-pragmatic factors comes the fact that ti-tlings and headings clearly occur at the boundaries of discourse units, en-capsulating and ushering in the text within their scope, while titlings, labelsand captions relating to images are not directly part of a larger language-text. These matters of discourse organization and coherence make them-selves apparent to readers, who are thus able draw upon them as a guide tointerpreting what they perceive.Contextual factors, too, play their part. For instance, suppose that a web

page contains a series of button icons, each of which contains a freestand-ing NP. In this case the context makes it fairly clear that each NP should betaken as a label applying to the button on which it is displayed. To takeanother example, the interpretation of the NP within Fig-ure 1 as a caption rather than a label depends not only on the content of NPbut also on the context of the whole figure. It is only because the otherfreestanding NPs in the context of the same image (namely the titling andthe labels of the axes of the chart) serve to indicate what information isconveyed by the curve that we can appreciate why the NP should be inter-preted as a caption rather than as yet another label.Not surprisingly, different types of factors can combine and intertwine.

For instance, suppose that a document contains a left-justified boldface NP,followed by a left-justified italic NP, followed by a paragraph in Romantypeface. The interpretation of these NPs intended by the author (it will beassumed) is as follows:

(23) a. left-justified boldface NP.b. left-justified italic NP.c. Roman-type paragraph.

Suppose, for the sake of argument, that the first of these NPs has alreadybeen interpreted by the reader as a layer-2 heading. What cues will thenhelp in the interpretation of the second NP as a layer-3 heading? The factthat it is some kind of heading will be suggested by its non-Roman type-face and by its position on the page, separate from the paragraph that en-sues and hence clearly located at the starting boundary of a discourse unit.(Thus, graphetic and discourse-pragmatic cues combine.) The fact that it isin a similar position on the page to the layer-2 heading that precedes it, butin a different style of typeface, suggests that it a heading of the next layerdown, namely layer-3. (Here, graphetic and contextual factors work to-gether, since the significance of the italic typeface is affected by the con-text, given that the interpretation of the italic style is conditioned by the

281

interpretation of the boldface style which occurred earlier in the samedocument.)In the light of the above, we may now address the question of how to

formulate pragmatic rules aimed at incorporating the principles behind theprocess of interpretation in cases like the one just outlined. A thoroughgo-ing solution to this problem is not possible at present. Nonetheless, it willbe instructive to make a first attempt. The following will serve as an illus-trative example:

(24)a. Expression (xi) precedes a paragraph of text ( )b. Expression (xi) is freestanding ( )c. Expression (xi) is an NPd. Expression (xi) is in a non-Roman style of typeface

Expression (xi) is a heading

This pragmatic rule consists of a set of conditions, of which at least a sub-set needs to be satisfied in order to give rise to the interpretation. Condi-tions (24a) and (24b) seem to be mandatory, while the rest serve to cor-roborate them, and the more that are fulfilled, the more confidence theremay be in the interpretation. Hence, the interpretation has the status of ahypothesis or conjecture. Of course, as stated above, (24) constitutes onlyan illustration of one possible pragmatic rule, and much work remains to bedone before a reasonably definitive set of rules can be proposed.Nevertheless, if we were, indeed, able to develop an accurate set of

pragmatic rules, then this would represent an important contribution to thetheory of verbal interaction. It would also have practical applications infields such as the automatic analysis and interpretation of documents bycomputer; see further Casey (1997) for an overview of this field, or Bunkeand Wang (1997) for a more detailed coverage.The formalization of pragmatic rules will be aided by the new architec-

ture proposed for FDG by Hengeveld (2004b), which provides for:

(25) a. A grammatical component, with interpersonal, representationaland structural levels.

b. An output component: acoustic in the case of the spoken lan-guage, while the written language has a counterpart in the opticalpatterns discerned in the shapes of characters and so on.

c. A contextual component and a cognitive component.10

282

This architecture affords the advantage of accommodating all the variousinformation sources upon which the pragmatic rules draw. Accordingly,referring back to (22) above, we see that information in the linguistic ex-pression (22a) is derived from the grammatical component (25a), whileinformation at the graphetic level (22b) is associated with the output com-ponent (25b). Discourse-pragmatic factors (22c) also lie within the scopeof the grammatical component of FDG in conjunction with the cognitive(25c) activity of discourse management (that is to say, coherent arrange-ment by the author and interpretation by the audience), while contextualfactors naturally find their home in the contextual component (25c).

The pragmatically-oriented, FDG-based treatment of freestanding NPswithin documents that has been offered in the present paper leads to twomain conclusions. Firstly, it seems feasible to extend the coverage of theFDG framework to incorporate such extra-clausal NPs, by introducing asmall number of additional pragmatic functions, including d-ing of course,encompassed all types of block language within our treatment here, butnevertheless, it is hoped that we have enhanced the FDG apparatus in amanner which is useful and which has reasonably broad applicability.Secondly, it is possible to expand the pragmatic treatment of freestand-

ing NPs beyond the grammar into the field of verbal interaction. A key stephas been to address the task of formulating pragmatic rules, as envisagedby Dik (1997a: 3-4). However, this is a field where a great deal of further,interesting research remains to be undertaken, within the supportiveframework provided by FDG.

1. This paper has benefited from comments by Chris Butler and three anonymousreferees.

2. See Hengeveld (this volume) in relation to matters of representation.3. It may be helpful to make clear the position taken here on the description of

discourse in terms of smaller units (for instance, chapters, sections and sub-sections) within FDG. Historically, the standard FG approach has been to rec-ognize units only if they have some reflex in the grammar. However, in my

283

own work, notably Connolly et al. ((1997)) and Connolly (2004), I have found(unsurprisingly, I feel) that if a comprehensive description of discourse is de-sired, then it is necessary to treat discourse in its own terms, and to employwhatever analytical categories appear to be illuminating, whether they havegrammatical reflexes or not. For instance, in Connolly et al. (1997) a purelydiscourse-oriented notion ofsome transcripts of dialogues between product-designers in an experimentalsituation. This aspect of the analysis revealed an element of cyclicity (as op-posed to hierarchy) inherent in the structure of the dialogues concerned. Thiswas a significant finding, which would have been missed if we had confinedourselves to grammatically-realized units only. I shall, therefore, persist herewith the stance that I have taken in my earlier writings.

4. A rhetorical question, to be answered by the author rather than the reader.5. This style of discourse representation is inspired by Dik (1997b: 409-441). It

assumes that discourse (D) can be analysed into smaller units, including trans-actions (T), moves (M) and acts (A). Each of these units is preceded by anENTER line and followed by a LEAVE line, which serve as brackets to en-close them. Each act is described in terms of its function in communicating amessage from its author to its audience and in terms of its illocutionary force,both from the point of view of the speaker (ILLS) and from the point of viewof the hearer (ILLA). For further details, see Connolly (2004).

6. Explanatory comments /* like this */ are intended as aids to readability, butare not part of the discourse-representation proper.

7. More precisely, a d-ing under the bridge. However, I retain the termfor the sake of consistency with my earlier writings. It is acknowledged thatfor other drivers, the road-sign would merely be informative, but these are notthe principal addressees; the sign would not have been erected for their bene-fit.

8. It is possible that another discourse analyst might regard the commonality ofcurrent interest as sufficient to treat all the articles as related, especially in thecase of a newspaper with an overall theme, such as sport. However, for ex-pository purposes, we shall here take the view that the articles constitute dif-ferent discourse strands.

9. If the numerals calibrating the axes of the graph are regarded as freestandingNPs, then they, too, should be classified as labels.

10. Whether these should be two components or one single component is a matterof debate which, however, lies outside the scope of the present paper; but seeButler (this volume), Garc (this volume), Keizer (this volume) andRijkhoff (this volume) for relevant discussion.

284

Barthes, Roland, tr. Heath, S.1977 London: Fontana Press.

Bunke, H. and Wang, P.S.P. (eds.)1997

Singapore: World Scientific.Butler, Christopher S.this vol. Interpersonal meaning in the noun phrase.

Casey, Richard G.1997 Document image analysis. In: Ronald A. Cole, Joseph Mariani, Hans

Uszkoreit, Giovanni B. Varile, Annie Zaenen, Antonio Zampolli andVictor Zue, (eds.),

, 68 71. Cambridge: Cambridge University Press.Connolly, John H.2000 Diagrams as components of multimedia discourse: a semiotic ap-

proach. In: M. Anderson, P. Cheng, and V. Haarslev (eds.),

, 479 482. Berlin: Springer.2004 The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and MarG -Gonz (eds.),

, 211 242. (Functional Grammar Series 24.) Berlin/New York:Mouton de Gruyter.

2007 Mental context and the expression of terms within the English clause:An approach based on Functional Discourse Grammar. In: MikeHannay and Gerard J. Steen, (eds.),

, 193-208. Amsterdam/Philadelphia: John Benja-mins.

Connolly, John H., Clarke, Anthony A., Garner, Steven W. and Palm1997 A functionally oriented analysis of spoken dialogue between indi-

viduals linked by a computer network. In: John H. Connolly, Roel M.Vismans, Christopher S. Butler and Richard A. Gatward (eds.),

33 58. (FunctionalGrammar Series 18.) Berlin/New York: Mouton de Gruyter.

Cook, Guy2001 2nd edition. London: Routledge.

Dik, Simon C.1997a

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

285

1997b(Functional Grammar Series 21.) Edited by Kees

Hengeveld. Berlin/New York: Mouton de Gruyter.Garcthis vol. Functional Discourse Grammar and extraction from (complex) noun

phrases.Hengeveld, Kees2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzie and Mar -Gonz, 1 21. (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzieand Mar

mez-Gonz ,365 378. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

this vol. Prototypical and non-prototypical noun phrases in FDG.Keizer, Evelienthis vol. Reference and ascription in FDG: an inventory of problems and some

possible solutions.Kosslyn, S.M.1994 New York: Freeman.

Kress, Gunther and van Leeuwen, Theo1996 London: Rout-

ledge.Mackenzie, J. Lachlan and Mar eles G -Gonz (eds.)2004 Berlin/New York:

Mouton de Gruyter.Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik1985 . London:

Longman.Rijkhoff, Janthis vol. Layers, levels and contexts in FDG.

Rush, Susan1998 The noun phrase in advertising English. 29:

155 171.

1

Although order phenomena within the noun phrase and the semantics un-derlying them have been discussed quite extensively in Functional Gram-mar (Dik 1997: 127ff.; Rijkhoff 2002), as yet not much work within the FGframework has been devoted to morphological and syntactic aspects of thenoun phrase. We are not aware of any fully worked out example of theprocess of term expression in the literature. It may be expected that theexpression rules, as they are presented in Dik (1997), will not be found fitto do the job. In Bakker (2001), it is argued that the expression rule com-ponent as it stands has several fundamental shortcomings, especially withrespect to agreement phenomena, and a new version of the FG expressionrules is proposed. Bakker (2001) takes the ingredients of the original rulesprimary and auxiliary operators, templates, and placement rules as a

point of departure. However, as opposed to the original proposal, he inte-grates these components in a dynamic framework, in which matters of formand order are interleaved. In Bakker (2005), it is shown that these dynamicexpression rules can handle complex cases of agreement at the level of theclause. On the basis of some notorious examples from Arabic, it is demon-strated that the dynamic rules may not only provide a descriptively ade-quate treatment of the cases at hand; they may also give insight into theprocessing aspects of the data, and thus potentially add an explanatoryflavour to the expression component. In other words, an attempt is made toprovide arguments for the psychological reality of theoretical constructs(Matthews 1991).In this article, we seek to implement dynamic expression at the level of

the noun phrase (NP), thus filling the above-mentioned gap in the formalcomponent of the theory. We will do this on the basis of agreementphenomena within the German NP, which is noted for its relativecomplexity. Another reason for choosing that language is the availability ofa corpus of spontaneous German speech errors, which includes a fairnumber of agreement errors within NPs. As is commonly assumed in

288

ment errors within NPs. As is commonly assumed in speech error research,we take these agreement errors, i.e. cases of rule misapplication, to shedsome extra light on the way the same rules work when they do not misfire.We will argue that it is precisely the dynamic nature of the expression

component which allows us to describe the possible mechanisms behindcertain types of speech errors which occur in German NPs. We will con-centrate on gender errors since gender is the morphosyntactic feature mostfrequently responsible for the NP-internal agreement errors we have at ourdisposal. We hope to show that these errors may be described in terms of aslight deviation in the execution of the expression rules during the produc-tion of formal structures (NPs) deriving from underlying semantic andpragmatic representations (URs). This deviation may well be explained onthe basis of the organization of the (mental) lexicon in terms of semanticand phonological networks and local activation patterns, combined withoccasional slips of attention during the complex process of sentence pro-duction.The article is organized as follows. In section 2, we will give a short ac-

count of dynamic expression rules, an elaboration of the standard FG ex-pression rules first presented in Bakker (1999). We will argue that theoriginal rules are inadequate in the sense that they both overgenerate andundergenerate. On the basis of the expression of a simple German NP itwill be shown that the dynamic version of the expression rules does nothave these shortcomings and furthermore is in full compliance with thediscourse model of FG. In section 3, we will turn to the speech error dataand discuss some of the more common error patterns as attested in GermanNPs. In section 4, we will study the implications for the expression rules.Finally, section 5 will present some conclusions.Before starting out on our journey we have to say a few words about the

status of the notionperspective, this notion refers to a specific type of syntactic structure, i.e. itis a formal notion. If we want to define it in terms of FG, two ways areopen to us, at least in principle. A strictly formal approach, typically takenby syntactic theories, would define NPs as syntactic structures with anominal head. From a more functional perspective, NPs may be defined asthe (infinite) set of forms in a language which are the expression of a term.2

In the latter case, the term would cover not only prototypical noun phrasessuch as the one from English in (1), but also the mpleteDutch in (2), the discontinuous one from Polish (Anna Siewierska, p.c.; forextraction from the NP see Garc this volume) in (3) and even theadpositional phrase from Abkhaz (Nichols 1992: 51) in (4).3

289

(1)

(2)2.SG have-2.SG ART.INDEF big house and 1.SG ART.INDEF small

(3)five buy-PAST.1.SG pencil-GEN.PL

(4)ART-river 3.SG-at

For the purpose of this article we will interpret the notion Noun Phrasemore or less dynamically as the structure that results from the expressionof a term provided that it does not have an adposition for its syntactic head.This negative characterization leaves open the possibility of includingheadless and discontinuous structures such as the ones in (2) and (3) andalso those that have pronouns for their heads. However, it excludes adposi-tional phrases such as the one in (4). As we will see, this fits in nicely withthe dynamic expression rules to which we will now turn.

As was mentioned in the introduction, the standard FG expression rules(ER) have a number of shortcomings, which fall in two general categories:overgeneration and undergeneration. We speak of overgeneration when amodel of grammar produces utterances which are not well-formed in alanguage and of undergeneration when it fails to produce certain utteranceswhich are considered to be well-formed. In cases where form and order aremutually dependent, the strategy of first generating the right morphologicalform of words and then linearizing them will run into problems. We willgive an example from Koegu, a Surmic language from Ethiopia. In thislanguage, the verb optionally agrees with the subject when it is postverbal(5a and b), and obligatorily when it is preverbal (5c and d).

290

(Hieda 1998:367)(5) a.

1.SG-drink-3.SG 1.SGb.drink-3.SG 1.SG

c.1.SG 1.SG-drink-3.SG

d. *1.SG drink-3.SG

The standard ER rules will first generate the verb form and then determinethe placement of the pronoun vis- -vis the verb. Thus, there is no way inpreventing the standard ER rules from generating (5d), unless we also giveup on (5b), which is grammatical. As shown in Bakker (2001), form-orderinterdependency is very common in the languages of the world. Thus, in amore constrained mode, the ER module will systematically miss out on anumber of well-formed constructions, such as the one in (5b), which makesit descriptively inadequate. In a less constrained, morewill produce a number of ungrammatical utterances, which will make itinadequate in the explanatory sense.

2.1. Some properties of Dynamic Expression Rules

A potential solution to this problem is to integrate into one dynamic proc-ess the rules which generate the forms and the ones that order them. In thisprocess, a tree is produced with traditional order templates as intermediatestructures (i.e. constituents) and word forms at the end of the branches. Asa result of this organization, the right form of a word may be computedprecisely when enough information is available, about both its semanticand pragmatic aspects and its eventual position in the linear ordering, orrather, in the syntactic structure. The resulting tree should not be seen as astatic morphosyntactic structure associated with a specific underlying rep-resentation URi, as in most syntactic theories both of a formal and a func-tional orientation (cf. Haegeman 1994 for the former and Van Valin and LaPolla 1997 for the latter). Rather, it is the result of a dynamic process start-ing out from URi, and in fact reflects its own history. This process takesplace in a top-down fashion, from left to right and depth first. Functionaland formal features may be inherited from higher nodes, or percolate up-

291

wards from lower nodes. In Bakker (2005) it is shown that a limitedamount of in order tocater for cases of anticipatory gender agreement. As we will show in sec-tion 3, this extra feature will be very helpful in explaining certain speecherrors.As in the case of other constituent structures, dynamic expression trees

consist of a number of nodes, which should be interpreted as the constitu-ents of the tree. However, in this case, they are considerably more thanmere labels. Each node expresses a specific part of the combined prag-matic-semantic underlying representation URi, including its lexical mate-rial, operators, functions, and layered structure. A node explicitly specifiesthe primary and auxiliarythe expression process. In fact, some nodes may only containand no lexical material at all. On the basis of its complex of lexical andgrammatical material, a node selects a template from the grammar, whichspecifies the linear order of the actual elements of the URi element handledby the node. For each of the slots in that template a new daughter node iscreated, in a left-to-right fashion, which then will deal with the element ofthe mother node that has been assigned to it, and so on recursively. In thatsense, a node can be seen as a device for mapping a specific set of underly-ing functional material (e.g. a term) onto a formal expression (e.g. a nounphrase). This process continues until each lexical element resulting fromthe original URi has been assigned its own terminal node.In contrast to the standard model of expression, grammatical elements

such as articles, auxiliaries and plural suffixes are not created before butrather during the process of linearization, at the moment at which preciselyenough information is available to determine the right form and in time forthem to introduce features that derive from them, such as case. Typically,grammatical markers such as articles and auxiliaries are the expression of aterminal node, which contains onlywhatsoever. The corresponding grammatical element is then selected bythe grammar. Some terminal nodes may even be prespecified for a form,which is then considered to be completely grammaticalized. This may bethe case for bound forms such as plural suffixes, which only have phonol-ogically determined allomorphs.

292

2.2. Exemplary derivation of a term

Let us now illustrate the relevant processes with a simple example. Theaforementioned top-down process starts with the creation of a node for thefully specified underlying representation being expressed, typically aclause, but it may also be a term or any other well-formed underlying con-struct. For the German term in (6) below, the top node for its expressionmay appear as in (7). In node specifications, i-fies the node in the grammar. It corresponds to the notion of functional slotin the traditional order templates. e-gory of the constituent to be expressed via this slot.that prespecifies the type of configuration to be expressed by this type ofnode, in this case a term. The (infinite) set of all possible configurationsthat fit this description could be seen as the recursive definition-by-enumeration of the func-tional (i.e. primary) and formal (i.e. auxiliary)node, respectively. Finally,ofthe subconstituents will be expressed. It could be seen as the (recursive)definition-by-enumeration of the i-tion of morphosyntactic constituency in the corresponding language.4

(6) (def pl x1 : Katze [n, fem] : schwarz [a])Patient(

(7) Node k (uninstantiated)Slot: termCat: npConfig: (DEF NUM VAR: PRED [n, GEN]: ( RESTRn [CATn] ) )

SEMFNCFncFtrs: SEMFNC, DEF, NUMFrmFtrs: GEN, CASESubCat:

(7) gives the node in its abstract, uninstantiated form, as it appears in thelanguage-specific grammar. It is selected on the basis of certain formalaspects of the UR that is being expressed, in this case a noun phrase. Theconfiguration field of this node will now be filled by the term under (6),which may or may not be part of a more complex structure being ex-pressed, such as a clause. This leads to the partially instantiated version in

293

(8) below. Note that the two types of feature sets are instantiated at thesame time, to the extent that their values are available.

(8) Node 1 (partially instantiated)Slot: termCat: npConfig: (def pl x1 : #katze [n, fem] : #schwarz [a])PatientFncFtrs: patient, def, plFrmFtrs: fem, CASESubCat:

In the next step, the formal features which are still unbound will be instan-tiated. Since there is no higher node from which to inherit, CASE will begiven a value by local rules. In this case, since the semantic function isPatient and the term does not have Subject function, the default value willbe accusative.5

(9) Node 1 (partially instantiated)Slot: termCat: npConfig: (def pl x1 : #katze [n, fem] : #schwarz [a])PatientFncFtrs: patient, def, plFrmFtrs: fem, accSubCat:

In the last step, the right template will be selected for the expression in theSubCat field of Node 1. This will give us its fully instantiated version in(10).

(10) Node 1 (fully instantiated)Slot: termCat: npConfig: (def pl x1 : #katze [n, fem] : #schwarz [a])PatientFncFtrs: patient, def, plFrmFtrs: fem, accSubCat: det, adjectival, nominal

The process now continues in a left-to-right, top-down fashion with theexpansion of the determiner slot, rnativenodes will be available, a situation comparable to the alternative placement

294

rules for the P1 slot in the standard theory (cf. Dik 1997: 421). The rightcandidate node will then be selected, and the corresponding element of theConfig field of node 1 will be inserted into the Config field of this newlycreated node 2 in this case the information pertaining to the article. Thishappens to be a grammatical node, i.e. one that leads to the direct expres-sion of a grammatical morpheme without further expansion. Before andafter instantiation, node 2 appears as in (11) below. Grammatical node 2 isprespecified for the categoryfrom the mother node 1. With the relevant feature values instantiated, theright form will be retrieved from the grammatical subsection of the lexi-con.6

(11) Node 2 (uninstantiated) Node 2 (fully instantiated)Slot: det detCat: art artConfig: FORM

[art, DEF, GEN, CASE, NUM] [art, def, fem, acc, pl]FncFtrs: DEF, NUM def, plFrmFtrs: GEN, CASE fem, accSubCat: FORM

The next slot that is expanded is the adjectival one. This node may becomplex, and have subcategories for the adjective and its adverbial modifi-ers, as in ( ewill assume that the expression rules will makenodes catered for, in this case to the node for the expression of a singleadjective. In other words, dynamic expression trees do not have any super-fluous structure, only precisely what is relevant. In the grammar of Ger-man, the node which expands an adjective may be morphologically com-plex, in order to cater for the inflectional case/number suffixes (as well asfor derivational affixes, such as the comparative and superlative). This isnode 3 under (12).

295

(12) Node 3 (uninstantiated) Node 3 (fully instantiated)Slot: adjectival adjectivalCat: adjcmp adjcmpConfig: PRED [a] #schwarz [a]FncFtrs: DEF, NUM def, plFrmFtrs: GEN, CASE fem, accSubCat: adjhead adjsfx

This leads to the following expansions, firstthen

(13) Node 4 (uninstantiated) (fully instantiated)Slot: adjhead nomheadCat: adj adjConfig: PRED FORM [a] #schwarz -SubCat: FORM -

(14) Node 5 (uninstantiated) (fully instantiated)Slot: adjsfx adjsfxCat: asfx asfxConfig: FORM -en

[asfx, GEN, CASE, NUM] [asfx, fem, acc, pl]FncFtrs: NUM plFrmFtrs: GEN, CASE fem, accSubCat: FORM -en

Subsequently, the rightmost element of the NP template the position ofthe nominal head of the term will be expanded. This will not be a termi-nal node either, since it has internal morphological complexity as a resultof the values for CASE and NUM. Thus, this node is expressed via a nodefor the stem and one for the plural suffix. (15) (17) below give the threenodes necessary for this process. Node 6 is the top node of this local struc-ture. Only after partial instantiation with the corresponding nominal predi-cate and inspection of noun specific information is it possible to expandthis node.7 In this case, the regular rules for the German plural femininenoun apply.

296

(15) Node 6 (uninstantiated) (partially inst.) (fully instantiated)Slot: nominal nominal nominalCat: nouncmp nouncmp nouncmpConfig: PRED [n, GEN] #katze [n, fem] #katze [n, fem]FncFtrs: NUM pl plFrmFtrs: GEN, CASE fem, acc fem, accSubCat: nomhead nounsfx

(16) Node 7 (uninstantiated) (fully instantiated)Slot: nomhead nomheadCat: noun nounConfig: PRED FORM [n] #katze -SubCat: FORM -

(17) Node 8 (uninstantiated) (fully instantiated)Slot: nounsfx nounsfxCat: nsfx nsfxConfig: FORM [nsfx, GEN, CASE, NUM] -nFncFtrs: NUM plFrmFtrs: GEN, CASE fem, accSubCat: -n -n

The result of the whole process can be represented as the tree in figure 1below, being the expression of the term in (6) above. Note again that this isnot the (static) tree stipulated for the term structure but a result of the ex-pression process as such. The figures next to the branches indicate the or-der in which node expansion has taken place.No special provisions are necessary to fit the Dynamic Expression Rule

module (DER) into the extended model of FG, Functional DiscourseGrammar (FDG; Hengeveld 2004, this volume). Both for the old and thenew model, expression starts off when all relevant pragmatic and semanticmaterial is available. In the old model, all information, with the inclusionof pragmatic and even syntactic functions, is compiled in the so-calledunderlying clause. In the FDG model, this information is distributed overthe interpersonal and the representational layers, respectively. In principle,all this pragmatic and semantic information is considered to be accessibleto the expression rule component, but it is an empirical question what theconstraints on this accessibility could be. Bakker and Siewierska (2004)discuss more extensively the implications of dynamic expression for thestructure of the interface between the underlying and expression levels of

297

the model. They also demonstrate how the dynamics of the expressionrules may be added to the construction of the underlying representations ofthe interpersonal and representational layers as well, giving the model thefull flavour of a model of the speaker, including the explanatory aspectsthat such a model may provide.

Node 1

13 4 11 12 19

Node 2 Node 3 Node 6

5 7 8 10 13 15 16 18

2 Node 4 Node 5 Node 7 Node 5

6 9 14 17

-en -n

. Full tree expansion

This concludes our example of the expression of the German NP. In orderto further motivate the structure-building processes proposed in this sec-tion, let us now have a look at some spontaneous German speech errors inwhich structural regularities within the noun phrase are disrupted in oneway or another.

In a comprehensive study of the processing of morphosyntactic categoriesand features in language production, Pfau (2000) investigates, amongstother things, the manipulation of morphosyntactic features in on-line pro-duction, above all by studying the role these features play in spontaneousGerman speech errors. The theoretical perspective taken is that of Distrib-uted Morphology (Halle and Marantz 1993; Harley and Noyer 2003). Ul-

298

timately, Pfau (2000) tries to show how the derivational processes assumedin Distributed Morphology (merger, morpheme insertion, feature copy,vocabulary insertion, phonological readjustment) can be mapped onto atop-down multi-level model of language production (e.g. Fromkin 1971;Garrett 1980; Levelt 1989) in order to account for a wealth of speech errordata.

3.1. The error corpus

As is well-known, the collection of speech errors has a very long history.Anwar (1979) points out that evidence exists that Arab linguists were col-lecting and analysing slips of the tongue as far back as the eighth century.More than one thousand years later, Rudolf Meringer, a linguist from Vi-enna, collected his famous and extensive error corpus (Meringer andMayer 1895). Apart from some scattered references, however, it was onlyin the late sixties of the 20th century that linguists regained interest in theinvestigation of speech errors (Fromkin 1968; Nooteboom 1969).For his study, Pfau (2000) compiled a corpus of speech errors the cur-

rent version of which contains 612 errors. This corpus is biased albeitdeliberately in the sense that it contains only errors which are relevant inthe context of that study. On the one hand, it includes relevant slips fromthe existing Frankfurt speech error corpus compiled by Prof. Helen Leun-inger and her team since the early 80fair number of slips himself. These were sampled in a paper-and-pencilfashion with as much of the error context as possible (e.g. embedding ofthe error element(s) in a sentence or discourse as well as possible non-linguistic influences). Subsequently, the data were classified according to anumber of criteria (type of error, error element, relevant feature, accom-modation, self-correction, and so on) and fed into a database.Of the 612 slips in Pfau rrors due to fea-

ture mismatches, be they agreement errors (due to person, number, gender,or case mismatch) or subcategorization errors (due to case mismatch). Outof a total of 252 agreement errors, 155 are errors in subject-verb agreement(see Pfau 2003 and Pfau and Bakker 2004 for discussion). In this article,we will draw upon 97 slips from the corpus that involve a feature mismatchwithin the noun phrase. These can be categorized as follows.

299

Types of feature mismatch within NP

Error in: N %

1. GENDER 86 88.7 %

2. NUMBER 6 6.2 %

3. GENDER + CASE 2 2.1 %

4. GENDER + NUMBER 2 2.1 %

5. CASE 1 1 %

3.2. Errors involving gender mismatch

Let us first look at mismatches that involve the gender feature. In the Ger-man noun system, three genders occur, viz. Masculine (M), Feminine (F)and Neuter (N). The assignment of a particular noun to a gender is highlyidiosyncratic. None of the three genders is particularly rare. Looking atTable 1, we observe that gender is by far the most frequently affected fea-ture in slips.8 Three representative examples involving gender, each beingof a different error type, are given in (18) to (20).In (18), we are dealing with a form-based noun ; the intrud-

ing noun ( ( o-nologically similar but are of a different gender. Example (19) is a oftwo competing noun phrases, which contain the (near) synonymous nouns

( ( inal( m-

peting noun, which is not part of the utterance. Note that in the nominative,the indefinite article is homophonous between masculine and neuter gen-der; therefore, the gender mismatch is only visible on the adjective. In (20),the of two nouns gives rise to gender conflicts in both positions.In this example, the self-correction reveals that we are in fact dealing witha noun exchange and not with an exchange of definite articles.9 Note thatall three slips contain self-corrections.10

(18)always the.M same chaos(N), er, case(M)

same case.

300

(19)that be.FUT a.M/N very exciting-N competition(M), er, final(N)

exciting-N final(N) // exciting-M competition(M)e a very exciting final // competition.

(20)I have-1.SG the.F Lothar(M) with the.M Rosi(F) PPT-meet-PPT

the.F Rosi(F) with the.M Lothar(M).

In all, gender is involved in 93% of all errors (types 1, 3, and 4 in table 1).However, it is not always clear whether we are in fact dealing with a

nvolved. As shown inTable 2 below, there are a great many instances of (vertical) homophonywithin the three paradigms of the German definite article (in bold), as wellas (horizontal) homophony between the genders (in italics).

Homophony in the German definite article.

DEF

MASC

DEF

NEUT

DEF

FEM

SG NOM

SG GEN

SG DAT

SG ACC

PL NOM

PL GEN

PL DAT

PL ACC

Example (21) below is a case in point. Actually, this error is ambiguous intwo respects. First, the neuter definite article is homophonous in thenominative and accusative. It can therefore not be decided whether theobserved mismatch is due to gender only or rather to gender and case. If we

301

assume that the wrong choice of article for the first noun (caused by anticipating the gender of the second noun ( k-endtoo (the preposition ( wever, due to thehomophony mentioned above, the only error we can be sure of is the wrongchoice of gender. All such ambiguous cases were conservatively treated asgender only errors (type 1 in Table 1). Secondly, since the error is self-repaired by the speaker immediately after the mismatch occurs, it is notclear whether we are dealing with an or with an incompleteexchange. The error in (22), in turn, is one of the only two instances whichunambiguously involve a combination of gender and case (type 3). Notethat proper names can take definite determiners in German.

(21)now the.N.NOM/ACC look(M), the.N.NOM look(M)

at.the.N.ACC weekend(N)

(22)the.M.ACC Bettina(F), er, the.F.DAT Bettina(F)

have.COND-2.SG you the.M.ACC also give-INF can-INF

Finally, to more or less complete the picture, we give one slip in (23) ablend in which the mismatch is only due to the number feature (type 2):the definite article is singular while the noun (plural.

(23)the.M.NOM people, er, the.M.NOM man(M) has PPT-say-PPT

the.M.NOM man(M) // the.PL people

The gender feature is inherent in nouns and prominent across noun phraseerrors while number is a term feature and case a contextually specifiedfeature. For the sake of clarity, we will therefore restrict our attention tothe 86 examples of type 1, with gender unambiguously being the only fea-

302

ture responsible for the error. As the examples given above make clear,gender errors in German are typically manifested in the choice of article, insome cases such as (19) also in adjectives, and in yet other cases in posses-sive and relative pronouns.

3.3. Distribution of gender errors

Let us now have a look at the distribution of gender mismatches acrosserror types. Note that slips which are ambiguous between anticipation andexchange (e.g. (21)) were treated as anticipations. Actually, most of theerrors classified as anticipations below are of this type. Apart from the fourtypes of gender errors already exemplified above, we found seven gendererrors due to a , one of which is given in example (24) below.In this case, the gender of a noun phrase seems to be carried along to anoun phrase that appears later in the expression.

(24)the.F.NOM two(F) spades(N) is next.to the.F.DAT ace(N) hearts(N)

the.N.DAT ace(N) hearts(N)

In table 3 below, we give the frequency distribution across the five errortypes of the 86 gender agreement errors which, in one way or another, in-volve two nominal predicates. Blends and substitutions we call paradig-matic errors since they make the wrong choice between an intended nounand a noun which is semantically or phonologically related to it, but whichis not part of the intended utterance. In contrast, exchanges, anticipationsand perseverations we call syntagmatic because they imply some sort ofinteraction between two elements which are both part of the intended utter-ance. In the table, the figures in brackets represent the number and the per-centage of self-corrections for each of the five types. For instance, 20 outof 42 blends (i.e. 47.6 % of all blends) are self-repaired.

303

Distribution of gender mismatches across error types

Gender mismatch due to: N %

PARADIGMATIC ERROR

1. Blend 42 (20) 48.8 (47.6) %

2. Substitution 12 (8) 14.0 (66.7) %

SYNTAGMATIC ERROR

3. Exchange 12 (7) 14.0 (58.3) %

4. Anticipation 13 (7) 15.1 (53.8) %

5. Perseveration 7 (1) 8.1 (14.3) %

Almost half of the gender errors (48.8%) are the result of a blend. As theself-repairs show, these errors are typically caused by the fact that two(near) synonyms, both compatible with the intention of the message, areselected in parallel. Gender information, and in connection to it the form ofthe article, is based on the intended noun. When a (near) synonym is se-lected as the nominal head of the noun phrase, which has a different gen-der, this will lead to a gender conflict.11

As mentioned above, substitution errors are considered paradigmatic,too, in that the error element is not part of the intended utterance but ratherphonologically or semantically related to one of the elements contained inthe utterance, in this case the intended noun. Interestingly, all the gendermismatches in the corpus are due to formal substitutions, as in example(18), while all semantic substitutions in the corpus that could in principlehave given rise to a gender conflict involve accommodation, that is, theerroneous utterance is brought into line with the grammatical requirements.An example of this is found in (25) below. Note that in contrast to the for-mal substitution in (18), an ungrammatical string ( N door(F) )does not surface.

304

(25) ,you must the.F.ACC door(F) then hold-INF,

rubbish, the.N.ACC window(N)then.

Pfau (2000, 2007) claims that this well-formedness is due to the fact thatsemantic substitutions take place early during the language productionprocess (Garrett 1980; Levelt 1989) while formal substitutions take placeonly after gender agreement has been computed and can therefore not berepaired anymore.In contrast, exchanges, anticipations, and perseveration are syntagmatic

in nature in that the source of the error is part of the intended string. It iseither an abstract feature that is shifted or a word, be it a noun or an article.While the latter is true for the examples in (20) and (22), the former holdsfor the slip in (26) below.12 Clearly, this error does not involve the antici-pation of a phonological word. Rather, it might be the female gender fea-ture of the noun (the definiteness of ( .13

(26)1.SG.DAT stands the.F.NOM sweat(M) on the.F.DAT forehead(F)

the.M.NOM sweat(M) on the.F.DAT forehead(F)

Of the 86 errors in table 3, 69 involve nouns, 11 involve misplaced articles,and six are clear instances of gender feature shifts (as in (26)). However,all examples involving articles are ambiguous in the sense that it cannot bedecided whether the article itself, i.e. a phonological form, has been mis-placed or whether it is only an abstract feature which is responsible for themismatch. Hence, in (27) either the whole definite article of the sec-ond NP may have been anticipated or only its masculine feature.

(27) ,I want-PAST only DEM with the.M.DAT time(F),

with the.F.DAT time(F) and the.M.DAT space(M) know-INF

305

If we opt for the second interpretation, then all 17 cases of errors in articlescan actually be considered the result of a feature misplacement.Note that in exactly 50% of the cases, self-corrections provide informa-

tion about the intended utterance. This shows that speakers are also hearersin the sense that they monitor their own output, and regularly repair it inthe case of an error (Levelt 1983, 1989). As a result, we can tell with cer-tainty in cases like (18) and (19) what the intended noun was, while with-out self-repairs we could only speculate about that.Let us now see how these different types of speech errors may be fitted

into the model of FDG, and implemented in the dynamic expression rulesas discussed in section 2.

Let us first try and define the five types of errors in terms of the FDGmodel. Thus, we distinguish between terms in the underlying representa-tion (UR) on the one hand and noun phrases as part of the actual expres-sion on the other hand. In the case of paradigmatic errors, only one NP isrelevant, i.e. the one that is the locus of the gender error. This will becalled NP1, its article ART1 and its nominal head NOUN1. When a secondNP is relevant, expressed before or after NP1, as in the case of syntagmaticerrors, this will be called NP2, with ART2 and head NOUN2. The term inthe underlying representation of which NP1 is the expression will be calledT1 and its head noun Nt1. In case of an NP2, its underlying term will becalled T2 with head noun Nt2.In an error-free expression, we would have the situation shown in Fig-

ure 2. GND = Nt1 should be read as t1 1 asdefiniteness value of T1errors can be accounted for in such a model, while in section 4.2, we willdiscuss syntagmatic errors in more detail.

306

UR

T1 (Nt1) T2 (Nt2)

NP1 NP2

ART1DEF = T1GND = Nt1

NOUN1 = Nt1ART2DEF = T2GND = Nt2

NOUN2 = Nt2

Relation between terms and noun phrases

4.1. Paradigmatic errors

In the case of paradigmatic errors, we have only NP1 and T1. Furthermore,feature DEF derives its value from T1 and feature GND derives its valuefrom Nt1. However, NOUN1 is not the form of nominal predicate Nt1.Rather, it is the form of some nominal predicate Nx in the lexicon which iseither phonologically similar (form-based substitution, as in example (18)repeated below) or synonymous to Nt1 (blend, as in example (19) repeatedbelow). Interestingly, in 8 of the 42 blends, the two competing nouns arealso phonologically related. The slip in (28) exemplifies this phenomenon(see Dell and Reich 1981 and Harley 1984 for discussion of such multi-causal errors and the implications for the assumed independence of proc-essing levels in top-down models of speech production).14

(18)always the.M same chaos(N),er, case(M)

307

(19)that be.FUT a.M/N very exciting-N competition(M), er, final(N)

//exciting-N final(N) // exciting-M competition(M)

(28)the.M chin(N), the.N chin(N), er, the.M jaw(M)The chin, the chin, eh, the jaw.

The expression process of NP1 may run more or less as follows. Initially,reference will be made to the correct nominal predicate, Nt1. If we considerthe expression of both (18) and (19) above we see that gender for theerroneous NP is taken into consideration not only for the expression of thearticle, the leftmost node of the NP, but also for the expression of the ad-jectival modifiers. This suggests that all prenominal modifiers express acommon auxiliary operator, which is retrieved only once on the basis of theinspection of the lexical entry of the intended Nt1. This availability at arelatively early stage of expression of the semantic and lexical aspects ofpredicates is fully compatible with the findings of Garrett (1980) andLevelt (1989) referred to above.For blends like the one in (19), we assume that when the process back-

tracks to fetch the phonological shape of the nominal head, something goeswrong, arguably along the following lines. Since the lexical entry of Nt1has been accessed at an earlier stage in order to provide the gender infor-mation, we may assume that its immediate lexical vicinity has been acti-vated and has therefore become more accessible as a whole. There is ahigher chance now for the retrieval of lexical items which are part of thesame semantic domain since we may assume that these semanticneighbours have received some activation due to their proximity to theentry for Nt1 in the semantic lexicon. This explains why the vast majorityof erroneous NOUN1 in the corpus are of this type: 42 out of 54, or 78%have a semantic similarity. Crucially, phonological similarity cannot play arole since lemmas within the semantic lexicon have no phonological speci-fication.In the remaining 22%, the intruding noun has a more or less clear pho-

nological similarity to Nt1. The nouns interacting in formal substitutionsbelong to the same phonological cohort in the phonological lexicon, i.e. thesubset of lexical items which start out with the same sequence of pho-nemes. Phonologically similar forms therefore receive activation from the

308

intended noun once the initial phonemes of the form under selection areavailable. Under certain circumstances, speakers may then stray to anotherentry in the cohort, and select the wrong form, as in (18).15 As we haveseen above, in example (28), these two error strategies may strengthen eachother: in 8 out of 42 semantic substitutions, the intended and the intrudingnoun are also phonologically related.We may implement the two types of paradigmatic errors as follows in

the model of dynamic expression. When the node for NP1 is created on thebasis of T1, its grammatical features will be instantiated, gender being oneof them. For this, a first inspection of the lexical entry of Nt1 will takeplace. We then proceed to the expression of the article and, possibly, othermodifiers of the nominal head, all of which will get the functional andformal features stemming from T1. When we backtrack to the NP node andreach the slot for the nominal head, the lexicon has to be accessed again tofetch the corresponding phonological material. Assuming that the organiza-tion of the FG lexicon is in keeping with psycholinguistic evidence for it(i.e. spreading activation from target noun to semantically and phonologi-cally related nouns), this will lead to blends and substitutions.

4.2. Syntagmatic errors

In the case of errors of a syntagmatic nature matters are quite different.Here we are dealing with two terms, T1 and T2 and their respective nominalheads, Nt1 and Nt2. For the three types of errors in this category, we estab-lished the following error numbers, mainly on the basis of self-correctionsand logic.

. Nature of syntagmatic errors

NOUN dislocated ART dislocated

Exchange 9 3

Anticipation 5 8

Perseveration 1 6

309

Where semantic or phonological similarity in the lexicon may provide agood explanation for the paradigmatic errors, we assume that for the syn-tagmatic ones, there are two related factors which may cause them. Firstly,terms T1 and T2 are both elements of the UR being expressed. The seman-tic/syntactic link between them is typically quite tight. They may, for ex-ample, be arguments of the same predicate, be part of the same complexterm, have parallel positions in a nominal or clausal coordination, or areotherwise intimately related, for instance, in an idiomatic expression. Sec-ondly, as a result of the relatedness of their underlying terms, NP1 and NP2may be sister nodes or otherwise end up as immediate neighbours in theresulting syntactic structure. Indeed, this is the case for the vast majority ofthese errors. (29) is an example.

(29)fetch-2.SG you me in the.F.DAT towel(N)

the.N.ACC kitchen.towel(N)

in the.F.DAT kitchen(F) the.N.ACC kitchen.towel(N)get me the kitchen towel from the kitchen?

We will assume that gender errors of the syntagmatic type may be ex-plained by a combination of the two factors mentioned above. Thus, thesemantic closeness of the two terms in the underlying representation makesthem candidates for confusion, especially during the retrieval of their verycentral nominal heads, more or less in the manner of lexical entries in thecase of paradigmatic errors. And this confusion may be heightened by thedynamics of the syntactic process in which the NPs are created. For thelatter we will invoke the mechanism of parallel processing, briefly intro-duced in section 2 as k-ker (2005) in order to explain the following cases of agreement in Arabic.The examples under (30) and (31) below are taken from Mohammad(2000: 112ff).

310

(30) a.the-girls and-the-boys read.3.PL.M book

b.read.3.SG.M the-boys and-the-girls book

c.read.3.SG.F the-girls and-the-boys book

As exemplified by (30a), in the case of preverbal compound NPs, genderagreement on the verb is always masculine, irrespective of the order of thecoordinated NPs, provided that one of them is masculine. However, in thecase of a postverbal subject, the gender is taken from the leftmost of thecoordinated NPs, irrespective of the gender of the rightmost NP (30b/c).This may be explained by assuming for the V-S case that during the ex-pression of the verb, processing of the next constituent, i.e. the subjectterm, has already started in parallel. Apparently, the gender of the leftmostdaughter of the Subject noun phrase is available before the final form ofthe verb has been established.16 Further evidence for an analysis of Arabicsubject-verb agreement along these lines comes from examples such asthose in (31) where we have a V-X-S situation, i.e. there is a constituentbetween the verb and the postverbal subject. In such configurations genderagreement is either with the subject, as in (31a), or we get the default mas-culine agreement, as in (31b). Apparently, the intermediate locative con-stituent renders the gender information of the subject inaccessible, at leastin some cases.

(31) a.PRT was-3.SG.F settle-3.SG.F the-Iraq-ACC

people.F.PL different

b.PRT was-3.SG.M settle-3.SG.M the-Iraq-ACC

people.F.PL different

311

Now let us see to what extent these mechanisms may be applied to thesyntagmatic gender errors. The five examples of nominal anticipation, as inexample (29), seem to be the most straightforward cases. As in the case ofthe paradigmatic errors, first comes the expression of the (correct) article,and then, assuming that T2 is either pragmatically prominent or alreadybeing processed in parallel, Nt2 is selected for expression in NP1. The sin-gle case of nominal perseveration is the mirror image of this: T1 is moreprominent than T2 and as a result its nominal head is retrieved when NP2 isbeing expressed despite the fact that T1 has already been expressed. Nomi-nal exchanges may be seen as anticipations which are not interrupted half-way by a self-correction, but which are continued and n-serting the non-expressed Nt1 in NP2.If taken literally, the dislocated articles are somewhat problematic for

the dynamic expression rules as they stand. Articles, and most other gram-matical elements come into existence very late in the derivation of an ut-terance, i.e. when they are about to be expressed. This would mean thatthey are not available for exchange with grammatical entities earlier on,unless we assume that parallel processing may make them available prema-turely. We would predict, then, that these phenomena happen more often inlanguages with prenominal articles than in those with postnominal ones.However, we have no evidence for this. Therefore, it seems to be safer toassume that all cases of article dislocation are in fact cases of gender dislo-cation, as has already been suggested above. This would be more in accor-dance with the model and what we observed above in relation to example(26). Further evidence for this analysis may be derived from errors like theone in (21) above, repeated below for convenience. In this slip, the genderof NP2 is anticipated in NP1 rather than the article as such since it ismerged with the preposition ( + ).

(21)now the.N.NOM/ACC look(M), the.N.NOM look(M)

at.the.N.ACC weekend(N)

The difference between noun dislocation and article dislocation boilsdown, then, to a difference between the of inspection. So for an-ticipations, selecting Nt2 instead of Nt1 leads to noun dislocation when donewhile NOUN1 is about to be expressed and to article dislocation whileexpressing ART1. To the extent that any statistically robust conclusions

312

may be drawn on the basis of the rather low figures in Table 4, they seemto suggest that both scenarios are more or less equally likely for anticipa-tions. A corollary of this analysis is that article dislocation would affect thegender but not the definiteness of the respective NPs. Indeed, we have notfound one example of an error in definiteness in the available data.17

Although article dislocations are relatively rare for exchanges, they dooccur. We may speculate that, while noun exchanges are in fact the mostcommon error, it is not very likely that speakers willwrong choice of gender in NP1 by expressing the gender of Nt1 on NP2,since their contribution to the semantics of the utterance is close to zero.For perseverations, articles are by far the most common error location.

For these cases, we assume that when the gender for NP2 is retrieved inorder to compute the right form for ART2, T1 is still in the last stage ofexpression, which in German typically concerns the head noun, Nt1. At thisstage it may compete with the local Nt2 and provide the wrong gender fea-ture.

The dynamic expression rules were initially introduced to implement mor-phosyntax in the original FG model without the obvious problems inherentin the original expression rules as discussed briefly in section 2. In Bakkerand Siewierska (2004) it was shown that they fit very well into the FDGmodel, especially when this is seen as a model of language production, orrather a model of the Speaker. The diachronic implications of the proposalwere first introduced in Bakker and Siewierska (2002), where the trajec-tory from relational noun via the two types of adposition to case affix wasdiscussed. In the current article, we hope to have shown that the dynamicexpression rules may be applied to the noun phrase as well, where theytake care of the order of the lexical and morphological material and theseveral types of agreement in these constituents.In order for a theory to be cognitively adequate, it should obviously be

in harmony with existing models of language production and perception.However, given two theories which fare equally well at these points, wethink that the more adequate of the two is the theory which also tells uswhat can go wrong during speaking, and why. Thus, a theory should notonly tell us how to derive all grammatical utterances of a specific language,it should also tell us which speech errors are more likely to be made thanothers. Thus, a theory should not be based solely on assumptions concern-

313

ing the well-formedness of utterances. As an empirical basis, all humanlinguistic output should be taken seriously, and this includes frequentlymade speech errors and, importantly, the self-corrections that speakersmake and do not make. Only then may we arrive at a fully fledged theoryof human linguistic behaviour rather than a theory about linguistic compe-tence. The former happens to be the final objective of FG. The latter is atbest only part of such a theory. It is the mechanism which is activated inlaboratory experiments about well-formedness and, probably more interest-ingly, in the case of self-corrections. It is the experience of linguists, for-malists and functionalists alike, that grammaticality judgements often varybetween native speakers, and are often not of a clear-cut yes/no nature. It isprecisely here where grammaticality shades off into ungrammaticality, andsynchrony into diachrony.One way of measuring the strength of an empirically oriented theory is

the number of phenomena it successfully describes (and explains), i.e. itsempirical coverage. Arguably more interesting, however, are predictionsconcerning phenomena which may not be encountered, i.e. to what extentthat theory is falsifiable. Put in the slightly more modest terms of a theoryabout probabilities rather than possibilities, this is the question of the(un)likelihood of certain structures occurring in linguistic output, in a spe-cific language or in languages at large. Restricting ourselves to agreementerrors in German noun phrases, and taking the static FDG model and ourcurrent view on dynamic expression as points of departure, we would argueas follows.18 Definiteness and number are determined during term forma-tion, at the interpersonal and representational levels, respectively. In thecurrent view on node instantiation during expression, these values are as-signed as functional features upon the creation of the NP node which ex-presses the term, presumably with semantic features being in the scope ofpragmatic features. They are established just once and never reassessedduring the expression of that node, only copied to the relevant daughternodes via inheritance. Therefore, it seems to be less likely that they changeduring the expression of the noun phrase under that node. Indeed, as isclear from the figures in Table 1, errors with number as a factor are rela-tively rare, and definiteness is absent altogether. As for gender, not a fea-ture of the term but of the nominal head, matters are quite different. Ini-tially, gender is established as a formal feature of the NP node byinspecting the lexical entry of the nominal head. As such it is inherited bythe prenominal modifiers, and expressed. However, once we turn to theexpression of the nominal head, the lexicon is accessed again, now in orderto fetch the phonological and morphological specifications of the noun. As

314

we have argued, this complex process may sometimes result in fetching theinformation from a lexical entry or a term next door. In this process, nodirect inspection takes place of the gender features of the noun, whichwould prevent the noun from being uttered. The clash may become onlyevident after the utterance, and this may lead to self-correction, or not.Although case errors are extremely rare as the single cause for mistakes(only 1 instance), there are no fewer than 21 instances where they combinewith gender errors. On face value, this is not what one would expect on thebasis of the theory. Case is a formal feature of the NP node, either inheritedfrom the mother node or established locally. Just like number and definite-ness, it has scope over the whole NP and is never reassessed after beingestablished. It is therefore quite unlikely that its value will change whilethe term is in the process of being expressed. However, given the high de-gree of homophony between the German article paradigms, 20 of thesemight as well be interpreted as instances of just the wrong gender ratherthan a combination of wrong gender and case.19

Thus, the model seems to u-tion of the different types of possible errors in German. Another predictionwhich could be made is that one would expect fewer errors of this type inlanguages which have gender distinction in their articles but N-Art order.In such languages, the optimal way of determining gender in the NP seemsto be by percolating the gender feature after expressing the nominal head.Under such conditions there seems to be less reason to re-inspect the lexi-cal entry, with a possible change of the feature value. Until a corpus ofspeech errors in such a language is available we rest our case.

1. The authors are grateful to three anonymous reviewers who proposed a num-ber of improvements to the original text. Only we, however, are to blame forany mistakes and inconsistencies that remain. Authors -mail addresses:R.Pfau@uva.nl and D.Bakker@uva.nl.

2. Here and below we will refer to FG as the theory most recently presented inDik (1997), and to FDG as the extended model for FG proposed in Hengeveld(2004). In the latter, a distinction is made between the pragmatic (Interper-sonal) and semantic (Representational) levels. As a result, the elements whichtraditionally belong to a specific term are assigned to their own levels. Below,we will use a term structure which contains all and only the relevant semanticinformation. Pragmatic information, such as topicality and focality, is derived

315

from the interpersonal level. Syntactic functions are assigned by the expres-sion component.

3. Throughout this article, the following abbreviations are used: 3 = third person,ACC = accusative case, ART = article, COND = conditional, DAT = dative case,DEF = definite, F = feminine (gender), FUT = future tense, GND = gender, INF =infinitive, M = masculine (gender), N = neuter (gender), NOM = nominativecase, NP = noun phrase, PPT = past participle, PL = plural, SG = singular, UR =underlying representation.

4. In the formalizations below, upper case notions indicate feature names(NUMBER) and lower case notions indicate feature values (sg). Symbols pre-ceded by # represent a link to an entry in the lexicon (#girl). Forms in quotesindicate phonological representations in the object language (there may be ambiguity among feature names, they are made more specific byway of an extension: e.g. NUMBER.subject means the Number value of theSubject term. All these notations are relatively informal, and meant only to il-lustrate the points under discussion. A formally more rigorous version, basedon feature-value logic and unification, was introduced in Bakker (1989). Acomputer implementation is discussed in Bakker (1994).

5. Obviously, this is too simplistic. Verbal predicates may directly assign specificcases to their arguments which deviate from the standard marking on the basicsemantic functions.

6. For a discussion of the contents of the lexicon in relation to dynamic expres-sion see Bakker and Siewierska (2002). In contrast to the standard view in FG,it is assumed that grammatical elements such as auxiliaries and adpositions arealso located in the lexicon, albeit with a grammatical rather than a lexicalmeaning. They are not accessible during the construction of underlying repre-sentations, only during expression.

7. In terms of standard FG, this implements the procedure of lexical priority.8. In another study on German speech errors, Berg (1987) presents similar NP-

internal agreement errors. In his corpus, too, most of the errors are due to agender mismatch. Two comparable slips from Spanish (Berg 1987: 287) andFrench (Rossi and Peter Defare 1995: 28), respectively, are given in (i) and(ii):(i)the.M.PL jacket(F)-PL, the.M.PL vest(M)-PL

(ii)that.is the.M same line(F), the.M same book(M) to the.F

line(F) exact

316

9. Exchanges of articles also occur, as is exemplified by the error in (i) where,again, we find a gender clash in two positions:(i)now have I myself the.M.ACC hand(F)

in.front.of the.F.ACC mouth(M),

er, the.F.ACC hand(F) in.front.of the.M.ACC mouth(M) PPT-hold-PPThave put the hand in front of my mouth myself.

10. In the examples below, we use the following notational conventions: the utter-ance containing the error is always given first followed by the intended utter-ance on the right hand side of the arrow. Whenever there is no arrow in an ex-ample, the error was self-corrected by the speaker. The error elements, i.e. theloci of gender mismatch, are in bold type. For blends, even if the error wasself-corrected, the two competing noun phrases are given in the form NP //NP. Except for blends, a translation is given only for the intended utterance.

11. Obviously, there is a certain chance that the wrong choice of noun would notlead to a gender conflict in such cases. Given the semantic compatibility of thetwo nouns, such errors would become apparent via self-corrections only. Sincethey do not lead to speech errors in the grammatical sense, speakers may leavethem uncorrected much more often than the cases that do lead to errors. In thespeech error database, such errors are not available, so we can only speculateabout the total number of them.

12. Possibly, (21) should be added to (20) and (22). However, in that case itwould not be the copy of a concrete form, since in this case we have a mergedform of the article with a preposition: . This suggests that an-ticipations are probably based on the wrong transfer of features rather thanforms. We will come back to this issue in the next section.

13. Of course, we may see these errors simply as a derHowever, one would then expect a random choice for any of the two otheravailable genders, or consistently a choice for some default gender. In all ourexamples, however, it is the gender of the noun which heads the NP in theimmediate context.

14. Note that (28) is particularly interesting since it also contains an erroneousself-correction, i.e. first the article is adapted to Nx and then finally the in-tended Nt1 is uttered.

15. Possibly, speakers make an internal shift to hearer mode here, i.e. they retrievea lexical item on the basis of incoming phonological material rather than re-trieving the phonological material belonging to a specific preselected lexicalentry.

16. The interesting fact that in postverbal situations number agreement is neutral-ized does not play a role in our analysis.

317

17. A hypothetical example could look like the one given in (i) where only thespecification for (in)definiteness is exchanged:

(i)I give-1.SG INDEF.SG.N.DAT pig(N) DEF.SG.F.ACC cherry(F)

I give-1.SG DEF.SG.N.DAT pig(N) INDEF.SG.F.ACC cherry(F)

18. But see Bakker and Siewierska (2004), who argue for a dynamic interpretationof the pragmatic and semantic modules of FG as well.

19. Another explanation may be that the speaker is not fully confident about theformal aspects of the respective paradigms. Unlike the gender system, caseseems to be on its way out in at least some dialects of German (Hans den Be-sten, p.c.).

Anwar, Mohamed S.1979 Remarks on a collection of speech errors.

6: 59 72.Bakker, Dik1989 A formalism for FG expression rules. In: John H. Connolly and

Simon C. Dik (eds.), , 4564. Dordrecht: Foris.

1994. Amsterdam: IFOTT.

1999 FG Expression Rules: from templates to constituent structure.67.

2001 The FG expression rules: a dynamic model.42: 15 54.

2005 Expression and agreement: Some more arguments for the dynamicexpression model. In: Casper de Groot and Kees Hengeveld (eds.),

, 1 40. (Func-tional Grammar Series 27.) Berlin/New York: Mouton de Gruyter.

Bakker, Dik and Anna Siewierska2002 Adpositions, the lexicon and expression rules. In: Ricardo Mairal

Us Mar z Quintero (eds.),, 125 178. (Functional

Grammar Series 25.) Berlin/New York: Mouton de Gruyter.2004 Towards a speaker model of Functional Grammar. In: J. Lachlan

Mackenzie and Mar eles G Gonz (eds.), 325364.

318

Berg, Thomas1987 The case against accommodation: Evidence from German speech

error data. 26: 277 299.Dell, Gary S. and Peter A. Reich1981 Stages in sentence production: An analysis of speech error data.

20: 611 629.Dik, Simon C.1997 (Functional Grammar Series

20.) Second, revised edition. Edited by Kees Hengeveld. Berlin/NewYork: Mouton de Gruyter.

Fromkin, Victoria A.1968 Speculations on performance models. 4: 47

68.1971 The non-anomalous nature of anomalous utterances. 47:

27 52.Garcthis vol. Functional Discourse Grammar and extraction from (complex) noun

phrases.Garrett, Merrill F.1980 Levels of processing in sentence production. In: Brian Butterworth

(ed.), , 177 220. London: AcademicPress.

Haegeman, Liliane1994 . Oxford: Blackwell.

Halle, Morris and Alec Marantz1993 Distributed Morphology and the pieces of inflection. In: Kenneth

Hale and Samuel J. Keyser (eds.),, 111 176.

Cambridge, M.A.: The MIT Press.Harley, Heidi and Rolf Noyer2003 Distributed Morphology. In: Lisa Cheng and Rynt Sybesma (eds.),

, 463 496. Berlin/New York: Mouton de Gruyter.Harley, Trevor A.1984 A critique of top-down independent levels models of speech produc-

tion: Evidence from non-plan-internal speech errors.8: 191 219.

Hengeveld, Kees2004 The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzieand Mar -Gonz 1 21.this vol. Prototypical and non-prototypical noun phrases in FDG.

319

Hieda, Osamu1998 A sketch of Koegu grammar. In: Gert Jan Dimmendaal and Marco

Last (eds.), , 345 373. K i-ger K

Levelt, Willem J.M.1983 Monitoring and self repair in speech. 14: 41 104.1989 . Cambridge, M.A.: The

MIT Press.Mackenzie, J. Lachlan and Mar -Gonz ez (eds.)2004 . (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.Matthews, Robert J.1991 Psychological reality of grammars. In: Asa Kasher (ed.),

, 182 199. Oxford: Blackwell.Meringer, Rudolf and Karl Mayer1895 .

Stuttgart: G (1978),2 . Amsterdam: Benjamins].

Mohammad, Mohammad A.2000

. Amsterdam: John Benjamins.Nichols, Johanna1992 . Chicago: The University of

Chicago Press.Nooteboom, Sieb G.1969 The tongue slips into patterns. In: A.G. Sciarone, A.J. van Essen, and

A.A. van Raad (eds.),, 114 132. The Hague: Mouton.

Pfau, Roland2000 Features and categories in language production Ph.D. diss., Johann

Wolfgang Goethe-University, Frankfurt am Main.2003 Defective feature copy and anti-agreement in language production.

53, 95 108 (Special Issue

; available at: http://uts.cc.utexas.edu/~tls/2001tls/Pfau.pdf).

2007 Cheap repairs: A Distributed Morphology toolkit for sentence con-struction. In: Carson T. Sch and Victor S. Ferreira (eds.),

(53), 9-33. Cambridge, M.A.: MITWPL.

320

Pfau, Roland and Dik Bakker2004 The implementation of agreement in the FDG model: Evidence from

anti-agreement phenomena. Paper presented at the. University of Oviedo, Gij

22 25 September, 2004.Rijkhoff, Jan2002 . (Oxford Studies in Typology and Theoretical

Linguistics.) Oxford: Oxford University Press.Rossi, Mario and Evelyne Peter Defare1995 Lapsus linguae: Word errors or phonological errors?

11: 5 38.Van Valin, Robert and Randy LaPolla1997 . Cambridge: Cambridge

University Press.

1

Within formalist models of grammar, much research has been devoted tothe conditions under which constituents can be extracted out of their natu-ral syntactic domains. One of the specific restrictions widely cited in theliterature is known as the Complex Noun Phrase Constraint (CNPC), whichforbids displacement of constituents out of clauses if they are dominatedby a Noun Phrase (NP). As noted by Van Valin (1996), extraction con-straints2 have been formulated in terms of syntactic movement and, there-fore, they are taken as evidence in favour of an autonomous syntactic mod-ule whose rules and principles do not derive from discourse-pragmaticfactors. Together with the fact that most work on NPs in both FunctionalGrammar (FG) and Functional Discourse Grammar (FDG) has so far cen-tred on the simple NP,3 the CNPC and related principles provide an inter-esting challenge to the theory from both a theoretical and an empirical per-spective.In this article, I will take the opposite route and argue that the extraction

of a syntactic unit out of a complex NP causes a conflict in the pragmaticorganization of the utterance and thus gives rise to anomalous expressionswith different degrees of acceptability across languages.4 In so doing, I willexamine the compatibility of FDG with Lambrecht n-formation structure, which views the informational organization of utter-ances in direct relation to the speech participants gmatic knowledge.This article is organized as follows. In section 2 I will introduce the

main aspects of Lambrecht ation structure and itspotential compatibility with FDG. Section 3 will be devoted to presentingthe original formulation of the CNPC in classical generative grammar. Iwill also discuss data from six languages which indicate that the CNPC andrelated constraints cannot be explained on the basis of configurational re-strictions alone. In section 4 I will show how a careful study of the infor-mational organization of complex noun phrases contributes to a better un-

322

derstanding of extraction restrictions. Finally, I will show how the previousobservations can be accommodated in FDG.

There is a long tradition in the functionalist literature of the study of howpragmatic factors influence the syntactic structure of linguistic expressions.FG has mostly concentrated on the formal impact of the status of referentson linguistic expressions through the assignment of the pragmatic functionsTopic and Focus. Dik (1997a: 310) defines pragmatic functions as c-tions which specify the informational status of the constituents in relationto the wider communicative setting in which they are used r-izes the notioninteraction based on the speaker imates of eachother m-pact on the pragmatically relevant structuring of linguistic expressions(1997a: 11), the truth is that, as noted by Siewierska (1991: 148), prag-matic functions in FG are only recognized if they have a formal impact onthe linguistic expression. This justifies Cornishthat the form-oriented aspect of pragmatic functionsover the discourse-cognitive dimension a: 311) iswell aware of the relevance of the notions given and new information andtopicality and focality in his model of verbal interaction, but the relationbetween these notions and the intraclausal pragmatic functions Topic andFocus remains unclear. This problem has long been noted in the FG tradi-tion (see e.g. Siewierska 1991: 155ff.; Mackenzie and Keizer 1991; Han-nay 1991; Bolkestein 1998; Cornish 2004) and different solutions havebeen suggested to bridge the gap between grammar and discourse in FG.However, possibly due to the priority granted to the theoretical problemitself, to my knowledge no studies of specific linguistic processes havebeen offered in the FG tradition in which formal marking is linked to thecognitive status of referents. This is unfortunate, for if Topic and Focusassignment relates to the informational status of referents with respect tothe pragmatic knowledge of both speaker and addressee in a given commu-nicative setting, constraints on pragmatic function assignment can only bediscovered if the role of the cognitive status of referents is taken into ac-count in grammatical description.One of the most important differences between FDG and standard FG,

as presented in Dik (1997a, b), is the top-down orientation of the former,

323

which now intends to construct its internal architecture in such a way thatit becomes compatible with well-established findings in the field of psy-cholinguistics. In particular, Levelt (1989) views linguistic behaviour as aprocess running from communicative intention to actual speech generation.In accordance with this top-down organization, FDG takes the

rather than the clause as the basic unit of linguistic description. Dis-course Acts can be decomposed into Subacts of Reference and Ascriptionto which, following the practice in classical FG, the pragmatic functionsTopic and Focus are assigned.Another important novelty in FDG is the introduction of both a cogni-

tive and a contextual (also named a-cent modules to the linguistic generator. Hengeveld (2004a: 3) character-izes them as follows:

The cognitive component represents the (long-term) knowledge of thespeaker, such as his communicative competence, his knowledge of theworld, and his linguistic competence. (...) The communicative componentrepresents the (short-term) linguistic information derivable from the preced-ing discourse and the non-linguistic, perceptual information derivable fromthe speech situation.

Even though these two components are not part of the grammatical systemproper, they are introduced into the theory in the belief that some gram-matical processes may be best understood if reference is made to them. Inmy view, the cognitive and contextual components can be useful in under-standing the grammatical coding of referents in discourse and in discover-ing significant correlations between the syntactic expression of Topic andFocus and the status of referents in the pragmatic knowledge of bothspeaker and addressee.5

An articulated pragmatic theory along those lines is presented in Lam-brecht (1994), who argues that the information structure of an utterancedepends on the mental states of the interlocutors and thus is not a constantfunction associated with a sentence but, rather, a function of the communi-cative situation in which the expression is uttered. Two information struc-ture categories that will be relevant in this article are Identifiability andActivation, which are characterized as follows (Lambrecht 1994: 76):

IDENTIFIABILITY, which has to do with a speakera discourse representation of a particular referent is already stored in thehearer ACTIVATION, which has to do withthe speaker

324

referent as already nactive in themind of the hearer at the time of the speech act.

The cognitive distinction between identifiable and unidentifiable referentsis coded in many languages through the formal distinction between definiteand indefinite noun phrases, although other means such as word order andcase markers are also relevant. Unidentifiable or brand-new referents canbe of two types: anchored or unanchored, a distinction taken from Prince(1981), who claims that an entity is anchored if the NP representing it islinked to another referent contained in it. The crucial thing here is thatanchored entities should contain at least one anchor that it is not itselfbrand-new or it wouldn 6 Prince (1981:236) offers the following examples to illustrate the contrast:

(1) a.b.

(1b) is anomalous because the anchoring entity is marked as indefinite and,therefore, assumed not to be known to the addressee. Note, additionally,that anchoring constructions may allow definite descriptions of entitieswithout any prior mention of them in discourse (cf.

work(s) with ding to Hawkins (1991:411fn), an anchoring construction makes a context available in which thereferent can be interpreted as unique by the hearer, and, therefore, the useof a definite article in the anchored NP is acceptable.Lambrecht (1994: 95) notes that activation has correlates in grammar:

phonological attenuation and zero coding in morphosyntax are propertiesof activated referents or presuppositional structure, whereas prosodicprominence and full lexical coding typically entail the inactive status of agiven referent (see also Chafe 1994).7 Accessible or semi-active referents(those which are peripheral, but not directly focused upon) can be of threetypes: deactivated or ccessibleschema or ally accessible -externally accessible

ation between the areas of identifi-ability and cognitive activation is given in the following table (Lambrecht1994: 109):

325

Identifiability and Activation according to Lambrecht (1994)

IDENTIFIABILITYUnidentifiable

AnchoredUnanchored

IdentifiableACTIVATION

InactiveAccessible

TextuallySituationallyInferentially

Active

What is the role of the topic and focus functions in this model? Accordingto Lambrecht (1994: 115 116) the activation states of referents should beseen as preconditions for topic and focus function. Thus, the same activa-tion state may be reflected in topic or focal status. This is illustrated withthe following contrast in Italian and French (emphasis in original):

(2) a. IO PAGO / MOI je PAYE.I PAY.

b. Pago IO / C MOI qui paye.I

The pronominal expressions in (2) have the same activation state in bothexamples, but receive focus function only in (2b). This expression is likelyto be uttered in a context in which it is presupposed that someone has topay something and two or more people intend to do so. Therefore, the sen-tences differ only in the pragmatic function of the pronominal expressionsbut their referents are necessarily active since they are speech participantsthemselves. Topic and focus thus relate to the status of a referent with re-spect to a given proposition, whereas identifiability and activation relate tothe mental representation of referents. Obviously, there are important rela-tions among both parameters to the extent that, for instance, a referent isinterpretable as the topic of a sentence only if it has certain activationproperties or a significant degree of pragmatic accessibility. In Lam-brechttopic of the expression. However, before making this communicative deci-

326

sion he also evaluates the status of that referent in his interlocutorand decides upon the formal properties of the sentence. Activation proper-ties act, therefore, as a precondition for topic status and, consequently,(Lambrecht 1994: 165) iently accessible topic ref-erents must pose certain difficulties of interpretation, hence will tend to beperceived as ill formed llowing scale:

Lambrecht

TOPIC ACCEPTABILITY SCALE

Active Most acceptableAccessibleUnusedBrand-new anchoredBrand-new unanchored Least acceptable

Lambrecht (1994: 262) notes the complementarity of the topic accessibilityscale with focal prominence: those referents which are low on the scale arelikely to be coded as non-topics, i.e. focal constituents. Hence, the scalepredicts that the more inactive a given referent is, the more likely it is tofunction as a focal constituent.Activation and Identifiability are mental notions which are undoubtedly

dependent upon properties of the human mind. Different authors havestressed the relevance of cognitive processing in the efficient identificationof referents. Deane (1992) claims that topicality and focality are discourseconcepts which can be defined in cognitive terms as both can be subsumedunder the more general concept of (see also Chafe 1994, whointroduces the term ); according to Deane (1992: 30), a-pacity to attract attention is a logical prerequisite of topic status, since itwould be difficult to talk about something to which the audience could notbe expected to attend nformation, by virtue of being new ormore important, is likely to attract attention immediately. New topics andfocal (i.e. new) information are introduced on-line according to currentcommunicative needs and decisions. Consequently, active topics by defini-tion require little attention and hence little processing effort. New topics(typically introduced via special constructions) should require more atten-tion since they demand a greater effort on the addresseelinguistic expression with the right referent.

327

Given that the human mind has limitations as to the amount of informa-tion that can be processed at a given time, we expect to find relevant re-strictions on the attention load that linguistic expressions can demand. Onesuch limitation has been formulated in various ways in the literature:Giv -chunk-per-clause

or Lambrecht), all of which roughly amount to claiming that each

unit of discourse or proposition cannot contain more than one chunk ofnew information. In terms of the Activation notion introduced above, thisamounts to saying that there can only be oneactivation applies to a single referent, event or state, but not to more thanoneout to be crucial in a pragmatic account of the CNPC; it is to the CNPCthat I now turn.

3.1. The CNPC in classical Generative Grammar: A brief history

In the early days of Generative Grammar, restrictions on the application ofmovement operations were built into the structure of the rules themselves,which led to the postulation of rules of enormous complexity and obscuredthe fact that different constructions were subject to similar constraints. Inhis dissertation, observedthat certain structural configurations function as slands blocking theoperation of syntactic rules, since their constituents cannot be extractedand moved to other syntactic positions. One of those syntactic environ-ments immune to extractability is the so-calledwhich Ross defined as an NP dominating a clausal unit. The behaviour ofthis environment as a syntactic island can be observed in the followingexamples (Ross [1967] (1986): 75):

(3) a.b.

(4) ab. *

328

As noted by Ross, these sentences differ only in that the sentential objectof is embedded in an NP headed by in (4a), but not in (3a).This difference must then be connected to the fact that the complement ofthe verb can be extracted in (3b) but not in the corresponding (4b). Asimilar situation obtains if the dominated clausal constituent is not a com-plement but a relative clause, as in example (6a). As relative clauses arenecessarily embedded within NPs, extraction appears to be blocked, asshown in (6b) (Ross 1986: 71):

(5) a.b.

(6) a.b.

But it is not only relativization that is impossible out of a complex NP.Questioning elements in this syntactic context also yield ungrammaticalsentences, as in (8b):

(7) a.b.

(8) a.b. *

Again, a similar paradigm is obtained if we try to question an elementwithin a relative clause modifying a lexical noun:

(9) ab.

(10) a.b. *

These facts led Ross to propose the following principle (1986: 76), whichhe named

329

No element contained in a sentence dominated by a noun phrase with a lexi-cal head noun may be moved out of that noun phrase by a transformation.

In a number of subsequent publications, Chomsky ([1973] 1977a, 1977b)pursued this line of inquiry. He suggested that the CNPC be derived from amore general principle of , a locality constraint, which had theeffect of banning all long distance movements. Roughly, the concept ofSubjacency prohibited movement out of a constituent if it meant crossingtwo bounding (cyclic) categories. The number and syntactic category of thebounding nodes seem to be subject to language variation, but in the case ofEnglish, the data suggest that the NP and the Sentence (S) impede syntacticmovement.Chomsky (1977b) examines further properties of the rule of wh-

movement and assumes that the process of topicalization presents essen-tially the same syntactic features. Thus, it is not possible to topicalize aconstituent if that entails a violation of Subjacency, as can be observed in(11):

(11) a. *

This also explains why it is not possible to have both processes at the sametime: both extracted constituents must end up in the same syntactic posi-tion, resulting in ungrammatical examples such as those in (12b and c):

(12) a.b.c.

One of the consequences of identifying NPs as bounding nodes is thatleftward wh-movement out of NP is banned, given the fact that themoved constituent has to cross two bounding nodes (S and NP). Thisseems to be correct for a great number of examples, such as the ungram-matical extractions in (13):8

330

(13) a.b. *c. *

However, there are cases which may seem rather problematic. This is thecase of so called cture

(14) a. .b.

In order to explain the grammaticality of (14b), Chomsky suggests a re-analysis process in which the PP is extraposed out of the NP, thusallowing extraction without violating Subjacency (see Bach and Horn 1976for the underlying reasoning). However, as Chomsky (1977b: 114) himselfacknowledges, the conditions on the application of the rule are rather ob-scure. For example, it does not seem to apply to the verb , witnessthe ungrammaticality of (15b), where the bracketing indicates the resultingstructure after the application of the reanalysis rule:9

(15) a. .b. *

A second source of problems is exemplified with the ungrammatical con-struction (16b):

(16) a. .b. *

(Chomsky 1977b: 115; ex. 206)

(16b) is ungrammatical, in spite of the fact that, after the application of thereanalysis rule, the moved element only crosses one bounding node.Clearly, an additional explanation is needed to account for these cases;Chomsky proposes the , which prohibitsmovement out of structures with an overt constituent in subject position(e.g. in (16b)).To summarize so far, we have seen that it is not possible to extract an

element from a complex NP, and that the structural machinery introducedin Generative Grammar to account for this fact, the Subjacency conditionin particular, leads to banning movement out of any NP. There are two

331

exceptions to this generalization:Consequently, two solutions have been proposed to account for each ofthem: the reanalysis of nominal PPs as verbal PPs and the Specified Sub-ject Condition.10

In the following section I will present data from a number of languagesto defend the view that island constraints cannot be fully accounted for onthe basis of configurational restrictions only. This will lead me to proposean alternative analysis in section 4.

3.2. The CNPC in six languages

In the generativist tradition long distance displacement is generally formu-lated in terms of syntactic movement. However, a brief look at (part of) thetypological data available seems to suggest that wh-movement and subja-cency effects do not necessarily go hand in hand. The existence of islandviolations in several languages is well-known in the generativist traditionand different authors, both formalists and functionalists, have remarked onthe importance of pragmatic factors in arriving at a full understanding ofthis phenomenon. In what follows I will present some relevant data from anumber of languages as well as the most interesting aspects of the theoreti-cal analyses proposed to explain them.

Japanese does not show visible wh-displacement in relative clause forma-tion and topicalization. Although Generativists claim that this languageobeys subjacency in non-visible movements at the level of Logical Form,Haig (1996) maintains that this position is simply incorrect. The authorexamines a wide range of extraction processes and concludes that subja-cency is for the most part inoperative in Japanese. However, he also admitsthe existence of a number of examples which prove that subjacency effectsare visible in Japanese to some extent. In a proposal reminiscent of Kuno(1987) analysis, Haig argues that relative formation and topicalization areexamples of topic/focus-comment structures which are subject to an

equires the comment portion of the sentenceto be about the focus or topic equent variation in grammaticalityjudgements is then attributed to speakers

332

comment portion of a sentence as appropriate in a given context. Considerthe following example (1996: 84):

(17) e1 e2taking-care-of was person NOM died dog NOM

every-night station to master ACC greet to came2 that the person1 who e1 was keeping e2 died came to the

station every evening to greet his master.

According to Haig, (17) is an acceptable to his informants, even thoughstructurally similar expressions are considered ungrammatical. Haig be-lieves that the acceptability of the example derives from the fact that thecontext (a popular Japanese story of a faithful dog) makes it clear that thewhole sentence should be interpreted as being about the dog. In a similarvein, Shimojo (2002) elaborates upon Haig g-nostic test for the aboutness condition. He furthermore argues that theaboutness condition is supported on cognitive grounds.

Van Valin s-tion formation in Lakhota, a Siouan language which does not displace wh-elements in interrogative constructions. He argues that if movement is anessential factor in the account of subjacency phenomena, languages inwhich wh-elements remain should not show these effects. However,Lakhota, and reportedly other languages as well, do show the relevant ef-fects, from which it follows that movement is not the mechanism at issue inthe explanation of subjacency (see Dik 1997b: 362 for the same reason-ing).11 Consider the following examples (Van Valin 1996: 36-37):

(18) a.Man the [NP [S dog a cat many bite ]

the this] saw

333

b.Man the[NP [S dog a *what/something bite ]

the this] saw Q

According to Van Valin, replacing the object within the relative clause in(18a) (i.e. ) with the interrogative element yields an accept-able sentence, but one which can only be interpreted as a yes-no question.Hence, this is an example of a language in which subjacency effects are notlinked to overt movement. The author proposes the following condition onextraction phenomena:

the displaced element mustbe linked to an argument position in the semantic representation of a clausewithin the Potential Focus Domain (PFD) of the Illocutionary Force (IF) op-erator.

As he acknowledges, for this account to be satisfactory it must be possibleto predict where the PFD lies in a sentence. He proposes a general struc-tural principle which governs the PFD in complex sentences:

A subordinate clause may be within the PFD if and only if it is a directdaughter of (a direct daughter of, and so on iteratively) the clause nodewhich is modified by the IF operator.

Consequently, this proposal relies on a structural definition of the PFD,which is in accordance with Van Valinextraction phenomena should involve a careful examination of the interac-tion between structure and function.

Just like Lakhota and Japanese, Korean is an language, but it doesnot seem to obey subjacency, with the exception of the wh-island con-straint. The author mentions the following contrast between English andKorean as regards the CNPC (Hong 2003: 102):

334

(19) a. ?b.Mary-NOM what-ACC buy-REL person-ACC meet-Q

As Hong operates within a generativist framework his proposal to accountfor the different behaviour between both languages as regards the CNPC isnot based on extragrammatical factors. He argues that the explanationshould rely on the different nature of English wh-elements and its Koreancounterparts, which require different processes to be correctly interpreted.

According to these authors, Danish generally obeys the CNPC, but theynote a number of counterexamples in extraction processes of relative for-mation which they attempt to explain in terms of their theory of i-nance -Shir and Lappin believe that extraction isonly possible out of those relative clauses that can be interpreted as beingdominant. They define the notion of dominance as follows:

A constituent c of a sentence S is dominant in S if and only ifthe speaker intends to direct the attention of his hearers to the intension of c,by uttering S.

Relative clauses, they claim, are generally used non-dominantly, the domi-nant part of the sentence being in the matrix clause, which accounts for theimpossibility of extracting linguistic material out of them.12 However, thefollowing sentences are grammatical precisely because the relative sen-tences in them are used dominantly (1979: 55):13

(20) a.That are there many who can like

ke that.

b.That know I many who can like

The crucial question, then, is how to establish the circumstances underwhich relative clauses can be used dominantly. The authors believe that

335

this is connected to the low information status of the matrix clause; in theirown words, when the matrix functions as anrest of the sentence is used dominantly. For example, if the matrix containsan existential operator or if the main verb is semantically neutral, the rela-tive clause becomes dominant and extraction of constituents may be possi-ble. The following examples, in which the matrix predicate sequentiallyincreases its complexity, illustrate the relevance of this factor:

(21) a. ?That have I asked many who have done

b. *That have I made-fun-of many that have done

made fun of many that have done that.

c. *That house know I a man who has bought

A similar paradigm is obtained in the following English examples (1979:58):

(22) a.b. ?c. *

As will be shown in section 4.1.2., the nature of the matrix predicate willturn out to be a relevant factor in accounting for the informational proper-ties of extraction processes.

D ctionfrom complex noun phrases is not allowed. He offers the following exam-ples:

336

(23) a.To-me bothers thefact of that not have.3PL built

a tube in this cityhave not made a tube in this city.

b.This is the city in thewhich to-me bothers

thefact of that not have.3PL built a tube? y in which it bothers me the fact that they havenot made a tube.

c.This is the city in thewhich to-me bothers

that not have.3PL built a tubewhich it bothers me that they have not made a

tube.

In compliance with the CNPC, the ungrammaticality of the examples in(23) must be attributed to the presence of the NP ( ), whichintroduces a bounding node for movement. However, the author himselfstates that in some varieties of Spanish, expressions which violate subja-cency seem acceptable to most speakers:

(24) a.Not can.1SG not even conceive the idea

of that leave.3PL that that city REFL burns

b.This city which not can.1SG not even

conceive the idea of that leave.3PLthat REFL burns

is more beautiful than Florence

would let it burn, is more beautiful than Florence.

337

And there are even examples in which the extracted element crosses two Snodes and one NP, resulting in a grammatical sentence. This is somethingwhich, as Dexplain:

(25) a.Not know.1SG to whomREFL him occurred

the very bad idea of reconstructing this theatre

with wood oldructing this

theatre with old wood.

b.This theatre which not know.1SG to whomREFL him

occurred the very bad idea of reconstructing

with wood old is to point of fall.REFL

reconstructing with old wood, is about to collapse.

He attributes the difference in grammaticality to the fact that in(23) is a factitive verb, whereas in (24) and in (25) arenot. He furthermore argues that expressions of the type

, etc. should be taken as being complex verbssyntactically.14

Even in English, Deane (1992: 9) notes that there are cases of what hecalls ement is extracted across two, threeand even four levels of embedding within a NP. According to the author,the following example, which is to be understood as uttered in a sarcastictone of voice, illustrates the case:

(26)

338

In his study on the CNPC in Swedish, Allwood (1976) notes that the prin-ciple is systematically violated in that language. He offers a battery of ex-amples in Swedish containing extractions from complex NPs: Topicaliza-tions, Question formation and Relative clause formation. The followingexamples illustrate extraction of the object within an embedded intransitiverelative clause (1976: 4):

(27) a.I know a man who is strong

ow a man who is strong.

b.A man know I who is strong

c.Who know I who is strong?

d.A man who I know who is strong

Although the acceptability of these and other extractions of different typesof constituents out of complex NPs varies among speakers, Allwood be-lieves that kers for us to beable to claim that, for example, the constraints formulated by Ross andHorn against movement from NPs cannot hold msthat the differences and regularities in acceptability judgements must takesemantic and pragmatic factors into account. As far as semantic factors areconcerned, Allwood notes the relevance of the definiteness of the extractedNP (demonstratives in particular contribute to making the extraction moreacceptable), and the choice of verb in the matrix clause, which also con-tributes to acceptability if it focuses on the state of affairs as a whole.Moreover, since expressions containing a topicalized or focalized unit re-quire a context to be interpreted as natural utterances, they can only beaccepted in isolation if speakers can imagine a reasonable context in whichthey might be used (cf. Kuno 1987; Shimojo 2002). If that is not the case,

339

acceptability is reduced. Anticipating the aboutness condition and similarobservations, Allwood (1976: 17) states thatexcept for the topicalized constituent must be interpretable as a reasonablepredication (comment) of the referent of the topicalized constituent.To sum up, this short survey of the cross-linguistic manifestation of the

CNPC shows that the degree to which this structural constraint is reflectedin the grammar of languages is anything but uniform and appears to beinfluenced by a number of intervening factors of different types. Since theextraction processes considered revolve around the notions of topicaliza-tion and focusing, most functional analyses, in accounting for the CNPCeffects, have centred on aspects of the organization of information withinexpressions, but the lack of uniformity in the characterization of pragmaticfunctions in the functionalist literature has somehow undermined theirpotential as explanatory notions for syntactic constraints. In the followingsection I will defend the view that the CNPC effects emerge from a conflictbetween the activation states of the referents of the NPs involved in themind of the speaker and the communicative function of the extractionprocesses (topicalization and focusing). It is hoped that this will show thata full understanding of the phenomenon requires the careful integration ofinsights stemming from both syntactic and pragmatic theory.

The preceding discussion has made clear that there are more factors in-volved than just configurational restrictions in an account of the CNPC.15

In particular, the following features have been identified by several au-thors: (in)definiteness, the informational load of the matrix predicate, thecommunicative setting and topical or focal attention, and the pragmaticstructuring of the expression. The number of factors involved is obviouslyan added complication in providing a full account of the phenomenon.Moreover, the relative weight of each factor is different according to thetype of construction involved and possibly also the particulars of the com-municative situation. Consequently, my aim in this section is not to providea full explanation of the problem, but rather to suggest a general line ofinquiry on the basis of Lambrecht ation structureas laid out in section 2 above.

340

4.1. Towards a pragmatic account of the CNPC

The CNPC relies on a definition of the complex NP as a syntactic unitheaded by a noun which dominates a clausal constituent. The clausal con-stituent may be either a complement to the head noun or a relative clause.The nouns that can take clausal complements in English are mostly lexi-cally derived from verbal predicates ( , etc.) aswell as a few other items such as or . As pointed out tome by Jan Rijkhoff (pc), nouns like and are special in that theyintroduce a referent which can be definite on first mention. Hawkins (1991:432) argues that NPs headed by nouns like make an e-ness claimindefinite article (cf.

). The precise role of the complement is toanchor the specific referent for the noun and, therefore, make it identifiablefor the addressee. Similarly, Rijkhoff (2002: 177), who refers to Lehmann(1984), notes that restrictive relative clauses typically serve a localizingfunction, as when they express the situation in which the head NP partici-pates and therefore localize it in time or cognitive space. This explains whythe typical relative clause modifies a definite matrix NP and, in many lan-guages, this seems to be the only possibility.16

Another important aspect of extraction which is not usually discussed insufficient detail in the relevant literature is the fact that the extracted ele-ment serves different pragmatic functions in spite of the formal similarityof the displacement processes. Therefore, an explanation on the basis ofthe pragmatic status of the displaced element is likely to be different fortopicalization and relativization, which involve topic displacement, asagainst wh-movement, which involves focal displacement.17 Ideally, how-ever, both explanations should rely on similar principles in order to ac-count for the parallel syntactic behaviour of the two processes.We are now in a position to understand some relevant properties of ex-

traction from complex NPs. Let us begin with topicalization processes; Ithink we can safely assume that the role of topicalization and relativizationis to promote to active status a referent which otherwise would present alow activation state. As noted by Chafe (1994: 73), converting an idea fromthe inactive to the active state is a costly mental effort. It follows that ac-tive topics will typically not receive special grammatical marking to signaltheir pragmatic status whereas non-active topics may be more readily sub-ject to special grammatical treatment. At the same time, if a constituent isassigned a special pragmatic status within an utterance, it is to be expected

341

that the extracting domain in the expression should contain presupposedinformation. Let us illustrate the discussion with Ross xamplesin (3) and (4), repeated here for convenience:

(3) ab.

(4) a.b.

The difference caused by the presence of in (4) is that a newreferent is introduced which is anchored by the complement clause. Hence,the complex NP relates to a referent which is brand-new anchored (i.e.unidentifiable), or, in Hawkins logy, inherently unique.However, what is displaced in (4b) is another NP ( ) within the com-plement clause. If the sequence contributesto identifying the referent of it must necessarily be understood asgiven or shared knowledge. Promoting to topic status is thus com-municatively anomalous since it must be assumed to be already active inorder to anchor . The presence of the demonstrative in Rossoriginal examples (3a) and (4a) is significant in this respect, as well as itsabsence in both (3b) and (4b). If the sentences are to be related via a trans-formation, one would not expect them to have different determiners. I as-sume that the demonstrative in the embedded clauses indicates a prominentactivation status of the referent, which is incompatible with the role of arestrictive relative clause as an anchoring unit in both (3b) and (4b). AsHawkins (1991: 414) observes, demonstratives usually requirephysical perception of entities in the world or explicit textual mentioningof entities ivation status of the referent is crucialin the topicalization process.The advantage of using the categories Activation and Identifiability in

explaining these facts is seen by the fact that they do not correlate with thenotions topic and focus in a one-to-one fashion. As mentioned above, ref-erents with similar activation states may be coded as focus or topic, some-times with significant formal differences.The analysis of wh-extraction goes along similar lines. Independently of

the traditional interpretation of the wh-element as the focus of the utter-ance, what is relevant in this context is its activation status. Typically,though not necessarily always, focal elements denote new information and

342

are, therefore, brand-new and unidentifiable (i.e. inactive). Hence, wh-displacement would serve to promote them to active status. The problemrelates to the previous discussion: the constituents present in the comple-ment clause must be regarded as shared knowledge to identify the referentof , etc. Therefore, questioning them makes no sense communica-tively. It would be impossible to identify the referent of the head noun ofthe complex NP if the sequence used for that purpose (i.e. the complementclause) itself contained an unidentified referent. Thus, both processes en-tail a violation of Chafe s-trate an example of expressions in which two referents compete for cogni-tive attention and exceed the capacity of the human processor.In short, I believe we can tentatively conclude that complex NPs consti-

tute a syntactic configuration which by virtue of the nature of the elementsinvolved introduces a conflict in the information structure of the expres-sion, giving rise to a communicatively anomalous utterance. The data fromsix languages examined in section 3 corroborate this observation: lan-guages and speakers differ in the degree of acceptability that can be as-signed to extractions form complex NPs, a fact which is to be expected ifthe felicity of the expression depends on pragmatic and syntactic factorsrather than on syntactic factors alone. In the following section I will try toextend this explanation in order to account for extraction from possessiveNPs.

In section 3.1 I mentioned the problem posed by so-calledin the classical generative grammar account of the CNPC. In short, exam-ples such as (14b) and (16b), repeated here for convenience,

(14b)(16b) *

forced Chomsky to introduce a rather reanalysis rule to account forcases like (14b) and the Subject Specified Condition for those like (16b). Iwill postpone discussion of the former to the following section and concen-trate now on examining the information structure organization of posses-sive NPs in search of an alternative explanation.In Rijkhoff relative clauses

typically serve a localizing function in a definite matrix NP. Thus, they

343

both help the addressee identify the intended referent of the matrix NP.This observation is confirmed by the fact that in many languages there is aformal relation between both categories (Rijkhoff 2002: 194fn.). More-over, Haspelmath (1999: 232) observes that, cross-linguistically, possessedNPs overwhelmingly tend to be definite. In his view, this is connected tothe anchoring function of the possessor.Now, if the possessor noun serves as an anchor, it follows that the ref-

erent of the sequencespeaking. Extraction of the most embedded NP is thus pragmaticallyanomalous: the identifiability of the head noun n-chora sequence denoting an unidentifiable referent. Such a conflict does notarise if the focused constituent is an adjunct to the matrix verb (e.g.

) as the focused constitu-ent is outside the relevant NP.Unfortunately, the problem is a bit more complex. Kuno (1987) notes

that extraction from NPs with a genitive possessor in English may be pos-sible if certain pragmatic conditions are met. Consider the extractions in(28) and (29):18

(28) a.b.

(29) a.b.

c.

Kuno claims that examples (29b) and (29c) are perfectly acceptable, eventhough they entail a violation of the Specified Subject Condition. The au-thor observes a crucial difference between the NPs

in (28a) and in (29a), in thatthe latter, even without any prior context, immediately suggests that thereare other peoplein that the expression , together with ourknowledge of the world, suggests that there may be other TV channelsreportings of the same event. From these facts, Kuno hypothesizes thatextraction from noun phrases with a specified subject may be possible ifthe possessor NP is readily interpretable contrastively (i.e. if it is the focusof the expression).19

344

But the question now is why extraction from possessive noun phrases isonly possible when the possessor noun is in focus. Kuno claims that thismay be connected with a conflict in topic-focus organization. Considerexample (30):

(30) The man bought the woman

According to Kuno, this sentence can be interpreted as being a statementabout and next about , but not about

, who does not qualify as a potential topic. Consequently, it is possibleto form relative clauses out of every NP in the sentence but , asshown in the examples in (31):

(31) a.

b.

c.

d. *

The question is then what makes extraction in (29b) possible. Kuno claimsthat in (29b) does not qualify as a topic, because of the nature of thenoun , which fosters its interpretation as a contrastive focal con-stituent. As a consequence, becomes eligible as the topic and canundergo relative formation. The problem with Kuno ysis, however,lies in its rather obvious circularity. The author claims that only those con-stituents that qualify as potential topics can undergo extraction processes.Yet, he does not offer a definition of topic function and one is left to be-lieve that the possibility of being extracted is a natural condition for topicalstatus.Obviously, in those cases in which the possessor noun is likely to be in-

terpreted as a focal constituent, the information structure of the expressionis significantly different. Kuno (1987: 13) notes that in examples (29b andc) both and receive focus intonation. He thus seems to assumethat prosodic marking indicates contrastive focus. If that is the case,and should be understood as active referents, since they are contrastedwith a shared alternative in the form of a potential set of candidates for thedifferent versions or reportings. Following Lambrecht (1994: 323), I will

345

assume that the role of accentuation on an active referent as in Kunoexamples isargument in a pragmatically structured propositionof the possessor NP in (29b and c) is not that of an anchor to help the ad-dressee identify the referent of the possessed NP as in the non-contrastivereading of genitives, but, rather, to direct the addresseewhat he considers the most relevant piece of information in that particularcommunicative setting. Given the contrastive flavour of the possessor NP,both the possessed noun and its complement should be understood as hav-ing identifiable referents. Hence, extraction of the embedded NP is accept-able as there is no interference with the identifiability of the referent of thepossessed noun.Undoubtedly, the acceptability of examples like (29b and c) very much

depends on the possibility of interpreting the possessor NP contrastively(which in turn depends on the nature of the matrix noun), and on the acti-vation status of the referent of the embedded NP. Presumably, discourse-active referents are not likely candidates for topicalization or focusingstrategies, but it seems sensible to assume that the contrastive reading ofthe possessor NP contributes to analysing the referent of the embedded NPas peripheral in a personextraction. Significantly, the pragmatic marking of an active referent in-volves no in Chafe n-tribute to violating the One New Idea Constraint by exceeding thespeaker e-quency of examples such as those in (29b and c) is related to their anoma-lous pragmatic configuration, which, in Gricean cooperative spirit, forcesspeakers to come up with a context in which they can be interpreted ascontrastive NPs.

The previous section illustrated the fact that the information structure mostreadily associated with complex NPs may be modified through intonationalmeans or on the basis of the inferences drawn from the lexical items em-ployed in a particular utterance. In this section I will show that the choiceof lexical items in the matrix clause can also have important consequencesfor the possibility of extracting constituents out of complex NPs.As mentioned in section 2, several scholars have noted the relevance of

the informational value of the matrix predicate in the grammaticality of

346

extraction processes (Allwood 1976; Grosu 1982). Van Valin (1996) ob-serves that extraction is possible out of neutral verbs of saying, such as ,but the same does not seem to hold in the case of -of-speakingverbs, as shown in (32):

(32) a.b. *

He believes that there is a straightforward explanation for this contrast: theselection of a neutral verb (together with an unmarked intonation pattern)indicates that the focus of the utterance is the content of the message,which is encoded in the verbal complement. However, when a manner-of-speaking verb is selected the focus shifts to the verb itself, because its in-troduction only makes sense if it is relevant to the main point of the ex-pression. Otherwise, Van Valin claims, we would have a violation ofGrice the main verb,extraction out of the verbal complement is blocked.A tentative attempt at classifying verbal predicates in terms of their in-

formational value is found in Chafe (1994: 110 ff.), who introduces a cate-gory of -content which are characterized by the fact that theydo not present an idea of their own, but are subservient to the idea ex-pressed by the object. Verbs in this category include the unaccented lightpredicates , , , etc., as well as a miscellaneous class of itemsexpressing typical ways in which an object is used (e.g. ), intro-ducing whatever is expressed by the object (e.g ) or speci-fying how a referent is perceived (e.g. ). Although Chafe leavesopen the question of the relation between verbal predicates and activationcost, his observation of the interaction of the verb and its object in thepragmatic organization of expressions is deeply connected with the discus-sion on extraction from (1977b: 114) admit-ted, the conditions for the application of his reanalysis rule seemed to de-pend on the nature of the main verb in obscure ways. Kuno (1987)observes the contrasts in (33):

(33) a.b. *

and those in (34), taken from Bach and Horn (1976):

347

(34) a.b. *

One might think that the grammaticality of some extractions is connectedto the fact that the PP can be interpreted as a verbal adjunct. Indeed, thisdoes seem to be the case for the concept . However,this does not explain the acceptability of the expression (34a), since theconcept does not make sense. Kuno (1987: 23) observesthat in the examples in (35):

(35) a.b.

the fact that the book was about John Irving is much more relevant in (35a)than in (35b), because one does not lose books because of their content.Hence, it may be possible to claim that the sequence requiresmuch more attention load than since it involves a rather unex-pected lexical combination in that context. It thus seems plausible to as-sume that the focus of the expression falls naturally on the verbal predicateand, again, extraction of the wh-element would imply the existence of twoentities in the expression competing for focus status and, consequently, aviolation of Chafe nstraint.As expected, if one can invoke a context in which the use of an infor-

mationally rich predicate sounds natural and does not require a focal inter-pretation, extraction may be possible. Consider Kuno xample with theverb :

(36) A:

B:

According to Kuno, the reason why (36B) is a possible sentence derivesfrom the fact that the speaker knows that the destruction of the pictureswas carried out on the basis of who they portrayed. This makes the embed-ded noun a relevant entity to the entire utterance.However, this does not account for all cases, as the same verb may pre-

sent conflicting results:

348

(37) a.b. *c.

The reason may be due to the different information loads of the expressionvs . Intuitively one feels that pictures are there to

be seen whereas books are not there to be seen but to be read (cf. (37c)).Hence in (37b) demands more attention and the focalization ofthe embedded unit results in an anomalous sentence. This means that lexi-cal influence on extraction processes is not to be seen purely as a matter oflexical choice of the matrix predicate, but as a consequence of the natural-ness of the inferences that can be drawn from the verb plus object rela-tion.20 Significantly, Dsame lines in Spanish. He considers the data in (38):

(38) a.have.1SGwritten / bought a book on the chimps

/bought a book on chimps.

b.Thoseare the chimps on the which have.1SG

written a book

c. *?Those are the chimps on the which

have.1SG bought a bookLit.: imps on which I have bought a book.

D can be considered an adjunctof , but not of , for obvious semantic reasons, which ac-counts for the ungrammaticality of (38c). But personally, I see nothingwrong with example (38c) as long as it is uttered in the right context. Thiscontext could be one in which the addressee has been previously informedthat the speaker had bought a book on chimpanzees and at the moment ofspeaking are both walking by the primate enclosure in a zoo. In such asituation, the expression is sharedknowledge, and cannot be interpreted as the focus of the expression. What

349

is more, in this situation even less natural lexical combinations seem to bepossible:

(39)Those are the chimps on the which

have.1SG seen / destroyed / introduced a bookLit.: yed / intro-duced a book.

Admittedly, though, some combinations are downright ungrammatical:

(40) *Those are the chimps on the which

have.1SG loved / adored / hated a bookLit.:book.

It thus seems that verbs of psychological perception do not readily acceptextraction in Spanish (see Lorenzo 1995: 180 for some discussion), whichalso reveals the need for a careful study of the lexical-semantic propertiesof the matrix predicates and their interaction with focus structure if we areto achieve a better understanding of extraction phenomena.21

I believe the conclusion that can be drawn from this section is that amodel of information structure which relies on the cognitive status of ref-erents in discourse may offer interesting insights into the nature of syntac-tic processes and the CNPC in particular. Moreover, not only is syntacticstructure sensitive to pragmatic demands, but lexical choice, too, can beinfluenced by general principles of verbal interaction, which, in turn, mayaffect the possibility of topicalizing or focusing linguistic units. It shouldalso be noted that the syntactic properties of individual languages may alsohave an influence on the acceptability of extractions from complex NPs.Thus, speakers of languages in which topicalizations and focusing struc-tures are frequent will possibly react in different ways to CNPC effectsthan speakers of languages with a more rigid word order. The CNPC ap-pears to be not simply a constraint on syntactic structures; rather it is thecase that complex NPs should be seen as syntactic configurations which, incombination with pragmatic and lexical factors, favour the construction ofcommunicatively anomalous expressions. Therefore, linguistic models

350

must at the very least provide the tools for analysts to make reference tothese different factors and provide integrated accounts of grammatical phe-nomena. In the following section, I will explore the extent to which FDGcan meet this requirement.

In section 2 I claimed that classical FG is to be understood as a grammati-cal component which aims at being compatible with or integrated within amodel of verbal interaction. However, given the fact that the general work-ings of this model have never been explored in detail by FG practitioners,it is no wonder that the search for correlations between syntactic processesand the activation states of referents in discourse has not been a major lineof research in the theory. Unsurprisingly, then, the exclusively syntacticnature of the CNPC is simply taken for granted; Dikof accessibility, defined asof some grammatical operationintrinsic, functional, and hierarchical. Hierarchical constraints involve theposition of a term within the clause structure in which it occurs. As a mat-ter of fact, Dik (1997b: 361) specifically mentions Rossa type of hierarchical constraint on the application of grammatical rulesand therefore accepts the existence of configurational restrictions in extrac-tion processes which, in his view, can be accommodated in the FG layeredstructure of the clause. However, no detailed indications are given as tohow this can be implemented and no attempt is made at relating this re-striction to extragrammatical factors.In his presentation of FDG, Hengeveld (2004b: 370) follows a similar

route in describing FDG as mmatical component of a wider theoryof verbal interaction contex-tual and a conceptual component reflects the recognition of the complexinterplay between syntactic, semantic and discourse-pragmatic factors inthe construction of linguistic expressions. Consequently, as Hengeveldhimself observes, a necessary line of research in the theory relates to work-ing out the internal structure of these two components and their relationwith the grammatical one.22

The introduction of the cognitive and contextual components, togetherwith the top-down orientation of the model, offers a natural environment inwhich to integrate the categories of Activation and Identifiability as pre-sented above. Moreover, it allows us to ground the pragmatic functions

351

Topic and Focus, as reflected in the interpersonal level, in the activationstatus of referents at the moment of speech or, to put it more accurately, thespeakerquote from Lambrecht (1994: 162 163) is illustrative:

In selecting a topic for a sentence, a speaker makes a communicative deci-sion as to the n-tity that she wishes to convey information about. But before making thiscommunicative decision, the speaker must make certain hypotheses concern-ing the status of the referent of the topic in the mind of the addressee at thetime of the utterance. On the basis of these hypotheses, the speaker then de-cides upon the form of the sentence in which the topic is to be coded.

This position has a natural correlation with the general organization ofFDG, which is presented in Figure 1 (Hengeveld 2005: 61; see alsoHengeveld, this volume).Communicative decisions and the relevant conceptualizations are made

in the conceptual component, which serves as a trigger for the grammaticalcomponent to start the linguistic generation process (Hengeveld 2004b:369). As noted by Lambrecht in the quote above, speakersdecisions are made on the basis of an assessment of their addresseepragmatic knowledge (see also Dik 1997a: 11), whether this is part of thespeaker -term knowledge or inferred from the current speechsituation. The next step will be the creation of a communicative intentionand the construction of a conceptual preverbal structure that will serve asinput to the Formulator. The process of Formulation translates this prever-bal conceptual message into a linguistic structure which, if relevant, willencode the cognitive status of referents through the assignment of prag-matic functions at the interpersonal level.We can assume that linguistically relevant aspects of the speech situa-

tion are captured at the contextual level, which serves as a source of infor-mation for the speaker to construct a felicitous utterance. Rijkhoff (thisvolume) divides the contextual component into two parts: the situationalcomponent (E-context), which relates to the extralinguistic setting and isthus potentially infinite, and the discourse or textual component (D-context), which is finite and purely linguistic (i.e. the co-text). Each one ofthese subcomponents defines a context which may be potentially relevantfor the construction of utterances, although, according to Rijkhoff, only thelatter is amenable to linguistic description.

352

General layout of FDG

FramesLexemesPrimaryoperators

Interpersonal Level

Phonological Level

Structural Level

Representational Level

Morpho-syntactic Encoding

Phonological Encoding

ProsodicpatternsMorphemesSecondaryoperators

Articulation

Utterance

Formulation

TemplatesAuxiliariesSecondaryoperators

353

There is a significant difference between the contextual and the concep-tual components, at least in the way they are understood here: the contex-tual component represents an objective reality, which is then subjectivelyinterpreted by the speaker in the conceptual component in order to evaluatereferents as identifiable for his addressee or not. In other words, thespeaker constructs a mental model of the objective situational and linguis-tic context,23 on the basis of which he makes his communicative decisionsand choices. This obviously allows for the possibility of wrongly evaluat-ing active referents as inactive or vice versa, and thus producing pragmati-cally anomalous utterances.Thus, it is possible to adapt Lambrecht ation and

Identifiability to Rijkhoff ontextual component in FDG.This gives us the following possibilities for identifiable referents:24

E-activeE-inactive

Situationally Accessible

D-activeD-inactive

Textually Accessible

Cognitive status of identifiable referents

A referent is E-active if the speaker assumes that it is active in the mind ofthe hearer thanks to evidence from the E-context. Similarly, a referent is D-active if the evidence for the activation is assumed to stem from the D-context. Note that both statuses may coincide in a single referent as incases in which the verbal interaction deals with a referent which is thefocus of the perceptual attention of the speech participants (e.g. two peoplecontemplating a painting at a museum and talking about it). This would bean example of the maximum activation of a referent.Conversely, referents may be E-inactive if the speaker assumes that

there is no evidence in the surrounding context to conclude that the refer-ent is active in the addressee As expected, D-inactive referents arethose which, in the speakerlinguistic material into the interaction. Again, both parameters may coin-cide in a single referent, giving rise to the highest degree of inactivenesspossible. Thus, Rijkhoffto further specify the Activation status of the referent according to the loca-tion of the referent itself and to established finer-grained distinctions inLambrecht ories.

354

Note that this approach shares a number of assumptions with Hannay(1991) modes of message management. Hannay intends to account forword order variation in English declarative sentences through the identifi-cation of a number of planning strategies available to speakers. The selec-tion of one particular strategy may have significant effects on the syntax ofexpressions. For example, his Topic Mode defines an option for thespeaker to single out topical information for special treatment by placing itin clause-initial (P1) position. In some cases, however, speakers need not

rmation status of a given unit, which may be imposedby the linguistic context itself. Thus, as usually assumed, a response to awh-question will typically introduce a focal (active) entity into the dis-course. This would be an example of Hannay eaction Mode. At thesame time, languages may offer specific constructions or syntactic posi-tions to encode a pragmatically salient unit. Hannay esentative Moderefers to the use of special constructions to introduce a new discoursetopic, typically through the presence of a -initial position.In a similar vein, the approach introduced here enables us to view

pragmatic functions as grammatical notions of cognitive origin, whoseassignment in the linguistic generation process is sensitive to the commu-nicative intentions and contextual relations represented in both the concep-tual and contextual components. Moreover, a speaker p-tions (including Hannayrepresentation of the contextual component and, in particular, on the as-sumed cognitive status of referents.Let us see how this model works in actual practice with the analysis of

D

(38) c. *?Those are the chimps on the which

have.1SG bought a bookLit.: are the chimps on which I have bought a book.

As mentioned above, Daccounts for that status on syntactic grounds only. I have claimed, how-ever, that it can be felicitous in a particular communicative setting. Thissetting may be formalized in a very simplified manner as follows:25

355

: Speaker and ad-dressee are visiting a zoo. They walkby the monkey enclosure.

(x1)Property: chimpsStatus: perceptually visible tospeech participants

Discourse (i.e.verbal exchange) does not deal withthe relevant chimpanzees.

Example of communicative setting for (38c).

On the basis of the properties of the communicative setting, the speakerassesses the status of referents in the addresseerelevant assumptions on the addresseefollowing conceptual component (Rijkhoff s C-context) at the time ofspeaking (again, only relevant details are given):

2)on chimps (x3)

1 (in E-context) and x3 refer to the same type ofentities

1 is situationally accessible and D-inactive for Addressee

2 is both E-inactive and D-inactive forAddressee

3 is D-inactive and situationally accessi-ble for Addressee

: promote referent x3 to active status: Inform addressee that x1 and x3 refer to the

same type of entities.

Example of Speaker -context) for (38c).

Of course, figure 4 exemplifies only one of the different potential estima-tions of the addressee r-ent n-active for the addressee. Hence, all in all, it shows a rather low activation

356

status, which makes it a good candidate for a topicalizing or focusing strat-egy in case the speaker decides to activate it.The assumption is that the D-context deals with a different discourse

topic at time t1, which means that the speaker will force a rather abrupttopic-shift in the flow of the verbal interaction. The use of a marked lin-guistic construction is therefore grounded on the communicative need toshift the focus of attention in discourse. This communicative decision willbe transmitted to the interpersonal level, where the referent x1 will becoded as the topic of the sentence and promoted to clause initial positionby the morphosyntactic encoder. Obviously, the felicity of the expressionwill very much depend upon the status of Proposition 1 (P1) in the ad-dressee ddressee may have been informed in the past of P1,but, even though the referent is situationally accessible, the radical shift inthe topic of the discourse may severely slow down the retrieval of thatpiece of information from his mental database. If that is the case, the inac-tive content 3 n-tion and thus the activation of x3 will imply a violation of Chafeconstraint. Thus, body language (e.g. pointing) or expressions likethatspeaker to promote a referent from situationally accessible to E-active inthe hearer ne described so as to reduce theactivation cost in the identification of the referent.

In this article, I have argued that the cognitive status of referents is crucialin accounting for restrictions on pragmatic function assignment to con-stituents within complex NPs. Although much more research is obviouslyneeded to evaluate the extent to which this approach can account for islandeffects in diverse languages, I hope to have shown that a full understandingof the nature of extraction constraints requires the integration of insightsstemming from various linguistic angles. From an FDG viewpoint, extrac-tion constraints point up the need to refine the concepts of Topic and Focusin the theory. Standard FG only assigns pragmatic functions to those con-stituents which have received special syntactic or morphological treatment.Thus, there is no way to predict which constituent can be extracted out of agiven configuration; rather, the theory can only assign Topic or Focusstatus to those constituents which have been extracted. The present articlehas shown that the assignment of Topic and Focus is deeply dependent

357

upon the status of referents in the speakerverbal interaction and that an adequate theory of pragmatic function as-signment can only be proposed if reference is made to them (a similar con-clusion is arrived at in Cornish 2004: 144, who also refers to G -Gonz 2001: 165 for the same point). In that sense, the introduction ofboth a conceptual and a contextual component in FDG is a positive steptowards understanding the relation between grammar and discourse.

1. I am extremely grateful to Jos , Jan Rijkhoff and ananonymous reviewer whose very detailed comments on a first draft of this ar-ticle have meant significant improvements in both its content and general or-ganization. However, I have been obstinate enough not to follow their advicein all cases and, therefore, all remaining errors are my sole responsibility.

2. The termdoes not make use of movement operations.

3. Cf. Rijkhoff (2002: 2), who notes the lack of typological data on the order ofconstituents within the NP.

4. In the same line, Goldberg (2006: ch. 7) shows that both information structureand processing constraints underlie an adequate account of island effects.

5. Butler (this volume) notes that the interpretation of the conceptual and contex-tual components as mere adjacent modules to the grammar is in line with thetraditional reluctance of FG to fully undertake the description of the discourseand psychological factors which contribute to shaping language.

6. But see Lambrecht (1994: 91) for discussion of some exceptions to this.7. Although active referents can also receive special pragmatic treatment in cer-

tain circumstances (see section 4.1.1 below).8. Obviously, examples (13b and c) are grammatical if the prepositional phrase

is interpreted as a verbal adjunct (a second layer satellite in FGterms), and not as a nominal adjunct, which is the intended reading here.

9. Extractions from icalityjudgements among speakers and, therefore, I have restricted myself to thejudgements given in the original sources. Undoubtedly, not all readers willagree with them. As a matter of fact, an anonymous reviewer, a native speakerof English, considers both (15b) and (16b) marginally acceptable. Chomsky(1977b: 115) claims that this is to be expectedsuch as this.

10. I do not wish to claim that this is the end of the picture in the generativisttradition. Quite on the contrary, a great deal of research has been devoted tosubsuming island constraints and related effects under more general and ex-

358

planatory rules, such as the so-called Empty Category Principle (see e.g.Hornstein and Weinberg 1995). Yet, these proposals, apart from the technicalimplementation details, do not modify the general formalist strategy presentedin the previous lines (see Carnie 2006 for a brief sketch of the treatment of is-lands constraints up to the Minimalist Program).

11. Note that the existence of CNPC effects in languages not showing wh-displacement is a serious problem for characterizations of island violations asstructures difficult for the human parser to process due to the distance betweenthe extracted unit and the gap. For some examples of this strategy see Pritchett(1991), Kluender (1992), and Hawkins (1999).

12. It is important to observe that (i) Erteschik-Shir and Lappin do not equate thenotion of dominance with that of focus and (ii) dominance is defined as a dis-course property which is assigned to a constituent in a given context. Hence,the same relative clause may be used dominantly or non-dominantly in differ-ent settings.

13. Erteschik-Shir and Lappin do not use glosses consistently and some of theirEnglish translations of Danish examples may sound odd to the native ear.However, I have preferred to keep the examples as in the original source in thebelief that this does not affect the point to be made.

14. Note that similar problems appear in English in cases like, etc. which are also treated as reanalysed structures by generativists (see

the discussion in Erteschik-Shir and Lappin 1979: 60ff.).15. Of course, pragmatic notions such as Topic and Focus have also been em-

ployed in Generative Grammar and they are assumed to be relevant in the mo-tivation of movement operations. Yet there has been an unfortunate tendencyto reinterpret them as abstract syntactic features, possibly in an attempt at pre-serving the autonomy hypothesis (Lambrecht 1994: 27). Zubizarreta (1998) isa recent approach to focus accentuation as a syntactic marker.

16. Significantly, some of Ross uralif the head noun is turned into an indefinite NP: ?

17. Van Valin (1996) notes that topicalized constituents can also be interpreted asfoci. I will not take that possibility into account here.

18. Hawkins (1999: 272) argues that the introduction of a possessor modifiercontaining a referring expression in (16b) increases the amount of semanticprocessing and the structure becomes difficult to process. Although this expla-nation is in line with the one advocated here, it does not account for the con-trasts in (28) and (29), which indicates that other factors may well be involved.

19. As pointed out to me by Jan Rijkhoff (p.c.), the nouns in example (29) () refer to abstract non-first order entities and, therefore, require

different identification strategies. This may well be behind the contrastive in-terpretation that Kuno observed.

359

20. This could be easily formalized by assuming that retrieving a lexical itemactivates a number of properties related to its prototypical use or function (cf.Pustejovskywould most likely demand focal status.

21. Kluender (1992: 244-247) notes the relevance of the aspectual class of thematrix verb in extraction processes. Building upon Kluender(1999: 272) proposes a hierarchy of semantic specificity for matrix verbs.Roughly, if the meaning of a verb accepting an extraction process entails themeaning of another (e.g. entails ), then the latter should also acceptthe process. His assumption is that less specific verbs are more likely to allowextractions, since they require less semantic processing.

22. See Butler (this volume), Connolly (2004), Escribano (this volume) andKeizer (this volume) for discussion on the nature of the contextual and con-ceptual components in FDG.

23. Hence, the World of Discourse, understood as the interpretation constructedby the speech participants as the interaction unfolds, can be conceived of aspart of the conceptual component.

24. It is not clear to me whether Lambrechtrelate to the E-context, D-context, to the conceptual component (Rijkhoff -context) or to all of them. I assume this may be related to the nature of theframe or schema which gives access to a referent or even to one ptionof the notion

25. Of course, I do not wish to suggest that the communicative setting presentedhere is the only one possible. One could say that the characterization of thegeneral and verbal situation in Figure 3 does not contain necessary but suffi-cient features for the felicitous use of (38c).

Allwood Jens S.1976 The complex NP constraint in Swedish.

2.Bach, Emmon and George M. Horn1976 Remarks on 7:

265 299.Bolkestein, A. Machtelt1998 What to do with Topic and Focus? Evaluating pragmatic information.

In: Mike Hannay and A. Machtelt Bolkestein (eds.),, 193 214. Amsterdam / Philadel-

phia: John Benjamins.

360

Butler, Christopher S.this vol. Interpersonal meaning in the noun phrase.

Carnie, Andrew2006 Island constraints. In: Keith Brown (ed.)

2nd Edition, 48 51. Oxford: Elsevier.Chafe, Wallace1994 . Chicago. The University of

Chicago Press.Chomsky, Noam1977a Reprint. Conditions on transformations. In:

, 81 160. Amsterdam: North-Holland. Original editionP. Kiparsky and S. Peters (eds.), . TheHague: Mouton, 1973.

1977b On wh-movement. In: Peter W. Culicover, Thomas Wasow andAdrian Akmajian (eds.), , 71 132. New York: Aca-demic Press.

Connolly, John H.2004 The question of discourse representation in Functional Discourse

Grammar. In: J. Lachlan Mackenzie and MarG -Gonz

, 211 242. (Functional Grammar Series 24.) Berlin/New York:Mouton de Gruyter.

Cornish, Francis2004 Focus of attention in discourse. In: J. Lachlan Mackenzie and Mar

, 117 150. (Functional Grammar Series 24.) Berlin/NewYork: Mouton de Gruyter.

2006 Discourse anaphora. In: Keith Brown (ed.)2nd Edition, 631 638. Oxford: Elsevier.

Deane, Paul D.1992 . (Cognitive Linguistics Research 2.)

Berlin/New York: Mouton de Gruyter.Dik Simon C.1997a

(Functional Grammar Series 20.) Second, revised edition.Edited by Kees Hengeveld. Berlin/New York: Mouton de Gruyter.

1997b(Functional Grammar Series 21.) Edited by Kees

Hengeveld. Berlin/New York: Mouton de Gruyter.D2001 . Madrid:

C

361

Erteschik-Shir, Nomi and Shalom Lappin1979 Dominance and the functional explanation of island phenomena.

6: 41 85.Escribano, Josthis vol. On noun phrase structure in F(D)G: Some conceptual issues.

Giv1984 Amsterdam/Phila-

delphia: John Benjamins.Goldberg, Adele E.2006 .

Oxford: Oxford University Press.G -Gonz2001 . Amsterdam/

Philadelphia: John Benjamins.Grosu, Alexander1982 The extragrammatical content of certain

9: 17 67.Haig, John H.1996 Subjacency and Japanese grammar: a functional account.

20: 53 92.Hannay, Mike1991 Pragmatic function assignment and word order variations in a Func-

tional Grammar of English. 16: 131 155.Haspelmath, Martin1999 Explaining article-possessor complementarity: economic motivation

in noun phrase syntax. 75-2: 227 243.Hawkins John A.1991 On (in)definite articles: implicatures and (un)grammaticality predic-

tion 27-2: 405 442.1999 Processing complexity and filler-gap dependencies across grammars.

75-2: 244 285.Hengeveld, Kees2004a The architecture of a Functional Discourse Grammar. In: J. Lachlan

Mackenzie and Mar -Gonz, 1 21. (Functional Grammar

Series 24.) Berlin/New York: Mouton de Gruyter.2004b Epilogue. In: J. Lachlan Mackenzie and Mar

mez-Gonz ,365 378. (Functional Grammar Series 24.) Berlin/New York: Mou-ton de Gruyter.

2005 Dynamic expression in Functional Discourse Grammar. In: Casper deGroot and Kees Hengeveld (eds.),

362

53 86. Functional Grammar Series 27.) Ber-lin/New York: Mouton de Gruyter.

this vol. Prototypical and non-prototypical noun phrases in FDG.Hengeveld, Kees and J. Lachlan Mackenzie2005 Interpersonal functions, representational categories, and syntactic

templates in Functional Discourse Grammar. In: MarG -Gonz

, 9 27. (Linguistic Insights 26.) Bern: Pe-ter Lang.

2006 Functional Discourse Grammar. In: Keith Brown (ed.),2nd Edition, Vol. 4, 668 676. Oxford:

Elsevier.Hong, Sun-ho2003 On island constraints in Korean. In: Gregory K. Iverson and Sang-

Cheol Ahn (eds.), 107 125.Seoul: Hankook Publishing Co.

Hornstein, Norbert and Amy Weinberg1995 The Empty Category Principle. In: Gert Webelhuth (ed.),

, 241 296.Oxford: Blackwell.

Keizer, Evelienthis vol. Reference and ascription in FDG: An inventory of problems and

some possible solutions.Kluender, Robert1992 Deriving island constraints from principles of predication. In: Helen

Goodluck and Michael Rochemont (eds.),223 258. Dordrecht: Kluwer.

Kuno, Susumu1987 . Chicago: The University of Chicago Press.

Lambrecht, Knud1994 . Cambridge: Cambridge

University Press.Lehmann, Christian1984

. T ngen: Narr.Levelt, Willem J.1989 . Cambridge, M.A.: The

MIT Press.Lorenzo, Guillermo1995 . Oviedo: Departamento de

Filolog

363

Mackenzie, J. Lachlan and Evelien Keizer1991 On assigning pragmatic functions in English. 1: 169

215Prince, Ellen F.1981 Toward a taxonomy of given-new information. In: P. Cole (ed.),

s, 223 256. New York: Academic Press.Pritchett, Bradley L.1991 Subjacency in a principle-based parser. In: Robert C. Berwick, Ste-

ven P. Abney, and Carol Tenny (eds.),, 301 345. Dordrecht: Kluwer.

Pustejovsky, James1995 . Cambridge, M.A.: The MIT Press.

Rijkhoff, Jan2002 . (Oxford Studies in Typology and Theoretical

Linguistics.) Oxford: Oxford University Press.this vol. Layers, levels, and contexts in Functional Discourse Grammar.

Ross, John R.1986 Reprint. . Norwood: Ablex Publishing Company.

Original edition . Ph. D. diss.,MIT, 1967.

Shimojo, Mitsuaki2002 Functional theories of island phenomena. The Case of Japanese.

26: 67 123.Siewierska, Anna1991 . London: Routledge.

Van Valin, Robert D.1996 Toward a functionalist account of so-called extraction constraints. In:

Betty Devriendt, Louis Goossens and Johan van der Auwera (eds.),, 29 60. (Functional

Grammar Series 17.) Berlin/New York: Mouton de Gruyter.Zubizarreta, Mar1998 . Cambridge, M.A.: The MIT Press.

Ahrens, Kathleen, 104Akman, Varol, 106Allan, Keith, 215, 216Allwood, Jens S., 338, 346Anstey, Matthew P., 2, 5 6, 9, 20,

34 35, 181Anwar, Mohamed S., 298Auwera, Johan van der, 51

Bach, Emmon, 164, 330, 346Bakker, Dik, 14, 16, 30, 32, 168,

287, 288, 290 291, 296, 298,309, 312, 314 315, 317

Barbosa, Pilar, 170Barthes, Roland, 276Barwise, Jon, 136Bazzanella, Carla, 106Beale, Stephen, 240Behrens, Leila, 104Berg, Thomas, 315Bernstein, Judy, 160Biber, Douglas, 232, 249Bolinger, Dwight, 49, 256Bolkestein, A. Machtelt, 4, 322Brousseau, Anne-Marie, 26, 73B Karl, 22Bunke, H. 281Butler, Christopher S, 2, 4, 27, 29,

30, 32, 34 35, 49, 63, 75 76,90, 97, 102 103, 125, 165166, 173, 174, 214, 225, 243,256, 282 283, 357, 359

Bybee, Joan L., 78

Carlson, Robert, 53Carnap, Rudolph, 214Carnie, Andrew, 358Casey, Richard G., 281Chafe, Wallace, 324, 326 327, 340,

342, 345 347, 356

Chappell, Hillary, 105Chomsky, Noam, 2, 144, 167, 174,

329 330, 342, 346, 357Chung, Sandra, 104Churchward, C. Maxwell, 105Cinque, Guiglielmo, 123, 160, 164,

170Clark, Eve V., 104Clark, Herbert H., 104Coates, Richard, 60, 215Comrie, Bernard, 76Connolly, John H., 6, 30, 32, 34, 90,

98, 197, 203, 226, 240, 268,277 278, 283, 359

Cook, Guy, 274Cooper, Robin, 136Cornish, Francis, 322, 357Coseriu, Eugenio, 167, 174Craig, Colette Grinevald, 73Crisma, Paola, 160Cruse, David Alan, 105

Dasher, Richard B., 104Davidson, Donald, 123, 164, 215Deane, Paul D., 326, 337Declerck, Renaat, 186 187, 189,

196, 213Defare, Evelyne Peter, 315Dell, Gary S., 306Deutsch, Georg, 242Dik, Simon C., 1 3, 5, 13 15, 17

20, 22, 23, 27, 30 32, 34 35,51, 58, 60, 74, 84 86, 90 91,93, 98 99, 101 102, 118, 122,124 133, 136 145, 149 150,152, 154 155, 159, 161 169,174, 181, 183 188, 197, 238,263, 265 266, 270, 278, 282283, 287, 294, 314, 322, 332,350 351

366

D , Francesco, 335 337, 348,354

Dixon, Robert M.W., 86Dowty, David, 164

Ernst, Thomas, 159Erteschik-Shir, Nomi, 334, 358Ervin-Tripp, Susan, 90Escribano, Jos 19, 30 31,

34 35, 60, 66, 118, 125, 128,130, 134, 137, 142, 144 148,154, 157 161, 164, 168, 172174, 212, 214, 357, 359

Faber, Pamela, 230Farsi, A.A., 82 83, 104Foley, William A., 48, 185Fortescue, Michael, 97Frajzyngier, Zygmunt, 79Frege, Gottlob, 106, 120, 136, 168

169, 189 191, 214Fries, Peter H., 104Fromkin, Victoria A., 298

Garc Daniel, 16, 30, 3334, 103, 163, 214, 283, 288

Garnier, K. von, 79Garrett, Merrill F., 298, 304, 307

Geach, Peter T., 214Gerds, Donna B., 83Giegerich, Heinz J., 83Gildea, Spike, 79Giusti, Giuliana, 160Giv Talmy, 106, 327Goddard, Cliff, 257Goldberg, Adele E., 357Golla, Victor, 50G oli Jos 4G -Gonz Mar

357Goulet, Pierre, 242Greenberg, Joseph H., 78Groot, Casper de, 13

Grosu, Alexander, 346

Haegeman, Liliane, 290Haig, John H., 331 332Halle, Morris, 297Halliday, Michael A.K., 22, 76, 83,

97, 102, 105, 225, 230, 243Hanna, Patricia, 201, 215Hannay, Mike, 4, 210, 256, 322, 354Hannequin, Didier, 242Harder, Peter, 214 215Harley, Heidi, 297Harley, Trevor A., 306Harrison, Bernhard, 201, 215Haspelmath, Martin, 343Hawkins, John A., 78, 157, 324, 340,

341, 358, 359Heim, Irene, 122, 164Hendrick, Randall, 105Hengeveld, Kees, 2 4, 6 7, 9 15,

18, 20, 22 23, 27 32, 34 35,43, 48, 52, 54, 58, 60, 63 64,75, 79, 86, 88 89, 91 92, 9599, 102, 104 106, 118, 125130, 133 135, 137 138, 141,158, 161, 163 167, 169, 174,181, 183, 185 189, 192 197,200 201, 207, 212 214, 221227, 232, 238, 243, 246, 250,255 257, 265, 281 282, 296,314, 323, 350 351

Hetzron, Robert, 104Heusinger, Klaus von, 104Hieda, Osamu, 290Higgins, F. Roger, 186, 213Hong, Sun-ho, 333 334Horn, George M., 330, 338, 346Hornstein, Norbert, 358Huang, Chu-Ren, 104Huddleston, Rodney, 232

Jackendoff, Ray, 164Joanette, Yves, 242Jurafsky, Daniel, 74

367

Kamp, Hans, 104Karttunen, Lauri, 71Kayne, Richard S., 147, 159, 162,

164, 171Keizer, Evelien, 19, 30 31, 35, 58,

60, 75, 88, 104, 118, 124 126,128 131, 133, 135, 137 138,141, 147, 160 161, 164 166,168, 170 174, 181, 183, 186190, 192, 195, 209 210, 213216, 226, 283, 322, 359

Kluender, Robert, 358 359Koptjevskaja-Tamm, Maria, 81Kosslyn, S.M., 276Kratzer, Angelika, 122, 164Kress, Gunther, 264Kripke, Saul, 215Kroon, Caroline, 4, 10, 99Kuno, Susumu, 331, 338, 343 344,

346 347, 358Kutsch Lojenga, Constance, 79, 80

Lambrecht, Knud, 321, 323 327,339, 344, 351, 353, 357 359

LaPolla, Randy, 2, 77Lappin, Shalom, 334, 358Lee, K., 83Lefebvre, Claire, 26, 73Lehmann, Christian, 340Levelt, Willem J., 5, 6, 298, 304

305, 307, 323Levi, Judith N., 82Levinson, Stephen C., 96, 257Locke, John, 121, 201Locke, John L., 242Longobardi, Giuseppe, 160Lorenzo, Guillermo, 349Loux, Michael J., 121Lyons, John, 34 35, 60, 202, 214

Mackenzie, J. Lachlan, 2 4, 6, 7, 913, 20, 27, 29, 30, 33 35, 43,52, 54, 60, 86, 99, 103, 105,118, 125 128, 130, 133 134,

137, 138, 158, 163 167, 177,192, 193 194, 196, 212, 214,222, 238, 256 257, 322

Mahesh, Kavi, 240Mairal Us Ricardo, 230Malinowski, Bronislaw, 90Marantz, Alec, 297Martin, Laura, 73Mart Mingorance, Leocadio, 230Matthews, Robert J., 287Matthiessen, Christian, 225, 230Mayer, Karl, 298McCawley, James D., 201, 215McGee Wood, Maty, 164McGregor, William, 105Meringer, Rudolf, 298Merlan, Francisca, 83Mill, John Stuart, 214 215Miner, Kenneth L., 83Mithun, Marianne, 57, 79, 83Mohammad, Mohammad A., 309Montague, Richard, 136, 164

Newman, Paul, 79Ney, J.W., 104Nichols, Johanna, 288Nirenburg, Sergei, 240Nooteboom, Sieb G., 298Noyer, Rolf, 297Nuyts, Jan, 6, 34, 257

Olbertz, Hella, 35

Parsons, Terence, 123Partee, Barbara H., 122Pederson, Eric, 257Pesetsky, David, 169Pfau, Roland, 16, 30, 32, 168, 297,

298, 304, 314Prince, Ellen F., 324Pritchett, Bradley L., 358Pullum, Geoffrey, 232Pustejovsky, James, 161, 164, 171,

359

368

Quirk, Randolph, 32, 78, 82, 105,264, 266

Reich, Peter A., 306Reichenbach, Hans, 133, 168Rijkhoff, Jan, 1, 19 20, 22 27, 29

32, 34 35, 48, 56, 60, 63 67,69, 77 78, 84 85, 90, 101,104, 106, 125 126, 128 130,133 143, 146 147, 150, 153161, 163 174, 181, 191 192,200, 214, 221 226, 230, 243,255, 283, 287, 340, 342, 351,353, 355, 357 359

Roberts, John R., 70 71Ross, John R., 327 328, 338, 341,

350, 358Rossi, Mario, 315Rush, Susan, 274Russell, Bertrand, 120, 215

Sadock, Jerrold M., 102Saeed, John I., 15Schmid, Hans-J , 91Schumann, John H., 242, 257Scott, Gary J., 160, 170Searle, John, 90, 183Seibt, Johanna, 26, 35, 63, 67Seiler, Hansjakob, 78Shimojo, Mitsuaki, 331 332, 338Sichelschmidt, Lorenz, 104Siewierska, Anna, 4, 14, 288, 296,

312, 315, 317, 322Smit, Niels, 56Smith, John Charles, 104, 267

Solms, Mark, 242, 257Springer, Sally P., 242, 284Staden, Miriam van, 95Steedman, Mark, 164Stokes, Bronwyn, 72Strawson, Peter F., 120, 183, 184Stroomer, Harry, 84

Taylor, John R., 36Timberlake, Alan, 104Tomasello, Michael, 105Traugott, Elizabeth C., 104Turnbull, Oliver, 242, 257

Ungerer, Friedrich, 91

Van Valin, Robert D. Jr., 2, 48, 77,185, 290, 321, 332 333, 346,358

Vendler, Zeno, 34Vet, Co, 4, 18, 19, 197

Wang, P.S.P., 281Wechsler, Stephen, 105Weinberg, Amy, 358Wierzbicka, Anna, 257Wilkins, David P., 82Willemse, Peter, 36Wittgenstein, Ludwig, 121, 215Wray, Alison, 242, 257Vries, Lourens de, 5

Zubizarreta, Mar 358Zwicky, Arnold, 102

act (see also ), 4, 6, 9,10, 18, 21 22, 27, 32, 43, 60,70, 90 91, 97 99, 106, 118,123, 126 127, 129 130, 135,160, 165, 167, 181, 183 184,187, 191, 193, 196, 199, 209,210, 212 222, 227, 238, 241242, 249, 256, 267 270, 276277, 283, 323 324, 326

activation, 33, 288, 307, 308, 323327, 339, 340 341, 343, 345346, 350, 351, 353, 355 356

activation cost, 326, 345 346,356

addressee, 9, 10, 22, 32, 43, 60, 70,82, 94, 96, 98, 102, 166, 193,197, 222, 226 227, 266, 322324, 326, 340, 343, 345, 348,351, 353, 355 356

adicity, 122 123, 144 145, 153,163, 171

adjective, 16 17, 23 24, 35, 49 50,60, 64 66, 75 77, 79, 82 83,86, 96, 105, 109, 127, 147,150, 154, 156 157, 163, 171,173, 184, 196, 223 224, 232,237, 251, 254, 294, 299, 302

advertisement, 270, 274 275, 278agreement, 32, 145, 226, 287 288,

298, 309 310, 312 313, 315316

Aktionsart, 23, 84anaphor, 216anaphoric relation, 188, 192anaphoricity, 205anchor, 324, 340 341, 343, 345anti-symmetry, 69, 72, 74appositive construction, 182, 207,

212article, 26, 55, 64, 72, 85, 104, 150,

166, 172 173, 236 237, 255,291, 294, 301 305, 307 308,311 312, 314, 316

ascription, (see also ) 10, 31,39, 111, 181 182, 192, 208,211 212

attitude, 25, 28, 32, 48, 54, 74, 77,80, 98, 101 103, 221, 223,225, 228 229, 235, 237, 239241, 243, 246, 249, 251, 254255, 257

attitudinal meaning, 31, 225, 228229, 243

augmentative, 228

binary branching, 31, 141 142, 150152, 154 155, 157, 159 160,169

block language, (see also)

264, 278, 297, 282British National Corpus, 227 231,

233 238, 241, 246 247, 249250, 256 257

competence, 167, 313, 323Complex NP Constraint (CNPC),

321, 327, 329, 331, 333 334,336, 338 339, 340, 342, 349350, 358

componentaffective/interactional, 243conceptual, 6, 34, 91, 94 95,103, 106, 193, 214, 222, 225,240 241, 255, 257, 350 351,353, 355, 359

contextual, 6, 31, 33, 63, 88, 90,92, 182, 193, 197, 200, 203204, 206, 212, 214, 222, 226,231, 240 241, 246, 249, 251,

370

254 255, 257, 281 282, 323,350 351, 353 354, 357

C-context, 88, 91 92, 94 95,103, 355

D-context, 88, 90, 92, 94 96,103, 226, 351, 353, 355 356,359

E-context, 31, 63, 88, 90, 92, 94,96 97, 103, 226, 351, 353,355, 359

G-context, 88, 91 95, 97, 103grammatical, 6 9, 13, 27, 31, 6364, 77, 91 99, 102 103, 193,199 200, 204, 211 212, 222,231, 238, 250, 255, 281 282,350 351

situational, 31, 90, 98, 103, 351Compositionality, 159, 178conceptual content, 241, 242contextimmediate context, 214, 226, 316situational context, 90, 92, 96,226

contextual factor, 63, 88, 103,231, 249, 280, 282

copular construction, 182, 184, 186188, 194, 199, 205, 211 213

classificational sentence, 199classifying construction, 186identificational sentence, 182,196

identity statement, 182, 187, 189,191, 196 199, 206 208

co-text, 88, 90, 92, 95, 98, 103, 221,226, 351

definiteness, (see also ) 17,23 24, 66 69, 71 74, 104,168, 170, 186, 213, 224 225,304 305, 312 313, 317, 338339

demonstrative, (see also) 65, 76 77,

109, 236, 244 245, 341diachrony, 313diminutive, 74, 76, 228Discoursediscourse context, 32, 88, 120,221, 226 227, 254 255, 257

Discourse Act, 4, 10, 35, 323discourse entity, 190, 198, 204,206 207, 212, 214

discourse organization, 266, 270,280

discourse world, 69, 73 74, 101,104, 197 198, 200, 203, 227

discourse-strand, 272Domain Integrity (DI), 117, 143,

146 147, 150, 153, 156 157,159, 172

Dynamic Expression, 32, 287 289,290, 296, 305, 311 312

Encoding, 9, 45expression rule, 12, 14, 93, 143, 152,

158, 173, 287 289, 294, 296,312

extra-clausal constituent (ECC), 263,265

extraction processes, 288, 321 322,328 331, 333 350, 356 359

focusing, (see also) 339, 345, 349, 356

topicalization, (see also) 150, 329,

331, 339, 340 341, 34extraction constraint, 321, 356 (see

also )

feature mismatch, 298 299form-function relationship, 14 16,

23 27, 65 66, 80 83

371

Formulation, 6, 9, 45, 351Functional Lexematic Model, 230genderagreement, 291, 302, 304, 310feature, 239, 299, 301, 304, 312,314

mismatch, 299, 302 303, 315,316

genitive, (see also ) 83, 343

Head Proximity (HP), 123, 143,153 163, 165, 177 179

headline, 264, 270 272

iconicity, 5, 48, 75, 77 78, 147, 153,159, 243

identifiability, 214, 223 224, 323325, 343, 345

Image, 32, 149, 263, 267, 276 280,311

indefiniteness (see also ),67, 225

information structure, 321, 323, 339,342, 344 345, 349, 357

interpersonal meaning, 32, 221, 225231, 234, 249, 255 256

interpersonal level (see )interpersonal modifier (see )intonation, 213, 230, 242 243, 344,

346Irrealis, 26, 67 74, 104island, 327, 331 333, 350, 356, 357,

358, 361, 363island constraint, 331 333, 357

Layout, 279levelinterpersonal level, 6, 9, 11, 4344, 47 48, 53, 59, 125, 165166, 181 182, 185, 193 196,199, 203, 205 206, 208 212,214, 221 222, 240 241, 249,254, 264, 315, 351, 356

representational level, 6, 11, 13,22, 27, 30 31, 45, 47 48, 5360, 86, 99, 102, 118, 120,129, 163, 165 166, 181 182,185, 193 212, 222, 225, 250251, 264, 313

lexical entry, 144, 251, 307 308,313 314, 316

lexicon, 5, 52, 105, 202, 230, 246,249, 250 251, 255, 288, 294,306 309, 313, 315

Linear Co-occurrence Axiom (LCA),147 162

linearization, 291

mentalextension set, 31, 129, 201 211,215

lexicon, 202modifier, 15 18, 20, 22, 25 26, 29

35, 47 49, 54 57, 60 68, 7576, 79, 84 86, 97 106, 120127, 132 137, 141 142, 152174, 186, 194, 209 210, 223232, 238, 243 246, 249, 264,294, 307 308, 313, 342, 358

attitudinal, 63, 74 75classifying, 24 26, 63 64, 80,101

discourse-referential, 24, 26, 31,35, 63 64, 66, 99, 10

illocutionary, 25, 101interpersonal, 29, 32, 64, 67, 76,97, 221, 227, 231, 234, 238,246, 249, 255

localizing, 15, 22, 25 26, 63, 66,77, 99, 101

proposition, 101qualifying, 16, 22, 25, 77, 80, 86,101, 105

quantifying, 22, 25, 66 67, 101Move, 10, 106multimodal, 264, 274, 276, 278

372

node, 241, 290 296, 307 309, 313,329, 330, 333, 336 337

terminal, 291, 29nominalization, 51 52nounincorporation, 56possessive, 344proper, 16, 32, 54 55, 60, 120,165, 182, 199, 207 209, 212,215, 232, 235, 301

noun phraseascriptive, 59possessive, 342, 344prototypical, 16, 31, 34, 43, 195,213, 223, 288

referential, 56standard, 43, 46, 50, 57, 59

noun stripping, 83numeral, (see also

) 65, 75, 84,168, 173, 184, 236, 245

ontology, 1, 20, 31, 117, 120 123,126, 162 167, 174, 239 241,251

operator, 9, 12, 17 31, 40, 45, 4748, 54, 59, 65 68, 77, 81, 8487, 91, 93 94, 99, 103 106,117, 123 124, 126 127, 131142, 146, 149 152, 154, 158,160, 163, 165 166, 168 170,174, 183, 190 192, 194, 199,201 206, 215, 223 229, 232,287, 291 292, 307, 333, 335

classifying, 27, 84, 87discourse-referential, 26, 66 67,77, 87, 94, 99

localizing, 23, 65, 66 67qualifying, 27, 66, 84 86, 103,224

quantifying, 23, 66, 168R-operator, 228

P1 position, 9, 10, 13, 34, 43, 54, 98,119, 139 140, 193, 222, 294,354 356

parallel, 22, 31, 41, 79, 97, 127, 132,138, 141, 165, 168, 224 226,250, 255, 303, 309 311, 340

possessor, (see also ,) 17, 65,

343 345, 358pragmatic function (freestanding NP)Caption, 266, 276 278, 280, 282Finish, 266 267Heading, 17, 264 275, 277, 279282

Label, 266, 276, 277, 278, 280,282, 292

Prelim, 266, 267Titling, 266 270, 272, 274 280,282

pragmatic function, 10 11, 32, 208,256, 269 270, 277 278, 282,322 323, 325, 339 340, 350351, 354, 356

Focus, 11, 34, 77, 150 151, 172173, 256, 322 323, 333, 351,356, 358

Topic, 11, 14, 34, 150 151, 172,256, 322 323, 325 326, 351,354, 356, 358

pragmatic rule, 32, 278, 281 282pragmatics, 4, 32, 77, 105, 242, 265,

276predication, 10, 12, 17 18, 21, 51,

123, 181, 183 184, 188, 212,214, 339

embedded, 18, 51, 154, 156 157,164, 166, 173, 238, 240, 264,328, 338, 341, 343, 345, 347348

priority, 148, 315, 322pronoun, 15, 19, 20, 32, 53 55, 58

59, 96 97, 153, 182, 188, 19192, 195, 199, 204 205, 207

373

209, 212, 216, 232, 235 236,264, 289 290, 302

prototypicality, 213quantifier, (see also

) 119 122,136, 146, 163, 184

Realis, 26, 67 74, 98, 101, 104reference (see also ), 3, 10, 15,

20, 26, 31, 48, 50, 54, 56, 58,60, 67, 69 73, 104, 118, 124,126 129, 136, 141, 152, 160,163, 165 167, 172, 181 182,185, 191 194, 196, 198 200,203 204, 208 212, 222 223,230, 237, 250, 254, 256, 307,323, 350, 357

definite reference, 26, 55, 69, 71,105, 121, 237, 255, 299 301,304, 324

indefinite reference, 192, 299,324, 340

non-specific reference, 72 73,104

reference modification, 49, 50,54, 223, 254, 256

specific reference, 9, 12, 14 15,45, 71 72, 74, 104, 118, 124,154, 156 157, 162, 165, 204205, 214, 239, 257, 269, 276,288, 290 292, 295, 312 316,322, 340, 354, 359

referent, 17, 20, 26, 31 35, 47 49,54, 57, 63 64, 66, 69 72, 74,76, 82, 87, 94, 103 105, 118119, 125, 128 129, 132 133,145, 161, 166, 182 188, 190,191, 193 195, 199 200, 202211, 214 215, 221, 223 224,227, 238 240, 244 245, 249257, 322 327, 339 346, 349351, 353 357, 359

accessible, 69, 146, 210, 296,307, 315, 324, 326, 355 356,359

active, 324 326, 340, 341 342,344 345, 353 357

inactive, 353prototypical, 190, 214referent modification, 48, 54,223, 254, 256

relative clause, 16 18, 23 24, 51,64, 66, 154, 160, 168, 176,328, 331, 333 334, 338, 340342, 344, 358

Role and Reference Grammar, 77,243

satellite, 15, 21, 23, 25, 34 35, 66,78, 80, 83 87, 99, 105 106,134, 139 140, 142, 147, 149,156, 168, 171, 214, 224, 357

satisfaction, 144 145, 156, 161, 23, 35, 84

scope, 9, 23, 29, 30 31, 48 50, 67,75, 117 118, 131, 136 139,142 143, 148 155, 158 160,165, 168 169, 173, 186, 202,224, 238, 265, 280, 282 283,313

speaker, 6, 9 10, 21 22, 32, 43, 4649, 54, 60, 64, 68 77, 80, 82,85, 94 96, 98, 102 105, 118,125, 126, 129 130, 132, 190193, 196, 198, 200 203, 206212, 221 223, 225 226, 228,230, 238 241, 243, 249, 251,254 257, 283, 297, 301, 316317, 322 323, 325, 334, 339,345, 347 348, 351, 353 357

speech error, 32, 287 288, 291, 297,298, 305, 312, 314 316

agreement, 288, 298, 313, 315blend, 299, 301, 303, 306anticipation, 301 302, 304, 311

374

exchange, 10, 25, 63, 97, 99102, 299, 301 302, 311, 35

paradigmatic, 302, 305 306,308 309, 311

perseveration, 302 304, 308, 311self-correction, 298 299, 302,305, 308, 311, 313 314, 316

syntagmatic, 305, 308Stepwise Term Construction (STC),

131, 134, 139, 141 148, 151152, 159, 162, 169

strong domain, 146Subact, 10, 28, 44, 46, 193 196,

199, 203, 205, 209 210, 222,256, 323

ascriptive, 28, 29, 46, 58, 60,106, 181, 193, 196, 199, 205,210, 222

referential, 28 29, 46 47, 53, 58,106, 127, 207, 222 223

Subjacency, 329, 330subject-verb agreement, 298, 310symmetry, (see also )

27, 69, 71 72, 74Systemic Functional Grammar, 105,

225, 231, 243, 256

typology, 1, 2, 40, 45, 79, 187, 256,331, 357

underlying representation, 12 13,77, 91, 186, 193, 203, 290292, 297, 305 306, 309, 315

variable, 1, 10 12, 14, 17 20, 22,26, 28 31, 34 35, 51, 53, 60,63, 65 66, 86 87, 99, 102,117 142, 145, 152, 158, 161,163 169, 174, 182 186, 188,191 194, 199, 200 204, 208213, 224, 251

f-variable, 11, 25, 60, 183, 185186, 188, 205, 208, 210

individual, 122, 124unbound, 134, 136, 163

x-variable, 11, 19, 35, 66, 100,182, 188, 200, 205, 210 212

zero sense, 126, 133, 135variable-binding, 133 135, 152, 167,

174verbal interaction, 3 4, 32, 197, 238,

265, 278, 281 282, 322, 349350, 353, 356 357

vocative, 4, 55

web page, 278, 280WordSmith Tools, 231, 256

Abkhaz, 288Aghem, 78Alamblak, 78Amele, 70, 71, 113Arabic, 293, 309, 319, 319Arrernte, 82, 115

Bambara, 78Basque, 78Berbice Dutch Creole, 78Bisaya, 90Bislama, 78Burushaski, 78

Caddo, 57

Danish, 334, 358Dutch, 15, 34, 42, 48, 75, 76, 78, 97,184, 288

English, 13, 16, 19, 32, 52, 72, 78-80, 82, 91, 93-95, 117, 120, 137,147 148, 151 160, 172, 182,210, 212, 217 217, 221, 227228, 231, 239, 241, 246, 248,254 257, 263 264, 285, 288,329, 333 335, 337, 340, 343,354, 357 358.

Fa d , 78Fongbe, 26, 73, 77French, 78, 147, 315, 325

Georgian, 78German, 32, 287 288, 292 317Guaran , 78

Hmong Njua, 78Hungarian, 78Hupa, 50 51

Jacaltec, 73, 77Japanese, 331 333

Kayardild, 78Ket, 78Kikuyu, 78Koegu, 289 290Kombai, 5Korean, 90, 333 334Kusaiean, 83

Lakhota, 332 333Lithuanian, 81

Nahuatl, Modern, 83Nama Hottentot, 78Ngiti, 79 80Noni, 78Nyigina, 72

Ojibwa, 78Oromo, 78, 84

Panare, 79Pipil, 78Polish, 288Portuguese, Brazilian, 54 55

Quechua, Imbabura, 78

Sango, 78Spanish, 96, 147, 166, 221, 227230, 256, 315, 335 337, 348349

Sranan, 78Supyire, 52 53Swedish, 81, 338

Tamil, 78Tidore, 95

376

Tongan, 105Turkish, 78

Zande, 78