Pronouns: Grammar and Representation

Pronouns – Grammar and Representation Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA...

Author: Horst J. Simon and Heike Wiese (Editors)

149 downloads 2306 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Pronouns – Grammar and Representation

Linguistik Aktuell/Linguistics Today Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

Series Editor Werner Abraham University of Vienna, Austria

Advisory Editorial Board Guglielmo Cinque (University of Venice) Günther Grewendorf (J.W. Goethe-University, Frankfurt) Liliane Haegeman (University of Lille, France) Hubert Haider (University of Salzburg) Christer Platzack (University of Lund) Ian Roberts (Cambridge University) Ken Safir (Rutgers University, New Brunswick NJ) Lisa deMena Travis (McGill University) Sten Vikner (University of Aarhus) C. Jan-Wouter Zwart (University of Groningen)

Volume 52 Pronouns – Grammar and Representation Edited by Horst J. Simon and Heike Wiese

Pronouns – Grammar and Representation

Edited by

Horst J. Simon Heike Wiese Humboldt University Berlin

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Pronouns – grammar and representation / edited by Horst J. Simon and Heike Wiese. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 52) Includes bibliographical references and index. 1. Grammar, Comparative and general--Pronoun. I. Simon, Horst J. II. Wiese, Heike. III. Linguistik aktuell ; Bd. 52. P279 P75 2002 415-dc21 isbn 902722773X (Eur.) / 1588112519 (US) (Hb; alk. paper)

2002074428

© 2002 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

SUBJECT "Linguistik Aktuell/Linguistics Today, Volume 52"

KEYWORDS ""

SIZE HEIGHT "220"

WIDTH "150"

VOFFSET "4">

Table of contents

List of contributors

vii

Preface

ix

Glossary

xi

Grammatical properties of pronouns and their representation: An exposition Heike Wiese and Horst J. Simon

1

Structuring the bundle: A universal morphosyntactic feature geometry Heidi Harley and Elizabeth Ritter

23

‘We’ rules: The impact of an inclusive/exclusive opposition on the paradigmatic structure of person marking Michael Cysouw

41

Reference devices in Sinhala Neelakshi Chandrasena Premawardhena Indefinite pronouns: Morphology and syntax in cross-linguistic perspective Helmut Weiss

63

85

Reference and representation of pronouns Klaus von Heusinger

109

The dynamics of syntax: Anaphora, relative pronouns and crossover Ruth Kempson and Wilfried Meyer-Viol

137

The third person pronoun tripartite verbless clauses of Qumran Hebrew Jacobus A. Naudé

161

< /R/TREARGET E FF

"pan"> "mue"> "pin"> "roo"> "si"> "toc">

vi

Table of contents

Pronominal nouns Phoevos Panagiotidis

183

Harmonic alignment and the hierarchy of pronouns in German Gereon Müller

205

Cortical reflections of two pronominal relations Maria Mercedes Piñango

233

Pronoun omission in Dutch and German agrammatic speech Esterella de Roo

253

Language index

285

Subject index

287

List of contributors

Michael Cysouw Zentrum f. Allg. Sprachwissenschaft, Typologie und Universalienforschung Jägerstr.10/11 10117 Berlin, Germany

Wilfried Meyer-Viol Philosophy Department King’s College London The Strand London, WC2R 2LS, UK

[email protected]

[email protected]

Neelakshi Chandrasena Premawardhena Department of Modern Languages University of Kelaniya Dalugama Kelaniya, Sri Lanka

Gereon Müller Institut für Deutsche Sprache Postfach 101621 68916 Mannheim, Germany

[email protected]

Heidi Harley Department of Linguistics University of Arizona Douglass 200E Tucson, AZ 85721, USA [email protected]

Klaus von Heusinger Fachbereich Sprachwissenschaft Universität Konstanz Fach D185 78457 Konstanz, Germany [email protected]

Ruth Kempson Philosophy Department King’s College London The Strand London WC2R 2LS, UK [email protected]

[email protected]

Jacobus A. Naudé Department of Near Eastern Studies University of the Free State P. O. Box 339 9300 Bloemfontein, South Africa [email protected]

Phoevos Panagiotidis Department of Foreign Languages & Literatures University of Cyprus Kallipoleos 75 PO Box 20537 1678 Nicosia, Cyprus [email protected]

Maria M. Piñango Department of Linguistics Yale University P. O. Box 208236 New Haven CT 06520, USA [email protected]

Elizabeth Ritter Department of Linguistics University of Calgary

viii List of contributors

Calgary, Alberta T2N 1N4, Canada [email protected]

Esterella de Roo University of Leiden Centre for Linguistics (ULCL) P. O. Box 9515 2300 RA Leiden, The Netherlands [email protected]

Horst J. Simon Institut f. dt. Sprache und Linguistik Humboldt-Universität zu Berlin Unter den Linden 6 10099 Berlin, Germany [email protected]

Helmut Weiß Allgemeine Sprachwissenschaft Universität Regensburg Universitätsstr. 31 93053 Regensburg, Germany [email protected]

Heike Wiese Institut f. dt. Sprache und Linguistik Humboldt-Universität zu Berlin Unter den Linden 6 10099 Berlin, Germany [email protected]

Preface

This book presents selected papers from a workshop at the Annual Conference of the German Society for Linguistics (DGfS) in March 2000. We would like to thank all participants for their stimulating discussion which laid the foundations for this volume. Special thanks go to Paul Law, our co-organiser of the workshop. The selection process would not have been possible without the invaluable help of a number of reviewers, whom we would like to thank here anonymously — and unanimously. We would also like to thank Werner Abraham, the editor of this series, for his support. Finally, we want to thank Allison Schmidt and Tatjana Danilewitsch for their help in preparing and proof-reading the manuscript. This work was supported by a research grant provided by the Institut für deutsche Sprache und Linguistik of Humboldt-University Berlin.

Horst Simon, Heike Wiese Berlin, November 2001

Glossary

1 2 3 acc aph art comp cond conj dat dem dist do emph excl fem fut gen imp impf incl indef inf inj

1st person 2nd person 3rd person accusative anaphoric pronoun article complementiser conditional conjunctive dative demonstrative distal pronoun direct object marker emphatic exclusive feminine future tense genitive imperative imperfect inclusive indefinite infinitive interjection

inst interr loc masc mod neg nom opt part past perf pl poss pres pron prox1 prox2 ptcl quot refl sg spec

instrumental interrogative marker locative masculine modal negation nominative optative participle past tense perfect plural possessive present tense pronoun 1st proximal pronoun 2nd proximal pronoun particle quotative reflexive singular specific

Grammatical properties of pronouns and their representation An exposition Heike Wiese and Horst J. Simon Humboldt University Berlin

1.

Overview

This volume brings together a cross-section of recent research on the grammar and representation of pronouns, centering around the typology of pronominal paradigms, the generation of syntactic and semantic representations for constructions containing pronouns, and the neurological underpinnings for linguistic distinctions that are relevant for the production and interpretation of these constructions. In this introductory chapter we first give an exposition of our topic (Section 2). Taking the interpretation of pronouns as a starting point, we discuss the basic parameters of pronominal representations, and draw a general picture of how morphological, semantic, discourse-pragmatic and syntactic aspects come together. In Section 3, we sketch the different domains of research concerned with these phenomena and the particular questions in which they are interested, showing how the papers in the present volume fit into the picture. Section 4 summarises the individual papers and gives a short synopsis of their main points of convergence.

2. Basic parameters of the grammar and representation of pronouns One of the features that make pronouns a special class of linguistic items is the way in which they contribute to the meaning of sentences (or other constructions

2

Heike Wiese and Horst J. Simon

in which they occur). On the one hand, they can pick out the same kinds of objects as full lexical nominals when they enter interpretation. On the other hand, they lack a comparable descriptive content. This gives them a borderline status within the linguistic system, between lexical categories like nouns and functional categories like complementisers. Nominals are like pronouns in that they identify objects, but unlike pronouns they do so based on their descriptive content. Complementisers are like pronouns in that they lack a descriptive content, but unlike pronouns they do not pick out objects in discourse. Figure 1 illustrates this in-between status of pronouns. lack descriptive content

pick out objects

e.g. nouns

PRONOUNS

e.g. complementisers

Figure 1.Pronouns as a borderline case.

Take for instance a pronoun like she. In the context of an utterance, she can pick out the same individual as a nominal like Mick’s sister without contributing a predicate like ‘sister of Mick’ to support the task of identification. The denotation of pronouns is crucially dependent on other elements in the discourse, drawing on the linguistic and the non-linguistic context. Taking a general approach to this phenomenon, we can distinguish morpho-semantic, discourse-pragmatic, morpho-syntactic and syntactic means that serve to establish the link between a pronoun and an object. In the following paragraphs, we discuss some core examples from the pronominal domain in order to illustrate these different aspects and to show how they come together in the representation of pronouns. Morpho-semantic means which support the interpretation of pronouns draw on features that are contributed by the pronoun itself. Pronouns are part of a paradigm whose positions are defined by a more or less elaborate system of morphological features. These features can identify members out of a selected set of conceptual distinctions, for instance in English, ‘number’ as indicated by singular/plural

Grammatical properties of pronouns and their representation

distinctions, ‘role in the speech act’ (such as ‘speaker’, ‘addressee’, or ‘other’ i.e. non-speech-act-participant) as indicated by person distinctions, or classifications like ‘male/female/inanimate (or non-human)’ as indicated by the distinctions realised in he vs. she vs. it. From a cross-linguistic point of view, person and number appear to be the basic pronominal categories that are involved here. Universally, paradigms of personal pronouns seem to distinguish at least some speech act roles and to give at least some indication of whether one or more than one entity is involved. Many languages manifest further distinctions in their paradigms: Most notably gender (correlated with sex or with other conceptual or non-conceptual classifications), but also distinctions according to, for example, considerations of politeness (‘respect pronouns’).1 In addition, pronominal paradigms in some languages make distinctions with respect to less wide-spread categorisations, such as ‘protagonist status’ (obviation; cf. for instance Mithun 1999: 3.1.3 on Algonquian languages), or ‘generation of persons involved’ as in, e.g., Lardil (cf. Hale 1966). In the utterance of a sentence like (1), number and person distinctions pick out the speaker as the object which the pronoun identifies: (1) Rose asked me about the movie.

In (2), the features that the pronoun contributes delimit the range of possible referents by excluding both speaker and addressee, and identifying a single male human (using ‘single’ in the sense of ‘one’ — versus many —, not in the sense of ‘unmarried’, of course). The identification of one particular person within this range can then be accomplished via discourse strategies, for instance by interpreting an indicating act, like a gesture or certain eye movements, that might accompany the utterance in (2): (2) Oh dear — look at him!

The pragmatic strategy necessary to interpret the pronoun in an utterance like (3) combines the interpretation of the linguistic context (‘Elizabeth married’) with general world knowledge (marrying is a ceremony between two persons). This way, from the range of possible objects that he identifies (i.e. male persons) we can single out Elizabeth’s husband: (3) Elizabeth married last Tuesday. He is Italian.

In all three cases, the pronoun itself determines a specific choice from (a restricted set of) conceptual distinctions. It selects members out of pairs of

3

4


corresponding features, for instance ‘one’ (vs. ‘many’), ‘male’ (vs. ‘female’) and ‘human’ (vs. ‘non-human’). This gives us the basic parameters for a conceptual representation, for instance ‘individual male person’ in (2) and (3). In contrast to a nominal like a man, a pronoun like he does not provide such a conceptual representation by virtue of its descriptive content, but contributes the respective conceptual distinctions via grammatical features that draw on a morphosemantic paradigm. These conceptual distinctions provide the mould into which a referent can then be fitted via pragmatic knowledge and general discourse strategies. Whereas for a 1st person singular pronoun like me in (1), the bulk of the job is done by the pronoun — in an arbitrary utterance the features contributed by me suffice to identify a particular person, namely the speaker —, the division of labour can also be the other way round. This is, for instance, the case in topic-drop languages where a sentence-initial pronoun can be dropped if pragmatic reasoning allows us to pick out a referent without the support of an explicit pronoun — and hence without morphological devices that specify a value for ‘speech act role’ or ‘number’ as a starting point. (4) gives an example from German: (4) Wo ist der Kuchen? — where is the cake ‘Where is the cake?’ [Ø] Hab ich aufgegessen [pron] have I up.eaten ‘I have eaten it up.’

Between the two extremes we illustrated in (1) and (4) are cases like (2) and (3) where the pronoun’s morpho-semantic features interact with pragmatic reasoning in the fixing of reference. Figure 2 brings together the different options we have discussed thus far: When identifying a discourse referent, one can rely on discourse-pragmatic strategies exclusively, or make use of morpho-semantic features of the pronoun that identify conceptual distinctions. These distinctions delineate the range of possible referents; discourse-pragmatic strategies then allow us to pick out one element from this range. A peculiarity of 3rd person pronouns is the option to use them anaphorically. In anaphoric usage, pronouns identify an object indirectly, as illustrated in (5): (5) Steve’s aunt married last Tuesday. She is Italian.


morphosemantic features

PRONOUN discoursepragmatic strategies

CONCEPTUAL DISTINCTIONS discoursepragmatic strategies

Figure 2.How to identify a discourse object for a pronoun.

In this case, the pronoun she does not pick out a referent directly, but is linked to another linguistic item: a nominal (Steve’s aunt) that serves as its antecedent. It is via the link to this antecedent that the pronoun gets its meaning. The connection between a pronoun and its antecedent can be signalled by syntactic agreement based on the morphological distinctions that the pronoun contributes (e.g. in (5): 3rd person, singular, feminine). In addition, this link is supported by pragmatic strategies, as the German example in (6) illustrates (coindexation marks the link between a pronoun and its preferred antecedent). (6) Charles hat Himbeeressigi mitgebracht. — Charles has raspberry.vinegar.masc with.brought ‘Charles has brought raspberry vinegar with him.’ Ich habe ihni für den Salat benutzt. I have pron.3sg.masc for the salad used ‘I used it for the salad.’

Based on syntactic agreement alone, both Charles and Himbeeressig make suitable antecedents for a 3rd person masculine singular pronoun like ihn. However, our world knowledge suggests that the vinegar, and not Charles, went into the salad, hence in the preferred reading of (6), Himbeeressig is the antecedent for ihn (that is, it would normally be ‘it’ in the English paraphrase). The influence of pragmatic reasoning becomes even more obvious when one compares (6) to (7):

5

6


(7) Charlesi hat Himbeeressig mitgebracht. — Charles has raspberry.vinegar.masc with.brought ‘Charles has brought raspberry vinegar with him.’ gefragt, was das soll. Ich habe ihni I have pron.3sg.masc asked what that shall ‘I asked him what that is supposed to be good for.’

With regard to the pronoun, the context is here the same as in (6), but then the sentence goes on with a verb gefragt (‘asked’) which suggests that Charles, rather than the vinegar, is the recipient since our experience is such that one does not talk much to vinegar. Accordingly, in the preferred reading of (7), Charles, and not Himbeeressig, is the antecedent for ihn (and accordingly, it would be ‘him’ in the English paraphrase). Apart from pragmatic strategies, the syntactic configuration may be relevant for determining an antecedent, as illustrated by the different binding restrictions for the reflexive and the personal pronoun in (8a) vs. (8b):2 (8) a. Kareni read a letter to herselfi. b. Kareni read a letter to herj/*i.

The link is here established via agreement features plus the syntactic configuration, leading to the identification of Karen as the antecedent of the reflexive pronoun in (8a) and to the rejection of Karen as an antecedent for the personal pronoun in (8b). In this case, the syntactic structure can suggest a certain reading independently of our world knowledge. This means that we can get counter-intuitive interpretations as in (9): (9) Rose told me that Kareni is going to visit herselfi.

In this sentence, Karen is identified as the antecedent of herself, while Rose is excluded — even though, based on pragmatic reasoning, Rose would be a much better candidate since our world knowledge suggests that it is anyone but Karen that Karen would visit. Figure 3 includes the link to an antecedent into our picture of pronominal representation and the (grammatical and pragmatic) sources it draws on. Dotted lines indicate additional information (in particular, conceptual distinctions that draw on morpho-semantic paradigms) that can enter the computation and support the identification of a discourse referent, in accordance with Figure 2 above. As the case of relative clauses shows, the syntactic structure can also support a link between an antecedent and a phonologically empty position, i.e. without


syntactic configuration

ANTECEDENT descriptive content

discoursepragmatic strategies

morphological features:

agreement

PRONOUN CONCEPTUAL DISTINCTIONS

Figure 3.How to identify a discourse object via an antecedent.

the additional support of the agreement features of an overt pronoun, a phenomenon parallel to the topic-drop case in (4) above. (10) gives an example of a phonologically empty position in an English relative clause (represented as [proni]; this is an empty position or a trace, depending on the syntactic analysis one assumes for that-relative clauses in English). (11) gives two examples from Persian. In (11a), an overt resumptive pronoun may optionally appear, while in (11b) the resumptive pronoun is obligatorily overt: (10) the booki that Karen bought [proni] ke (ui) mı¯raqsad zan-ı¯i woman-spec comp pron.3sg danced.3sg ‘the woman that danced’ b. zan-ı¯i ke u-ra¯i dı¯dam woman-spec comp pron.3sg-acc saw.1sg ‘the woman that I saw’

(11) a.

In our overview so far, we have not yet discussed a class of words that are traditionally also included as a pronominal subclass, namely wh-words like who or what. How do they fit into the picture? (12) Who is this? (13) What is this?

7

8


In (12) and (13), the wh-words occur in an interrogative context; they mark the constituent that is asked for. However, wh-words are not confined to interrogative clauses but appear in a wide range of sentence types. They occur systematically in exclamative and declarative contexts, and introduce clausal attributes (relative clauses) and complements (that is, embedded clauses that constitute interrogative, exclamative, and declarative contexts). (14) through (17) illustrate some of these contexts: (14) What a nice day it was!

[exclamative]

(15) Anna hört wen. Anna hears whom Anna hears someone.

[declarative; German]

(16) the woman who called

[relative clause]

(17) She told him who called.

[declarative; embedded clause]

As argued in Wiese (in press), the different usages can be captured by a unified semantic representation of wh-words as lexically underspecified elements, which do not gain interrogative, exclamative or indefinite-referential force before they enter interpretation.3 What is crucial for our discussion here is that in all of these cases, the wh-word does not introduce a referent into the interpretation via a descriptive content, but can contribute conceptual distinctions via morpho-semantic features, along the lines we sketched above (for instance, English who vs. what in (12) and (13) support the distinction [±human]). Based on syntactic and discourse-pragmatic devices, these distinctions can then provide the basis for the pronoun’s denotation. The entity that is picked out can be left unspecific (as in (15), similar to indefinite pronouns); it can be identified via an antecedent (as for the relative pronoun in (16)); it can be marked as a degree above a (contextual) norm (as in the exclamative context in (14)), or it can be left open (as in the interrogative contexts in (12) and (13)), signalling the addressee to identify an entity that fits into the conceptual distinctions the pronoun provides and which satisfies the context set up by the interrogative. Let us sum up our exposition. We started with the observation that in the case of pronouns, one faces the task of identifying an object in the absence of a descriptive content. Two options are available: a pronoun can pick out an object directly (no linguistic antecedent) or indirectly (the pronoun is linked to a linguistic antecedent). The interpretation can draw on the following means:


–

– –

–

Morpho-semantic devices: Via morphological paradigms, pronouns can determine choices within a restricted set of conceptual distinctions (e.g. speech act role, ‘one’ vs. ‘many’, ‘human’ vs. ‘non-human’), which limit the range of possible referents or possible antecedents. Discourse-pragmatic devices: The discourse context and world knowledge contribute to the identification of a referent or an antecedent. Morpho-syntactic devices: The link between a pronoun and its antecedent can be supported by morpho-syntactic agreement (with respect to e.g. person, number, gender). Syntactic devices: The syntactic structure can identify (or exclude) possible antecedents.

This outlines the basic parameters relevant for our topic and makes clear how the different aspects in the grammar and representation of pronouns are related. In the following section, we illustrate the different kinds of research questions that evolve from this, within typology, formal syntax and semantics, and psycho- and neurolinguistics, and show how the contributions to the present volume fit into the picture. The final section then provides individual summaries of the papers.

3. Research questions Pronouns are relatively easy to identify cross-linguistically which makes them an ideal candidate for typological investigations. Hence, it is probably no accident that one of the first books to discuss a grammatical category in a genuinely typological way, Forchheimer (1953), focused on pronouns. Typological approaches to pronouns frequently deal with the grammatical categories that organise pronominal paradigms and with the factors that govern their development over time. As we have illustrated above, these paradigms can play a role both for the identification of conceptual distinctions and for the morpho-syntactic agreement that supports the link between a pronoun and its antecedent. Typological approaches contribute to our understanding of the basic pattern underlying such paradigms, making clear which properties are chosen for grammaticalisation from a potentially infinite pool of conceptual features, i.e. which are the distinctions that languages employ to pick out a referent in the interpretation of pronouns.

9

10


Four contributions to the present volume — Harley and Ritter, Cysouw, Chandrasena Premawardhena, Weiß — are concerned with pronominal paradigms, the features organising them, and the way they contribute to the interpretation of pronouns (cross-linguistically or in a particular language). The issues they deal with can be subsumed under the following questions: – – –

How are pronominal paradigms structured, and which constraints govern their organisation? Which grammatical categories are involved, and how do they split up into different features? How do these categories interact? Are there possible neutralisations in certain grammatical environments? What is the influence of extra-grammatical parameters?

Semantic and syntactic analyses aim to account for the way pronouns enter hierarchical structures and how this integration effects their interpretation. Much of semantic research has centered on elaborations of the somewhat simplified picture we sketched above, discussing amendments necessary to account for more complex relationships between a pronoun and its antecedent;4 most notably in cases where a pronoun has a quantifier expression as its antecedent or where the pronoun can have an interpretation as in the ‘sloppy identity’ reading illustrated in (18):5 (18) Only Karen takes her cat to the beer garden.

A major line of research on the semantics-syntax interface that is of direct relevance for our topic focuses on two pronominal subclasses, 3rd person pronouns and reflexive pronouns, and the way they are related to an antecedent. While the binding principles aiming to account for these relationships were originally formulated as syntactic constraints on the placement of pronouns vs. reflexives, later versions approached the problem from the point of view of the interpretational rules that access syntactic structures.6 In some recent approaches, the rules system is largely reduced to pragmatic principles.7 Other questions pertaining to our topic concern the categorial status of pronouns and their projections (e.g. ‘Are pronouns lexical or functional elements?’), and the way we can account for the correlation between different linear and hierarchical structures in the generation of sentences containing pronouns (‘Do we have to assume movement and traces to account for the syntactic behaviour of pronouns? If so, should we provide counterparts for syntactic traces in our semantic representations?’).


Results from these research areas contribute to our understanding of how the syntactic structure is organised that is involved in the derivation of interpretations for pronouns, how pronominal paradigms provide a basis for morpho-syntactic agreement, and how on the other hand the semantic side of such paradigms is realised, and what pronouns contribute to the interpretation of sentences. Within the present volume, five papers — von Heusinger, Kempson and Meyer-Viol, Naudé, Panagiotidis, Müller — present semantic and/or syntactic analyses for pronouns. The following list summarises the main questions they address: –

–

–

How can we define semantic representations for the lexical entries of pronouns? How is reference constituted, and which pragmatic strategies are involved? How do syntactic and semantic phenomena interact for the interpretation of pronominals? How do pronouns contribute to the representation of sentences? What are the specific syntactic features of pronouns, and what distinguishes pronominal subclasses? What is their status within hierarchical syntactic structures?

Psycho- and neurolinguistic approaches to pronouns investigate, among others, the psychological reality of representations assumed for pronouns and their neurological implementation. In particular, the fact that the link between a pronoun and its antecedent can be based on a certain kind of syntactic configuration makes pronouns a promising topic for investigations concerned with the impairments that are involved in Broca’s aphasia. In the last decades, at least three features relevant for our topic have been discussed as characteristic for the performance of patients suffering from Broca’s aphasia: (i) the omission of functional elements, (ii) problems with certain syntactic configurations that might be related to a deficit in the representation of syntactic traces (cf. Swinney et al. 1996), and (iii) problems with the interpretation of bound pronouns (cf. Grodzinsky 1990). In particular, the latter two phenomena have been interpreted as an indication that Broca’s area is crucially involved in the task of constructing syntactic dependencies in the normal time-course (e.g. Zurif et al. 1993). The investigation of Broca’s patients’ performance might therefore help us to identify dissociable aspects relevant in the processing and interpretation of pronouns, and in particular, it can support the distinction of syntax- versus discourse-based strategies. In the present volume, two papers — Piñango and

11

12


De Roo — present neurolinguistic approaches to pronouns which pertain to the following research questions: – – –

What is the nature of the impairments in the comprehension and production of pronouns in Broca’s aphasia? What kind of deficits can be observed? Which linguistic systems or subsystems are involved? What can the deficits observed in aphasia tell us about the organisation of the unimpaired system?

4. The papers In the remainder of this chapter, we present an overview of the contributions to this volume and illustrate their interrelations. We first provide summaries of individual papers in the order they appear in the book; on this basis we briefly point out the major areas of convergence in the final subsection. 4.1 Summaries of the individual papers Heidi Harley and Elizabeth Ritter present a generalised pattern for the paradigms of personal pronouns in the languages of the world. Inspired by phonological theories that employ a geometric arrangement of features, they propose a hierarchically organised geometry of morphological features, in conjunction with the notion of underspecification and the assignment of default values. Subtrees of the geometry represent the grammaticalisation of natural cognitive categories. Features are organised into three main groups: Participant, Individuation, and Class, accounting for person, number, and classifications like gender (and other class information), respectively. Although Harley and Ritter adduce additional evidence from first language acquisition, their main concern is the fact that different languages may vary enormously, yet systematically, in the make-up of their pronominal paradigms. They make explicit how their proposal can capture the most elaborate personnumber-paradigms by fully exploiting the array of features available in Universal Grammar, as well as paradigms where only a small fraction of the features is active. They characterise the features and their geometric arrangement as innate. That is, these patterns are assumed to be provided by Universal Grammar. Accordingly, this approach makes strong predictions about possible pronominal paradigms. This means that in contrast to a lot of previous studies its claims are falsifiable and can therefore serve as a starting point for further investigations.


Michael Cysouw investigates implicational relations between different properties of pronominal paradigms, drawing on a sample of more than 230 genetically and areally diverse languages. While he takes into account inflectional paradigms as well as those manifested by independent pronouns, he restricts his analysis to ‘simple’ person-number-paradigms, leaving aside those which employ additional categories such as gender or paradigms with special values for the number category (for instance, trials and some duals). Cysouw starts off from maximally eight possible distinctions, based on different combinations of values for ‘number’ and ‘person’. Since only a few paradigms seem to exploit the full range of possible distinctions, he investigates which neutralisations occur in his sample. A main point concerns paradigms with an inclusive-exclusive differentiation, which Cysouw dubs paradigms with ‘pure person’ marking. He shows that these paradigms never neutralise ‘person’ in the singular but always distinguish the roles of ‘speaker’, ‘addressee’, and ‘other’. Pure person paradigms also tend to keep apart the reference to groups including the speaker and to groups without the speaker. In contrast, there seems to be no (positive or negative) correlation between pure person marking and the neutralisation of number. Cysouw proposes an Explicitness Hierarchy that orders pronominal paradigms with respect to the number of distinctions they neutralise. He shows that the more explicit ones tend to consist of independent pronouns, whereas neutralisations seem to occur more often in inflectional paradigms. Cysouw explains this correlation in terms of different degrees of linguistic awareness speakers have with regard to free morphemes and affixes. The study presented by Neelakshi Chandrasena Premawardhena can be seen as complementary to the first two contributions: She provides an in-depth study of a single language displaying the full range of options that are available for its speakers when they want to refer to some non-speech-act-participant. Her object of study, the Indo-Aryan language Sinhala, is particularly interesting in this context since it is spoken in a sociolinguistically complex situation in Sri Lanka involving a high degree of diglossia (she concentrates on the spoken variety). Her main findings concern the fact that the choice of a reference device in Spoken Sinhala is not only determined by categorial distinctions like [±human], but also by considerations of sociolinguistic appropriateness. When it comes to pronominal forms, speakers select one from a list of items according to morphosemantic features. In addition to 3rd person pronouns, speakers can choose to employ a nominal element (a noun such as a kinship term or a

13

14


professional title) or use no overt expression at all (zero anaphora). Crucially, however, not all speakers have access to the same inventory of forms. In particular, Chandrasena Premawardhena shows that there is a difference in the way the gender systems subdivide the pronominal paradigms used by different social groups (defined by educational level, sometimes correlated with sex). This has the effect that for instance the same lexical form is classified as [−human, +animate] in one social group and only as [+animate] in the other group, due to neutralisation of the [±human] feature. Hence, depending on sociological parameters of the discourse, certain pronouns might or might not be understood as [−human], and accordingly take on a pejorative meaning when used for a person. Helmut Weiß discusses weak indefinite pronouns like someone, something, or no-one, with a focus on negative forms, ‘n-indefinites’. He defines n-indefinites as elements which either incorporate a morpheme which is synchronically or diachronically related to a negative expression (nobody, no-one), or which are suppletive forms that substituted such indefinites diachronically. Weiß distinguishes ‘NC (Negative Concord) languages’ and ‘Non-NC-languages’, depending on whether n-indefinites are used in the scope of negation or not, and investigates the occurrence of different forms of indefinites in three kinds of contexts: normal negation (NEG), negative polarity (NPI), and positive polarity (PPI). Based on a cross-linguistic sample, he argues for the following distribution: If a language distinguishes only two kinds of indefinites, then in NC-languages, indefinites in PPI-contexts pattern with those in NPI-contexts (and are distinct from those in NEG contexts), whereas in Non-NC-languages, indefinites in NPIand NEG-contexts are the same, and are distinct from those in PPI-contexts. Discussing the status of the negative morpheme in n-indefinites, Weiß points out that the semantic import of n-indefinites is the same as that of nonnegated existentials in Non-NC-languages. He suggests a semantic analysis of n-indefinites as elements that do not contribute negation, but carry only a formal, uninterpretable Neg-feature. According to this analysis, negation is contributed by the head of NegP, while the negative morpheme marks the n-indefinite as belonging to the scope of negation, but does not carry any negative force by itself. Klaus von Heusinger discusses the contribution of 3rd person pronouns to the semantic representation of sentences. Based on an overview of philosophical and linguistic approaches to the semantics of these pronouns, von Heusinger argues for a unified analysis that allows us to assume one lexical entry for 3rd


person pronouns in different usages. His analysis is based on the notion of a salience hierarchy; objects of different kinds get their positions in this hierarchy with respect to their contextual prominence. Choice functions identify for each set of entities its most salient element in a given context (for instance for the set of women, a choice function would identify the contextually most salient woman). Drawing on these notions of salience hierarchy and choice function, von Heusinger puts forward an analysis of 3rd person pronouns as terms that refer to the most salient entity with a particular property P. The predicate P is, for instance, identified as ‘female’ for a pronoun like English she, or as ‘male’ for he. Hence, we can regard this predicate as the semantic part of the morphosemantic features underlying pronominal paradigms; P is the form in which these features enter semantic representations. According to this analysis, the relevant salience hierarchy for a pronoun like she is that for female persons. An antecedent like Rose could contribute to this hierarchy, updating the respective choice functions so that they will yield Rose as the most salient woman in the given context. In non-anaphoric usages of pronouns, only non-linguistic factors contribute to the salience hierarchy, for instance, the most salient female person could be a woman that was just pointed out. Ruth Kempson and Wilfried Meyer-Viol present an analysis of anaphoric personal pronouns and relative pronouns that accounts for the derivation of the semantic and syntactic representations for the constructions these pronouns enter, and the way these representations establish a link between a pronoun and its antecedent. Their analysis is situated in their framework of ‘Dynamic Syntax’, which accounts for natural language understanding as a process of building up an interpretation (formalised within a typed lambda-calculus) based on tree structures that are updated in a way that follows the sequence of words in an utterance. Hence, this model combines syntax and semantics in one structural component as the dynamic projection of progressively enriched (partial) trees. Within this approach, nodes in a tree can be initially unfixed (e.g. in instances of left dislocation) and bring with them requirements that can be fulfilled later. Among others, this means that the model does not require any traces for the representation of relative pronouns. Kempson and Meyer-Viol account for the referential dependency of anaphors and relative pronouns by introducing pronouns as underspecified elements that contribute a metavariable to the semantic representation. In the process of tree growth, this variable is substituted by a copy of a selected term, the antecedent.

15

16


Crucially, this substitution is a pragmatic process restricted only by syntactic locality considerations that rule out certain formulae as possible copies. Such a fixing of nodes could, hence, be supported by salience hierarchies as assumed by von Heusinger. Discussing data from English, Arabic and Hebrew, Kempson and Meyer-Viol show how phenomena like resumptive pronouns and crossover in relative clauses can be accounted for within this approach. Jacobus A. Naudé discusses an unusual construction which exists in a number of Semitic languages: Some clauses lack a verb, but have a (seemingly pleonastic) 3rd person pronoun in addition to the subject. (19) gives an illustration from Qumran Hebrew (Naudé’s example (5b)): (19) hqryh hyãã yrwšlm. the.city pron.3sg.fem Jerusalem ‘The city is Jerusalem.’

Naudé investigates the status of the pronoun in such constructions in Qumran Hebrew. He provides evidence suggesting that in Qumran Hebrew this pronoun is not generated freely, but is obligatory in verbless sentences with a definite or ‘specificational’ NP in predicate position. He argues that the pronoun is neither a suppletive form of the copula nor a resumptive pronoun, and suggests an analysis of pronouns in these constructions as subject clitics that support agreement features and thus yield grammatical (verbless) clauses. In particular, he argues that the pronoun insertion in these cases is a Last Resort strategy necessary to prevent the sentence from being ill-formed. According to this analysis, the pronoun marks the sentence as specificational, thereby indicating that there is a relation between two argument positions; this triggers the generation of a well-formed predicate-argument structure necessary for the interpretation of the sentence. Such an analysis, then, suggests an extreme case of lexical reduction for pronouns: In verbless clauses of the kind Naudé discusses, pronouns seem to not pick out a referent anymore, but are reduced to the function of mere sign posts that indicate the presence of well-formed argument positions. Phoevos Panagiotidis is concerned with the syntactic categorisation of personal pronouns. He argues against an account of personal pronouns as intransitive determiners, drawing on two kinds of evidence. First he points out, with the example of Thai and Japanese, that pronouns in some languages do not form a closed class and, like nouns, can be modified and can be the complement of a demonstrative pronoun. Drawing on German and French evidence, he secondly


emphasises that gender features are not contributed by determiners but by nouns, and are realised on a determiner via agreement. Based on this discussion, Panagiotidis suggests a unified account of pronouns and articles as transitive determiners with a nominal complement. According to this analysis, the complement of pronouns is an ‘elementary noun’ that does not denote a concept but contributes a categorial feature N and morpho-semantic features like gender or politeness. In languages like English and German, this elementary noun is usually phonetically null (however, it can be overtly realised as one in English); in languages like Thai and Japanese the pronoun itself is analysed as the elementary noun, while the determiner can be phonetically null. Gereon Müller examines personal pronouns in German, distinguishing ‘strong’, ‘unstressed’, ‘weak’ and ‘reduced’ pronouns. While the elements of the four subclasses can be characterised by different phonological and semantic properties (namely [±stress], [±reduced], and [±animate]), Müller shows that the relevant classification can also be achieved on purely syntactic grounds. He argues that the classes form a hierarchy of strength that allows implicational generalisations: If a given pronoun has a particular syntactic property (namely, ‘must undergo Wackernagel movement’, ‘permits R-pronoun formation’, ‘cannot undergo coordination’, ‘cannot undergo topicalisation’), then all weaker pronouns share this property, too. Müller gives an account for the syntactic properties of the different pronominal subclasses within the framepersonal pronouns + stress (‘strong’)

– stress + animate (‘unstressed’)

– animate – reduced (‘weak’)

+ reduced (‘reduced’)

undergo Wackernagel movement RPF possible RPF obligatory

Figure 4.Characterisation of pronouns via semantic and phonological vs. syntactic properties.

17

18


work of Optimality Theory, based on a ranking of constraints that relates to this hierarchy. Figure 4 contrasts the semantic and phonological properties of pronominal subclasses with the syntactic characterisation Müller suggests (subclasses are ordered from left to right with respect to strength according to the hierarchy he assumes; ‘RPF’ stands for ‘R-pronoun formation’).8 Maria Mercedes Piñango analyses the mechanisms linking a pronoun with its antecedent in view of their neurological underpinnings, drawing on experimental evidence from on-line and off-line studies on the interpretation of 3rd person pronouns and reflexives in Broca’s aphasia. She argues that the performance of patients suffering from Broca’s aphasia supports the distinction of coindexation as a syntactic process based on binding relations, and coreference as a discourse-level process. While both mechanisms can establish the link between a pronoun and its antecedent, coindexation is obligatory for reflexives, whereas personal pronouns can undergo either coindexation or coreference. Piñango shows that coindexation relies crucially on the integrity of Broca’s area, while coreference is unimpaired in Broca’s aphasia, suggesting that this second mechanism is not syntactically governed but belongs to an independent module that is presumably part of the discourse level. She gives a unified account of the patterns observed in Broca’s aphasics in terms of a slowed-down syntactic processor that prevents the construction of syntactic structure within the normal time-course (Slow-Syntax Hypothesis). According to this account, Broca’s patients base their interpretation of personal pronouns on coreference, even in constructions where coindexation should take place, while in the case of reflexive pronouns (where coreference is not an option), the system waits for the slow syntactic tree to emerge, accounting for the difference in on- and off-line performance for reflexives (but not personal pronouns). While Piñango discusses the distinction of discourse processes and syntactic processes in the interpretation of pronouns, Esterella De Roo presents evidence for a similar distinction in the production of pronouns. In particular, she argues that pronoun omission in agrammatic aphasia does not result from a specific syntactic impairment, but reflects the overuse of a pragmatically driven option that is also available in normal grammar. She bases her argument on an investigation of German and Dutch aphasic speech by Broca’s patients who were diagnosed as agrammatic. De Roo’s analysis of the production data suggests that the pronoun omission in the


speech of these patients follows a similar pattern as that in non-impaired speech, where in certain contexts pronouns can be dropped if the interpretation can be discourse-based (this is illustrated, for instance, in (4) above). De Roo argues that in agrammatic speech, this option is overused in order to reduce the processing load of an utterance. According to this account, the impairment observed in agrammatic aphasia is not due to a lack of syntactic knowledge but to a limitation in the capacity to process syntactic information. As a result, agrammatic patients rely more on non-syntactic, discourse information in their production of pronouns than non-aphasic speakers do. This overuse of discourse strategies in agrammatic speech emphasises the availability of these strategies (in addition to syntactic strategies) in the unimpaired linguistic system. 4.2 Synopsis of main points of convergence The papers in this volume offer a kaleidoscope of studies united by the common topic of pronouns as a domain of language that exemplarily shows the interaction of different components responsible for computational (syntactic and semantic), lexical, and discourse-pragmatic processes. The different contributions converge on (at least) two main points, one concerning patterns in the make-up of morphological paradigms, the other touching upon the relationship between syntax and semantics/pragmatics. A common concern of the typologically oriented contributions are the implicational relations that hold within morphological paradigms (Harley and Ritter: relations between nodes in a hierarchically organised feature geometry; Cysouw: impact of inclusive/exclusive distinction on person or number (non-)neutralisations; Weiß: possible polarity contexts for indefinites and n-indefinites). Several of the papers emphasise the role of discourse-pragmatic (and semantic) strategies that complement syntactic processes. This concerns, for instance, the introduction or omission of pronouns (Kempson and Meyer-Viol: resumptive pronouns in relative clauses; De Roo: phenomena like topic drop); the way a link between a pronoun and its antecedent is established (Kempson and Meyer-Viol: fixing of syntactic-semantic tree nodes; von Heusinger: recourse to updated salience hierarchies; Piñango: coreference vs. coindexation under binding conditions), and the way a pronoun picks out an entity (von Heusinger: choice functions; Chandrasena Premawardhena: sociolinguistic factors).

19

20


Notes 1. For an account of the status of ‘respect’ and its relation to person, number, and gender, cf. Simon (in press), who discusses the emergence of ‘respect’ as a grammatical category in the German pronominal paradigm. 2. Cf. the contributions in Frajzyngier and Curl (2000) for a cross-linguistic overview of phenomena pertinent to constructions with reflexive pronouns. 3. Cf. also Bhat (2000) for cross-linguistic evidence from interrogative and indefinite pronominal contexts supporting this analysis. 4. Cf. Reinhart (1991) for an introduction. 5. Under a strict reading of the pronoun, Karen is the only person who takes Karen’s cat to the beer garden; under a sloppy reading, nobody takes their respective cats to beer gardens, except for Karen. — Similarly, the following joke makes use of the difference between these two readings: Karen to her husband: “John always kisses his wife good-bye when she leaves for the office. Why do you never do that?” Husband: “But I hardly know his wife.” 6. Cf. Chomsky (1981) vs. Chomsky (1995). Cf. also Jackendoff (1992) for a rejection of a syntactocentric view of binding; Reuland (2001) for a recent discussion of syntactic and semantic issues involved in binding. 7. For example Levinson (2000: Ch. 4), Huang (2000). 8. An interesting side aspect here is the status of the semantic feature [±animate] that distinguishes ‘unstressed’ pronouns, which cannot undergo R-pronoun formation, from ‘weak’ pronouns, which optionally undergo R-pronoun formation. In German, pronouns are not lexically specified for [±animate], the way for instance English third person pronouns are specified for the [±human] feature (he/she vs. it); the interpretation is context-dependent. Accordingly, the same pronoun can count as [+animate] or [−animate], depending on the object it picks out. For instance, er (‘he’) in our salad examples (6) and (7) counts as [−animate] with an antecedent Himbeeressig (‘raspberry vinegar’) in (6), but as [+animate] in the preferred reading of (7) where Charles is its antecedent. As Müller’s discussion illustrates, these contextually given, interpretational differences of pronouns can go together with the (non-)availability of syntactic operations.

References Bhat, D. N. S. 2000. “The indefinite-interrogative puzzle”. Linguistic Typology 4: 365–400. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Frajzyngier, Z. and Curl, T. (eds). 2000. Reflexives: Forms and Functions [Typological Studies in Language 40]. Amsterdam and Philadelphia: John Benjamins. Grodzinsky, Y. 1990. Theoretical Perspectives on Language Deficits. Cambridge, MA: MIT Press.


Groenendijk, J. and Stokhof, M. 1992. Interrogatives and Adverbs of Quantification [ILLC Prepublication Series for Logic, Semantics and Philosophy of Language LP-92–14]. Amsterdam: University of Amsterdam. Forchheimer, P. 1953. The Category of Person in Language. Berlin: Walter de Gruyter. Jackendoff, R. S. 1992. “Mme. Tussaud meets the Binding Theory”. Natural Language and Linguistic Theory 10: 1–31. Hale, K. L. 1966. “Kinship reflections in syntax: Some Australian languages.” Word 22: 318–324. Huang, Y. 2000. Anaphora: A Cross-Linguistic Approach. [Oxford Studies in Typology and Linguistic Theory]. Oxford and New York: Oxford University Press. Levinson, S. C. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge, MA: MIT Press. Mithun, M. 1999. The Languages of Native North America. Cambridge: Cambridge University Press. Reinhart, T. 1991. “Pronouns”. In Semantik/Semantics: Ein internationales Handbuch der zeitgenössischen Forschung / An International Handbook of Contemporary Research [Handbücher zur Sprach- und Kommunikationswissenschaft 6], A. von Stechow and D. Wunderlich (eds), 535–548. Berlin and New York: Walter de Gruyter. Reuland, E. 2001. “Primitives of binding”. Linguistic Inquiry 32: 439–492. Simon, H. J. in press. “From pragmatics to grammar: Tracing the development of ‘respect’ in the history of the German pronouns of address”. In Diachronic Perspectives on Address Term Systems [Pragmatics and Beyond. New Series], A. H. Jucker and I. Taavitsainen (eds). Amsterdam and Philadelphia: John Benjamins. Swinney, D., Zurif, E., Prather, P. and Love, T. 1996. “Neurological distribution of processing resources underlying language comprehension”. Journal of Cognitive Neuroscience 8: 174–184. Wiese, H. in press. “WH-words are not ‘interrogative’ pronouns: The derivation of interrogative interpretations for constituent questions”. In Questions: Multiple Perspectives on a Common Phenomenon [Language and Discourse Series], M. Hoey and N. Scott (eds). Liverpool: University of Liverpool Press. Zurif, E., Swinney, D., Prather, P., Solomon, J. and Bushell, C. 1993. “An on-line analysis of syntactic processing in Broca’s and Wernicke’s aphasia”. Brain and Language 45: 448–464.

21

Structuring the bundle A universal morphosyntactic feature geometry* Heidi Harley and Elizabeth Ritter University of Arizona / University of Calgary

1.

Introduction

While there is widespread agreement that both syntax and phonology should be represented by a formal hierarchical system which accounts for many crosslinguistic generalizations, morphology has, in general, not received similar treatment until recently. At least since Greenberg’s important typological work in the 1960s, it has been recognized that person, number and gender features are systematically organized cross-linguistically. Most morphological theories, however, do not address this fundamental observation. Ritter and Harley (1998) have proposed that morphological features are organized in a feature geometry, and that this explains both the observed regularities, and the possible variations, in the organization of such features. Our assumption that this geometry is provided by Universal Grammar makes strong predictions about both the possible syncretisms in person paradigms and the acquisition of personal pronouns. In the first part of this paper, we outline the basic proposal and briefly discuss how an acquisition study by Hanson (2000) supports the notion of a morphosyntactic feature geometry. We then go on to apply the proposal to several relatively complex systems of personal pronouns.

2. A morphosyntactic feature geometry In phonological theory, phonological feature geometries are used to account for the interdependencies observed among phonological features (cf., e.g., Sagey 1986, Calabrese 1988, and many others). Phonological rules (or, in more

24

Heidi Harley and Elizabeth Ritter

modern terms, optimality-theoretic phonological constraints) can make reference to features that dominate subtrees in the geometry to capture generalizations that apply to all features in the subtree — for instance, to [Sonorant] to refer to the natural class that includes nasals and liquids. Some nodes in a phonological feature geometry correspond to physical elements of the vocal tract, and hence the organization of the vocal tract can be seen as imposing structure on the geometry. Finally, the geometry predicts possible variations in the complexity of phonological systems cross-linguistically. It also predicts a child’s phonological acquisition path intralinguistically: less complex, and hence less-marked, geometries are acquired before more marked geometries. All of the reasons listed above for positing a formal phonological feature geometry obtain in the morphology of pronominal systems as well. Morphological rules make reference to certain classes of features, and not to others. All morphological features are drawn from a limited set of types, which we claim correspond to basic subparts of the human cognitive apparatus; cognition thus imposes structure on the morphology. Finally, the variation between different morphological systems is constrained in certain ways, and the acquisition path, as far as it is known, shows regularities in order that suggest a universally provided template. 2.1 Cross-linguistic patterns in morphological features Let us consider, for instance, the claims of Greenberg (1966) about typological universals in the organization of morphosyntactic features, some of which are listed in example (1): (1) Universal patterns in morphosyntactic feature organization: Universal 32: Whenever the verb agrees with a nominal subject or object in gender, it also agrees in number. Universal 36: If a language has the category of gender, it always has the category of number. Universal 37: A language never has more gender categories in nonsingular numbers than in the singular. Universal 45: If there are any gender distinctions in the plural of the pronoun, there are some gender distinctions in the singular also. Greenberg (1966)

If these universals are accurate, the general approach to morphosyntactic features evident in the literature provides no account of them. The morphosyntactic features characterizing a given form are generally grouped into

Structuring the bundle

amorphous bundles, whose potential combinations, if restricted at all, are characterized by constraints external to the representation, see e.g. Noyer (1997) for a successful example of this type of approach, employing a Universal Feature Hierarchy. Other patterns emerge in acquisition and diachronically. Some patterns in acquisition uncovered in Hanson (2000) in a survey of ten acquisition studies of typologically and historically unrelated languages are listed in (2): (2) Patterns in acquisition: a. the first pronoun to emerge is either 1st sg. or 3rd sg. neuter/inanimate b. the relative order of acquisition of 2nd person and 3rd (non-neuter) and singular and plural, varies considerably Hanson (2000)

As we will demonstrate in Section 2.3 below, these acquisition patterns lend themselves well to a geometric treatment, but pose significant problems for other approaches. 2.2 The proposal: A morphosyntactic feature geometry The geometric organization of morphosyntactic features we propose can be seen in (3) below: Referring Expression (=Agreement/Pronoun)

(3)

PARTICIPANT

Speaker

Addressee

INDIVIDUATION

Minimal Group CLASS Augmented Animate

Inanimate/Neuter

Masc. Fem

In this geometry, all nominal features are dependent upon a root node which we call [Referring Expression]. Very approximately speaking, we divide these morphological features into three groups, identified by the nodes in small caps. The Participant node and its dependents are used to represent person, specifically, 1st and 2nd person (3rd person being unmarked, cf. the discussion below). The Individuation node and its dependents Group (plural) and

25

26


Minimal (marked singular, dual in combination with Group), as well as Augmented (trial/paucal) are used to represent number systems. Finally, the Class node encodes gender and other class information. In this paper, we focus on how languages use the Participant and Individuation nodes to represent person and number and the interactions between them. We will not address the content of the Class node here. Two of the particular formal properties of the feature geometry will be important to us here. First, along the lines of the phonological geometries proposed by Archangeli (1988), and Avery and Rice (1989), we represent only positive values for features in the geometry; if a feature is not active, it is not present. Second, nodes may be underspecified, lacking dependents entirely, in which case they receive a default interpretation. Above, the default, underspecified interpretation for each of the three major nodes is the underlined dependent. The notion of a default will be important in our discussion of acquisition below. The primary division of the root Referring Expression (RE) node is into person and number features, corresponding to Participant and Individuation. It is worth noting that we exclude an explicit feature for 3rd person from the geometry entirely. First and second person, as participants in the discourse, have a status significantly different from that of third person. Some quotations from various researchers who have argued this point are listed in (4) below. (4) “‘Person’ belongs only to I/you, and is lacking in he.” Benvéniste (1971: 217) “1st & 2nd persons are personal, 3rd person is definite.” Bloomfield (1938: 225f.) “Whoever does not act a rôle in the conversation either as speaker or as addressed remains in the great pool of the impersonal, referred to as ‘third person’.” Forchheimer (1953: 5f.)

3rd person, we claim, is simply the absence of person, and represented by the presence of the Individuation node alone. The special status of discourse participants is an example of how the structure of the morphological geometry is at least partially imposed by external factors, in the same way that the physical features of the vocal tract impose some structure on the phonological geometry. Following Ritter (1997), we claim that the external factors which play a role in morphology are conceptual in nature. Specifically, notions such as deixis, countability and taxonomy constrain and motivate the relationships which are


apparent among morphological features and represented in the formal geometry. Subtrees of the geometry represent the grammaticization of natural cognitive categories, accounting for the apparent yet tenuous relationship between grammatical features and meaningful concepts. In the person geometry, 1st person is represented by a bare, underspecified Participant node, which receives a default interpretation of Speaker (in more complex systems, the Speaker node may be overtly represented, of course). Second person is represented by a Participant node with an Addressee dependent. Inclusive is represented when both Speaker and Addressee nodes appear as dependents on Participant. Number, specified by the Individuation node and its dependents, is encoded as follows: Singular is encoded by a bare Individuation node, which receives its interpretation as if it had a Minimal dependent, or (in a more complex system) by an Individuation node with an overt Minimal dependent; Plural is encoded by an Individuation node with a Group dependent. Dual occurs when both Group and Minimal are present (two being the ‘minimal group’), and trial or paucal number when Group, Minimal and Augmented are present. 2.3 Accounting for Greenbergian and acquisition universals Let us consider how the above geometry, if provided by UG, might account for some of the generalizations we saw in Section 2.1. We will need to make some additional assumptions about morphological operations, but the general research strategy should be clear. Greenberg’s universal 32, which notes that agreement for gender entails agreement for number, can be understood if we treat agreement as a Copy operation applying to the root node of the feature geometry. Any agreement which includes the Class node (distinguishing gender features) must also include the Individuation node (distinguishing number features), since Class is dependent upon Individuation. Universal 36, according to which the presence of gender categories in a language entails the presence of number categories, is similarly accounted for by the dependence of Class upon Indiv. Finally, universals 37 and 45 can be accounted for if we treat the notion of markedness as a node-counting or degree-of-embedding metric (cf. Harley 1994). Any singular representation will be less marked than a plural representation, given the geometry above, since plural requires an additional Group node in its representation. Hence, if a Class node may co-occur with Group, it follows that it may occur without Group, since a representation without Group is less marked.1

27

28


To conclude this section, let us consider how the geometry proposed above allows Hanson (2000) to account for the universals of pronoun acquisition which she enumerates. Building on the work on the acquisition of the phonological feature geometry by Rice and Avery (1995), and Brown (1997), she adopts the general constraints listed in (5) below: (5) a.

The Structure Building Hypothesis UG provides a minimal initial structure, which is added to in response to contrasts detected in the input. b. Constrained general learning path Acquisition proceeds from the top down; a given node must be acquired before its dependents. In this way the geometry captures the global uniformity apparent in acquisition. c. Free specific learning path The available paths may be elaborated in any order. In this way the geometry captures the local variability in acquisition.

UG provides the root RE node, and, in response to positive evidence, children build structure incrementally. Consider the initial pronoun acquired by the child. In Hanson’s study, 1st singular or 3rd singular inanimate is the first pronoun to appear in a child’s inventory (see (2a)). Whichever of these pronouns comes first, the other follows immediately after. Now, consider our geometry. If UG provides a default of [Speaker] for Participant, a default of [Minimal] for Individuation and a default of [Inanimate] for Class, the acquisition pattern above falls out. Any other pronoun will be more marked and hence acquired later by elaboration of structure, with the appropriate triggers. In particular, it is worth noting that other approaches to morphological universals cannot account for the early acquisition of 3rd singular inanimate pronouns. Noyer’s (1997) Feature Hierarchy, for instance, places 1st and 2nd person pronouns above 3rd, and masculine and feminine above neuter. If the early emergence of 1st person pronouns indicates the acquisition of pronouns higher in the hierarchy, 3rd singular inanimate should be among the last to be acquired. Only a geometric approach like that outlined here can provide a principled account of the early appearance of these two pronouns (for further discussion, see Hanson et al. 2000, and Harley and Ritter, in press). 2.4 Exploiting the full person/number geometry: Boumaa Fijian Before going on to discuss some paradigms with particularly interesting properties for the system we have so far proposed, let us illustrate a language


which exploits the full person/number feature geometry to familiarize the reader with the workings of the geometry. In (6) below is the paradigm for the cardinal pronouns of Boumaa Fijian: (6) Boumaa Fijian cardinal pronouns Singular plural 1st ex. yau ’eimami 1st in. — ’eta 2nd i’o ’emunuu 3rd ’ea (i) ra Dixon (1988: 54f.)

dual ’eirau ’eetaru ’emudrau (i) rau

paucal ’eitou ’etatou ’emudou (i) ratou

In (7), we illustrate each pronominal form with its associated geometry. Person distinctions are made under the Participant node, number distinctions under the Individuation node. Notice that the most geometrically complex, and hence most marked form, intra- and cross-linguistically, is the 1st inclusive paucal pronoun. (7)

sg.

1st excl. a.

b.

RE

RE

c.

RE

d. INDIV

PART INDIV PART INDIV PART Spkr

Min

paucal

dual

pl.

Spkr Group Spkr

Min

RE INDIV

PART Spkr

Gr

Min

Gr

Aug yau

1st incl. e.

Spkr

f.

RE

PART

’eirau

’eimami

INDIV

RE PART

Addr Group Spkr

’eitou

g.

RE

INDIV

Addr Min

Gr

INDIV

PART Spkr

Addr Min Aug

’eta

’eetaru2

’etatou

Gr

29

30


sg. 2nd

h.

pl.

RE

i.

dual

RE

j.

RE

PART INDIV PART INDIV PART Addr

Min

Addr Group Addr

i’o 3rd

l.

’emunuu RE

m.

RE

INDIV

INDIV

Min

Group

paucal RE

k. INDIV Min

Gr

PART Addr

’emundrau n.

INDIV Min

’emundrou o.

RE

RE

INDIV Min

Gr

Gr

INDIV Min

Gr

Aug ’ea

(i)ra

(i)rau

(i)ratou

3. Person without number: The effect of empty paradigm space In theory, a language with a full set of distinctions in the person geometry and none in the number geometry could exist, given that person and number are independent and equal nodes under RE.3 Exploiting just the features which are dependent on the person geometry, there are four logical geometric possibilities (including the unmarked Participant node), illustrated in (8): (8) Person: four logical possibilities Participant Speaker

Addressee

Participant

Participant

Addressee

Speaker

Participant

It is our claim that such languages, without an Individuation node, may use these person geometries to represent number in the first person only. 3.1 Three person geometries: Berik First let us consider the case of Berik, a language which contains no number marking on most pronouns, full nouns, or in the agreement system. This we take to indicate that the Individuation node is not active in this language;


hence the lack of number contrasts. However, it does show number marking in the first person only. If we put aside the 3rd person pronoun, which we assume is represented by a bare RE node, the paradigm contains just three forms: 1st sg., 1st pl., and 2nd person. The paradigm is illustrated in (9). (9) Berik (Nominative) singular plural 1st ai (ajam) ne (nejam) 2nd aame (ijam) 3rd je (jam) Westrum and Wiesemann (1986: 39)

We claim that, in the absence of an Individuation node, a 1st person singular/ plural contrast can be represented by the geometry usually reserved for 1st person inclusive forms. The geometries we propose to capture the Berik 1st and 2nd forms are illustrated in (10): (10)

a. 1st person pl.

b. 1st person sg.

RE

RE

RE

Participant

Participant

Parcipitant

Speaker

Addressee

c. 2nd person

Addressee

How can such an interpretation arise? Here, we claim that this is another instance in which the external nature of the group represented may influence the interpretation of a geometry. We observe of 1st person plural forms in general that they denote a mixed group consisting of the speaker and other individuals. This is in marked contrast to 2nd and 3rd person plural forms, which may denote a group of addressees or a group of other individuals, respectively. In natural language, there is no genuine 1st person plural — we never speak in choruses.4 The inclusive geometry denotes such a mixed group, consisting of the speaker and another entity. In the more usual situation, where this geometry co-occurs with features indicating number, this geometry gets its usual inclusive interpretation: the other entity is the Addressee. In the absence of number features, a language may resolve the interpretation of the more complex geometry in an unusual direction, taking advantage of the empty paradigm space: the “Addressee” node is generalized beyond its specific 2nd person role to indicate simply “another entity besides the speaker”.5

31

32


3.2 Four person geometries: Maxakalí and Kwakiutl In previous work, we discussed the facts of Maxakalí and Kwakiutl, both of which have inclusive pronouns. These languages, whose paradigms are reproduced in (11) and (12), are notable, like Berik, in that they appear to manifest number distinctions in 1st person only. Unlike Berik, however, both languages have three 1st person forms: a 1st singular pronoun, a 1st person exclusive plural pronoun, and a 1st person inclusive pronoun. (11) Maxakalí (Absolutive)

1st excl. 1st incl.

singular

plural

’u˜g

’u˜gmu˜g yu˜mu˜g

2nd ’ã 3rd ’u˜ Popovich (1986: 352) (12) Kwakiutl (Nominative)

1st excl. 1st incl.

singular

plural

-en

-enu7x» u -ens

2nd -es 3rd Boas (1911: 529)

Again exploiting the fact that 1st person plurals do not refer to a group of speakers, but rather to one speaker and either one or more addressees, or a speaker and one or more other individuals, we argue that the pronouns in these languages contrast only in person, and that in fact they do not express number at all. Support for this approach comes from the observation that these languages, like Berik, normally make no morphological number or gender distinctions on nouns or verbs. In order to account for the facts, we propose that Maxakalí and Kwakiutl make full use of the four different Participant subgeometries available in the system. These sub-geometries are depicted in (13):


(13)

1st sg. PART

2nd PART

1st excl. pl. PART

Addr Maxakalí Kwakiutl

’ug « -en

’ã -es

Spkr ’ugmug « « 7 -enu x. u

1st incl. PART Spkr

Addr

yumug « « -ens

The fact that these languages make no use of number distinctions elsewhere in the grammar strongly suggests that number features are simply not present in their feature inventories. Thus, the only way to capture the contrasts among the four distinct 1st and 2nd person pronouns is by means of person features. In order to distinguish it from the 1st person singular, the 1st exclusive plural must be represented with a dependent Speaker node6 (note that this entails a particular approach to underspecification: ‘default’ interpretations must contrast with fully specified one). 3.3 Over-riding the defaults: 2nd person inclusive Finally, we turn to a much-discussed question: that of whether or not an inclusive pronoun is ever truly a second person form, rather than a first person form. That is, since both the Speaker and Addressee nodes are active in an inclusive form, one might expect to see the possibility that the morphological shape or other properties of the inclusive pattern with second person forms rather than first person forms. In our geometry, however, the Speaker and Addressee nodes are sisters, meaning that when they are both active as in an inclusive form, it is at least theoretically possible for either to be morphologically salient. Given our approach to universal defaults, according to which first person forms in general are less marked than second, this situation would presumably be rare, requiring robust morphological evidence for acquisition. Nonetheless, it is a possibility in a system like that presented here, but not in a system which identifies the Addressee merely as [−Spkr. +Part]. We argue that robust morphological evidence for a second person inclusive7 exists, however, in three distinct languages. Consider the paradigms in (14) below, from Yokuts, Ojibwe and Nama:

33

34


(14) The marked 2nd person inclusive/exclusive distinction: a. Yokuts pronouns Singular Dual 1st na’ na’ak’ 2nd incl. * mak’ 2nd excl. ma’ ma’ak’ 3rd ’ama’ ’amak’ Newman (1944: 127)

Plural na’an may ma’an ’aman

b. Ojibwe pronouns Singular Plural 1st n-iin n-iin-awint 2nd incl. * k-iin-awint 2nd excl. k-iin k-iin-awaa 3rd w-iin w-iin-awaa Schwartz and Dunnigan (1986: 296) c.

Nama pronouns8 Singular fem 1st tita 2nd incl. * 2nd excl. sas 3rd //îs Hagman (1977: 44)

Dual

Plural

masc

fem

masc

fem

masc

tita * sats //îb

si-m sa-m sa-ro //î-ra

si-kho-m sa-kho-m sa-kho //î-kha

si-se sa-se sa-so //î-di

si-ge sa-ge sa-go //î-gu

Note that in each case, the inclusive form patterns morphologically in the prefix with the 2nd person forms. Despite the shape of these pronouns, some theorists, including Zwicky (1977) and Noyer (1997), have argued that this similarity is not evidence that the inclusive form is 2nd person. However, Déchaine (1999) gives two convincing syntactic arguments that, in fact, the inclusive forms for Ojibwe in (14b) should be analyzed as 2nd person. First, the 2nd inclusive form may be used in the imperative, which is otherwise restricted to 2nd person forms. Second, there are two types of argument agreement on the Ojibwe verb, direct and inverse. The direct form occurs when the subject is 2nd person and the object is 1st person, and the inverse form occurs when the object is 2nd person and the subject is 1st person. The pattern is laid out in the table in (15). Crucially, the inclusive pronoun triggers the same agreement pattern as


the 2nd person (exclusive), rather than the 1st person. That is, when the subject is an inclusive form, the marking is direct, as it is when the subject is a second person pronoun, and when the inclusive is the object, the inverse form occurs, as when the object is a second person pronoun.9 (15) Verb forms in Ojibwe

1st object 2nd object

1st subject

2nd subject

~ inverse

direct ~

In at least these two respects, then, it is the active Addressee node that is the crucial element of the pronoun to the morphosyntax, rather than the Speaker node. We consider that these paradigms constitute robust evidence for a separate Addressee feature.

4. Conclusions Evidently, this proposal represents a research program barely past the beginning stages. However, we hope to have demonstrated at least that generalizations about morphosyntactic feature groupings may be treated in a principled formal system, and that a deeper level of explanation for such phenomena than is usually presented is possible. With respect to the specifics of our proposal, we argued for the following points: 1. Morphosyntactic features are arranged in a formal geometry whose shape is partially constrained by their conceptual content. 2. The structure of the geometry constrains acquisition. 3. Some variation is allowed in that particular geometries may be mapped onto different subparts of conceptual space in different languages, as long as a) the mapping does not conflict with the conceptual content of the geometries, and b) the overall paradigm space allows it. 4. Both a Speaker feature and an Addressee feature are necessary to capture the range of variation in person paradigms cross-linguistically. Considerably more extensive investigation is needed to test the particular geometry we have argued for here. However, we hope to have demonstrated that the project of developing a robust morphosyntactic feature geometry is both feasible and explanatorily useful.

35

36


Notes * We would like to thank Rebecca Hanson, Andrew Carnie, Horst Simon, Heike Wiese, Phoevos Panagiotidis, Rolf Noyer, Diana Archangeli and the participants in the Workshop for valuable discussion and suggestions. This work was supported in part by Social Sciences and Humanities Council of Canada grant number 410–98–0328 to Ritter. All remaining errors are, of course, our own. 1. Horst Simon (p.c.) draws our attention to work which attacks the validity of these last two Greenbergian universals, Plank and Schellinger (1997). For example, in Spanish, the nominative and prepositional case forms of 1st pl. pronouns are marked for gender, while the singular pronouns are not. On closer inspection we find that, unlike other members of the paradigm, these pronouns are bi-morphemic, nos + otros/otras and vos + otros/otras ‘we + others(m/f)’ and ‘you + others(m/f)’ respectively. Thus, these pronouns consist of one morpheme expressing person and number, but not gender (nos or vos), and a second morpheme expressing number and gender, but not person (otros or otras). It is the combination of these two morphemes which gives rise to this unexpected feature specification. Crucially, these Spanish pronouns do not contain a morpheme whose feature content consists of 1st or 2nd person and gender, but not number. The systems of Windesi and Wandamen (Austronesian) are amenable to a similar treatment. For further discussion of these and other systems discussed by Plank and Schellinger, see Harley and Ritter (in press). 2. Note that an interesting issue arises with the 1st inclusive dual. Many traditional grammar writers automatically label the 1st inclusive a “dual” form — even without morphological evidence — as it must minimally refer to two persons. However, in the present treatment, it is conceivable that a 1st inclusive singular form could exist, as the person and number markings are not interdependent. Such a form, while referring to a group of two, can be associated with the singular (‘minimal’) number geometry. Here, however, given that the morphological shape of this particular 1st inclusive form shares elements with the other duals, we treat it as having an explicit morphological dual marking. Cf. the discussion in Cysouw (this volume), and for an example of a singular inclusive (in Kalihna), see Harley and Ritter (in press). 3. Presumably, the reverse situation could not exist, for reasons having nothing to do with feature geometry: A language cannot exist without speakers and addressees. 4. A reviewer points out, almost never. 5. Diana Archangeli (p.c.) points out that there is a natural analogue to this situation in the phonological literature. Consider the case of a phoneme that is specified for [+low] and [+front] features simultaneously. The [+low] feature pushes the tongue body back, meaning that it conflicts somewhat with the [+front] feature. Depending on the structure of the rest of the phonological inventory, different languages will resolve this conflict in different ways, resulting in different phonetic realizations of the same geometry. For an alternative treatment, see Harley and Ritter (in press). 6. This system makes a prediction that a language which makes a five-way distinction in 1st and 2nd person, but yet has no number marking in the 3rd person should not exist, as there are only four possible geometries available using only the Participant node. Unfortunately, this prediction appears to be problematic. The description of Guaraní pronouns given by


Croft (1990: 111) appears to be exactly that: five personal pronouns including 1st sg., 1st incl., 1st excl., 2nd sg. and 2nd pl., exist in the language, but there are no number distinctions in third person. We have discovered, however, that there does appear to be a plural morpheme in the language, exemplified below: (i) pe mit½a h-as¾e-má hikw½ai that boy he-cry-all pl ‘Those boys are all crying.’ Gregores and Suárez (1967: 155) For further discussion of a similar paradigm in Koasati, see Harley and Ritter (in press). 7. Horst Simon brings to our attention that the term “2nd person inclusive” has been used elsewhere (e.g. Plank 1985, Mel’cˇuk 1994, Simon 2001) to indicate “Addressee + Others”, in contrast to a “2nd exclusive” that indicates “Addressee + Addressee”. We use the term rather to refer to an inclusive pronoun with the Addressee + Speaker reference where the Addressee is the morphologically salient category. Simon (2001) argues that the “Addressee-only” vs. “Addressee + Others” type of 2nd inclusive/exclusive distinction is not in fact attested, a result which conforms to the predictions of our geometry: we can encode no such contrast. 8. // denotes a lateral click in these examples. 9. Moreover, neither the direct nor the inverse agreement form is acceptable when the noninclusive argument is 2nd person; this is presumably for the same reason that two first person arguments or two third person arguments specified on the same verb are ungrammatical: the reflexive form is required. We attribute this ungrammaticality to a constraint against the overlapping syntactic reference of the 2nd person (excl.) and the inclusive form, along the lines of Guéron’s (1984: 44) Nondistinctness Constraint.

References Archangeli, D. 1988. “Underspecification in phonology”. Phonology 5: 183–207. Avery, P. and Rice, K. 1989. “Segment structure and coronal underspecification”. Phonology 6: 179–200. Benvéniste, E. 1971. Problems in General Linguistics. Miami: University of Miami Press. Bloomfield, L. 1938. Language. New York: Holt, Rinehart and Winston. [Reprinted 1984. Chicago: University of Chicago Press.] Boas, F. 1911. “The Kwakiutl Indian language”. In Handbook of American Indian languages, 427–557. Reprinted 1971. Seattle, WA: The Shorey Book Store. Brown, C. 1997. Acquisition of Segmental Structure: Consequences for Speech Perception. Ph.D. Diss., McGill University Montreal. Calabrese, A. 1988. Towards a Theory of Phonological Alphabets. Ph.D. Diss., MIT, Cambridge, MA. Croft, W. 1990. Typology and Universals. Cambridge: Cambridge University Press. Cysouw, M. this volume. “‘We’ Rules: The impact of an inclusive/exclusive opposition on the paradigmatic structure of person marking”.

37

38


Déchaine, R. 1999. “What Algonquin morphology is really like: Hockett revisited”. In Papers from the Workshop on Structure and Constituency in Native American Languages [MIT Occasional Papers in Linguistics 17]. L. Bar-el, R. Déchaine and C. Reinholtz (eds), 25–72. Cambridge, MA. Dixon, R. M. W. 1988. A Grammar of Boumaa Fijian. Chicago: University of Chicago Press. Forchheimer, P. 1953. The Category of Person in Language. Berlin: Walter de Gruyter. Greenberg, J. H. 1966. “Some universals of grammar with particular reference to the order of meaningful element”. In Universals of Language, J. H. Greenberg (ed.), 73–113. Cambridge, MA: MIT Press. Gregores, E. and Suárez, J. A. 1967. A Description of Colloquial Guarani. The Hague: Mouton. Guéron, J. 1984. “Inalienable possession, PRO-inclusion and lexical chains”. In Grammatical Representations, J. Guéron, H.-G. Obenauer and J.-Y. Pollock (eds), 43–86. Dordrecht: Foris. Hagman, R. S. 1977. Nama Hottentot Grammar. Bloomington: Indiana University Press. Hanson, R. 2000. “Pronoun acquisition and the morphological feature geometry”. In Calgary Working Papers in Linguistics 22, S. Atkey, J. Carson, and M. Dobrovolsky (eds), 1–14. Calgary: Department of Linguistics, University of Calgary. Hanson, R., Harley, H. and Ritter, E. 2000. “Underspecification and universal defaults for person and number features”. In Proceedings of the 2000 Annual Conference of the Canadian Linguistic Association [Cahiers Linguistiques d’Ottawa], J. T. Jensen and G.van Herk (eds), 111–122. Ottawa, ON. Harley, H. 1994. “Hug a tree: Deriving the morphosyntactic feature hierarchy.” In Papers on Phonology and Morphology, A. Carnie, H. Harley and T. Bures (eds), 289–320. Cambridge, MA: MIT Working Papers in Linguistics. Harley, H. and Ritter, E. in press. “Meaning in morphology: Motivating a feature-geometric analysis of person and number”. Language. Mel’cˇuk, I. 1994. Cours de morphologie générale (théorique et descriptive), II: Significations morphologiques. Montréal: Presses Universitaires de Montréal, Centre National de la Recherche Scientifique. Newman, S. 1944. The Yokuts Language of California. New York: The Viking Fund. Noyer, R. 1997. Features, Positions, and Affixes in Autonomous Morphological Structure. New York: Garland. Plank, F. 1985. “Die Ordnung der Personen”. Folia Linguistica 19: 111–176. Plank, F. and Schellinger, W. 1997. “The uneven distribution of genders over numbers: Greenberg nos. 37 and 45”. Linguistic Typology 1: 53–101. Popovich, H. 1986. “The nominal reference system of Maxakalí”. In Pronominal Systems, U. Wiesemann (ed.), 351–358. Tübingen: Narr. Rice, K. and Avery, P. 1995. “Variability in a deterministic model of language acquisition: A theory of segmental elaboration”. In Phonological Acquisition and Phonological Theory, J. Archibald (ed.), 1–22. Hillsdale, N. J.: Lawrence Erlbaum. Ritter, E. 1997. “Agreement in the Arabic prefix conjugation: Evidence for a non-linear approach to person, number and gender features”. In Proceedings of the Canadian Linguistic Association, L. Blair, C. Burns and L. Roswell (eds), 191–202. Calgary: Department of Linguistics, University of Calgary.


Ritter, E. and Harley, H. 1998. Meaning in morphology: A feature-geometric analysis of person, number and gender. Paper presented at the 1998 meeting of GLOW, University of Tilburg, Tilburg, The Netherlands. Sagey, E. C. 1986. The Representation of Features and Relations in Non-linear Phonology. Ph.D. Diss., MIT, Cambridge, MA. Schwartz, L. J. and Dunnigan, T. 1986. “Pronouns and pronominal categories in Southwestern Ojibwe”. In Pronominal Systems, U. Wiesemann (ed.), 285–307. Tübingen: Narr. Simon, H. 2001. Only you? On the alleged inclusive-exclusive-distinction in the second person plural. Paper presented at the Fourth Meeting of the Association for Linguistic Typology (ALT IV), Santa Barbara, 2001. Westrum, P. N. and Wiesemann, U. 1986. “Berik pronouns”. In Pronominal Systems, U. Wiesemann (ed.), 37–45. Tübingen: Narr. Zwicky, A. 1977. “Hierarchies of person”. In Papers from the Regional Meetings, Chicago Linguistics Society, W. A. Beach and S. E. Fox (eds), 714–733. Chicago: Chicago Linguistics Society.

39

‘We’ rules The impact of an inclusive/exclusive opposition on the paradigmatic structure of person marking* Michael Cysouw ZAS Berlin

1.

Introduction

Pronominal elements in language are traditionally analysed on the basis of a division into three persons and two numbers. Other categories that are found regularly, like duals or an inclusive/exclusive opposition, are often seen as additions to this pronominal core. This perspective might seem natural from the point of view of the structure of European languages, but it is rather Eurocentric to impose this analysis on the structure of other languages in the world (Forchheimer 1953; Greenberg 1963; Ingram 1978; Mühlhäusler and Harré 1990). This paper reports on a cross-linguistic study into the paradigmatic structure of pronominal elements. Specifically, the marking of person will be investigated. Person markers, either independent pronouns or pronominal inflection, are not considered as individual items but as bound in a paradigm. The structure of the paradigm determines the precise value of the person markers. One of the goals of investigating the cross-linguistic variety of grammaticalised categories in paradigms of person is to check empirically which kinds of homophony are accidental, and which are systematic — the assumption being that accidental homophony will only sporadically turn up in the world’s languages, but systematic ones will occur regularly. An important result will be to show that an inclusive/exclusive opposition in a paradigm of person is a typologically salient feature — not a secondary addition. If an inclusive/ exclusive opposition is present, many paradigmatic structures are significantly less likely to occur. I will explain these restrictions by introducing the notion ‘pure person’. Paradigms with an inclusive/exclusive opposition qualify as pure

42

Michael Cysouw

person paradigms. I will claim that pure person paradigms are specialised to mark person categories. They are the real prototypical examples of person marking, and, consequently, the inclusive/exclusive opposition is not a secondary subcategory of pronominal marking. In Section 2, the definition of the concept ‘paradigm of person’ adopted in this study is explicated. Some restrictions on its scope are necessary to keep the investigation within workable limits. The sampling method is discussed (see also the appendix) and the frame for the cross-linguistic classification will be presented. In Section 3, some general typological generalisations are discussed. I argue that the difference between accidental and systematic homophony is a continuum, not a clear dichotomy. Also in this section, the first signs are found that there is something special about paradigms that have an inclusive/exclusive opposition: these paradigms show less ‘exceptional’ paradigmatic structures. Section 4 investigates some possible correlates of the inclusive/exclusive opposition. The characteristics of paradigmatic structure that correlate significantly with the inclusive/exclusive opposition form a hierarchy. This Explicitness Hierarchy is described in Section 5. In that section, it is also noted that this hierarchy correlates with the morphological status of the paradigm. Highly explicit paradigms are more often marked with independent pronouns; barely explicit paradigms are more often inflectionally marked. This is precisely as would be expected from the perspective of grammaticalisation. However, diachronic change does not follow a unidirectional path along the hierarchy. Therefore, a different explanation is needed to extend the grammaticalisation account. Conscious awareness of the referential value of independent words (as opposed to less awareness of inflectionally marked categories) is invoked. Awareness of referential value favours more explicit marking of person categories. Section 6 summarises the central results and presents implications for future investigations.

2. Definitions and method The domain of inquiry of this investigation is the category of person. Person is defined prototypically as the linguistic marking of the speech roles ‘speaker’ and ‘addressee’. For English, this includes words like I and you. These words specifically refer to the speaker and addressee — they can not be used for any other reference. An important definitional aspect is that only such specialised marking of person is included in this study. Many linguistic items can be used to refer to

‘We’ rules

the principal actors of a speech act besides such specialised person marking. Language is an interactional device, fine-tuned to the conversational setting, and, consequently, all kinds of nouns can be used to refer to the speaker or addressee. For example, when speaking to his child, a father can refer to himself by using the noun ‘daddy’ as in Daddy is going upstairs. Or in a school-class setting, a teacher can refer to a pupil by using a proper name, as in How would John deal with this problem? For English, such usage is restricted to specific settings and registers, yet in many other languages the usage of nouns and proper names to refer to speaker or addressee is much more widespread, e.g. Thai (Cooke 1968). Nonetheless, such non-specialised person markers are excluded from this investigation. Cross-linguistically, the exclusion of these markers amounts to a rather strong restriction of the investigation. Yet, such a formal criterion is necessary to assure the comparability of linguistic marking through the very diverse structures that are found among the world’s languages. It is beyond the scope of this study to discuss the overall structure of whole languages. This inquiry focuses specifically on the structure of the paradigm. A paradigm is defined as a group of morphemes that is syntagmatically identical. The morphemes that belong to a paradigm have, so to speak, the same place within the overall structure of the language. A paradigm of person markers, then, is a paradigm that includes at least an opposition between speaker and addressee.1 For English, the person paradigm as defined by the primary opposition I versus you also includes he, she, it, we and they.2 The cases that are counted in the present paper are individual person paradigms. One language can have (and usually has) more than one such paradigm. Paradigms that mark person can be morphologically independent as well as inflectionally bound. I do not place any morphological or functional restrictions on the paradigms included in the sample, since in many languages pronominal inflection is functionally comparable to the independent pronouns of other languages.3 For some languages, pronominal inflection is probably best qualified as agreement; for others the independent pronouns are optional adjuncts. In between these extremes, there is a large variety of functional possibilities attested for person marking. It would be of great interest to make a cross-linguistic investigation of this continuum. The present investigation is only a first step towards that goal. While the method of sampling adopted here allows only for an investigation of paradigmatic structure, the resulting classification can guide any future investigation into the syntagmatical structure of person marking. As a first step in this direction, I will show that morphological status turns out to be a significant parameter of variation in paradigmatic structure (see Section 5).

43

44

Michael Cysouw

The data presented in this article are counts based on an extensive investigation of the variability of the paradigmatic structure of person marking by the present author (Cysouw 2001). The objective of that investigation was to present a survey of the variation in structure of person paradigms. Some regions or families turn out to be more interesting than others, leading to some genetic skewing in the data. However, the sample of 265 paradigms of person from 234 languages for the present study is large and genetically varied enough to qualify as a typological sound database (see the appendix for further details). Due to space limitations, I cannot discuss in detail the actual data used in this study, but only present the generalisations. I should point out that it would be rather easy to expand the number of the most common paradigmatic structures. I decided to stop collecting examples of a particular paradigmatic structure when I had about 40 cases, as this was enough to establish that the structure was clearly common among the world’s languages. This decision implies that the current database of paradigmatic structure is more variable than the actual linguistic variation among the world’s languages. For the generalisations that will be presented in this article, this methodological decision possibly results in less strong correlations and trends than would have been found using a stricter procedure of genetic sampling. It is highly improbable that the sample used for this paper leads to more significant correlations than would characterize the actual world’s linguistic diversity. In this paper, only paradigms with singular and general non-singular (i.e. ‘plural’) forms are included. All paradigms that distinguish gender, honorifics or specialised number categories (like dual, trial etc.) are excluded. However, a few paradigms with gender in the third person singular are included because otherwise some unusual paradigmatic structures would have been missing. Contrary to these omissions, some instances of the category that is traditionally called ‘first person inclusive dual’ are included in the present domain. The first person inclusive dual is an ambiguous category (Greenberg 1988). In some paradigms it is clearly a dual form (these are excluded here), but in other paradigms it is a specific kind of person marking (these are included in this paper). The criterion that is used to discriminate between these two possibilities is the presence of other dual morphemes in the paradigm. If there are other dual person markers in the paradigm, then the first person inclusive dual is interpreted as a dual and excluded from the discussion here. In contrast, if there are no other dual morphemes in the paradigm, except for this first person inclusive dual, this morpheme is interpreted as a specific kind of person marking, called the ‘minimal inclusive’ (McKay 1978).4 A minimal inclusive is a category that

‘We’ rules

includes the present speaker and addressee (i.e. inclusive), but nothing else (i.e. minimal). That duality is epiphenomenal to this category is shown by the fact that in almost all instances, this apparent dual is the only dual in the whole language (Plank 1996: 130–131). Incorporating all these restrictions, eight referentially distinct person categories remain as the subject of the present paper. Other person categories are theoretically feasible, yet are unattested in the world’s languages (Cysouw 2001: 66–73).5 These eight person categories are shown in Figure 1. The present investigation includes only paradigms that distinguish maximally eight different morphemes, one for each of these categories. The crux of this paper is paradigms that distinguish less than these eight person categories. The main question to be dealt with is which categories are combined into the reference of a single morpheme. The eight person categories are shown here in the representational frame that will be used throughout this paper to display the structure of paradigms of person. The outline as presented in Figure 1 is chosen as a working compromise among various language-specific structures. The representational frame is kept identical throughout to enhance comparability, possibly at the expense of language-specific preferences. singular

non-singular 1+2 1+2+3

minimal inclusive augmented inclusive

speaker

1

1+3

exclusive

addressee

2

2+3

second person plural

other

3

3+3

third person plural

inclusive

first person plural

Figure 1.The paradigmatic frame with the eight basic person categories.

3. Typological characteristics: Accidental vs. systematic homophony Not all the paradigmatic structures which are theoretically possible are attested among the world’s languages. In the present sample of 265 paradigms, 62 different paradigmatic structures are attested. These 62 different paradigms include each and every paradigmatic structure that I have been able to find.

45

46

Michael Cysouw

This means, for example, that I have also included the perfective suffixes of Kunimaipa, as shown in Figure 2, although they show a different structure from the imperfective suffixes in the same language, and they are also different from the independent pronouns (Pence 1968: 110; Geary 1977: 26). The structure, as shown in Figure 2, is, as far as I know, unique among the world’s languages. …-ho

1+2 1+2+3

1

…-ho

…-gi

2

…-ngi

2+3

3

…-ha

3+3

1+3

Figure 2.An uncommon paradigmatic structure: Kunimaipa perfective suffixes.

Many of the 62 paradigmatic structures are only attested in one, or in a very small number of cases. A restricted set of paradigmatic structures is more widespread. Typically, such common paradigmatic structures are found throughout the world’s languages, independent of areal and genetic bonds. An example of a common paradigmatic structure is the paradigm of the present suffixes of Latin, as shown in Figure 3.

1+2 …-mus

1+2+3

1

…-o

2

…-s

…-tis

2+3

3

…-t

…-unt

3+3

1+3

Figure 3.A common paradigmatic structure: Latin present suffixes.

There is no clear-cut division between the common and uncommon paradigmatic structures. Rather, a continuous cline from common to uncommon paradigmatic structures is found. The Latin-type paradigm is one of the most

‘We’ rules

common patterns world-wide, yet it is surely not the only common paradigmatic structure. The structures of the nine most frequent paradigmatic structures are shown in Figure 4. These nine paradigmatic structures account for almost 70% of the sample. They form a rather coherent group, which I will refer to as the ‘common’ paradigmatic structures. The kinds of homophony that are found in these paradigmatic structures are clearly systematic. The other 53 paradigmatic structures that I have found (and other curious examples that probably exist, but that I missed) are classified as ‘uncommon’. However, this does not imply that all the special kinds of homophony in these paradigms are accidental. The distinction between common and uncommon is a rather ad hoc division of a continuum of frequency. The most frequent uncommon paradigm is found almost equally frequent as the least frequent common paradigm (Cysouw 2001: 158f.).

1 2 3

1 2 3

1+2 1+2+3 1+3 2+3 3+3

1+2 1+2+3 1+3 2+3 3+3

Figure 4.The nine most frequent paradigmatic structures.

A central characteristic of the structure of a pronominal paradigm is the way in which the categories of the first person plural are marked. The opposition between 1+2/1+2+3 (inclusive) and 1+3 (exclusive) turns out to be a good predictor for some other characteristics of the paradigmatic structure. A general typological observation is that the amount of paradigmatic variation is less for paradigms that have an inclusive/exclusive opposition, when compared to the paradigms that do not have this opposition. In other words, if there is an inclusive/ exclusive opposition, then much fewer uncommon structures are attested. The frequencies from the sample are shown in Table 1. From these frequencies, a statistically valid implication can be extracted (Fisher’s Exact p = .001):6

47

48

Michael Cysouw

(U1)

Inclusive/exclusive opposition Æ less uncommon paradigmatic structures

Table 1.Paradigms with an inclusive/exclusive distinction show less uncommon paradigmatic structures Inclusive vs. Exclusive No

Yes

Total

Common Paradigmatic Structures Uncommon Paradigmatic Structures

87 (32.8%) 57 (21.5%)

97 (36.6%) 24 (9.1%)

184 81

Total

144

121

265

4. The impact of the inclusive/exclusive opposition: Pure person While, generally speaking, the amount of uncommon paradigmatic structures is less if there is an inclusive/exclusive opposition, there are still many of these cases. The implication (U1) only stipulates that there is something interesting going on. In this section, I will try to unravel what causes the statistical skewing found in Table 1. The conclusion will be that paradigms can be more or less explicit in the marking of person. The inclusive/exclusive opposition is a sign of a highly explicit interest in the marking of person in the paradigm. Once this opposition is present, homophony between different persons in the paradigm is rarely found, and, consequently, much fewer uncommon paradigmatic structures are attested. Three different kinds of homophony, which I will call ‘singular’, ‘vertical’ and ‘horizontal’ homophony, are used to analyse the structure of a paradigm. The term homophony is used purely descriptively. Homophony simply means that some of the eight referential person categories (from Figure 1) are marked by the same morpheme. All kinds of homophony are treated alike — only after their distribution has been investigated there might be made a claim for a difference between accidental and systematic homophony. Singular homophony is characterised by a homophony between two of the three singular categories. For example, in the Dutch present suffixes, the second and third person singular are marked identically, as shown in Figure 5. Other kinds of singular homophony that are attested are combinations of the first and

‘We’ rules

second person (as, for example, in the English present suffixes, marked by zero) and combinations of the first and third person (as, for example, in the Spanish ‘pretérito imperfecto’, marked by -ba).

1+2 1+2+3 1 2 3

…-Ø …-t

…-en

1+3 2+3 3+3

Figure 5.Dutch present suffixes, without inversion (singular and vertical homophony).

Vertical homophony is characterised by a morpheme that marks one of the categories of the first person plural together with the second or third person plural (or both the second and third person plural). This is, for example, found for the Dutch suffix -en, as shown in Figure 5. A more interesting case of vertical homophony is exemplified by the independent pronouns from Sanuma, a language spoken on the border of Venezuela and Brazil, as shown in Figure 6 (Borgman 1990: 149). In this paradigm, a vertical homophony is attested between the second person plural and the inclusive. Note that the definition of vertical homophony is independent of the presence of an inclusive/exclusive distinction. The Sanuma pronouns show that vertical homophony is also possible with an inclusive/exclusive distinction.

49

50

Michael Cysouw

makö

1+2 1+2+3

1

sa

samakö

1+3

2

wa

makö

2+3

(classiWers)

3

3+3

Figure 6.Sanuma independent pronouns (vertical homophony).

1+2 we 1

I

3

1+3 you

2 he/she/it

1+2+3

2+3 they

3+3

Figure 7.English subject pronouns (horizontal homophony).

Finally, horizontal homophony is characterised by a morpheme that marks both a singular and a non-singular category. This is exemplified here by the English independent pronouns, as shown in Figure 7. The pronoun you is used both for the singular and the plural address: a case of horizontal homophony. Horizontal homophony is a commonly occurring phenomenon, as can be seen from the structures shown in Figure 4 above. The vast majority of cases show a ‘real’ horizontal homophony between the first person singular and plural, the second person singular and plural or the third person singular and plural. Yet, there is a small set of cases that have a ‘diagonal’ homophony between singular and plural (seven cases in the sample, which is 2.6%). This is found, for example, in the German present suffixes, as shown in Figure 8 (see p.52). In this paradigm, a homophony is found between the second person singular and the second person plural, marked by -t. Note also that a vertical homophony between the first and third person plural is found in this paradigm, marked by -en.

‘We’ rules

The presence or absence of these three kinds of homophony is now crosscut with the occurrence of an inclusive/exclusive opposition within the same paradigm. First, the inclusive/exclusive opposition is inversely correlated with singular homophony, as shown in Table 2. As can be seen in the table, paradigms with an inclusive/exclusive opposition do not show any singular homophony at all. This results in a very strong implication (Fisher’s exact p = .000): (U2)

inclusive/exclusive opposition Æ no singular homophony

Table 2.Inclusive/exclusive opposition and singular homophony Inclusive vs. Exclusive No

Yes

Total

No Singular Homophony With Singular Homophony

119 (44.9%) 25 (9.4%)

121 (45.7%) 0 (0.0%)

240 25

Total

144

121

265

The inclusive/exclusive opposition is also inversely correlated with vertical homophony, as shown in Table 3. Paradigms with an inclusive/exclusive opposition seldom show vertical homophony. However, there are 14 (5.3%) exceptions to the following implication, which is still highly significant (Fisher’s exact p = .001): (U3)

inclusive/exclusive opposition Æ less vertical homophony

Table 3.Inclusive/exclusive opposition and vertical homophony Inclusive vs. Exclusive No

Yes

Total

No Vertical Homophony With Vertical Homophony

103 (38.9%) 41 (15.5%)

107 (40.4%) 14 (5.3%)

210 55

Total

144

121

265

The inclusive/exclusive opposition implies a low occurrence of vertical homophony. Formulated differently, the absence of vertical homophony is a precondition for the presence of an inclusive/exclusive opposition. The precondition

51

52

Michael Cysouw

1+2 …-en

1+2+3

1

…-e

2

…-st

…-t

2+3

3

…-t

…-en

3+3

1+3

Figure 8.German present suffixes (diagonal and vertical homophony).

can be understood by considering vertical homophony as a kind of fused marking of the non-singular categories. This fusion should be absent for an inclusive/exclusive opposition to be feasible. Within the group of 14 exceptions to this generalisation, by far the most common variants of vertical homophony combine inclusive with second person plural (five cases) or exclusive with third person plural (four cases). These patterns can be referentially summarised by noting that the inclusion of the addressee is relevant for both the inclusive and the second person plural; oppositely, the exclusion of the addressee is relevant for both the exclusive and the third person plural. These exceptions are still neatly organised along the lines of the person categories. Yet, it is the addressee that is the crux of this marking instead of the speaker. The remaining five really exceptional cases are a collection of random kinds of homophony that can be characterised as accidental. Finally, the inclusive/exclusive opposition is not at all correlated with horizontal homophony, as is shown in Table 4 (Fisher’s exact p = .388). Statistically speaking, there is no reason to draw any conclusions from the slight skewing that is found in the table. The amount of skewing falls within the range of a random distribution. In summary, it turns out that the inclusive/exclusive opposition correlates significantly with singular and vertical homophony, but it does not correlate significantly with horizontal homophony. The two implications (U2) and (U3), that were extracted from the two significant correlations, point to the same direction. Singular and vertical homophony are preconditions for the presence of an inclusive/exclusive opposition. The person distinctions within the singular and within the non-singular have to be clearly separated before an inclusive/ exclusive distinction is possible. A mix between singular and non-singular of

‘We’ rules

Table 4.Inclusive/exclusive opposition and horizontal homophony Inclusive vs. Exclusive No

Yes

Total

No Horizontal Homophony 69 (26.0%) With Horizontal Homophony 75 (28.3%)

65 (24.5%) 56 (21.1%)

134 131

Total

121

265

144

the same person (i.e. horizontal homophony) is indifferent to the presence of an inclusive/exclusive opposition. These correlations can be understood as an effect of the essence of an inclusive and exclusive opposition. A first person plural is in fact a fusion of these two referentially highly different categories. Homophony between the inclusive and exclusive amounts to a form of indifference to the status of the addressee. In contrast, paradigms that mark an inclusive/exclusive opposition do not show this indifference; they mark ‘pure person’. Pure person marking refers to the fact that paradigms with an inclusive/exclusive opposition consequently distinguish all different referential possibilities that can be made with the two most important interactional categories: the speaker and the addressee. The reference to ‘speaker and others’ is distinguished from the reference to ‘addressee and others’, but also from reference to a group that includes ‘speaker and addressee’. The correlations as established in this section show that in a paradigm with an inclusive/exclusive distinction, the person reference is of such central importance that it is not possible to fuse person reference elsewhere in the paradigm. Only when the inclusive/exclusive opposition is not present, are other referential fusions possible.

5. The Explicitness Hierarchy The marking of pure person is better seen as a continuum between a more or less explicit marking of person. This continuum involves four different characteristics of paradigmatic structure. Three of these four characteristics have been discussed in the previous section: the inclusive/exclusive opposition, vertical homophony and singular homophony. The fourth characteristic that belongs to this continuum is the opposition between minimal inclusive and augmented inclusive. The minimal inclusive is a category that refers to the pairing of speaker and addressee.

53

54

Michael Cysouw

The augmented inclusive is a category that refers to a group that includes the speaker, addressee and others. This opposition is a subdivision of the general inclusive category. These four characteristics form a hierarchy, as shown in Table 5. At the top of this Explicitness Hierarchy, paradigms maximally distinguish all eight referential categories by different morphemes. These paradigms have a minimal/augmented opposition, an inclusive/exclusive opposition, no vertical homophony and no singular homophony. At the bottom of the hierarchy are paradigms that barely distinguish person categories. Table 5.The Explicitness Hierarchy Frequent Paradigmatic Structures

Infrequent Paradigmatic Structures

Minimal vs. Augmented Inclusive Inclusive vs. Exclusive No Vertical Homophony No Singular Homophony

+ + + +

– + + +

– – + +

– – – +

– – – –

+ – + +

– + – +

– – + –

+ + – +

+ – – +

Number of cases

26

78

99

20

21

3

12

4

1

1

(244 cases, 92% of total)

(21 cases, 8% of total)

This hierarchy can be thought of as showing more and more homophony from left to right, as exemplified in Figure 9. In the first stage, all eight person categories are distinguished. Starting from the first person complex, but invading the rest of the paradigms sequentially, more and more categories are taken together in the marking of one morpheme. In this figure, a prototypical example of each stage is shown. In the second stage of the hierarchy, the minimal and augmented versions of the inclusive are combined into one general inclusive category. Next, in the third stage, the inclusive and the exclusive are combined into a general first person plural. Only then is it possible for vertical homophony to occur. And just in case vertical homophony is present in the paradigm, singular homophony can eventually be found. Note that many possibilities for horizontal homophony exist, cross-cutting this hierarchy (not shown in Figure 9). Also, there are many more possible kinds of vertical and singular homophony than shown here in the last two stages of the hierarchy. The paradigms presented in Figure 9 are only prototypical exemplars of the various stages of the Explicitness Hierarchy. As noted in section 2, in collecting the paradigms for this investigation, I have not a priori restricted the sample to either independent pronouns or

‘We’ rules

1+2 1+2+3 1+3 2+3 3+3

1 2 3

Figure 9.Exemplars of the Explicitness Hierarchy.

inflectional ‘person agreement’. Including both kinds of paradigms in the current sample allows for an interesting conclusion. It turns out that the Explicitness Hierarchy is correlated with the morphological status of the paradigms. Highly explicit paradigms are more often independently marked; barely explicit paradigms are more often inflectionally marked, as shown in Table 6. The ‘exceptional’ cases, which fall outside the hierarchy, are more often inflectional than independent. Note that the mean percentage of inflectional paradigms over all 265 cases is not 50% but 55.8% (148 inflectional versus 117 independent). Table 6.Morphological status of the Explicitness Hierarchy Explicitness Hierarchy

Others

Minimal vs. Augmented Inclusive Inclusive vs. Exclusive No Vertical Homophony No Singular Homophony

+ + + +

– + + +

– – + +

– – – +

– – – –

Independently marked Inflectionally marked

21 5

41 37

42 57

10 10

0 21

3 18

% Inflectional

19.2%

47.4%

57.6%

50.0%

100%

85.7%

The purer the marking of person, the more likely it is that the paradigm is marked as an independent morpheme. The differences are not very obvious in the middle range of the hierarchy, but to the extremes the skewing is quite clear. The paradigms high on the Explicitness Hierarchy are almost all independent pronouns, and the paradigms low on the hierarchy are all inflectionally marked. Surprisingly, in many cases low on the hierarchy, the paradigm is the basic way to refer to an argument — it is a ‘pronominal argument’ so to speak. The same holds for those inflectionally marked paradigms, which fall outside the hierarchy (labelled as ‘Others’ in Table 6). It is not necessary for these paradigms to be a form of agreement, although it might be the case that the proportion of

55

56

Michael Cysouw

agreement is relatively high in these two classes of the hierarchy. This hypothesis has yet to be investigated in more detail. In general, the correlation between low explicitness and inflectional marking is exactly as one would expect from the perspective of grammaticalisation. A reduction in the explicitness of marking is paralleled by a reduction in morphological independence. However, there is no indication that the hierarchy describes stages in a unidirectional change, as would be predicted by grammaticalisation. Changes in both directions of the hierarchy are possible (Cysouw 2001: 197–217). Yet, if a paradigm cliticises, then it is more likely that this paradigm will subsequently be reduced rather than be enlarged. This directional priority explains, at least partly, the skewing as found in Table 6. It can explain why there are more inflectional paradigms on the lower side of the hierarchy. However, it cannot explain why there are more independent paradigms at the higher side of the hierarchy. Furthermore, grammaticalisation cannot explain why horizontal homophony is not part of this hierarchy. Speculatively, I propose the varying level of awareness that speakers have of linguistic elements as a subsidiary explanation for these observations. Independent pronouns (like all independent words) are real things to a linguistically naïve speaker of a particular language. As speakers are consciously aware of the referential properties of independent words, unusual mixtures of referential categories are disfavoured. Highly explicit paradigms are, therefore, predominantly found as independent pronouns. At the other end of the spectrum, there is a lesser degree of awareness of the inflectional morphemes themselves, which opens up the possibility for unusual combinations of referential value for such elements, possibly as the result of some historical merger (cf. the history of the Germanic inflection). Consequently, paradigmatic structures that fall outside the Explicitness Hierarchy (labelled ‘Others’ in Table 6) show a large proportion of inflectionally marked cases. Horizontal homophony is not an unusual kind of mixture, as person marking remains constant; there is only a difference in number marking.

6. Conclusion and prospects The main finding of this cross-linguistic survey is the importance of the concept of ‘pure person’ marking. For pure person marking, all possible references to speaker and addressee (in different combinations) have to be distinguished in the person paradigm. The central characteristic of pure person marking is the

‘We’ rules

opposition between inclusive and exclusive first person plural. Paradigms that distinguish between inclusive and exclusive first person plural show much less variation than other paradigms. Paradigms with an inclusive/exclusive opposition rarely allow vertical homophony; singular homophony is not found at all. The restrictions on the possible kinds of homophony can be captured in the Explicitness Hierarchy. This hierarchy is correlated with the morphological status of the paradigms. Paradigms that are highly explicit are more commonly found to be morphologically independent; paradigms that are barely explicit are more commonly found to be inflectional. Exceptions to this hierarchy also show a high percentage of inflectional cases. The correlation between the Explicitness Hierarchy and morphological status follows from grammaticalisation theory. Inflectional, i.e. highly grammaticalised, marking is less content oriented. Yet, historically the Explicitness Hierarchy does not seem to be followed unidirectionally — as would be expected from grammaticalisation theory. A subsidiary explanation for this difference in morphological status of the various paradigms can be found in the amount of conscious awareness people have of a linguistic element. Independent words are like ‘things’ to a language user. The conscious categories are structured by the difference in referential value, as this is important in the daily usage of a language. In contrast, inflectional categories are often opaque for the language user. In inflectional paradigms, a fusion of referential values is thus possible. The Explicitness Hierarchy describes a continuum of more or less rich marking of person categories in a paradigm. Interpreted this way, the Explicitness Hierarchy could be a formalisation of the concept of ‘richness’, as it is used in the pro-drop hypothesis (cf. Harbert 1995: 222). Only rich inflectional paradigms, so the hypothesis says, allow for pro-drop. Unfortunately, the notion of explicitness as developed in this paper does not seem to work as an explanation for pro-drop. Even on the lowest rung of the explicitness hierarchy, there are languages that allow for pro-drop (Cysouw 2001: 51–53). The generalisations as put forward in this paper could help to build a theoretical framework of linguistic generalisations that is informed by the diversity as attested in the world’s languages. Both the usage of features (Ingram 1978; Noyer 1992: 147–155) and the somewhat more powerful principle of feature geometries (Williams 1994: 21–29; Harley and Ritter, this volume) are attempts to capture these regularities in the world’s languages. Although nothing is wrong with such an attempt, the problem is that paradigmatic constellations that do not fit into such a framework will be described as accidental exceptions. However, it is still an open question as to whether there

57

58

Michael Cysouw

is really a distinction between accidental homophonies and structural ones, “which cannot be ascribed to paradigm inconsistency, and which are not attributable to phonological factors either” (Carstairs-McCarthy 1998: 330). From my investigations, it appears that the actual division between what counts as an accidental exception and what as a structural homophony is not clear-cut. From the generalisations put forward in this paper, a cline of predictability along the Explicitness Hierarchy can be abstracted. On top of the hierarchy is the minimal-augmented opposition.7 If there is such an opposition between minimal and augmented inclusive, then the structure of a paradigm is strongly predictable — I would classify the few deviations as real accidental exceptions. If there is no minimal/augmented opposition but still an inclusive/exclusive opposition, then there are more deviations possible, and these deviations are partly internally regular. This means that these deviations are still exceptions, yet not as clearly accidental as in the case before. Further down, the possibilities of homophony between different referential categories become greater still. From this point onward, a homophony between first and second person, for example, is not an accidental exception anymore, but a structural possibility — although it would be an accidental exception when it would be found higher up the hierarchy. I leave it as a topic for future research to propose a framework in which such gradual regularities can be accounted for. For now, I want to make a plea for more attention to the (very large) actual diversity of the world’s languages when developing a theoretical framework. A framework uninformed by the various linguistic possibilities may sound initially plausible, but is easily falsified. Finally, I will present some other distinctions that seem to pattern with the generalisations presented. First, paradigms that distinguish number marking (dual, trial, paucal) seem to follow the Explicitness Hierarchy. Vertical homophony is rarely attested when there is an inclusive/exclusive opposition in paradigms with dual marking (Cysouw 2001: 270–273). Second, gender marking is occasionally found in categories that include the speaker or the addressee. For instance, in Spanish, the first and second person plural have different forms according to gender (nosostros vs. nosotras and vosotros vs. vosotras). A more intricate example is found in the Semitic languages, which have gender in the second and third person singular and plural. In the Imperfect, the second person masculine is generally identical to the third person feminine (e.g. in Cairene Arabic, Gary and Gamal-Eldin 1982: 100). More examples of gender marking in the first and second person can be found in Plank and Schellinger (1997). A major prediction that can be derived from the pure person hypothesis is that natural gender (in the first and second person) will not be found when

‘We’ rules

an inclusive/exclusive opposition is present in the paradigm. Gender marking does not qualify a paradigm for pure person. In a pure person paradigm only person marking counts. No other referential content (like gender characteristics) is allowed to mingle with this. There can only be gender marking in the first or second person if there is no inclusive/exclusive opposition. The only apparent counterexample to this claim is the Khoisan language Nama, but the gender marking and the inclusive/exclusive marking are probably two different systems (Haacke 1977). I do not know of any other counterexamples. However, this claim is open for falsification.

Appendix The sample of the 265 paradigms that has been used for the analyses in this paper is a rather ad hoc collection of paradigms. The selection was primarily driven by the urge to collect as many as possible different paradigmatic structures. Because of this intention, some linguistic families are represented more often than others are. The 265 paradigms are collected from 234 different languages. From most languages, only one of the different available paradigms has been selected. A few languages turned out to have multiple paradigms of interest, all of which were included in this study. A few criteria have been used for the selection. First, every paradigm different from the nine common paradigms as shown in Figure 4 has been included. In contrast, I stopped collecting of these common paradigms after I had found 40 cases from all over the world. Also, paradigms with specialised number marking (dual, trial, etc.) and with (natural) gender marking in the first or second person are disregarded. Because of this decision, some linguistic families are underrepresented in this sample. Yet, the resulting collection is a good representation of the linguistic possibilities of person marking as attested among the world’s languages. It might be the case that the amount of variation in this sample is slightly larger than would have been the case if I had used a perfectly balanced genetic sample of the world’s languages. However, the claims that have been made in this paper would only have been much stronger if I had included all paradigms. The interesting variation would probably be lost in an even larger proportion of ‘common’ paradigms as shown in Figure 4. The complete set of data is described elsewhere in detail (Cysouw 2001: Ch. 4). Due to reasons of space, I give here only a list of the linguistic families that are represented in this collection to assure that the sample consists of paradigms from a well-dispersed set of languages. The 265 paradigms come from the following families (ordered roughly by geographic region): Africa:

Eurasia:

Niger-Congo (Adamawa, Atlantic, Bantoid, Dogon, Kru, Mande), NiloSaharan (Central Sudanic, Nilotic, Nubian, Saharan, Surmic), AfroAsiatic (Semitic, Chadic, Cushitic, Omotic); Indo European (Germanic, Romance, Celtic, Slavic, Iranian, IndoAryan), Uralic, Altaic (Turkic, Mongolian, Tungusic), Dravidian, NakhDagestanian, South Caucasian, Chukotko-Kamchatkan;

59

60

Michael Cysouw

Southeast Asia/Pacific: Chinese, Japanese, Tai, Ainu, Siau, Philippine Austronesian; New Guinea: Sko, Isumrud, Central and South New Guinea, East New Guinea Highlands, Binanderean, Suki, Central and South-eastern New Guinea, Sentani, Border, Nimboran; Australia: Bunaban, Nyulnyulan, Tiwi, Daly; North America: Athabascan, Siouan, Salish, Muskogean, Sahaptin, Keres, Yuman, Algonquian, Miwok, Wakashan, Eskimo-Aleut; Middle America: Uto-Aztecan, Oto-Manguean, Mixe-Zoque, Huavean; South America: Chibchan, Choco, Gé, Arawakan, Carib, Aymaran, Quechuan, TupíGuaraní, Tucanoan, Mascoian, Guiacuran, Paezan, Zamucoan, Yanumam, Murã Pirahã.

Notes * This paper has been written partly at the Institute for General Linguistics of the University of Nijmegen, and partly within the context of the project on pronominal clitics at the Zentrum für Allgemeine Sprachwissenschaft in Berlin. At those places respectively, I thank Leon Stassen and Paul Law for continued interest in these matters. Furthermore, I thank the present editors and two anonymous reviewers for their comments and additions to this paper. Special thanks go to Laura Downing for quick and thorough proofreading and for the many suggestions that substantially improved my English. 1. As a boundary case of person marking, I have included a few paradigms that have an opposition of the combination ‘speaker/addressee’ versus ‘other’, like the opposition zero versus -s in the English inflection. 2. Impersonal markers, reflexives, and reciprocals are excluded from this investigation. 3. Following Jelinek (1984), these are nowadays often referred to as ‘pronominal argument’ languages. 4. This kind of paradigmatic structure is also known as the ‘Ilocano-type’, referring to the influential analysis of the pronouns of the Philippine language Ilocano by Conklin (1962: 135) — based on original observations by Thomas (1955). An earlier account of this phenomenon, which did not have any repercussions in the literature, is found in Foster and Foster (1948) for the Mixe-Zoque language Sierra Popoluca, spoken in Mexico. 5. Basically, it is argued there that multiple occurrences of 1 (i.e. speaker) and 2 (i.e. addressee) are not necessary to describe the attested cross-linguistic variation of person marking. This means that categories like 1+1, 2+2 or 1+2+2 do not have to be included in the frame as depicted in Figure 1. See also the discussion in Studies in Language by Greenberg (1988, 1989) and McGregor (1989). 6. The statistical measure ‘Fisher’s Exact’ describes whether the distribution found differs significantly from a random distribution. For instance, in case of the distribution of Table 1, the cross-section of an inclusive/exclusive opposition and uncommon paradigmatic structures is found in 24 cases (9.1% of all 265 cases). The two parameters that are cross-sected at this

‘We’ rules

point in the table are attested respectively in 121 cases (45.7% of 265) and 81 cases (30.6% of 265). From these frequencies, it can be computed that — in a random distribution — the cross-section of these two parameters would be expected to occur in 37 cases (14.0% of 265 cases). Fisher’s Exact now tells us that the actual number of 24 cases is only to be expected with a chance of .001 — which is small enough to warrant the claim that the deviation is not due to chance. 7. Note that the terms minimal and augmented are used here in the sense of McKay (1978), not in the sense of Harley and Ritter (this volume).

References Borgman, D. M. 1990. “Sanuma”. In Handbook of Amazonian Languages, D. C. Derbyshire and G. K. Pullum (eds), 15–248. Berlin and New York: Mouton de Gruyter. Carstairs-McCarthy, A. 1998. “Paradigmatic structure: Inflectional paradigms and morphological classes”. In The Handbook of Morphology, A. Spencer and A. M. Zwicky (eds), 322–334. Oxford: Blackwell. Conklin, H. C. 1962. “Lexicographical treatment of folk taxonomies”. In Problems in Lexicography, F. W. Householder and S. Saporta (eds), 119–142. Bloomington: Indiana University Press. Cooke, J. R. 1968. Pronominal Reference in Thai, Burmese, and Vietnamese [University of California Publications in Linguistics 52]. Berkeley: University of California Press. Cysouw, M. 2001. The paradigmatic structure of person marking. Ph.D. Diss., University of Nijmegen. Forchheimer, P. 1953. The Category of Person in Language. Berlin: Walter de Gruyter. Foster, M. L. and Foster, G. M. 1948. Sierra Popoluca Speech [Smithsonian Institution Publications 8]. Washington D. C.: U. S. Government Printing Office. Gary, J. O. and Gamal-Eldin, S. 1982. Cairene Egyptian Colloquial Arabic [Lingua Descriptive Studies]. Amsterdam: North Holland Publishing Company. Geary, E. 1977. Kunimaipa Grammar [Workpapers in Papua New Guinean Languages 23]. Ukarumpa: Summer Institute of Linguistics. Greenberg, J. H. 1963. “Some universals of grammar with particular reference to the order of meaningful elements”. In Universals of Language, J. H. Greenberg (ed.), 73–113. Cambridge, MA: MIT Press. Greenberg, J. H. 1988. “The first person inclusive dual as an ambiguous category”. Studies in Language 12: 1–18. Greenberg, J. H. 1989. “On a metalanguage for pronominal systems: A reply to McGregor”. Studies in Language 13: 452–458. Haacke, W. H. G. 1977. “The so-called ‘personal pronoun’ in Nama”. In Khoisan Linguistic Studies, A. Traill (ed.), 43–62. Johannesburg: African Studies Institute. Harbert, W. 1995. “Binding Theory, control, and pro”. In Government and Binding Theory and the Minimalist Program, G. Webelhuth (ed.), 177–240. Oxford: Blackwell. Harley, H. and Ritter, E. this volume. “Structuring the bundle: a universal morphosyntactic feature geometry”.

61

62

Michael Cysouw

Ingram, D. 1978. “Typology and universals of personal pronouns”. In Universals of Human Language, J. H. Greenberg (ed.), 213–248. Stanford: Stanford University Press. Jelinek, E. 1984. “Empty categories, case, and configurationality”. Natural Language and Linguistic Theory 2: 39–76. McGregor, W. B. 1989. “Greenberg on the first person inclusive dual: Evidence from some Australian languages”. Studies in Language 13: 437–451. McKay, G. R. 1978. “Pronominal person and number categories in Rembarrnga and Djeebbana”. Oceanic Linguistics 17: 27–37. Mühlhäusler, P. and Harré, R. 1990. Pronouns and People: The Linguistic Construction of Social and Personal Identity. Oxford: Blackwell. Noyer, R. R. 1992. Features, Positions and Affixes in Autonomous Morphological Structure [MIT Working Papers in Linguistics]. Cambridge, MA: MIT Press. Pence, A. R. 1968. “An analysis of Kunimaipa pronouns”. Kivung 1: 109–115. Plank, F. 1996. “Domains of the dual, in Maltese and in general”. Rivista di Linguistica 8: 123–140. Plank, F. and Schellinger, W. 1997. “The uneven distribution of genders over numbers: Greenberg Nos. 37 and 45”. Linguistic Typology 1: 53–101. Thomas, D. D. 1955. “Three analyses of the Ilocano pronoun system”. Word 11: 204–208. Williams, E. 1994. “Remarks on lexical knowledge”. Lingua 92: 7–34.

Reference devices in Sinhala Neelakshi Chandrasena Premawardhena University of Kelaniya

1.

Introduction

This study examines various reference devices employed by Sinhala native speakers in order to keep track of participants in the discourse. Sinhala is the majority language spoken in Sri Lanka. Besides Divehi, the language of the Maldives, Sinhala is the only Indo-Aryan language geographically separated from the other members of its family, such as Hindi, Bengali, Gujarati spoken in India. Disanayaka (1998: 118f.) argues that “Sinhala occupies a unique position among the languages of South Asia because of its close affinities with two of the major linguistic families of the Indian Subcontinent, Indo-Aryan and Dravidian”. Tamil, the oldest of the Dravidian languages and Sinhala have coexisted for generations interacting with each other, thus bearing an impact on the phonology, morphology and syntax of Sinhala (Karunatillake 1984; Disanayaka 1991). There exists a vast difference between the spoken and the written variety of modern Sinhala. The Sinhala diglossia situation has made many linguists refer to the two forms of written and spoken language as ‘Literary Sinhala’ and ‘Spoken Sinhala’. The latter makes a further distinction between Formal and Colloquial varieties (see Karunatillake 1992; Disanayaka 1998; Gair 1998). Formal Spoken Sinhala is usually used in speeches, lectures or in the media and Colloquial Sinhala is the language of ordinary conversation. In this study I will refer to Spoken Sinhala, where in fact the language of normal day-to-day conversation is meant. Due to the vast differences between the two varieties of Sinhala, this study concentrates on data available from Spoken Sinhala only. The aim of the study is to discuss reference devices employed by Sinhala native speakers in discourse cohesion. The identity of the speaker (first person) and the addressee (second person) is generally evident in discourse and it does

64

Neelakshi Chandrasena Premawardhena

not cause much difficulty to decode who is being referred to. Hence the focus will be mainly on devices employed to identify third persons referred to in discourse, i.e. third person pronouns and other devices such as nouns, zero anaphora, honorific and kinship terms. It is important to note that sociolinguistic factors play a significant role in identifying referents in Sinhala (see Chandrasena Premawardhena 2001: 21f., 150–156). Social norms are as important as formal grammatical rules in this process. Thus this study will also provide a sociolinguistic interpretation of reference devices employed in Spoken Sinhala. There are a number of personal pronouns used in Spoken Sinhala (as illustrated in Table 2 below). However, due to social norms and customs such as respecting one’s elders by not addressing or referring to them with proper names or pronouns, the Sinhalese use nominal reference devices such as kinship terms or honorific terms instead. Hence this study gives an insight into both pronominal and nominal reference devices. Reference devices in Spoken Sinhala have not been the subject of attention in many extensive studies apart from Chandrasena Premawardhena (2001). In Gair (1998) and Gair and Karunatillake (2000), the focus has been mainly on the so called “neutral” pronouns, i.e. the third person pronouns used by the educated urbanites, while categorising the rest of the third person pronouns used in everyday speech by the not-so-educated rural folk as derogatory. This paper will give an insight into pronominal and nominal reference devices in Spoken Sinhala and sociolinguistic factors which determine the use of the above mentioned reference devices in day-to-day conversation. Further, various problems arising in decoding the referents such as ambiguity and how various reference devices used in Sinhala give an insight into the social status, age and sex of the referent in question and how the speaker’s attitude towards the referent is expressed through the choice of particular reference devices will be discussed. The neutral and derogatory use of the third person pronouns in Spoken Sinhala will be discussed in detail here and it will be argued that this is mainly determined by the social status, age, sex of the speaker/addressee and referent in question as well as the context of the situation and not solely by the pronouns used in discourse.

2. Reference devices in Sinhala Sinhala falls into the category of languages which Foley and Van Valin describe as one where “assignment of coreference is often determined by the subtle use

Reference devices in Sinhala

of sociolinguistic variables and is not directly signalled in the linguistic form” (1984: 324).1 There is the gender difference of animate/inanimate evident in Spoken Sinhala. However, their use in speech demands a wider knowledge of the language, its speaker and his environment, than that of only of its syntax and semantics. This system termed as inference system by Foley and Van Valin is employed in Spoken Sinhala in discourse cohesion. Honorific speech levels as well as “rules of conversational inference based on cultural knowledge” (Foley and Van Valin 1984: 324) are heavily used in Spoken Sinhala. Foley and Van Valin add further: All languages use inference to a significant degree in their assignment of coreference among NPs, but in these languages it has been elevated to the status of a fine art. These features may be correlated with the high value the cultures in which these languages are spoken place upon indirectness and ambiguity in speaking (1984: 324).

In Sinhala noun forms functioning as pronominals as well as zero anaphora, the recurrent use of proper nouns, honorific terms and kinship terms are heavily used in discourse apart from the personal pronouns (see Chandrasena Premawardhena 2001: 133–135). In the case of third person referents in discourse, Sinhala has the animate/inanimate categories of third person pronouns; the animate category being further divided into human/animal. They can be used neutrally or derogatorily depending on the speaker/addressee and the situation as well as the referent in question. In the category of human neutral, there is no gender marking based on sex (eya¯ ‘he/she’; pl. eya¯la/e¯gollo), nor in the category of non-human neutral (u¯ ‘it’; pl. u]). If the third person in question lies high in degree of respect, i.e. parents, teachers, Buddhist monks or elderly people, either zero anaphora or honorific terms such as mahatt6ya ‘gentleman/sir/master’, no¯na ‘lady’ or kinship terms are evident in discourse instead of the occurrence of pronouns. Zero anaphora is heavily used in Sinhala in reference to the speaker, the addressee as well as the third person in question. The following example illustrates how reference is established in discourse. (1) A: Siri mahatt66ya1 de] ged6r6 inn6va-d6? Siri master now home be.pres-interr ‘Is Siri at home now?’ B: næ:. (eya¯) Ø1 ta¯m6 kanto¯ru-ve. neg. (he) Ø still office-loc ‘No, he is still at the office.’

65

66


A: Ø1 kı¯y6»t6-d6 ged6r6 enne? Ø when-interr home come.fut ‘What time does he come home?’ B: Ø2 kanto¯ruv6 vahanne hat6r6-»t6. Ø office close.fut four-dat ‘They close the office at four.’ paha-»t6 vit6r6 (eya¯) Ø1 ged6r6 ¯evi. five-dat about Ø home might.come ‘He is likely to come home at about five.’ æy, Ø3 mahatt66ya-v661 hamb6-venn6 o¯n6-d6? why Ø master-acc meet-inf want-interr ‘Why, do you want to see him?’ A: ov, ma-»t6 mahatt66ya-v661 yes pron.1sg-dat master-acc t» ikak hamb6-venn6 o¯n6. a-bit meet-inf want ‘Yes, I want to meet (him) for a little (while).’ (eya¯) Ø1 hariy6»t6-m6 paha-»t6 ¯evi-d6? Ø exactly-emph five-dat might.come-interr ‘Is he likely to come at five?’ B: Ø4 kiyann6 bæ:. Ø say.inf neg. ‘(I) can’t say.’ sam6har6 davasv6l6 Ø2 en6-ko»t6 ræ: ven6va some days.loc Ø come-when night become ‘Some days when he comes it’s late.’ (eya¯v6) Ø1 hamb6venn6 o¯n6 na] Ø3 t» ikak inn6. (him) Ø meet.inf want if Ø a-little stay.inf ‘If you want to meet him, wait a while.’ (source: Gair 1998: 127f.)

In this example, ‘A’ (Referent3) wants to meet Siri (Referent1). ‘B’ (Referent4) provides him with information about Referent1. Referent1 is continuously being referred to by ‘B’ either using zero anaphora or the honorific term mahatt6ya ‘master’. During the course of the conversation ‘A’ refers to Referent1 with zero anaphora. Note that only ‘B’ who apparently lives in Referent1’s house, uses the honorific term mahatt6ya to refer to Referent1. This gives an indication that he might be an employee in this household and not of the same social level as Referent1. ‘B’ also avoids using a second person pronoun or a nominal term to


address ‘A’. He uses zero anaphora instead, which is, as mentioned above, very common in Spoken Sinhala. 2.1 The significance of sociolinguistic factors in identifying referents in Spoken Sinhala Before discussing the various nominal and pronominal reference devices in Sinhala in detail, it is important to look into various sociolinguistic factors that play a significant role in day-to-day conversation. Subtle social differences such as educated vs. not-so-educated as well as urbanites vs. rural folk are portrayed in the use of Spoken Sinhala in ordinary conversation. Whether the pronouns used are neutral or derogatory depends on the social status of the speaker/ addressee/referent. Hence it is necessary to identify the two groups in order to understand the degrees of respect implied in discourse. I will make the distinction between the educated middle class urbanites, and the not-so-educated urbanites as well as rural folk (see Disanayaka 1998: 59–62, Chandrasena Premawardhena 2001: 139f.). Table 1.Social Groups Group 1

Group 2

speaker/addressee/referent: educated urbanites/rural folk, women without age difference

speaker/addressee/referent: not-so-educated urbanites/rural folk, men among peers irrespective of any social/ educational difference (sometimes young women among themselves)

(Source: Chandrasena Premawardhena 2001: 140)

These two groups will henceforth be referred to as Group 1 and 2. There is a difference of educational level evident in day-to-day conversation. I avoid the term ‘uneducated’ since the literacy rate in Sri Lanka is above 90% (Sri Lanka Central Bank, 1999). By classifying the second group as ‘not-so-educated’ I mean the speakers who can read and write, but do not possess educational qualifications such as a secondary school certificate. The third person pronouns used by the speakers differ according to these two groups as illustrated in Table 2. 2.2 Pronominal and nominal reference devices in Sinhala Sinhala has a set of first, second and third person pronouns to identify referents

67

68


in discourse. Furthermore, kinship terms, zero anaphora, honorific terms and proper nouns are used to refer to the addressee as well as a third person. Superior kin is in general addressed or referred to by kinship terms. Due to a closely knit social unit that exists in the Sinhalese community, kinship terms such as akka ‘elder sister’, ayya ‘elder brother’, na]gi ‘younger sister’, malli ‘younger brother’ or puta¯ ‘son’ and duv6 ‘daughter’, nænda ‘aunt’, ma¯ma ‘uncle’ etc. are used not only to address or to refer to close relations in the family but also to refer to distant cousins, aunts and uncles, nephews, nieces as well as others who are not related.2 In this case as well as in reference to superiors, the use of second and third person pronouns are generally avoided. Thus, more often than not, kinship and honorific terms take the place of these pronouns. The social and educational differences discussed in Section 2.1 do not apply to the first person pronouns mam6 (1sg) and api (1pl). However, there is a general tendency among Sinhala native speakers to refer to oneself in the plural form api ‘we’ instead of the singular form mam6 ‘I’. Even if the speaker is all by himself, he generally refers to himself as api ‘we’, thus referring to his house as ape¯ ged6r6 ‘our house’, ape¯ amma/ta¯tta ‘our mother/father’ etc. This can be attributed to collective thinking and closeness of the social unit and the virtual non-existence of the individual ‘I’ in the society (see also Chandrasena Premawardhena 2001: 133f.). With regard to the role of the addressee, the social as well as the educational level of the speaker and the addressee will have to be considered.3 These differences apply to the use of both pronominal and nominal terms. If the speaker and the addressee belong to the same group (Group 1) or to two different groups (Group 1 and 2) the terms of address differ as illustrated below. i.

Speaker — Group 1, addressee — Group 1: between equals: oya¯ (2sg) oya¯la (2pl) (see Karunatillake 1992: 120); children to elders or vice versa: kinship terms — amma ‘mother’, ta¯tta ‘father’, puta¯ ‘son/daughter’, duv6 ‘daughter’; zero anaphora ii. Speaker — Group 2, addressee — Group 2: peers, children to elders or vice versa: u~b6 (2sg) u~b6la (2pl); kinship terms as mentioned above; zero anaphora iii. Speaker — Group 2, addressee — Group 1: honorific terms — mahatt6ya/ s6: ‘sir’, no¯na/mis ‘lady’; profession + honorific term — dosta¯r6 + mahatt6ya ‘doctor’; u~b6 (2sg), u~b6la (2pl) (derogatory/neutral)4 iv. Speaker — Group 1, addressee — Group 2: proper nouns (first name or surname depending on the familiarity, u~b6 (2sg), u~b6la (2pl) (derogatory)5


2.2.1 Discourse deixis and third person pronouns in Sinhala Spoken Sinhala has a four-member set of determiners indicating deictic properties. These determiners vary according to the proximity to the object in discourse in both singular and plural, irrespective of its animate/inanimate status. These determiners indicate the degree of proximity of speaker/addressee to the referent in discourse. “Sinhala third person pronouns are morphologically complex and are composed of one of four deictic elements plus a nominal element” (Gair 1998: 127). Each personal pronoun in Sinhala is thus preceded by one of these determiners whether they are first, second or third person pronouns, or possessive pronouns. Further, there is no distinction in Spoken Sinhala between the demonstrative and third person pronouns. The demonstrative pronouns function as the third person pronouns in discourse. They are further divided into animate/inanimate and human/animal categories. Gair (1998: 127) describes the four way deictic system in Spoken Sinhala as follows: i.

First Proximal: me¯ ‘this, these’: proximity to speaker or both speaker and addressee ii. Second Proximal: oy6 ‘that, those (by you)’: proximity to addressee iii. Distal: ar6 ‘that, those (over there)’: distal from both speaker and addressee, generally in sight iv. Anaphoric: ¯e ‘that, those (in question)’: reference to something in discourse Each pronoun is preceded by one of these determiners, thus indicating the proximity of speaker/addressee to the referent. They function as the demonstrative pronouns in Sinhala. Although only the fourth category preceded by ¯e is described here as anaphoric, all four proximate forms function as anaphoric in reference to a participant in discourse (see examples (2), (3), (16)). Gair/ Karunatillake (2000: 721, 729) also mention the use of “other deictic pronominal forms” for coreference in discourse “with different pragmatic effects” other than “null proform or one of the anaph forms”. Let us examine these four different proximate forms of human, non-human animate and inanimate categories in their function as demonstrative/third person pronouns. I will divide the pronouns into Group 1 and 2 as used by the two different social groups in Table 1 so that their derogatory and neutral functions are also clearly indicated. Table 2 illustrates the third person pronouns used by Sinhala native speakers. In Group 1 there is no gender marking based on sex either in human or nonhuman animate categories. The pronouns in the inanimate category are common to both groups. In Group 2 there is no difference between the

69

70


Table 2.Third person pronouns in Spoken Sinhala Group 1

Group 2

m/f

m

f

human

PROX1 PROX2 DIST APH

meya¯ pl. meya¯la oya¯ pl. oya¯la ar6ya pl. ar6y6la eya¯ pl. eya¯la/e¯gollo

mu¯, me¯ka pl. mu] o¯ka pl. o¯kuη aru¯, ar6ka pl. aruη ¯eka, u¯ pl. u], evuη

me¯ki pl. me¯kila o¯ki pl. o¯kila ar6ki pl. ar6kila ¯eki pl. ¯ekila

nonhuman animate


mu¯ pl. mu] u¯ pl. u]/o¯ka pl. o¯ku] aru¯ pl. aru] u¯ pl. u]

mu¯ pl. mu] u¯ pl. u]/o¯ka pl. o¯ku] aru¯ pl. aru] u¯ pl. u]

me¯ki pl. me¯kila o¯ki pl. o¯kila ar6ki pl. ar6kila ¯eki pl. ¯ekila

inanimate


me¯k6 pl. me¯va o¯k6 pl. o¯va ar6k6 pl. ar6va ¯ek6 pl. ¯eva



(Source: Chandrasena Premawardhena 2001: 141)

pronouns used for human and non-human animate and they are used with masculine and feminine gender marking. The PROX2 oya¯ also occurs as the familiar form of second person pronoun in Spoken Sinhala as discussed in Section 2.2. It is also noteworthy here that the speakers belonging to Group 2 do not use the above mentioned pronouns in reference to highly respected people in the society. Either honorific terms or proper nouns and zero anaphora are preferred in this context. All the above mentioned pronouns, irrespective of their various degrees of proximity to speaker and addressee, are used in discourse to keep track of references, thus implying their anaphoric function in Spoken Sinhala. This proximity could be physical or conversational. Thus, for instance, the PROX2 oya¯/o¯ka/o¯ki are used not only to refer to someone physically present near the addressee, but also to refer to someone who was just mentioned in discourse. Let us consider the following examples: (2) apoy miyuri] ] akka-d6? inj Miyurin elder.sister-interr o¯ki tamay ape¯ hen6 bo¯mbe. prox2.sg.fem emph our big bomb ‘Oh, Miyurin? She is indeed a bomb (She is a dangerous woman).’ (Kiriwattuduwa 1987: 97)


In this example Miyurin is not in the vicinity of either the speaker or the addressee. However, as she has been mentioned earlier in the conversation, the PROX2 is used to refer to her. (3) bal6-paη aru¯ ma-»t6 k6r6pu væde » ¯. look-imp dist.sg pron.1sg-dat do.past thing ‘See what that one did to me!’ (Kirwattuduwa 1987: 134)

The referent is not in the vicinity of the speaker/addressee in this example either. However, since the speaker assumes that the addressee identifies the referent, the distal form aru¯ is used here instead of anaphoric u¯ or eya¯. In Sinhala the human +/−masculine third person pronouns eya¯ ‘he/she’ (pl. eya¯la/e¯gollo), non-human animate u¯ ‘it’ (pl. u]) and inanimate ¯ek6 ‘it’ (pl. ¯eva) exist, which do not carry a negative meaning or denote disrespect. (This will be referred to here as the neutral category.) However, in ordinary conversation there is a choice of several other third person pronouns available to the Sinhala native speaker (see Table 2). The unique feature of these pronouns is that they do not carry a derogatory meaning, contrary to many arguments (see Matzel 1983: 31; Karunatillake 1992: 259; Jayawardena-Moser 1993: 26; Gair 1998: 112; Gair/Karunatillake 2000: 718). Whether they denote disrespect or not is determined by the social status of the speaker/addressee and the referent in question and the context. When these pronouns are used by the educated urbanites, they denote disrespect. On the other hand, the rural folk use them in their day-to-day conversation neutrally. Hence they cannot be categorised as “in derogatory speech they get extended to refer to humans as well” (Karunatillake 1992: 259). For example, the third person pronoun for non-human animate u¯ (pl. u]) is often used in reference to humans in Spoken Sinhala. It can denote disrespect only when used by an educated urbanite in reference to humans. However, when used by the rural folk or men among peers in reference to humans there is no derogatory meaning indicated. The following two examples illustrate how the use of ¯eki/me¯ki ‘she’ by two speakers of different social backgrounds in reference to female referents could be interpreted as neutral or derogatory. (4) bal6-ko me¯ki-ge kon» de » dig6. look-imp prox1.sg.fem-poss hair long ‘See what long hair (this one) she has!’ (Karunatillaka 1993: 125)

71

72


(5) ra¯ni hita¯gen6 æti me¯ eηg6lantey kiy6la. Rani think might dem England comp ‘Rani must be thinking this is/(we are in) England.’ ma-»t6-naη dæη e¯ki pe¯nn6 bæ pron.1sg-dat-ptcl now aph.sg.fem see.pres neg ‘I simply can’t stand her now.’ (Rajapaksa 1986: 20)

In (4) the speaker is from Group 2, from a rural background. Thus his reference to the third person in conversation by me¯ki is neutral. However, the speaker in (5) is an educated urbanite. Hence the use of ¯eki denotes disrespect. Thus, social norms, customs and traditions, age, sex, social status of the speaker/addressee/referent play a significant role in Spoken Sinhala. This is well portrayed in its pronominal system. As discussed above, a set of pronouns used neutrally by one party (Group 2) will be interpreted as extremely disrespectful when used by the educated urbanites and rural folk (Group 1) in discourse. As indicated in Table 1, the social status or educational level of younger men does not play a significant role in their choice of pronouns in speech even if they use the pronouns of Group 2 since these forms are considered as neutral when used among peers. For example let us take a part of a dialogue between two males. (6) A: nimal he»t6 en6-va kivva-d6? Nimal tomorrow come-fut say.past-interr ‘Did Nimal say that he is coming tomorrow?’ B: u¯-»t6 bæri-lu. aph/pron.sg-dat can.neg-say.ptcl ‘He says he can’t.’ (Chandrasena Premawardhena 2001: 153)

The use of u¯ in reference to Nimal (a male) does not carry any negative meaning as the conversation is between two male friends. 2.3 Respect degrees denoted by reference devices Social factors play a significant role in spoken Sinhala in identifying the referents in discourse. As discussed above, zero anaphora, proper nouns or the use of honorific terms and kinship terms to refer to referents respected in society is a marked feature. These terms are used in reference to the addressee as well as to the third person in discourse. Further, the speaker has the ability to express his respect or disrespect by the choice of second and third person


pronouns he uses in discourse. Let us take the example where the speaker says to the addressee ‘I will say that to the boss’. Note that the honorific term s6:/s6:r6 ‘master’ is a loan-word from English which is used in Sinhala both to address a superior or to refer to him as a third person. The use of s6: in example (7) could also mean ‘master’ or ‘teacher’ as it is often used as an honorific term in Sinhala in reference to ‘boss’, ‘master’ or ‘teacher’. What the speaker is going to say to the referent is not indicated here. The examples (7a), (7b) and (7c) indicate the different levels of degrees of respect expressed by the speaker by the choice of different third person pronouns in each occasion: With a mark of respect towards the referent: (7) maη gihiη Ø s66:r-t6 »6 kiyan-naη. pron.1sg go.fut Ø (it) boss-dat say-fut ‘I will tell the boss/teacher(about it).’ (7) neutral: a. maη gihiη Ø kiyan-naη. pron.1sg go.fut Ø (him) say-fut ‘I will tell (him about it).’ b. with less respect: maη gihiη Ø undæ-t6 » 6 kiyan-naη. pron.1sg go.fut Ø (it) 3sg-dat say-fut ‘I will tell him (about it).’ c. derogatory (indicating anger and disrespect): maη gihiη Ø o¯ka-t6 » 6/u¯-t» 6 kiyan-naη. pron.1sg go.fut Ø (it) pron.3sg-dat say-fut ‘I will tell him [the scoundrel] (about it).’ (Chandrasena Premawardhena 2001: 169f.)

In the case of (7b) the use of undæ (vaguely ‘fellow’ in English) in reference to an employer or a teacher does not indicate much respect. Note that undæ is a noun generally used to describe an elderly person which also functions as a pronominal in Spoken Sinhala (see also 2.3.2). This denotes disrespect when used in reference to a referent from Group 1. In (7c) the disrespect and anger of the speaker towards the referent is even more clearly expressed by the use of u¯/o¯ka as the speaker is from Group 1, an educated urbanite. When discussing the respect degrees in reference tracking in Sinhala, it is also noteworthy to mention how Buddhist priests are referred to in discourse. The honorific terms unna¯]se or ha¯muduruwo ‘reverend’ are used by the Sinhalese in reference to Buddhist monks.

73

74


(8) A: pans6l-e¯ loku ha¯muduruwa] ]-t6 » 6 a¯ra¯d6na¯.k6rann-ay api temple-poss chief monk-dat invite-fut 1pl hita] inne. think be.pres ‘We are hoping to invite the chief priest of the temple.’ B: ¯ek6 ho~da-y. ha¯muduruwo bæ: kiy6nekan-næ:. pron.3sg good-be monk can.neg say.fut-neg ‘That’s good. The priest won’t say no.’ (Chandrasena Premawardhena 2001: 171)

When elderly relatives are being referred to in discourse, mostly kinship terms are used (see Karunatillake 1992: 120). The occurrence of zero anaphora is also a common feature. (9) ta¯tta-»t6 sanı¯p6 næ:. ¯enisa¯ ta¯tta (Ø) be¯t gann6 giya¯. father-dat well neg therefore father Ø medicine take.inf go.past ‘Father is not well. (Therefore) So father (he) went to the doctor.’ (Karunatillake 1992: 120)

In reference to superiors and the highly respected in society apart from older relatives where kinship terms are used as in (9), the use of mahatt6ya ‘master/ gentleman’ or no¯na ‘lady’ is common instead of the use of third person pronouns (Karunatillake 1992: 120; Gair/Karunatillake 2000: 729). (10) ad6 dost66r66 mahatt66ya enne næ:. today doctor master come.pres neg ‘The doctor will not come today.’ ad6 ¯e mahatt66ya ven6 gam6nak yan6va. today dem gentleman another trip go.pres ‘He is going somewhere else today.’ (Karunatillake 1992: 120)

2.3.1 Zero Anaphora in Spoken Sinhala The occurrence of zero anaphora is a common feature of Spoken Sinhala. As discussed above, zero anaphora occurs in Sinhala in reference to speaker, addressee as well as the third person. Gair/Karunatillake (2000: 729) argue that “a null pronominal is generally preferred unless a lexical form is required by the structure”. The use of pronominals is not common in reference to the highly respected in society (see, for example, Chandrasena Premawardhena 2001: 167–169).


(11) A: Ø1 amma2 inn6va-d6? (your) mother be.pres-interr ‘Is (your) mother at home?’ »¯ giya¯. B: (amma) Ø2 kade (mother) Ø boutique go.past ‘Mother went to the boutique.’ A: ehenaη Ø2 kiyann6 maη ævit giya¯ kiy6la. then Ø say.imp pron.1sg come.past go.past comp ‘Then tell her that I came.’ (Chandrasena Premawardhena 2001: 165)

Here the possible recurrence of the noun form amma (Referent2) in the discourse is avoided due to redundancy. A third person pronoun such as eya¯ is also not used, as the mother is a respected person in society and thus not generally referred to by similar pronouns. However, the pronominal use of the noun undæ is often used by the not-so-educated rural folk in reference to mother or father or other elderly people where its use does not denote disrespect. Educated urbanites and rural folk generally avoid this term in reference to highly respected persons in society. Note that in the above example reference to the addressee (Referent1) is also substituted by zero anaphora. 2.3.2 Pronominal use of nouns in Spoken Sinhala In Spoken Sinhala there are some nouns functioning as pronominals in discourse. The derogatory or neutral use of them is determined once again by the social status of the speaker and the referent. If Group 1 uses them, they are generally considered derogatory. When used by Group 2, they denote no disrespect. I will mention here three such noun forms which function as third person pronouns in Spoken Sinhala: undæ (pl. und6la/undæla) ‘fellow’ — unmarked for gender, usually used by Group 2 in reference to elderly people and by Group 2 in reference to anyone to indicate disrespect; miniha ‘man’ — in reference to males, gæ:ni ‘woman’ — in reference to women, both terms used by Group 2 neutrally and by Group 1 derogatorily. The latter mentioned two pronominals generally occur in the singular form only. In reference to babies and children, the noun forms baba¯ ‘baby’, daruva ‘child’ are often used pronominally by Group 1 (see Section 2.2.3). (12) A: ko¯ daya¯? where Daya? ‘Where’s Daya?’

75

76


B: eya¯-ge ya¯luwo ka»tt» iyak ekk6 giya¯. aph.sg-poss friends some with go.past ‘(He) went out with some of his friends.’ miniha driηk » ekak ehem6 gahala passe ey. 3sg (man) drink some ptcl drink.past later come.fut ‘He will come after drinking a couple of glasses (of alcohol).’ (Rajapaksa 1986: 69)

In the above example the referent ‘Daya’ is once referred to as eya¯ and once as miniha. The speaker ‘B’ is a male in this instance and thus the use of miniha does not denote disrespect, as he belongs to Group 2, referring to one of his peers named Daya. The next example shows the use of undæ by a speaker from Group 2, from a rural area, where this term is often used in reference to elderly people. (13) sub66te¯ris ba¯ppa ¯ek6 æd6la-dey undæ-ge karatt-e]. Subateris uncle aph.sg pull-fut pron.3sg-poss cart-inst ‘Uncle Subateris will deliver it in his cart.’ (Karunatillake 1993: 85)

If this pronominal is used by Group 1, it could be used in reference to anyone irrespective of age and sex. However, it is used only derogatorily (see example (7b)). 2.3.3 Reference to children, animals and inanimate objects In this chapter a few special features of Spoken Sinhala in referring to children, animals and inanimate objects will be discussed. Cultural and social norms also play a major role in this respect (see Chandrasena Premawardhena 2001: 159–163). Children are a very significant part of Sinhalese society. Hence it is important to see how they are referred to in discourse. In this respect too, there exists a difference between the reference devices used by Groups 1 and 2. Generally the noun forms baba¯ ‘baby’ and daruva ‘child’ are used by Group 1 with no gender marking in reference to babies and children. Group 2 generally uses the terms podi » eka¯ (masc), podi » ekı¯ (fem) with gender marking meaning ‘the little one’ or the third person pronouns u¯, ¯eka, ¯eki. With reference to animals, Group 1 generally uses the gender neutral third person pronoun u¯ (pl. uη). Depending on emotional involvement, even the third person pronoun for humans eya¯ (pl. eya¯la/e¯golla) is used in reference to animals if the referent is for example a household pet. Group 2 uses the pronouns with gender marking ¯eka (masc) and ¯eki (fem) in addition to gender neutral u¯.


As shown in Table 2, the same set of third person pronouns ¯ek6 (pl. ¯eva) etc. are used by the speakers of both Group 1 and 2 in reference to objects. However, if it is an object highly respected and venerated by the Sinhalese, these pronouns are avoided. For instance, the Bo tree (ficus religiosa) or Buddha statues in a temple are objects of utmost veneration to the Sinhalese Buddhists. Thus, these objects are considered as humans and honorific terms such as the suffix (honorific) -vaha]se are added when referring to them as in the case of Buddha and Buddhist monks: bo¯dı¯n-vaha]se ‘highly respected Bo-tree’, pilimavaha]se ‘highly respected statue’. Note that vaha]se does not mean ‘holy’. Hence the translation given here as ‘highly respected’ (see Chandrasena Premawardhena 2001: 162f.).

3. Identifying referents in discourse 3.1 A question of ambiguity Due to common ground existing between the speaker and the addressee, difficulties that occur in identifying the referents in discourse are minimal. According to Clark (1996: 14) “common ground is the foundation for all joint actions, and that makes it essential to the creation of speaker’s meaning and addressee’s understanding as well.” However, there can be a few instances where a question of ambiguity occurs in Spoken Sinhala due to the fact that third person pronouns for animals (eg. u¯, ¯eka ‘it’) are also used in reference to humans. In the following case the use of u¯ for both male humans as well as animals causes ambiguity: (14) nimal isko¯le ærila ged6r6 yan6-ko»t6 Nimal school after home go-while har66k-ek passeη el6v6-la. bull-indef behind chase-past ‘While Nimal was going home after school a bull chased him/he chased a bull.’ u¯ pane¯pa¯kiy6la duv6la. aph.sg scared run.past ‘He/it ran away.’ (Chandrasena Premawardhena 2001: 176)

Since there is no subject verb agreement in Spoken Sinhala or in many occasions a fixed position for subject, object or verb (see Gair 1998: 51), this sentence could represent either of the following meanings:

77

78


While Nimal was going home after school, i. a bull chased him and he ran away or ii. he chased after a bull and it ran away. How native speakers avoid such a situation of ambiguity needs to be discussed. It could either be resolved by intonation or by using attributes which indicate whether (a) Nimal is usually known for such deeds as chasing after bulls or (b) whether he is a helpless/poor victim of an errant bull. (14) a.

ar66 vis66kodiya Nimal (…) ged6r6 yan6-ko»t6 dem rascal Nimal home go-while har6k-ek passen el6v6-la. bull-indef behind chase-past ‘That rascal Nimal has chased after a bull on his way home.’ u¯ pane¯pa¯kiy6la duv6la. aph.sg scared run.past ‘It ran away.’ (Chandrasena Premawardhena 2001: 178)

Since it is clearly indicated by the use of the word vis6kodiya » ‘rascal’ that Nimal is a boy who is capable of chasing after innocent bulls and cows he sees on his way home, the speaker indicates to the addressee that the third person pronoun u¯ in the following utterance, who ran away, refers to the bull. However, if there is any indication in the utterance signifying the helplessness of poor Nimal as in (14b), then the addressee identifies the subject u¯ as Nimal; i.e. it was the bull who chased Nimal and it was Nimal who ran away. (14) b. ar66 ahiηs66k66 Nimal koluwa ged6r6 yan6-ko»t6 dem innocent Nimal boy home go-while har6kek passeη el6v6la. bull-indef behind chase.past ‘A bull chased that poor boy Nimal on his way home.’ u¯ pane¯pa¯kiy6la duv6la. aph.sg scared run.past ‘He ran away.’ (Chandrasena Premawardhena 2001: 178)

As discussed in Section 2.3 kinship terms as well as honorific terms and zero anaphora are used in reference to both the addressee and the third person in discourse. Thus there might be difficulties arising in identifying the referent as in the case of (15).


(15) akka he»t6 kol6~b6 yan6va-d6? elder.sister tomorrow Colombo go.pres-interr ‘Is elder sister going to Colombo tomorrow?/Elder sister, are you going to Colombo tomorrow?’ (Chandrasena Premawardhena 2001: 179)

If the addressee in the above situation corresponds to the kinship term akka ‘elder sister’, there may be a question of ambiguity arising, since the term can also refer to another elder sister who is not present during the conversation. The occurrence of oya¯ in reference to 2sg (see Section 2.2) as well as PROX2 (see 2.2.1) in the case of reference to someone just mentioned in discourse can similarly cause ambiguity in Spoken Sinhala. (16) ma] oya¯-t6 »6 kivva ma-»t6 vela¯ pron.1sg pron.2/3sg/fem/masc-dat say.past pron.1sg-dat time næhæ kiy6la. neg comp ‘I told you/him/her that I don’t have time.’ (Chandrasena Premawardhena 2001: 179)

The use of oya¯ in this example could thus refer to the addressee as well as a male or a female referent mentioned earlier in discourse. 3.2 Human/non-human vs. neutral and derogatory As discussed above, it is common in Spoken Sinhala that the pronouns used for non-human animate are also used in reference to humans. Their use causes not only ambiguity as seen in (14) but also leads them to being labelled as derogatory by many linguists (see Section 2.2.1). I have argued in this paper that their neutral or derogatory sense is determined by the situation, the context and the social status, age and sex of the speaker/addressee and the referent in discourse. I have divided the native speakers of Sinhala into two groups (Table 1) and argued that the third person pronouns used by Group 2 that are considered as derogatory are, in fact, used in a neutral sense in discourse by the speakers belonging to this group. These pronouns are used in their ordinary conversation with no negative meaning attached to them. However, these same pronouns certainly carry a negative meaning denoting disrespect when speakers from Group 1, as illustrated in examples (5) and (7), use them. Thus it is not fair in my opinion, to regard these so called derogatory pronouns as such without considering their neutral use in the discourse. The majority of the

79

80


population of Sri Lanka is based in rural areas (Disanayaka 1998: 59) and their use of these pronouns in discourse in the neutral sense cannot be ignored. Table 3 illustrates the significant features of the reference devices in Spoken Sinhala (only singular forms). Table 3.Reference devices in Spoken Sinhala Group 1 educated urbanities, educated rural folk

Group 2 not-so-educated urbanities, not-so educated rural folk, men among peers

m/f

m

f

m/f

animate human

Ø, proper nouns, honoric and kinship terms, meya¿ , oya¿ , eya¿ , ar6ya

undœ, miniha, me¿ka, o¿ ka, ar6ka, e¿ka

undœ, gœ:ni, me¿ki, o¿ ki, ar6ki, e¿ki

Ø, proper nouns, honorific and kinsip terms

animate non-human

u¿

u¿ , me¿ka, o¿ ka, ar6ka, e¿ka

me¿ki, o¿ ki, ar6ki, e¿ki

inanimate

me¿k6, o¿ k6, ar6k6, e¿k6

me¿k6, o¿ k6, ar6k6, e¿k6

(Source: Chandrasena Premawardhena, 2001: 173)

While both groups use zero anaphora and proper nouns as well as honorific and kinship terms, it should be noted that according to the data available it is Group 1 that mostly uses these. The educated use the pronouns of Group 2 to denote disrespect towards the referent in discourse. Since Group 2 uses all these pronominals in a neutral sense, it is interesting to consider how they express their disrespect towards the referents. In the data available from Spoken Sinhala, it is evident that they generally use noun forms and adjectives (with negative meanings) such as bat6l6ya ‘the fat one’, pissa ‘the mad one’, yaka¯ ‘devil’ and nouns for animals bu¯ruwa ‘donkey’, balla ‘dog’ (Chandrasena Premawardhena 2001:174f). The speakers of Group 1 use the same terms too, to indicate disrespect. The difference is that they have a set of third person pronominals as well to use in a derogatory sense, whereas the speakers of Group 2 have to resort to various nouns as mentioned above to achieve the same purpose. (17) oy bu¯ruwa-»t6 ma] kivva o¯k6. dem donkey-dat pron.1sg say.past dem ‘I told that donkey about this.’ (Chandrasena Premawardhena 2001: 174).


Thus the choice of the pronouns used by Sinhala native speakers varies according to their degree of respect towards the referent. This can depend upon the context, situation or the current mood of the speaker/addressee and his relationship to the referent.

4. Conclusions In the first and second sections of this paper I have introduced the various reference devices available in Spoken Sinhala. Since I have concentrated on pronominal and nominal reference devices, I have not gone into analysing other inherent grammatical and syntactical features such as gender, singular/ plural marking, case, position of subject etc. which obviously contribute a lot to identifying the referents in discourse in any language. However, due to their significance in identifying the referents as well as the attitude of the speaker towards the referent, the role of social factors in tracking discourse participants in Spoken Sinhala was discussed in detail. Thus, I have argued that especially the speakers of Group 1 heavily use kinship terms or zero anaphora in Spoken Sinhala when addressing or referring to a superior kin. When addressing or referring to a superior other than kin, honorific terms such as mahatt6ya ‘gentleman/master’, no¯na ‘lady’ or in the case of Buddhist monks ha¯muduruwo, sva¯mı¯n-vaha]se ‘reverend’ are used. Karunatillake (1992: 120) argues that “it may be noted that specially with the younger generation the form eya¯ (3sg) is gaining currency as a colourless third person demonstrative pronoun”. It was also discussed as to how the use of different pronominal and nominal devices as well as various degrees of proximity to the referent indicates respect degrees. Through its pronominal system not only the referent, but also the distance to the referent from speaker and addressee as well as the various degrees of respect towards the referent indicated by the speaker could be identified. In the third section I have illustrated a few possibilities of difficulties that might arise in identifying the referents in discourse and how ambiguity is avoided when the same pronoun is being used to refer to humans as well as animals. Further, I have attempted to point out that a set of pronouns so far categorised as ‘derogatory’ in many studies can only function derogatorily when they are used by the educated urbanites/rural folk. I have argued that the social status, age, and sex of the speaker/addressee and referent determine the neutral or derogatory use in question as well as the context of the situation. Thus reference devices in Sinhala reveal not only who is being referred to in discourse, but also how he is

81

82


being referred to. This could also explain the heavily used zero anaphora and noun recurrence where the speaker/addressee is on safer ground than when using pronouns which may be interpreted as derogatory.

Notes 1. See also Head (1978). 2. See Disanayaka (1998: 63–66) for a detailed list of kinship terms. 3. See Chandrasena Premawardhena (2001: 136–138) for a more detailed account of reference to the addressee in Spoken Sinhala. 4. ‘mis’ is a loan-word from English (‘Miss’). However, in Spoken Sinhala it is used to refer to both married and unmarried women. 5. ‘u~b6’ used among the speakers in Group 2 does not denote disrespect. However, when used by Group 1 in reference to a person in Group 2, it is derogatory. When this pronoun is used by Group 2 in reference to a member of Group 1, the situation decides whether it is derogatory or neutral because more often than not, the speakers use it without any disrespect as they use it in neutral context among themselves and do not consider it as derogatory. 6. ‘s6:r’ is used by monolingual speakers as against s6: by bilingual speakers.

References Chandrasena Premawardhena, N. 2001. Referenz-Indizes im Deutschen und im Singhalesischen. Ph.D. Diss., Universität Siegen. Clark, H. H. 1996. Using Language. Cambridge: Cambridge University Press. Disanayaka, J. B. 1991. The Structure of Spoken Sinhala 1: Sounds and their Patterns. Colombo: Lake House Investments. Disanayaka, J. B. 1998. Understanding the Sinhalese. Colombo: Godage. Foley, W. and Van Valin, R. D. Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge University Press. Gair, J. W. 1998. Studies in South Asian Linguistics: Sinhala and Other South Asian Languages. Oxford and New York: Oxford University Press. Gair, J. W. and Karunatillake, W. S. 2000. “Lexical anaphors and pronouns in Sinhala”. In Lexical Anaphors and Pronouns in Selected South Asian Languages: A Principled Typology, B. Lust et al (eds), 715–773. Berlin: Mouton de Gruyter. Head, B. F. 1978. “Respect Degrees in Pronominal Reference”. In: Universals of Human Language. Vol. 3: Word Structure, J. H. Greenberg (ed.), 151–211. Stanford: Stanford University Press. Jayawardena-Moser, P. 1993. Grundwortschatz Singhalesisch-Deutsch: Mit grammatischer Übersicht. Wiesbaden: Harrassowitz.


Karunatillaka, M. 1993. Siriyaawathige kataawa [Siriyaawathi’s Story]. Colombo: Godage. Karunatillake, W. S. 1984. Aytihaasika vaagvidyaa praveeshaya [Introduction to Historical Linguistics]. Colombo: Gunasena. Karunatillake, W. S. 1991. An Introduction to Spoken Sinhala. Colombo: Gunasena. Kiriwattuduwa, S. 1987. Isuru Soyaa [Seeking Treasures] Colombo: IPC Printers. Matzel, K. 1983. Einführung in die singhalesische Sprache. Wiesbaden: Harrassowitz. Rajapaksa, A. 1986. Gal boralu nopeegu paayuga [Feet that never felt stones]. Colombo: Lake House Investments. Sri Lanka Central Bank. 1999. Socio-Economic-Data. Colombo: Government Printing Press.

83

Indefinite pronouns Morphology and syntax in cross-linguistic perspective* Helmut Weiß University of Regensburg

1.

Introduction

This paper cross-linguistically investigates the morphology and syntax of socalled weak indefinite pronouns (cf. Diesing 1992) which semantically correspond to German jemand ‘someone’, etwas ‘something’ and (the indefinite article) ein ‘a’. The underlying assumption (due to Discourse Representation Theory, cf. Diesing 1992) is that weak indefinites are in general restricted variables which get bound by existential closure or by another quantifier. In addition, I purport that indefinites can have an additional feature depending on the syntactic context in which they appear. These additional features like [wh] or [Neg] can be thought of as formal features in the sense of the Minimalist Program (MP) (Chomsky 1995, 1999) which are thus subject to feature checking as conceived in MP. Informally speaking, indefinites are semantically decomposable into three parts, which is — in a very simplistic fashion — shown in Table 1 for jemand, wer, and niemand. Note that the existential quantifier is no proper part of the indefinite itself, but delivered by the context, e.g. via existential closure (as for wh-indefinites, cf. Bayer 2000). The underlying intuitive idea is that weak indefinites, despite being morphologically distinct in some languages, are not different lexemes but only allomorphs, as proposed for some and any by Musolino et al. (2000). I would like to extend this analysis to negative indefinites as well as wh-indefinites, too. In what follows, I will restrict my attention to three syntactic contexts in which the morphological design of indefinites will be observed. These are (1)

86

Helmut Weiß

Table 1

jemand ‘someone’ wer ‘who’ niemand ‘nobody’

Feature

Quantifier

Restriction

[αSpec] etc. [+wh] [+Neg]

$ $ $

Person (x) Person (x) Person (x)

positive sentences, (2) negative sentences, and (3) negative polarity contexts like adversative predicates, conditionals, questions and so on (cf. Suñer 1995). I will refer to these contexts as PPI-context (where Positive Polarity Items occur), NEG-context (i.e. direct negation) and NPI-context (where Negative Polarity Items occur), respectively. (1)–(3) are paraphrases of rather ‘idealized’ instances of the three contexts (here German is a more suitable metalanguage than English because it allows one to use jemand in all three contexts): (1) ‘Es ist der Fall, daß ich jemanden getroffen habe’ ‘It is the case that I have met X’ (2) ‘Es ist nicht der Fall, daß ich jemanden getroffen habe’ ‘It is not the case that I have met X’ (3) ‘Wenn es der Fall ist, daß ich jemanden treffe’ ‘If it is the case that I will meet X’

Before going on to investigate the morphological system of indefinites in several languages, which is my first aim, I would like to confess that I am particularly interested in negative indefinites (n-indefinites). The semantic nature of n-indefinites like nobody or nothing is still a matter of controversy in literature (cf. Suñer 1995, Hoeksema 1997 and others). The discussion mostly centers on the question of whether they are negative quantifiers or negative polarity items. Those who have argued for them being quantifiers still disagree on the matter of whether they are negated existentials or universals. On the contrary, defenders of the NPI hypothesis maintain that n-indefinites are akin to indefinite polarity items such as anybody. Still other scholars argue for a dual nature of n-indefinites: some assert that they are quantifiers or NPIs depending on the syntactic context, some maintain that they are negative universals or non-negative existentials. The most economical assumption to capture this diversity of behaviour seems, in my view, to be that in natural languages n-indefinites are no quantifiers at all, but weak indefinites in the sense of DRT, for which a certain kind of ambiguity between weak and strong readings was shown to hold in general.

Indefinite pronouns

2. Cross-linguistic patterns As for the inventory of indefinites, there are many differences between languages (cf. Haspelmath 1997), but the most obvious and presumably most basic one is that some languages possess n-indefinites and some do not. N-indefinites are defined here as those which incorporate a negative morpheme: therefore, English no-one is made up of the indefinite expression one and the negative morpheme no. The defining criterion is thus a formal one, namely the morphological make up: n-indefinites incorporate a morpheme which historically goes back to a negative expression (most often meaning ‘neither’, cf. Appendix A in Haspelmath 1997). In most languages, n-indefinites are nowadays lexicalised so that the morphological relation between them and negation is no longer transparent: e.g. Italian nessuno ‘no one’ coming from Latin nec uno (Haspelmath 1997: 262) may not be recognized now as containing a negative morpheme by Italian speakers. However, in some languages there are remnants of morphological transparency. For example, Russian n-indefinites, which consist of the negative morpheme ni and the wh-indefinite, are split up in prepositional phrases: ni s kem ‘with no one’, lit. ‘no with who’ (cf. Payne 1985: 236f.). Note that in languages like Spanish and others there can be indefinites like nadie ‘no one’ or nada ‘nothing’ which do not contain a negative morpheme: they nevertheless count as n-indefinites since their emergence was due to a development called suppletion in which they substituted n-indefinites (cf. Bernini and Ramat 1996: 156–158). The crucial point is that suppletion only occurred in languages which already had ‘regular’ n-indefinites. N-indefinites are also called inherently negative quantifiers (Payne 1985). However, many languages (see Haspelmath 1997) do not have n-indefinites: in Malayalam (see below), for instance, expressions like no-one do not occur. Therefore, the working hypothesis is that the (non-)occurrence of n-indefinites is the typologically relevant parameter causing the other differences which will be investigated in this paper. The following cross-linguistic investigation does not try to be exhaustive in a quantitative manner because the sample on which my investigation is based is too small. It comprises seven languages (Bavarian, Hungarian, Italian, Malayalam, Polish, Slovene, Turkish) for which I have collected data from native speakers,1 and eight other languages (Albanian, Armenian, Catalan, Greek, Hindi, Korean, Serbo-Croatian, Spanish) for which literature supplies appropriate and sufficient data. These languages were selected in order to represent both groups of languages mentioned above with several instances.

87

88

Helmut Weiß

The goal of my investigation is to elicit representative patterns which reveal interesting aspects of the syntax and semantics of indefinites. In what follows, I will completely neglect n-indefinites of standard languages which are not compatible with negative particles expressing clausal negation, e.g. Standard English nobody or Standard German niemand. They emerged from semi-natural language change as defined in Weiß (2001a), so their relevance for typological as well as theoretical research is minor. Due to lack of space, the reader is referred to Weiß (2001a) for further details.2 2.1 NC- vs. Non-NC-languages When we consider indefinites in negated sentences we can see two possibilities to express existential quantification in the scope of negation: languages use either n-indefinites or non-negated ones. The Bavarian sentence in (4a) represents the first possibility: the indefinite is morphologically marked for negativity in addition to the particle expressing clausal negation, whereby the two negations do not cancel one another.3 This phenomenon is called double or multiple negation with negative concord and I, therefore, refer to languages exhibiting this as NC-languages. Besides Bavarian (Weiß 1998, 1999, 2002) as an instance of a Germanic NC-language, NC appears in Romance (Zanuttini 1997) and Slavic languages (Progovac 1994) as well as in Hungarian (Tóth 1999) or Albanian (Turano 1998), for example.4 (4) a.

Gesdan han’e neamd ned gseng. (Bavarian) yesterday have-I nobody not seen ‘Yesterday, I saw nobody.’ b. Innale ñaa²n aar-e-um kaND-illa. (Malayalam) yesterday I who-acc-um see-not ‘Yesterday, I saw nobody.’

Apart from NC-languages, there are languages which do not possess negated indefinites. I refer to them as Non-NC-languages. In Non-NC-languages like Chinese, Greek (Giannakidou 1997, Klidi 1998, Tsimpli and Roussou 1996), Hindi (Lahiri 1998), Korean (Lee 1996) or Malayalam, existential quantification in the scope of negation is expressed by indefinites which are not morphologically marked for negativity. Sentence (4b) is an example from Malayalam (see above). Let us first consider NC-languages.


2.2 NC-languages For NC-languages it generally holds that n-indefinites are restricted to negative sentences, whereas in the PPI- or NPI-context, non-negated ones are used, as we see in (5a–d) for Italian. (5) a.

Non è venuto nessuno. not is come nobody ‘Nobody came.’ b. Non credo che qualcuno sia così stupido. not believe.1sg that someone be.3sg.conj so stupid ‘I do not believe that anyone is that stupid.’ c. Il testimone contesta che qualcuno abbia perso qualcosa. the witness denies that someone have.3sg.conj lost something ‘The witness denies that anyone has lost anything.’ d. Se/quando qualcuno vuole qualcosa, deve dirlo. if/when someone will something must say.it ‘If/when anyone wants anything, he must say it.’

This restriction to the context of direct negation holds in many NC-languages without any exception, e.g. in Bavarian, Hungarian or Slovene. In some NC-languages, n-indefinites could be additionally licensed by the preposition without, but this seems to differ from language to language in a non-predictable way, cf. (6a–d). (6) a.

senza niente without nothing b. bez niczego without nothing c. brez necˇésa without anything d. ohne ebbs without something ‘without anything’

(Italian) (Polish) (Slovene) (Bavarian)5

However, in some NC-languages there appears to be a strong tendency towards using n-indefinites like ordinary NPI-indefinites, i.e. occurring in sentences without overt negation and — this is the crucial point — without negative meaning at all. This is the case in Spanish (Suñer 1995) where n-indefinites are not only licensed by sentence negation (7a), but also by an adversative predicate like inútil as in (7b), and certain quantifiers like last (7c).

89

90

Helmut Weiß

(7) a.

No vio a nadie. not saw do nobody ‘S/he didn’t see anybody.’ b. Es inútil hacer nada por él. is useless do nothing for him ‘It is useless to do anything for him.’ c. Es la última vez que te digo nada. is the last time that you tell.1sg nothing ‘It is the last time that I tell you anything.’

Although Spanish is not a singular case (cf. Vallduví 1994 for Catalan), it is nevertheless the marked case. The unmarked one is that n-indefinites are restricted to direct negation, whereas in the PPI- and NPI-contexts non-negated indefinites appear. As for indefinites in the NPI-context, there are two types in NC-languages: they are either the same ones as in the PPI-context, or they are different from them. Polish is an example for a language which does not distinguish NPI-indefinites from PPI-indefinites: ktos´ ‘someone’ occurs in positive sentences like (8a,b) as well as in NPI-contexts like superordinate negation, questions, adversative predicates and conditionals (9a–d). A very interesting effect can be obtained in negated conditionals (9e,f ): the form of the indefinites varies depending on scopal requirements. Note that in (9e) the indefinite, though being outside the scope of the negation, does not require a specific reading, as would be the case in a root sentence. (8) a.

Wczoraj ktos´ zadzwonił (i pytał o pokój). yesterday who phoned (and asked for room ‘Yesterday, someone phoned (and asked for room).’ b. Znam kogos´, kto che˛tnie czyta ksiøz˙ki. know.1sg who.acc who willingly reads books ‘I know someone who willingly reads books.’

(9) a.

Nie sødze˛, ˙ze ktos´ jest tak głupi. not believe.1sg that someone is so stupid ‘I do not believe that anyone is that stupid.’ b. Kto mnie o cos´ pytał? who me for something asked ‘Who asked me anything?’ c. To nie jest konieczne, z˙e ktos´ zostaje w domu. it not is necessary that someone stays at home ‘It is not necessary that anyone stays home.’


d. Jes´li ktos´ cos´ chce, musi (to) powiedziec´. if someone something wants must (it) say ‘If anyone wants anything, he must say it.’ e. Jes´li Jan kogos´ nie chce widziec´ nie zauwaz˙a go. if John someone not wants see not sees him ‘If John does not want to see someone, he overlooks him.’ f. Jes´li Jan nikogo nie chce widziec´, zamyka drzwi. if John no one not wants see locks door ‘If John does not want to see anyone, he locks the door.’

Special NPI-indefinites exist in Hungarian, Serbo-Croatian (Progovac 1994) and Slovene. In Slovene (Derbyshire 1993), for example, we have three paradigms of indefinites: kdó ‘someone’, níkdo ‘no one’, and nékdo ‘anyone’. The latter form is used in the context of questions (10a), adversative predicates (10b) and in conditionals (10c, d), but it can be substituted by the PPI-forms in most cases. The contrast between (10d) and (10e) exhibits the scope effect already mentioned with regards to Polish. (10) a.

Bi lahkó (né)kdo zaprl okno? (Slovene) cond please (any/someone closed window?6 ‘Could any/someone close the window?’ b. Upiral se je nékaj podariti. refused refl is anything give.away ‘S/he refused to give away anything.’ c. Cˇe nékdo nékaj hoce, mora rêcˇi. if anyone anything wants must say ‘If anyone wants anything, he must say it.’ d. Cˇe Hans (né)koga ne želi videti, ga spregleda. if John (any/someone not wants see him overlook ‘If John does not want to see someone, he overlooks him.’ e. Cˇe Hans níkogar ne želi videti, zapre vrata. if John no one not wants see locks door ‘If John does not want to see anyone, he locks the door.’

Exactly the same distribution is found in Hungarian (Tóth 1999), where NPI-indefinites are derived from PPI-forms by adding the particle is which means ‘also, even’. Examples are given in (11a–e); ((11a,b) are Tóth’s (12) and (15)). (11) a.

Pál nem mondta, hogy Mária valakit is látott. Paul not said that Mary anybody saw ‘Paul did not say that Mary saw anybody.’

91

92

Helmut Weiß

b. Kétlem, hogy ezt valaki is megértette. doubt.1sg that this anybody understood ‘I doubt that anybody understood this.’ c. Ha János valakit (is) akar látni, … if John any/someone wants see ‘If John wants to see anybody, …’ d. Ha János valakit nem akar látni, … if John someone not wants see ‘If John does not want to see anyone, …’ e. Ha János senkit nem akar látni, … if John noone not wants see ‘If John does not want to see anyone, …’

In summary, the distinctive trait of NC-languages is the existence of n-indefinites which are restricted to the scope of direct negation in the unmarked case. In languages with only two indefinite paradigms, the PPI- and NPI-contexts nearly always share the same forms, with the exceptions mentioned above. In some languages there is a third indefinite paradigm occurring in NPI-contexts alone. What seems to be lacking are languages paralleling PPI- and n-indefinites. 2.3 Non-NC-languages In Non-NC-languages we find remarkable differences as well as parallels. First, Non-NC-languages can also be divided into two- and three-paradigm languages. In contrast to NC-languages, those with two indefinite paradigms use identical forms in NEG- and NPI-contexts. This is the case, for example, in Hindi (Lahiri 1998) where these forms are derived from PPI-indefinites by adding the particle bhii which means ‘also, even’ like the Hungarian is. In (12a–c) — Lahiri’s (6b), (10a), (29b) — we can see that these bhii-indefinites appear in direct negation, conditionals and adversative predicates. (12) a.

Koii bhii nahiiN aayaa. (Hindi) anyone not came ‘No one came.’ b. Agar raam kissi-ko bhii dekhegaa to tumheN bataayegaa. if Ram anyone see.fut then you tell.fut ‘If Ram sees anyone, he will inform you.’


c.

Mujhe is baat par aaScarya huaa ki koii bhii tumhaare me this fact on surprise be that anyone your ghar gayaa. house went ‘It surprised me that anyone went to your house.’

Besides this pattern there are also Non-NC-languages which have separate indefinites in all three contexts. This is the case in Malayalam and Korean. In Malayalam, indefinites are made up of an interrogative stem and three suffixal ‘indefinite markers’ (Hany Babu 1999, 2000).7 Indefinites with the suffix -oo are used in positive sentences as in (13a–b), whereas indefinites in the scope of negation are suffixed with -um (14).8 In NPI-contexts, indefinites are marked with the suffix -enkilum, as for example in the conditional in (15a). The contrast between the negated conditionals in (15b) and (15c) reveals that the morphology of indefinites is primarily due to whether they belong to the scope of negation or not, even in a language without n-indefinites. (13) a.

Innale aar-oo foon ceyt-irunnu. yesterday who-oo phone do-perf ‘Yesterday someone phoned.’ b. Innale ñaa²n aar-e-oo kaNDu yesterday I who-acc-oo saw ‘Yesterday I saw someone.’

(14) Innale ñaa²n aar-e-um kaND-illa. yesterday I who-acc-um saw-not ‘Yesterday, I did not see anybody.’ (15) a.

Aar-kk-e˙nkilum ent-e˙nkilum wee-Nam e˙nkil, … who-dat-enkilum what-enkilum want-opt if ‘If anyone wants anything, …’ b. Hans-in6 aar-e-e˙nkilum kaaN-eeNta e˙nkil, awan kaND-illa Hans-dat who-acc-enkilum see-not.opt if he saw-not enn6 naTikkum. quot pretend.will ‘If John does not want to see someone, he overlooks him.’ c. Hans-in6 aar-e-um kaaN-eeNta e˙nkil, awan waatil atak’k’-um. Hans-dat who-acc-um see-not.opt if he door close-mod ‘If John wants to see nobody, he locks the door.’

93

94

Helmut Weiß

3. Summary of the cross-linguistic data I now sum up the results of the brief cross-linguistic investigation. The first distinction I made was one between NC- and Non-NC-languages depending on whether they use n-indefinites or non-negated ones in the scope of negation. As we have seen, this really seems to be the fundamental distinction because the two types of languages behave rather differently in both of the other contexts investigated. Although in both types there are languages with two and three indefinite series, as we see in Table 2, the crucial difference is found between the 2-paradigm languages of both types. In Non-NC-languages the NEG and the NPI-context go together, i.e. they share the same indefinites. This is exactly what one would expect given the semantic properties (cf. Zwarts 1996) that the two contexts have in common. The somewhat surprising (and until now surprisingly unnoticed) result is then that in the unmarked case NC-languages tie together the PPI and the NPI context with regards to indefinites. Table 3 shows that Catalan and Spanish are really exceptions in that they pattern with Non-NC-languages although they possess n-indefinites. This result sheds light on at least two points worth discussing in some detail: the analysis of non-negated and n-indefinites in NC-languages. As for the first ones, vanden Wyngaard (1999) argues for English indefinite NPs with Table 2a +NC

PPI NPI NEG Languages

a

−NC

3 paradigms

2 paradigms

3 paradigms

2 paradigms

A B C Albanian, Hungarian, SerboCroatian, Slovene

A/B

A B C Malayalam, Korean

A B/C

C Bavarian, Italian, Polish, Turkish

Armenian, Greek, Hindi

Thanks to one of the reviewers for suggesting this form of the table.

Table 3

+NC −NC

PPI & NPI

NEG & NPI

Bavarian, Italian, Polish, Turkish

Catalan, Spanish Armenian, Greek, Hindi


the determiner a being weak PPIs because they are only ungrammatical within the semantic scope of sentential negation (16a), whereas they can occur in antiadditive (16b) and monotone decreasing contexts (16c).9 (16) a. *Sam didn’t greet a woman. b. All students who kissed a girl should go home now. c. Few surgeons had a beer after work.

This holds true for the non-negated indefinites of NC-languages with two indefinite paradigms, in general. They show exactly the same behaviour: e.g. Polish cos´ ‘some-/anything’ is grammatical in the complement of the adversative verb refuse — an anti-additive context — and within the scope of quantifiers like few — a monotone decreasing context, cf. (17a,b). Therefore, they can be classified as weak PPIs. (17) a.

Wzbraniała sie, z˙eby cos´ podarowac´. refused.3sg.fem refl to something give.away ‘She refused to give away anything.’ b. Niewielu studentów zrobiloby cos´ dla ciebie. few students do.conj something for you ‘Few students would do anything for you.’

In NC-languages with three paradigms, non-negated indefinites could be either strong PPIs if they occur only in monotone increasing contexts, or NPIs if they occur in NPI contexts (modulo sentence negation). However, this differentiation appears not to be very strict since, as we have seen for Slovene and Hungarian, in the NPI-contexts (modulo sentence negation) both forms are at least sometimes mutually exchangeable without differences in grammaticality or meaning.10 The second point worth discussing is the fact that n-indefinites in NC-languages only exceptionally occur in NPI-contexts (modulo direct negation), which obviously falsifies their analysis as NPIs. Given that negative polarity is an existing and uniform semantic property of specific contexts (cf. Zwarts 1996), and that clausal negation is a negative polarity context, it makes no sense to classify items occurring only under negation as NPIs. Even to classify them as special NPIs (weak ones?) seems to be inappropriate because then we have to put indefinites like German jemand or Italian qualcuno, which also occur in NPI-contexts, into this class, too (see above). I think it would be wiser to apply the term only to those items which appear in both the negative and the negative polarity context or, at least, only in the NPI-contexts. This means that n-indefinites in Catalan and Spanish are indeed NPIs as well as, for example, the nonnegated bhii-indefinites in Hindi because they appear in both contexts.

95

96

Helmut Weiß

On the other hand, none of the results above speaks in favour of analysing n-indefinites as negative quantifiers. The semantic import of n-indefinites in negated sentences as (18a) does not differ from that of non-negated ones as in (18b): both sentences can be best paraphrased as in (19). A rather natural analysis seems to be that even n-indefinites are indefinites in the sense of DRT, i.e. they introduce restricted variables which get existentially bound. Thus, they are no quantifiers.11 (18) a.

Gestern han’e neamd ned gseng. (Bavarian) yesterday have-I nobody not seen ‘Yesterday, I did not see anybody.’ b. Innale ñaa²n aar-e-um kaND-illa. (Malayalam) yesterday I who-acc-um saw-not ‘Yesterday, I did not see anybody.’

(19) ‘It is not the case that I saw anybody yesterday.’

There is a vast amount of empirical evidence corroborating the assumption that n-indefinites are semantically non-negated existentials. For instance, it seems to be rather common in NC-languages for lexical items without a negative morpheme to acquire a ‘negative meaning’, i.e. they can be used in elliptical negative answers like true n-indefinites and they are restricted to NEG contexts. This process is called ‘suppletion’ (cf. Bernini and Ramat 1996) and it concerns items like French personne or — presumably — German kein. As we have seen with regards to Spanish, the change of meaning can also turn in the other direction: lexical items containing an overt negative morpheme can almost completely lose their negative feature. There are also languages which have changed from NC- to Non-NC-languages — namely Greek — or from NonNC- to NC-languages — namely Turkish.12 All these changes could not be expected to happen so easily if n-indefinites were indeed negated quantifiers.

4. Towards an explanation of NC What remains to be explained is why n-indefinites contain negative morphemes. If the assumptions about the non-quantificational nature of n-indefinites are correct, then the widespread functional explanations can no longer be maintained: the ‘doubling’ negative particle can not be interpreted as either reinforcing the negative feature (Horn 1978, Haspelmath 1997) or merely doubling it (van der Wouden 1997). The following sketch of a possible explanation is twofold:


first it tries to locate the cause for doubling the negative morpheme on the indefinite in the semantics and syntax of negation (and not indefinites), and second to propose how to implement it in a Minimalist framework operating with a purely syntactic mechanism. 4.1 Semantics: Negation as quantifier over events Let us assume that the meaning of negated sentences can be paraphrased as in (20). (20) It is not the case that p.

One possibility to give this intuition an appropriate formalization is to treat negation, as Krifka (1989) or Aquaviva (1994) among others have proposed, as a quantifier binding the event-variable. To modify this and comparable conceptions, I will assume that the negative quantifier — like any other quantifier (Larson and Segal 1995) — has a tripartite structure as in (21a), i.e. it entails a restriction in addition to its scopal domain.13 Thus, a sentence like (21b) can be transformed into a semantic form like (21c) where the event-variable is bound in the restriction. Negation is therefore comparable to quantificational adverbs (cf. Chierchia 1995): both quantify over events. This result is conceptually very attractive since it corresponds to the intuition that linguistic negation has more in common, say, with frequency adverbs than with linguistic expressions of the logical operators and and or. (21) a. Neg [restriction] [scope] b. John did not insult Henry. c. Neg(x) [event (x)] [insult(J,H)]

The cross-linguistic investigation indirectly supports the quantificational approach to negation in two ways. First, it has shown that the semantic import of n-indefinites is the same as that of non-negated existentials in Non-NC-languages. The best way to put them in a formula as the one in (21a) is, thus, to map the variables introduced by them into the scope of negation where they get existentially bound. Therefore, the semantic form of an NC-sentence like (22a) is the one given in (22b). (22) a.

Non è venuto nessuno. (Italian) not is come nobody ‘Nobody came.’ b. Neg(x) [event (x)] ($y)[person(y) Ÿ come (y)]

97

98

Helmut Weiß

Further evidence comes from the if/when-clauses mentioned above. The morphological design of indefinites varies in dependence of whether they belong to the scope of negation or not, as in (23a) vs. (23b) from Italian. (23) a.

Se/quando Gianni non vuole vedere nessuno, … if/when John not want.to see nobody ‘If/when John does not want to see anybody, …’ b. Se/quando Gianni non vuole vedere qualcuno, … if/when John not want.to see somebody ‘If/when John does not want to see somebody, …’

Following Kratzer (1995) and others, I will treat conditionals as quantifiers — and additionally assume that the conditional operator binds the event-variable in its restriction (compare (3) above) — so that in negated conditionals, indefinites can scope under negation or conditional. Scoping under negation yields n-indefinites (23a) and scoping under conditional (i.e. above negation) results in non-negated ones (23b). As we have seen above, even in a NonNC-language like Malayalam, indefinites behave morphologically distinct in these two scopal environments (cf. also Hany Babu 2000). Therefore, it seems justified to treat negation on a par with quantificational adverbs or conditionals, i.e. as a quantifier. Ladusaw (1992, 1994) has proposed a similar treatment of NC, which differs from the one assumed here in some respects, however. Ladusaw identifies the specifier of NegP with the restriction of the negative operator, e.g. in order to capture the behaviour of Standard Italian nessuno which is incompatible with the negative particle non when occurring pre-verbally, but ungrammatical without it when occurring post-verbally. In the former case the n-indefinite is thought of as strongly construed, i.e. establishing the restriction of the negative operator, in the latter case it is understood as a nuclear scope variable (i.e. construed weakly). This strong-weak construal difference also holds for languages like Catalan where n-indefinite subjects in preverbal position are compatible with negation (Ladusaw 1992: 247 is, thus, forced to see Catalan no “as not expressing negation”). In my approach the specifier of NegP is part of the scope of negation, thus allowing a uniform analysis of n-indefinites independent of their surface position. Their semantics remain the same, whether they uniformly precede the negation (as in Slavic languages or Bavarian) or follow it (as negative indefinite objects in Romance languages). Following the logic of the Ladusaw approach, Slavic or Bavarian n-indefinites would always be strongly construed, which


seems to me highly questionable. This semantic inappropriateness is circumvent in my approach. The assumption that the specifier belongs to the scope of its head may seem a little strange at first, but it is nevertheless coherent with the MP analysis proposed here because the specifier of a functional projection is primarily a position where features get checked. This conception seemingly departs from the widely accepted theory of Beghelli and Stowell (1997) according to which the specifier of functional projections is the position from where quantifiers take scope. However, in my approach n-indefinites are no quantifiers, so they cannot take scope at all! In the concept developed above, the function of the negative morpheme is, intuitively speaking, to mark the indefinite as belonging to the scope of negation. This is exactly the same as what we found in Non-NC-languages: to make up the special form of indefinites used in negative sentences, particles or morphemes with the meaning ‘even, also’ are often attached to some bases like wh-indefinites. The common function of these particles in all contexts where they occur is best captured by analysing them as inclusion markers (as Lee 1996 assumes for Korean, but contra Gil 1993). A rather straightforward consequence of weak indefinites’ non-quantificational nature is that they alone must be marked in these ways when occurring under negation: The variables introduced by them are, in principle, open to be bound by any unselectively binding operator. If we conceive negation as a ‘true’ binding operator, we can intuitively interpret the negative as well as the inclusion marking as a means to prevent them from getting bound by negation.14 4.2 Syntax: NC as feature checking My assumptions on NC, mainly based on semantic considerations and the cross-linguistic findings, can be fairly well implemented in a syntactic framework. We have to assume the existence of a Negation Phrase (cf. Haegeman and Zanuttini 1991, Haegeman 1995, Ouhalla 1997) whose head hosts the negative particle and to whose specifier the negative indefinite must move to check its Neg feature, which is thereafter deleted — thereby receiving Haegeman and Zannutini’s (1991) NEG-absorption for free (see also Brown 1999 for a similar proposal). If we conceive of movement as copy and delete in a Minimalist sense, this derivation guarantees that the Neg feature of the negative indefinite is checked away and that the indefinite gets VP-internally interpreted (cf. Bayer 1997, Weiß 1999). The derivation in (24a–e) should illustrate how a mechanism like the one sketched below would work.

99

100 Helmut Weiß

(24) a.

dass’e neamd ned gseng hob (Bavarian) that-I nobody not seen have ‘that I haven’t seen anybody’ b. [NegP neamd [Neg0 ned] [VP … neamd …] c. [NegP neamd [Neg0 ned] [VP … neamd(x) …] pre-Spell-Out deletion d. [NegP neamd [Neg0 ned] $x [VP … neamd(x) …] existential closure e. [NegP neamd [Neg0 ned] $x [VP … neamd(x) …] post-Spell-Out deletion after feature checking

In order to get this mechanism to work within the Minimalist Program, we have to assume that the n-indefinite contains an uninterpretable Neg feature and Neg0 an interpretable one, and that Neg0 attracts the n-indefinite which moves to SpecNegP (24b) to get its uninterpretable feature checked away (24e).15 The copy which the n-indefinite leaves behind is deleted before Spell-Out (24c) and its variable existentially closed (24d). Note that for the n-indefinite to bear the interpretable Neg feature (as would be a rather natural consequence of analysing them as negated quantifiers, cf. Haegeman 1995, Beghelli and Stowell 1997) and for Neg0 to bear the uninterpretable one would not be possible because then sentences with only the negative particle would be ungrammatical in any case: The uninterpretable feature of Neg0 could not be checked and would force the derivation to crash on LF. As the contrast between Romance languages and Germanic or Slavic languages shows, this movement can be done either post- or pre-Spell-Out. One might be forced to reduce this difference to a different parametrisation in feature strength: The Neg feature of Germanic/Slavic n-indefinites is strong, and therefore must be checked away before Spell-Out, whereas Romance n-indefinites possess a weak Neg feature which allows post-Spell-Out checking. However, since in Chomsky (1998: 49) the concept of feature strength is abandoned, I would like to propose another scenario which keeps along the lines of the Minimalist Program as well, but explains the core data. The two basic patterns are the Italian-type (25a) and the Bavarian-type (25b):16 (25) a.

Non ho visto nessuno. not have.1sg seen nobody ‘I did not see anybody.’ b. dass’e neamd ned gseng hob that-I nobody not seen have ‘that I did not see anybody’


Suppose for the moment that checking of the Neg feature need not take place before Spell-Out in general, as indicated by Italian. The question to be answered is what forces n-indefinites in Bavarian to move before Spell-Out if it were not feature strength.17 One major difference between Bavarian and Italian is that the former is an OV language, which if sentence negation is present would yield the order Neg > O > V (for ease of argumentation, I omit subjects in what follows). Recall now that semantically we have conceived negation as an operator binding the event-variable, which is located in or associated with V0, and that binding means being syntactically c-commanded by an appropriate antecedent. This amounts to Neg0 binding V0. Now consider what happens when the negated indefinite object stays in place as in (26): Here the Neg-operator binds the object which intervenes between Neg0 and V0, thus yielding a ‘logical’ double negation reading (i.e. both negations cancel each other out). (26) wei’a eam ned nix schengd because-he him not nothing give.as.a.present ‘because he does not give nothing to him’

This reading follows immediately from the analysis of negation as a variable binding operator in the assumed syntactic framework. The problem to solve now is that the pre-Spell-Out movement of negative indefinites, as observed in Bavarian, cannot be forced by any semantic reason since movement is “semantically myopic” (Hornstein 1995: 69). However, there is a way to derive the (c)overt difference if we apply Chomsky’s (1999) concept of derivational phases or Uriagereka’s (1999) Multiple Spell-Out system. Note that the crucial difference with regards to the structural position of NegP is that it is above TP in Italian-type languages (Zanuttini 1997), but below it in Bavarian-type languages (Weiß 1998, 1999, 2002). So it seems plausible to assume that NegP belongs to different phases (in the sense of Chomsky 1999) in both language types: In Bavarian NegP is the edge of VP so that Neg0 can access items inside VP, whereas this is impossible in Italian. VP containing the arguments (which is a Larsonian double VP in Bavarian as well, cf. Weiß 2001b) is a Chomskyan phase (or a derivational cascade in Uriagereka’s sense), i.e. a subarray of the numeration which constitutes a more or less closed derivational cycle. The consequence thereof is that — according to Chomsky’s (1999) Phase Impenetrability Condition (PIC) — its accessibility from outside is restricted to what Chomsky calls the edge of a phase. This means that a functional head H0 with a feature [F] can attract a lexical item L with the same feature before Spell-Out only if HP is on the edge of an XP containing L.

101

102 Helmut Weiß

Applying this to our problem, we can assume that a NegP which immediately dominates VP can overtly attract an n-indefinite from inside VP to check its Neg feature, but not if NegP is above TP as is the case in Italian type languages. If this analysis (which needs to be worked out in greater detail, cf. Weiß, in press) is correct, then the (c)overt difference can be explained on structural grounds without the concept of feature strength.

5. Conclusion In this paper, I first tried to show that the difference between NC and Non-NC languages is typologically the most basic one: The pre-/absence of n-indefinites has far-reaching consequences for the structuring of the indefinite system of a language. If a language with two indefinite paradigms possesses n-indefinites, it has no NPI indefinites, which in contrast occur in Non-NC languages with two paradigms, because the NEG context and NPI contexts share the same indefinites. In NC languages with two paradigms there is no difference between PPI and NPI contexts with regards to indefinites. As far as I know, this has been unnoticed so far and it may have important consequences for theorizing as well. Based on these typological findings which show n-indefinites to be neither NPIs nor quantifiers, I attempted to give a new Minimalist explanation of NC in Section 4 which ties together the semantic and syntactic side of negation. The basic semantic assumptions were that negation is a tripartite quantifier (quantifying over events) and that n-indefinites get existentially closed. I further tried to implement this in a Minimalist framework: The NEG feature of n-indefinites is conceived as a formal feature (like the wh-feature of interrogatives) which has to be checked and gets thereafter deleted. Under this conception, NC loses most of its traditional senses: It is neither a form of reinforcement (Horn 1978, Haspelmath 1997) nor is it a negative polarity phenomenon (Progovac 1994) nor does the n-indefinite introduce the negative force (e.g. Beghelli and Stowell 1997, Haegeman and Zanuttini 1991, Haegeman 1995).

Notes * This paper presents the material and results of a research project on ‘Double Negation: Insights into the syntax-semantics-interface’ for which I am grateful for the financial support provided by the Deutsche Forschungsgemeinschaft (DFG). I would like to thank the

Indefinite pronouns 103

audiences of the Colloquium ‘Generative and Cognitive Linguistics’, University of Jena, 3.2.2000, and the Pronoun workshop in Marburg. Many thanks to the anonymous reviewers of this volume for their helpful comments and especially to Heike Wiese and Horst Simon for their suggestions and encouraging interest in this work. Special thanks to Janna Lisa Zimmermann for checking and improving my English. 1. Very special thanks to Ildikó Tóth, Ilaria Cicchetti, Sara Dassatti, Hany Babu, Mariola Sadowska, Magorzata Szłaga, Andrea Koncnik, Darja Lesnik and Dueniz Guerdal for supplying me with the data. The informants had to translate sentences from German into their native tongue. 2. A simple possibility to integrate standard languages of the English-German type is to analyse them as ‘hidden’ negative concord languages, that is to assume that the negative particle gets deleted at PF. This solution was suggested by one of the reviewers. Weiß (2001a) proposes the same mechanism for Romance languages like Italian where preverbal n-indefinites force PF deletion of the negative particle, but post-verbal ones do not. 3. Bavarian shows that n-indefinites and their positive counterparts need not come from the same root: neamd ‘no one’ comes from Old High German neo/nioman, whereas ebba ‘someone’ goes back to et(t)ewer (Braune and Eggers 1975: 250, 251). 4. The glosses throughout this paper will be as sparse as possible. 5. Other German dialects pattern with Italian and Polish: in some Swiss German dialects, e.g. ooni nüüt ‘without nothing’ is used with the meaning ‘without anything’, as one of the reviewers brought to my attention. 6. The present conditional in Slovene is formed by using the participle in -l (here zaprl) in addition to the particle bi, lahkó which literally means ‘easily’ (Derbyshire 1993). 7. In many languages, indefinite pronouns are derived from or identical to interrogative pronouns (cf. Haspelmath 1997), not surprising given their semantic parallel (i.e. being variables) discussed above. 8. The suffix -um meaning ‘also, even’ has three additional functions: (i) it is a conjunctive connective (e.g. John-um Mary-um ‘John and Mary’), (ii) it forms universal/distributive quantifiers (ella aaNkuTTikaL-um [all boys-um] ‘all the boys’), (iii) together with -aal- it marks conditionals (aar6 paraññ-aal-um awar sammatikk-illa [who say-COND-um they agree not] ‘They won’t agree (to it) who ever says (it)’) (Hany Babu 1999). 9. For the relevant definitions of anti-additive, monotone de-/increasing etc. in terms of Boolean algebra semantics cf. Zwarts (1996) or van der Wouden (1997). Note that (16a) becomes grammatical when interpreted specifically, i.e. outside the scope of negation. 10. Though this is a matter highly relevant for the present investigation, I leave it for further research. But note that, e.g. in Slovene, there are clear cases where no difference in meaning arises from the substitution of nekaj ‘anything’ through kaj ‘something’, cf. (i): (i) Cˇe Hansu nekaj/kaj ne ugaja, tega ne bere. if John any-/something not likes, that not reads ‘If John doesn’t like something, he doesn’t read it.’ In (i) even the known-unknown-difference which is relevant in other languages (Haspelmath 1997) as well as in Slovene does not play any role. For such cases I propose an analysis which

104 Helmut Weiß

could be called the ‘((irgend)et)was’-phenomenon: German irgendetwas could appear as well as etwas or as was in certain contexts, but not vice versa, cf. (ii) and (iii): (ii) Hast du heute schon ((irgend)et)was gegessen? (iii) Ich habe ((*irgend)et)was gegessen, das mir sehr geschmeckt hat. 11. Interestingly, Ladusaw (1994: 227), though he does not deny the quantificational force of Standard English nobody, analyses a sentence like ‘nobody left’ as a thetic judgement (in one reading), thus treating the strong quantifier nobody on par with the weak determiner three, i.e. being ambiguous between a strong and a weak construal. 12. A reviewer points to French as a language which is changing from a NC to a Non-NC language, where the negative clitic ne is lost in the spoken substandard (e.g. J’vois rien ‘I see nothing’), but multiple occurrences of n-indefinites harmonize (e.g. Il voit jamais rien ‘he never sees anything’). The loss of the negative particle is part of Jespersen’s cycle (cf. Horn 1989) and modern French will presumably one day arrive at the stage where Haitian Creole already is, cf. Jan pa we pèsòn, lit. John not see nobody, ‘John doesn’t see anybody’ (from Déprez 1997). Therefore, French is not changing to a Non-NC language at all. There seems to be a widespread misunderstanding to relate Jespersen’s cycle and NC (cf. Horn 1978, 1989), but both phenomena are independent of one another (cf. Weiß 1998). 13. The analysis proposed here is based on insights from Aquaviva (1994) and Ladusaw (1992, 1994). For the differences between their and my proposals see below. Pafel (1998) argues against analysing natural language negative items as quantifiers: his main objections are the phenomena of cohesion (e.g. no man spelling out not plus any man) and of pleonastic negation (i.e. NC). Since cohesion and NC in the traditional sense do not exist under my proposal, I take it as confirmation. 14. This statement is far beyond being a principled explanation, which I am not able to give at the moment. In addition, I have nothing substantial to say about why so many languages lack n-indefinites, but I am convinced this is neither accidental nor does it have anything to do with semantic variation. 15. The same seems to hold for wh-questions: It is the wh-phrase which has the uninterpretable feature, as is (tentatively?) assumed by Chomsky (1999). In an analysis of DP internal agreement, Carstens (2000) also argues that targets containing interpretable features can attract items with uninterpretable ones. 16. I think that it is substantial for a linguistic analysis of NC to have a story to tell about these two patterns, whereas the incompatibility of n-indefinites with the negative particle is an artificial trait which should not attract too much interest. For an extensive discussion of different types of naturalness in natural languages the reader is referred to Weiß (2001a). 17. The following explanation is intended to comprise other Germanic NC languages (e.g. West Flemish) as well.


References Acquaviva, P. 1994. “Representation of operator-variable dependencies in sentential negation”. Studia Linguistica 48: 91–132. Anderwald, L. 1999. Negation in Non-Standard British English. Ph.D. Diss., University of Freiburg i.Br. Bayer, J. 1997. Word order in Bavarian multiple negation. Paper presented at the 9th Wuppertaler Kolloquium, 1997. Bayer, J. 2000. Wh-in-situ. Ms., University of Jena. Beghelli, F. and Stowell, T. 1997. “Distributivity and negation: The syntax of each and every”. In Ways of Scope Taking, A. Szabolcsi (ed.), 71–107. Boston, Dordrecht and London: Kluwer. Bernini, G. and Ramat, P. 1996. Negative Sentences in the Languages of Europe: A Typological Approach. Berlin and New York: Mouton de Gruyter. Braune, W. and Eggers, H. 1975. Althochdeutsche Grammatik. 13th ed. Tübingen: Niemeyer. Brown, S. 1999. The Syntax of Negation in Russian. A Minimalist Approach. Stanford: CSLI Publications. Carstens, V. 2000. “Concord in minimalist theory”. Linguistic Inquiry 31: 319–355. Chierchia, G. 1995. Dynamics of Meaning: Anaphora, Presupposition, and the Theory of Grammar. Chicago and London: University of Chicago Press. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, N. 1998. Minimalist Inquiries: The Framework [MIT Occasional Papers in Linguistics 15]. Cambridge, MA: MITWPL. Chomsky, N. 1999. Derivation by Phase [MIT Occasional Papers in Linguistics 18]. Cambridge, MA: MITWPL. Déprez, V. 1997. “A non-unified analysis of negative concord”. In Negation and Polarity. Syntax and Semantics, D. Forget, P. Hirschbühler, F. Martineau and M.-L. Rivero (eds), 53–74. Amsterdam and Philadelphia: John Benjamins. Derbyshire, W. W. 1993. A Basic Reference Grammar of Slovene. Columbus, OH: Slavica Publishers. Diesing, M. 1992. Indefinites. Cambridge, MA: MIT Press. Giannakidou, A. 1997. The Landscape of Polarity Items. Ph.D. Diss., University of Groningen. Gil, D. 1993. “Conjunctive operators in South-Asian languages”. In Papers from the Fifteenth South Asian Language Analysis Roundtable Conference 1993, A. Davison and F. M. Smith (eds), 82–105. Iowa City: University of Iowa. Haegeman, L. 1995. The Syntax of Negation. Cambridge: Cambridge University Press. Haegeman, L. and Zanuttini, R. 1991. “Negative heads and the Neg criterion”. The Linguistic Review 8: 233–251. Hany Babu, M. T. 1999. Indefinite pronouns and correlatives. Paper presented at the Colloquium ‘Generative and Cognitive Linguistics’, University of Jena, 1999. Hany Babu, M. T. 2000. What the indefinites tell us about clause structure: Evidence from Malayalam. Paper presented at GGS 2000, Potsdam. Haspelmath, M. 1997. Indefinite Pronouns. Oxford: Clarendon Press.

106 Helmut Weiß

Hoeksema, J. 1997. “Negation and negative concord in Middle Dutch”. In Negation and Polarity: Syntax and Semantics, D. Forget, P. Hirschbühler, F. Martineau and M.-L. Rivero (eds), 139–156. Amsterdam and Philadelphia: John Benjamins. Horn, L. R. 1978. “Some aspects of negation”. In Universals of Human Language. Vol. 4: Syntax, J. H. Greenberg (ed.), 127–210. Stanford: Stanford University Press. Horn, L. R. 1989. A Natural History of Negation. Chicago and London: University of Chicago Press. Hornstein, N. 1995. Logical Form. From GB to Minimalism. Oxford and Cambridge, MA: Blackwell. Klein, J. S. 1997. “Indefinite pronouns, polarity, and related phenomena in Classical Armenian: A study based on the Old Armenian Gospels”. Transactions of the Philological Society 95: 189–245. Klidi, C. 1998. “Negative polarity items and negative quantifiers in Modern Greek: Two in one or one in two?”. The Linguistic Review 15: 209–213. Kratzer, A. 1995. “Stage-level and individual-level predicates”. In The Generic Book, G. N. Carlson and F. J. Pelletier (eds), 125–175. Chicago and London: University of Chicago Press. Krifka, M. 1989. “Nominal reference, temporal constitution, and quantification in event semantics”. In Semantics and Contextual Expressions, R. Bartsch et al. (eds), 75–115. Dordrecht: Foris. Labov, W. 1972. “Negative attraction and negative concord in English grammar”. Language 48: 773–818. Ladusaw, W. A. 1992. “Expressing Negation”. In SALT II Proceedings, C. Baker and D. Dowty (eds), 236–259. Ohio State University, OH. Ladusaw, W. A. 1994. “Thetic and categorial, stage and individual, weak and strong”. In SALT IV Proceedings, M. Harvey and L. Santelmann (eds), 220–229. University of Rochester, N. Y. Lahiri, U. 1998. “Focus and negative polarity in Hindi”. Natural Language Semantics 6: 57–123. Larson, R. and Segal, G. 1995. Knowledge of Meaning: An Introduction to Semantic Theory. Cambridge, MA: MIT Press. Lee, C. 1996. “Negative polarity items in English and Korean”. Language Sciences 18: 505–523. Lewis, G. L. 1967. Turkish Grammar. Oxford: Oxford University Press. Musolino, J., Crain, S. and Thornton, R. 2000. “Navigating the quantificational space”. Linguistics 38: 1–32. Ouhalla, J. 1997. “The structure and logical form of negative sentences”. Linguistic Analysis 27: 220–244. Pafel, J. 1998. Skopus und logische Struktur: Studien zum Quantorenskopus im Deutschen [Arbeitspapiere des Sonderforschungsbereiches 340: Sprachtheoretische Grundlagen für die Computerlinguistik Nr. 129]. Universities of Stuttgart and Tübingen. Payne, J. R. 1985. “Negation”. Language Typology and Syntactic Description, vol. 1: Clause Structure, T. Shopen (ed.), 197–242. Cambridge: Cambridge University Press. Progovac, L. 1994. Negative and Positive Polarity. A Binding Approach. Cambridge: Cambridge University Press. Suñer, M. 1995. “Negative elements, island effects and resumptive no”. The Linguistic Review 12: 233–273.


Tóth, I. 1999. “Negative polarity licensing in Hungarian”. Acta Linguistica Hungarica 46: 119–142. Tsimpli, I.-M. and Roussou, A. 1996. “Negation and polarity items in Modern Greek”. The Linguistic Review 13: 49–81. Turano, G. 1998. “Overt and covert dependencies in Albanian”. Studia Linguistica 52: 149–183. Uriagereka, J. 1999. “Multiple Spell-Out”. In Working Minimalism, S. D. Epstein and N. Hornstein (eds), 251–282. Cambridge, MA: MIT Press. Uriagereka, J. 2000. Comments on N. Chomsky’s Derivation by Phase. Ms., University of Maryland. Vallduví, E. 1994. “Polarity items, n-words and minimizers in Catalan and Spanish”. Probus 6: 263–294. Weiß, H. 1998. “Logik und Sprache: der Fall der doppelten Negation im Bairischen”. Linguistische Berichte 175: 386–413. Weiß, H. 1999. “Duplex negatio non semper affirmat. A theory of double negation in Bavarian”. Linguistics 37: 819–846. Weiß, H. 2001a. “On two types of natural languages. Some consequences for linguistics”. Theoretical Linguistics 27: 87–103. Weiß, H. 2001b. “Information structure meets Minimalist syntax. On argument order and case morphology in Bavarian”. To appear in Making Sense: From Lexeme to Discourse in Honor of Werner Abraham at the Occasion of his Retirement [GAGL Nr. 44], G. van der Meer and A. G. B. ter Meulen (eds), 21–34. Groningen: Center for Language and Cognition. Weiß, H. 2002. “Three types of negation. A case study in Bavarian”. In Syntactic Microvariation [Meertens Institute Electronic Publications in Linguistics (MIEPL) 2], S. Barbiers, L. Cornips and S. van der Kleij (eds). (http://www.meertens.knaw.nl/projecten/sand/synmic/). Weiß, H. in press. A quantifier approach to negation in natural languages. Nordic Journal of Linguistics 25. Wouden, T. van der. 1997. Negative Contexts. Collocation, Polarity and Multiple Negation. London and New York: Routledge. Wyngaerd, G. vanden. 1999. “Positively polar”. Studia Linguistica 53: 209–226. Zanuttini, R. 1997. Negation and Clausal Structure. A Comparative Study of Romance Languages. New York and Oxford: Oxford University Press. Zwarts, F. 1996. “A hierarchy of negative expressions”. In Negation. A Notion in Focus, H. Wansing (ed.), 169–194. Berlin and New York: de Gruyter.

Reference and representation of pronouns* Klaus von Heusinger University of Konstanz

1.

Introduction

Personal pronouns in the third person, such as English he, she, it, they, exhibit a great variety of referential properties for different functions in a sentence or a discourse. However, the difference in their referential behavior is not reflected in their lexical, morphological, or syntactic appearance and properties. This is illustrated by one of the most prominent contrasts in the referential behavior of pronouns: the contrast between their deictic and anaphoric uses. There are basically two options for the semantic representation: The first option remains close to the view that pronouns vary in their referential behavior; hence, they are represented in different ways. This is the assumption of most semantic theories (Geach 1962; Lasnik 1976; Evans 1980, etc.) which represent deictic pronouns as free variables, while anaphoric pronouns are copies of their antecedents, bound variables, or impoverished definite descriptions. The second option assumes that there is only one semantic representation reflecting the unity of the morphological form. The different referential properties are analyzed by using additional contextual parameters which determine the interpretation; e.g. the difference between the deictic and the anaphoric use of a pronoun can be reduced to the influence of contextual parameters on the interpretation process. In this paper, I defend the second option with a detailed analysis of crosssentential pronouns. I assume that pronouns are interpreted as referring to the most salient element given so far: a deictic pronoun refers to an object that was raised to salience by the extra-linguistic context; an anaphoric pronoun refers to an item that was made salient by the linguistic context, e.g. by having been previously mentioned. The pronoun itself can be represented as an indexed epsilon term εix [P(x)] that is interpreted as referring to the most salient

110

Klaus von Heusinger

element with the property P. The argument is built on observations from the research tradition, systematic considerations, and on a close comparison with the analysis of definite NPs. The paper is organized as follows: Section 2 gives a very short overview of the treatment of deictic and anaphoric pronouns throughout history. Since classical times, deictic and anaphoric pronouns have been distinguished from one another. The latter were regarded as copies of their antecedents, even though this view has been heavily criticized ever since it was formulated. In the last 30 years, new differences were discovered in the context of new semantic theories. Section 3 provides a systematic comparison of the properties of the mentioned groups of pronouns. Depending on the theoretical background, different classes are formed. Furthermore, there is no clear cut distinction between classes since features are shared across different classes. Starting from the three main principles for pronoun interpretations — binding, being an independent term, and salience — it will be shown that only the last principle is essential. Section 4 presents a similar situation which holds for definite NPs. There are different uses of definite NPs, similar to the uses of personal pronouns. Other than with pronouns, most semantic theories assume a uniform representation of definite NPs. Once again, salience is presented as the basic principle of definiteness, and epsilon terms are introduced as the formal representation for definite NPs. They are interpreted by context dependent choice functions, which reconstruct the salience structure of the discourse. Section 5 applies this analysis to personal pronouns and shows that their different referential behavior can be explained by assuming that pronouns are represented by context dependent epsilon terms. It will be shown that the different uses of pronouns can be captured by different aspects of the salience structure of a text. Section 6 provides a short summary of the argument developed in this paper.

2. History of the analysis of pronouns 2.1 The Greeks: Deictic vs. anaphoric From the early investigation into linguistics and the categories of words, it became clear that personal pronouns in the third person have two quite different functions, although they have only one form. One of the earliest statements is the following by Apollonios Dyskolos of Alexandria (2nd cent. AD)

Reference and representation of pronouns

in περ ντωνυµ ας (peri antônymias “on pronouns”, p. 10 B, ed. J. Bekker, quoted according to Schneider 1965):1 Πσα ντωνυµ α δεικτικ ’στιν ναφορικ pâsa antônymia ê deiktikê ’stin ê anaphorikê every pronoun either deictic is or anaphoric “Every pronoun is either deictic or anaphoric”

Since then several approaches have been undertaken to unify the two notions. To give one example, Bühler (1934: 385ff. (§26 “Anaphora”)) tries to distinguish between different domains in which one and the same pronominal pointing-function (“Zeigefunktion”) operates: in one domain (the context) they result in deictic pronouns, in the other one (the symbolic field) they result in anaphoric pronouns. 2.2 Bloomfield: Substitution The structuralist Bloomfield (1933: 251 (§ 15.4)) took pronouns — as their Latin name suggests — as standing for nouns. He describes the different functions of the pronoun he in Chapter 15 “Substitution”, where he distinguishes between two types of substitution: anaphoric and deictic use, which he calls “limitation”. He compares deictic pronouns with definite NPs: The meaning of the substitute he may be stated thus: (…) B Substitution-types 1.Anaphora: he implies, in nearly all its uses, that a substantive designating a species of male personal objects has recently been uttered and that he means one individual of this species, say ‘recently mentioned’; 2.Limitation: he implies that the individual is identifiable from among all the individuals of the species mentioned; this element of meaning is the same as that of the syntactic category of definite nouns (§12.14) and can be stated, say, as ‘identified’.

Following Bloomfield, American structuralism assumed a substitution function of pronouns (“pronominalization”), which was then inherited by transformational grammar as a “substitution transformation”. In this view, the pronoun replaces a copy of the syntactic expression of its antecedent. Such a pronoun is sometimes called “pronoun of laziness”. This view works for proper names as in (1a): the pronoun her refers deictically to some salient female individual, while the pronoun he in the second sentence refers to its antecedent John. Both can be understood as substitutes for the full proper names as illustrated in (1b).

111

112

Klaus von Heusinger

(1) a. John met her for dinner. He had a wonderful evening. b. John met Mary for dinner. John had a wonderful evening.

The substitution theory of pronouns has already been criticized by the Stoics. They illustrated this with the “nobody-paradox”. One cannot replace the pronoun he in (2a) by its antecedent somebody, as in (2b) (cf. Egli 2000: 19ff.): (2) a. If somebody is in Athens, it is not the case that he is in Rhodes. b. If somebody is in Athens, it is not the case that somebody is in Rhodes.

A second problem is illustrated by sentences with cross-reference, as in the famous Bach-Peter sentences. If we substitute the pronouns in (3a) by their antecedents, we get sentence (3b) where each substitute contains another pronoun which must be replaced by the other substitute etc. By the structure of these examples, we will always end in a recursive structure which will never come to an end (cf. Bach 1970, Karttunen 1971). (3) a. {The man who deserves [it]2}1 will get [the price, {he}1 desires]2. b. {The man who deserves [the price, {he}1 desires]2}1 will get [the price, {the man who deserves [it]2}1 desires]2.

2.3 Quine and quantifiers Quine (1960: 113 (§23)) criticizes the substitution method along the same line as the “nobody-paradox” of the Stoics. The pronoun it cannot be supplanted by its antecedent if this is an indefinite term. The pronoun it “remains a definite singular term whether its antecedent is or not”. Quine (1960: 103 (§21)) also notes that pronouns are short-hands for definite descriptions: Often the object is so patently intended that even the general term can be omitted. Then, since ‘the’ (unlike ‘this’ and ‘that’) is never substantival, a pro forma substantive is supplied: thus ‘the man’, ‘the woman’, ‘the king’. These minimum descriptions are abbreviated as ‘he’, ‘she’, ‘it’. Such a pronoun may be seen thus as a short singular description, while its grammatical antecedent is another singular term referring to the same object (if any) at a time when more particulars are needed for its identification.

In §28 “Some ambiguities of syntax”, Quine discusses another set of cases involving pronouns, as illustrated in (4). (4) Everything has a part smaller than it.


Quine calls this sentence ambiguous since the pronoun it can be co-referential — or “cross-referential” in Quine’s terminology — to the universal quantifier everything or to the indefinite a part. However, we cannot disambiguate the example by substituting the pronouns with their antecedent as argued before. Therefore, Quine suggests that such pronouns are represented by variables that can be bound. So he (1960: 137 (§28)) replaces the concept of co-reference with the principle of binding from predicate logic: Logicians happily have another terminology for talking of cross-reference, where variables are concerned: they speak of binding. The introductory or appositive occurrence of ‘x’ is said to bind the various recurrences of ‘x’, insofar as they hark back to that apposition and not to some independent use of the letter.

Generalizing Quine, the classical view assumes that all pronouns are represented as variables of predicate logic: Deictic pronouns are like unbound variables which must be given a value by contextual information as in (1c); and anaphoric pronouns are bound variables, which get the same value as their antecedent due to the formal rule of quantifier interpretation, as illustrated by the semantic representations (4b) and (5b) for the sentences (4a) and (5a). For the bound pronoun cases we additionally need co-indexing at the level of syntactic form, as in (4a) and (5a): (1) a. c.

John met her for dinner. Met_for_dinner(john, x) with x contextually determined

(4) a. Everythingi has a part smaller than iti. b. "x [Thing(x) Æ $x [Part(y) & Smaller(y,x) & Has(x,y)]] (5) a. Nobodyi thought that hei was culprit. b. ¬$x [Thought(x, (Culprit(x))]

The contrast between deictic pronouns as free variables and anaphoric pronouns as bound variables has been the classical semantic view since Montague. Anaphoric pronouns stand for the same semantic objects or referents as their antecedents, rather than for the same syntactic expression as in the structuralist and early transformational view presented in Section 2.2. Note that this position does not reflect Quine’s description of pronouns since for a certain class of pronouns he assumes that they are impoverished definite descriptions, rather than variables.

113

114

Klaus von Heusinger

2.4 Geach: Cross-sentential anaphoric pronouns Geach (1962) shows that there are serious problems with the binding approach of the classical view for cross-sentential anaphoric pronouns since the scope of the binding operator is essentially defined in the limit of a sentence. This can be illustrated with examples like (6a), where the anaphoric pronoun he occurs in a separate sentence. In the representation (6b), the pronoun is represented as the third occurrence of the variable x. In this position, the variable cannot be bound by the existential quantifier since it is outside of the scope of the quantifier: (6) a. A mani walks. Hei whistles. b. $x [Man(x) & Walk(x)] & Whistle(x) [problem of scope]

Given that the problem of binding can be solved in some way or other, there is a second problem of cross-sentential anaphoric pronouns, namely that the referent of the indefinite is determined before the anaphoric pronoun is resolved. This can lead to wrong interpretations — therefore, I call it the “problem of interpretation”. Assume a situation with two walking men: man1 is not whistling and man2 is. Intuitively, sentence (6a) is true in this situation since there is a man who is walking and whistling. For the interpretation of the two sentences we start with the first sentence, as illustrated in (6c). It becomes true if one man walks — let us say man1. Now we have fixed the referent for the indefinite NP, and the pronoun must get the same referent, namely man1. Since man1 does not whistle, the second sentence becomes false and therefore the whole conjunct — contrary to our intuitions. (6) c.

(6b) = 1 iff $x [Man(x) & Walk(x)] = 1 & Whistle(x) = 1 Situation: man1 Œ Walk , man2 Œ Walk , man2 Œ Whistle Man(x) & Walk(x) = 1 for x = man1 Whistle(x) = 0 for man1; therefore (6b) = 0 [problem of interpretation]

Geach argues that we have to replace the classical existential quantifier, which is restricted to a sentence, with a textual existential quantifier. There are two steps in interpreting cross-sentential anaphoric pronouns: (i) the representation of the anaphoric link by using the same variable, and (ii) the binding of this variable at the level of a text with a new kind of existential operator $text:


(6) d. i. anaphoric resolution: Man(x) & Walk(x) & Whistle(x) …(x)… ii. binding at text-level: $textx [Man(x) & Walk(x) & Whistle(x) …]text

2.5 Dynamic semantics Geach’s analysis of cross-sentential anaphoric pronouns conflicts with the principle of compositionality since Geach can only interpret the whole text, but not a particular sentence without its context. Two ways to solve this problem have been suggested: (i) dynamic semantics and (ii) E-type semantics, which I discuss in Section 2.6. Dynamic semantics appears in different guises, e.g. as File Change Semantics (Heim 1982), Discourse Representation Theory (Kamp 1981; Kamp and Reyle 1993), and Dynamic Predicate Logic (Groenendijk and Stokhof 1991). In the following, I sketch the principles of dynamic semantics by means of the example of Dynamic Predicate Logic (DPL). The basic idea of DPL is that the meaning of a sentence is not a truth value, but a contribution to change the context. The meaning can be expressed as a relation between two information states, an input and an output state. An information state can be constructed as a set of assignment functions. The basic idea is the same as in the Heim-Kamp theories, namely that an indefinite NP determines an assignment function for the variable that is introduced by the indefinite. The formalism passes this assignment function to the following text, and all subsequent occurrences of the variable receive the same value by the determined assignment function. DPL employs the traditional syntax of predicate logic, but interprets the symbols in a different way. In particular, the conjunction and the existential quantifier receive a dynamic interpretation that makes it possible to “bind” variables across the syntactic scope of the existential quantifier. Sentence (6a) is represented by the formula (6e), which is equivalent to (6f) due to the particular definition of the interpretation process, explained below. (6) e. f.

$dynx [Man(x) &dyn Walk(x)] &dyn Whistle(x) $dynx [Man(x) &dyn Walk(x) &dyn Whistle(x)]

In the following I demonstrate the dynamic interpretation of Groenendijk and Stokhof (1991: 46ff.) with the interpretation of the atomic formula (7b), the conjunction (7c), and the existential quantifier (7d). In general, an expression α is interpreted as the set of pairs of assignment functions g and h, such that g is the input assignment function and h is the output assignment function. Additional restrictions hold for these pairs — representing the information of

115

116

Klaus von Heusinger

the sentence. Thus an informative sentence reduces the set of input-output pairs. The interpretation (7b) of an atomic formula is static, i.e. the input function is the same as the output function. The interpretation only “tests” if the input function holds of the atomic sentence. The interpretation (7c) of the conjunction is dynamic: it introduces a new assignment function k that serves as output of the first conjunct and as input of the second. The interpretation (7d) of the existential quantifier is also dynamic: it changes the input function g to a new function k that maximally differs in the assignment of the variable x, written as k[x]g. In other words, the existential quantifier “updates” a given assignment function g for the value of the variable x it binds. The updated function is then the input function for the next expression (for convenience, I will suppress the index M, g on the interpretation function α M,g): (7) a. b. c. d.

α = {·g, hÒ | restriction } R(t1, … tn) = {·g, hÒ | h = g & · t1 , …, tn Ò Œ R } &dyn ψ = {·g, hÒ | $k: ·g, kÒ Œ & ·k, hÒ Œ ψ } $dynx = {·g, hÒ | $k: k[x]g & ·k, hÒ Œ }

In this way the existential quantifier can “bind” across sentence boundaries, as it is illustrated for the interpretation of (6e) in (6g). We introduce a new assignment function k that is the output of the first conjunct and the input of the second; the first conjunct updates the input function g to l[x]g, i.e. g modified for the value of x, and passes this function to the next conjunct such that the x in the second conjunct is assigned the same individual as the two occurrences of x in the first conjunct. Thus, the updated assignment function l[x]g assigns to all occurrences of x the same value or individual. Therefore, the interpretation of (6e) is equivalent to the one of (6f): (6) g.

(6e) = {·g, hÒ| $k: ·g, kÒ Œ $dynx [man(x) &dyn walk(x)] & ·k, hÒ Œ whistle(x) } = {·g, hÒ| $k: ·g, kÒ Œ {·g, hÒ | $l: l[x]g & ·l, hÒ Œ man(x) &dyn walk(x)] }& ·k, hÒ Œ whistle(x) } = {·g, hÒ| l[x]g & ·g, lÒ Œ man(x) &dyn walk(x) & ·l, hÒ Œ whistle(x) } = {·g, lÒ| l[x]g(x) Œ man & l[x]g(x) Œ walk & l[x]g(x) Œ whistle }

Summarizing, dynamic semantics assume that anaphoric pronouns are variables that must be bound dynamically. Dynamic binding is reconstructed as the interaction of the context-change potential of certain expressions with the interpretation of discourse items depending on the updated context.


2.6 Evans: E-type pronouns Evans (1977; similar by Cooper 1979, later Neale 1990 and Heim 1990) discusses another problem of Geach’s analysis and its reformulation in dynamic semantics. Evans argues that the representation of the pronoun they as a bound variable in (8b) does not give the correct truth conditions of the two sentences in (8a). Rather, (8b) is the representation for sentence (8c), where the quantifier is restricted by the content of the first sentence. However, in the intuitive interpretation of (8a), the quantifier phrase few MPs is not restricted by came to the party (otherwise the sentence would not make sense). The pronoun they should be rephrased as all the MPs who came to the party as in (8d), and not represented as a bound variable. (8) a. b. c. d.

Few MPs came to the party, but they had a wonderful time. [Some: x MP: x] (came to the party (x) & had a wonderful time (x)) Few (of the) MPs who came to the party had a wonderful time. Few MPs came to the party, but all the MPs who came to the party had a wonderful time.

The same problem arises for an analysis of sentence (9a) with a singular pronoun. Dynamic semantics represents it as (9b), where the pronoun is captured by the bound variable. However, this representation stands for sentence (9c), which intuitively expresses a different meaning than sentence (9a). Evans argues that a pronoun that is outside the syntactic scope of its antecedent quantifier — a so-called “E-type pronoun” — has to be represented by an independent term that refers to all the individuals that make the antecedent sentence true. Thus the pronoun in (9a) can be rephrased as the man who drank champagne in (9d). The best representation for this is a definite description as in (9e). Russell (1905) defines a definite description by the uniqueness of its descriptive material. In (9d) the pronoun he in (9a) is replaced by the definite description the unique man who drank champagne, which is represented with the corresponding iota-term ιx [Man(x) & Drank_champagne(x)]: (9) a. b. c. d. e.

Just one man drank champagne and he was ill. $!x [Man(x) & Drank_champagne(x) & Ill(x)] Just one man both drank champagne and was ill. Just one man drank champagne and the (unique) man who drank champagne was ill. $!x [Man(x) & Drank_champagne(x)] & Ill(ιx [Man(x) & Drank_champagne(x)]

117

118

Klaus von Heusinger

2.7 Functional pronouns Another problem for binding approaches is the interpretation of so-called “paycheque pronouns” (cf. Karttunen 1969: 114), which I call ‘functional pronouns’. The pronoun it in (10a) does not refer to the same object as its antecedent his paycheque, rather it refers to a different paycheque. The representation (10b) as a bound variable predicts a reading according to which we are talking about one and the same paycheque. The pronoun it stands for his paycheque and represents a function f from men into their paycheques as in (10c); this function can be spelled out by a definite description as in (10d): “the paycheque of y” with y a free variable that must be either locally bound or supplied by the immediate context.2 (10) a.

The man who gave his paycheque to his wife was wiser than the man who gave it to his mistress. b. There is an x such that x is a paycheque and it holds that the man who gave x to his wife was wiser than the man who gave x to his mistress. c. There is an f such that f(y) is a paycheque and it holds that the man1 who gave f(man1) to his wife was wiser than the man2 who gave f(man2) to his mistress. f: function from men into their paycheques d. ιx [Paycheque(x) & Owned(x,y)]

Functional pronouns can be easily captured in theories that represent pronouns as definite descriptions. The difference between anaphoric pronouns and functional pronouns is that the former are represented by a definite description whose descriptive material is a one-place predicate, while for the latter, the descriptive material contains (at least) a two-place predicate, as in (10d). To sum up, E-type approaches assume that pronouns are terms, i.e. definite descriptions, rather than bound variables. The descriptive material of the definite description must be provided by the linguistic environment and the pragmatic context. The two main problems of E-type theories are (i) the uniqueness condition of the Russellian definite description standing for the pronoun. The representation with the Russellian description requires that there is only one such object in the relevant context. However, this is very often not the case; and (ii) there is no convincing description for the mechanism of copying the descriptive material of the definite description (cf. Evans 1977, Cooper 1979, Neale 1990, Heim 1990).


2.8 Coreference and coindexing Anaphoric pronouns (except for functional pronouns) are coreferential with their antecedents. This is explained either by binding the same variable (see Section 2.4) or by some principle of copying the descriptive material of the antecedent into the pronoun (see Section 2.6). However, in most semantic representations, coreference must already be assumed. It does not follow from the representation as such. Both families of approaches assume that the antecedent and the anaphoric expression are marked by the same index indicating coreference. Often, this is not necessary since in simple examples there are not enough potential antecedents to see this problem. However, in the fragment (11) from Hemingway it becomes crucial. (11) A clean, well-lighted place It was late and every one had left the café except an old man who sat in the shadow the leaves of the tree made against the electric light. […] The two waiters inside the café knew that the old man was a little drunk […]. “Last week he tried to commit suicide,” one waiter said. “Why?” […] The younger waiter went over to him. […] The old man looked at him. The waiter went away. […] The waiter took the brandy bottle and another saucer from the counter inside the café and marched out to the old man’s table. […] The waiter took the bottle back inside the café. He sat down at the table with his colleague again. (Hemingway 1925)

The pronoun he in the last sentence could refer to any of the discourse items that are masculine: the two waiters or the old man. Neither dynamic theories nor E-type approaches can explain why the pronoun refers to the last mentioned discourse item the waiter. For E-type theories, there is an additional problem: Once we have copied the material of the waiter into a definite description representing the pronoun, we get problems with the uniqueness condition since there are two waiters in the immediate context. The case clearly favors an analysis that is based on salience, which has to be constructed out of recency and other parameters. Here, we can neglect other parameters and reduce salience to recency: the last item mentioned is the most salient, and therefore picked up by the pronoun.3 This idea is worked out in Section 3.4.

119

120 Klaus von Heusinger

3. Systematic classifications of pronouns The historical survey has provided a great variety of ways how pronouns can refer, and along with it a similar great variety of representations. For a clear classification we have to account for three main parameters: In Section 3.1, I discuss cases, uses, or functions of pronouns, often illustrated by clear contrasts such as the contrast between deictic and anaphoric. In Section 3.2, I describe the properties of these different uses of pronouns and assign them to the different groups. The main properties or principles according to which pronouns refer are: the principle of binding, the principle of being an independent term, and the principle of salience. In Section 3.3, I show how the different uses are organized in groups or classes depending on the representational structure of the given theory. I compare dynamic approaches with E-type approaches. Finally in Section 3.4, I argue that a third alternative is necessary: a theory that is based on the principle of salience. 3.1 Varieties of pronouns Throughout the history of describing and analyzing third person pronouns, different groups of pronouns have been distinguished. We can summarize the groups following Evans (1980: 337) except for the functional pronouns (v), which Evans would have included into group (iv).4 (12) Different uses or groups of pronouns: i. Deictic pronouns Pronouns used to make a reference to an object (or objects) present in the shared perceptual environment, or rendered salient in some other way. ii. Pronouns of laziness Pronouns intended to be understood as being coreferential with a referring expression occurring elsewhere in the sentence. iii. Pronouns as bound variables Pronouns which have quantifier expressions as antecedents, and are used in such a way as to be strictly analogous to the bound variables of the logician. iv. E-type pronouns Pronouns which have quantifier expressions as antecedents, but are standing outside of the scope of the quantifier. They are used as referring expressions.


v.

Functional pronouns Pronouns that do not refer to the same object as their antecedents. They “go proxy” for a functional concept with one open argument that must be filled by the immediate context.

3.2 Properties of pronouns The characterization of the different groups includes functional and representational contrasts. The different functions of pronouns are distinguished by various properties. In Table 1, I list some of the more prominent features: One of the main distinctions is deictic vs. anaphoric (a), while the distinction into coreferential pronouns vs. non-coreferential pronouns (b) is rarely mentioned. Pronouns can receive their referent by the principle of salience as is the case with deictic pronouns or pronouns of laziness or they (seem to) get their referent by (dynamic) binding (c). Pronouns can also differ in the dependency on the context (d). This feature differs from (c) since it also includes functional pronouns (v), which need to be fed by some argument to be capable of reference. Furthermore, we can distinguish between pronouns inside or outside the scope of their antecedents (e). This criterion is not applicable for deictic pronouns since they do not have a linguistic antecedent. Table 1.The properties of pronouns a. deixis b. coreference c. salience/binding d. context/binding e. scope

deictic: (i) coreferential with antecedent: (i)–(iv) value determined by salience: (i), (ii) depending on contextual information: (i), (ii), (v) in the scope of the antecedent: (ii), (iii)

anaphoric: (ii)–(v) not coreferential with antecedent: (v) value determined by binding: (iii), (iv) binding/copying material: (iii), (iv) outside of the scope of the antecedent: (iv), (v)

3.3 Categorization of pronouns Depending on the theory, the uses (or groups) of pronouns are categorized differently. The classical theory of pronouns (Geach 1962, Lasnik 1976, etc.) groups (i) and (ii) into one category that is governed by the principle of

121


salience, while the pronouns in groups (iii) and (iv) are represented as bound variables; group (v) is not regarded as a prominent case. The distinction deictic vs. anaphoric cuts across this structure. conferential pronouns

(13) salience (i)

bound variable (ii)

deictic

(iii)

(iv)

?? (v)

anaphoric

Comparing the two main families of approaches to pronouns, we can summarize the characteristics in Table 2. Deictic pronouns are represented by a free variable that must be interpreted according to a contextual parameter (which mirrors the salience structure of the context) (a). Alternatively, in dynamic semantics, deictic pronouns can be represented by accommodation of the nonlinguistic antecedent and then by binding. Coreference must be indicated by indexing in both kinds of approaches (b). The main means of describing the anaphoric link is dynamic binding in dynamic semantics and the use of definite descriptions in E-type theories (c). For pronouns standing outside the scope of their antecedents (and binders), dynamic binding as well as the representation by definite descriptions works well (d) — it was exactly the case for which the two approaches were designed. However, as we saw in Section 2.6, there are cases where the binding approach results in readings which are too strong (cf. examples (8) and (9) and Table 2 (e)). Furthermore, dynamic semantics does not provide a representation for functional pronouns (f), as discussed in Section 2.7. On the other hand, E-type theories show problems with the mechanism of copying the material into the definite description (g) and with the uniqueness condition of the Russellian definite description (h). 3.4 The salience approach The picture we get is that both families of approaches have some serious problems: the binding approach produces representations that predict incorrect truth conditions (see the discussion of (8) and (9) above) and no representations for functional pronouns, while the question of how to copy the material for the definite description and the uniqueness condition is highly problematic for E-type theories. Therefore, a third type of semantic theory is called for: This theory is based on the third important principle of pronoun interpretation, the


Table 2.Main characteristics of dynamic semantics and E-type theories with respect to the representation of pronouns Characteristics a. deictic pronouns b. coreference c. main means for describing the anaphoric relation d. pronouns outside the scope of the antecedent e. restriction of the binder (cf. (8) and (9)) f. functional pronouns g. copy of the descriptive material of the antecedent h. uniqueness condition

Dynamic semantics

E-type theories

by a free variable by indexing dynamic binding

by a free variable by indexing definite descriptions

dynamic binding definite description works well works well too strong a representation OK no representation not applicable

OK no clear mechanism

not applicable

highly problematic

principle of salience. This principle is necessary for explaining the reference of deictic pronouns and pronouns of laziness, (see (14)). And I will show that it can be extended to the other uses of pronouns as well. This view has been suggested at various times, e.g. Kripke (1991: 95, n. 32): […] ‘he,’ ‘she,’ ‘that,’ etc. can, under various circumstances, refer to anything salient in an appropriate way. Being physically distinguished against the background is a property that may make an object salient; having been referred to by a previous speaker is another.

A similar statement was given by Bach (1994: 315): I argue […] that anaphoric reference, the use of pronouns to refer to individuals previously mentioned, deserves no special semantic or syntactic consideration: being previously mentioned is simply one way of being salient. Unless the anaphoric reference is explicitly marked, by dedicated anaphors like reflexives and reciprocals, it is not a syntactic phenomenon.

The principle of salience is extended to the discourse: A discourse item can be salient due to the physical context or due to another linguistic expression that has been previously mentioned. However, this intuitively attractive view has never been worked out. Before such a theory is developed in Section 5, I insert a short excursion to the treatment of definite NPs, which shows several analogous features. Furthermore, Bloomfield and Quine have already noted that

123


anaphoric pronouns are a kind of “impoverished definite NPs” (see Sections 2.2–2.3). The treatment of anaphoric pronouns as E-type pronouns also showed the close relation between pronouns and definite NPs.

4. Definite NPs and their representations 4.1 Varieties of definite NPs There are different uses or functions of definite NPs, just as there are different functions of pronouns. Among other uses, the following four are the most prominent ones — they can be paired with the uses of pronouns mentioned in (12) above. (14) Different uses of definite NPs and uses of pronouns: i. situationally salient def. NPs deictic pronouns ii. anaphoric def. NPs anaphoric pronouns iii. functional def. NPs functional pronouns no direct correspondence iv. unique def. NPs

In (14i) the definite NP refers to a situationally salient object, as illustrated in sentence (15a) and (15b). Depending on the extra-linguistic context, the definite term refers to different objects. In (14ii), the definite NP is used anaphorically, i.e. it refers to an already established discourse item, as illustrated in (16a), (16b), or in the fragment (11) above. Here, the reference of the NP depends on the linguistic context and may change with the discourse, e.g. the referent of the term the waiter does change in the fragment (11) above. In (14iii), the definite NP is a functional expression and therefore refers to exactly one item (17a). The reference can be determined by other contextual factors, such as time (president at what time?), but this use is less contextually dependent than the uses in (14i) and (14ii). In (14iv), the reference of the definite NP is exclusively determined by the lexical material: There is only one object that fulfills the descriptive content of the NP, as in examples (18a) and (18b): (15) a. The island is beautiful. b. The train left two minutes ago. (16) a. Once upon a time, there was a king, … and the king … b. … except an old man who sat in the shadow […] the old man was a little drunk (see (11)).


(17) a. The father of Bertrand Russell was a rich man. b. The president of the US is white. (18) a. The sun is shining. b. The first man on the moon was an American.

This very short survey of some of the most prominent uses of definite NPs illustrates that the reference of definite NPs is established by different principles: salience, anaphoric linkage, functional concepts or unique descriptions. Different families of approaches to definite NPs have been developed. Each approach takes one of the four principles as an essential principle and extends it to all other cases. In the following section I only present the classical theory of definite description developed by Russell (1905).5 4.2 The theory of definite NPs Russell’s (1905) Theory of Descriptions is the clearest and the best understood approach to the semantics of definite NPs. It gives a clear formal representation of definite and indefinite NPs as quantifier phrases. The definite article and its indefinite counterpart are defined in the context of a sentence as quantifiers — their lexical meaning can be extracted via lambda-abstraction, as in (19). The indefinite article expresses existence, while the definite article expresses existence and uniqueness. The complex quantifier representing the definite article can be abbreviated as the iota-operator “ι”: (19) The lexical meaning of the articles according to Russell (1905): a = λQ λP $x [Q(x) & P(x)] the = λQ λP $x "y [(Q(y) ´ x = y) & P(x)] = λQ λP [P(ιx [Q(x)])] (according to a context-definition)

Examples (20)–(24) illustrate how the semantics of the articles combine in a sentence with the descriptive material of the NPs. In (20a) the indefinite NP a girl is translated into the existential quantifier, while the definite NP the girl is translated into a iota-term in (20b) that includes an additional predicate RESTR. This predicate is necessary since there is more than one girl. Such a restriction must be explicitly or implicitly added to nearly any definite NP. The indefinite NP a girl in (21a) shows scope interactions with the universal quantifier in (21b) and (21c), while the definite NP the girl with the red hair can only receive wide scope with respect to the universal quantifier. However, the definite NP the boy that brings her flowers in (23a) has narrow scope with respect to the universal quantifier. In this example it is not even clear that there is a

125


unique boy for each girl. The representations (24b) of the two definite NPs in (24a) raise additional questions: the Queen is a queen of some country — here indicated by the constant c, that must be given by the context. The definite NP the island on Lake Constance is not well-formed since there are at least three islands on that lake (see below for a detailed discussion): (20) a. A girl laughs. The girl has red hair. b. $x [Girl(x) & Laugh(x)] & Red_hair(ιx [Girl(x) & RESTR(x)]) (21) a. Every boy dates a girl. b. "x [Boy(x) Æ $y [Girl(y) & Date(x,y)]] c. $y [Girl(y) "x [Boy(x) Æ & Date(x,y)]] (22) a. Every boy dates the girl with the red hair. b. "x [Boy(x) Æ Date(x, ιy [Girl(y) & Red_hair(y)])] (23) a. Every girl dates the boy that brings her flowers. b. "x [Girl(x)ÆDate(x, ιy [Boy(y) & Bring_flower(y,x)])] (24) a. The Queen visits the island on Lake Constance. b. Visit(ιx[Queen(x,c)], ιy [Island_on_Lake_Constance(y)])

4.3 Definiteness and Salience Russell’s Theory of Descriptions cannot analyze all uses of definite NPs, even with the modification of adding an additional domain restriction. Therefore, a more general approach is necessary, which takes the situational use (14i) as the central one of definite NPs. The salience approach essentially incorporates contextual information into the representation of a definite expression. The contribution of the context to the interpretation of the definite NP consists of a salience hierarchy. It is assumed that each context can be associated with an ordering among the elements of subsets of the domain of discourse. The definite NP the F denotes the most salient F according to the situation i. This representation completes the ideas of discourse representation theories by producing a more comprehensive picture: a definite NP is not only linked to an already introduced discourse referent, rather it is linked to the most salient discourse referent of the same kind so far. The idea of a salience hierarchy as part of contextual information can be illustrated by the following. Let us assume that there are three islands on Lake Constance: Mainau, Reichenau, and Lindau. Depending on my location, the three islands are ranked differently. The definite NP the island always refers to the highest ranked element. For example, located in Constance, I probably refer to the Mainau with the definite NP the island,

Reference and representation of pronouns 127

while the same term refers to the Reichenau if uttered in Allensbach (a small village opposite to that island), etc. (25) The islands on Lake Constance: Mainau, Reichenau, Lindau Location

Salience hierarchy

Constance Mainau > Reichenau > Lindau Allensbach Reichenau > Mainau > Lindau Friedrichshafen Lindau > Reichenau > Mainau

Reference Mainau Reichenau Lindau

The salience theory of definiteness has three historical sources: First, Lewis (1979) criticizes Russell’s Theory of Descriptions and sketches an alternative theory using a salience ranking instead of Russell’s uniqueness condition. Secondly, the investigations of the Prague School (cf. Sgall et al. 1973; Hajicˇová et al. 1995) developed an information structure of a sentence the pragmatic background of which is a hierarchy of “activated” referents. Thirdly, research in artificial intelligence has shown that discourse models need a structure or hierarchy of referents that is very similar to Lewis’ concept of salience (cf. Grosz et al. 1995). In the following, I refer to Lewis’ basic ideas: Lewis (1970: 63) mentions the concept of salience in the philosophical and linguistic discussion of the Russellian Theory of Descriptions: Second, consider the sentence ‘The door is open’. This does not mean that the one and only door that now exists is open; nor does it mean that the one and only door near the place of utterance, or pointed at, or mentioned in previous discourse, is open. Rather it means that the one and only door among the objects that are somehow prominent on the occasion is open. An object may be prominent because it is nearby, or pointed at, or mentioned; but none of these is a necessary condition of contextual prominence. So perhaps we need a prominent-objects coordinate, a new contextual coordinate independent of the other. It will be determined, on a given occasion of utterance of a sentence, by mental factors such as the speaker’s expectation regarding the things he is likely to bring to the attention of his audience.

Lewis (1979: 178) rejects Russell’s uniqueness condition for definites or any modified version of it: “It is not true that a definite description ‘the F’ denotes x if and only if x is the one and only F in existence. Neither is it true that ‘the F’ denotes x if and only if x is the one and only F in some contextually determined domain of discourse.” The definite NP must refer uniquely according to another and more general principle. Lewis (1979:178) names this principle ‘salience’:


The proper treatment of description must be more like this: ‘the F’ denotes x if and only if x is the most salient F in the domain of discourse, according to some contextually determined salience ranking.

4.4 Salience, epsilon terms, and choice functions The concept of salience was never formally reconstructed although it was informally regarded as an essential part of fixing the referent of definite expressions. In this section, I develop a formal reconstruction of salience by means of indexed epsilon terms that are interpreted as context dependent choice functions. Choice functions have recently become a fashionable tool for representing indefinites (cf. Reinhart 1992, Winter 1997, Kratzer 1998, von Heusinger 2000). I will not discuss this aspect of choice functions here, but concentrate on the application of choice functions to definite expressions. As semantic characterization for the definite article I use the epsilon operator that was introduced into metamathematics by Hilbert and Bernays (1939). The epsilon operator corresponds to a choice function that assigns to each nonempty set one element of this set. An empty set will be assigned an arbitrary element. Like the iota operator, the epsilon operator forms a term from a predicate. Unlike the iota operator, it does not express the uniqueness condition. The main difference may be shown by the formalization and the paraphrase of the description the island, as given in (26a) and (26b): (26) a. ιx [island(x)] the unique x, such that x is an island b. εx [island(x)] the selected x, such that x is an island

Definite descriptions as epsilon terms do not require any uniqueness condition like the Russellian definite descriptions. Since the uniqueness condition has raised several problems for definite descriptions, as well as E-type pronouns, epsilon terms are the more adequate representation for definite descriptions. However, in the representation (26b), they are not flexible enough. Depending on a physical context or a certain discourse, a definite NP can refer to different objects. Therefore, we assume different salience structures and incorporate different choice functions into our model: M = ·D, I, Φ1,Φ2, Φ3, … ΦnÒ. A choice function Φi is a function that assigns to a set s one of its elements, as in (27). On the side of the syntactic representation, we mark the particular contextually given salience structure by an index on the epsilon term. The definite NP the F is represented as an epsilon term which depends directly on the given context (standing for a salience hierarchy). It is interpreted as the

Reference and representation of pronouns 129

operation of the given or actual choice function (= salience ranking) applied to the set of Fs. It yields the most actually salient F: (27) Φ(s) Œ s if s π Ø and Φ(s) Œ D if s = Ø (28) the F: εix Fx

M,g

= Φi( F

M,g

) with i contextually determined

The salience hierarchy can also be changed by the linguistic context as assignment functions are dynamically updated (see Section 2.5). An NP like a girl updates the current choice function Φ (representing the salience structure) by promoting the referent d of a girl to the most salient individual of the set of girls. This can be formalized as Φ[ Girl ,d], meaning that this function selects out of the set of girls the object d; for all other sets it selects the same element as the original function Φ. The context change potential of linguistic expressions is modeled along the same lines as in dynamic semantics except for the assumption that the choice function Φ is updated, rather than the assignment function g; see Peregrin and von Heusinger (in press) for a detailed analysis. With this representation for definite NPs in general, we can account for the particular cases mentioned in (20)–(24) by one and the same representation: an indexed epsilon term. Here I present only the cases that are problematic for the Russellian analysis with the uniqueness condition: The anaphoric definite NP the girl in (20a) is represented as an indexed epsilon term. The interpretation depends on an updated choice function Φ[ Girl ,d] such that the referent for the indefinite is also the referent for the definite. Thus we do not need to insert any additional restriction, as in (20b) above. The situational dependency of the definite NP the island in (24a) is also resolved by the dependency on the given context i in (24c). (20) a. c.

A girl laughs. The girl has red hair. $x [Girl(x) & Laugh(x)] & Red_hair(εix [Girl(x)])

(24) a. c.

The Queen visits the island on Lake Constance. Visit(εix[Queen(x,c)], εiy [Island_on_Lake_Constance(y)])

In this section, I have argued that most semantic theories assume a uniform representation for definite NPs even though the same theories do not give a uniform representation for pronouns, which show a similar variety of uses. I presented the Russellian uniqueness theory and argued that it only covers a restricted set of definite NPs. Alternatively, I developed the salience theory of definiteness as a formal reconstruction of definiteness. The reconstruction of the principle of salience was presented with choice functions, and definite NPs received a uniform representation as indexed epsilon terms. The problematic


uniqueness condition was replaced by the principle of choice, and the contextual influence on the interpretation of definite NPs was acknowledged by introducing the situational index on the epsilon term, indicating different possible salience structures.

5. Pronouns and salience Anaphoric pronouns in third person singular are “impoverished” definite NPs. This claim can be traced back through literature. Bloomfield (1933: 251 (§ 15.4)) notes that “[deictic, KvH] he implies that the individual is identifiable from among all the individuals of the species mentioned; this element of meaning is the same as that of the syntactic category of definite nouns […]” (see also the discussion in Section 2.2). Quine (1960: 103 (§21)) observes that “minimum descriptions are abbreviated as ‘he’, ‘she’, ‘it’. Such a pronoun may be seen thus as a short singular description […]” (see the discussion in Section 2.3). Finally, the discussion of E-type pronouns in Section 2.6 has shown that (at least) certain pronouns are best represented as definite NPs. In this final section, I apply to pronouns the salience semantics developed in the last section for definite NPs. In the previous section, it was shown how this salience structure is reconstructed using choice functions as part of the model. There are two sides of this salience structure: on the one hand, NPs change or update a given salience structure, and on the other hand, definite NPs are interpreted according to the (updated) salience structure. In this view, the salience structure is that part of the context that is necessary to interpret definite expressions. We can now introduce a uniform representation of pronouns as indexed epsilon terms with minimal descriptive content. For example, in English the pronoun he expresses that it is the most salient object of the set of male individuals, she the most salient one of the set of female individuals, etc. (29) a. he: εix [Male(x)] = Φi( Male ) with i contextually determined b. she: εix [Female(x)] = Φi( Female ) with i context. determined c. it: εix [Neuter(x)] = Φi( Neuter ) with i context. determined6

This representation of pronouns depends on the salience structure of the context. In order to allow for anaphoric links to other expressions, the salience change potential of NPs must be modified. In our first version, an NP such as a girl updates the salience for the set of girls. However, for an anaphoric link


with a pronoun, the NP must also change the salience structure of the set of all female individuals. So it is assumed that an NP not only updates the salience structure for the set it describes, but also for supersets. This is illustrated by the following situation (for a more detailed account, see Peregrin and von Heusinger, in press). Assume that we have three girls: Ann, Babs and Chris, and two women: Mary and Nancy. Our starting choice function Φi selects Babs out of the set of girls (for the girl), Mary out of the set of women (for the woman), and Chris out of the set of all female individuals (for she), as illustrated in (30). Now we utter the sentence (20a), repeated as (31a). The salience change potential of a girl updates the salience structure for the set of girls and the set of all female individuals — for both sets the referent of a girl, let’s say Ann, becomes the most salient one. (30) situation:

Girl = {ann, babs, chris} Woman = {mary, nancy} Female = {ann, babs, chris, mary, nancy} the girl: Φi( Girl ) = babs the woman: Φi( Woman ) = mary she: Φi(εix [ Female(x) ]) = chris

(31) a. A girl laughs. Φi is updated to Φi+1 = Φi[ Girl ,ann][ Female ,ann] b. She has red hair. she: Φi+1( Female ) = ann c. The girl has red hair. she: Φi+1( Girl ) = ann

With this uniform representation for pronouns and the salience change potential of NPs, we can account for the different uses discussed in Section 3.1. In the case of a deictic pronoun as in (1a), repeated as (32a), the index of the epsilon term is determined by the context, as in (32b). Thus the term refers to the most salient female individual in the context according to the given salience structure. The pronoun of laziness he in the second sentence of (1a), repeated as (33a), is also represented as an indexed epsilon term. Here the index i is marked with “+1” indicating that the salience structure was updated by previously mentioned John: John has become the most salient man such that the pronoun refers to John. The E-type pronoun case works similarly, except that here the indefinite NP a man has raised its referent to salience such that the updated salience structure takes it as the most salient male individual. The pronoun refers to it via its representation as an epsilon term. (32) a. John met her for dinner. b. Met_for_dinner(john, εix [Female(x)]) the most salient female in the context

deictic pronoun

131

132

Klaus von Heusinger

(33) a. He had a wonderful evening. b. Have_a_wonderful_evening(εi+1x [Male(x)]) (34) a. A man walks. He whistles. b. … & Whistle(εi+1x [Male(x)]) i+1 the salience hierarchy updated by a man

pronoun of laziness E-type pronoun

With this very short sketch on how to present deictic and anaphoric pronouns with indexed epsilon terms, I conclude the argument. The overall picture of the different groups of pronouns can be given as in (28): All different uses are governed by the principle of salience, depending on contextual or discursive means. Functional pronouns take functional concepts as the domain of choice. pronouns

(35)

salience Contextual salience (i) deitic

anaphoric salience (ii)-(iv) anaphoric

functional salience (v) anaphoric

6. Summary In this paper I have argued that pronouns have a uniform semantic representation, even though they show quite different referential behavior. Pronouns are understood as referring to the most salient individual in the context so far, and are represented as indexed epsilon terms interpreted by a choice function. The choice function reconstructs the salience structure of the context. NPs change the salience structure by updating the choice function for the set they describe and some supersets. Interpreted by the updated choice function, the pronoun refers to the same referent as its antecedent. Thus, the semantics of pronouns in the third person is explained along the same lines as the semantics of definite NPs.

Notes * I would like to thank the workshop-organizers as well as the audience for helpful comments on the paper. I am also grateful to two anonymous reviewers for their detailed and constructive comments and requests for clarification. Many thanks to Ruth Kempson for her clarifying


comments, and to Miriam Butt for proofreading. The research was supported by a Heisenberg fellowship of the Deutsche Forschungsgemeinschaft (German Science Foundation). 1. I focus on pronouns in the third person singular. Pronouns in the first and second person do not exhibit such a rich referential behavior as they are primarily deictic expressions. Plural pronouns show additional complications of forming sets from explicitly or implicitly given discourse referents (cf. Kamp and Reyle 1993: Ch. 4). However, this process of set-formation is independent from the referential properties of the pronouns. Therefore, I concentrate on third person singular pronouns in this study. 2. There is no formal account of how this free variable is instantiated — by binding or by contextually given information. However, there seems to be some restriction that derives from the antecedent expression: The it stands for his paycheque such that the missing argument must be male. 3. One reviewer noted that pronouns are resolved according to different pragmatic principles like parallelism, grammatical function of the antecedent, level of embedding etc., among which salience is only one parameter. I do not assume that salience is the unique principle for resolving pronouns, but I assume that it is a semantic parameter that can be formally reconstructed, as will be shown below. Ruth Kempson made me aware of the fact that salience is not just “recency” and has to be computed (in some sense) in conjunction with derivability of appropriate inferential effects. See Sperber and Wilson (1986) for an informal description of this interaction. For the present study I abstract away from this additional complication. 4. Purely syntactic pronouns such as reflexives or reciprocals as well as non-referential pronouns are not included in this list. I do not have anything to say about non-referential pronouns (but see Bosch 1983: Ch. 4). For reflexives and reciprocals, I assume that they are treated according to class (iii) as bound variables with some additional restrictions. The focus of this investigation are the referential properties of pronouns — so I concentrate on the cross-sentential occurrence. 5. Besides Russell’s classical Theory of Descriptions, the most influential theories are: Heim (1982) and Kamp’s (1981) Familiarity Theory, and Löbner’s (1985) functional approach to definite expressions. Although the theories are confronted with the multiple uses of definite NPs, they assume that there is only one underlying meaning of the definite NP that can be found in all of its uses. However, each of the theories chooses a different use of definite NPs as the primary one and gives an adequate analysis of this use. The analysis is then extended to the other uses. The three theories are successful in their primary area, but they cannot convincingly describe other uses of definite NPs. Therefore, a more general approach becomes necessary, which is presented in Section 4.3. 6. A reviewer noted that the use of the English pronouns it indicates that the referent is [−human], rather than the grammatical category neuter.

133


References Bach, E. 1970. “Problominalization”. Linguistic Inquiry 1: 121–122. Bach, K. 1994. Thought and Reference. Oxford: Clarendon. Bloomfield, L. [1933] 1967. Language. London: Allen and Unwin. Bosch, P. 1983. Agreement and Anaphora. New York: Academic Press. Bühler, K. [1934] 1982. Sprachtheorie. Die Darstellungsfunktion von Sprache. Mit e. Geleitw. von Friedrich Kranz. [Ungekürzter Neudr. d. Ausg. Jena: Fischer 1934]. Stuttgart and New York: Fischer. Cooper, R. 1979. “The interpretation of pronouns”. In Syntax and Semantics 10: Selections from the Third Groningen Round Table, F. Heny and H. S. Schnelle (eds), 61–92. New York: Academic Press. Egli, U. 2000. “Anaphora from Athens to Amsterdam”. In Reference and Anaphoric Relations, K. von Heusinger and U. Egli (eds), 17–29. Dordrecht: Kluwer. Evans, G. 1977. “Pronouns, quantifiers and relative clauses (I)”. Canadian Journal of Philosophy 7: 467–536. Evans, G. 1980. “Pronouns”. Linguistic Inquiry 11: 337–362. Geach, P. 1962. Reference and Generality. An Examination of Some Medieval and Modern Theories. Ithaca, NY: Cornell University Press. Groenendijk, J. and Stokhof, M. 1991. “Dynamic predicate logic”. Linguistics and Philosophy 14: 39–100. Grosz, B., Joshi, A. and Weinstein, S. 1995. “Centering: A framework for modeling the local coherence of discourse”. Computational Linguistics 21: 203–225. Hajicˇová, E., Hoskovec, T. and Sgall, P. 1995. “Discourse modeling based on Hierarchy of Salience”. The Prague Bulletin of Mathematical Linguistics 64: 5–24. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Diss., University of Massachusetts, Amherst. Heim, I. 1990. “E-type pronouns and donkey anaphora”. Linguistics and Philosophy 13: 137–177. Hemingway, E. [1925] 1966. “A clean, well-lighted place”. In The short stories of E. Hemingway, 379–383. New York: C. Scribner. von Heusinger, K. 2000. “The reference of indefinites”. In Reference and Anaphoric Relations, K. von Heusinger and U. Egli (eds), 247–265. Dordrecht: Kluwer. Hilbert, D. and Bernays, P. [1939] 1970. Grundlagen der Mathematik. Vol. II. 2nd ed. Berlin, Heidelberg and New York: Springer. Kamp, H. 1981. “A theory of truth and semantic interpretation”. In Formal Methods in the Study of Language, J. Groenendijk, T. Janssen and M. Stokhof (eds), 277–322. Amsterdam: Mathematisch Centrum. Kamp, H. and Reyle, U. 1993. From Discourse to Logic. Introduction to Model Theoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer. Karttunen, L. 1969. “Pronouns and variables”. Papers from the Fifth Regional Meeting of the Chicago Linguistic Society, 108–115. Chicago: University of Chicago Press. Karttunen, L. 1971. “Definite descriptions with crossing coreference”. Foundations of Language 7: 157–182.


Kratzer, A. 1998. “Scope or pseudoscope? Are there wide-scope indefinites”. In Events and Grammar, S. Rothstein (ed.), 163–196. Dordrecht: Kluwer. Kripke, S. [1977] 1991. “Speaker’s reference and semantic reference”. In Pragmatics: A Reader, S. Davis (ed.), 77–96. Oxford: Oxford University Press. [First published in: Midwest Studies in Philosophy 2: 255–276.] Lasnik, H. 1976. “Remarks on coreference”. Linguistic Analysis 2: 1–22. Lewis, D. 1970. “General semantics”. Synthese 22: 18–67. Lewis, D. 1979. “Scorekeeping in a language game”. In Semantics from Different Points of View, R. Bäuerle, U. Egli and A. von Stechow (eds), 172–187. Berlin, Heidelberg and New York: Springer. Löbner, S. 1985. “Definites”. Journal of Semantics 4: 279–326. Neale, S. 1990. Descriptions. Cambridge, MA: MIT Press. Peregrin, J. and von Heusinger, K. in press. “Dynamic semantics with choice functions”. In Context Dependence in the Analysis of Linguistic Meaning, H. Kamp and B. Partee (eds) Amsterdam: Elsevier Science. Quine, W. V. 1960. Word and Object. Cambridge, MA: MIT Press. Reinhart, T. 1992. “Wh-in-situ: An apparent paradox”. In Proceedings of the Eighth Amsterdam Colloquium, P. Dekker and M. Stokhof (eds), 483–491. ILLC, University of Amsterdam. Russell, B. 1905. “On denoting”. Mind 14: 479–493. Schneider, R. [1878] 1965. Apollonii Dyscoli quae supersunt. Scripta minora a Richardo Schneider edita continens [reprint of the ed. 1878/1902] [Grammatici graeci 2,1]. Hildesheim: Olms. Sgall, P., Hajicˇová, E. and Benesová, E. 1973. Topic, Focus and Generative Semantics. Kronberg, Taunus: Scriptor. Sperber, D. and Wilson, D. 1986. Relevance. Oxford: Blackwell. Winter, Y. 1997. “Choice functions and the scopal semantics of indefinites”. Linguistics and Philosophy 20: 399–467.

135

The dynamics of syntax Anaphora, relative pronouns and crossover* Ruth Kempson and Wilfried Meyer-Viol King’s College London

In this paper, we show how incorporating into syntax the concept of structural underspecification and the dynamics of its resolution enables us to characterise the interaction between construal of anaphoric expressions and relative clauses displayed by crossover and resumptive pronoun phenomena without any stipulation specific to either. The account will be developed within the Dynamic Syntax (DS) framework (Kempson et al. 2001) which models the process of interpreting a string as a left-right process of logical form construction, where the resulting logical form is represented as a tree structure. What we shall argue is that general concepts of tree growth, together with anaphora resolution as a process of term substitution, provide a basis for characterising crossover effects in an integrated way while sustaining a unitary account of anaphora as a pragmatic process. Resumptive pronoun construal emerges as epiphenomenal, not requiring structure-specific stipulation.

1.

The flow of language understanding

According to Dynamic Syntax, the process of natural language understanding is a monotonic process of building up a logical form following the left-right sequence of words, where the goal is to establish some propositional formula as interpretation (= ?Ty(t)) as the root of a tree. The various nodes of this tree are decorated with sub-terms of this propositional formula. These are not words of the language under analysis; they are expressions of a typed lambda calculus — representations of content, each a value of the Fo predicate (Fo for ‘formula’). The process is one of incrementally building up interpretation in two stages: first, setting out the tree structure as directed by the overall goal and additional

138

Ruth Kempson and Wilfried Meyer-Viol

subgoals (so-called ‘requirements’); second, annotating nonterminal nodes in the tree by steps of labelled type deduction defined over values of the Ty predicate (Ty for ‘logical type’) once there are appropriate decorations on the terminal nodes. Intrinsic to this process are three forms of syntactic underspecification, each of which has to be resolved during the construction process. First, nodes are introduced with requirements to be fulfilled later, as displayed in Figure 1, starting with a single node decorated a requirement indicating the overall goal to be achieved — ?Ty(t). Each node then introduced has a requirement (the node marked with the pointer is the node currently being developed). Notice how in Figure 1 each partial tree in the sequence is a development of the previous one. It is characteristic of the requirements that drive this development process that they may be satisfied substantially later in the construction 1)

?Ty(t), ◊

2)

?Ty(t) ?Ty(e), ◊

4)

?Ty(e → t)

PAST: ?Ty(t) Fo(John)

?Ty(e → t)

Fo(lxly[Upset(x)(y)]) 5)

?Ty(e), ◊

PAST: ?Ty(t) Fo(John)

?Ty(e → t), ◊

Fo(λxλy[Upset(x)(y)]) 6)

Fo(Mary)

Fo(PAST: Upset(Mary)(John)), ◊ Fo(John) Fo(λy[Upset(Mary)(y)]) Fo(λxλy[Upset(x)(y)])

Note

Ty(e) Ty(e → t)

Fo(Mary)

the semantic type of DP the semantic type of VP.

Figure 1.Parsing John upset Mary.

?Ty(t)

3) Fo(John)

?Ty(e → t), ◊

The dynamics of syntax 139

process than the point at which they are introduced. To pick out just the most extreme case, notice that the goal ?Ty(t) introduced at step (1) in Figure 1 is not met until the final step. A second form of underspecification is that nodes can be introduced into a tree as initially unfixed, characterised only as dominated by the top node, their position in the tree being fixed later in the construction process. This is the analysis proposed for the core cases of long-distance dependency.1 Again, we display the phenomenon graphically — in Figure 2. (1) Mary, John upset (a) After processing Mary:

?Ty(t)

Fo(Mary), ◊ (b) Having processed John: Fo(Mary)

?Ty(t)

Fo(John)

(c) After processing upset: Fo(Mary)

?Ty(e → t), ◊

?Ty(t)

Fo(John) Fo(Upset)

?Ty(e), ◊

(d) Merging the unWxed node and the object node: ?Ty(t), ◊ Fo(John)

?Ty(e → t)

Fo(λxλy[upset(x)(y)])

Fo(Mary)

Figure 2.Parsing left-dislocation structures.

At step (a), the node annotated by the word Mary is introduced as dominated by the top node without fuller specification of its relation to the top node (hence at that juncture ‘unfixed’ within that tree). At step (d), this node is merged with some further node whose relation to the top node is determined, hence establishing its precise role in the logical structure. As Figures 1 and 2 jointly display, different strings may lead to the same decorated tree, the

140 Ruth Kempson and Wilfried Meyer-Viol

difference between them arising from the different sequence of steps leading to that result. As the sequence of trees in Figure 2 shows, the structure at the endpoint in the growth process involves simpler resulting trees than in frameworks in which all syntactic properties are assigned a direct reflex in the configuration itself, for the burden of explanation is split between the final tree and the process of building it up. There are no functional heads, and no analogue to the CP/IP distinction. Instead there is progressive introduction and decoration of nodes in a tree, any one node in the tree possibly being decorated by more than one node; and there is also the licence to introduce initially unfixed nodes that get merged with some fixed point in the tree structure subsequently. Interacting with both these structural forms of underspecification is the third form of underspecification: the interpretation lexically provided by a pronoun. Reflecting the fact that the lexical content of a pronoun underspecifies its interpretation in context, a pronoun is defined to annotate a node with a meta-variable as Fo value to be substituted by a copy of some selected term. This process is taken to be pragmatic, restricted only in so far as locality considerations distinguishing individual anaphoric expressions preclude certain formulae as putative antecedents for any such copy process:2 (2) Q: Who upset Mary? Ans: John upset her.

Thus in processing the pronoun in (2), the object node is first decorated with a meta-variable U within Fo(U), this being replaced by a copy of some other term, e.g. Mary, copied from the structure constituting the interpretation of the previous sentence. Notice that the substituend for the meta-variable is not the English word Mary but the term taken to represent the individual referred to by that word in the given context. This reflects the “representationalist” view that natural language interpretation is projected as a structured representation (see Fodor 1981, 1983; and see Kempson et al. 1999, Kempson et al. 2001: Ch. 1 for evidence for this account of anaphora). Terms such as pronouns that underspecify their interpretation as assigned in context are accordingly analysed as projecting from the lexicon a form that is enriched in such contexts by a process of substitution. 1.1 The formal framework 1.1.1 Decorated partial trees As Figures (1)–(2) have informally displayed, decorated partial trees are progressively constructed, each node of which is initially decorated with

The dynamics of syntax

requirements and subsequently with annotations. Such trees are described by a modal logic of finite trees (LOFT — see Blackburn and Meyer-Viol 1994), with three basic operators (denoting daughter and mother relations, and a relation between one tree and another) and further operators defined in terms of these: ·ØÒ, ·Ø0Ò, ·Ø1Ò, ·≠Ò, ·Ø*Ò, ·≠*Ò, ·LÒ, ·L−1Ò, ·DÒ, ·UÒ Each operator is interpreted by a discrete relation between nodes in a tree: e.g. the modality ·ØÒ is evaluated over the daughter relation — ·ØÒTy(eÆt) ‘holds’ on a node n if there is a daughter where Ty(eÆt) holds; ·≠Ò over the mother relation. More specifically, LOFT has ·Ø0Ò ·Ø1Ò interpreted over first and second daughters respectively corresponding to nodes decorated with argument and functor types, ·Ø*Ò is interpreted over the dominance relation (the reflexive transitive closure of the daughter relation), ·≠*Ò over the inverse of dominance, ·LÒ over a relation of LINK between trees (see Section 1.2 of this chapter), ·L−1Ò over its inverse, and finally ·DÒ, the weakest relation, is interpreted as picking out any relation between nodes (the reflexive transitive closure of the union of daughter and LINK relations). The decorations that may hold at a node include specification of a value for the formula predicate Fo, a type specification, expressed as an argument of the predicate Ty, a tree node position — represented as an argument of the predicate Tn. Thus if ·DÒFo(Run) holds at a node n, there is some node m that can be reached from n following daughter and link relations arbitrarily far, and Fo(Run) holds at m. Included within possible specifications are meta-variables, being place-holders for some fixed value to be provided. As a tree description language, LOFT can express the types of tree relations between nodes expressible in other formalisms. For example, the concept of locality associated with the construal of pronouns can be defined as a relation between some node and a dominating node Tn(a) such that the following holds at the node to be decorated by the pronoun: ·≠0Ò·≠*1ÒTn(a) That is, a node is local to some top node Tn(a) if the path between this top node and the node in question lies along a (possibly empty) chain of functor relations and one argument-daughter relation. Choice of a value for some meta-variable projected by a pronoun can then be precluded should the putative antecedent be at some node with address Tn(a) along such a chain. The specific and novel advantage of LOFT emerges from the use of the LOFT operators in combination with a generalisation of the concept of requirement ?X to any decoration X. This combination makes it possible to describe

141


partial trees which have requirements on a tree node which are modal in form, which will be fulfilled by some other node having a given annotation: a lexically defined restriction may be imposed upon one node that to be satisfied must involve a decoration of some distinct node possibly arbitrarily far. The requirements that may be imposed are thus by no means restricted to nonmodal requirements such as ?Ty(e), or simple modal requirements, such as ?·Ø1ÒTy(eÆt). To the contrary, any formula may be used to express a requirement. For example, while ·Ø*ÒFo(α) decorating a node n as an annotation implies that n dominates a node where Fo(α) holds, ?·Ø*ÒFo(α) decorating node n implies that Fo(α) is required to hold at a node dominated by n (literally, there is a requirement on successful completion of node n that there be a node dominated by n annotated by Fo(α)). By this means, requirements may constrain subsequent development of the tree from a node at some arbitrary remove; and this provides an additional mechanism for pairing noncontiguous expressions according as one expression imposes some requirement on a node which is secured by a decoration on some discrete node by the other. This gives a much greater flexibility than is standard. In particular, complementisers may impose complex requirements on the top node of a newly introduced tree, to be met in virtue of some annotation on a node in the subsequent construction of that tree. 1.1.2 The dynamics of tree growth LOFT is a language for describing (partial) trees. To describe the tree growth process, we define transitions between partial trees. There are three types of action: computational actions, which are general (albeit possibly languagespecific); lexical actions, which are associated with individual words; and pragmatic actions, which are substitution operations, using terms/structure antecedently available. To exemplify the pattern of computational actions defined, we list *Adjunction which licenses the introduction of an unfixed node: *Adjunction {{Tn(a), …?Ty(t), }} {{Tn(a), …, ?Ty(t)}, {·≠*ÒTn(a), …, ?Ty(e), }} A rule like this should be read as follows: this transition is defined as starting from a partial tree (described as a structured set of nodes) containing only one node (described by the set of formulae holding at that node), here some arbitrary tree node a with requirement ?Ty(t). The transition then adds to that one-member set the node described as being dominated by a, ·≠*ÒTn(a),


requiring a type e decoration, with the pointer indicating that it becomes the node currently under development. This rule conforms to what is an invariant pattern of defining information-preserving transitions from partial tree to partial tree. Any node introduced by this rule has ultimately to be assigned a fixed tree position by a process, Merge, which unifies tree nodes. Characteristically Merge takes place (as displayed in Figure 2), where co-present in a tree are an unfixed node annotated with a formula of a certain type and a fixed node requiring that type. Lexical actions defining the contribution of individual words are, equally, procedures for updating partial tree descriptions. The lexical specifications are of the form ·IF , THEN α1, ELSE α2Ò, with the ‘IF’ condition specifying that must hold on the node at which the pointer resides if the possible compound action given by ‘THEN’ is to be carried out — viz. α1. If does not hold, then the action α2 is carried out. For example, the conditions for the actions induced by the English verb upset require the pointer to be at a node decorated with the requirement ?Ty(eÆt), from which it initiates the addition of a subtree — a daughter node annotated with Fo(Upset) — and the addition to its mother of the requirement for a daughter decorated by an argument:3 upset IF {?Ty(eÆt)} THEN go(·≠Ò), put(Tns(PAST)), go(·Ø1Ò), make(·Ø1Ò);go(·Ø1Ò); put(Fo(Upset),Ty(eÆ(eÆt)), [Ø]^); go(·≠1Ò); put(?·Ø0Ò(Ty(e))) ELSE ABORT If the condition is not met, the sequence of actions aborts. Pronouns illustrate how the structure projected by a word may underdetermine content. They supply a meta-variable which has to be replaced by some fixed value to yield a well-formed output.4 Notice the specification of case, here nominative, as imposing a constraint on tree position that the mother node in the resulting tree be of type t: he IF {?Ty(e)} THEN put(Fo(U), Ty(e), ?·≠0ÒTy(t), [Ø]^) ELSE ABORT


Both lexical specifications determine through the annotation ‘[Ø]^’ that the annotated node in question is the terminal node of a tree, a general property of lexical items. It is the interaction of computational, lexical and pragmatic processes which determines the interpretation of a string. A well-formed string is one for which at least one logical form can be constructed from the words in sequence within the context of a given class of computational and pragmatic actions with no requirements outstanding. In consequence, as we shall see, the imposition of requirements and their subsequent satisfaction are central to explanations to be given. 1.2 Linked structures and relative clauses The Dynamic Syntax framework also licenses the construction of pairs of trees in tandem connected by a ‘LINK’ relation, described by the operator ·LÒ. This adjunction introduces the top node of a new tree and copies information from one tree to the other. Taking nonrestrictive relatives as the most transparent case, consider the steps involved in projecting the construal of (3) (displayed in Figure 3): (3) John, who I like, chain-smokes.

Having processed the word John to yield a partial tree in which the formula Fo(John) annotates a subject node in some tree (the ‘head’ node), a transition is licensed which builds a LINK relation from that node, introducing a new (LINKed) tree with top node decorated with the requirement ?Ty(t) plus the requirement for an occurrence of the formula Fo(John) at some node, without further specification as to where in the newly introduced tree that might be:5 Link Adjunction (English)

{{

head

{ .. {X, Fo(a), Ty(e), ◊} } { .. {X, Fo(a), Ty(e)}, head

{

{

{〈L–1〉X, ?Ty(t)} {〈↑∗〉〈L–1〉X, ?Fo(a), ?Ty(e), ◊} } linked node

unWxed node


The relative pronoun who duly provides the necessary copy, in (3) construed as Fo(John): whorel IF {?Ty(e), ·≠*Ò·L−1ÒFo(x)} THEN put(Fo(x), Ty(e), [Ø]^) ELSE ABORT {Tn(a), ?Ty(t)} {〈↑0 〉 Tn(a), Fo(John)}

{?Ty(e → t)}

{〈L–1〉〈↑0〉 Tn(a), ?Ty(t)} {〈↑* 〉〈L–1〉〈↑0 〉Tn(a), ?Fo(John), ◊}

Figure 3.Building a LINK transition with LINK Adjunction.

Notice how the rule feeds the update provided by the relative pronoun. The process of tree construction then proceeds as in the simpler case, with the initially unfixed node having its position in that tree established in due course through the process of Merge. This rule also applies to restrictive relative clause construals. The internal structure resulting from construal of NP sequences contains two nodes of type e: a node annotated by a variable, (which is introduced in parsing the noun), and a node projected by the NP as a whole (see Figure 4).6 In both restrictive and nonrestrictive construals, the wh-relativiser serves the anaphoric function of ensuring the presence of the copy in the LINKed structure. Restrictive relatives involve a copy of the variable, nonrestrictive relatives involve a copy of the formula decorating the containing node of type e, a formula built up from the subparts of that NP sequence. Figure 4 displays the process of Link Adjunction for the case of restrictive relative construal, imposing a requirement on an unfixed node in the newly introduced linked structure, as in Figures 3 and 4, a requirement fulfilled by the relative pronoun who which copies the variable as the head. (4) the mani whoi Sue likes ei

The output of any such construction process is a pair of linked structures, the secondary LINKed structure once complete leading to the construction of a complex predicate decorating the type cn node as in Figure 5.7

145


{?Ty(e)} {Fo{λP(the P)), Ty(cn → e)}

{?Ty(cn)} {Fo(U), Ty(e)}

{Fo(λY (Y, Man(Y))), Ty(e → cn)}

{〈L–1〉 X, ?Ty(t), ◊}

{〈↑* 〉〈L–1〉X, ?Fo(U), ?Ty(e)}

Figure 4.Structure resulting from LINK-Adjunction in (4). ?Ty(e), ◊ Fo(λP(the, P))

Fo((U, Man(U)) ∧ Like(U)(Sue)) Fo(U)

Fo(Man)

Fo(Like(U)(Sue)) Fo(Sue)

Fo(Like(U)) Fo(U)

Fo(Like)

Figure 5.Completing interpretation for a restrictive relative.

An alternative form of construction process for relative clause construal involves the obligatory resumptive use of pronouns, as is displayed by Arabic. In (Egyptian) Arabic, a pronoun is essential in all non-subject positions for the strings to be well-formed:8 (5) il mudarris illi Magdi darab-u the teacher who Magdi hit-him ‘the teacher who Magdi hit’ (Egyptian Arabic)


To reflect this distribution, we propose an analysis in which the complementiser induces the introduction of the required linked tree with its associated requirement for a copy (see Figure 6): it does not provide either an unfixed node or the required copy.9 This requirement, which singularly lacks any locality restriction requiring the copy to occur in some subtree of the total structure, is expressed using the ·DÒ operator, which is satisfied by an arbitrary sequence of daughter or LINK relations. illi IF {Fo(x), Ty(e), Def(+), [Ø]^}, THEN make(·LÒ); go(L); put(?Ty(t), ?·DÒ(Fo(x), Ty(e))), ELSE ABORT Host tree

{?Ty(t)} {Fo(Za‘al), Ty(e → t)}

{?Ty(e)}

{Fo(lP(e, P)), Ty(cn → e)}

{?Ty(cn)}

{Fo(U), Ty(e)}

Linked tree

{Fo(Mudarris), Ty(e → cn)}

{?Ty(t), ?〈D〉(Fo(U))} {Fo(Magdi), Ty(e)}

{?Ty(e → t)}

{Fo(Darab), Ty(e → (e → t))}

{Fo(U), Ty(e), ◊}

Figure 6.Projection of LINKed trees in Arabic with resumptive pronoun.

The consequence of defining illi in this way is that, with no particular lexical item defined in the language as projecting the required copy in such structures,10 the only way of meeting this requirement is to use the regular substitution process of the language, selecting as interpretation for some pronoun (-u in (5)) the value of the formula provided by the head node. Such a construal is essential, since any other substituend (replacing the pronoun’s meta-variable


with some independently available term e.g. Tom, Dick …) will leave the LINKed structure with a requirement outstanding, hence not well-formed. In consequence, a pronominal must occur in the subsequent string in a position from which an argument to the predicate can be directly constructed, and, moreover, must be interpreted as providing a copy of the formula annotating the head. Thus, Arabic provides a case indicated earlier, in which the complementiser imposes a complex requirement on the top node of a newly introduced tree which is met in virtue of some annotation to a node by a pronoun whose position in the string may be indefinitely far away from the complementiser. This obligatory occurrence of a resumptively construed pronominal needs no separate stipulation, and the substitution process updating the pronominal remains a purely pragmatic process. It is merely the interaction of this process with the modal form of requirement on the top node of the LINKed tree which determines the result (see Kempson et al. 2001 for analyses in detail, and all formal specifications of the framework). From the different patterns displayed in English and Arabic, we can abstract the common pattern across all linked structures. All pairs of linked trees can be characterised as sharing a common element, and all transitions from one tree to the other involve the imposition of a requirement on the second for a copy of the selected formula in the first (in head-initial languages, the formula corresponding to the head). What varies is the mode of projection onto this structure — according to whether an unfixed node is or is not projected that carries the required copy; or as varying locality restrictions are or are not imposed upon where within such structures the copied term is required to occur. What we might now expect is that some languages might display mixed effects. And indeed, Hebrew does. Hebrew allows both a ‘gap’ in relative clause sequences as in English, and also a resumptive pronoun as in Arabic. This can be explained as a combination of the two patterns. Supposing that the complementiser še can project merely the weak form of modal requirement imposed on the top node as in Arabic, and furthermore that *Adjunction is freely available (and not as in Arabic restricted to wh-questions — see footnote 11), the introduction of the root node of a LINKed structure can be introduced from the lexical actions of še, and *Adjunction can then apply creating an unfixed node requiring type e, a node which a pronoun can annotate with a copy of the head formula. This unfixed node can then be updated by merging with a fixed node:11 (6) ha-‘-iš še xašavt še Dani pagaš (‘oto) (Hebrew) the.man that you.thought that Dani met (him) ‘the man that you thought Dani met’


(7) ha-‘-iš še ‘ani xošev še ‘alav ‘amarta še Sara katva šir the.man that I think that about.him you.said that Sara wrote poem ‘the man about whom I think you said that Sara wrote a poem’ (8) ha-‘-iš še ‘alav ‘ani xošev še ‘amarta še Sara katva šir the.man that about.him I think that you.said that Sara wrote poem ‘the man about whom I think you said that Sara wrote a poem’ (9) ha-‘-iš še ‘ani xošev še ‘amarta še ‘alav Sara katva šir the.man that I think that you.said that about.him Sara wrote poem ‘the man about whom I think you said that Sara wrote a poem’

The sequence of updates is displayed as:12 Step I by *Adjunction and Substitution (anaphora resolution): INPUT {Tn(a), ?Ty(t), ?·DÒ(Fo(α), ?Ty(e)), } OUTPUT {{Tn(a), ?Ty(t), } … {·≠*ÒTn(a), Fo(α), Ty(e), ?$xTn(x)}} Step II by Merge: INPUT {{Tn(a), ?Ty(t), } … {·≠*ÒTn(a), Fo(α), Ty(e), ?$xTn(x)}} OUTPUT {{Tn(a), ?Ty(t), } … {Tn(b), Fo(α), Ty(e), }} As in all other cases, notice how each individual step in the growth process adds information to the tree. The only information ever to be removed are requirements, which go once they are fulfilled. The result is a characterisation of resumptive pronouns in languages such as Arabic and Hebrew without any special grammar-internal stipulation of a particularised form of resumptive construal.13 The challenge is now to see whether in other languages there is need of construction-specific use of such pronouns. And, perhaps surprisingly, the evidence that no such construction-specific itemisation is required emerges in considering crossover phenomena in a language such as English, in which resumptive use of pronouns is peripheral.

2. Crossover in relatives Crossover phenomena have been reported in the literature to be a ‘mystery’ (Postal 1993),14 apparently bifurcating into strong and weak crossover, but also potentially subdividing into two further subcategories - ‘extended strong’ and ‘weakest’ crossover. The issue in relatives is the specification of the circumstances under which a pronoun may intervene between the relativising element and its ‘gap’, the normal perspective being to consider the phenomenon


exclusively in terms of the hierarchical relationship between pronoun and gap. In these terms, if pronoun and gap are both arguments and the pronoun ccommands the gap, an interpretation in which the pronoun and wh-expression are taken to denote the the same entity is not possible (strong crossover — (10)). However, if the pronoun precedes but does not c-command the gap, then under some circumstances they may be construed as co-referring (weak crossover — (12)). And if the pronoun c-commands the gap but the binder is complex, then the judgements coincide with those of weak crossover allowing co-referring interpretations, despite their classification with the much stronger ‘strong crossover’ restriction — the so-called extended strong crossover restriction (13):15 (10) *Johni, whoi Sue thinks hei said ei was sick, has stopped working. (11) Johni, whoi Sue thinks ei said hei was sick, has stopped working. (12) Johni, whoi Sue said hisi mother worries about ei, has stopped working. (13) Johni, whosei motherj hei worries about ej far too much, has stopped working.

The first thing to note is that all strings involving crossover effects are grammatical: the issue solely concerns what interpretations can be assigned to the strings in question. What is less often pointed out (though see Safir 1996) is that the crossover phenomena overlap with resumptive pronoun phenomena, since (14), in which resumptive pronouns are used throughout the English string, are acceptable: (14) Johni, whoi Sue thinks hei said hei was sick, has stopped working.

The debate about crossover effects is extensive, and an integrated analysis remains a challenge (see e.g. Lasnik and Stowell 1991, Postal 1993, Safir 1996). An alternative perspective is to look at the process of setting up the requisite interpretation within a given structural context, rather than just the output logical form. We observe that if either pronoun or the wh-element is nested within a containing structure, then an interpretation in which the pronoun and the wh-expression can be construed identically is acceptable. This suggests that it is the relation between the wh-element and pronoun that is critical rather than that between pronoun and ‘gap’. With this in mind, we seek to explain these data in terms of the sequence of transitions in building up the relevant tree structure, and the particular information that is available at intermediate steps of the interpretation process.


We take first the strong crossover cases as exemplified by (10) as demonstration of how both construal of long-distance dependency involve updating one tree representation with another, in long-distance dependency fixing a tree relation, in anaphora fixing a formula value. Given that in English the relative pronoun itself annotates an unfixed node within the emergent linked structure, a primary subsequent task is to identify the position within the tree for this initially underspecified node. The question that arises then is whether this node can be merged with that decorated by the pronoun presented in (10) as the subject node of the subordinate clause, as a process of update for the metavariable provided by the pronoun? And the answer is “Yes”: nothing prevents the node which the pronoun has decorated with a meta-variable, Fo(U), itself in a fixed tree relation to the root, being merged with the initially unfixed node annotated by Fo(John). Fo(John) is a possible update for Fo(U), and ·≠0Ò·≠0Ò ·≠1ÒTn(0) (the full description of the embedded subject position in a tree of top node Tn(0)) is a possible update for ·≠*ÒTn(0) (the specification that Tn(0) dominates that node). Moreover, this use of Merge to unify these partially specified nodes is the only way of updating the meta-variable projected by the pronoun under this interpretation. This is because, subsequent to its introduction into a partial tree, an unfixed node is successively evaluated against the set of nodes subsequently introduced in the tree until the point at which it is merged with some node, and its Formula value cannot be selected as a putative antecedent from which a value for the meta-variable can be simply copied, since the unfixed node will invariably count as falling in the class of nodes which the locality constraint on identifying possible antecedents for the meta-variable determines may not provide a copy.16 An immediate consequence of merging the unfixed node and that decorated by the pronoun is that when at a subsequent stage in the construction process, the pointer is at a node requiring type e but the next word is of inappropriate type, there will be no extant unfixed node to provide the necessary step of Merge and the sequence of steps will abort, with no well-formed result on that interpretation of the pronoun. Of course if the pronoun is identified as referring to some other individual, and the term that replaces the meta-variable is not that of the head, then there will be no interaction between fixing the tree relation for the initially unfixed node decorated by the relative pronoun, and the result will be well-formed. The identification of a pronoun with the head by means of an application of Merge provides the interaction between the process of anaphora construal and the fixing of interpretation for the wh-expression within the emergent structure that underlies the strong crossover effects.

151

152


On this analysis, there will be no interaction between such an unfixed node and the construal of the pronoun if either the wh-expression or the pronominal are contained within some larger environment. If, in (15) the pronoun is identified as identical to that of the wh-expression, this has no consequences for identifying the tree position of the unfixed node because the wh-expression projects merely a subterm in the formula decorating the unfixed node. That is, construal of he as ‘John’ will not provide a basis for establishing the tree position for the node annotated by a formula constructed from whose mother: (15) Johni, whosei motherj Sue says hei worries about ej far too much, is at the hospital.

To similar effect, in the case of weak crossover, no pronoun internal to a determiner can provide update for the unfixed node projected by the relative pronoun who. In this case, with the wh-expression not being complex, the account turns on the analysis of genitive constructions. These are analysed as projecting a distinct linked structure whose internal arguments cannot provide an update for a node unfixed within the structure initiated by the relativising particle. Hence, as with extended strong crossover effects, there is no interaction between the construal of the wh-expression and the pronominal:17 (16) Johni, whoi Sue said hisi mother was very worried about ei has stopped working.

An immediate consequence of this analysis, given the distinct analyses of Arabic and English forms of relativiser, is that, in Arabic, the phenomenon of crossover will not arise: (17) irra:gil illi nadja ftakkarit inn-u ‘a:l inn-u aiya:n the.man who Nadya thought that-he said that-he was.sick ‘the man who Nadya thought he said he was sick’

This is because, by analysis, the particle illi merely imposes a modal requirement on the top node of the linked structure it introduces into the overall description, and, with no introduction of any unfixed node (see footnote 10), there is no possible interaction between *Adjunction and updating some metavariable introduced by processing a pronoun. Furthermore, as before, a pronoun appropriately construed is the only way of meeting this modal requirement, and so the data display, as we would expect, a lack of ‘gap’ positions in the structure or any initially unfixed node.


A second consequence is that the framework leads us to expect that English pronouns used with resumptive construal will be well-formed in all positions in relative clauses:18 (18) The head of the department, who he admits that he needs a holiday, is coming to the conference. (19) That dreadful professor, who I took great care to ensure that I didn’t get him as a tutor, is complaining to the head that I don’t go to his classes. (20) The house at the corner, which they’ve called it Pandemonium, is a monstrosity.

In English, using a pronoun to ensure a fixed tree position for an unfixed node gives every appearance of being an optional device, with markedness effects: (21) The man who Sue says he is no longer coming to the conference has written us an excellent paper. (22) Our Head of Department, who even he now realises he should be taking a rest, is nevertheless coming to the conference.

Despite the common assumption that the only relative clause strategy in English is a ‘gap’ creating one,19 it appears that unacceptability of resumptive pronouns in some contexts is a pragmatic filter. Amongst various factors contributing to the acceptability of such resumptive construals are whether there are extra contextually-based implications, such as ‘focus’ effects, which cannot be conveyed without the use of a pronoun (needed to be the locus of contrastive stress) as in (22), and whether the pronoun occurs in a subordinate clause sequence (as in (21)), which might be taken to indicate a level of complexity in parsing more than presented by a single clause. Such manipulation of pronouns to clarify the interpretation to be conveyed or encourage other conversational effects is suggestive of Relevance Theory assumptions, in which it is claimed that all utterance interpretation is constrained by a balancing of cognitive effort with some appropriate degree of cognitive effect (no cognitive effort without commensurate cognitive effect — Sperber and Wilson 1995). We do not pursue this here, merely noting that our account, in anticipating the availability of resumptive pronoun construal in relative clauses, has to turn to pragmatics to explain the commonly made assumption that English is not a language that allows free use of resumptive pronouns.20 Evidence that the restriction on resumptive pronoun use in English is indeed pragmatic comes, perhaps surprisingly, from other languages, where analogous restrictions are imposed, but in different structural environments.

153

154


For example, in Arabic, a pronoun is obligatory in all non-subject positions, with no markedness effects. However in subject position, where a pronoun is unnecessary (Arabic being a subject pro-drop language),21 the use of a lexical form of pronoun is subject to exactly the same form of markedness considerations as in English. That is, in both relative clauses and left-dislocation structures, the pronoun is not necessary, reported to be debarred, but in fact fully acceptable either if the form is stressed, or if it occurs in the presence of a particle equivalent to e.g. English even/only: (23) irra:gil illi mabsu:t the.man who happy ‘the man who is happy’ (24) ‘irra:gil illi huwwa mabsu:t the.man who he happy’ ‘the man who he is happy’ (25) irra:gil illi (hatta) huwwa mabsu:t the.man who (even) he happy ‘the man who (even) he is happy’ (26) irra:gil mabsu:t the.man happy ‘The man is happy.’ (27) ‘irra:gil, huwwa mabsu:t the.man he happy ‘The man, he is happy.’ (28) irra:gil, hatta huwwa mabsu:t the.man even he happy ‘The man, even he is happy.’

Contrary to an increasingly prevalent view (see e.g. Bresnan’s optimalitytheoretic analysis — Bresnan 1999), we accordingly assume that the explanation of such data involves interaction between grammar-internal and pragmatic constraints rather than evidence for additional grammar-internal restrictions. On this view, while well-formedness according to the grammar-internal rules tout court is defined in terms of availability in principle of at least one sequence of steps leading from a given sequence of words onto a complete propositional structure, full acceptability in use of a string involves both grammar-internal processes and pragmatic considerations determining successful recovery of the interpretation of an utterance in the particular context provided. The effect is


that pragmatic considerations provide an additional filter affecting acceptability judgements by speakers. So the result — with a promissory note granting the need of an accompanying pragmatic explanation — is a framework within which properties of relative clauses and crossover (strong/weak) can be explained in terms of tree growth. Notice that it is neither exclusively the linear order of words that determines interpretation, nor is it merely the resulting structure that might be said to follow some completed process of reconstruction. Rather, it is the intermediate steps in the process of building up interpretation which determine the availability of the appropriate interpretations. It is the interaction between fixing of initially unfixed nodes, establishing the value of some incomplete formula, and the use of information from one partial structure to initiate another, that determine the full range of data comprising not merely resumptive pronoun construal but also the puzzling crossover data. It might seem that such a characterisation of resumptive pronoun data is at best only partial, since there is broad cross-linguistic variation in this area; and setting aside the fact that full acceptability of resumptive pronoun construal may depend on pragmatic factors, there are conditions under which resumptive pronouns are clearly precluded in English. For example, resumptive pronouns are debarred in wh-questions, a restriction not apparently captured by our analysis given that application of Merge is a necessary transition step of update in the interpretation of such strings. However, we can see straightforwardly what is required to express the necessary restriction by a graphic display of the needed step of Merge in such cases.22 Suppose we presume upon the availability of a process of update for a string such as: (29) *Which book did John read it? {?Ty(t)} {?Ty(e)}

{Fo(John)}

{Fo(WH)} {Fo((x), Book(x)), Ty(cn)}

{Fo(Read)}

{Fo(U)}

Figure 7.Resumptive pronouns in questions.

As Figure 7 makes plain, in order for the node projected by the pronoun to be unified with the node projected from which book, there must be no restriction

155


on the decorations projected by the pronoun that the node it decorates be a terminal node in some tree. In general, lexical items project exactly such a restriction, and if pronouns can be used in wh-questions (for which the leftperipheral expression has to be taken as projecting an initially unfixed node within a single tree), they must be assumed to have lost this terminal-node restriction.23 In the case of relative clause construal, no such assumption is necessary, because the initiation of linked structures is defined to involve the imposition of a requirement for a copy of a Formula value, and not any copying of structure. The significance of these data for capturing the availability of resumptive pronouns in relatives, but not, in general, in wh-questions is, then, that as long as we define wh-expressions as invariably projecting complex internal structure (with determiner node, type cn node, variable and nominal nodes), even from simple expressions such as who and what, the nonavailability of resumptive pronouns in English questions will be expected, for in these but not in relatives, the application of Merge would have to have involved merging the node decorated by the pronoun with a nonterminal node, which the lexically imposed restriction that the set of actions decorate only a terminal node precludes.24 Thus, notwithstanding the liberality of the Dynamic Syntax framework in characterising resumptively construed pronouns in English relatives as well-formed, there is nonetheless a natural basis for characterising the availability of resumptively construed pronouns.25 In sum, the processing-oriented perspective articulated in Dynamic Syntax provides the basis for a characterisation of cross-linguistic variation in relative clause construal and resumptive use of pronouns, using only the interaction between general concepts of tree update and the meta-variable replacement which defines anaphoric resolution. Such an account of the resumptive use of pronouns without structure-specific stipulation is notably not available in any other framework. In other frameworks, the fact that resumptive pronouns play a role in long-distance dependency is taken as evidence that they are a syntactic phenomenon, in contrast to discourse anaphora, which is pragmatic (with bound-variable anaphora being characterised either as syntactic or as semantic, depending on the theoretical framework). It is only by, on the one hand, construing syntax as the dynamic projection of progressively enriched partial structures, and, on the other hand, by also construing interpretation, and in particular interpretation of pronouns as equally the progressive enrichment of partial structures, that a unitary explanation can be provided. As a bonus, confirming this analysis, the Dynamic Syntax account yields an integrated account of the notoriously intransigent crossover data.26 This two-fold success


in providing unitary characterisations of both anaphora and crossover strongly suggests that there are new perspectives to be explored by a shift in perspective for grammar design. According to this new perspective, syntactic formalisms for natural language grammars should be defined in terms of tree growth following the dynamics of the parsing process.

Notes * We are grateful to K. von Heusinger and two anonymous referees for detailed comments on earlier versions of this paper. 1. This process bears close formal resemblance to the concept of ‘functional uncertainty’ of Kaplan and Zaenen (1989), articulated within LFG, but that framework lacks the dynamics of updating such uncertainty as part of the structural characterisation. 2. The analysis of pronoun construal as a process of substitution is argued for in several different frameworks: see Kamp and van Eijck (1997), Ranta (1994), Piwek (1998), Fernando (in press), Kempson et al. (1999, 2001). 3. The detailed specification of condition and actions in the lexical specifications of verbs, including numbers of nodes to be constructed, varies from language to language, and indeed from verb to verb. This lexical description highlights the various basic actions DS uses to construct trees: put, go, make… 4. The pragmatic process of substitution is also used to model the incremental way in which some scope choices are established (see Kempson et al. 2001). 5. The rule here does not extend to pied-piping cases, but this minor simplification is for purposes of exegesis. See Kempson et al. (2001) for a fuller definition which applies to such more complex cases: (i) A Givenchy shirt, the collar of which was faded, was in the sale. 6. Nonstandardly all NPs are taken to project expressions of type e with quantified expressions characterised as variable-binding term operators (see Kempson et al. 2001: Ch. 4,7). We suppress here all details concerning the intrinsic content of the determiner. 7. This rule reflects the semantics of restrictive relative construals which, as standardly, we take to constitute a complex restriction on the domain of the variable bound by the determiner. Quantification is represented by term-binding variable operators (see Kempson et al. 2001: Ch. 7). 8. -u is a (reduced) pronoun that in all semantic respects functions as a full pronoun, with a regular array of interpretations when occurring in contexts in which no resumptive construal is forced. 9. As a subject pro-drop language, we take verbs in Arabic to have a condition for lexical action of ?Ty(t) from which node the subject node is constructed as well as the predicateinternal structure, annotating that subject node with a meta-variable needing substitution exactly as a lexical pronominal.

157

158


10. In variants of Arabic except classical, wh-questions are the only form which license a leftdislocated expression with no resumptive pronoun. This suggests that *Adjunction in Arabic should be defined to be sensitive to the presence of a +Q feature relative to which the unfixed node is required to have a +WH feature (see Kempson et al. 2001: Ch. 5). 11. As (6)–(9) indicate, še is also the bare subordinating device. The challenge to reduce all uses of še to a single unambiguous particle remains: postulated ambiguity is a common property of analyses of še (see Shlonsky 1993). 12. For simplicity here, we take ‘alav to annotate a node of type e: there are asymmetries between relativising oblique positions and indirect/direct objects which we ignore here. These involve adding extra constraints on the relative position of the unfixed node which do not affect the general dynamics. 13. Whether such projected meta-variable-containing formulae should, in the Arabic case, be taken to be defined with a separate lexical entry for the suffix as part of the characterisation of their phonological host is not a question we address here. Our sole concern is whether the resumptive use of such expressions needs to be distinguished from regular discourse use of a pronoun as fixed in context. And even taking into consideration the greater complexity imposed by asymmetry between relativisation of oblique NP positions and subcategorised positions, the increased complexity would lie in the characterisation of the relativiser and not in the pronominal element. 14. Safir (1996) purports to provide a solution to the phenomenon of weak crossover in response to this challenge of Postal’s, but his analysis is problematic in not extending to the French data observed by Postal, as Safir himself points out. 15. The debate is often discussed in connection with wh-question data, but here we restrict our attention to crossover phenomena in relatives. See Kempson et al. (2001) for a fuller discussion which includes wh-questions. 16. For the formal details of both Merge and the independent substitution process which copies a term from a selected antecedent as a Formula value, see Kempson et al. (2001). 17. The analysis of genitives as ‘implicit relatives’ is well attested in the literature (see Baker 1996, Kayne 1994, etc.). Notice that this form of analysis provides a basis for characterising their island-like status, with dependencies between external terms and terms internal to the genitive being in general not available. 18. As Tami Kaplan has noted (personal communication), such resumptive use of pronouns occurs regularly in conversation, and we are grateful to her for bringing her collection of such examples to our notice. 19. See Sag (1997), who, despite an otherwise comprehensive analysis of English restrictive relatives, makes no mention of the resumptive use of pronouns in relatives. 20. The Kaplan collection contains many examples where there is no obvious pragmatic motivation, and the result is nothing worse than slightly reduced level of acceptability: see (20), which is taken from that collection. According to a relevance-theoretic characterisation of such judgements, the judgement would be a reflection of less than optimal relevance in cases in which more than minimal effort is imposed without any offsetting benefit in the form of additional interpretive effects.


21. In our framework, the subject pro-drop phenomenon is modelled by defining verbs to project a propositional structure in which the subject node is directly annotated with a metavariable, as though decorated by a lexical pronoun. 22. As elsewhere in this paper, we ignore the tense specification. 23. See Kempson et al. (2001) for an alternative argument that this restriction has a pragmatic basis. 24. In some languages, e.g. Greek, left-dislocation structures optionally allow case-matching between the left-dislocated element and a resumptively construed pronoun. Given our analysis of case as a requirement on the output structure, such examples must be taken to necessitate application of *Adjunction to create an unfixed node which subsequently unifies with the node annotated from the actions of the pronoun with a meta-variable. In such languages, the resumptive pronoun must be analysed as having lost the terminal-node restriction. Greek, notably, is a language in which resumptive pronouns are used in some forms of wh-questions, see Tsiplakou (1998). 25. Although in this article, we have not covered the full range of data displayed in all of Safir (1996), Aoun and Benmamoun (1998), Demirdache (1991), Suñer (1998), it is nonetheless striking that in order to characterise when pronouns are construed resumptively, we have never needed to invoke the stipulations attendant on movement analyses, with such stipulations as morphological realisation of trace in addition to coindexing (Safir 1996, Aoun and Benmamoun 1998), postulation of clitic as an operator and as a variable within a single derivation (Demirdache 1991), or other such stipulations (see also Suñer 1998). 26. As with resumptive pronouns and a more general characterisation of anaphora, there is standardly no attempt to define a uniform characterisation of wh and anaphora interaction, simply presuming on analyses of strong and weak crossover in entirely distinct terms. The only exception to this is that of Dowty (1993), in categorial terms.

References Aoun, J. and Benmamoun, E. 1998. “Minimality, reconstruction and PF movement”. Linguistic Inquiry 29: 569–598. Baker, M. 1996. The Polysynthesis Parameter. Oxford: Oxford University Press. Blackburn, P. and Meyer-Viol, W. 1994. “Linguistics, logic, and finite trees”. Bulletin of Interest Group of Pure and Applied Logics 2: 2–39. Bresnan, J. 1999. The emergence of the unmarked pronoun. Ms., Stanford University. Demirdache, H. 1991. Resumptive Chains in Restrictive Relatives, Appositives, and Dislocation Structures. Ph.D. Diss., MIT, Cambridge, MA. Dowty, D. 1993. “Variable-free syntax, variable-binding syntax, the natural deduction Lambek calculus and the crossover constraint”. West Coast Conference of Formal Linguistics 11: 161–176. Fernando, T. in press. “Three processes in natural language interpretation.” In Festschrift for Salomon Feferman, ASL Lecture Notes. Fodor, J. A. 1981. Representations. Cambridge, MA: MIT Press.


Fodor, J. A. 1983. Modularity of Mind. Cambridge, MA: MIT Press. Kamp, H. and Eijck, J. van. 1997. “Representing discourse in context”. In Handbook of Logic and Language, J. van Benthem and A. ter Meulen (eds), 179–239. Amsterdam: Elsevier. Kaplan, R. and Zaenen, A. 1989. “Long-distance dependencies, constituent structure, and functional uncertainty”. In Alternative Conceptions of Phrase Structure, M. Baltin and A. Kroch (eds), 17–42. Chicago: University of Chicago Press. Kayne, R. 1994. The Anti-Symmetry of Syntax. Cambridge, MA: MIT Press. Kempson, R., Meyer-Viol, W. and Gabbay, D. 1999. “VP Ellipsis: Towards a dynamic structural account”. In Fragments: Studies in Ellipsis and Gapping, S. Lappin and E. Benmamoun (eds), 227–290. Oxford: Oxford University Press. Kempson, R., Meyer-Viol, W. and Gabbay, D. 2001. Dynamic Syntax: The Flow of Language Understanding. Oxford: Blackwell. Lasnik, H. and Stowell, T. 1991. “Weakest crossover”. Linguistic Inquiry 22: 687–720. Piwek, P. 1998. Logic, Information and Conversation. Ph.D. Diss., University of Eindhoven. Postal, P. 1993. “Remarks on weak crossover effects”. Linguistic Inquiry 24: 539–556. Ranta, A. 1994. Type-Theoretic Grammar. Oxford: Clarendon Press. Safir, K. 1996. “Derivation, representation and resumption”. Linguistic Inquiry 27: 313–340. Sag, I. 1997. “English relative clauses”. Journal of Linguistics 33: 431–483. Shlonsky, U. 1993. “Resumptive pronouns as a last resort”. Linguistic Inquiry 23: 343–368. Sperber, D. and Wilson, D. 1995. Relevance: Communication and Cognition. Oxford: Blackwell. Sunˇer, M. 1998. “Resumptive restrictive relatives: A cross-linguistic perspective”. Language 74: 335–364. Tsiplakou, S. 1998. Focus in Greek: Its Structure and Interpretation. Ph.D. Diss., School of Oriental and African Studies, London.

The third person pronoun in tripartite verbless clauses of Qumran Hebrew Jacobus A. Naudé University of the Free State, Bloemfontein

1.

Introduction

Qumran Hebrew (QH) is associated with the vast quantities of textual material which were brought to light from the eleven caves in the area around Khirbet Qumran at the north-west end of the Dead Sea.1 This textual material is related to the same group of users or speech community, labelled the Qumran Community, who occupied the buildings for a period of 200 years up to 68 CE. Chronologically, QH stands between Biblical Hebrew (BH) and Mishnaic Hebrew (MH).2 QH escaped the grammatical adaptation of transmitters (e.g. the Tiberian Masoretes) who could have influenced BH, and it could well represent the oldest form of Hebrew (apart from Archaic Hebrew). A grammarian is often confronted with structural elements that are lacking in his native language. Their presence compels the grammarian either to ignore them as pleonastic or to afford them some semantic interpretation. The purpose of this paper is to determine the status and interpretation of such an element, namely the independent third person pronoun which is found between the subject and predicate of some verbless clauses of QH.3 For each pronominal construction, grammatical and ungrammatical sentences from QH will be constructed. Grammatical sentences will be constructed directly from the literature corpus of QH. Ungrammatical sentences will be constructed on the basis of technical familiarity with the language on the one hand, and on the basis of occurrence in a particular sentence position on the other hand. If an independent personal pronoun does not occur in a particular sentence position in the literature corpus, it will be assumed that if an independent personal pronoun should occur in that particular position then that sentence would be ungrammatical (see for example (18) below). (For the taxonomy all instances of the independent personal pronoun have been gathered.)

162 Jacobus A. Naudé

The term verbless clause is used in traditional grammars of QH (and other Semitic languages, e.g. Arabic (see Kempson and Meyer-Viol, this volume), Aramaic, Ethiopic, Modern Hebrew) to refer to sentences with predicates lacking an overt verb form.4 Sentence (1) illustrates this phenomenon. (1) 1QS XI.95 wãny lãdm ršäh and.I [PP to.humankind evil]6 ‘However, I belong to evil humankind.’

In QH, verbless clauses are always interpreted as being in the present tense, equivalent in meaning to corresponding sentences in other languages which contain a copular verb (e.g. to be in English). Present tense verbless clauses in QH lack a copula, but in some cases contain an optional independent third person pronoun in addition to the subject, as shown in (2). This so-called pleonastic or dummy pronoun is indicated in (2) by the label pron. In this paper a verbless clause consisting of the string noun phrase — third person pronoun — noun phrase is called a tripartite verbless clause. (2) 4Q393 [4QLiturgical Work] 2.67 ãth hwãã yhwh you.masc pron.3sg.masc Lord ‘You are the Lord.’

In traditional grammar, various analyses have been advanced for tripartite structures like the one in (2).8 One approach takes pron to be a copula, denying the possibility that it is an ordinary pronoun which is used for, conceivably the purpose of expressing emphasis.9 According to another view, pron can be analysed as the resumptive pronoun of a dislocation construction. In the case of (2), the predication would then be semantically associated either with hwã yhwh SUBJECT — PREDICATE (in which case ãth is taken as the dislocated element of a left-dislocation construction) or with ãth hwã SUBJECT — PREDICATE (in which case, yhwh is taken as the dislocated element of a right-dislocation construction) (Driver 1892: 271). Such analyses would suggest the glosses As for you, he is the Lord or You are he: the Lord. Respectively, a third approach is to analyse pron as a pleonastic element (Brockelmann 1956: 26–28). The ostensibly haphazard variation in the occurrence of gaps and pron, in languages like Hebrew, has been forwarded as an argument to support the existence of an optional pron strategy employed in the formation of verbless clauses such as (2) alongside the more normal strategy illustrated in (1), in which pron is absent.

The third person pronoun in tripartite verbless clauses of Qumran Hebrew 163

A close study of QH data strongly suggests, however, that there is no such option in QH. On the contrary, for a specificational interpretation of verbless clauses with definite NPs in predicate positions, the presence of pron seems obligatory; by contrast the absence of pron will lead to a predicational interpretation of such verbless clauses.10 A verbless clause is specificational if the term that is the subject represents a variable for which the predicate nominal specifies a value. Such clauses are identifying in that the specification of a value makes it possible to identify the variable, that is, to pick out the referent represented by the variable from the set (e.g. identity statements and descriptionally identifying clauses). Instead of specifying a value for a variable (i.e. identifying a referent) predicational clauses merely predicate something of the referent of the subject term. An analysis is proposed infra which seems to provide an explanation for the presence of pron marking a verbless clause as specificational. The thrust of this paper is to show that pron only occurs as a last resort, when other avenues to yield a grammatical output of a certain construction are not available. Although the focus will be on QH verbless clauses, it is suggested that pron is never freely generated and that its distribution is universally regulated by last resort considerations. If prons occur only as a saving device for an otherwise ungrammatical derivation, then an investigation into their distribution should focus not on where prons are allowed to occur, but rather on cases where a grammatical derivation is blocked. In other words, the approach advocated here is that the explanation for the distribution of pron should be sought in a comparison of the internal syntax of structures containing pron with those in which pron is lacking. The paper is organised as follows: Section 2 presents the basic QH data and outlines the problem of the categorial status of pron and its restricted distribution in verbless clauses. The hypothesis that pron represents a copula, forming part of the sentence predicate or that it represents a resumptive pronoun in a left-dislocation construction is critically examined in Sections 3 and 4. In Sections 5 and 6 an alternative analysis of QH verbless clauses is presented. More specifically, it is argued that pron is a clitic which is selected as a last resort for the checking of agreement features. The conditions under which pron may, should, or may not be selected, are related to the various co-occurrence relationships that hold between subjects and predicates.


2. The problem The predicate of a verbless clause in QH can be classified as a projection of a noun (N), an adjective (A) or a preposition (P), as illustrated by the examples in (3a–c), respectively. 4Q417 [4QSap. Workc] 2.I.16 whwãh hzwn » hhgwy lspr zkrwn and.he [NP vision the.meditation.gen to.book memorial.gen] ‘This is the vision of the meditation of the book of memorial…’ b. 1QH XIV.15. ky ãth sdyq » for you [ap just] ‘For you are just…’ c. 1QH IV.29 whwã bäwwn mrhm » and.he [PP in.sin from.womb] ‘He is in sin from the womb.’

(3) a.

If sentences of the type in (3) express past or future tense, the verbal root hyh (be), is obligatory as indicated in (4).11 (4) a.

1QS III.5 t» mã yhyh kwl ywmy mwãsw bmšp»ty ãl defiled be.impf.3sg all days spurn.inf.his on.decrees God ‘Defiled shall he be all the days he spurns the decrees of God.’ b. 1QM VII wkwl ãyš ãšr lwã yhyh t» hwr and.every man that not be.impf.3sg.masc pure ‘Every man that will not be pure…’ c. 1QS II.24 hkwl yhyw byhd » ãmt the.all be.impf.3pl.masc in.community truth.gen ‘All shall be in a community of truth.’

As mentioned supra, an extra or additional pronoun, pron, may occur in present tense nominal sentences. pron is only attested in verbless clauses, the predicates of which are projections of N and seems to be obligatory in sentences like those in (5). As illustrated in (5), pron ranges over all the independent nominative third person independent personal pronouns and may agree in number (and gender) with the subject (see (36)–(39)).


(5) a.

1QpHab XII.3 kyã hlbnwn hwãã äs» t hyhd » because Lebanon pron.3sg.masc council community.gen ‘Because Lebanon is the Council of the Community.’ b. 1QpHab XII.7 hqryh hyãã yrwšlm the.city pron.3sg.fem Jerusalem ‘The city is Jerusalem.’ c. 4Q169 [4QpNah] 3–4.II.1 wmlãkyw hm syrw » and.messengers.his pron.3pl.masc envoys.his ‘His messengers are his envoys.’

Schematically, the construction in question exhibits the pattern in (6). (6) X pron Y, where pron stands for a third person pronoun, and X and Y for other syntactic elements.

In QH, X can be an independent personal pronoun, an interrogative pronoun, a demonstrative pronoun, a free relative pronoun or definite noun (including proper names). This is illustrated by the examples in (7)–(13). Y in (6) stands for a definite noun/referring NP.12 (7) 4Q266 [4QDa] 18.V.9 ãt hw[ãã] hkwl you.masc.sg pron.3sg.masc the.everything ‘…you are everything…’ (8) 1Q27 [1QMyst] 1.II.3 mh hwãã hywtr l[ what pron.3sg.masc the.advantage for ‘What is the advantage for…?’ (9) 4Q186 [4QHoroscope] 1.II.8 wzh hwããh hmwld and.this pron.3sg.masc the.sign ‘And this is the sign…’ (10) 4Q162 [4QpIsab] II.7 ãšr byrwšlym hm ãšr mãsw those in.Jerusalem pron.3pl.masc those reject.perf.3pl.masc


ãt twrt yhwh acc law Lord ‘Those are they who have rejected the law of the Lord.’ (11) 4Q169 [4QpNah] 3–4.III.9 whyãrym hm gd[w]ly [mn]šh and.the.rivers pron.3pl.masc nobles Manasseh.gen ‘And the rivers are the nobles of Manasseh.’ (12) 4Q169 [4QpNah] 3–4.I.10 rwbkh hm gdwdy hylw » abundance.your pron.3pl.masc gangs soldier.his.gen ‘Your abundance is his gangs of soldiers.’ (13) 4Q169 [4QpNah] 3–4.III.9 ãmwn hm mnšh Amon pron.3pl.masc Manasseh ‘Amon is Manasseh.’

At least three questions follow from the pattern in (6). Firstly, why is pron present in verbless clauses (see (5)) and what is its categorial status? Secondly, for what reason is pron attested in verbless clauses, predicates which include projections of definite nouns but not other elements (see (5))? Thirdly, what is the cause for the absence of pron in the verbless clauses (14)–(16), the predicates of which also comprise definite nouns and adjectival and prepositional phrases? (14) 4Q504 [4QDibHama] 1–2 recto.V.8 kyã ãth ãl hy » for you.sg.masc [NP God life] ‘… for you are a living God…’ (15) 1QH XIV.15 ky ãth sdyq » for you.sg.masc [ap just] ‘For you are just…’ (16) 1Q28b [1QSb] IV.24 wãth kmlãk pnym and.you.sg.masc [PP like.angel face.gen] ‘And may you be like an angel of the presence/face…’

In Sections 3 and 4 two hypotheses about the categorial status of pron that have been presented in the literature, are investigated. In Sections 5 and 6 an alternative analysis is presented to account for the presence and distribution of pron in QH verbless clauses.


3. The third person pronoun as verb It is claimed in literature that pron, when appearing in sentences like (5), has the same categorial status and grammatical function as the copular verb hyh in sentences like (4). To put it differently: pron is nothing other than the suppletive form of hyh in present tense sentences. This hypothesis is found in several traditional grammars of Hebrew (e.g. Muraoka 1985: 67–82, Waltke and O’Connor 1990: 297 and Gibson 1994: 1). In this section two arguments based on distributional facts are presented against the hypothesis that pron, on a par with hyh, represents a copula forming part of the sentence predicate. The first argument concerns the co-occurrence of hyh and present participles. Hyh may appear in constructions with verbal present participles as shown in (17). (17) 1QM IX.1 whkwhnym yhyw mryäym and.the.priests be.impf.3sg.masc blow.part bh» sw » srwt » hhllym » on.trumpets the.destruction.gen ‘…but the priests shall continue blowing the trumpets of destruction.’

If pron is analysed as V and hyh may appear in (17), it should be possible for pron to replace hyh in such a construction without changing the appropriate grammatical structure or the interpretation (cf. (18)). (18) whkwhnym hmh mryäym and.the.priests pron.3pl. blow.part bh» sw » srwt » hhllym » on.trumpets the.destruction.gen ‘…but the priests — they will continue blowing the trumpets of destruction.’

The sentence in (18) is grammatical, but only if interpreted as a left-dislocation like the example in (19). (19) 4Q226 [4QpsJubb] 4.1 ky yhwšã b]n nwn hwãã äwbr lp[ny for Joshua son Nun.gen pron.3sg.masc cross.part before ‘for Joshua, the son of Nun, he is crossing before…’


In short, the above examples illustrate that pron cannot fill the position normally occupied by hyh. Since pron and hyh do not share the same distribution, it follows that pron cannot be construed as the suppletive form of hyh, that is, it cannot be analysed as a copular verb. The second argument has to do with word order. A verb in QH may precede the subject (20)–(21). This does not hold good for pron, as is shown by the ungrammaticality of (22). perfective verb form: (20) 1QM I.6 wsrh mmšlt ktyym and.be.end.perf.3sg.fem rule Kittim.gen ‘The rule of the Kittim will come to an end…’ imperfective verb form: (21) 1QS II:23 wlwã yšpwl ãyš mbyt mämdw and.not be.lower.impf.3sg.masc man from.house standing.his ‘And a man will not move down from his allotted position.’ 4Q163 [4QpIsac] 2–3.4 htrh hyãã r»swn the.law pron.3sg.fem Rezin ‘The law is Rezin…’ b. *hyãã htrh r»swn pron.3sg.fem the.law Rezin ‘The law is Rezin…’

(22) a.

In contrast to pron, however, the copular verb hyh may precede the subject, as is shown in (23). This difference in distributional properties provides additional evidence that pron does not represent the suppletive form of hyh, and hence cannot be analysed as a copular verb. (23) 1QS I.18 yhyw hkwhnym whlwyym be.impf.3pl.masc the.priests and.the.levites mbrkym ãt ãl yšwäwt bless.part acc God salvation.gen ‘…the priests and Levites shall bless the God of salvation.’


4. The third person pronoun as a resumptive in a left-dislocation construction Consider the left-dislocation construction (24). (24) 4Q396 [4QMMTc] 1.III.413 wmäs´r hbqr whswn » lkwhnym hwãã and.tithe the.herd and.the.flock.gen to.priests pron.3sg.masc ‘The tithe of the herd and the flock — it is for the priests.’

As illustrated in (24), a dislocation construction consists of a dislocated constituent (wmäs´r hbqr wh»swn (‘and the tithe of the herd and the flock’)) and a clause that is linked to the dislocated constituent (lkwhnym hwã (‘it is for the priests’)). The linked clause contains a resumptive pronoun (hwã (he)), an expression which is semantically related to the dislocated constituent. A similar analysis is proposed by Gross (1987: 123–30) for verbless clauses containing pron, with pron taken to be a resumptive pronoun. In this section several considerations will be presented which indicate that such a resumptive pronoun analysis of pron is not feasible for QH. The first consideration concerns definiteness. A subject followed by a pron is always definite (25), whereas a left-dislocated NP may be indefinite (26). (25) 4Q219 [4QJubd] II.17;14 11Q19 [11QTa] LIII.6 ky hdm hwãã hnpš for the.blood pron.3sg.masc the.life ‘…for the blood is the life…’ (26) 1QH XI.8 wkwl kbwd ãtkh hwã and.all glory.indef with.you pron.3sg.masc ‘And all the glory — it is with you.’

The second consideration concerns the fact that it is not possible to construe questions out of sentences with dislocated constituents, i.e. the dislocated constituent cannot be in the form of a question. The example in (27) illustrates this phenomenon; (27b) is ungrammatical with the same interpretation as (27a). (27) a.

1QH III.23 wãny ysr » hhmr » mh ãny and.I creature the.clay.gen what I ‘And I, creature of clay, what am I?’


b. *mh ãny ysr » hhmr » ãny and.what I creature the.clay.gen I ‘And I, creature of clay, what am I?’

If a verbless clause with pron is analysed as a component of a dislocation construction, it should be impossible to reformulate it as a question. This consequence is not borne out by the facts. As (28) shows, a verbless clause with pron can assume the form of a question. (28) 1QH VII.32 wmh hwãã ãyš thw … and.what pron.3sg.masc man empty lhtbwnn bmäsý plãk to.understand.inf in.works wonder.your ‘What is empty man … to understand your wondrous deeds?’

The third consideration against the dislocation analysis involves agreement facts. The resumptive pronoun in dislocation constructions agrees with the dislocated element in terms of gender, person and number. This is illustrated by the examples in (29)–(35). masculine third person singular: (29) 4Q226 [4QpsJubb] 4.1 ky yhwšä b]n nwn hwã äwbr lp[ny for Joshua son Nun.gen pron.3sg.masc cross.part before ‘for Joshua, the son of Nun, he is crossing before…’ feminine third person singular: (30) 1QH IV.13 wästkh » hyã and.counsel.your pron.3sg.fem tqwm stand.up/last.impf.3sg.fem ‘And your counsel — it will last…’ first person singular: (31) 1QH II.28 wãny bmws lby kmym and.pron.1sg when.melt.inf heart.my like.water ‘And as for me, when my heart is melted like water…’

The third person pronoun in tripartite verbless clauses of Qumran Hebrew

second person singular: (32) 4Q418 [4QSap Work Aa] 81.6 wãth hlwã lkh t» wbw and.pron.2sg.masc not.interr to.you goodness.his ‘And you, is not his goodness for you?’ third person singular: (33) 1QS XI.21 whwãh mäpr mgblw and.pron.3sg.masc from.dust shape.his ‘And as for him, his shape is from dust.’ first person plural: (34) 4Q491 [4QMa] 14–5.3 ]wãnw hnnw äwmdym lhtqrb[ and.pron.1pl behold.we stand.part to.approach ‘And we, behold we take up position to approach…’ third person plural: (35) 1QM XVII.4 [kyã]15 hmh lthw for pron.3pl.masc towards.chaos wlbhw tšwqtm and.emptiness desire.their ‘As for them, their desire is towards chaos and emptiness.’

By contrast, as shown in (36)–(39), it is not necessary for pron to agree always in person (and gender) with the NP to its left. In fact, a corresponding leftdislocated interpretation will be ill-formed. second person and third person: (36) 4Q393 [4QLiturgical Work] 2.6 ãth hwãã yhwh you.sg.masc pron.3sg.masc Lord bhrth » bãbwtynw choose.perf.2sg.masc to.fathers.our ‘You are the Lord. You have chosen our fathers.’ feminine and masculine: (37) 11Q13 [11QMelch] II.15 hzwãt hwããh ywm h[šlwm this.sg.fem pron.3sg.masc day the.peace.gen ‘This is the day of peace…’

171


masculine and feminine: (38) 4Q251 [4QHalakhaha] 5.4 wlhm] » bkwrym and.bread.sg.masc first.fruits.gen hyãã hlwt » hhm» » s pron.3sg.fem loaves the.leaven ‘and the bread of the first fruits is the leavened loaves…’ singular and plural: (39) 4Q169 [4QpNah] 3–4.III.9 ãmwn hm mnšh Amon pron.3pl.masc Manasseh ‘Amon is Manasseh.’

Furthermore, the claim that pron is a resumptive pronoun fails to account for the fact that pron, together with an NP subject, has a restricted distribution. It does occur with predicates which contain definite NPs, but is absent from predicates which are projected from NP, AP or PP (e.g.(14)–(16)). In short, there is ample evidence that predicate verbless clauses with pron cannot be analysed as the linked clause in left-dislocation constructions, and that pron does not represent a resumptive pronoun in such constructions. Against this background, it is argued below that pron should rather be analysed as a clitic which is selected as a last resort for the checking of agreement features. Prior to that, however, the structure of verbless clauses is to be investigated in greater detail in Section 5.

5. The structure of verbless clauses One fundamental assumption which is followed in the present paper is that sentence structure is to a large extent a reflection of the properties of lexical items. On this assumption constituents are taken to be hierarchically organised around a head, with the latter determining the structural and thematic properties of the constituents that are selected. For example, the question as to whether the VP should (not) contain one or more argument NPs is determined by the properties of the specific verb, the head of the VP. If a VP has a transitive verb as its head, (at least) a direct object NP is required as the complement of the verb. If a VP contains an intransitive verb as its head, then no NP complement is allowed. Whether a given verb belongs to the group of transitive or intransitive verbs is treated as an idiosyncratic property of that verb.


In the Chomskyan framework, notions like transitive and intransitive have been formalised on the basis of the approach commonly adopted in logic. For example, in the notation of formal logic (40a) is assigned the representation (40b): (40) a. Mary imitates John. b. A (m, j) where A = imitate, m = Mary and j = John

(40a) contains the NPs Mary and John, two referring expressions or arguments, that is, expressions which serve to pick out one entity from the universe of discourse. It also contains a predicate, imitate. The predicate does not refer to a person or thing, but rather defines some relation between the referring expressions. Predicates requiring two arguments are called two-place predicates. The transitive verbs of traditional syntax correspond approximately to the twoplace predicates of logic. The arguments of a predicate are realised by noun phrases in our example: In (40a) the subject NP is one argument and the object NP is the second of the verb imitate. Intransitive verbs correspond to one-place predicates: They take one argument, as in (41). (41) a. Mary stumbled. b. S (m) where S = stumbled and m = Mary

Neither auxiliaries nor the copula have an argument structure (Pollock 1989). Other lexical categories, besides the verb, that have an argument structure are adjectives, prepositions and nouns. Consider in this regard the examples in (42). (42) a. John is restless. b. John is in London. c. John’s analysis of the data was superfluous.

The adjective restless in (42a) is a one-place predicate since it takes one argument (the NP John). The preposition in in (42b) takes two arguments (John, London). So does the noun analysis in (42c), which takes the arguments John and the data. To summarise, the argument structure of a predicate determines which elements of the sentence are obligatory. If a predicate expresses an activity involving two arguments, there will have to be two constituents in the sentence which correspond to these arguments.16 Let us now examine against this background the structure of verbless clauses in QH. As was illustrated above in (4), the predicate of a verbless clause

173


can be a projection of a noun, an adjective or a preposition, parallel to verbal predicates, which are projections of verbs. Support for the claim that N, A and P like V, have their own projections and that they can each function as a predicate comes from the fact that N, A and P can occur in front of the subject, as illustrated by the examples in (45), (46) and (47) respectively. In this respect, N, A and P behave exactly like V (compare (43), (44)). perfective verb form (43) 1QM I.6 wsrh mmšlt ktyym [VP and.be.end.perf.3sg.fem] rule Kittim.gen ‘The rule of the Kittim will come to an end…’ imperfective verb form (44) 1QS II:23 wlwã yšpwl [VP and.not be.lower.impf.3sg.masc] ãyš mbyt mämdw man from.house standing.his ‘And a man will not move down from his allotted position.’ (45) 1QH X.3 wãdmh hwã[ [NP and.earth] he ‘He is earth…’ (46) 4Q218 [4QJubc] 1.2 ky] qdwš hwã [m]kl hymym for [ap holy] he in.comparison.to.all the.days ‘…for it is holier than all the other days.’ (47) 4Q160 [4QVisSam] 3–4.II.6 kyã ämkh hwãh[ for [PP with.you] he ‘…for with you is he…’

On the basis of these examples, it could be claimed that Ns, As and Ps in QH are (verbless) predicates in their own right rather than the complements of some phonetically empty verb. The predicate is a projection of N, A or P, parallel to verbal predicates that are a projection of V. Given this hypothesis, the verbless predicate in (48) would lack a VP node.


(48) 1QH XIII.14 whwã mbnh äpr and.he [NP building dust.gen] ‘He is a building of dust…’

The syntactic derivation of the verbless clause in (48) starts with the selection of the fully inflected noun forms hwã (he) and mbnh äpr (‘building of dust’). As a noun predicate mbnh äpr requires an argument with the thematic role of Experiencer. The predicate position (clause final position) of a verbless clause without pron is not an argument position. The predicate is a projection of N, A or P. Thus, a P, an A or N that appears in predicate position in a verbless clause lacks argument status, but rather represents the predicate.17 In verbless clauses without pron a quality or property specified by the predicate is attributed to the NP subject. This type of verbless clause is sometimes referred to as predicational.18

6. The role of the third person pronoun in the derivation of verbless clauses 6.1 The third person pronoun as clitic pron in QH exhibits at least the following two properties that Zwicky (1977) lists as characteristic of clitics. Firstly, pron cannot occur in isolation. As illustrated in (49a), pron must be preceded by a subject or antecedent. (49b), where pron is preceded by a predicate and (49c) where pron is followed by its subject are ungrammatical with the same interpretation as (49a). (49) a.

4Q169 [4QpNah] 3–4.I.11 ]w»trpw hwãã hhwn and.prey.his pron.3sg.masc the.wealth ‘His prey is the wealth…’ b. *]w»trpw hhwn hwã c. *hwã t» rpw hhwn

Secondly, pron never occupies the first position in a sentence. If an independent personal pronoun takes the first position in a sentence, it is a subject topic and cannot be interpreted as pron (50).

175


(50) 4Q321 [4QCalendrical Doc Ba] 2.II.2; 2.III.9 hwããh ywm hzkrwn pron.3sg.masc day remembrance.gen ‘It is the memorial day…’

The fact that pron is in a second position derives from the fact that it cliticised to a preceding NP, i.e. it cannot stand on its own. It is not related to the second position phenomena. These facts suggest that pron should be analysed as a clitic. 6.2 The third person pronoun as a saving device In verbless clauses with pron, the NP in the predicate position does not specify a property of the subject. Rather, like the NP subject, the NP in the predicate is referential in that it denotes a specific entity in the universe of discourse.19 Such sentences are sometimes referred to as specificational. Specificational NPs are arguments and accordingly must be assigned functions (see endnote 16). Therefore, the derivation of a verbless clause will pose a problem whenever a specificational NP occupies a predicate position (a clause final position). A verbless clause including a specificational NP (i.e. a string of the form NP-det-NP) will of necessity be ill-formed, on account of there being no predicate to assign functions (e.g. EXPERIENCER, IDENTIFIER or SPECIFIER) to the NPs. Since the specificational NP cannot function as a predicate, it will be impossible to assign a function to the subject in this case. Thus, it would seem that specificational NPs should be barred from occurring in the predicate position (a clause final position). Languages do, however, have various saving devices for allowing a referring NP to occupy a predicate position (a clause final position). English and many other languages use a copula. This point is obfuscated by the fact that English has a copula, even if predicates are not necessarily referring.20 In QH pron seems to be used to yield a grammatical specificational verbless clause. Consider sentence (51), which contains a verbless clause with pron. (51) 1QpHab XII.7 hqryh hyãã yrwšlm the.city pron.3sg.fem Jerusalem ‘The city is Jerusalem.’

The derivation of (51) starts with the selection of Ns hqryh (‘the city’) and yrwšlm (‘Jerusalem’), which are referring NPs. The pron hyã is inserted as a saving device, since functions must be assigned to the two arguments to create


a specificational structure. In short, then, pron serves to mark the verbless clause as specificational. The absence of pron would yield an ungrammatical sentence (with a specificational interpretation). As argued in Section 6.1 pron is a clitic, which is cliticised to the first NP in (51). As a clitic pron has the same categorial status as the element to which it is cliticised (Borer 1983: 35). The agreement features in verbless clauses with pron do not become agreement affixes as they do in sentences with verbs. Rather those features are realised as pronominal clitics on the noun subject of verbless clauses.

7. Conclusion In this paper an analysis is proposed of tripartite verbless clauses in QH with a third person pronoun. It is argued that the pronoun in question, pron, has the status of a pronominal clitic. The conditions under which a pronominal clitic may, must, or may not be used in verbless clauses follow from the various subcategorial relations that hold between subjects and predicates. pron is obligatory in verbless clauses with definite/specificational NPs in predicate positions where it serves to assign functions to the referring NPs. pron insertion might best be construed as a type of last-resort strategy (like do-support in English; cf. Müller (this volume) on the idea that R-pronouns in German and English are formed as a syntactic repair in contexts where regular pronouns are blocked). Thus, pron is never freely generated. pron is necessary for the semantic interpretation of specificational verbless clauses and must be interpreted as such.

Notes 1. Although the biblical texts discovered at Qumran reflect peculiarities inherent to QH (see Kutscher 1974), it is normally accepted that QH is mainly reflected by the non-biblical texts of Qumran. 2. Mainly for practical reasons, Hebrew is normally divided into periods corresponding to the different linguistic corpora. However conventional and unadventurous this classification might seem, it does serve as a framework for providing a diachronic view of the language, while at the same time implying acceptance of the argument that QH is clearly distinguished from BH and MH, especially with regard to aspects of phonology, morphology and syntax.


3. The understanding of verbless clauses propounded in this paper benefits from the expositions of Doron (1983, 1986) and Rapoport (1987) on Modern Hebrew, executed within the GB-framework. 4. The term verbless clause is used by Andersen (1970), Blau (1972), Huehnergard (1987: 215) and Miller (1999). The term nominal sentence is used by Albrecht (1887, 1888), Kraus (1984) and Hoftijzer (1973). Geller (1991) classifies certain types of the verbless clause as cleft sentences. 5. The designation of a text is as follows: The first number (1–11) indicates the cave at Qumran (Q) where the text (scroll) was discovered. It is followed by a letter or number indicating the specific text, the column number and line number. The source of all texts is García Martínez and Tigchelaar (1997) except if stated differently. This source offers a fresh transcription and an English translation of all the relevant non-biblical texts found at Qumran, arranged by serial number from Cave 1 to Cave 11. 6. Lexical categories are N (= Noun), V (= Verb), A (= Adjective/Adverb) and P (= Preposition). The lexical categories are projected to the following syntactic categories: NP (= Noun Phrase), VP (= Verb Phrase), AP (= Adjectival Phrase) and PP (= Prepositional Phrase). 7. pron is marked bold in the example sentences. 8. Linton (1987) gives a critique of the views of Albrecht, Hoftijzer, Andersen and Muraoka on the verbless clause in Biblical Hebrew. 9. Cowley (1910: 141) and Muraoka (1985: 67–77) both recognise the emphatic force of the copula. 10. The terms specificational and predicational are suggested by Declerck (1988). Rapoport (1987: 28) uses the terms predicative and equative. 11. Bartelmus (1982) gives a description of the distribution and function of hyh in Biblical Hebrew. 12. A comparison between the QH data and the Modern Hebrew data in Berman and Grosu (1976: 265–85) shows some differences in the distribution of the third person pronoun. According to the data, Y may be indefinite and includes adjectives (i). (i) a.

b.

c.

ha-yeled hu student. the-boy he student ‘The boy is a student’. ha-yeled hi pikx-it. the-girl she smart-fem ‘The girl is smart’. david ve-tali hem nexmad-im. David and-Tali they nice-pl ‘David and Tali are nice’.

13. The reference in the composite text of Qimron and Strugnell (1994: 54) is B64. 14. The restoration is supported by VanderKam and Milik (1992: 73), Attridge et al. (1994: 47) and Wacholder and Abegg (1995: 16). 15. The restoration is according to Lohse (1971: 218) and Yadin (1962: 339).


16. The thematic requirements/argument structure associated with lexical items must be saturated in the syntax. In GB-theory the assignment of thematic roles is regulated by the theta criterion (i). (i) Theta criterion a. Each argument is assigned one and only one theta role. b. Each theta role is assigned to one and only one argument. This criterion has been abandoned in MP, its function taken over by the principle of Full Interpretation (Chomsky 1992: 37). The assumption is that all morphological features have to be saturated at LF. Should an obligatory argument be absent at this level, morphological features cannot be checked and the derivation will crash. For purposes of this paper the argumentation will be based on the assignment of functions to arguments by predicates and not on feature checking. 17. Compare Muraoka (1991: 143–51) for the distribution of PPs in Biblical Hebrew. 18. Semantically, the predicate in this type of verbless clause does not specify a value for a variable. It merely predicates a property of the subject NP, for example class-membership or class-inclusion. 19. It was noted in Section 2 that the obligatoriness of pron in a sentence correlates with the sentence having as its predicate a definite NP, that is, a referring NP or argument. According to Declerck (1988) specificational sentences are of two types. The first type is used to answer an overt question and contains only the material necessary for specifying a value for the variable in the question. The second type consists of two constituents, one representing the variable, the other the value. 20. English has a copula in cases like (i), where the predicate is not referring. (i) a. John is handsome. b. *John handsome.

References Albrecht, C. 1887. “Die Wortstellung im hebräischen Nominalsatze I”. Zeitschrift für alttestamentliche Wissenschaft 7: 218–224. Albrecht, C. 1888. “Die Wortstellung im hebräischen Nominalsatze II”. Zeitschrift für alttestamentliche Wissenschaft 8: 249–263. Andersen, F. I. 1970. The Hebrew Verbless Clause in the Pentateuch. Nashville, TN: Abingdon Press. Attridge, H. et al. 1994. Qumran Cave 4. VIII. Parabiblical Texts. Part 1. [Discoveries in the Judaean Desert 13]. Oxford: Clarendon Press. Bartelmus, R. 1982. HYH. Bedeutung und Funktion eines hebräischen ‘Allerweltswortes’. St. Ottilien: EOS Verlag. Berman, R. and Grosu, A. 1976. “Aspects of the copula in Modern Hebrew”. In Studies in Modern Hebrew Syntax and Semantics, P. Cole (ed.), 265–285. Amsterdam: North Holland.


Blau, J. 1972. “Review of Andersen (1970)” [in Hebrew]. Leshonenu 37: 69–74. Borer, H. 1983. Parametric Syntax. Dordrecht: Foris. Brockelmann, C. 1956. Hebräische Syntax. Neukirchen: Neukirchener Verlag. Chomsky, N. 1992. “A minimalist program for linguistic theory”. MIT Occasional Papers in Linguistics 1: 1–76. Cowley, A. E. 1910. Gesenius’ Hebrew Grammar. Oxford: Clarendon Press. Declerck, R. 1988. Studies on Copular Sentences, Clefts and Pseudo-clefts. Leuven: Leuven University Press. Doron, E. 1983. Verbless Predicates in Hebrew. Ph.D. Diss., University of Texas at Austin. Ann Arbor, MI: University Microfilms International. Doron, E. 1986. “The pronominal copula as agreement clitic”. Syntax and Semantics 19: 313–32. Driver, S. R. 1892. A Treatise on the Use of the Tenses in Hebrew and Some Other Syntactical Questions. Oxford: Clarendon Press. García Martínez, F. and Tigchelaar, E. J. C. 1997. The Dead Sea Scrolls Study Edition. Leiden: Brill. Geller, S. A. 1991. “Cleft sentences with pleonastic pronoun: A syntactic construction of Biblical Hebrew and some of its literary uses”. Journal of Ancient Near Eastern Studies 20: 15–33. Gibson, J. C. L. 1994. Davidson’s Introductory Hebrew Grammar — Syntax. 4th ed. Edinburgh: T & T Clark. Gross, W. 1987. Die Pendenskonstruktion im Biblischen Hebräisch. St. Ottilien: EOS Verlag. Hoftijzer, J. 1973. “The nominal clause reconsidered”. Vetus Testamentum 23: 446–510. Huehnergard, J. 1987. “Stative, predicative form, pseudo-verb”. Journal of Near Eastern Studies 47: 215–32. Kempson, R. and Meyer-Viol, W. this volume. “The dynamics of syntax: anaphora, relative pronouns and crossover”. Kraus, F. R. 1984. Nominalsätze in altbabylonischen Briefen und der Stativ. Amsterdam: Noord-Hollandsche Uitgevers Maatschappij. Kutscher, E.Y. 1974. The Language and Linguistic Background of the Isaiah Scroll. Leiden: Brill. Linton, J. A. 1987. Four Views of the Verbless Clause in Biblical Hebrew. Ph.D. Diss., University of Wisconsin, Madison. Lohse, E. 1971. Die Texte aus Qumran. München: Kösel-Verlag. Miller, C. L. (ed.). 1999. The Verbless Clause in Biblical Hebrew. Linguistic Approaches. [Linguistic Studies in Ancient West Semitic 1]. Winona Lake, IN: Eisenbrauns. Müller, G. this volume. “Harmonic alignment and the hierarchy of pronouns in German”. Muraoka, T. 1985. Emphatic Words and Structures in Biblical Hebrew. Leiden: Brill. Muraoka, T. 1991. “The Biblical Hebrew nominal clause with a prepositional phrase”. In Studies in Hebrew and Aramaic Syntax, K. Jongeling, H. L. Murre-van den Berg and L. Van Rompay (eds), 143–151. Leiden: Brill. Pollock, J.-Y. 1989. “Verb movement, Universal Grammar, and the structure of IP”. Linguistic Inquiry 20: 365–424. Qimron, E. and Strugnell, J. 1994. Qumran Cave 4. V (Miq»sat Maäasé Ha-Torah) [Discoveries in the Judaean Desert 10]. Oxford: Clarendon Press. Rapoport, T. R. 1987. Copular, Nominal, and Small Clauses: A Study of Israeli Hebrew. Ph.D. Diss., MIT, Cambridge, MA.


VanderKam, J. C. and Milik, J. T. 1992. “A preliminary publication of Jubilees Manuscript from Qumran Cave 4: 4QJubd (4Q219)”. Biblica 73: 62–83. Wacholder, B. Z. and Abegg, M. G. 1995. A Preliminary Edition of the Unpublished Dead Sea Scrolls: The Hebrew and Aramaic Texts from Cave Four. Fascicle Three. Washington, DC: BAS. Waltke, B. K. and O’Connor, M. 1990. An Introduction to Biblical Hebrew Syntax. Winona Lake, IN: Eisenbrauns. Yadin, Y. 1962. The Scroll of the War of the Sons of Light against the Sons of Darkness. Oxford: Oxford University Press. Zwicky, A. 1977. On Clitics. Bloomington, IN: Indiana University Linguistics Club.

181

Pronominal nouns* Phoevos Panagiotidis University of Cyprus

According to the standard syntactic account of pronouns, they are intransitive determiners, that is, determiners with no NP complement (Abney 1987). In this contribution, I attempt to demonstrate that such an approach is undesirable on both empirical and conceptual grounds. Moreover, I will argue that the source of pronominal reference is a closed class of nouns, grammatical nouns (Emonds 1985), that is, nouns that do not denote a concept; grammatical nouns can be phonologically null but are not necessarily so. The first section of this chapter looks at the syntactic nature of pronouns in Japanese and Thai and provides evidence that they are nouns (‘N-pronouns’). In Section 2 the relationship of such pronouns to the English noun one is examined; in Section 3 the source of pronominal reference in Syntax is examined as well as whether all D-pronouns are ‘intransitive determiners’. In Section 4 the impossibility of a determiner without a nominal complement is discussed and the chapter concludes with Section 5.

1.

Japanese and Thai: Pronouns as nouns

Throughout this chapter Abney’s term ‘intransitive determiner’ will be informally used to designate the hypothetical kind of determiners that take no (NP) complement. Now, a first class of empirical evidence against the analysis of pronouns as intransitive determiners comes from the syntactic behaviour of pronouns in languages like Japanese and Thai. We shall examine some characteristics of the pronominal systems in these two languages in the rest of this section. 1.1 Japanese Both Fukui (1987) and Noguchi (1997), although departing from slightly different assumptions, argue that pronouns in Japanese behave syntactically

184 Phoevos Panagiotidis

more like nouns rather than determiners. As the following examples (from Noguchi 1997: 777) illustrate, pronouns in Japanese can be modified by adjectives (1), be a possessum (2), and follow demonstratives (3), apparently being in their complement. (1) tiisai kare small he (2) watasi-no kare my he (3) kono kare this he

Examples (1)–(3) are surprising. If the pronoun kare is itself a determiner, then it is odd that it follows adjectives, as in (1): we expect D to be the highest position inside the nominal phrase. Moreover, if Noguchi (1997) is correct that Japanese demonstratives are determiners, the DP under (3) is even more odd if two determiners co-occur in it. Finally, although kare normally means ‘he’, in (2) it can be glossed as ‘boyfriend’ (Noguchi 1997: 778).1 Apart from the untypical co-occurrences of Japanese pronouns lower than adjectives (1) and with determiners (3), there is another interesting fact about them: Nolan (2000) and Noguchi (1997: 778) show that pronouns in Japanese seem to abide by quite stringent stylistic restrictions. The situation appears similar to the stylistic status of groups of nouns like freedom-fighter/rebel/ terrorist and, although ‘polite’ pronominal forms are quite common, most of them seem to reflect language usage rather than some sort of grammatically encoded feature [±Politeness] or [±Honorific] — exactly the situation in Japanese as postulated by Nolan (2000). Hence German Sie, French vous and Italian Lei are not exclusively ‘polite’ pronouns but, rather, ordinary pronouns that happen to be used in order to convey politeness. A consequence of the stylistic restrictions regulating the use of Japanese pronouns is the fact that there are quite a few of them. Consider the following variants from Noguchi (1997: 778): (4) First person singular: watasi, watakusi, ore, temae, boku, etc. Second person singular: anata, kimi, omae, temae etc.

What is expressed by I/me in English, can be expressed by at least five different forms in Japanese. Each of them is appropriate in specific contexts only,

Pronominal nouns

although all of them mean ‘the Speaker’. A similar state of affairs is true for the second person, a situation far more complex than the tu–vous distinction in French. Another interesting point in (4) is that the pronoun temae is actually ambiguous between a first and a second person specification — a very peculiar property for pronouns when viewed from an Indo-European point of view. 1.2 Thai Another East Asian language, Thai, displays properties similar properties to Japanese, although the two languages are typologically unrelated. Cooke (1968) has examined the pronominal system of Thai in an exhaustive fashion and points out that Thai pronouns can be modified, although to a lesser extent than Japanese ones, usually by clauses (Cooke 1968: 10).2 As in Japanese, a large number of pronominal forms is available for each person specification and choice among all the different forms is sensitive to style and discourse factors such as the formality of the situation — towards these two points, Table 1 which lists all the forms for 1st person singular in Thai is very illuminating.3 Moreover, like Japanese temae, there are also pronominal forms that are ambiguous for person: raw’ can be either 1st or 2nd person, khaw can be either 1st or 3rd person. An additional oddity of the Thai pronominal system is the fact that some pronouns are actually loanwords from English and Chinese, like ‘aj and ‘ua’ respectively. This suggests that at least these words belong to lexical categories. Of course, it can be the case, albeit rarely, that functional items can be borrowed, like English they from Norse. 1.3 Japanese and Thai pronouns are not determiners The data presented so far will be shown to constitute a severe blow to the explanatory capacity of an analysis of pronouns as intransitive determiners. In order to make clear why this is so, let me summarise the relevant characteristics of Japanese and Thai pronouns: a. in Japanese and Thai a large number of pronouns is available for each person specification (1st, 2nd etc.); b. some Japanese and Thai pronominals are actually either ambiguous between different person markings or unspecified for person; c. there exist Thai pronouns that are loanwords; d. Japanese and Thai can have their pronominals modified or as complements of D heads such as demonstratives.

185


Table 1.Thai 1st person pronounsa Semantic features Speaker

‘aadtamaa’ ‘aj ‘ua’ (1) ‘ua’ (2) chan’ (1) chan’ (2) dichan’ kan’ kraphom’ kuu’ khaa (1) khaa (2) khaaphacaw khaw phom’ raw’ (1) raw’ (2) raw’ (3)

Addressee

Attitude

Male

Adult

Male

Adult

Superior Intimate Nonrestraint

+ ?

/+ /+

?

/+ /+

/– / – / – /– /+ / /+ / – /

+ + – – + + +

+ + +

+

+ + /+ + /+ + /–

? +

+ +

/+

+

/+ /–

/ (/+) – – /

– +

– +

+

+

– + – ++ + ++ – ++ (–) – – +

– – ++ ++ – – – –

Cooke’s terms and notation: ‘++’ very marked, ‘/’ neutral. Other combinations for intermediate settings. a

1.3.1 Open pronominal classes From points (a) to (c), I conclude that the Japanese and Thai pronominal systems display some crucial characteristics of open classes. Hence, there is a great number of pronominal elements, at least thirteen for a 1st person specification in Thai, listed in Table 1. They can be introduced in the language with relative ease, through borrowing from Chinese and English in the case of Thai; choosing the ‘right’ pronominal form goes beyond Case or other grammatical restrictions but largely depends on (subtle) stylistic considerations, context and the discourse. Recall now that determiners, being members of a functional category, must form a closed class. In fact, they cannot but form closed classes (Abney 1987: 64f), this being one of the hallmarks of a class of functional elements. As to why functional heads like determiners necessarily belong to closed classes, Emonds (1985: Ch. 4) provides an account on the basis of formal

Pronominal nouns 187

features available by UG being restricted in number. Let us assume that mental grammars use a (proper?) subset of the UG pool of formal features. Let us further assume that functional heads are just combinations of formal features that denote no concept. If different functional heads are simply different combinations of formal features and formal features are drawn from a restricted subset, there can only exist a limited number of functional heads per grammar, differentiated from each other only by virtue of their different feature specifications. In order to clarify why this is important, compare the state of affairs for lexical heads: lexical heads can be differentiated from each other by virtue of their descriptive (concept-denoting) features (i.e. features that are interpreted as cat, thermometer, play, sweet etc.). It would be fairly uncontroversial to claim that pen (‘writing instrument’) and pen (‘enclosed space’) have identical formal features, but they are two different lexical entries by virtue of their different descriptive features that encode different concepts. Crucially, functional heads can encode only formal features, whereas lexical ones avail themselves to concept-denoting ones as well as secondary ‘semantic’ features, such as classificatory ones (like gender or conjugation class) or other. In order to illustrate why functional heads (and consequently, determiners) form closed classes, let us entertain the following thought experiment: Suppose there were only two formal features (apart from the categorial one) available in a particular language for determiners: α and β. Due to the unavailability of descriptive features for functional heads, there would be just four lexical entries (‘A–D’) for this category: Table 2.Possible lexical entries for formal features α and β

+β −β

+α

−α

A [α, β] C [α]

B [β] D

Having then in a grammar more than one functional element — like a determiner is — specified as, say, [+α] [+β] and as many as thirteen is hence impossible: there would be no way to distinguish between multiple occurrences of determiners, all encoding the formal feature combination meaning ‘1st person’. Therefore, if Japanese and Thai pronouns form open classes, they cannot be intransitive determiners. Finally, if determiners, like all functional heads, are each unique combinations of formal features, there can be no ‘space’ for loanwords, such as ‘aj and


‘au in Thai: pronouns as determiners are incompatible with loanword status, unless, of course, the loanword replaces an existing functional element as was the case with the introduction of they in English. 1.3.2 Pronominal features Perhaps we could still salvage an analysis in which Japanese and Thai pronouns are intransitive determiners if, according to the discussion above, we can show that every member of the paradigm, for example 1st person in Japanese and Thai, encodes unique formal feature combinations. For concreteness, let us follow Nolan (2000) and hypothesise that Japanese pronouns encode features such as [±Politeness] and [±Honorific]. For instance, different combinations of three such features would yield 23 = 8 different feature combinations per person, cf. (6). The apparent stylistic factors for the choice of, say, watasi instead of watakusi (or instead of nothing at all: Japanese is a null subject language) would now reduce to the feature specification of particular pronominal determiners for features such as [±Politeness] and [±Honorific]. Hence, differences between pronouns marking a particular person specification need not be identical to differences between nouns like freedom-fighter/rebel/terrorist. A first, practical problem with an analysis like the one above is that it is not easy to show that all the different first person pronouns in (4) or Table 1 encode different specifications of grammatical features. So, in Table 1, what grammatical feature could possibly distinguish kuu’ and khaa (2)? Even in Cooke’s profligate feature system, they appear as identically specified. Moreover, the difference between chan’ (1) and chan’ (2) seems to be like that between ‘2nd person plural’ vous and ‘polite’ vous. Do ‘2nd person plural’ vous and ‘polite’ vous — or chan’ (1) and chan’ (2) for that matter — differ from each other in terms of a feature? I would doubt so. Even if the above practical difficulties could be overcome, which would not be straightforward, there remains another problem. If all or most of the different pronominal forms in (4) and Table 1 encode different combinations of features like [±Politeness] and [±Honorific] are these the kind of features that functional categories can encode? In other words, are these formal features? Apparently, such features are not at all as ‘basic’ as Person, and the provisional answer to the questions above will be ‘no’, for reasons to be discussed and clarified in detail in Section 3.3. For the time being accept without discussion that not all features on Japanese and Thai pronouns are formal features. As far as the person ambiguity is concerned of forms like temae (Japanese; between 1st and 2nd person) and khaw (Thai; between 1st or 3rd person), this


indeed makes Japanese and Thai pronouns very untypical determiners, should they be determiners: nouns are typically unspecified for person, and person seems to be encoded on determiners.4 Person features are associated with ostension/deixis and they seem to belong to the same natural class with definiteness and deictic features. Given that under standard assumptions, definiteness and deixis are located on determiners, it can safely be assumed this to be the case for Person, as well. An additional argument towards this assertion could well be Ritter’s (1995: 421) observation that person entails definiteness, but not the other way round.5 If Japanese and Thai pronouns are heads, their occasional ambiguity for person could be a strong indication that they are either not determiners or determiners unmarked for person, like the. Towards this, according to Harley and Ritter (1997, this volume), a pronominal determiner that encodes ‘intimate attitude’ or ‘addressee status’ — like raw’ under Table 1 does — but is ambiguous for Person is impossible, as there exists a feature hierarchy inside the pronominal determiner according to which a lexical item cannot be specified for Class (gender, politeness, status etc.) when underspecified for Participant (person). 1.3.3 The syntactic behaviour of Japanese and Thai pronouns Compare the point illustrated in (1)–(3), namely that Japanese and Thai pronouns can appear in the complement of demonstrative determiners, after adjectives and as a possessum, to the situation in languages like English or German where pronouns occupy the syntactic positions of determiners (7a), resist modification by adjectives (7b), unless nominalised (e.g. this is your new you), and are incompatible with possessors (7c). These are some of the characteristics of a solid correlation between articles and pronominals that renders the inclusion of the latter in the class of determiners natural for a large number of languages:6 (7) a.

[DP The politicians] do not earn enough. [DP We politicians] do not earn enough. b. *[DP Desolate we] do not earn enough. c. *[DP Our them] do not earn enough.

1.4 Japanese and Thai pronouns are nouns I have shown that Japanese and Thai pronouns do not behave like (intransitive) determiners: they seem to form open classes; they encode features untypical of determiners — or other functional categories. As should be quite obvious by


now, a sound solution to the puzzles of Japanese and Thai pronouns is to admit that they are not (intransitive) determiners, but nouns. This would capture their belonging to open classes (or at least, the fact that they do not encode exclusively formal features). Moreover, the only obvious solution for their distribution inside the DP — they can be modified and appear in the complement of a determiner — is that they are nouns. Of course, this is not an original proposal; it has already been argued by Fukui (1987) and Noguchi (1997). Noguchi dubs Japanese pronouns N-pronouns, partly on grounds of the discussion above. For reasons of succinctness, the term N-pronoun is adopted here to signify pronouns that are of the category N, like in Japanese and Thai. At the same time, pronouns that appear to be determiners, like in English and German, are dubbed D-pronouns.

2. D-pronouns and N-pronouns 2.1 ‘One’ as an N-pronoun Interestingly, N-pronouns are hardly exotic produce and at least one N-pronoun seems to exist in English. This is one, a “pronominal count noun” according to Radford’s very felicitous characterisation (1993: 102). One can appear inside the complement of a demonstrative determiner (8a), be modified by an adjective (8b) or by a prepositional phrase (8c).7 (8) a. b. c. d.

[This one] is sometimes a challenge. You should carefully file [the new ones]! I find it annoying she lost [the one from New Jersey]. I have found [a new one].

A comparison between (8) and (1)–(3) is especially enlightening: one behaves exactly like N-pronouns, pace the presence of an overt determiner the.8 It is a normal noun, not only in terms of its distribution but, also, as far as morphology is concerned, as it can take plural inflection (8b). At the same time, one does not denote a concept (like nouns such as cat, thermometer, toy, chocolate etc. do): the DP containing it is a semantic variable restricted by the features inside the DP at LF (Heim and Kratzer 1998: Ch. 5, Ch. 9). For instance, a new one in (8d) can refer to anything new — or someone new as well, a reading pragmatically not particularly felicitous. A last point here is that most pronouns are generally argued to be phrases, that is DPs (see Cardinaletti and Starke 1999 for an in-depth discussion). Such

Pronominal nouns

an analysis becomes more evident when the DPs containing one or Japanese pronominal expressions such as ano hito (lit. ‘this person’, i.e. ‘s/he’) are included in the picture: these are pronominal expressions and both their determiner and the noun constituents are visible. 2.2 Postal (1969) and ‘one’ Paul Postal, to the best of my knowledge, was the first to notice the pronominal nature of one that co-exists with its morphosyntactic status as a normal noun. In a paper dating back to at least 1966, cited here as Postal (1969), he observed the following ‘twist’ regarding phrases like these under (7), adapted here as (9) below: (9) a. [DP The politicians] do not earn enough. b. [DP We politicians] do not earn enough. c. [DP We] do not earn enough.

In (9b) there is nothing pronominal about the DP we politicians — this phrase refers to a group of politicians that the author of the speech event (as in Halle 1997) is a member of. In terms of interpretation, it is closer to the definite description the politicians in (9a) than the pronoun we in (9c). Postal correctly argues that this is so because ‘pronominal determiners’ like we are mere articles, albeit marked for Person, 1st plural in (9b). In other words, we is the 1st person plural version of the — not a pronoun as such. He then goes on to argue that when articles like we appear with no visible complement — as is the case in (9c) — they are in fact complemented by one in Deep Structure, which then gets deleted by Surface Structure. Whether one accepts the possibility of deletion operations like the one posited here or not, the essence of Postal’s proposal is the following: ‘pronominal reference’ is triggered not by determiners (‘articles’), even if they encode Person, but by a special noun in their complement, whether visible (like one) or not.9 I attempt to recast and expand this insightful intuition in what follows.

3. Are D-pronouns determiners without a complement? 3.1 Why bare D-pronouns are attractive It is exactly the idea that pronouns consist of a determiner and a null noun that Abney (1987: 281–284) challenges. The spirit of his ‘intransitive determiner’

191


hypothesis for pronouns is essentially predicated on simplicity. A determiner complemented by a pronominal noun (like one or a null version thereof) will have the same semantic import with a determiner that radically lacks a nominal complement: in both instances, there will be no concept denoted by the noun of the DP; this can be either because of the ‘pronominal’, non-descriptive nature of the noun or because there is no noun there at all! Why then go for a representation like [DP [D we] [NP N non-descriptive]] for (9c), when [DP [D we]] will have exactly the same LF import? On grounds of parsimony and simplicity, the intransitive determiner hypothesis for pronouns is indeed an extremely attractive one. 3.2 Two ways to pronominal reference? If all pronouns were determiners, then an analysis positing that they are determiners without a nominal complement would indeed be the optimal hypothesis. Given the discussion of N-pronouns and one, however, it should be evident that pronominal reference triggered by the absence of a noun in a DP cannot be the whole story. Pronominal nouns, that is nouns that denote no concept, do exist: one as well as Japanese and Thai N-pronouns seem to be some of them. One then would be led to assume that pronominal reference of a DP can be triggered in either way (either by a non-descriptive noun or by the absence of a noun) and formulate the following disjunctive statement: Pronominal reference of a DP at LF can be triggered by a. either the radical absence of an included noun — in languages like English and German, b. or a non-descriptive included noun — one and N-pronouns in languages like Japanese and Thai. As far as learnability is concerned, it could be claimed that D-pronouns, being intransitive determiners, will not tolerate any other determiners or modifiers and will form a closed class. On the other hand, N-pronouns will be nouns that do not denote concepts and will be used in DPs with pronominal reference; they will co-occur with adjectives and be able to be headed by determiners such as articles or demonstratives. Now, disjunctive statements are not in general theoretically desirable and, moreover, the particular one above seems not to capture the point: pronominal reference at LF is not an idiosyncratic property of a class of DPs that happens to include case (a.) and (b.) above but more of a coherent phenomenon that seems to have a unique syntactic trigger. Let us then recast it as under (10).


(10) LF condition on pronominal reference (provisional) When an extended nominal projection not headed by a concept-denoting noun reaches LF, it is interpreted as a semantic variable.10

The statement in (10) derives the disjunctive formulation offered earlier, while offering a single trigger for pronominal reference at LF in Syntax: lack of a concept-denoting noun. Naturally, the obvious cases in which an ‘extended nominal projection’ (in the sense of Grimshaw 1991) will not be headed by a concept-denoting noun are when this Extended Projection is a. either headed by a non-descriptive N (N-pronoun or one), b. or when there is no noun there at all: the (trivial) DP will be an intransitive determiner. Prima facie, (10) appears to be a statement that satisfactorily captures all the instances in which a nominal phrase will be interpreted as pronominal at LF cross-linguistically. Nevertheless, in the following sections I examine evidence that points towards (10) being too strong in terms of expressive power, as well as that it predicts the existence of empirically unattested entities: ‘intransitive determiners’. 3.3 The question of gender A first sign that pronouns even in languages like English and German cannot be determiners without a nominal complement is the fact that 3rd person pronouns in these, and other languages, are marked for gender. This was no problem for Abney because he assumes gender, as well as the rest of features like person and number, to be encoded on D heads (Abney 1987: 283). He partly bases this argument on the “full range of inflectional distinctions” that German pronouns — ‘intransitive determiners’ according to him — display, in contrast with German nouns. For Abney, this is enough and clear indication the determiner is “the actual site” of features (ibid.). This cannot be entirely accurate, though, for reasons both empirical and conceptual. Focusing on gender, I am going to show that it is not an inherent feature of determiners and that it is marked on them only as a result of agreement, “concord” more specifically (cf. Carstens 2000) with something else. To illustrate this, take French, a language that does not unambiguously mark gender on nouns but does so on determiners and agreeing adjectives.11 To start by stating the obvious, no one would wish to claim that the locus of gender is on the adjective, although some adjectives are marked for gender in a way


clearer than nouns are in French. At the same time, adjectives are marked for gender due to concord in French or any similar language, even when determiners are not marked for gender or when determiners are altogether absent; consider the following: (11) a.

pas de forêts vertes (/vert/) not ptcl forests green.pl.fem b. pas de bois verts (/ver/) not ptcl woods green.pl.masc ‘no green forests/woods’

In both instances of (11), there is no (overt) determiner to inflect for gender. At the same time, the adjective vert does inflect for gender, apparently as a result of concord. To keep assumptions to a minimum, let me just point out that morphology is not a secure guide towards deciding (a) which the locus of a feature in a nominal phrase is, and (b) what elements are marked for this feature as a result of concord. Dwelling a bit more on French, consider nouns like poste, which have a totally different meaning according to which gender they are specified for, as in (12a). Similar facts hold for the Spanish nouns libro and libra, as in (12b): (12) a.

le poste, la poste the.masc appliance the.fem post b. el libro, la libra the.masc book the.fem pound

If gender is located on determiners, articles in this case, the following paradox ensues: the gender specification on D can actually decide the lexical semantics of nouns. Note that this is not a contrast of the lion–lioness sort — there is no relationship between appliances and the post, or books and pounds. The pairs in (12) are minimal pairs and the only feature that tells their meaning apart is gender. The only reasonable explanation would then be to posit that gender is encoded on nouns: determiners, when marked for gender, show concord with them, just like adjectives.12 Such a conclusion is in accord with the idea that gender is a noun-classificatory feature (cf. Corbett 1991). Harris (1991: 34) has reached the same conclusion on independent grounds, again with reference to concord inside the nominal phrase: “like determiners and other noun modifiers, adjectives have no inherent gender; they do, however, show gender concord with the noun they modify.” Similar assumptions were held by traditional grammarians, as well.


The idea that gender is part of the lexical entry for nouns is also supported by the fact that it has to be learnt — the vast majority of nouns in languages with gender distinctions have a fixed gender specification. When the specification can be variable, as in (12), there is a change of meaning as a consequence. Moreover, in nouns denoting concepts with ‘prototypical’ sexual gender, like German Mutter (‘mother’) and Vater (‘father’), grammatical gender reflects sexual gender, i.e. grammatical gender is intimately related to properties of the concept the noun denotes. This contrasts gender with number: number specifications such as mass (as in Delfitto and Schroten 1991), singular or plural are available for all nouns, even when this entails a forced shift in interpretation (sugars meaning ‘teaspoons’ or ‘sachets’ or ‘cubes’ of sugar). To keep things simple: gender is learnt with the noun, number is not; gender features are part of the lexical entry for a noun. The careful reader would have by now noticed that a contradiction emerges: if gender is a feature encoded on nouns, then pronouns marking gender cannot be ‘intransitive determiners’ — gender concord on D must be with some noun. In other words, German 3rd person pronouns er (‘he’) and sie (‘she’), for instance, cannot be determiners without a nominal complement because gender is marked on them, masculine and feminine respectively; if they are agreeing determiners then they must agree with something, and given that it is gender they agree for, this ‘something’ must be a noun encoding the gender feature. To put it plainly, gender marking on a pronoun is evidence (for both the language learner and the linguist) that the pronoun cannot be an ‘intransitive determiner’ but that it must have some sort of nominal complement, as nouns are the locus of gender features. See also Felix (1990), Rouveret (1991), Zwarts (1993), Cardinaletti (1994), Ritter (1995), Chomsky (1995: 337–8), Uriagereka (1995), Corver and Delfitto (1999), Cardinaletti and Starke (1999) and Koopman (2000) for analyses of pronouns as determiners with a (silent) nominal complement. To summarise, gender features are tell-tale signs of the presence of a noun because we do not want noun-classificatory features located on functional heads like D, exactly the way we would not like conjugation class morphology to be primarily encoded on a verbal functional head, say v/Voice, Asp or T. It is also most likely that these noun-classificatory features, always encoded on nouns, include features such as [±Politeness] and [±Honorific], the Class features of Harley and Ritter (this volume). A residual issue is the following: in most languages that possess them, at least some D-pronouns are morphologically very different from the determiners.


This has led Koopman (2000) to argue that such pronouns are Num(ber) P(hrases) and not DPs, but I believe that an alternative exists. Siding again with Postal (1969) we can reverse the argument and say that pronominal forms that do not look like other determiners, like articles, are allomorphs of exactly these determiners. Take the, uncontroversially a determiner, and the pronoun it in English: they appear to encode the same features and it is the case that they are in complementary distribution. Moreover, the complementary distribution of the and it is compounded by whether these two forms, allomorphs if we are right, can tolerate a phonologically overt complement. Consider (13) below, where e is a null empty noun. (13) a. [DP the [N car]] b. *[DP the [N e]]

*[DP it [N car]] [DP it [N e]]

Hence, the simplest explanation for ‘intransitive’ determiners like it is that they are allomorphs of determiners like the when the complement is phonologically covert, as in (13b), and it is for this reason that *it car in (13a) is not ungrammatical. The same holds for possessives like mine e vs. my ones — whether mine/my is a D or at SpecDP — and similar facts are responsible for the only partial overlap between the demonstrative and article paradigm on the one hand and the pronominal paradigm on the other.

4. Eliminating the redundancy In the previous section I reviewed a number of empirical arguments against an analysis of 3rd person pronouns that are marked for gender as intransitive pronominal determiners. Moreover, I argued for a noun complement, phonologically null apparently, for these pronominal determiners; the locus of gender is on this null noun and the determiner displays concord for gender with it. Nevertheless, a serious question remains unanswered: can a determiner that does not mark gender, like I or du (‘you’) stand alone? If it can stand alone, it will be an instance of an ‘intransitive determiner’ and, by (10), receive a pronominal interpretation at LF. 4.1 On the peculiar semantic status of ‘intransitive determiners’ Let us now informally discuss the semantic function of determiners. Intricacies concerning definite descriptions and related matters aside, determiners are of


the semantic type ··e,tÒ,eÒ: they take a nominal predicate (the nominal expression) and yield an individual (Heim and Kratzer 1998: 52–53). A determiner like the takes a nominal predicate like key and yields the individual referred to by the expression the key, thus the function ··e,tÒ,eÒ that the determiner denotes is satisfied at LF.13 This has also been expressed in studies like Longobardi (1994) as a statement that determiners encode referentiality. Alternatively, we can follow Higginbotham’s (1985) account of the relationship between a noun and its determiner: a noun like key is a predicate that has an open position, a θ role it needs to assign. The determiner the receives this θ role. The two accounts, namely that the determiner needs a nominal predicate to satisfy the function it denotes and that the noun has an open position satisfied by the determiner, are incompatible with each other.14 Nevertheless, both have as a result that a determiner cannot stand on it own: ‘intransitive determiners’ should not exist. Contrary to this, let us assume that a determiner can stand on its own, that it can be without a nominal complement. This would result in a violation of principles in either analysis. According to Heim and Kratzer’s story, the function an intransitive determiner denotes would remain unsaturated due to the radical absence of a predicate (the ·e,tÒ bit), that is, a noun. According to Higginbotham, the determiner would not be θ marked by a noun and thus not licensed. Whichever the most satisfactory treatment, an intransitive determiner would be an offending element at LF.15 Thus, intransitive determiners must be impossible. Let us see now why an elementary noun, whether specified for features like gender or radically impoverished — like D in Table 2 — enables a seemingly ‘intransitive’ determiner to stand alone. Let us further make the straightforward hypothesis that an elementary noun is a non-concept denoting noun specified for at least a categorial feature [N], a grammatical noun (Emonds 1985). In a Heim and Kratzer style analysis, an elementary noun would denote a ‘trivial’ function of the type ·e,tÒ that would take any individual and yield a ‘true’ value for it — it is easy to see why the presence of such a noun in a DP would consequently trigger pronominal reference at LF. Similarly, according to Higginbotham’s story, an elementary noun would assign just the one θ role by virtue of its categorial feature [N] and, thus, license a seemingly ‘intransitive’ determiner like they. The grammatical noun’s lack of descriptive content and, consequently, lexical semantics would trigger a ‘pronominal’ reference at LF, as by (10). Again, the non-descriptive noun would denote a ‘trivial’ predicate that would be true for any individual.


The elimination of the concept of ‘intransitive determiners’ from a theory of grammar presents another conceptual advantage: Abney (1987:285) notices that determiners differ from other functional elements [as] determiners sometimes appear without a complement […] the appearance of functional elements as “intransitives”, in a pronominal usage, constitutes a systematic exception to the otherwise general requirement that they take an obligatory complement.

If determiners without complements do not exist, determiners are no different from the rest of functional categories that obligatorily take a lexical head somewhere in their complement.16 Hence, a source of anomaly caused by just a handful of lexical entries within the functional category of Determiner, as not all Determiners (can) appear without overt complement, can be dispensed with. 4.2 A (non-descriptive) noun inside every pronoun If the discussion above is on the right track, then ‘intransitive determiners’ are not an option available in UG. The presence of a non-descriptive noun is forced in any case: its lack of denotation triggers the pronominal reference of the DP that contains it. This noun can be either overt, like kare, khaw or one, or null, like the complement of we in (9c). It can encode either semantic and formal features, like gender on the noun in the complement of er or features like [±Politeness] and [±Honorific] on some Japanese pronouns under (4), or just a categorial [N] feature, as it seems to be the case with indexicals like I and du. A more restrictive, and more accurate, version of (10) can now be offered: (14) LF condition on pronominal reference (revised) When an extended nominal projection headed by a non concept denoting noun reaches LF, it is interpreted as a semantic variable.

The condition above unifies N-pronouns and our more familiar D-pronouns. If the D category is indeed universal,17 as Noguchi (1997), Progovac (1998), Cinque (1999) among others would argue, all pronouns cross-linguistically have the form of [DP D [NP N ]]. This in turn would mean that, as far as the PF representation of pronouns is concerned, there are the following possibilities: a. The D is null and the N overt: this is the case of Thai pronouns as well as Japanese pronouns under (4), which we descriptively dubbed N-pronouns after Noguchi (1997);18 b. The D is overt but the N is not: these are the familiar cases of Indo-European pronouns as well as pronominal clitics (Uriagereka 1995; Corver and Delfitto 1999; Cardinaletti and Starke 1999);


c. Both D and N are overt: this is the case of the new one in English or ano hito in Japanese; d. Both D and N are null, hence the whole DP is phonologically null. This null DP has been argued by Radford (1993) and Koopman (2000) to be pro, but see arguments against this in Panagiotidis (2002: Ch. 4). It is more likely that a DP containing a null D and a null N is the representation for nonvariable null indefinite arguments19 in languages like Brazilian Portuguese, Quiteño Spanish, Modern Greek and Bulgarian (Dimitriadis 1994a; 1994b). See Giannakidou and Merchant (1996) and Panagiotidis (2002: Ch. 2) for an account. In all the four cases above the unifying factor is the lack of concept-denoting features on N — the grammatical trigger of ‘pronominality’ expressed under (14).

5. Conclusion In this chapter it has been shown that there exist languages which possess pronouns that are not determiners (without a nominal complement), these languages including Japanese and Thai. Arguments against the determiner status of the pronouns in such languages were drawn from the fact that they form open classes, from the kind of features they encode and from the positions they occupy within the nominal phrase. It was consequently shown that such N-pronouns exist also in languages that do possess paradigms of D- pronouns, one is a case in point. It was further shown that gender marking on D, gender being a feature of N, is always a tell-tale sign of the presence of an N inside the nominal phrase, including those pronominal noun phrases that mark gender. Subsequently, the impossibility of ‘intransitive’ determiners was argued on the basis of the function that determiners denote. Finally, the LF condition on pronominal reference was formulated as a semantic variable interpretation, triggered by the presence of a non-concept denoting noun in a nominal projection at LF.

Notes * I am grateful to Heidi Harley, Stephen Nolan, Roger Hawkins, Bob Borsley, Karl Johnston, Heike Wiese, Horst Simon and the audience at the 2000 DGfS meeting in Marburg for stimulating discussion and interesting comments. I also wish to thank two anonymous


referees for their helpful, detailed and constructive comments. All errors and misconceptions, as is obvious, remain mine. 1. A similar phenomenon occurs in Italian. Cardinaletti (1994) discusses examples like: (i) la mia lei the my.fem she ‘my she’ (i.e. ‘my girlfriend’) Cardinaletti, nevertheless, points out that the process yielding examples like the one above is not productive. 2. Adjectival modification, like in (3), is also possible in Thai: phom’ nîi (‘me this’). 3. Table from Cooke (1968: Ch. 1). I cannot endorse all the features and feature specifications in Table 1 as grammatical features; e.g. instead of positing two almost identical lexical entries for chan’, one of which encodes a ‘male speaker’ value and the other a ‘non-male speaker’ one, we could just state that in terms of chan’ being underspecified for ‘speaker gender’. Moreover, diacritics such as ‘++’, standing for ‘very marked’ don’t make much sense in a feature system with binary values. 4. As an anonymous reviewer observes, Japanese and Thai pronouns seem to be closer to nouns like committee than determiners: if one is a member of the committee and utters “it’s the committee’s opinion that the matter should be adjourned”, the DP the committee is, in a sense, a first person plural expression. 5. See Lyons (1999: Ch. 5) for an excellent review of the relationship between person and definiteness. 6. Such correlation is even more evident in the paradigm of pronominal clitics in Romance languages like Italian and Spanish: clitics and definite articles are morphologically identical for most feature combinations. 7. The ban on expressions like John’s one, that is where one is a possessum, probably has to do with some sort of “lexical economy”, as the alternative elliptical John’s is also available. In any case, expressions like John’s ones and my ones are fine in a lot of dialects of English (Karl Johnston, p.c.). Of course, these are not the only complications regarding the occurrence of one, consider *a one and the one *(that I saw). An anonymous reviewer argues, in the face of such complications, that one is a classifier, approximately as in Kester (1996). See also Panagiotidis (2002: Ch. 2) for arguments against this and for relevant discussion. 8. Whether N-pronouns are complements of a null (and possibly underspecified) determiner in article-less languages is a matter for debate. Noguchi (1997: 786) claims that only demonstratives are determiners in Japanese. See also Progovac (1998) for determiners in SerboCroatian, also an article-less language. 9. Or any other N-pronoun, we can add. 10. This term should be neutral regarding the controversial issue of whether DPs or other functional projections are universal (Cinque 1999) or not (Thráinsson 1996). See also footnote 8. 11. See also Corbett (1991: 57–62). I wish to thank an anonymous reviewer for suggesting an optimal example to illustrate the point made here.


12. Ritter (1993) argues that a degree of parametrisation exists: gender can be encoded on either N or Num (an intermediate functional head between D and N). Even if this is accurate, gender would not be located on D. 13. All functions must be satisfied at LF — this is Heim and Kratzer’s (1998:49) Principle of Identification. 14. I wish to thank an anonymous reviewer for pointing this out to me. 15. Panagiotidis (2002: Ch.5) attempts to capture this in terms of syntactic feature checking: determiners encode an uninterpretable categorial [N] feature which leads to crash at LF unless checked by a noun. 16. This requirement can perhaps be formulated along the lines of Grimshaw’s (1991) Extended Projection. 17. See also footnote 8. 18. If we are correct in assuming that person is encoded on D, then the unambiguous Thai pronouns under (5) should be DPs headed by a null 1st person D, while forms such as raw’ could be headed either by a 1st or a 2nd person D, or an underspecified one, all null. Similar facts would hold for Japanese. For person being encoded on N, see Rouveret (1991) and Shlonsky (1997: 122–3). 19. These null arguments are not restricted by locality, unlike Portuguese null variable arguments (Raposo 1986).

References Abney, S. P. 1987. The English Noun Phrase in its Sentential Aspect. PhD. Diss., MIT, Cambridge, MA. Cardinaletti, A. 1994. “On the internal structure of pronominal DPs”. Linguistic Review 11: 195–219. Cardinaletti, A. and Starke, M. 1999. “The typology of structural deficiency; a case study of the three classes of pronouns”. In Clitics in the Languages of Europe, H. van Riemsdijk (ed.), 145–233. Berlin and New York: de Gruyter. Carstens, V. 2000. “Concord in Minimalist Theory”. Linguistic Inquiry 31: 319–355. Chomsky, N. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Cinque, G. 1999. Adverbs and Functional Heads. Oxford: Oxford University Press. Cooke, J. R. 1968. Pronominal Reference in Thai, Burmese and Vietnamese [University of California Publications in Linguistics 52]. Berkeley: University of California Press. Corbett, G. 1991. Gender. Cambridge: Cambridge University Press. Corver, N. and Delfitto, D. 1999. “On the nature of pronoun movement”. In Clitics in the Languages of Europe, H. van Riemsdijk (ed.), 799–861. Berlin and New York: de Gruyter. Delfitto, D. and Schroten, J. 1991. “Bare plurals and the number affix in DP”. Probus 3: 155–185. Dimitriadis, A. 1994a. “Clitics and island-insensitive object drop”. Studies in the Linguistic Sciences 24: 153–170.


Dimitriadis, A. 1994b. “Clitics and object drop in Modern Greek”. MIT Working Papers in Linguistics 23: 95–115. Emonds, J. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris. Felix, S. 1990. “The structure of functional categories”. Linguistische Berichte 125: 46–71. Fukui, N. 1987. The Theory of Projection in Syntax. Stanford: CSLI publications. Giannakidou, A. and Merchant, J. 1996. “On the interpretation of null indefinite objects in Greek”. Studies in Greek Linguistics 17: 141–155. Grimshaw, J. 1991. Extended Projection. Ms., Brandeis University, Waltham, MA. Halle, M. 1997. “Distributed morphology: Impoverishment and fission”. MIT Working Papers in Linguistics 30: 425–449. Harley, H. and Ritter, E. 1997. Meaning in morphology: Motivating a feature-geometric analysis of person and number. Ms., University of Calgary. Harley, H. and Ritter, E. this volume. “Structuring the bundle: a universal morphosyntactic feature geometry”. Harris, J. W. 1991. “The exponence of gender in Spanish”. Linguistic Inquiry 22: 27–62. Heim, I. and Kratzer, A. 1998. Semantics in Generative Grammar. Oxford: Blackwell. Higginbotham, J. 1985. “On semantics”. Linguistic Inquiry 16: 547–593. Kester, E.-P. 1996. The Nature of Adjectival Inflection. Utrecht: OTS. Koopman, H. 2000. “The internal and external distribution of pronominal DPs”. In The Syntax of Specifier and Heads, H. Koopman, 77–118. London: Routledge. Longobardi, G. 1994. “Reference and proper names: A theory of N-movement in Syntax and Logical Form”. Linguistic Inquiry 25: 609–665. Lyons, C. 1999. Definiteness. Cambridge: Cambridge University Press. Noguchi, T. 1997. “Two types of pronouns and variable binding”. Language 73: 770–797. Nolan, S. 2000. The semantic features of pronouns in Standard Japanese. Paper presented at the 2000 Meeting of the DGfS, Marburg. Panagiotidis, P. 2002. Pronouns, Clitics and Empty Nouns: ‘Pronominality’ and Licensing in Syntax. Amsterdam and Philadelphia: John Benjamins. Postal, P. 1969. “On so-called ‘pronouns’ in English”. In Modern Studies in English, D. Reibel and S. Schane (eds), 201–224. New York: Prentice-Hall. Progovac, L. 1998. “Determiner phrase in a language without determiners”. Journal of Linguistics 34: 165–179. Radford, A. 1993. “Head-hunting: on the trail of the nominal Janus”. In Heads in Grammatical Theory, G. Corbett et al. (eds), 73–113. Cambridge: Cambridge University Press. Raposo, E. 1996 “On the null object in European Portuguese”. In Studies in Romance Linguistics 24, O. Jaeggli and C. Silva-Corvalán (eds), 373–390. Dordrecht: Foris. Ritter, E. 1993. “Where is gender?”. Linguistic Inquiry 24: 795–803. Ritter, E. 1995. “On the syntactic category of pronouns and agreement”. Natural Language and Linguistic Theory 13: 405–443. Rouveret, A. 1991. “Functional categories and agreement”. Linguistic Review 8: 353–387. Shlonsky, U. 1997. Clause Structure and Word Order in Hebrew and Arabic. Oxford: Oxford University Press. Thráinsson, H. 1996. “On the (non)universality of functional categories”. In Minimal Ideas: Syntactic Studies in the Minimalist Framework, W. Abraham (ed.), 253–281. Amsterdam and Philadelphia: John Benjamins.


Uriagereka, J. 1995. “Aspects of the syntax of clitic placement in Western Romance”. Linguistic Inquiry 26: 79–123 Zwarts, J. 1993. “Pronouns and N-to-D movement”. In OTS Yearbook 1993, M. Everaert et al. (eds), 93–112. Utrecht: Research Institute for Language and Speech.

Harmonic alignment and the hierarchy of pronouns in German* Gereon Müller IDS Mannheim

1.

Introduction

Different types of (personal) pronouns behave differently with respect to syntactic operations like Wackernagel movement, R-pronoun formation, coordination, and topicalization. This motivates the existence of pronoun classes. I assume that pronouns are ordered according to strength; the hierarchy that emerges for German is given in (1). (1) Personal Pronoun Scale: Pronstrong > Pronunstressed > Pronweak > Pronreduced (> Pronclitic) ihn[+stress] ihn[+anim] ihn[–anim] es (> ’s)

The phonological feature [±stress] distinguishes between strong pronouns (which are capitalized here and henceforth) and others. The semantic feature [±animate] distinguishes between what I call unstressed pronouns and weak pronouns (cf. Cardinaletti and Starke 1996). The phonological feature [±reduced] distinguishes between weak pronouns and reduced pronouns; the only reduced pronoun in Standard German is es. Finally, the phonological feature [±clitic] distinguishes between reduced pronouns and clitic pronouns. In contrast to what is the case with the other pronoun classes, the syntactic behaviour of clitic pronouns is mainly governed by phonological (rather than syntactic) constraints related to the distinction between pro- and enclisis, properties of potential clitic hosts, and so on (cf. Gärtner and Steinbach 2000, Zifonun 2001). In what follows, I abstract away from this last class. Tableau 0 shows how the remaining four pronoun classes behave with respect to the four syntactic operations mentioned above. Interestingly, the pattern is not arbitrary, but obeys strict implicational generalizations: If a given

206 Gereon Müller

pronoun undergoes Wackernagel movement, all weaker pronouns also undergo this operation; if a given pronoun undergoes R-pronoun formation (optionally or obligatorily), then all weaker pronouns do so as well; and if a given pronoun resists coordination and topicalization, all weaker pronouns share this property. Tableau 0.Pronoun classes and syntactic operations Pronouns

Wackernagel movement

R-pronoun formation

coordination

topicalization

strong unstressed weak reduced

– obligatory obligatory obligatory

– – optional obligatory

optional optional – –

optional optional – –

The main goal of this paper is to account for this pattern. The basic assumption is that for all four pronoun classes there is a constraint that forces a pronoun to show up in the Wackernagel position. All these Wackernagel constraints can in principle be violated; but the stronger a pronoun, the more likely the violability. Since this requires ranking and violability of constraints, I will develop the approach within optimality theory.1 Furthermore, I will employ the optimality-theoretic mechanism of harmonic alignment in order to derive a fixed ranking among the four Wackernagel constraints on the basis of the pronominal hierarchy in (1). Variable behaviour with respect to the three remaining syntactic operations will then be shown to result from the way other constraints are interspersed among the Wackernagel constraints.2 I will proceed as follows. In Section 2, I introduce the concept of harmonic alignment. In Section 3, harmonic alignment is applied to the scale in (1), which accounts for the distribution of Wackernagel effects with pronouns in German. Based on this, Sections 4, 5, and 6 address R-pronoun formation, coordination, and topicalization. A conclusion is drawn in Section 7.

2. Harmonic alignment Prince and Smolensky (1993) develop harmonic alignment as a meta-principle that takes two indepedently motivated scales (hierarchies) as input and generates constraint hierarchies with a fixed internal order on this basis. The operation is defined in (1).

Harmonic alignment and the hierarchy of pronouns in German 207

(2) Harmonic Alignment (Prince and Smolensky 1993: 136): Suppose given a binary dimension D1 with a scale X > Y on its elements {X,Y}, and another dimension D2 with a scale a > b > … > z on its elements {a,b,…,z}. The harmonic alignment of D1 and D2 is the pair of harmony scales HX, HY: a. HX: X/a X/b … X/z b. HY: Y/z … Y/b Y/a The constraint alignment is the pair of constraint hierarchies CX, CY: i. CX: *X/z … *X/b *X/a ii. CY: *Y/a *Y/b … *Y/z

Thus, given two scales, one of them binary, two harmony scales are derived by harmonic alignment. The first harmony scale results from combining the first member of the binary scale with the members of the other scale from left to right. The second harmony scale is created by combining the second member of the binary scale with the members of the other scale, this time proceeding from right to left. In the (abstract) case at hand, the harmony scale HX states that it is most harmonic for an X to be (more generally, to be associated with) an a, and least harmonic for an X to be a z; and HY implies that a Y is ideally a z, and the worst case is for a Y to be an a. The constraint hierarchies with their fixed internal order, CX, CY, are then derived from the respective harmony scales HX, HY by constraint alignment. Constraint alignment consists of reversing the order, replacing the symbol (“is more harmonic than”) with the standard constraint ranking symbol (“is ranked higher than”), and prefixing each member of the harmony scale with a star * that tells us to avoid a given configuration. Accordingly, the fixed order in, e.g., the constraint hierarchy CX implies that the ban on an X being a z is always highest-ranked, and that the ban on an X being an a is always lowest-ranked. The empirical domain for which Prince and Smolensky (1993) develop harmonic alignment concerns syllable structure. The constraint H-Nuc (“Nuclear harmony constraint”) requires syllables to have the most sonorous nucleus. This constraint differs from other optimality-theoretic constraints in that it is not local (i.e. in order to find out how well a candidate fulfills the constraint, it does not suffice to only consider the candidate itself), but rather translocal (the candidate must be compared with other candidates so as to find out which candidate has the most harmonic nucleus). Thus, evaluation of a candidate with respect to H-Nuc cannot consist of a simple yes/no decision (respected or violated); it requires an additional (albeit trivial) optimization procedure (much like transderivational constraints in some versions of the

208 Gereon Müller

minimalist program). Prince and Smolensky (1993) note that this is theoretically unattractive and suggest deconstructing H-Nuc by deriving its effects by harmonic alignment on the basis of two simple scales, viz., the syllable position hierarchy in (3a), and the sonority hierarchy in (3b). (3) a.

Syllable Position Prominence: P > M (Peak > Margin) b. Sonority Hierarchy: a>i>…>t

Harmonic alignment creates two harmony scales. One of these, HP, states that it is most harmonic for a syllable peak to be a low vowel, and least harmonic for a peak to be a voiceless alveolar stop. (4) Harmonic Alignment: a. HP: P/a P/i … P/t b. HM: M/t … M/i M/a

Finally, constraint alignment produces two hierarchies with fixed order. CP implies that the constraint blocking voiceless stops in the syllable peak always outranks the constraint blocking voiced stops in this position, that all constraints blocking consonants in this position outrank those blocking vowels, and that the lowest-ranked constraint is the one that blocks low vowels in the syllable peak. (5) Constraint Alignment: a. CP: *P/t … *P/i *P/a b. CM: *M/a *M/i … *M/t

Given that every syllable must have some segment in its peak, it is clear that the CP constraint hierarchy forces the choice of the most sonorous syllable peak in the optimal candidate, and thereby derives the effects of H-Nuc. Hence, given harmonic alignment, this translocal constraint can be dispensed with. In addition, the system is now more flexible because other constraints can be interspersed among the constraints of the syllable peak constraint hierarchy. Harmonic alignment has been applied to syntax by Aissen (1999) and Artstein (1999). Among other things, these authors discuss the influence of a person scale (1./2. Pers. > 3. Pers.) on constraints that regulate the mapping from arguments to grammatical functions (Aissen), and on constraints that govern pro-drop (Artstein). In the next section, I show how harmonic alignment can account for Wackernagel movement in German.


3. Wackernagel movement 3.1 Empirical evidence for Wackernagel movement It has often been noted that certain types of pronouns must show up in a leftperipheral (“Wackernagel”) position following C in German.3 In this position, the pronouns precede non-pronominal arguments (with one exception, see below) and adverbs of all types. The dividing line is between strong pronouns on the one hand, and unstressed, weak, and reduced pronouns on the other hand. As shown in (1b), strong pronouns do not have to undergo Wackernagel movement. The question arises of whether they may undergo Wackernagel movement. (1a) is fairly deviant, although perhaps not completely ill-formed. I take this to mean that whereas Wackernagel movement is not an option with strong pronouns, these pronouns behave in many respects like non-pronominal NPs and can thus undergo scrambling (see below for the distinction). Under this view, (1a) is deviant for the same reason that scrambling of stressed NPs generally is in German; see Lenerz (1977). (6) a. ?*daß ihr1 der Fritz gestern t1 ein Buch geschenkt hat that her.dat art Fritz.nom yesterday a book.acc given has b. daß der Fritz gestern ihr1 ein Buch geschenkt hat that art Fritz.nom yesterday her.dat a book.acc given has

In contrast, other personal pronouns obligatorily undergo Wackernagel movement. This is shown for unstressed (animate) pronouns in (7), for weak (inanimate) pronouns in (8), and for reduced pronouns in (9). (7) a.

daß ihr1 der Fritz gestern t1 ein Buch that her.dat art Fritz.nom yesterday a book.acc geschenkt hat given has b. *daß der Fritz gestern ihr1 ein Buch geschenkt hat that art Fritz.nom yesterday her.dat a book.acc given has der Fritz gestern der Maria t1 daß sie1 that her.acc art Fritz.nom yesterday art Maria.dat geschenkt hat given has b. *daß der Fritz gestern der Maria sie1 geschenkt hat that art Fritz.nom yesterday art Maria.dat her.acc given has

(8) a.

210 Gereon Müller

(9) a.

daß es1 der Fritz gestern der Maria t1 gegeben hat that it.acc art Fritz.nom yesterday art Maria.dat given has b. *daß der Fritz gestern der Maria es1 gegeben hat that art Fritz.nom yesterday art Maria.dat it.acc given has

There is one exception to the generalization that pronouns in Wackernagel positions must precede all clause-internal arguments. Subjects, and only subjects, may precede pronouns in Wackernagel positions clause-internally. Compare, e.g. (9a) with (10a) (where the subject NP precedes the reduced pronoun es) and (10b) (where es is preceded by the indirect object NP). (10) a.

daß der Fritz es1 t gestern der Maria t1 gegeben hat that art Fritz.nom it.acc yesterday art Maria.dat given has gestern t t1 gegeben hat b. *daß der Maria es1 der Fritz that art Maria.dat it.acc art Fritz.nom yesterday given has

I assume that the well-formedness of (10a) alongside (9a) is due to the fact that subject raising from the vP-internal base position to SpecT is optional in German (cf. Grewendorf 1989:Ch. 3), and that SpecT precedes the Wackernagel position.4 Abstracting away from the question of how the optionality of subject raising is derived, it seems clear that the phenomenon in (10a) is independent of Wackernagel movement per se. 3.2 Analysis of Wackernagel movement One might take the evidence in (6)–(10) to suggest that unstressed, weak and reduced pronouns undergo Wackernagel movement to a specific functional projection that intervenes between TP and vP. Here, I will adopt a slightly different approach. Under this view, Wackernagel movement of pronouns is formally indistinguishable from scrambling of non-pronominal NPs; both operations involve substitution in an outer specifier of vP (see Chomsky 1995, 2000, 2001 on multiple specifiers). The question then arises as to why Wackernagel pronouns must precede all other clause-internal material (except for subjects in SpecT). The answer is that the trigger for Wackernagel movement to Specv is different from the trigger for scrambling. I would like to suggest that Wackernagel movement is not feature-driven and targets a specific domain that is defined in both hierarchical and linear terms: the left edge of vP. In contrast, scrambling can be assumed to be triggered either by certain kinds of formal features (see Grewendorf and Sabel 1999 and Sauerland 1999, among others),

Harmonic alignment and the hierarchy of pronouns in German

or by linearization requirements that concern the relative order of items (e.g., [+def] precedes [–def], [+animate] precedes [–animate]), not any fixed position (like the left edge) (see e.g. Choi 1999 and Büring 2001).5 The analysis relies on harmonic alignment of the two scales in (1). (11) a.

Position Scale: vPi(nternal) > vPe(xternal) b. Personal Pronoun Scale: Prons(trong) > Pronu(nstressed) > Pronw(eak) > Pronr(educed)

The position scale is binary. It distinguishes between vP-internal positions and vP-external positions and thereby reflects a fundamental dichotomy in syntax: Given the predicate-internal subject hypothesis, vP is the minimal clausal domain containing the verb and its arguments; all higher clausal projections are functional. A crucial assumption now is that there is exactly one type of position that is dominated by a vP node but does not count as a vP-internal position, viz., the edge of vP, understood here as a left-peripheral specifier position of v:6 (12) Edge of vP: a. A category α is at the edge of vP if (i) α is a specifier of v, and (ii) there is no β preceding α that is dominated by (a segment of) vP. b. If vP dominates α, α is vP-external if it is at the edge of vP, and vPinternal otherwise.

The personal pronoun scale reflects the relative strength of pronouns and has already been discussed above (see (1)).7 Harmonic alignment creates two harmony scales HvPi and HvPe. HvPi states that to show up vP-internally is more marked for reduced pronouns than for weak pronouns, more marked for weak pronouns than for unstressed pronouns, and more marked for unstressed pronouns than for strong pronouns; HvPe states the opposite for personal pronouns in vP-external positions. (13) Harmonic alignment: a. HvPi: vPi/Prons vPi/Pronu vPi/Pronw vPi/Pronr b. HvPe: vPe/Pronr vPe/Pronw vPe/Pronu vPe/Prons

Finally, constraint alignment produces the constraint hierarchies in (14). (14) Constraint alignment: a. CvPi: *vPi/Pronr *vPi/Pronw *vPi/Pronu *vPi/Prons b. CvPe: *vPe/Prons *vPe/Pronu *vPe/Pronw *vPe/Pronr

211

212 Gereon Müller

As in the account of Prince and Smolensky (1993), the analysis is based on one of the two constraint hierarchies thus derived: CvPi. Since the constraints of this hierarchy will turn out to act as triggers for Wackernagel movement, I will also refer to them as Wackernagel constraints.8 The constraints of CvPi all take the form “Avoid a personal pronoun of a certain type in a vP-internal position” and thereby force personal pronouns to leave the vP-internal domain and show up somewhere else — either at the edge of vP, or in some other position. The weaker the pronoun is, the higher-ranked is the constraint that triggers displacement. The nature of this displacement operation must be movement rather than base-generation if we assume that all personal pronouns are generated in a vP-internal position.9 The constraints in CvPi trigger Wackernagel movement of all kinds of pronouns; however, strong pronouns do, in fact, not undergo Wackernagel movement. Hence, there must be a conflicting constraint that prohibits such movement. In the approach adopted here, pronoun movement to the edge of vP differs from other, well-established instances of feature-driven movement (like, e.g. wh-movement). Wackernagel movement is triggered by a conspiracy of constraints, rather than by a regular feature. In Chomsky (2001), such repairdriven movement violates a ban on the assignment of EPP-features to v. I will assume here (following Heck and Müller 2000b) that it violates the Last Resort constraint in (15) (see Chomsky 1995). (15) Last Resort (LR): Movement must be feature-driven.

The ranking in German must then be (16): (16) Ranking in German: *vPi/Pronr *vPi/Pronw

*vPi/Pronu

LR

*vPi/Prons

Based on these assumptions, we can now derive that strong pronouns do not undergo Wackernagel movement in German, whereas unstressed, weak, and reduced pronouns do. For the time being, let us focus on the vP domain and ignore all higher structure (TP, CP, matrix clause). The ranking LR *vPi/Prons ensures that with a strong pronoun in the input, the optimal output has the pronoun in the vP-internal domain; see Tableau 1, where output O2 is optimal. In contrast, the rankings *vPi/Pronu LR, *vPi/Pronw LR, and *vPi/Pronr LR trigger Wackernagel movement of unstressed, weak, and reduced pronouns to the left edge of vP. This is shown for unstressed pronouns in Tableau 2.


Tableau 1.vP optimization: strong pronouns & Wackernagel movement Input: ihr, ...

*vPi/ Pronr

*vPi/ Pronw

*vPi/ Pronu

O1: [vP ihr1 .. t1 .. V ]

LR

*vPi/ Prons

*!

O2: [vP .. ihr1 .. V ]

*

Tableau 2.vP optimization: unstressed pronouns & Wackernagel movement Input: ihr[+anim], ...

*vPi/ Pronr

*vPi/ Pronw

*vPi/ Pronu

O1: [vP ihr1 .. t1 .. V ] O2: [vP .. ihr1 .. V ]

LR

*vPi/ Prons

* *!

With weak and reduced pronouns, the only difference to Tableau 2 is that the in situ candidate O2’s fatal violation is one of *vPi/Pronw or *vPi/Pronr. At this point, an obvious question needs to be addressed: The constraints of the CvPi hierarchy force movement of pronouns out of the vP-internal domain, but they do not explicitly force movement to the edge of vP. In principle, these constraints could also be fulfilled by moving the pronoun to some higher position, e.g. to a specifier position in the TP or CP domains. Given that Wackernagel movement can in fact not target these positions, this option must be excluded. To this end, I will adopt a version of the system of local optimization that is developed in Heck and Müller (2000a,b). Under this view, syntactic structure is created derivationally, by iterated applications of Merge and Move (cf. Chomsky 1995). Each subpart of the derivation from one cyclic node to the next cyclic node is subject to input/output optimization. An optimal subderivation is one that respects all inviolable constraints (e.g. the Strict Cycle Condition), and that also best satisfies an ordered set of violable constraints. The optimal subderivation of one cycle then serves as (part of) the input for the next cycle, and so on, until the root is reached and the sentence is complete. The question arises as to what counts as a cyclic node. Following Chomsky (1995), Heck and Müller (2000a,b) postulate that all XPs are cyclic. In contrast, I assume here that only phases in the sense of Chomsky (2000, 2001), i.e. vP and CP, count as cyclic nodes that are subject to optimization.10 This is in line with Fanselow and Cávar (2001).

213

214 Gereon Müller

The local optimization procedure that is relevant in the present context concerns vP. Given local optimization, it is clear why *vPi/Pron can only be fulfilled by movement to the left edge of vP: At the stage of the derivation where vP optimization takes place, there is not yet any structure present above vP, and hence, no potential alternative landing site for Wackernagel movement. Of course, a subsequent optimization procedure (affecting CP) may require further pronoun movement (e.g. wh-movement; see below), but such an operation must then be triggered by a different constraint (most notably, a Feature Condition that demands checking of features by movement), and does not qualify as Wackernagel movement anymore: Once a *vPi/Pron constraint is fulfilled by movement to the left edge of vP, any further pronoun movement that is not triggered by some other constraint or feature will fatally violate LR. Similarly, if LR blocks Wackernagel movement to the left edge of vP (as with strong pronouns), it will also block Wackernagel movement to other positions in subsequent cycles. Note that, other things being equal, this result could not easily be obtained in a standard (global) optimization approach in which complete sentences are evaluated in one step (see Grimshaw 1997 and the vast majority of work in optimality-theoretic syntax). In such an approach, there would be no a priori reason why Wackernagel movement driven by *vPi/Pron targets the left edge of vP, and not some other position higher up in the tree; consequently, additional constraints would be necessary to predict the correct outcome. With Wackernagel movement now accounted for, I turn to R-pronoun formation.11

4. R-Pronoun formation 4.1 Empirical evidence for R-Pronoun formation Within German PPs, personal pronouns are sometimes replaced by R-pronouns with an adverbial origin ([–wh] da and [+wh] wo); whereas regular pronouns show up to the right of P, an R-pronoun attaches to the left of P in this construction.12 This process of R-pronoun formation is impossible with strong pronouns and unstressed pronouns, optional with weak pronouns, and obligatory with reduced pronouns.13 The ban on R-pronoun formation with strong and unstressed pronouns is illustrated in (17) and (18), respectively.


(17) a.

Ich habe gestern [PP mit ihr] telefoniert I have yesterday with her phoned b. *Ich habe gestern [PP da-mit] telefoniert I have yesterday there[+stress]-with phoned

(18) a.

Ich habe gestern [PP mit ihr1] telefoniert I have yesterday with her phoned b. *Ich habe gestern [PP da1-mit] telefoniert I have yesterday there[+anim]-with phoned

(19) shows that R-pronoun formation is optional with weak pronouns. (The pronoun here stands for an inanimate feminine NP like die Ausstellung (‘the exhibition’).) (19) a.

Maria hat noch oft [PP an sie1] gedacht Maria has ptcl often of her thought gedacht b. Maria hat noch oft [PP da1-r-an] Maria has ptcl often there[−anim]-of thought

Finally, R-pronoun formation is obligatory in the case of reduced pronouns: (20) a. *Maria hat noch oft [PP an es1] gedacht Maria has ptcl often of it thought b. Maria hat noch oft [PP da1-r-an] gedacht Maria has ptcl often there[+red]-of thought

The different behaviour of reduced, weak and unstressed/strong pronouns with respect to R-pronoun formation clearly shows that at least these three classes of pronouns must be recognized in the syntax of German; given the dividing line between reduced/weak/unstressed and strong pronouns with respect to Wackernagel movement, all four pronoun classes described in terms of features in (1) are now syntactically motivated. In what follows, I will derive the distribution of R-pronoun formation from the relative ranking of the CvPi Wackernagel constraints and two other constraints. 4.2 Analysis of R-Pronoun formation The basic observation is that PP-internal personal pronouns obey *vPi/Pron just like other personal pronouns. Therefore, we expect PP-internal unstressed, weak, and reduced pronouns to undergo Wackernagel movement. However, there is a problem for pronouns in this position: Personal pronouns are Casemarked, and PP is a barrier for extraction of Case-marked NPs in German.

215

216 Gereon Müller

Only Case-less R-pronouns can be moved out of PP (“postposition stranding”). Compare (21a) with (21b).14 (21) a. *Maria hat [vP sie1 noch oft [PP an t1] gedacht] Maria has her ptcl often of thought b. Maria hat [vP da1 noch oft [PP t1 an] gedacht] Maria has there ptcl often of thought

The locality constraint in question can be formulated as in (21); I assume that it is undominated among the constraints discussed in the present paper:15 (22) PP-Loc (“PP-Locality”): NPs that receive Case from P must not move out of PP.

Thus, R-pronouns can move out of PP, but they do not have to: They do not obey *vPi/Pron in the first place because they are adverbial pronouns, not personal pronouns. Hence, the R-pronoun in (21b) is not Wackernagel-moved (triggered by *vPi/Pron), but scrambled (triggered by some other factor). This view is reinforced by the fact that R-pronouns can show up in non-edge positions of vP (see (23a)) and cannot show up in front of any Wackernagel pronoun (see (23b)). (23) a.

Maria hat [vP mich gestern da1 nicht [PP t1 mit] beeindruckt] Maria has me.acc yesterday there not with impressed b. *Maria hat [vP da1 mich gestern nicht [PP t1 mit] beeindruckt] Maria has there me.acc yesterday not with impressed

Thus, in PP-internal contexts, the situation arises that unstressed, weak, and reduced personal pronouns must undergo Wackernagel movement because of *vPi/Pron; however, they can never do so because of a higher-ranked PP-Loc.16 Suppose now that we replace the selected personal pronoun with an unselected, adverbial R-pronoun, i.e. apply R-pronoun formation. This way, *vPi/Pron and PP-Loc can both be fulfilled. However, the cost is a violation of a faithfulness constraint Faith(Sel) that requires lexical selection to be realized in the output: (24) Faith(Sel): Selected material of the input must be realized in the output.

I would like to contend that the relative ranking of Faith(Sel) and the *vPi/ Pron constraints of the CvPi hierarchy derives the impossibility, optionality, and obligatoriness of R-pronoun formation with the various pronoun classes. More specifically, suppose that the ranking in (16) above is extended as in (25).


(25) Extended ranking in German: PP-Loc *vPi/Pronr *vPi/Pronw | Faith(Sel) *vPi/Prons

*vPi/Pronu

LR

As we have seen, *vPi/Pronu and *vPi/Prons differ in their ranking with respect to LR, but they do not differ in their ranking with respect to Faith(Sel). Given the order Faith(Sel) *vPi/Pronu, *vPi/Prons (and an undominated PP-Loc), strong and unstressed pronouns must remain in situ, within PP, in violation of the respective Wackernagel constraint. This is shown for unstressed pronouns in Tableau 3.17 Tableau 3.Unstressed pronouns and R-pronouns Input: [PP P ihr[+anim]], ..

PPLoc

*vPi/ Pronr

*vPi/ Faith *vPi/ Pronw (Sel) Pronu

O1: [vP .. [PP P ihr ]..] O2: [vP ihr .. [PP P t ]..]

LR

*vPi/ Prons

* *!

*

O3: [vP .. [PP da-P ]..]

*!

Next, *vPi/Pronw and Faith(Sel) are tied, i.e. equally ranked. This derives the optionality of R-pronoun formation with weak pronouns in PP-internal contexts. If the tie is resolved into the ranking *vPi/Pronw Faith(Sel), the optimal vP output has R-pronoun formation; if the tie is resolved into the ranking Faith(Sel) *vPi/Pronw, the optimal vP output retains the personal pronoun within PP. This is shown in Tableau 4. Tableau 4.Weak pronouns and R-pronouns Input: [PP P sie[–anim]], ...

PPLoc

O1: [vP .. [PP P sie ]..] O2: [vP sie .. [PP P t ]..] O3: [vP .. [PP da-P ]..]

*vPi/ Pronr


LR

*vPi/ Prons

*(!) *!

* *(!)

Finally, *vPi/Pronr outranks Faith(Sel). Hence, a pronoun es is obligatorily replaced with the R-pronoun da in PP-internal contexts; cf. Tableau 5.

218 Gereon Müller

Tableau 5.Reduced pronouns and R-pronouns Input: [PP P es], ...

PPLoc

O1: [vP .. [PP P es ]..] O2: [vP es .. [PP P t ]..] O3: [vP .. [PP da-P ]..]

*vPi/ Pronr


LR

*vPi/ Prons

*! *!

* *

This is the gist of the present account of R-pronoun formation in German.18 However, certain other issues remain to be clarified. A first question is why it is the adverbial pronoun da that is chosen as the non-selected “repair” or “last resort” form that vacuously fulfills the *vPi/Pron constraints of the Wackernagel hierarchy. I think the key to a solution is that da is the adverbial pronoun with the least specific semantics (an “Allerweltsproform”, as Altmann 1981: 241 puts it). Generality of meaning is characteristic of repair forms of this type; compare e.g. the verb do that is used in English do-support constructions, or the wh-item corresponding to English what that is used in wh-scope marking constructions in languages like German, Hindi, and Hungarian. With regard to do-support, Grimshaw (1997) argues that if an item that is not part of the input must be inserted in the optimal candidate, and a faithfulness violation is thus unavoidable, this violation must at least be kept minimal, and the relevant faithfulness constraint in her analysis is designed in such a way that items with more specific semantics will invariably violate it to a higher degree than items with less specific semantics. The minimal violation is incurred by the item with the least specific semantics that is appropriate in a given syntactic context; and in the case of verbs in English, this happens to be do. I assume that something similar can be said for the R-pronoun da in German. Second, it is not a priori clear why the non-selected R-pronoun da shows up to the left of P, even though a selected NP must normally show up to the right of P. One possible solution would be to assume that da must form a complex word with P if it is adjacent to it. Complex word formation would then result in the reverse order, either because the head must be right-peripheral in morphology (and the head of the complex word is P), or because da would explicitly be marked as proclitic. However, more would have to be said given the contrast between ab da (“from there/then”) and *da-r-ab (“from there/then”), where da is a selected adverbial that is adjacent to the preposition ab and still shows up to the right of it, like selected NPs and unlike non-selected da. Another solution


that evades this problem would be to assume a constraint outranking LR that demands that non-selected items must not show up in complement positions (this is essentially one part of Chomsky’s 1981 Projection Principle). Under this view, a non-selected da would have to move to a non-complement position, and the nearest such position would be a specifier to the left of P. For now, I will leave it at that, assuming that the linearization effect with R-pronoun formation can be accounted for in one of the ways envisaged here. Third and finally, the question arises of whether the same kind of analysis can be given for the [+wh] pronoun wo that optionally replaces the weak NP pronoun was within PPs: (26) a.

[PP An was]2 hast du t2 gedacht? of what have you thought b. [PP Wo-r-an]2 hast du t2 gedacht? where-of have you thought

Given that was is a weak pronoun that obeys *vPi/Pronw, the analysis is indeed identical to that of inanimate sie in Tableau 4: During vP optimization, the NP pronoun was must move to the left edge of vP but cannot do so because of a higher-ranked PP-Loc; the violation of *vPi/Pronw that results from leaving was within PP, and the violation of Faith(Sel) that results from replacing the NP pronoun with an R-pronoun wo are equally costly, and optionality arises. By this reasoning, we expect that the optimal derivation underlying a sentence like (27) involves Wackernagel movement of was (“what”) to the left edge of vP during vP optimization, and subsequent wh-movement to SpecC during CP optimization.19 (27) Was1 hat tÁ1 Fritz gestern t1 gemacht? what has Fritz yesterday done

5. Coordination Cardinaletti and Starke (1996) point out that (what I call) weak and reduced pronouns (their “weak pronouns”) cannot be coordinated in German, whereas unstressed and strong pronouns (their “strong pronouns”) can be. This is shown by the data in (28).

220 Gereon Müller

(28) a.

Maria hat [NP ihn [α und seinen Freund]] zur Krippe gebracht Maria has him and his friend to.the day.care taken b. Maria hat [NP ihn [α und seinen Freund]] zur Krippe Maria has him[+anim] and his friend to.the day.care gebracht taken [α und das Fahrrad]] zur c. *Maria hat [NP ihn Maria has him[−anim] and the bicycle to the Werkstatt gebracht garage taken d. *Maria hat [NP es [α und die Zeitung]] gekauft Maria has it and the newspaper bought

Particularly relevant here is the contrast between (28b) and (28c). If ihn is interpreted as animate (e.g. as referring to a young boy), coordination is possible; if ihn is interpreted as inanimate (e.g. as referring to a scooter), coordination is impossible. The analysis I would like to suggest is similar to that of R-pronoun formation. As we have seen, constraints of the type *vPi/Pron trigger Wackernagel movement. However, the Coordinate Structure Constraint (Csc) blocks extraction of (and out of) conjuncts (see Ross 1967). (29) Csc (“Coordinate Structure Constraint”): No item can undergo leftward movement out of a coordinate structure.

Given that the Csc is ranked high, like PP-Loc, the pronouns in (28) cannot be extracted out of the coordinate NP in order to move to the Wackernagel position.20 Thus, the situation is comparable to that found with PP-internal contexts. However, this time, R-pronoun formation does not apply; compare (30) with (28c). (30) *Maria hat [NP da(-r-) und das Fahrrad] zur Werkstatt gebracht Maria has him[−anim] and the bicycle to the garage taken

This means that R-pronouns in coordinate structures violate some higherranked constraint (various possibilities come to mind, one being the wellknown parallelism constraint on coordination, another one a lexical property of und (‘and’) that precludes R-pronoun formation). What, then, is the optimal candidate that blocks (28c) and (28d)? This is an instance of a standard question arising in optimality theory: How is absolute ungrammaticality (ineffability) derived? For the sake of concreteness, I will assume that the optimal candidate that blocks (28c), (28d) is one in which α moves out of the coordinate


structure by extraposition, i.e., rightward movement. This violates LR and the faithfulness constraint in (31), but not the Csc. (31) Faith(Lex): Lexical material in the output must be present in the input.

This constraint belongs to the class of so-called Dep (dependency) constraints which prohibit epenthesis in phonology. In the case at hand, Faith(Lex) prohibits the insertion of non-input auch (‘too’) which is, however, required to support the extraposed category α. The candidate that blocks (28c) is (32); α moves rightward and thereby destroys the coordinate structure, which enables ihn to undergo Wackernagel movement in accordance with the Csc. (32) Maria hat [vP ihn1 t1 tα zur Werkstatt gebracht] und [α das Maria has him[−anim] to the garage taken and the Fahrrad *(auch)] bicycle *(too

Faith(Lex) must be ranked lower than *vPi/Pronr,w, but higher than *vPi/ Pronu,s. The following two tableaux illustrate the two representative cases in (28b) and (28c). On the one hand, Tableau 6 explains why unstressed (hence, strong) pronouns can be coordinated. The optimal candidate violates *vPi/ Pronu in order to respect the higher-ranked Faith(Lex). Tableau 6.vP optimization: unstressed pronouns and coordination Input: [NP ihn[+anim] & NP], ..

Csc

*vPi/ Pronr

*vPi/ Faith *vPi/ Pronw (L) Pronu

O1: [vP .. [NP ihn α]..] O2: [vP ihn..[NP t α]..] O3: [vP ihn.. t..tα..] .. α auch

LR

*vPi/ Prons

* *!

* *!

**

On the other hand, Tableau 7 shows why weak (hence, reduced) pronouns cannot undergo coordination. The optimal candidate violates Faith(Lex) to respect the higher-ranked *vPi/Pronw by applying Wackernagel movement.21

222 Gereon Müller

Tableau 7.vP optimization: weak pronouns and coordination Input: [NP ihn[–anim] & NP], ..

Csc

*vPi/ Pronr

O1: [vP .. [NP ihn α]..] O2: [vP ihn..[NP t α]..]

*vPi/ Faith *vPi/ Pronw (L) Pronu

LR

*vPi/ Prons

*! *!

O3: [vP ihn.. t..tα..] .. α auch

* *

**

6. Topicalization Cardinaletti and Starke (1996) observe that the dividing line among pronouns with respect to coordination also exists in the domain of topicalization. Topicalization of unstressed and strong pronouns is possible; see (33a,b). In contrast, topicalization of weak and reduced (object) pronouns is impossible; see (33c,d) (on subject pronouns, see below). (33) a.

Ihn1 hat Maria t1 geküßt him has Maria kissed b. Ihn1 hat Maria t1 geküßt him has Maria kissed c. *Ihn1 hat Maria t1 repariert him has Maria fixed d. *Es1 habe ich t1 gelesen it have I read

Again, a striking contrast arises between (33b), where ihn is interpreted as animate (e.g. as referring to Fritz), and (33c), where ihn is interpreted as inanimate (e.g. as referring to a car). The data lend themselves to an analysis that is similar to the one developed for R-pronoun formation and coordination. The role of PP-Loc and Csc is played by Top-Edge, and the role of Faith(Sel) and Faith(Lex) by Faith(Top). (34) a.

Top-Edge: Topics must not be at the edge of vP. b. Faith(Top): A topic feature in the input must be present in the output.

Suppose that items are marked as topics by a [+top] feature in the input; that Top-Edge is ranked higher than all CvPi constraints; and that Faith(Top)


intervenes between *vPi/Pronw and *vPi/Pronu. Tableau 8 then shows that with a [+top]-marked unstressed (or strong) pronoun in the input, the optimal vP is not one in which the pronoun undergoes Wackernagel movement — either as a [+top] pronoun (which would fatally violate Top-Edge), or as a pronoun from which the [+top] feature has been removed (which would fatally violate Faith(Top)). Rather, the optimal vP output here is one in which the [+top] pronoun stays in situ, in violation of *vPi/Pronu. Tableau 8.vP optimization: unstressed pronouns & topicalization Input: ihn[+anim],[+top], ...

TopEdge

*vPi/ Pronr

*vPi/ Faith *vPi/ Pronw (Top) Pronu

O1: [vP .. ihn1[+top] .. V] O2: [vP ihn1[+top] .. t1 .. V]

LR

*vPi/ Prons

* *!

*

O3: [vP ihn1[+top] .. t1 .. V]

*!

*

If, on the other hand, the [+top]-marked pronoun in the input is weak (or reduced), the optimal vP output will involve deletion of the feature [+top] on the pronoun and movement to the edge of vP. This is shown in Tableau 9.22 Tableau 9.vP optimization: weak pronouns & topicalization Input: ihn[–anim],[+top], ...

TopEdge

O1: [vP .. ihn1[+top] .. V] O2: [vP ihn1[+top] .. t1 .. V] O3: [vP ihn1[+top] .. t1 .. V]

*vPi/ Pronr

*vPi/ Faith *vPi/ Pronw (Top) Pronu

LR

*vPi/ Prons

*! *!

* *

*

Thus far, Tableau 8 and Tableau 9 show that the optimal vP is different with unstressed/strong vs. weak/reduced pronouns. However, this, per se, does not yet tell us why topicalization is possible with the former class and impossible with the latter. Here, one additional assumption is necessary: Suppose that topicalization is regular, feature-driven movement that is triggered by a general constraint Feature Condition (FC) which implies that a [+top]-marked item must move to the specifier of a C head that has a matching [+top] feature (a C head with a [+top] feature also triggers V/2 movement in German). FC out-

224 Gereon Müller

ranks LR in German and becomes active in the CP cycle (see Vikner 2001). Now, an account of the topicalization asymmetry is straightforward: vP optimization maintains the [+top] feature of an unstressed (or strong) pronoun; hence, subsequent CP optimization can fulfill FC only by topicalizing the pronoun (assuming, contra Chomsky (2000, 2001), that movement to SpecC is possible from a VP-internal position). In contrast, vP optimization removes the [+top] feature of a weak (or reduced) pronoun; consequently, subsequent CP optimization can fulfill FC vacuously without topicalizing the pronoun, and such topicalization will be blocked by the low-ranked constraint LR. In a nutshell, the earlier vP optimization procedure has turned a topic pronoun into a non-topic pronoun (as yet another instance of input neutralization), and pronoun topicalization is thus blocked.23 Of course, there is again much more to be said about the construction at hand, and many issues have to be left unresolved. Let me just briefly address two issues that are potentially problematic for the approach developed here. First, Gärtner and Steinbach (2000: 35f) contend that topicalization of (what I call) weak and reduced pronouns should not be excluded syntactically, and they try to substantiate this claim by providing examples (partly from dialectal varieties of German) that are analogous to (33c) and (33d), but seem wellformed. Interestingly, though, all their examples either involve genuine phonological clitics, or can be argued to involve reductions of the demonstrative pronoun das rather than weak personal pronouns es. I conclude from this that the generalization that weak and reduced object pronouns cannot undergo topicalization is not undermined by the empirical evidence presented by Gärtner and Steinbach.24 Second, it has often been observed that weak and reduced subject pronouns can show up in what looks like a topic position in German: (35) Es hat mich beeindruckt. it.nom has me.acc impressed

Travis’s (1984) conclusion is that subjects can occur in a SpecV/2 position without being topicalized, due to a left-periphal T node in German. This would in principle be compatible with the analysis developed here. Alternatively, one might argue that subject pronouns do in fact not undergo Wackernagel movement to the left edge of vP but stay in situ, raising to SpecT in the following CP cycle. If that is so, all subject pronouns can survive vP optimization with a [+top] feature, and can therefore be NP-raised and topicalized during CP optimization. I will leave this question open.


7. Conclusion I have shown that by applying harmonic alignment to two simple hierarchies involving pronouns and positions, a set of *vPi/Pron constraints can be derived. These constraints trigger displacement of pronouns, and they have a fixed internal order. In interaction with the economy constraint LR, these constraints directly account for the distribution of Wackernagel effects in German. The Wackernagel constraints are dominated by certain constraints on movement (PP-Loc, Csc, Top-Edge), and they are interspersed with certain faithfulness constraints (Faith(Sel), Faith(Lex), Faith(Top)); this way, they indirectly also account for the distribution of R-pronoun formation, and coordination and topicalization of pronouns in German. Because of the fixed order resulting from harmonic alignment, implicational generalizations that hold in these domains (modulo intervening factors like, e.g. additional morphophonological constraints in the case of clitic pronouns) can be captured: If a given pronoun has property P, then all weaker pronouns also exhibit P (where P stands for properties like “undergoes Wackernagel movement”, “permits/forces R-pronoun formation”, “cannot undergo coordination”, and “cannot undergo topicalization”). We have seen that there are three distinct dividing lines, which necessitate four pronoun classes. First, in the case of Wackernagel movement, there is a dividing line between strong pronouns on the one hand and unstressed, weak, and reduced pronouns on the other. Second, R-pronoun formation involves two dividing lines: one between strong or unstressed pronouns and weak pronouns, and another one between weak pronouns and reduced pronouns. This already suffices to motivate the four pronoun classes. Third, in the domains of coordination and topicalization, there is a dividing line between strong and unstressed pronouns on the one hand, and weak and reduced pronouns on the other.25 Since the account of Rpronoun formation, coordination, and topicalization is dependent on the account of Wackernagel movement, and since the pronoun classes have been shown to behave in a non-uniform way with respect to these empirical domains, the present approach crucially presupposes that *vPi/Pron constraints are violable and ranked.26 Thus, the analysis supports an optimalitytheoretic approach; more specifically, one that is local (rather than global), and that relies on harmonic alignment. Local optimization can be shown to significantly reduce the size of candidate sets (see Heck and Müller 2000b), and harmonic alignment clearly reduces the factorial typology that results from free re-ranking of constraints. Hence, the present approach can also be viewed as

226 Gereon Müller

part of the more general enterprise to tackle the problem of complexity in optimality-theoretic syntax.

Notes * For comments and discussion, I would like to thank Judith Aissen, Marcel den Dikken, Fabian Heck, Winfried Lechner, Peter Sells, Wolfgang Sternefeld, Sten Vikner, Ralf Vogel, two reviewers for this volume, the editors, and audiences at the University of California at Santa Cruz, Universität Tübingen, IDS Mannheim, and Universität Marburg (Pronomina AG of the DGfS meeting). This research was supported by DFG grants MU 1444/1–2,2–2. 1. In what follows, knowledge of the basic concepts of optimality-theoretic syntax is presupposed; introductory texts include Legendre (2001) and Müller (2000b). 2. Throughout this paper, I assume the clausal architecture developed in Chomsky (1995, 1998, 1999): CP is headed by C, which selects TP (with SpecT the canonical subject position); T selects vP (where v is a light verb and external arguments are base-generated in Specv); and v selects VP (which contains the main verb and internal arguments). 3. See, e.g. Lenerz (1977, 1992), Thiersch (1978), Haftka (1981), Hoberg (1997), and Haider and Rosengren (1998). 4. As noted above, I assume that subjects are base-generated in a specifier of a light verb v that in turn embeds VP. — Note that the assumption of optional subject raising to SpecT makes a unified approach to German Wackernagel movement and Scandinavian object shift possible (see Thráinsson 2001 for an overview). On this view, the two operations target the same position, the main structural difference being the optionality vs. obligatoriness of subject raising. Needless to say, though, a complete identification of the two operations ultimately demands an account of further (apparent) differences, concerning, e.g. the A- vs. A-bar nature of pronoun movement (thus, Wackernagel movement behaves like whmovement with respect to the typical A-bar property of licensing parastic gaps, whereas object shift behaves like NP-raising in this respect), and the dependence on V movement (object shift can only take place if the main verb has moved out of the vP, a restriction that does not seem to hold for Wackernagel movement). 5. The fact that different triggers are involved can also be held responsible for certain asymmetries between Wackernagel movement and scrambling in the domain of remnant categories; see Müller (1998: Ch. 5). 6. This approach can straightforwardly be generalized from vP to the more general concept of “phase” in Chomsky (2000, 2001); that approach also employs the notion of “edge.” Note, however, that the notion of edge adopted here is closer to the notion of “phonological border” in Chomsky (2001), and to the notion of “edge” in phonology, than to the notion of “edge” in Chomsky (2001). 7. Do these simple scales directly reflect relative markedness, as is the case with the harmony scales below? This seems intuitively appealing, and at least for (a version of) (11b), it has been proposed by Cardinaletti and Starke (1996). However, Müller, Tiedemann and Schmitz


(2000) show that evidence from language acquisition does not support this hypothesis. Furthermore, as noted by a reviewer, it is by no means obvious from a theory-external point of view why the position scale should have to have the order in (11a); e.g. the reverse order vPe > vPi could be motivated on the basis of the canonical c-command relations in trees. Thus, I conclude that the scales in (11) are primarily motivated on an empirical basis by data of the type discussed in the present paper; and I will leave open the issue of whether they can ultimately be justified externally. 8. As for CvPe, for present purposes we can assume that the whole hierarchy is ranked low in German and therefore does not normally produce effects. That said, CvPe as a whole could actually take over most of the tasks of the economy constraint Last Resort that is introduced below — it blocks all instances of pronoun displacement except for those that are vP-internal. 9. This is generally presupposed; cf. the standard assumption that θ-role assignment obeys strict locality (see Chomsky 1981). In the present system, it can be derived from higherranked constraints that regulate the mapping from θ-grids to grammatical functions and that require arguments of the verb to show up in a fixed hierarchical order within vP. 10. I do so mainly for expository purposes. However, under the alternative definition of cyclic nodes, the approach to R-pronoun formation in Section 4 below would have demanded a slightly more intricate version of the Strict Cycle Condition than is standard, viz., one that permits an operation to exclusively affect PP in the structure [vP…[ PP…]…]. 11. The question arises of how multiple Wackernagel movement as in (i-a) fits into the approach. (i) a.

daß es1 ihr2 der Fritz gestern t1 t2 gegeben hat that it.acc her.dat art Fritz.nom yesterday given has b. *daß ihr2 der Fritz gestern es1 t2 gegeben hat that her.dat art Fritz.nom yesterday it.acc given has c. *daß ihr2 es1 der Fritz gestern t1 t2 gegeben hat that her.dat it.acc art Fritz.nom yesterday given has

First, (i-b) shows that unstressed, weak, and reduced pronouns all have to move in multiple pronoun constructions. Strictly speaking, they cannot all show up at the left edge of vP, though. Still, *vPi/Pron successfully triggers multiple Wackernagel movement if it is interpreted as gradient — the consequence then is that a pronoun moves as close to the edge as possible. Alternatively, (12a) can be revised in such a way that pronouns as phonologically light items do not count as interveners: A category α is at the edge of vP if (i) α is a specifier of v, and (ii) there is no non-pronoun β preceding α that is dominated by (a segment of) vP. Second, (i-c) indicates that the pronouns show up in a fixed order in the Wackernagel domain. It may be that this phenomenon shows the activitiy of a low-ranked faithfulness constraint demanding order preservation that also derives superiority effects; see Williams (1999), Müller (2001). Alternatively, fixed pronoun order might follow from phonological linearization constraints concerning (a) pronoun length, and (b) quantity and quality of the onset, nucleus, and coda of the syllables that are involved; see Wegener (1985), Hoberg (1997), and Zifonun (2000). 12. R-pronouns owe their name to the fact that an epenthetic r is inserted if the onset of the preposition starts with a vowel; compare da-r-an (“there-[r]-of”) with da-mit (“there-[]-with”).

228 Gereon Müller

13. See Helbig (1974), van Riemsdijk (1978), Wunderlich (1984), van Riemsdijk and Williams (1986), Breindl (1989), Trissler (1993). Certain exceptions, intervening factors, and instances of synchronic and diachronic variation that blur this simple picture are addressed in Müller (2000a). The approach to R-pronoun formation to be developed below is a radical reworking of the analysis in that paper, which relies on a specific functional projection for Wackernagel pronouns and involves neither harmonic alignment nor local optimization. 14. See, e.g. Ross (1967), van Riemsdijk (1978), Kayne (1981), Fanselow (1983, 1991), Grewendorf (1989), Trissler (1993), Bayer (1994). 15. PP-Loc can arguably be derived from the interaction of more general constraints. I will not pursue this matter here. 16. One might think that *vPi/Pron and PP-Loc could both be respected by PP pied-piping, as in the analogous wh-movement case in (i-a). [PP An [NP welche Frau]]2 hat Maria gestern t2 gedacht? of which woman has Maria yesterday thought b. *Maria hat [PP an [NP es1]]2 noch oft t2 gedacht Maria has of it ptcl often thought

(i) a.

(i-b) shows that this is not the case: PP pied-piping is not a way to fulfill *vPi/Pron. Possibly, this difference is related to the fact that wh-movement targets a position that is structurally defined (SpecC), whereas Wackernagel movement targets a position that is both linearly and structurally defined: Given (12), es in (i-b) is not in a vP-external position. (More generally, it is already observed in Ross 1967 that movement types may differ with respect to piedpiping options.) 17. Again, I confine myself to the relevant local optimization procedure of the complete derivation, viz., that affecting vP. Note that the variable shape of PP documented in O1, O2 vs. O3 supports the conclusion that PP itself is not yet an optimization domain; as remarked in footnote 10, a modification of the standard versions of the Strict Cycle Condition would be called for so as to permit backtracking if it were. 18. R-pronoun formation is not confined to arguments; it also takes place in adverbs. The account developed here thus presupposes that all adverbs are generated within vP. This conclusion does not strike me as particularly problematic as long as it is ensured that vP/VP offers a sufficient number of adverb and argument positions (which it does, given multiple specifiers), and that the c-command relations of different types of adverbs with respect to arguments and to each other can be accounted for without resort to designated functional projections related to adverb classes. 19. At least at first sight, multiple questions seem to pose a problem for this approach because a clause-internal was can in fact not occupy the left edge of vP, but must stay in situ, within vP: (i) Wann hat (?*was1) der Fritz gestern (was1) gemacht? when has (?*what art Fritz yesterday (what done However, there is an intervening factor. In general, a wh-in situ item in a multiple question in German must be stressed. If this is so, the ban on Wackernagel movement of was in (i) is accounted for.


20. This shows that the edge of vP cannot be defined in a purely linear way: Even scrambling of the coordinate NP to a Specv position does not fulfill *vPi/Pron. Under present assumptions, this follows because the pronoun in the coordinate NP is not moved to Specv in this case; cf. (12). 21. Note that O3 in Tableau 6 and O3 in Tableau 7 (with an unstressed and weak pronoun, respectively) are both optimal candidates in other competitions in which they do not violate Faith(Lex) or LR because a feature is present that triggers extraposition of α, and the additional lexical item auch is already part of the input. Thus, what happens in Tableau 7 is a neutralization of input specification: Regardless of whether features and lexical material that trigger α extraposition are present in the input in this context, α extraposition will take place in the output. See Legendre, Smolensky, and Wilson (1998), Bakovic´ and Keer (2001) and Vogel (2001), among others, for further applications of input neutralization as a means to account for ineffability. 22. In both competitions, there is another candidate O4 that deletes the [+top] feature and leaves the pronoun in situ. Since this output violates *vPi/Pron and Faith(Top), its constraint profile is always worse than that of O1; O4 is “harmonically bounded” by O1 (see Prince and Smolensky 1993). 23. A reviewer notes that examples like (33b) are not acceptable for some speakers. This can straightforwardly be accounted for by assuming microparametric variation in the form of a reranking of Faith(Top) and *vPi/Pronu in these varieties. This would leave all other reasonings intact; in particular, the implicational generalizations that can be derived by harmonic alignment remain valid. 24. That said, should cases that are truly analogous to (33c) and (33d) be acceptable in some varieties, this could be accounted for in the present system by ranking Faith(Top) higher than *vPi/Pronr,w; see the last footnote. 25. Thus, as long as one restricts the attention to these latter domains, one can do without a distinction between weak and reduced pronouns; see Cardinaletti and Starke (1996). The distinction between weak and reduced pronouns is solely motivated by the evidence from Rpronoun formation. 26. For instance, *vPi/Pronw cannot be violated by a well-formed sentence in simple clause, coordination, and topicalization contexts, but it can be violated by a well-formed sentence in a PP-internal context. Similarly, *vPi/Pronu cannot be violated by a well-formed sentence in simple clause contexts, but it can be violated in PP-internal, coordination, and topicalization contexts.

References Aissen, J. 1999. “Markedness and subject choice in optimality theory”. Natural Language and Linguistic Theory 17: 673–711. Altmann, H. 1981. Formen der Herausstellung im Deutschen [Linguistische Arbeiten 106]. Tübingen: Niemeyer. Artstein, R. 1999. “Hierarchies”. In Proceedings of Console VII, T. Cambier-Langeveld, A. Lipták, M. Redford and E. J. van der Torre (eds), 1–15. Leiden: SOLE.

230 Gereon Müller

Bakovic´, E. and Keer, E. 2001. “Optionality and ineffability”. In Optimality-Theoretic Syntax [A Bradford Book], G. Legendre, J. Grimshaw and S. Vikner (eds), 97–112. Cambridge, MA: MIT Press. Bayer, J. 1994. Barriers for German [SFB-Bericht Nr. 47]. Ms., Universität Stuttgart. Breindl, E. 1989. Präpositionalobjekte und Präpositionalobjektsätze im Deutschen [Linguistische Arbeiten 220]. Tübingen: Niemeyer. Büring, D. 2001. “Let’s phrase it! Focus, word order, and prosodic phrasing in German double object constructions”. In Competition in Syntax [Studies in Generative Grammar 49], G. Müller and W. Sternefeld (eds), 69–105. Berlin: Mouton de Gruyter. Cardinaletti, A. and Starke, M. 1996. “Deficient pronouns: A view from Germanic”. In Studies in Comparative Germanic Syntax, vol. II [Studies in Natural Language and Linguistic Theory 38], H. Thráinsson, S. D. Epstein and S. Peter (eds), 21–65. Dordrecht: Kluwer. Choi, H.-W. 1999. Optimizing Structure in Context. Scrambling and Information Structure [Dissertations in Linguistics]. Stanford: CSLI Publications. Chomsky, N. 1981. Lectures on Government and Binding [Studies in Generative Grammar 9]. Dordrecht: Foris. Chomsky, N. 1995. The Minimalist Program [Current Studies in Linguistics 28]. Cambridge, MA: MIT Press. Chomsky, N. 2000. “Minimalist Inquiries: The Framework”. In Step by Step, R. Martin, D. Michaels and J. Ursiagereka (eds), 89–155. Cambridge, MA: MIT Press. Chomsky, N. 2001. “Derivation by Phase”. In Ken Hale: A Life in Language, M. Kenstowicz (ed.), 1–52. Cambridge, MA: MIT Press. Fanselow, G. 1983. Zu einigen Problemen von Kasus, Rektion und Bindung in der deutschen Syntax. MA thesis, Universität Konstanz. Fanselow, G. 1991. Minimale Syntax. Habilitation thesis, Universität Passau. Fanselow, G. and Cávar, D. 2001. “Remarks on the Economy of Pronunciation”. In Competition in Syntax [Studies in Generative Grammar], G. Müller and W. Sternefeld (eds), 107–150. Berlin: de Gruyter. Gärtner, H.-M. and Steinbach, M. 2000. Conditions on the Scrambling and Fronting of Reduced Pronominals in Dutch and German. Ms., Universität Potsdam and Universität Mainz. Grewendorf, G. 1989. Ergativity in German [Studies in Generative Grammar 35]. Dordrecht: Foris. Grewendorf, G. and Sabel, J. 1999. “Scrambling in German and Japanese”. Natural Language and Linguistic Theory 17: 1–65. Grimshaw, J. 1997. “Projection, heads, and optimality”. Linguistic Inquiry 28: 373–422. Haftka, B. 1981. “Reihenfolgebeziehungen im Satz”. In Grundzüge einer deutschen Grammatik, K.-E. Heidolph et al. (eds), 702–764. Berlin: Akademieverlag. Haider, H. and Rosengren, I. 1998. “Scrambling”. Sprache und Pragmatik 49. Heck, F. and Müller, G. 2000a. “Successive cyclicity, long-distance superiority, and local optimization”. In Proceedings of WCCFL 19, R. Billerey and B. D. Lillehaugen (eds), 218–231. Somerville, MA: Cascadilla Press. Heck, F. and Müller, G. 2000b. Repair-Driven Movement and Local Optimization. Ms., Universität Stuttgart and IDS Mannheim.


Helbig, G. 1974. “Bemerkungen zu den Pronominaladverbien und zur Pronominalität”. Deutsch als Fremdsprache 11: 270–279. Hoberg, U. 1997. “Die Linearstruktur des Satzes”. In Grammatik der deutschen Sprache [Schriften des Instituts für Deutsche Sprache 7], G. Zifonun, L. Hoffmann and B. Strecker (eds), 1498–1680. Berlin: de Gruyter. Kayne, R. 1981. “On certain differences between English and French”. Linguistic Inquiry 12: 349–371. Legendre, G. 2001. “An introduction to optimality theory in syntax”. In Optimality-Theoretic Syntax [A Bradford Book], G. Legendre, J. Grimshaw and S. Vikner (eds), 1–27. Cambridge, MA: MIT Press. Legendre, G., Smolensky, P. and Wilson, C. 1998. “When is less more? Faithfulness and minimal links in wh-chains”. In Is the Best Good Enough?, P. Barbosa et al. (eds), 249–289. Cambridge, MA: MIT Press and MITWPL. Lenerz, J. 1977. Zur Abfolge nominaler Satzglieder im Deutschen [Studien zur deutschen Grammatik 5]. Tübingen: Narr. Lenerz, J. 1992. “Zur Syntax der Pronomina im Deutschen”. Sprache und Pragmatik 29. Müller, G. 1998. Incomplete Category Fronting [Studies in Natural Language and Linguistic Theory 42]. Dordrecht: Kluwer. Müller, G. 2000a. “Das Pronominaladverb als Reparaturphänomen”. Linguistische Berichte 182: 139–177. Müller, G. 2000b. Elemente der optimalitätstheoretischen Syntax [Stauffenburg Linguistik 20]. Tübingen: Stauffenburg Verlag. Müller, G. 2001. “Order preservation, parallel movement, and the emergence of the unmarked”. In Optimality-Theoretic Syntax [A Bradford Book], G. Legendre, J. Grimshaw and S. Vikner (eds), 279–313. Cambridge, MA: MIT Press. Müller, N., Tiedemann, C. and Schmitz, K. 2000. Starke, schwache und klitische Pronomina im Erwerb. Ms., Universität Hamburg. Prince, A. and Smolensky, P. 1993. Optimality Theory. Ms., Rutgers University. Riemsdijk, H. van. 1978. A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases [Studies in Generative Grammar 4]. Dordrecht: Foris. Riemsdijk, H. van and Williams, E. 1986. Introduction to the Theory of Grammar [Current Studies in Linguistics 12]. Cambridge, MA: MIT Press. Ross, J. 1967. Constraints on Variables in Syntax. Ph.D. Diss., MIT, Cambridge, MA. Sauerland, U. 1999. “Erasability and interpretation”. Syntax 3: 161–188. Thiersch, C. 1978. Topics in German Syntax. Ph.D. Diss., MIT, Cambridge, MA. Thráinsson, H. 2001. “Object shift and scrambling”. In The Handbook of Contemporary Syntactic Theory, M. Baltin and C. Collins (eds), 148–202. Oxford: Blackwell. Travis, L. 1984. Parameters and Effects of Word Order Variation. Ph.D. Diss., MIT, Cambridge, MA. Trissler, S. 1993. “P-Stranding im Deutschen”. In Extraktion im Deutschen I, F.-J. d’Avis et al. (eds), 247–291. Stuttgart and Tübingen: SFB 340. Vikner, S. 2001. Verb Movement Variation in Germanic and Optimality Theory. Habilitation thesis, Universität Tübingen.

231

232 Gereon Müller

Vogel, R. 2001. “Case conflict in German free relative constructions: An optimality theoretic treatment”. In Competition in Syntax, G. Müller and W. Sternefeld (eds), 341–375. Berlin: de Gruyter. Wegener, H. 1985. Der Dativ im heutigen Deutsch [Studien zur deutschen Grammatik 28]. Tübingen: Narr. Williams, E. 1999. Economy as Shape Conservation. Ms., Princeton University. Wunderlich, D. 1984. “Zur Syntax der Präpositionalphrase im Deutschen”. Zeitschrift für Sprachwissenschaft 3: 65–99. Zifonun, G. 2001. Grammatik des Deutschen im europäischen Vergleich: Das Pronomen [Amades 4/01]. Mannheim: IDS.

Cortical reflections of two pronominal relations Maria Mercedes Piñango Yale University

1.

Introduction

This paper is concerned with the nature of the interpretation of referential pronouns as revealed by the comprehension patterns of pronouns and reflexives observed in Broca’s aphasia. In off-line studies, Broca’s patients exhibit an asymmetry with respect to their comprehension of pronominals, in particular with the comprehension of referential pronouns and reflexives.1 These patients perform well in their comprehension of reflexives, but somewhat poorly in their comprehension of pronouns (e.g. Grodzinsky et al. 1993). By itself this observation is interesting, but its true importance lies in the fact that the Broca’s syndrome has lesion-localizing value. That is, this syndrome allows us to connect specific linguistic behaviors to specific cortical areas and not others (e.g. Goodglass and Kaplan 1972; Caramazza and Zurif 1976; Piñango, Zurif, and Jackendoff 1999). Localization in and of itself is not necessarily a fruitful object of research. After all, to know whether phenomenon Y can be associated with cortical region X does not in itself bring us much closer to knowing how language is implemented by the brain. However, what localization does offer is a window into how the brain distributes linguistic functions, and into how this distribution correlates with linguistic distinctions proposed in theories of language representation. It is this possibility that aphasia research (along with current imaging techniques such as ERP, PET or fMRI) makes available in the study of brainlanguage relations. The patterns of sparing and loss that it shows constitute a rich source of evidence which allows us to decide between competing characterizations of linguistic phenomena at all levels of description. The study of how pronoun interpretation emerges in the comprehension pattern of Broca’s patients constitutes one realization of this possibility.

234 Maria Mercedes Piñango

The paper is structured as follows. First, I present a characterization of the pattern of sparing and loss observed in Broca’s aphasia, which bears on theories of pronominal interpretation. Secondly, I discuss the main elements involved in the interpretation of pronominals from a representational perspective. Thirdly, I present two studies that investigate pronoun interpretation off- and on-line. Fourthly, I present an account which together with our theory of pronoun interpretation accounts for the seemingly disparate results. Finally, I discuss the implications of this account for our understanding of pronominal interpretation. To forecast, the experimental evidence will support the proposal that pronominal interpretation is composed of two dissociable mechanisms: one syntactic (i.e. coindexation) and one non-syntactic (i.e. coreference). Moreover, the results will show that despite the interaction of these two mechanisms, only the syntactic one depends on the integrity of the left anterior hemisphere (i.e. Broca’s area). It will be argued that a proper understanding of the language system (including how it relates to the brain) is obtainable only through the study of how models of language representation must be implemented in the mind of the language user. 1.1 Neuroanatomical considerations As mentioned, the connection to neuroanatomy I capitalize on turns on the lesion localizing value of the well-known syndrome of Broca’s aphasia. Broca’s aphasia patients are clinically identified by their nonfluent output (i.e. “telegraphic speech”) and relatively intact comprehension. In terms of cortical localization, Broca’s aphasia is associated with variable anterior left hemisphere lesion sites, often including but certainly not limited to the frontal operculum (Broca’s area). There is evidence suggesting that this variation can be partially explained cytoarchitectonically. However, what is important for present purposes is that this variability is restricted to the anterior region of the left hemisphere excluding other language relevant regions such as the superior posterior left hemisphere associated with Wernicke’s aphasia (for details see e.g. Amunts et al. 1999; Zilles 1997; Alexander et al. 1990; Benson 1985; Goodglass 1993; Naeser et al. 1989; Tonkonogy 1986; Vignolo 1988). 1.2 Linguistic considerations Investigation of the sentence comprehension of Broca’s patients reveals some well-defined comprehension problems: normal-like comprehension of so-called

Cortical reflections of two pronominal relations 235

agentive canonical structures — i.e. sentential constructions with verbs that license agents as their first argument (e.g. The girl chases the boy), and impaired comprehension of non-canonical structures (i.e. those in which the noun phrase preceding the verb is interpreted, not as the agent of the action, but as the entity acted upon or undergoing the action, e.g. The boy is chased by the girl or The boy who the girl chases is tall).2 It has been shown that even though this syntactically-capturable pattern may not be observed in each and every patient diagnosed with Broca’s aphasia it does emerge rather distinctly once we look at a set of patients with this clinical classification (e.g. Grodzinsky, Piñango and Zurif 1999; Drai and Grodzinsky 1999 and references therein). When Broca’s patients are considered as a group, therefore, it is found that their comprehension of canonical structures clusters tightly around a performance level that is well above chance (85% correct), whereas their comprehension for noncanonical structures clusters tightly around the 50% correct or chance level of performance. It would appear that, however apparent its role in production, the left anterior brain region associated with Broca’s aphasia seems also to be implicated in syntactic analysis for the purpose of comprehension.3 This generalization sometimes seems to apply beyond the neuroanatomical region defined by Broca’s aphasia since patients with other syndromes (e.g. Wernicke’s patients) also have problems comprehending noncanonical structures. Nevertheless, the canonical/non-canonical difference is not as clear for Wernicke’s patients as it is for Broca’s patients (Wernicke’s make a significant number of errors also in the canonical structures). This suggests that the contrast in this other group may not have the same underlying source as that for Broca’s. This conclusion is in fact suggested by experimental work where only Broca’s patients show limitations in their ability to construct syntactic dependencies in real-time (e.g. Zurif et al. 1993; Swinney et al. 1996). From this perspective, only the anterior region, associated with Broca’s aphasia, seems to be involved in the fast and automatic construction of these syntax-internal relations.4 The question arises as to what the source of the impairment for this syndrome is that can explain the different patterns of performance observed. Without such characterization it is not possible to formulate testable predictions with respect to the comprehension of pronominals. In Piñango (1999a, 1999b) I have proposed that the comprehension deficit in Broca’s aphasia involves a dampening of the language processing system that prevents it from carrying out on time the construction of syntactic structure. However, once this is accomplished, the syntactic representation is well-formed. That is, it is indistinguishable from that expected for the normal system. This generalization,


originally called the Argument Linking Hypothesis but which, given subsequent developments, is now termed the Slow-Syntax Hypothesis (SSH) tells us that in the presence of a “slowed-down” construction of syntactic structure, nonsyntactic information involved in sentence interpretation is allowed to emerge. Once this happens, there are two possible sources of interpretation: one that comes from the regular but slow system, and another that comes from bypassing the system altogether. It is the claim of the SSH that the availability of these two competing sources of interpretation is what we observe as chance performance in the comprehension of non-canonical constructions in Broca’s patients. The SSH accounts for the canonical/non-canonical contrast in the following way. Given a delay in the syntactic algorithm that assigns thematic roles to syntactic positions, the verb’s argument structure itself is allowed to emerge as a possible candidate for assignment of theta roles. Here a short explanation is necessary. For present purposes, “argument structure” refers to the ordered set of thematic roles (understood here as the set of semantic relations between a predicate and its arguments) licensed by the predicate, in this case, a verb. Following the thematic hierarchy (e.g. Jackendoff 1990; Grimshaw 1990), I assume that this order is uniform across verbs, such that regardless of the particular meaning of the verb, if the verb licenses an agent and a patient (as in The girl chases the boy), the agent will be the most prominent (first in the order) and if no other information is available (i.e. syntactic information) it will be linked to the highest grammatical function in the sentence. By the same token, if the verb licenses an experiencer and a theme (as in The movie frightens the boy), the experiencer will always be assigned to the highest grammatical function in the absence of syntactic information invoked to mediate that linking process. This information is present whenever a verb is used in a sentence. So, the distinction that is made here is that only in the abnormal case (the case where syntactic processing is delayed) is the ordering information that is contained in argument structure utilized for the purposes of linking arguments to thematic roles. In the normal case — as it is understood within the Government and Binding approach — such linking is done via syntactic mechanisms (i.e. thematic roles are not assigned directly to arguments but to the syntactic positions these arguments occupy in underlying representation). In cases where syntactic displacement has reversed the order of thematic roles in the sentence (i.e. the construction is non-canonical), potential competition arises between the order assigned via argument structure alone, and the order assigned via syntactic means (the correct assignment). It is this competition that results in chance performance in the comprehension of non-canonical


constructions (e.g. agentive passives and object relatives). In canonical constructions, by contrast, no competition emerges given that both the argument structure-based linking and the syntactically based linking produce the same order of thematic roles. Lack of competition results in above-chance performance for this kind of construction, as it is observed in agentive actives (e.g. The boy chases the girl) and subject relatives clauses (e.g. The girl who chases the boy is wearing a hat). In sum, the Slow-Syntax Hypothesis states that the source of impaired performance in the comprehension system of Broca’s patients is a slowed down syntactic processor. This factor allows non-syntactic mechanisms to emerge which are otherwise suppressed in the normal system, thus creating competition with the product of algorithmic processes. This competition arises only in those instances where there is a mismatch between the non-syntactic linking and the syntactic assignment. In this manner, the Slow-Syntax Hypothesis represents a characterization of the Broca’s impairment, one which is syntactic and minimal since once the syntactic structure is fully formed it is not distinguishable from the normal structure.5 This brief exposition of the linguistic relevance of the Broca’s comprehension deficit places us in a better position to investigate its relevance for pronoun interpretation. That is the focus of the next section.

2. Representation of pronominals: Coindexation vs. coreference As a general framework for our model of pronoun interpretation I follow the main ideas proposed in Reinhart (1983, 1986), (see also Reinhart and Reuland 1993, Grodzinsky and Reinhart 1993, and Avrutin 1999 among others). Even though the experimental evidence I present does not speak to the validity of the specifics of this proposal, it does support its main insight: referential relations must be distinguished from syntactically-governed dependencies. In traditional binding theory (e.g. Chomsky 1981) coindexation (the sharing of indices by at least two elements within a sentence) refers to two different concepts: coreference (e.g. a relation between a pronoun and a referential antecedent) as in Maryi thinks shei will win the competition and coindexation proper (e.g. the interpretation of a pronoun as a bound variable), as in Every girli thinks shei will win the competition. In contrast to coreference, bound variable anaphora is possible only when the antecedent c-commands the pronoun (e.g. Grodzinsky and Reinhart 1993: 72).


The important issue for our purposes is that whereas, under the standard view of binding theory, the syntactic constraints govern coreference relations (hence they are indistinguishable from coindexation), in the present approach, binding principles determine only the conditions allowing bound variable interpretation. Coreference relations are governed by an independent module which some have proposed to be part of a discourse level of representation (e.g. Heim 1982; Sells 1987; Avrutin 1999; Levinson 2000).6 One motivation for this distinction is that pronouns, in contrast to other pronominals, are the only referentially deficient elements that are free to choose their referent from the discourse (regardless of whether or not one is licensed in the sentence). Moreover, only pronouns can find an antecedent within the sentence in the absence of binding, that is, without being c-commanded by that antecedent, as in A party without Luciei annoys heri (Grodzinsky and Reinhart 1993:74). In this framework, pronouns fall under binding conditions but only with respect to their bound variable interpretation. In other words, the syntactic conditions on binding (coindexation) apply only to those instances where the pronoun is free in its governing category and behaves like a variable (i.e. it is bound by an antecedent in an argument position and lacks lexical content). Crucial for this approach, other cases of pronoun coindexation are noninterpretable. So, what are the principles that govern coreference? As mentioned, coreference is not syntactically-governed. It is defined as “the assignment of identical values to NPs with distinct syntactic indices” (Grodzinsky and Reinhart 1993: 77). Therefore, the sentence Maryi thinks that shei will win the competition is ambiguous between a bound variable interpretation: Mary (λx (x thinks x will win the competition) where it is the property of “thinking that one will win” that is attributed to Mary, and a coreference interpretation: Mary (λx (x thinks she will win the competition) — that is, it is the property of Mary that she thinks that she will win — where Mary and she can corefer. In this example, both interpretations are equivalent. However, it illustrates the point that even in the cases where coindexation is allowed, there is an alternative interpretation, coreference, that is not governed by syntactic constraints. Instead, such an interpretation is constrained by the rule known as Rule-I (e.g. Grodzinsky and Reinhart 1993: 80): “NP A cannot corefer with NP B if replacing A, at LF, with a variable bound by a trace of B yields an indistinguishable interpretation”. This means that in the sentence Mary loves her, Mary and her cannot corefer because in the LF representation Maryi (ei loves her), her, if we replaced it with a variable, would be c-commanded by e. Consequently, the interpretation would not be distinguishable from a bound variable interpretation (i.e. Mary loves herself).


Hence, coreference is ruled out. Notice that such is not the case with the coreference reading in Maryi thinks that shei will win the competition where the bound variable interpretation is distinguishable from the coreference one as I showed above. To summarize, the model of pronoun interpretation adopted here is based on Reinhart’s (1983, 1986) main observation that syntactically-governed binding processes (bound variable interpretation) are to be distinguished from coreference relations. The latter are taken to belong to “a broader process of semantic resolution” and are constrained by a rule on interpretation known as Rule-I. Even though all pronominals can have bound variable interpretation, only referential pronouns can also take part in coreference relations given their ability to choose their referent from the discourse. As it must be clear by now, the choice of this model of pronoun interpretation is no coincidence. Only a model that distinguishes reflexives from pronouns in this way can allow a connection with both the Slow-Syntax Hypothesis and the basic pattern of sparing and loss observed in Broca’s patients: bound variable interpretation (found in reflexives) is normal-like and interpretation of referential pronouns is impaired. In what follows, I present experimental evidence that discusses this observation in detail and which reveals a direct correlation with the coindexation-coreference distinction. This, in turn, will lead to a discussion of the interaction between the Slow-Syntax Hypothesis and our model of pronominal interpretation and to the presentation of testable hypothesis that results from this interaction.

3. Reflexives and pronouns in aphasia: Experimental evidence As mentioned, the pattern of comprehension of pronominals in aphasia is of interest among other reasons because it has not yet received a characterization that is consistent with the generalizations on aphasia comprehension proposed in the literature. This situation can be explained in two ways: either binding phenomena do not rely on the mechanisms that, for instance, the interpretation of subject and object relatives does, and therefore it is a coincidence that they be both affected by damage to Broca’s area, or the generalizations proposed to capture the canonicity contrast cannot account for these two sets of phenomena as they are presently formulated. I believe that the latter is correct. The rest of this paper is, thus, dedicated to presenting such a unified characterization.


3.1 Off-line evidence: Original and replication In order to better assess the comprehension pattern of Broca’s aphasic patients in reflexives and pronominals a study was conducted seeking to replicate Grodzinsky et al. (1993). It was a partial instantiation of that study because only Broca’s patients were tested (as opposed to Broca’s and Wernicke’s as in Grodzinsky et al.’s version), and only the pronoun and reflexive conditions were tested. Besides increasing the database, the main motivation for carrying out a new study was to assess performance with a larger number of tokens per type. This sought to sharpen the above-chance/chance distinction on which the interpretations are based. In what follows, I explain the experimental methodology that was used in both studies. 3.1.1 Methods Tasks. In both studies, the truth-value judgement task, an off-line technique was used (Crain and McKee 1995). In this technique, the experimenter uses toys to act out either a true or a false interpretation of a sentence. Afterwards, the subject responds by saying either “yes” (if sentence and interpretation match) or “no” (if sentence and interpretation don’t match). Subject scores are thus measured by the number of correct acceptances and the number of correct rejections per condition. The following illustrates the process (from Grodzinsky et al. 1993: 405): (1) a.

Experimenter: This is Mama Bear (pointing to a bear toy). This is Goldilocks (pointing to a doll). b. Experimenter acts out Mama Bear touching her own chest. c. Experimenter reads the sentence: “Is Mama bear touching herself?” (reflexive condition) “Is Mama bear touching her?” (pronoun condition)

(2) Subject responds with a “yes” or “true” if action and sentence match. She responds “false” or “no” if they don’t.

In Grodzinsky et al. (1993) each condition had six tokens for each reading (six correct acceptance vs. six correct rejection). In the replication, each sentence type had ten tokens per reading. In both the original study and the replication, all verbs were agentive, and all necessary characters were introduced for each sentence to create a context for the presentation of the action. Scenes had two characters present (even if in half of the cases represented by the reflexive condition only one character was


relevant for the action). This was done so as to introduce the necessary ambiguity in the picture to allow the subject to make mistakes. In this way the correct response (either correct rejection or correct acceptance) could only arise by interpreting the sentence heard using the expected linguistic mechanisms (as opposed to strategically from the acted out scene). Subjects. In Grodzinsky et al.’s study, fifteen subjects were tested: six Broca’s, four Wernicke’s, and three normal controls (matched for age and educational level); (two Broca’s subjects were later excluded from the dataset). In the partial replication five subjects were tested: two Broca’s (RD and JB) and three matching controls. In both studies, aphasic patients were selected based on their performance in diagnostic batteries (Goodglass and Kaplan 1972), radiological information, and neurological work-up. All Broca’s patients exhibited lesion sites around the perisyilvian region, in the anterior portion of the left hemisphere, consistent with our description of Broca’s aphasia. All lesions crucially excluded Wernicke’s area. Wernicke’s patients exhibited lesions circumscribed to the posterior superior temporal gyrus. All patients presented only unilateral lesions. 3.1.2 Results For both studies, normal controls performed at ceiling levels in all conditions suggesting that the task and sentences were accessible to any non-impaired native speaker of the language. Grodzinsky et al.’s (1993) results for Broca’s patients reveal above-chance performance in their interpretation of reflexives (mean correct = 4.9/6 for correct acceptance, and 5.5/6 correct for correct rejection). In the interpretation of pronouns, Broca’s performed above-chance in the correct acceptance or “true” condition (mean correct = 4.7/6), but their performance was no different from chance in the “false” condition (correct rejection): mean correct = 3.3/6. Wernicke’s patients patterned with Broca’s in that they, too, performed at above-chance levels on both readings of the reflexive condition. By contrast, they showed no distinction between readings in the pronoun condition. As for the replication, the results were as follows: Both patients exhibited the asymmetry between reflexives and pronouns — reflexives faring consistently better than pronouns. Both patients performed at above-chance levels on both readings of the reflexive condition (20/20 for JB and 19/20 for RD including both correct acceptance and correct rejection conditions). For the pronoun condition, one patient, RD, replicated Grodzinsky et al.’s results, performing above-chance on the acceptance condition as revealed by the binomial test


(p(8/10 correct) = .053), and chance on the correct reject condition (p(7/10 correct) = .172). The other patient, JB, however, showed less of a distinction with a performance of 5/10 correct for the correct acceptance condition (p >> .05) and with 7/10 correct for the correct rejection condition (p = .172). 3.1.3 Discussion As can be seen, the findings of Grodzinsky et al. (1993) have been partially replicated. Except for the performance of JB in the pronoun “true” condition, the second study renders some support to the original pattern reported which, therefore, calls for an explanation.7 The importance of this outcome is twofold. On the one hand, it supports the validity of the findings. On the other hand, it provides for interpretable results.8 Two similar accounts of these findings have been proposed independently by Grodzinsky et al. (1993) and Avrutin (1999). Both views invoke Reinhart’s (1983, 1986) approach to pronominal interpretation, including the existence of Rule-I. Those two proposals hypothesize that Broca’s aphasic patients (as well as children) have reduced processing resources. Impaired performance for pronouns in the correct reject condition (disjoint reference) is obtained by positing that the cost of determining whether or not coreference is allowed, that is, the cost of computing Rule-I is too high for the system. Grodzinsky et al. (1993) and Avrutin (1999) posit that Rule-I requires the hearer to do the following: while still holding the sentence in short-term memory, the system has to construct two representations — one involving the correct binding option (disjoint reference) and the alternative (and incorrect) coreference reading. The two representations have to be compared, relative to their context, in order for the system to decide whether or not they are distinguishable (recall that this is the crucial constraint that determines if coreference is allowed). If the two readings are distinguishable, coreference is allowed; if they are not, coreference is ruled out. Both Grodzinsky et al. (1993) and Avrutin (1999) argue that this process is just too costly for a system with such limited resources. The system, therefore, cannot properly carry out the required computation so chance performance results. They argue that chance performance will occur only for the correct reject condition for pronouns because, as proposed in the model, it is only in this case where a comparison of the two possible interpretations is required. This sort of explanation is interesting in that it connects fundamental representational distinctions such as coindexation and coreference with a specific pattern of impairment observed in Broca’s aphasia comprehension. It, therefore, makes neuropsychological evidence relevant for theories of language representation.


Nevertheless, as presented, this account leaves one important issue unresolved: It cannot serve as a characterization for the canonicity contrast (discussed above) which is nonetheless caused by damage to, arguably, the same component of the system (syntax). Hence, it is an undesirable situation where damage to a subcomponent of the system (in this case syntax) causes impairments for which different underlying cognitive sources are posited: in the case of the canonicity contrast, it is claimed that what has gone wrong in the Broca’s system is an inability to access or implement certain kinds of syntactic relations (e.g. Grodzinsky, 1995; Mauner, Fromkin, and Cornell 1993 among others). In the case of coreference relations, the source of the problem is a generalized processing limitation. This situation, although not untenable, makes the language-brain connection a lot more tortuous than expected and the evidence harder to interpret. However, there is another reason why this account is not as useful: as it is formulated it cannot capture all the available evidence even regarding pronominal interpretation. The next section presents and discusses that evidence. 3.2 On-line evidence: Real-time comprehension The observations on binding phenomena in relation to Broca’s aphasia do not stop with findings from off-line tests. Love et al. (1998) report on the performance of Broca’s patients in on-line comprehension of reflexives and pronouns. They carry out an experiment which examines the ability of the Broca’s and Wernicke’s systems to connect a possible antecedent and its pro-form (either a reflexive or a pronoun) as sentence comprehension unfolds. They do this via a cross-modal lexical decision task which works as follows: Uninterruptedly and over headphones, subjects hear a sentence containing both the antecedent and the pro-form (reflexive or referential pronoun). Right after the pronoun/reflexive has been heard, a letter-string (known as target) — representing either a word semantically related to the antecedent of the pro-form or a semantically unrelated foil — is flashed on a computer screen. The reaction time (RT) to the target — that is, the time it takes for the subject to indicate whether or not the letter-string presented is a word of English — is then compared to the RT for the semantically unrelated letter string. A significant difference between the RTs (where the RT for the related word is lower than that for the unrelated one) is interpreted as an instance where the pro-form has identified its antecedent in the sentence — that is, as an instance of binding.


Love et al. (1998) examined this process in three Broca’s and three Wernicke’s patients and matching controls in two conditions: (a) type of antecedent (correct vs. aberrant) and (b) type of pro-form (reflexive vs. pronoun). So, in the sentence The boxeri said that the skierj in the hospital had blamed himi/*j/ himselfj/*i * for the recent injury, right after either pro-form a letter-string signaled with “*” would flash on a computer monitor.9 When assessing binding for the reflexive, for instance, the letter string flashed would represent a word semantically related to skier. This represents the “correct antecedent” condition. In the “aberrant antecedent” condition, the target would be semantically related to boxer. These two related targets would be compared to two comparably unrelated ones (control targets). A facilitation effect is inferred when the RT for the related target is significantly lower than the RT for the unrelated target. That is, if at the point of the pronoun the related target elicited a lower RT than the unrelated target, we would conclude that facilitation for the related target has taken place. This is interpreted as an instance of binding. Love et al. (1998) focused on binding relations for reflexives in sentences containing either a reflexive or a pronoun. Therefore, the relation investigating correct and aberrant linking for the pronoun was not reported in that study. In personal communication, however, Love points out that subsequent testing of priming for boxer, reveals consistent results with those of Love et al. (1998). That is, only normal controls and Wernicke’s show activation when boxer acts as the antecedent for the pronoun, and only Wernicke’s and normals show no activation when boxer acts as antecedent for the reflexive. These findings reveal that normal controls as well as Wernicke’s patients show a pattern that reflects the formation of correct binding relations. Moreover, these two groups show appropriate activation of skier for the reflexive, and no activation for skier when the pro-form is the pronoun. Broca’s patients, by contrast, exhibit no priming for reflexives and only aberrant priming for pronouns — that is, the pronoun primes but for the wrong antecedent (skier in this case). As we can see, the off-line and on-line findings together present a seemingly inconsistent picture. On the one hand, in off-line tests, Broca’s exhibit good comprehension for reflexives and poor for pronouns. On the other hand, in real-time comprehension, these same patients show aberrant priming for pronouns and no priming for reflexives (Tables 1 and 2 below provide a summary of the results discussed). The question then arises: Can these two sets of findings be reconciled in a unified account? And if so, can such a characterization be consistent with accounts of the canonical/non-canonical contrast found in off- and on-line tasks?


Table 1.Performance pattern in Broca’s and Wernicke’s aphasia off-line comprehension Reflexives

Broca’s

Pronouns

True

False

True

False

Above-chance

Above-chance

Above-Chance* CHANCE

*as reported by Grodzinsky at al. (1993); recall however, that the second study replicated the results but only for one of the two subjects.

Table 2.Performance pattern in Broca’s and Wernicke’s aphasia on-line comprehension: activation of antecedent after hearing pro-form.*

Broca’s Wernicke’s

Reflexive

Pronoun

No priming Priming

Aberrant priming No priming

*as reported by Love et al. (1998).

I propose here that the findings can be captured in a unified fashion by the Slow-Syntax Hypothesis (SSH). In the remainder of the paper, I present an SSH-based characterization of the findings for Broca’s patients, and discuss the implications of this account for our understanding of language-brain relations with respect to pronominals.

4. The Slow-Syntax Hypothesis and pronominal interpretation in aphasia: Interpretation of the results In the case of the canonicity contrast, the SSH posits that in the presence of slowed down syntactic structure formation, non-syntactically mediated information such as the verb’s argument structure is used by the system as an alternative mechanism for linking thematic roles to arguments in the sentence. Chance performance arises whenever the order of arguments in the syntactic structure does not match with those given in the argument structure of the verb. For the case of pronominal interpretation, I propose that the interpretation of pronominals is affected whenever syntactic slowing allows non-syntactically-governed mechanisms (i.e. coreference) to emerge as alternatives for interpretation.


4.1 Reflexives: Off-line and on-line Above-chance performance is predicted for both the correct accept and correct reject conditions for reflexives in off-line techniques as follows. The only way a reflexive can be interpreted is as a bound-variable, which requires syntactic tree formation to be executed (i.e. the antecedent must c-command the reflexive). So, even though the syntactic tree is slowed down, the interpretive system is forced to wait for the syntactic structure to be formed since without it, no interpretation is possible. That is, contrary to the passive and object relative constructions (non-canonical structures), in anaphoric constructions of this sort no alternative interpretation is available given the nature of the constraints on reflexive interpretation. In on-line studies, absence of priming for reflexives is predicted as follows. Activation of an antecedent for a reflexive requires a fully formed syntactic representation since just like in the off-line case, c-command is a necessary prerequisite for bound-variable interpretation. At the time the experimental task taps the system — immediately after the reflexive has been heard — the syntactic tree is still not fully formed (since in the Broca’s system this process is slowed down). This means that at that moment the reflexive cannot be properly interpreted. In the cross-modal task, this translates into no facilitation of the antecedent. This is, in fact, what Love et al. (1998) report. 4.2 Pronouns: Off-line and on-line For the true condition of pronouns (e.g. Mama Bear is touching her, where her is Goldilocks), above-chance performance is predicted as follows. In the presence of slowed down syntactic-tree formation, coreference (between Mama Bear and her) is incorrectly allowed to emerge since there is no structure to rule it out. Such coreference, however, is disallowed in the interpretation acted out by the experimenter (since Mama Bear is, in fact, touching Goldilocks). Moreover, once the sentence is fully formed by the system, its interpretation supports the interpretation provided by the action in the acted out scene. These two sources of evidence (the possibility of disjoint reference allowed by Rule-I and supported by the act out scene) bias the interpretation towards disjoint reference (the correct interpretation) which yields a higher number of correct responses. In contrast, chance performance is predicted for the pronoun in the correct reject condition (Mama bear is touching her, where her is Mama Bear) in the following way. In the presence of a delayed syntactic tree coreference is as always


allowed to emerge. However, unlike the correct accept condition, by the time the syntactic-tree is formed a different interpretation is created (incorrect coreference interpretation). But this interpretation condition is not supported by the action in the scene. Therefore, no compensating bias takes place. Instead, the acted out scene supports a possibility not available in the normal interpretation, a coreference interpretation. This leads to two competing interpretations: disjoint reference — which allows the correct reject condition to be true — and coreference, which is incorrect, but which happens to be the one supported by the action presented by the experimenter (since Mama Bear is indeed touching herself). So, in this case, sometimes the coreference interpretation will (incorrectly) win, sometimes it will lose out to the syntactically ruled (and correct) disjoint reference; hence chance performance results. As can be seen, it is the interplay of three elements: slowed-down syntactic tree formation, emergence of nonsyntactic information (coreference), and the nature of the task (which introduces either compensating or competing bias) that predicts the reported results. In the case of the on-line evidence, aberrant priming for pronouns is similarly predicted. Just like in the off-line case, simple coreference (a nonsyntactic process) is always available as a possible source of interpretation for pronouns, as soon as the pronoun is heard. In the absence of a fully-formed syntactic tree, this mechanism of coreference becomes the only means of interpretation available to the system in an on-line task, which, in this case, only serves to mislead the system. In the cross-modal task this is observed in the form of facilitation for the wrong antecedent (the semantically related target for the wrong antecedent elicited a lower reaction time than the control target) or as Love et al. refer to it “aberrant” priming. As can be seen, the puzzling facts about binding in off- and on-line aphasic comprehension can now be viewed in a unified manner. As I have pointed out, the kind of dissociation observed between reflexives and pronouns is capturable once we separate syntactic (co-indexation) from non-syntactic mechanisms (coreference) and place it in a context of a compatible descriptive generalization.

5. Summary and conclusions In this paper I have evaluated the patterns of performance in on- and off-line tasks for Broca’s aphasia. Given its localizing value, this syndrome allows connections for the implementation of linguistic phenomena to specific cortical regions and no others. I have presented one linguistically-based characterization


for the comprehension deficit observed in Broca’s aphasia, the Slow-Syntax Hypothesis, that supports them. I have proposed that the pattern of performance presented is captured when one focuses on the implications of a “slowed-down” syntactic tree formation for sentence interpretation and the possibility for non-syntactic information to emerge as a potential competing source of sentence interpretation (e.g. coreference). This proposal, thus, unifies the pattern of performance in pronominals with those found for the active/ passive contrast and the subject/object relative contrast in this population. The fact that so many contrasts in performance can follow from this description, in combination with independently motivated models of representation, argues for its validity and, consequently for the notion that only certain specific syntactic aspects of sentence interpretation depend on the integrity of Broca’s area. The present results strongly suggest that even though the interpretation of pronominals per se is not dependent on the integrity of Broca’s area, other interacting and constraining mechanisms, such as availability of a boundvariable interpretation are. It is these mechanisms which depend on Broca’s area but which are observed only under specific experimental conditions (i.e. on-line comprehension). The above discussion has several implications: (1) The pattern observed in the behavior of binding relations can only be explained via representational theories where coreference and coindexation are independent processes belonging to different linguistic subsystems (semantics/discourse and syntax respectively). So, only theories that allow for this dissociation have neurological and psycholinguistic support; (2) Evidence from language breakdown represents a source of evidence in the investigation of the language system at all levels of description; and (3) The language system seems to be built in such a way that differences posited for representational models carry over not only to processing distinctions but can also be seen in dissociations in the cortical regions on whose integrity the implementation of these phenomena depends. This supports the observation also noted by others that the differences between a representational model of language and processing/neurological models of language are for the most part methodological (e.g. Jackendoff 1997; Wiese 2001). In the end, a valid architecture of the language system must have consequences with respect to its implementation and its neurology. The evidence presented here allows us to make such a connection by bringing together the various levels of language description (representational, processing, neurological) and in this way, it closes the gap between abstract models of language representation and their neurological realization.


Notes 1. Off-line studies are those where subjects are given time to reflect on their response after the stimulus sentence has been presented. In contrast, on-line studies are those where the experimental techniques require subjects to react to stimulus as quickly and as accurately as possible as comprehension of the stimulus sentence unfolds. 2. The observations regarding the canonical/non-canonical contrast have been made for a variety of languages such as English, Dutch, Spanish, Japanese, Mandarin, French, Hungarian, Italian, and German. Yet, it must be noted that all languages where this contrast has been attested place the subject before the object in their canonical form (relative position of the verb is not as relevant for these purposes). That is, all these languages are claimed to be Subject-Object languages. So, the claim that Broca’s patients exhibit this contrast is not universal. Its universality has yet to be confirmed by investigating aphasia comprehension in so-called Object-Subject languages. That is, languages where the subject does not precede the object in its canonical sentential form (i.e. agentive active declarative sentence). 3. As I have noted, the claims that will be made here are intended to account for the patterns of sparing and loss observed in the comprehension system of Broca’s aphasia. So, even though a descriptive generalization that encompasses both production and comprehension is attractive and indeed desirable, such an account is still a promissory note. Nevertheless, the fact that comparable patterns of sparing and loss between the two systems have been found suggests the possibility of such an overarching account (see De Roo, 1999 and this volume, for a counterpart to the present study in the Broca’s production system). 4. The differences alluded to here have to do with the instantiation of long-distance dependency relations such as wh-movement in real-time comprehension (the so-called “gapfilling” experiments) (see Zurif et al. 1993; Swinney et al. 1996). Using the cross-modal lexical decision paradigm (see methods section in main text), these studies investigate the facilitation of an antecedent at the point in the sentence where that antecedent gets interpreted (within the Government and Binding approach this is the position where that antecedent is base-generated in the underlying representation and has left behind a trace). Normal control subjects (i.e. non-neurologically impaired subjects) show such a facilitation. So, for instance, in the sentence The doctori from south Boston whoi/j the boxer (*) punched (tracej) wears black sneakers, subjects show facilitation for nurse as compared to noise only at the point of the trace (as opposed to the control position (*) which is not a syntactically licensed position for that argument). Studies report that Wernicke’s patients pattern after normal controls in that they also exhibit such activation. By contrast, Broca’s patients show no evidence of activation of the antecedent at the point of trace. It is this asymmetry in performance which suggests that even though in off-line experiments both populations (Broca’s and Wernicke’s) exhibit similar performance patterns, the sources of the impairment that yield those patterns differ in nature. 5. This said, it is important to keep in mind that all along what I have described is a linguistic process (formation of syntactic structure, i.e. the operation of merge) associated with a specific cortical regions, the left anterior frontal cortex. This region, however, is not intended to indicate a linguistic “center”. Rather, it should be understood as a processing area. That is,


the implementation (i.e. production and/or comprehension) of the identified phenomena seems to be dependent on the integrity of this area (as opposed to being located on these areas). 6. Bound variable interpretation captures in addition to the interpretation of bound pronouns, the interpretation of co-argument reflexives, wh-traces, NP-traces, and (nonarbitrary) PRO. In all these cases, the intended antecedent c-commands the variable and as a result it binds it. 7. As this is work in progress, more data is currently being gathered to solidify this finding. However, as it stands now, it would appear that the pattern first uncovered by Grodzinsky et al. (1993) is still valid and warrants an explanation. 8. One reviewer asks whether the source of the problem in the pronoun condition could not have been that the sentence might be interpreted as incomplete, with her being a possessive pronoun. This is a valid point given the evidence I presented and considering the performance of the second patient in the replication (JB). However, another condition also examined in Grodzinsky et al. (1993), but not presented here, suggests that it is not a likely explanation. This condition tested the comprehension of bound pronouns (pronouns which, like reflexives, have a bound variable interpretation), as in the sentence Is every bear touching her? Crucially, performance was above chance in this condition both for the correct accept and correct reject sides. This suggests that the misinterpretation by the Broca’s patients in the correct reject referential pronoun condition was not the result of an incomplete DP interpretation, but of a difference in the interpretive mechanism itself. 9. Example taken from Love et al. (1998).

References Amunts, K., Klingberg, T., Binkofski, F., Schormann, T., Sitz, R. J., Roland, T. E. and Zilles, K. 1999. “The cytoarchitecture of Broca’s region and its variability”. NeuroImage 7: 353. Alexander, M., Naeser, M. A. and Palumbo, C. L. 1990. “Broca’s area aphasias: Aphasia after lesions including the frontal operculum”. Neurology 40: 353–362. Avrutin, S. 1999. Development of the Syntax-Discourse Interface. Boston: Kluwer. Benson, F. 1985. “Aphasia”. In Clinical Neuropsychology, K. Heilman and E. Valenstein (eds), 17–47. New York and Oxford: Oxford University Press. Caramazza, A. and Zurif, E. 1976. “Dissociation of algorithmic and heuristic processes in sentence comprehension: Evidence from aphasia”. Brain and Language 3: 572–582. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Crain, S. and McKee, C. 1985. “Acquisition of structural restrictions in anaphora”. Proceedings of NELS 16: 94–110. Drai, D. and Grodzinsky, Y. 1999. Comprehension regularity in Broca’s aphasia? There’s more of it than you ever imagined. Brain and Language 70: 139–143. Goodglass, H. 1993. Understanding Aphasia. San Diego: Academic Press. Goodglass, H. and Kaplan, E. 1972. The assessment of aphasia and related disorders. Philadelphia, PA: Lea and Febiger. Grimshaw, J. 1990. Argument Structure. Cambridge, MA: MIT Press.

Cortical reflections of two pronominal relations

Grodzinsky, Y. 1995. “Trace deletion, theta roles and cognitive strategies”. Brain and Language 51: 469–497. Grodzinsky, Y., Piñango, M. M. and Zurif, E. 1999. “The constrained nature of Broca’s aphasia comprehension”. Brain and Language 67: 134–147. Grodzinsky, Y. and Reinhart, T. 1993. “The innateness of binding and coreference”. Linguistic Inquiry 24: 69–101. Grodzinsky, Y., Wexler, K., Chien, Y.-C. and Marakovitz, S. 1993. “The breakdown of binding relations”. Brain and Language 45: 371–395. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Diss., University of Massachusetts, Amherst. Jackendoff, R. 1990. Semantic Structures. Cambridge, MA: MIT Press. Jackendoff, R. 1997. The Architecture of the Language Faculty [Linguistic Inquiry Monographs 28]. Cambridge, MA: MIT Press. Levinson, S. C. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge, MA: MIT Press. Love, T., Nicol, J., Swinney, D., Hickok, G. and Zurif, E. 1998. “The nature of aberrant understanding and processing of pro-forms by brain-damaged populations”. Brain and Language 65: 59–62. Mauner, G., Fromkin, V. and Cornell, T. L. 1993. “Comprehension and acceptability judgements in agrammatism: Disruptions in the syntax of referential dependency”. Brain and Language 45: 340–370. Naeser, M. A., Palumbo, C., Helm-Estabrooks, N., Stiassny-Eder, D. and Albert, M. L. 1989. “Severe nonfluency in aphasia: Role of the medial subcallosal fasciculus and other white matter pathways in recovery of spontaneous speech”. Brain 112: 1–38. Piñango, M. M. 1999a. Syntactic and Semantic Operations and Their Neurological Underpinnings. Ph.D. Diss., Brandeis University, Waltham, MA. Piñango, M. M. 1999b. “Canonicity in Broca’s sentence comprehension: The case of psychological verbs”. In Language and the Brain, Y. Grodzinsky, L. Shapiro and D. Swinney (eds), 334–335. San Diego: Academic Press Piñango, M. M., Zurif, E. and Jackendoff, R. 1999. “Real-time processing implications of aspectual coercion at the syntax-semantics interface”. Journal of Psycholinguistic Research 28: 395–414. Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press. Reinhart, T. 1983. Anaphora and Semantic Interpretation. London: Croom Helm. Reinhart, T. 1986. “Center and periphery in the grammar of anaphora”. In Studies in the Acquisition of Anaphora, Vol. 1, B. Lust (ed.), 123–150. Dordrecht: Reidel. Reinhart, T. and Reuland, E. 1993. “Reflexivity”. Linguistic Inquiry 24: 657–720. Roo, E. de. 1999. Agrammatic Grammar: Functional Categories in Agrammatic Speech. Ph.D. Diss., University of Leiden. Roo, E. de. this volume. “Pronoun omission in Dutch and German agrammatic speech”. Sells, P. 1987. “Aspects of logophoricity”. Linguistic Inquiry 18: 445–481. Swinney, D., Zurif, E., Prather, P. and Love, T. 1996. “Neurological distribution of processing resources underlying language comprehension”. Journal of Cognitive Neuroscience 8: 174–184. Tonkonogy, J. 1986. Vascular Aphasia. Cambridge, MA: MIT Press.

251


Vignolo, L. 1988. “The anatomical and pathological basis of aphasia”. In Aphasia, F. C. Rose, R. Whurr and M. A. Wyke (eds), 227–255. London: Whurr. Wiese, H. 2001. Semantics as a gateway to language. Paper presented at the workshop on ‘The Syntax-Semantics Interface: Linguistic Structures and Processes’, Annual Meeting of the German Society for Linguistcs (DGfS), Leipzig 2001. Submitted for publication in: Structures and Processes at the Syntax-Semantics Interface, H. Härtl and H. Tappe (eds). Zilles, K., Schleicher, A., Langemann, C., Amunts, K., Morosan, P., Palomero-Gallagher, N., Schormann, T., Mohlber, H., Burgel, U., Steinmetz, H., Schlaug, G. and Roland, P. 1997. “Quantitative analysis of sulci in the human cerebral cortex: development, regional heterogeneity, gender difference, asymmetry, intersubject variability and cortical architecture”. Human Brain Mapping 5: 218–221. Zurif, E., Swinney, D., Prather, P., Solomon, J. and Bushell, C. 1993. “An on-line analysis of syntactic processing in Broca”s and Wernicke’s aphasia’. Brain and Language 45: 448–464.

Pronoun omission in Dutch and German agrammatic speech Esterella de Roo Leiden Centre for Linguistics

1.

Introduction

Agrammatism is a language disorder associated with Broca’s aphasia. Speech production of agrammatic patients is slow and effortful, while utterances are short and often incomplete. The main characteristics of agrammatic speech are the omission and/or substitution of bound and freestanding grammatical morphemes, and a reduced variety of sentential types (Menn and Obler 1990). Various theoretical explanations of agrammatic speech production can be found in the literature. These explanations can be subdivided into purely linguistic and psycholinguistic ones.1 The purely linguistic explanations assume a syntactic disorder that is related to the functional structure of sentences. Ouhalla (1993) states that agrammatic utterances are lexical projections only, functional projections being absent altogether. Friedmann and Grodzinsky (1997) have proposed that agrammatic utterances involve truncated functional structures; lower projections are present but higher projections are missing (cf. Hagiwara 1995). The psycholinguistic explanations do not assume a (partial) loss of syntactic representations but a limitation in the capacity to process syntactic information in agrammatic speech. Some authors claim that agrammatic patients — who have a limited capacity to generate sentences — opt for normal elliptical structures (Kolk et al. 1990, De Roo 1999) or linguistic structures which require less processing resources (Avrutin 1999). There are strong arguments in favour of a psycholinguistic explanation. First, there is severity variation: patients vary substantially in the number of functional elements they omit in their speech. If patients only omit a small number of functional elements, this cannot be due to a lack of syntactic knowledge.

254 Esterella de Roo

Second, within patients, task variation is observed. Hofstede and Kolk (1994) show that Dutch patients infrequently produce grammatical morphology in their spontaneous speech, but do so much more often in a picture description task. Moreover, patients who never produce passive constructions in their spontaneous speech, do so regularly after ‘priming’ of the passive construction (Hartsuiker and Kolk 1998). Third, the reduced utterances of agrammatic speakers are grammatically correct and comparable to reduced utterances in normal speech (e.g. Kolk and Heeschen 1992, Hofstede 1992, Avrutin 1999, De Roo 1999). After having briefly introduced the pattern of agrammatic speech production in general, I come to the main topic of this paper: the realization of pronouns in agrammatic speech. Many studies on agrammatic comprehension of pronouns are found in the literature,2 but considerably less on agrammatic production of pronouns. Problems with pronouns are usually related to their status as function words or closed-class words. Bradley et al. (1980) showed in an experiment with control subjects that there is a special retrieval routine for closed-class words. In addition, they demonstrated that this routine is disrupted in agrammatism. Also Friederici (1985) concludes that the ‘closed-class’ route, depending on syntactic information alone, seems to be disrupted in agrammatic speakers. Ouhalla (1993) formulates the same idea within the framework of linguistic theory.3 Kohn et al. (1997) discuss the results of a production experiment in which pronouns are avoided rather than omitted. Six out of nine English agrammatic patients produced fewer personal pronouns in favour of nouns as compared to a group of non-aphasic control subjects. Kohn et al. attribute the difficulty in using pronouns to the syntactic demands associated with pronoun selection — without specifying these syntactic demands, though. We can infer the demands of pronoun selection from a general model of lexical access of pronouns, as presented, for example, in Schmitt et al. (1999). This model includes three stages: (1) message encoding — a pronoun can only be selected if the corresponding lexical concept is marked as ‘in focus’ in the discourse; (2) grammatical encoding — the lexical selection of a pronoun includes all relevant grammatical information, such as gender, person and number features; (3) phonological encoding — a phonetic plan for the articulation of the pronoun is built up. If pronouns are not avoided, as in the study of Kohn et al. (1997), disturbances at the different stages of the models might lead to specific problems. A problem in grammatical encoding would lead to the misselection of pronouns, that is, substitution of gender, number and person features. A problem in phonological encoding would lead to misarticulation,

Pronoun omission in Dutch and German agrammatic speech 255

that is, phonematic paraphasias. It is not clear to me what the consequences of a disturbance at the stage of message encoding would be. Reports on spontaneous speech observe a fair amount of pronoun omission. Saffran et al. (1989) developed a system to analyse spontaneous speech (the Quantitative Production Analysis), which aims at finding the essential characteristics of the agrammatic speech pattern. Most agrammatics in their study are found to show a substantial reduction in pronoun usage (noun/ pronoun ratio) as compared to control speakers (cf. Rochon et al. 2000). The cross-linguistic study of Menn and Obler (1990), including 14 different languages, also concludes that pronoun omission is one of the characteristics of agrammatic speech. They claim that patients suffer from output syntactic processing difficulties, which mainly affect grammatical morphemes that have a relational meaning and require computations involving several words. Not much is known about the realization of different types of pronouns or the syntactic conditions that affect pronoun realization in agrammatic speech. In general, personal pronouns seem to dominate in agrammatic pronoun production, but also in pronoun omission and substitution (e.g. Kolk et al. 1990, Stark and Dressler 1990). The relevance of the grammatical status, especially the distinction between subject and object pronouns, is not yet settled. Kolk et al. (1990) found no clear difference between subject and object pronouns in the speech of their two Dutch patients. Kolk and Heeschen (1990) notice that if the verb is in the non-finite form in Dutch and German agrammatic utterances, the subject pronoun is usually missing. Friedmann and Grodzinsky (1997) also observe that most of the expected subject pronouns are missing in the speech of a Hebrew patient, who shows severe problems in correct tense marking of the verb. Hebrew is a restricted Pro-Drop language: it allows subject pronoun omission in first and second person contexts in the past and future tense, but omission in third person contexts and in present tense is not allowed. The Hebrew patient, nevertheless, omits subject pronouns in every context. Friedmann and Grodzinsky relate the omission of subject pronouns to a defective Tense node, which is not able to assign nominative case to the subject (cf. Cahana-Amitay 1997 for English and Dutch). In this paper, I present a detailed study on the realization of pronouns in the spontaneous speech of four agrammatic Broca’s aphasics. I discuss the overall omission rates and look for disturbances in any of the stages in the model of lexical access to pronouns. The main focus of the study is on variables potentially relevant to the realization of pronouns in spontaneous speech. Three variables are investigated. The first variable is tense: the distinction between


tensed and untensed utterances is investigated. I demonstrate that the pronoun omission rate is very high in untensed utterances as compared to tensed utterances. The second variable is grammatical status: the distinction between subject pronouns and object pronouns is investigated. I demonstrate that subject pronouns are almost completely lacking from untensed clauses. Combining these results, I will develop a general account of untensed clauses in agrammatic speech that explains the near absence of subject pronouns and the possible omission of object pronouns. The third variable concerns language specific properties, that is, the possibility of pronoun omission in the normal grammar of a language. Friedmann and Grodzinsky (1997) mention the fact that Hebrew is a restricted Pro-Drop language, but claim that the subject pronoun omissions in the speech of their patient are not regular Pro-Drop cases. I investigate the relevance of a Germanic omission phenomenon in the speech of the patients included (two Dutch and two German patients): the omission of topic pronouns from linear first position in tensed sentences known as Topic Drop and Diary Drop. I demonstrate that the patients show more omission from first position than from non-first position. In the Discussion section, I relate my results to the different types of explanation proposed for agrammatic speech.

2. Method 2.1 Speech data The agrammatic speech studied in this paper was obtained in different ways. The tests carried out differ in how much ‘discourse’ is presented to the patient. First, a spontaneous speech interview is conducted in which patients are encouraged to speak freely with as little interruption from the interviewer as possible (Aachen Aphasia Test (AAT), see Huber et al. (1983) for the original German version, see Graetz et al. (1992) for the Dutch version). A few general questions are asked to initiate spontaneous speech: ‘How did your aphasia start and how are you doing right now?’, ‘Can you tell me something about your family/(former) profession/hobbies?’, etc. The interviewer avoids yes/no questions. Secondly, a test is given that measures the verbal communicative abilities of aphasic patients (Amsterdam-Nijmegen Test voor Alledaagse Taalvaardigheid (ANTAT), see Blomert et al. 1995). Patients are presented with ten daily-life situations and asked what they would say in such a situation.


Examples of situations are going to the shoemaker to get your shoe repaired or cancelling an appointment with your doctor by phone. Thirdly, narrative retelling takes place, in this case the story of Little Red Riding Hood. Fourthly, different picture description tests are carried out: (a) a test including pictures showing one or more people involved in a specific situation or action; patients have to describe the picture in one sentence (AAT subtest); (b) a test in which patients have to tell a story based on one complex picture (Cookie Theft picture, Boston Diagnostic Aphasia Examination (BDAE), see Goodglass and Kaplan 1972); (c) a test in which patients have to tell a story based on a sequence of pictures (Wechsler-Bellevue picture sequences, see Wechsler 1949). 2.2 Patients The four patients included in this study are two Dutch (GS and HB) and two German (MU and ME) Broca’s aphasics, as classified by the Aachen Aphasia Test.4 All patients had suffered a stroke at least one year before the time of testing. The patients are three males and one female (GS), aged between 47 and 76. For each patient, a speech fragment is presented below. In addition to the aphasic patients, ten Dutch controls participated in the ANTAT test. Occasionally, the Dutch patients will be compared to the controls.

Patient GS GS is described in detail in De Roo (1999). Her speech was obtained by means of the AAT spontaneous speech interview, the ANTAT test and picture descriptions (AAT). Interviewer’s question: Can you tell me how your disease started and how you are doing now? Answer of GS: moeilijk — maar één arm — en niet meer praten — difficult only one arm and not more talk.inf een beetje veranderd — niet helemaal a bit change.part not quite wat praten — en sommetjes — a.bit talk.inf and sums en nog een andere schrijfgroep ook and also another writing.group too


Patient HB HB is described in detail in De Roo (1999). The speech studied in this paper was obtained by means of the AAT spontaneous speech interview and the ANTAT test. Scenario (ANTAT test): A cat has strayed in at your place. He has a collar around his neck with a phone number on it. You call up. What do you say? Response of HB: bij ons is aan komen lopen — with us is at come walk.inf had bandje om zijn nek — had.3sg small.collar around his neck en dus nummer is derbij — and so number is with.it nummer is gezet — belt de nummer op number is put.part call.3sg the number up

Patient MU MU is described in more detail in Hofstede (1992). The speech studied in this paper was obtained by means of a spontaneous speech interview. Interviewer’s question: Can you try to tell what happened when you became ill? Answer of MU: und dann Montag Dienstag frei — und dann Kopfschmerzen — and then Monday Tuesday off and then headache einkaufen gegangen — shop.inf go.part ich so runtergeschmissen — I like.this down.throw.part und Aschenbecher runtergeschmissen and ash.tray down.throw.part

Patient ME ME is described in detail (as Mr. Meyer) in Stark and Dressler (1990). The speech studied in this paper was obtained by means of a spontaneous speech


interview, narrative re-telling and picture description (Cookie Theft picture and the Wechsler-Bellevue Picture Sequences). Complete transcripts of ME’s speech are published in Stark and Dressler (1990); I re-analysed his speech with respect to features relevant to this paper (see section 2.3 Syntactic analysis). Sequence of pictures: A man wakes up too late. He has to hurry to work. In the office he continues sleeping. Description of ME: der Mann frühstück — und Eile — und eilt auf die Strasse — the man breakfast and rush and hurry.3sg onto the street und im Büro — and at.the.dat office im Büro hat nicht gearbeit at.the.dat office have.3sg not work.part sondern geschlafen but sleep.part 2.3 Syntactic analysis The aphasic speech was segmented on the basis of syntactic, prosodic and semantic criteria. Repetitions, automatisms and corrected parts were removed from the speech. In the agrammatic speech, the realization of pronouns was investigated, that is, omission and substitution of pronouns. Obligatory contexts for pronouns are defined as follows: the argument position(s) of a verb or a preposition. Pronouns are only considered to be missing if the verb or preposition is present. Such a conservative analysis crucially differs from the analysis of, for example, Stark and Dressler who reconstruct the agrammatic utterances first. For example, the Little-Red-Riding-Hood-story of patient ME contains the utterance Ohren ‘ears’. This one-word utterance is reconstructed as [es] [fragt] [nach] [den] Ohren ‘[she] [asks] [about] [the] ears’. All the elements between square brackets, including the subject pronoun, are then counted as omissions. According to my method of analysis, the one-word utterance Ohren ‘ears’ does not include an obligatory context for a pronoun, and hence, no pronoun omission is scored. Different syntactic contexts are distinguished. Tensed contexts (utterances which include a tensed verb) are compared to untensed contexts (utterances which include only a verb(s) that is (are) not overtly marked for tense). In


addition, (linear) first utterance position is compared to non-first utterance position. Comparisons between different syntactic contexts are statistically tested by means of a Chi-square test (χ2) with ranges for the 0.05 level of significance (p < 0.05) and one degree-of-freedom. In the case of very small numbers, occasionally a Fischer’s Exact test (F. E.) will be used, with a similar level of significance.

3. Results 3.1 Overall pronoun omission 3.1.1 Empirical results The overall pronoun omission rates of the four patients are shown in Table 1. There is a lot of variation between the patients in the amount of pronoun omission. The omission rates are relatively low: three of the patients produce pronouns in more than half of the obligatory contexts. Table 1.Overall pronoun omission

patient ME patient HB patient MU patient GS

pronouns present

pronouns omitted

% omission

50 184 43 39

10 54 34 194

17% 23% 44% 83%

3.1.2 Lexical access of pronouns In the process of lexical access to pronouns, three stages are distinguished: message encoding, grammatical encoding and phonological encoding. Problems at the latter two stages might have clear consequences for spontaneous speech. At the stage of grammatical encoding the grammatical features for number, person and gender are selected in accordance with the features of the pronoun’s antecedent. A problem at this stage would lead to pronoun substitution. Only very few pronoun substitutions are found. Some examples are given for Dutch in (1) and for German in (2).


(1) a.

dat heb ik veel aan that have.1sg I much about ‘I have much about that’ b. meneer wil ie maatregelen nemen mister want.3sg he measures take.inf ‘mister, does he want to take measures’

(2) a.

es sieht den Wecker it see.3sg the.acc alarm.clock ‘It sees the alarm clock.’ b. wie ihn eine Keks nimmt how he.acc a.nom/acc cookie take.3sg ‘how he takes a cookie’

The examples show the following types of substitution: in (1a) a regular demonstrative pronoun is substituted for an R-pronoun;5 (1b) shows a personsubstitution — a reduced 3rd person personal pronoun is used for a 2nd person pronoun; (2a) shows a gender-substitution — a neuter personal pronoun is used for a masculine pronoun; and (2b) shows a case-substitution — an accusative case is used instead of a nominative case. Problems at the stage of phonological encoding would result in the misarticulation of pronouns (phonematic paraphasias). I did not find clear examples of such misarticulations. I conclude that pronouns are not impaired as a category in the spontaneous speech of the four patients, and no specific stages of pronoun access are found to be disturbed. The general pattern in the realization of pronouns is this: a minority of pronouns are omitted (relatively low omission rates in three of the patients) while pronouns that are realized are usually correct (hardly any substitution errors). In the following sections I investigate the relevance of different syntactic contexts (presence of tense, grammatical status and linear position) for the realization of pronouns. As a starting point, I use the finding that subject pronouns are usually missing from untensed clauses (Friedmann and Grodzinsky 1997, Cahana-Amitay 1997, Kolk and Heeschen 1992). 3.2 Pronoun omission in untensed utterances 3.2.1 Empirical results Two of the patients (HB and GS) show an almost complete absence of subject pronouns in untensed utterances, as shown in Table 2. Patient MU also shows many subject pronoun omissions, but at the same time, subject pronouns are


Table 2.Omission of subject and object pronouns in untensed utterances subjects


objects

present

omitted

% omission

present

omitted

% omission

1 0 7 2

2 20 18 165

67% 100% 72% 99%

0 0 1 7

0 0 2 24

– – 67% 77%

sometimes present in his untensed utterances. Patient ME produces only very few untensed utterances, and therefore no conclusions about subject pronouns can be drawn from his speech. To demonstrate that subject pronoun omission is characteristic of untensed utterances, two types of comparison should be made. First, the omission of subject pronouns should be compared to the omission of other pronouns in untensed utterances. For this comparison, Table 2 includes the omission rates for object pronouns in untensed utterances. Three of the patients, however, show no or only few contexts for object pronouns, and therefore no conclusions can be drawn. Patient GS shows significantly less omission from object pronouns than from subject pronouns (χ2 = 22.846, p < 0.0001), although the object pronoun omission rate is as high as 77%. I will pursue on this issue later on. Second, subject pronoun omission in untensed utterances should be compared to subject pronoun omission in tensed utterances. See Table 3 below for the omission rates in tensed utterances. All patients show more subject omission in untensed utterances than in tensed utterances. This is a statistically significant difference in the speech of patient HB (χ2 = 68.54, p < 0.0001), patient MU (χ2 = 11.69, p = 0.0006) and patient GS (χ2 = 141.02, p < 0.0001). The number of untensed utterances in the

Table 3.Omission of subject and object pronouns in tensed utterances subjects


objects

present

omitted

% omission

present

30 144 33 24

7 26 12 4

19% 15% 27% 14%

11 16 3 6

omitted % omission 1 0 3 1

8% – 50% 14%


speech of patient ME is too low to reach a significant result. I conclude that subject pronoun omission is characteristic of untensed clauses. Three questions emerge from this conclusion: (1) how can subject pronoun omission from untensed utterances be explained? (2) how can the relatively high object pronoun omission rate from untensed utterances in the speech of patient GS be explained? (3) how can the presence of a certain number of subject pronouns in untensed utterances in the speech of patient MU be explained? These questions will be dealt with in the remainder of this section. 3.2.2 General account of untensed main clauses in agrammatic speech In this section I develop a general account of untensed main clauses in agrammatic speech, which covers the (non-)realization of subject pronouns. Since tense is not (overtly) realized in such utterances, it is plausible that the Tense projection is defective or not accessible in some way. Friedmann and Grodzinsky (1997) claim that Tense is impaired and the functional tree is pruned (that is, projections higher than Tense are missing). In their view, the impairment of Tense is purely a syntactic problem. As such, it should be an all-ornothing phenomenon. I demonstrate in the next subsection that the four patients in this study show variation with respect to the number of untensed utterances in their speech. Consequently, I will not present a purely linguistic explanation but a psycholinguistic processing account instead. 3.2.2.1 Full syntactic structure. I propose the verbal structure in (3) for Dutch and German, following Chomsky (1993). (3) [cp C [agrsp AgrS [tp T [agrop AgrO [negp Neg [vp V]]]]]]

If the Tense Projection was entirely missing or impaired in agrammatic speech due to a syntactic problem, patients would either produce only tensed utterances (TP being intact) or only untensed utterances (TP being impaired). Table 4 shows that this is not the case with the Dutch and German patients. In the speech of patient ME, HB and MU, the majority of utterances consist of tensed utterances, so I do not have to argue that the syntactic knowledge needed to build a functional tree including a Tense projection is available with these patients. The speech of patient GS shows a majority of untensed utterances. Nevertheless, there are both conceptual and syntactic arguments to assume the presence of TP in her utterances as well.


Table 4.Tensed and untensed utterances


tensed utterances

untensed utterances % untensed

37 168 45 28

3 20 25 167

8% 11% 36% 86%

First, the concept of time (tense) is available to patient GS: she actually produces time adverbials, also in untensed utterances, as shown in the examples in (4). (4) a.

vanavond op bezoek komen tonight on visit come.inf ‘pay a visit tonight’ b. pas geleden nog een cadeautje gehad short.time ago still a present get.part ‘just got a present a short time ago’

The use of time adverbials not only shows the availability of the concept of time, it can also be taken as evidence for the presence of the Tense projection (Bobaljik and Jones 1996). Not only time adverbials but also negation can be taken as evidence for the presence of the functional projection TP (Platzack 1996, Zanuttini 1997). Examples of negation in untensed utterances of patient GS are shown in (5).6 (5) a.

niet de deur voetballen niet not the door play.football.inf not ‘don’t play football (in front of) the door’ b. de hond niet meer blaffen the dog not longer bark.inf ‘the dog no longer barks’

Finally, although patient GS produces a vast majority of untensed utterances, she does produce 28 tensed utterances as well. Two of the tensed utterances contain a verb marked for past tense. 3.2.2.2 Underspecified Tense and discourse interpretation for an empty Tense element. I claim that in untensed main clauses Tense is present but underspecified. This means that Tense is syntactically ‘inactive’: verbs are not marked for


tense and nominative case is not available (cf. Friedmann and Grodzinsky 1997). Underspecification of Tense in agrammatic speech is not due to any cognitive or syntactic problem. I assume that patients sometimes do not overtly realize Tense in order to reduce the processing load of an utterance (cf. Kolk et al. 1990, Bastiaanse 1995, Avrutin 1999). The underspecified Tense is licensed by an empty element that receives discourse interpretation. Empty Tense does not have an inherent present/past feature, but is interpreted within the discourse. Therefore, empty Tense should not be restricted to a specific context, but show up in both past and present contexts. This can easily be demonstrated in the speech of patient GS. In (6a) a past tense context is introduced by the interviewer’s question. While telling about what happened in the past, GS uses untensed utterances (6b). Although there is no tense marking on the verb, the use of past participles in (6c) clearly indicates that the event happened in the past. It is already a long time ago, but can you tell me once again how your disease started? b. op bed nog ’t haar even doen on bed still the hair just do.inf ‘just wanted to do the hair while sitting on the bed’ en op bed zitten en wachten op hulp and on bed sit.inf and wait.inf for help ‘and sit down on bed and wait for help’ c. maatschappelijk werkster De Veste social worker De Veste me gevonden me find.part ‘social worker De Veste (has) found me’ de dokter gebeld the doctor call.part ‘(has) called the doctor’

(6) a.

In (7) a fragment from the ANTAT test is presented. The scenario in (7a) introduces a present tense context. Patient GS again produces untensed main clauses, as shown in (7b). You have to arrange dinner for a wedding. Right now you are in a restaurant. I am the owner. What do you say to me? b. voor twintig personen reserveren for twenty people reserve.inf ‘reserve for twenty people’

(7) a.


een hotel een restaurant eten met z’n allen a hotel a restaurant eat.inf with his all ‘a hotel a restaurant eat with all of us’

The examples in (6) and (7) show that the empty Tense occurs in both past and present tense contexts, as predicted by the discourse interpretation analysis. I follow Rizzi (1994) in assuming that discourse interpretation is only available for elements in the highest projection, which are not potentially governed by a clause-internal antecedent. The account of untensed main clauses in agrammatic speech which I develop here explains the following facts: (1) patients can differ in the number of untensed utterances they produce, since underspecification of Tense is an attempt to reduce the processing load of the utterance, and is not due to a specific syntactic problem regarding Tense; (2) the omission of subject pronouns in untensed utterances naturally follows the fact that an underspecified Tense, being syntactically ‘inactive’, cannot assign nominative case. Two observations from Table 2 still require an explanation: (1) the omission of object pronouns in the untensed utterances of patient GS; (2) the occasional presence of subject pronouns in untensed utterances, which is most prominently found in the speech of patient MU. 3.2.3 Omission of object pronouns in untensed main clauses I offer an explanation of object pronoun omission which is in line with the general account of untensed main clauses in agrammatic speech: object pronoun omission results from a further underspecification of the functional structure of the clause, namely underspecification of AgrOP. In an underspecified AgrOP, no accusative case is available, and hence, objects are missing. Underspecification of AgrOP is another way of reducing the processing load of the utterance, and as such, not a direct syntactic consequence of the underspecification of Tense. This means that almost all untensed utterances will include subject pronoun omission, (see the next section for a different pattern); object pronoun omission is “optional” though. This explains the significant difference between subject and object pronoun omission in untensed clauses in the speech of patient GS: subject pronoun omission occurs in 99% of the untensed utterances while object pronoun omission occurs in 77% of the untensed utterances (χ2 = 22.846, p < 0.0001).


3.2.4 Occasional presence of subject pronouns in untensed utterances In the previous section, I stated that no structural nominative case is available in untensed utterances. I will show, however, that grammar allows for default case in dislocated positions, also in untensed utterances. 3.2.4.1 Default nominative case. In untensed main clauses, default nominative subjects are found in commands and exclamations, as shown for Dutch in (8a) and (8b) respectively. (8) a.

niemand bewegen nobody move.inf ‘nobody move’ b. zij volgend jaar een boek schrijven? she next year a book write.inf dat geloof ik niet that believe I not ‘She will write a book next year? I don’t believe that.’

In addition, default case is found in dislocated position with tensed sentences, see example (9). (9) a.

die verraadster, ze heeft hem aangegeven bij de politie the traitor.fem she has him reported to the police ‘That traitor, she has reported him to the police.’ b. die verrader, ze heeft hem aangegeven bij de politie the traitor.masc she has him reported to the police ‘That traitor, she has reported him to the police.’

In (9a) the dislocated element is linked to subject position, while in (9b) the dislocated element is linked to object position. In Dutch, as opposed to German, morphological Case is not overtly expressed on determiners and nouns. Therefore, it is not clear what Case the dislocated element in the Dutch examples in (9) bears. The German example in (10) is conclusive in this respect: the dislocated argument is overtly marked for nominative Case. (10) der Hans, den mag ich the.nom Hans he.acc like I.nom ‘Hans, I like him’

The question arises what position allows for default nominative marking? It is not the subject position in untensed clauses; this would include embedded


clauses — which do not allow for default case — and exclude the dislocated position as in examples (9) and (10) — which do allow for default case. It might be the dislocated position; this would correctly rule out embedded clauses, but it needs to be demonstrated that subjects in untensed main clauses are in the dislocated position. An argument for the dislocated position of subjects in untensed clauses is the impossibility of topicalization. While adverbs can be topicalized in a Dutch tensed clause in (11a), this is not true for the untensed clause in (11b). (11) a.

volgend jaar schrijft zij een boek next year write.3sg she a book ‘next year she writes a book’ b. *volgend jaar zij een boek schrijven? next year she a book write.inf ‘next year she write a book?’

The impossibility of topicalization in (11b) follows the assumption that subjects in untensed main clauses are in the dislocated position, since no element can precede the dislocated position. I conclude that default nominative case is available for dislocated elements, including dislocated subjects in untensed main clauses. We now return to the speech of the patients. 3.2.4.2 Nominative subjects in the untensed utterances of patient MU. Patient MU produces seven subjects in untensed clauses. No structural nominative is available in untensed clauses, but a default nominative can be assigned in dislocated position. The question is whether the subjects in untensed main clauses in the speech of patient MU are in dislocated position. Three of the utterances have the following structure: Subject pronoun – object – nonfinite verb; this structure does not include sufficient elements to show the position of the subject. Another utterance includes an adverbial adjunct intervening between the subject and the verb, see (12a). This word order demonstrates that the subject is outside VP, not necessarily in dislocated position, though. The untensed clauses in (12b) through (12d) clearly show dislocated subjects. (12) a.

ich so runtergeschmissen I like.this down.throw.part ‘I thrown down like this’ b. fünfzehn Jahre Bus gefahren ich fifteen years bus drive.part I ‘fifteen years driven bus me’


c.

Schulbus gefahren ich ein Woche schoolbus drive.part I one week ‘driven schoolbus me one week’ d. vier Mal dagewesen hier ich four times there.be.part here I ‘been there four times here me’

3.2.4.3 Default case as a Last Resort strategy. I argued that structural nominative case is not available in untensed main clauses and, hence, subjects are usually missing. I further proposed a mechanism for default case assignment, which explains the occasional presence of subjects in untensed utterances, mainly found in the speech of the German patient MU. The introduction of a default mechanism, however, raises the question why subject pronouns do not always receive a default nominative case, so that they can be freely generated in untensed utterances. This is because default case marking is a Last Resort strategy. Therefore, only 28% of the untensed utterances in the speech of patient MU include a subject pronoun. I assume that the default case mechanism, being a syntactic Last Resort strategy, is only invoked when the presence of a subject is pragmatically required. In the spontaneous speech interview, a subject is usually easily recoverable since the patient is mainly talking about herself/himself. In this case, the presence of the subject is not pragmatically required. In the ANTAT test, most subjects refer to either the speaker or the hearer and are, therefore, easily recoverable as well. This explains the limited use of default subjects. In a picture description the presence of a subject might be pragmatically required. Especially when pictures are presented in isolation, there is no verbal context. Moreover, to arrive at a complete description of a picture, all participants (including the subject) should be mentioned. The performance of patient GS in the AAT picture description task clearly contrasts with her results from spontaneous speech (no data from this task are available for the other patients). She produces five untensed main clauses in this task, four of which contain a subject. The status of the initial argument in the fifth untensed clause is ambiguous between a subject and an indirect object. See an example of an untensed utterance from the picture description task in (13).7 (13) target GS

de man leest de krant ‘the man is reading the newspaper’ de krant lezen the newspaper read.inf


de man de krant lezen the man the newspaper read.inf ‘the man reads the newspaper’

Notice that GS produces the subject in (13) by self-correction. At first, she produces a regular untensed clause without a subject. This clause is grammatically correct: the subject is missing due to the unavailability of nominative Case. However, this is pragmatically incorrect, since a non-recoverable element is missing. Patient GS adds a subject in the second instance. We might interpret this ‘second instance subject’ as not belonging to the basic structure, but actually being adjoined to it in a dislocated position. 3.2.5 Summary In this section, I showed that subject pronouns are almost completely missing in untensed utterances in the speech of the Dutch and German patients. This sharply contrasts to object pronouns in untensed utterances and subject pronouns in tensed utterances, which both show a much lower omission rate. Patients do not show much variation in the subject pronoun omission rate in untensed utterances. I therefore interpret this type of omission as a direct consequence of the syntactic structure involved. Patients do vary, however, in the number of untensed utterances they produce: the use of untensed utterances is not due to a specific syntactic problem. I proposed that untensed main clauses in agrammatic speech involve an underspecified Tense. Underspecification of the Tense node is an attempt to reduce the processing load of the utterance. No structural nominative Case is available with an underspecified Tense and therefore subject pronouns are usually missing. Occasionally, subject pronouns receive a default nominative case in dislocated position, which is assigned as a Last Resort. A further underspecification of AgrO leads to object pronoun omission. The explanation for pronoun omission in untensed utterances cannot be extended to tensed utterances. Tensed utterances include a syntactically ‘active’ Tense projection that assigns nominative case to a subject pronoun. Consequently, no difference is found between subject and object pronoun omission in tensed utterances (see Table 3). In the next section I investigate whether sentence position plays a role in the realization of pronouns in tensed utterances.


3.3 Pronoun omission in tensed utterances 3.3.1 Empirical results Table 3 on pronoun omission in tensed utterances only includes subject and object pronouns. In Table 5, I compare pronouns in first sentence position to pronouns in non-first position. This table includes all types of pronouns that are allowed in first position: subjects, object and indirect object pronouns, R-pronouns that are the complement of a preposition and pronominal adjuncts (mainly place adjuncts), for example daar ‘there’. Table 5.Pronoun omission from first and non-first position in tensed utterances first position


non-first position

present

omitted

% omission

present omitted

16 95 21 20

6 24 12 5

27% 20% 36% 20%

33 89 14 10

2 10 2 0

% omission 6% 9% 13% –

All patients show more pronoun omission from first position than from non-first position. This is a significant difference for patient ME (Fisher’s Exact test, p = 0.0449). It is a clear trend in the speech of patient HB (χ2 = 3.43, p = 0.060). Patient GS only shows omission from first position, although the comparison does not reach a level of significance. Table 5 shows that the pronoun omission rate is relatively low in tensed utterances, both in first and non-first position. First sentence position does not have a syntactic property that ‘prohibits’ the realization of pronouns (as is the case for subject position in untensed utterances). I would rather talk about a preference for pronoun omission from first position as compared to non-first position in the speech of the patients. Such a preference might be related to the normal grammar of Germanic languages. In the next section, I discuss Topic Drop and Diary Drop, both normal grammar phenomena of Dutch and German. Thereafter, I come back to the agrammatic speech and show that pronoun omission from first position in the patients’ speech can be interpreted as regular instances of Topic Drop and Diary Drop. 3.3.2 Topic Drop and Diary Drop in normal language Pronoun omission is a well-known (spoken language) phenomenon in Germanic


languages (Cardinaletti 1990, Haegeman 1990, Rizzi 1994, Huang 1984). I discuss the grammatical status of omitted pronouns and the type of pronoun. 3.3.2.1 Grammatical status of missing pronouns. Both subject and object pronouns can be missing from first position (14c/d, 15c/d) but not from non-first position (14e/f, 15e/f). Dutch examples are given in (14); the corresponding German examples are in (15). (14) a.

b. c. d. e. f. (15) a.

b. c. d. e. f.

ik heb hem gekend I have he.acc know.part ‘I have known him’ hem heb ik gekend [e] heb hem gekend [e] heb ik gekend *ik heb [e] gekend *hem heb [e] gekend ich hab ihn gekannt I.nom have he.acc know.part ‘I have known him’ ihn hab ich gekannt [e] hab ihn gekannt [e] hab ich gekannt *ich hab [e] gekannt *ihn hab [e] gekannt

In addition to subject and object pronouns, pronouns that are the complement of a preposition can be omitted if they are in first position. Such omission concerns R-pronouns, a special class of pronouns that can be moved from the complement position to first sentence position.8 Dutch examples are given in (16): (16a) shows an R-pronoun in first position, (16b) shows omission from that position and (16c) shows that omission from non-first position is ungrammatical. Corresponding German examples are in (17). (16) a.

daar heb ik niet mee gespeeld that have I not with play.part ‘I did not play with that’ b. [e] heb ik niet mee gespeeld c. *ik heb [e] niet mee gespeeld


(17) a.

da habe ich nicht mit gespielt that have I.nom not with play.part ‘I did not play with that.’ b. [e] habe ich nicht mit gespielt c. *ich habe [e] nicht mit gespielt

3.3.2.2 Types of missing pronouns in Dutch. For Dutch, Jansen (1981) investigated what types of pronouns can be missing in normal spontaneous speech. He distinguishes personal pronouns from demonstrative pronouns. Personal pronouns can only be omitted if they are subjects. Usually, the 1st person singular ik ‘I’ is missing. Very marginally, omission of the 2nd person singular je/jij ‘you’, the polite form u ‘you’, the non-referential 3rd person pronoun het ‘it’ and the 1st person plural we ‘we’ is found. Omission of a personal pronoun object is excluded, as shown in (18). (18) *[mij] nodigde hij niet uit voor het diner [me invited he not ptcl for the dinner ‘he didn’t invite (me) for dinner’

Notice that I gave an example of object omission in (14)/(15). The (a)-sentences include a personal pronoun object, hem and ihn ‘him’ respectively. The (d)sentences show that the object pronoun can be missing from first position. If you compare the (d)-examples to the (a)-examples one might think that a personal pronoun is missing. It is more plausible, however, that the missing object is a demonstrative pronoun, as in (19a) and (19b). (19) a.

(die) heb ik gekend ‘(that one) did I know’ b. (den) hab ich gekannt ‘(that one) did I know’

The ambiguity with respect to the missing pronoun only arises in the case of an animate referent. If the referent is inanimate, you cannot put a 3rd person pronoun into first position. If the referent is animate, both a personal and a demonstrative pronoun are possible in first position.9 I don’t know of any linguistic diagnostic to determine what type of pronoun is missing in the latter case, which shows up in(14d)/(15d)/(19). Nevertheless, the intuition of native speakers of Dutch is clear: I asked ten native speakers, who all claim that in (14d)/(19a) the demonstrative pronoun die ‘that one’ is missing. Two native speakers of German confirm this intuition for (15d)/(19b). Demonstrative


pronouns that are subjects or moved complements of a preposition (R-pronouns) can also be omitted. 3.3.2.3 Topic Drop and Diary Drop. I arrive at the following description of possible pronoun omission from first position in normal speech. Demonstrative pronouns can be missing as subjects, objects and R-pronouns. I refer to this type of omission as Topic Drop. In addition, first person personal pronouns can be missing as subjects. I refer to this type of omission as Diary Drop (following Haegeman 1990). I assume that in the case of pronoun omission (Topic Drop and Diary Drop), the Determiner Phrase is underspecified. DP contains an empty pronoun, which receives a discourse interpretation. The need of discourse interpretation explains the restriction to first position. Dropped pronouns must be sufficiently salient in the context. Demonstrative pronouns and 1st person personal pronouns meet this criterion.10 3.3.3 Pronoun omission from first position in agrammatic speech I now return to the agrammatic data. In absolute numbers, patient HB shows the most pronoun omissions from first position. I therefore start by looking at his data. An overview of pronoun types and their grammatical function in the case of first position omission is given in Table 6. Table 6.Pronouns omitted from first position in the speech of patient HB subjects

personal pronouns

demonstr. pronouns objects R-pronouns

1st person sing. 1st person plur. 2nd person sing. 3rd person sing.

8 1 1 2 7 0 5

Table 6 shows that twenty of the omitted pronouns in the speech of patient HB are of the type and grammatical function as required for omission in normal speech:11 the 1st person subject pronoun ik ‘I’ (Diary Drop), demonstrative pronoun subjects dat/die ‘that’ (Topic Drop) and R-pronouns (Topic Drop). Two of the omitted pronouns are of a type marginally found in normal speech (Jansen 1981): the 1st person plural subject we ‘we’ and the 2nd person singular subject jij ‘you’. Only two omitted pronouns are of a type not found in normal first position omission — the 3rd person pronoun hij ‘he’. I conclude


that first position pronoun omissions in the speech of patient HB are instances of regular Topic Drop and Diary Drop. Examples are given in (20). (20) a.

[e] ben naar Ramzegerenhuis geweest [e] am to Ramzegerenhuis be.part ‘(I) have been to Ramzegerenhuis’ b. [e] zijn christelijk [e] are Christian ‘(we) are Christians’ c. [e] bent al vijf jaar buurman [e] be.2sg already five years neighbour ‘(you) are (my) neighbour for five years already’ e. [e] gaat aardig goed [e] goes pretty well ‘(that) is going pretty well’ f. [e] gebeuren ongelukken mee [e] happen accidents with ‘accidents happen with (that)’

Recall that the pronoun omission rate is only 20% in first position, see Table 4. Therefore, underspecification of DP is not due to a specific syntactic problem. I assume that the use of an underspecified DP is an attempt to reduce the processing load of the clause. The other Dutch patient, GS, shows a similar pattern. She omits five pronouns from first position and none from non-first position. Three of the missing pronouns are of the type and grammatical function as required for omission in normal speech: two demonstrative pronoun subjects and one demonstrative pronoun object (Topic Drop). One of the omitted pronouns is of a type marginally found in normal speech (Jansen 1981) — a non-referential 3rd person pronoun er ‘there’. Only one omitted pronoun is of a type not found in normal first position omission — the 3rd person pronoun het ‘it’. The pronoun omissions from first position in the speech of patient GS can be interpreted as regular instances of Topic Drop. Her speech does not show cases of Diary Drop (omission of ik ‘I’). If we turn to the German patients, we find a similar pattern. Patient ME has six pronoun omissions from first position. They are of the following types: two 1st person pronoun subjects ich ‘I’, one demonstrative pronoun object das ‘that’ and three 3rd person pronoun subjects sie ‘she’ (two times) and er ‘he’. Patient MU shows twelve pronoun omissions from first position. They are


of the following types: three 1st person pronouns ich, two demonstrative pronoun subjects, two demonstrative pronoun objects, two non-referential pronoun subjects es ‘it’ and two 3rd person pronoun subjects sie und er. One case (21a) is unclear between a 1st and a 3rd person personal pronoun subject. Other German examples, taken from the speech of patient MU, are shown in (21b) through (21e). (21) a.

[e] muss mal Arzt anrufen [e] must.1sg/3sg just doctor call.inf ‘(I) have/(he) has to call the doctor’ b. [e] wollte es mal machen [e] wanted that just do.inf ‘(I) just wanted to do that’ c. [e] is Geschmackssache [e] is matter.of.taste ‘(that) is (a) matter of taste’ d. [e] hat verheiratet [e] had married ‘(she) was married’ e. [e] weisst du wohl [e] know you certainly ‘you certainly know (that)’

3.3.4 Summary In this section I discussed pronoun omission from tensed utterances. I showed that pronouns are more often omitted from first sentence position than from non-first position. Pronoun omissions from first position were demonstrated to be instances of regular Topic Drop and Diary Drop — pronoun omission phenomena found in normal spoken language in Germanic languages. I assume that these omission phenomena involve an underspecified DP that hosts an empty pronoun. The empty pronoun is interpreted in the discourse, and hence, its distribution is restricted to first position.

4. Discussion The investigation into pronouns in the spontaneous speech of two Dutch and two German Broca’s aphasics reveals the following results. First, the majority of pronouns are correctly realized: overall pronoun omission rates are relatively


low, pronoun substitution hardly occurs. Secondly, subject pronouns are generally missing from untensed main clauses. Thirdly, in tensed utterances, there is a preference for pronoun omission from linear first position. Piñango (this volume) argues that patterns of language breakdown tell something about theories of sentence representation and processing, in particular theories on pronoun processing. She demonstrates that two types of information are relevant for the processing of pronouns: (1) purely syntactic information, and (2) non-syntactic information including coreference. These types of information can also be distinguished in pronoun production in spontaneous speech. Syntactic information is involved in morphological marking for number, gender and person features. Coreference (discourse information) plays a crucial role in the realization of pronouns. The Dutch and German agrammatic patients seem to deal differently with the syntactic and non-syntactic information in pronoun production. Roughly speaking, the patients do not show evidence for a problem with syntactic information, but they rely on non-syntactic information more often than non-aphasic speakers do. A problem in the syntactic information of pronouns — a disturbance at the level of grammatical encoding — would lead to substitution errors. Such errors hardly occur in the speech of my patients. We have to be cautious, however, in drawing conclusions on grammatical substitution on the basis of spontaneous speech alone. Hofstede and Kolk (1994) show that Dutch and German patients who frequently omit freestanding grammatical morphemes in their spontaneous speech make less omission errors but significantly more substitution errors with the same morphemes in a picture description task. I cannot exclude the possibility that my patients would show (some) substitution errors in a specific task on pronoun realization. The reliance on non-syntactic discourse information is evident in the pattern of pronoun omission in tensed utterances in the Dutch and German patients’ speech. The patients show a clear preference for the omission of topic pronouns from first position. Such omission depends on the referential properties of pronouns. The reliance on non-syntactic discourse information is a general pattern in the agrammatic speech of the four patients discussed in this paper: it is not only evident in the pattern of pronoun omission in tensed utterances, but it also underlies the use of untensed utterances. The use of untensed utterances, in turn, results in subject pronoun omission. The overall account of Tense omission (the use of untensed utterances) and Pronoun omission (omission of pronouns from tensed utterances) runs as follows. The Tense Phrase and the


Determiner Phrase (which hosts pronouns) both have a referential function. TP establishes the temporal reference of the utterance, while DP establishes the nominal reference of phrases within the utterance. Being referential projections, the interpretation of TP and DP can be fixed in syntax but also in discourse. TP and DP might be left syntactically underspecified, in which case the empty functional head has to be interpreted in discourse. The reference is then not provided by the speaker (the agrammatic patient) but should be determined by the hearer. The advantage for the patient is that she/he can leave the referential element syntactically unexpressed. I assume that syntactically unexpressed elements require less processing capacity. Underspecification of a referential projection is an attempt to reduce the processing load of the utterance. If this explanation is on the right track, it is expected that also other elements can be omitted as a result of underspecification of a referential projection. De Roo (1999) demonstrates this to be true for determiners in Dutch agrammatic speech. In this paper I discussed two types of pronoun omission — first, the omission of subject pronouns in untensed utterances. I argued that such omission is a direct consequence of the syntactic structure of untensed clauses, which lack structural nominative case for a subject pronoun. The use of untensed utterances, however, does not reflect a purely syntactic problem. It results from underspecification of the Tense Phrase as an attempt to reduce the processing load of the utterance. The same is true for pronoun omission from first position in tensed utterances, which results from underspecification of the Determiner Phrase. I developed a psycholinguistic account of agrammatic speech: there is no lack of syntactic knowledge but there is a limitation in the capacity to process syntactic information. Three arguments support my psycholinguistic account. First, the pronoun omission rates are relatively low. This means that pronouns are not disturbed as a category. The number of untensed utterances, which involve subject pronoun omission, is also relatively low and shows important variation between the patients. This means that the syntactic knowledge to generate tensed utterances should be available. Secondly, task variation was observed in the speech of patient GS. She leaves out almost all subject pronouns in untensed utterances in spontaneous speech, but not in a picture description task. Thirdly, the omission phenomena observed in agrammatic speech are part of the normal grammar of Dutch and German. Pronoun omission form first position in tensed utterances (Topic Drop and Diary Drop) is frequently discussed in linguistic literature on Germanic languages (see Section 3 for references). Untensed main clauses in normal language are discussed, for example, in De Roo (1999) for Dutch and Lasser


(1997) for German. A control group of Dutch speakers who participated in the ANTAT test also actually produced first position pronoun omissions in tensed utterances (22) and untensed clauses including subject pronoun omission (23). (22) a.

[e] ben helaas verhinderd [e] am unfortunately unable.to.come.part ‘unfortunately (I) am not able to come’ b. [e] kun je alleen maar kurken mee uithalen [e] can.2sg you only just corks with remove.inf ‘you can only remove corks with (that)’

(23) a.

dit vaststikken this stitch.inf ‘stitch this’ b. een groentemandje bestellen voor een kennis a vegetable.basket order.inf for a friend ‘order a vegetable basket for a friend’

The Dutch control subjects show less Topic Drop and Diary Drop constructions in the ANTAT test than the patients do: these omission phenomena occur in less than 2% of the tensed utterances in the control speech (De Roo 1999). Jansen (1981) found omission rates between 1% and 13% in a corpus of normal Dutch spontaneous speech. I have no quantitative data on Topic Drop and Diary Drop in normal German. The Dutch control subjects also produce less untensed main clauses in the ANTAT test than the patients do: the occurrence of untensed utterances is within the range [1–8%] in the control speech (De Roo 1999). Lasser (1997) found 3% of untensed main clauses on average in a German spontaneous speech corpus.12 I claim that agrammatic patients overuse grammatical options of the normal grammar (De Roo 2000). This Overuse Hypothesis resembles the Adaptation Theory (e.g. Kolk et al. 1990, Hofstede 1992) in that agrammatic patients overuse normal elliptical (untensed) structures. The Overuse Hypothesis incorporates the finding that normal omission phenomena are also found in tensed utterances in agrammatic speech. Notice that pronoun omissions in the speech of the agrammatic patients are not restricted to the overuse of grammatical options (Topic Drop or Diary Drop). The patients show some ‘illegal’ omissions as well, that is, omissions from non-first position. It is important to note that the control subjects also show illegal omissions, see the examples in (24).


(24) a.

zou [e] zo goed willen zijn would [e] so nice want.inf be.inf ‘would (you) be so kind’ b. dat [e] een fout ingeslopen is that [e] an error in.creep.part is ‘that an error has crept in (there)’

I consider such examples as speech errors. It is quite plausible that agrammatic patients produce more speech errors than normal speakers do.

Notes 1. For agrammatic comprehension, the same subdivision can be made. Agrammatics do not reliably assign semantic roles consistent with syntactic structure. Especially problematic are the following construction types: passive sentences (as opposed to active sentences), object relatives (as opposed to subject relatives), object cleft sentences (as opposed to subject cleft sentences). The problematic constructions involve the syntactic movement of a constituent. There is an asymmetry between constructions in which the object is moved (poor comprehension) and constructions in which the subject is moved (comprehension above chance). The purely syntactic theory of agrammatism explains agrammatic comprehension in terms of the inability to parse sentences, i.e. to construct syntactic representations of the input sentences. For example, Grodzinsky (2000) assumes that agrammatic representations lack traces. The subject-object asymmetry is due to the fact that object movement results in noncanonical word order. Grodzinsky presents additional evidence for his Trace Deletion Hypothesis from grammaticality judgement experiments and real-time processing experiments. Linebarger et al. (1983), however, demonstrated that agrammatic patients who showed impaired comprehension of object wh-questions did not show problems with the same syntactic structures in a grammaticality judgement experiment. Such results provide strong counter-evidence to a purely syntactic explanation of the comprehension deficit in Broca’s aphasia. Another asymmetry observed in agrammatic comprehension concerns reflexives and pronouns: performance on reflexives is consistently better than on pronouns (e.g. Piñango, this volume). Piñango presents a psycholinguistic explanation for the specific agrammatic pattern in the comprehension of pronouns and reflexives based on the assumption of a slowed down syntactic parser. 2. Since this paper is on speech production only, I do not give an overview of the literature on agrammatic comprehension. I just mention a few variables which were found to be relevant for the comprehension of pronouns, because these variables might have an equivalent in agrammatic production. Hickok and Avrutin (1996) assume that ‘discourse linking’ is a relevant variable in agrammatic comprehension. They show that patients have problems with discourse-linked which-questions but not with not discourse-linked who-questions (cf. Grodzinsky 2000, Frazier and McNamara 1995 for a different explanation). Pronominal type is a relevant variable in comprehension studies of e.g. Grodzinsky et al.


(1993) and Piñango (this volume): personal pronouns are relatively impaired in agrammatic comprehension. Jarema and Friederici (1994) demonstrate that agrammatic comprehension of French personal pronouns which carry grammatical gender information is relatively impaired as compared to personal pronouns which carry natural gender information. 3. Ouhalla presents a theory within the framework of the Minimalist Program (Chomsky 1993). He puts forward the hypothesis that functional categories (which largely overlap with closed-class words) are stored in a UG lexicon, distinguished from the grammatical lexicon, which contains the lexical representations of words. The impairment underlying agrammatism affects the ability to access the categories of the UG lexicon. Therefore, pronouns, which are generated in the functional projection DP, are missing from agrammatic speech. 4. The data of the Dutch patients GS and HB were made available to me by Leo Blomert. The data of the German patient MU were made available to me by Ben Hofstede. The data of the German patient ME are published in Stark and Dressler (1990). 5. The use of R-pronouns will be discussed in more detail in Section 3.3. 6. Some of the examples show double negation, which is not correct in Dutch. However, this does not reflect a specific problem with negation, since there are also double adjuncts, as shown in (a)–(c). (a) misschien een volgende misschien (b) niet meer badminton meer (c) al een paar maanden al op bed

‘maybe a next one maybe’ ‘not anymore badminton anymore’ ‘already a few months already in bed’

Both adjuncts or negations are in a possible position, however, they should occur only once in the utterance. 7. Notice that the subjects in the untensed utterances in the picture description test are full DPs and no pronouns. This is not important for the demonstration that default case is a Last Resort strategy, since full DPs and pronouns both need nominative case if they function as a subject. In the interview and the ANTAT test, GS also produces five DP subjects in untensed utterances: four of these are outside VP, but none can be demonstrated to be in dislocated position. 8. In Dutch the relevant pronouns end in -r; in German an -r is only invoked when the pronoun is left-adjoined to the preposition and the preposition starts with a vowel. See for example Van Riemsdijk (1978) for a discussion of R-pronouns in Dutch, Müller (this volume) for German. 9. Thanks to Horst Simon for this information. 10. This analysis contrasts with Cardinaletti (1990) and Rizzi (1994) who present a separate account for subject and object omission. They wrongly assume that subject drop can involve pronouns with any person specification. 11. A reviewer suggests that the trend towards omission of the element in first position could also be due to the fact that subjects are omitted more easily and subjects are usually more often in first position. This suggestion might be reinforced by the fact that patient HB does not omit any object pronouns, see Table 3. R-pronouns, however, do show the first-position effect:


five out of ten R-pronouns are omitted in first position against three out of twelve in nonfirst position. With such small numbers, statistical testing does not reach a significant result. 12. The corpus used by Lasser consists of conversations between two parents and their child. It is possible that the adults adapt their speech to that of the child, which might increase the number of untensed main clauses. The percentage mentioned by Lasser falls within the range reported for Dutch, though.

References Avrutin, S. 1999. Development of the Syntax-Discourse Interface. Dordrecht: Kluwer. Bastiaanse, R. 1995. “Broca’s aphasia: A syntactic and/or a morphological disorder? A case study”. Brain and Language 48: 1–32. Blomert, L., Koster, C. and Kean, M.-L. 1995. Amsterdam-Nijmegen Test voor Alledaagsetaalvaardigheid (ANTAT). Lisse: Swets and Zeitlinger. Blumstein, S. E., Goodglass, H., Statlender, S. and Biber, C. 1983. “Comprehension strategies determining reference in aphasia: A study of reflexivization”. Brain and Language 18: 115–127. Bobaljik, J. and Jonas, D. 1996. “Subject positions and the roles of TP”. Linguistic Inquiry 27: 195–236. Bradley, D., Garrett, M. and Zurif, E. 1980. “Syntactic deficits in Broca’s aphasia”. In Biological Studies of Mental Processes, D. Caplan (ed.), 269–286. Cambridge, MA: MIT Press. Cahana-Amitay, D. 1997. Syntactic Aspects of the Production of Verbal Inflection in Aphasia. Ph.D. Diss., Boston University. Cardinaletti, A. 1990. “Subject/object asymmetries in German null-topic constructions and the status of SpecCP”. In Grammar in Progress: GLOW Essays for Henk van Riemsdijk [Studies in Generative Grammar 36], J. Mascaró and M. Nespor (eds), 75–84. Dordrecht: Foris. Chomsky, N. 1993. “A minimalist program for linguistic theory”. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, K. Hale and S. Keyser (eds), 1–52. Cambridge, MA: MIT Press. Frazier, L. and McNamara, P. 1995. “Favor referential representations”. Brain and Language 49: 224–240. Friederici, A. 1985. “Levels of processing and vocabulary types: Evidence from on-line comprehension in normals and agrammatics”. Cognition 19: 133–166. Friedmann, N. and Grodzinsky, Y. 1997. “Tense and agreement in agrammatic production: Pruning the syntactic tree”. Brain and Language 56: 397–425. Goodglass, H. and Kaplan, E. 1972. The Assessment of Aphasia and Related Disorders. Philadelphia: Lea and Febiger. Graetz, P., Bleser, R. De and Willmes, K. 1992. De Akense Afasie Test, Ned. Versie. Lisse: Swets and Zeitlinger. Grodzinsky, Y. 2000. “The neurology of syntax: Language use without Broca’s area”. Behavioral and Brain Sciences 23: 1–71.


Grodzinsky, Y., Wexler, K., Chien, Y.-C. and Marakovitz, S. 1993. “The breakdown of binding relations”. Brain and Language 45: 371–395. Hagiwara, H. 1995. “The breakdown of functional categories and the economy of derivation”. Brain and Language 50: 92–117. Haegeman, L. 1990. “Non-overt subjects in diary contexts”. In Grammar in Progress. GLOW Essays for Henk van Riemsdijk [Studies in Generative Grammar 36], J. Mascaró and M. Nespor (eds), 167–174. Dordrecht: Foris. Hartsuiker, R. and Kolk, H. 1998. “Syntactic facilitation in agrammatic sentence production”. Brain and Language 62: 221–254. Hickok, G. and Avrutin, S. 1996. “Comprehension of wh-questions in two Broca’s aphasics”. Brain and Language 52: 314–327. Hofstede, B. 1992. Agrammatic Speech in Broca’s Aphasia: Strategic Choice for the Elliptical Register. Ph.D. Diss., University of Nijmegen. Hofstede, B. and Kolk, H. 1994. “The effects of task variation on the production of grammatical morphology in Broca’s aphasia”. Brain and Language 46: 278–328. Huang, C.-T. J. 1984. “On the distribution and reference of empty pronouns”. Linguistic Inquiry 15: 531–574. Huber, W., Poeck, K., Weniger, D. and Willmes, K. 1983. Aachener Aphasie Test (AAT). Göttingen: Hogrefe. Jansen, F. 1981. Syntaktische Konstrukties in Gesproken Taal. Ph.D. Diss., University of Leiden. Jarema, G. and Friederici, A. 1994. “Processing articles and pronouns in agrammatic aphasia: Evidence from French”. Brain and Language 46: 683–694. Kohn, S., Cragnolino, A. and Pustejovsky, J. 1997. “Pronoun production in agrammatic speakers: Patterns of use and avoidance in context neutral sentences”. Aphasiology 11: 157–175. Kolk, H. and Heeschen, C. 1990. “Adaptation symptoms and impairment symptoms in Broca’s Aphasia”. Aphasiology 4: 221–231. Kolk, H. and Heeschen, C. 1992. “Agrammatism, paragrammatism and the management of language”. Language and Cognitive Processes 7: 89–129. Kolk, H., Heling, G. and Keyser, A. 1990. “Agrammatism in Dutch: Two case studies”. In Agrammatic Aphasia: A Cross Language Narrative Sourcebook, L. Menn and L. Obler (eds), 179–259. Amsterdam and Philadelphia: John Benjamins. Lasser, I. 1997. Finiteness in Adult and Child German [MPI series in psycholinguistics 8]. Wageningen: Ponsen en Looyen. Linebarger, M. 1995. “Agrammatism as evidence about grammar”. Brain and Language 50: 52–91. Linebarger, M., Schwartz, M. and Saffran, E. 1983. “Sensitivity to grammatical structure in so-called agrammatic aphasics”. Cognition 13: 361–392. Menn, L. and Obler, L. (eds). 1990. Agrammatic Aphasia: A Cross Language Narrative Sourcebook. Amsterdam and Philadelphia: John Benjamins. Müller, G. this volume. “Harmonic alignment and the hierarchy of pronouns in German”. Ouhalla, J. 1993. “Functional categories, agrammatism and language acquisition”. Linguistische Berichte 143: 3–36. Piñango, M. M. this volume. “Cortical reflections of two pronominal relations”.


Platzack, C. 1996. “The initial hypothesis of syntax: A minimalist perspective on language acquisition and attrition”. In Generative Perspectives on Language Acquisition [Language Acquisition and Language Disorders 14], H. Clahsen (ed.), 369–414. Amsterdam and Philadelphia: John Benjamins. Riemsdijk, H. van. 1978. A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases. Dordrecht: Foris. Rizzi, L. 1994. “Early null subjects and root null subjects”. In Language Acquisition Studies in Generative Grammar, T. Hoekstra and B. Schwartz (eds), 151–176. Amsterdam and Philadelphia: John Benjamins. Rochon, E., Saffran, E., Berndt, R. and Schwartz, F. 2000. “Quantitative analysis of aphasic sentence production: Further development and new data”. Brain and Language 72: 193–218. Roo, E. de 1999. Agrammatic Grammar: Functional Categories in Agrammatic Speech. Ph.D. Diss., University of Leiden. Roo, E. de 2000. “Agrammatic speech production: The Overuse Hypothesis”. Brain and Language 74: 555–558. Saffran, E., Berndt, R. and Schwartz, M. 1989. “The quantitative analysis of agrammatic production: Procedure and data”. Brain and Language 37: 440–479. Schmitt, B., Meyer, A. and Levelt, W. 1999. “Lexical access in the production of pronouns”. Cognition 69: 313–335. Stark, J., and Dressler, W. U. 1990. “Agrammatism in German”. In Agrammatic Aphasia: A Cross Language Narrative Sourcebook, L. Menn and L. Obler (eds), 281–429. Amsterdam and Philadelphia: John Benjamins. Wechsler, D. 1949. Wechsler Intelligence Scale for Children. New York: Psychological Corporation. Zanuttini, R. 1997. Negation and Clausal Structure: A Comparative Study of Romance Languages [Oxford Studies in Comparative Syntax]. New York: Oxford University Press.

Language index A Albanian, 88, 94 Arabic (Egyptian), 58, 146–149, 152, 154, 157 n.9, 162 Armenian, 94 B Bavarian See German: Bavarian Berik, 30f Bulgarian, 199 C Catalan, 90, 94, 95, 98 Chinese, 88 D Dutch, 48f, 253–284 West Flemish, 104 n.17 E English, 3, 6, 17, 20 n.8, 49, 50, 86, 88, 94, 103 n.2, 104 n.11, 145, 150, 152f, 176f, 179 n.20, 189, 190, 196, 199, 218 F Fa¯rsı¯ See Persian Fijian (Boumaa), 28–30 French, 96, 104 n.12, 184, 188, 194 G German, 4–6, 8, 20 n.1, 50, 52, 86, 95, 96, 103 n.2, 104 n.10, 184, 193, 195, 196, 205–231, 253–284 Bavarian, 88, 89, 94, 96, 98, 100, 101, 103 n.3 Old High German, 103 Swiss German, 103 n.5

Greek, 88, 94, 96, 159 n.24, 199 Guaraní, 36 n.6 H Haitian Creole, 104 n.12 Hebrew Biblical, 161, 178 n.8, n.11, 179 n.17 Modern, 148f, 178 n.3, n.12, 255f Qumran, 16, 161–181 Hindi, 88, 92, 94, 95, 218 Hungarian, 88, 89, 91f, 94, 95, 218 I Ilocano, 60 n.4 Italian, 87, 89, 94, 95, 97, 98, 100, 101, 102, 103 n.2, 184, 200 n.1, n.6 J Japanese, 183–190, 191, 198, 199, 200 n.8, 201 n.18 K Kalihna, 36 n.2 Koasati, 37 n.6 Korean, 88, 93, 94, 99 Kunimaipa, 46 Kwakiutl, 32f L Lardil, 3 Latin, 46 M Malayalam, 87, 88, 93, 94, 96, 98, 103 n.8 Maxakalí, 32f N Nama, 34, 59

286 Language index

O Ojibwe, 34f P Persian, 7 Polish, 89, 90f, 94, 95, 103 n.5 Portuguese Brazilian, 199 European, 201 n.19 R Russian, 87 S Sanuma, 49, 50 Serbo-Croatian, 91, 94, 200 n.8 Sierra Popoluca, 60 n.4 Sinhala, 13, 63–83 Slovene, 89, 91, 94, 95, 103 n.6 Spanish, 36 n.1, 49, 58, 87, 89f, 94, 95, 96, 194, 200 n.6 Quiteño, 199

T Tamil, 63 Thai, 43, 185–190, 198, 200 n.2, n.3, 201 n.18 Turano, 88 Turkish, 94, 96 W Wandamen, 36 n.1 Windesi, 36 n.1 Y Yokuts, 34

Subject index A address (pronominal versus nominal terms) See pronouns: pronominal use of nouns agrammatism, 253–256, 265f, 270, 274, 277–280 See also Broca’s aphasia agreement, 5f, 9, 17, 43, 55f, 77, 170–172, 193, 195 agreement features See features: checking direct versus inverse, 34f negative concord See negation: negative concord anaphora, 4–7, 15, 69, 70, 113–115, 140 cross-sentential, 114f deictic vs. anaphoric pronouns, 110f, 122 zero anaphora, 65–67, 70, 74f animacy, 13f, 20 n.8, 65, 70, 76f, 205, 211 aphasia See Broca’s aphasia; Wernicke’s aphasia argument structure, 173, 236 See also b-roles awareness of linguistic elements, 42, 56f B Bach-Peter sentences, 112 binding, 6 Binding Theory, 10, 237–239, 244, 248 c-command, 237 coreference and binding, 113, 121, 237 See also coindexation versus coreference reflexives versus pronouns, 6, 246f Broca’s aphasia, 11f, 18f, 233–237, 247f, 249 n.2, n.3, 253, 280 n.1

limitation of processing resources, 19, 242f, 253 See also underspecification: as a means to reduce processing load Slow-Syntax Hypothesis, 18, 236f, 245–248, 280 n.1 See also agrammatism; omission of pronouns in speech C c-command See binding: c-command clitics, 175f, 177, 198, 200 n.6, 205, 224 coindexation versus coreference, 18f, 119, 237–239, 242 constraint alignment, 207f, 211 copula, 167f, 177, 178 n.9 crossover, 149–157 strong, 149–152 weak, 149f, 152, 158 n.14 D deixis, 121, 189 See also anaphora: deictic vs. anaphoric pronouns; pronouns: deictic pronouns demonstrative pronouns, 69, 165f, 224 determiners intransitive See pronouns: as (in-)transitive determiners semantic function, 196 Diary Drop, 271, 274–276, 278f diglossia, 63, 67f, 70 discourse-pragmatic information, 19, 277f for the interpretation of empty elements, 264–266, 274, 276–278 for the interpretation of pronouns, 3–6, 9, 65–67, 77–79, 277f formality of situation, 185

288 Subject index

restrictions on resumptive pronoun use, 153f sociolinguistic factors in referent identification, 67f dislocation (left- / right-), 167, 169–172, 267f, 159 n.24 do-support, 177, 218 D-pronouns See pronouns: N-pronouns and D-pronouns dynamic semantics, 115–117, 123 Dynamic Syntax, 137–149 analysis of 3rd person pronouns, 143 case, 143 computational action, 142f lexical action, 142, 143 linked structures, 145f, 148, 151f partial trees, 140–142 pragmatic action, 142 structural underspecification, 138–140 tree growth, 15, 142–144 unfixed nodes, 15, 140, 148, 151f well-formed strings, 144, 154 E edge of vP, 210f elementary nouns, 197f See also grammatical nouns EPP-feature, 212 E-type pronouns, 117, 120, 123, 132 exclusive versus inclusive See inclusive Explicitness Hierarchy See paradigms (pronominal): Explicitness Hierarchy F faithfulness constraints, 216, 221f, 225 feature geometry, 12, 23–39, 57 morpho-syntactic, 23, 25–27, 35 phonological, 23 feature hierarchy, 28, 189 features, 57 acquisition See paradigms (pronominal): acquisition checking, 99, 172, 179 n.16 Feature Condition (FC), 223f

f-features, 193 formal, 187 grammatical, 2–5, 8, 9, 10, 35, 188, 200 n.3 identifying conceptual distinctions, 3f morpho-semantic See features: grammatical morpho-syntactic See features: grammatical [Neg] feature See negation: [Neg] feature formality See discourse-pragmatic information: formality of situation functional pronouns, 118, 121 G gender, 3, 13, 25f, 36 n.1, 44, 58f, 65, 69f, 193–196 See also animacy grammatical nouns, 183 See also elementary nouns grammaticalisation, 42, 56f H Harmonic Alignment, 206–208, 211 homophony within pronominal paradigms See paradigms (pronominal): homophony; neutralisation honorific terms, 65f, 68, 70, 72–74, 77, 81 See also respect I inclusive, 31, 34f, 41–62 augmented, 45, 54, 58, 61 n.7 minimal, 36 n.2, 44, 45, 54, 58, 61 n. 7 See also person indefinite expressions, 85–107 negative indefinites, 14, 86f, 89f, 92, 94–102, 103 n.2, 104 n.12 NPI-indefinites See Negative Polarity Items (NPI) PPI-indefinites See Positive Polarity Items (PPI) weak indefinites, 14, 85, 86, 99

Subject index 289

wh-indefinites See wh-expressions: indefinite interrogative pronouns See wh-expressions; wh-questions K kinship terms See pronouns: pronominal use of nouns L language production (pronouns), 253–284 grammatical encoding, 254, 260, 277 lexical access, 254f, 260 message encoding, 254f, 260, phonological encoding, 254f, 260f See also omission of pronouns in speech Last Resort Strategy, 16, 163, 176f, 212, 218 default case, 267–270 left dislocation See dislocation (left- / right-) lexical access See language production (pronouns): lexical access long-distance dependencies, 139, 151, 156, 249 n.4 M markedness, 27, 33, 226 n.7 N NC-languages See negation: NC- versus Non-NC-languages negation 85–107 direct negation, 86, 89, 90, 92, 95 NC- versus Non-NC-languages, 14, 88–93 negative concord, 88, 97, 98, 99, 102, 104 n.13 negative quantifiers, 97 NEG-contexts, 86, 94, 96, 102 [Neg] feature, 14, 85, 94, 99, 100, 101, 102 NegP, 98f, 101f

See also indefinite expressions: negative indefinites; Negative Polarity Items Negative Polarity Items (NPI), 86, 95, 102 NPI-contexts, 93, 94, 95, 102 NPI-indefinites, 89, 90, 91 NEG-context See negation neutralisation within pronominal paradigms See paradigms (pronominal): neutralisation; homophony n-indefinite See indefinite expressions: negative indefinites nobody-paradox, 112 nominal sentences See verbless clauses Non-NC-languages See negation: NCversus Non-NC-languages N-pronouns See pronouns: N-pronouns versus D-pronouns Null Topic See Topic Drop number, 3, 12, 13, 23–39, 195 dual, 58 trial, 58 See also person O omission of pronouns in speech, 253–284 object pronouns, 255f, 262f, 266, 270–273, 275f, 278, 281 n.10, n.11 subject pronouns, 154, 159 n.21, 255f, 259, 261–263, 266, 270–279, 281 n.10, n.11 See also Diary Drop, Topic Drop open versus closed classes, 186–188 optimisation local, 214, 225, 228 n.17 standard (global), 214 P paradigms (pronominal), 3, 9f, 12, 13, 19, 41–62 accidental versus systematic homophony, 41, 45–48, 58 acquisition, 25, 28 diagonal homophony, 50 Explicitness Hierarchy, 13, 42, 53–58 horizontal homophony, 50, 52–54, 56

290 Subject index

Ilocano-type, 60 n.4 neutralisation, 10, 13, 14 ‘pure person’ marking, 13, 41f, 53, 56, 58 richness, 57 singular homophony, 48f, 51 vertical homophony, 49–54 paycheque pronouns, 118 person, 3, 12, 13, 23–39, 41–62, 184, 185, 189, 201 n.18 1st person inclusive dual, 44, 36 n.2 1st person inclusive paucal, 29 1st person plural, 31f 2nd person inclusive, 33, 37 n.7 discourse participants, 25f, 31, 32f, 35 paradigm of person markers, 43, 47 See also inclusive politeness See respect See also pronouns: neutral versus derogative use; social norms (relevance for pronoun choice) Position Scale, 211 Positive Polarity Items (PPI), 86, 91, 95, 102 PPI-context, 94 PPI-indefinites, 90, 92 postposition stranding, 216 processing load See Broca’s aphasia; underspecification pro-drop, 57, 188 See also omission of pronouns in speech Pronominal nouns See pronouns: N-pronouns versus D-pronouns pronouns as (in-)transitive determiners, 17, 183, 185, 191f, 197f as semantic variables, 190, 193, 198 See also Dynamic Syntax: unfixed nodes bound variable interpretation, 238f, 250 n.6, n.8 coordination with noun phrases, 205, 219–221 deictic pronouns, 120, 131f different uses, 120–122

in language production See language production (pronouns) lexical versus functional categories, 2 See also open versus closed classes loanwords, 185 N-pronouns and D-pronouns, 190, 192, 200 n.8 neutral versus derogative use, 67, 71f, 75, 79–81, 82 Personal Pronoun Scale, 211 pleonastic pronouns, 162 pronominal use of nouns, 65, 68, 72–76 pronoun movement, 226 n.4, 249 n.4 See also wh-expressions: wh-movement pronouns of laziness, 111, 120, 132 reduced pronouns, 205 reflexives See binding sloppy identity reading, 10, 20 n.5 strong pronouns, 205 substitution as a function of pronouns, 111f unstressed pronouns, 205 weak pronouns, 205 See also anaphora; clitics; demonstrative pronouns; Dynamic Syntax: analysis of 3rd person pronouns; E-type pronouns; functional pronouns; paradigms (pronominal); paycheque pronouns; R-pronouns; relative pronouns; resumptive pronouns; social norms (relevance for pronoun choice); wh-expressions proximity, 70 ‘pure person’ See paradigms (pronominal): ‘pure person’ marking R reference tracking, 63–83 relative clauses, 6f, 144–155 bound variable interpretation, 113, 120 nonrestrictive, 144f restrictive, 145f

Subject index 291

See also relative pronouns; resumptive pronouns; wh-expressions relative pronouns, 145f, 151, 165f See also relative clauses; wh-expressions Relevance Theory, 153, 158 n.20 respect, 3, 20 n.1, 44, 184, 188, 195 See also honorific terms resumptive pronouns, 7, 146–149, 150, 153–155, 159 n.25, 169–172 See also discourse-pragmatic information: restrictions on resumptive pronoun use R-pronouns, 17f, 20 n.8, 177, 205, 214–219, 227 n.12, 228 n.16, 271–274, 281 n.11 S salience analysis of pronouns, 15f, 121, 122–124, 126–132, 133 n.3 and definiteness, 126–128 choice functions, 15, 128 epsilon terms, 128 Scandinavian object shift, 226 n.6 scrambling, 210 Slow-Syntax Hypothesis See Broca’s aphasia social norms (relevance for pronoun choice), 13f, 64, 72–74, 75 Sonority Hierarchy, 208 Syllable Position Prominence, 208 syncretism within pronominal paradigms See paradigms (pronominal): homophony; neutralisation T b-roles, 197, 227 n.9 See also argument structure Topic Drop, 4, 271, 274–276, 278, 279 topicalization, 205, 222–224 U underspecification as a means to reduce processing load, 265f, 270, 275, 277f

lexical, 8 morphological, 26, 33 of AgrOP, 266, 270 of DPs, 274–276, 278 of Tense, 264–266, 270, 278 structural See Dynamic Syntax: structural underspecification See also paradigms (pronominal): homophony; neutralisation Universal Grammar, 12, 23, 28, 187 V variables event-variables, 97, 98, 101 bound variable interpretation of pronouns See pronouns pronouns as semantic variables See pronouns verbless clauses, 162–181 specificational versus predicational interpretation, 163, 175, 176f, 178 n.10, 179 n.19 W Wackernagel Movement, 17, 205, 206, 209–216, 226 n.4, 227 n.11, 228 n.16 Wernicke’s aphasia, 235, 244f, 249 n.4 wh-expressions, 7f, 145, 150, 152, 155, 156, 165f indefinite 85, 87, 99 multiple questions, 228 n.19 wh in situ, 228 n.19 wh-movement, 228 n.16 See also relative clauses; wh-questions wh-questions, 155f See also wh-expressions: multiple questions

In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titles have been published thus far, or are scheduled for publication: 1. KLAPPENBACH, Ruth (1911-1977): Studien zur Modernen Deutschen Lexikographie. Auswahl aus den Lexikographischen Arbeiten von Ruth Klappenbach, erweitert um drei Beiträge von Helene Malige-Klappenbach. 1980. 2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion und Beispielanalyse. 1982. 3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers from the 3rd Groningen Grammar Talks (3e Groninger Grammatikgespräche), Groningen, January 1981. 1983. 4. ABRAHAM, Werner & Sjaak De MEIJ (eds): Topic, Focus and Configurationality. Papers from the 6th Groningen Grammar Talks, Groningen, 1984. 1986. 5. GREWENDORF, Günther and Wolfgang STERNEFELD (eds): Scrambling and Barriers. 1990. 6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences. 1989. 7. ÅFARLI, Tor A.: The Syntax of Norwegian Passive Constructions. 1992. 8. FANSELOW, Gisbert (ed.): The Parametrization of Universal Grammar. 1993. 9. GELDEREN, Elly van: The Rise of Functional Categories. 1993. 10. CINQUE, Guglielmo and Guiliana GIUSTI (eds): Advances in Roumanian Linguistics. 1995. 11. LUTZ, Uli and Jürgen PAFEL (eds): On Extraction and Extraposition in German. 1995. 12. ABRAHAM, W., S. EPSTEIN, H. THRÁINSSON and C.J.W. ZWART (eds): Minimal Ideas. Linguistic studies in the minimalist framework. 1996. 13. ALEXIADOU Artemis and T. Alan HALL (eds): Studies on Universal Grammar and Typological Variation. 1997. 14. ANAGNOSTOPOULOU, Elena, Henk VAN RIEMSDIJK and Frans ZWARTS (eds): Materials on Left Dislocation. 1997. 15. ROHRBACHER, Bernhard Wolfgang: Morphology-Driven Syntax. A theory of V to I raising and pro-drop. 1999. 16. LIU, FENG-HSI: Scope and Specificity. 1997. 17. BEERMAN, Dorothee, David LEBLANC and Henk van RIEMSDIJK (eds): Rightward Movement. 1997. 18. ALEXIADOU, Artemis: Adverb Placement. A case study in antisymmetric syntax. 1997. 19. JOSEFSSON, Gunlög: Minimal Words in a Minimal Syntax. Word formation in Swedish. 1998. 20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Adverbs, pronouns, and clause structure in Romance and Germanic. 1998. 21. KLEIN, Henny: Adverbs of Degree in Dutch and Related Languages. 1998. 22. ALEXIADOU, Artemis and Chris WILDER (eds): Possessors, Predicates and Movement in the Determiner Phrase. 1998. 23. GIANNAKIDOU, Anastasia: Polarity Sensitivity as (Non)Veridical Dependency. 1998. 24. REBUSCHI, Georges and Laurice TULLER (eds): The Grammar of Focus. 1999. 25. FELSER, Claudia: Verbal Complement Clauses. A minimalist study of direct perception constructions. 1999.

26. ACKEMA, Peter: Issues in Morphosyntax. 1999. ° 27. RUZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study. 1999. 28. HERMANS, Ben and Marc van OOSTENDORP (eds): The Derivational Residue in Phonological Optimality Theory. 1999. 29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 1999. 30. BEUKEMA, Frits and Marcel den DIKKEN (eds): Clitic Phenomena in European Languages. 2000. 31. SVENONIUS, Peter (ed.): The Derivation of VO and OV. 2000. 32. ALEXIADOU, Artemis, Paul LAW, André MEINUNGER and Chris WILDER (eds): The Syntax of Relative Clauses. 2000. 33. PUSKÁS, Genoveva: Word Order in Hungarian. The syntax of È-positions. 2000. 34. REULAND, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. 35. HRÓARSDÓTTIR, Thorbjörg. Word Order Change in Icelandic. From OV to VO. 2000. 36. GERLACH, Birgit and Janet GRIJZENHOUT (eds): Clitics in Phonology, Morphology and Syntax. 2000. 37. LUTZ, Uli, Gereon MÜLLER and Arnim von STECHOW (eds): Wh-Scope Marking. 2000. 38. MEINUNGER, André: Syntactic Aspects of Topic and Comment. 2000. 39. GELDEREN, Elly van: A History of English Reflexive Pronouns. Person, ‘‘Self’’, and Interpretability. 2000. 40. HOEKSEMA, Jack, Hotze RULLMANN, Victor SANCHEZ-VALENCIA and Ton van der WOUDEN (eds): Perspectives on Negation and Polarity Items. 2001. 41. ZELLER, Jochen : Particle Verbs and Local Domains. 2001. 42. ALEXIADOU, Artemis : Functional Structure in Nominals. Nominalization and ergativity. 2001. 43. FEATHERSTON, Sam: Empty Categories in Sentence Processing. 2001. 44. TAYLAN, Eser E. (ed.): The Verb in Turkish. 2002. 45. ABRAHAM, Werner and C. Jan-Wouter ZWART (eds): Issues in Formal German(ic) Typology. 2002. 46. PANAGIOTIDIS, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. 47. BARBIERS, Sjef, Frits BEUKEMA and Wim van der WURFF (eds): Modality and its Interaction with the Verbal System. 2002. 48. ALEXIADOU, Artemis, Elena ANAGNOSTOPOULOU, Sjef BARBIERS and HansMartin GAERTNER (eds): Dimensions of Movement. From features to remnants. 2002 49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. 2002. 50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. 51. GERLACH, Birgit: Clitics between Syntax and Lexicon. n.y.p. 52. SIMON, Horst J. and Heike WIESE (eds): Pronouns Grammar and Representation. 2002. 53. ZWART, C. Jan-Wouter and Werner ABRAHAM (eds): Studies in Comparative

54. 55. 56. 57. 58. 59. 60. 61.

Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26-27, 2000)(Workshop). n.y.p. BAPTISTA, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. n.y.p. COENE, M. and Yves D'HULST (eds): From NP to DP. Volume 1: The syntax and semantics of noun phrases. n.y.p. COENE, M. and Yves D'HULST (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. n.y.p. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. n.y.p. DI SCIULLO, Anna-Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. n.y.p. DEHÉ, Nicole: Particle Verbs in English. Syntax, information structure and intonation. n.y.p. TRIPS, Carola: From OV to VO in Early Middle English. n.y.p. SCHWABE, Kerstin and Susanne WINKLER (eds.): The Interfaces. Deriving and interpreting omitted structures. n.y.p.

Pronouns: Grammar and Representation

Pronouns

German pronouns and prepositions

Descriptive Pronouns and Donkey Anaphora

Nouns and Pronouns in Sanskrit

Nias Unmutated Personal Pronouns

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Practice Makes Perfect: Italian Pronouns and Prepositions

The Secret Life of Pronouns

Representation

Diagrammatic Representation and Inference

Urban Space And Representation

Knowledge representation and reasoning

Fair and Effective Representation?

Representation and Realism

Diagrammatic Representation and Inference

Representation and Reality

Meaning and Mental Representation

Knowledge Representation and Reasoning

Lexical representation and process

Representation And Reality

Diagrammatic Representation and Inference

Urban Space And Representation

Idolatry and Representation

Sociology and Visual Representation

Culture, Capital and Representation

Representation and Expression

Lexical Representation and Process

Pronouns: Grammar and Representation

Pronouns

German pronouns and prepositions

Descriptive Pronouns and Donkey Anaphora

Nouns and Pronouns in Sanskrit

Nias Unmutated Personal Pronouns

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Representation and Reality (Representation and Mind)

Practice Makes Perfect: Italian Pronouns and Prepositions

The Secret Life of Pronouns

Representation

Diagrammatic Representation and Inference

Urban Space And Representation

Knowledge representation and reasoning

Fair and Effective Representation?

Representation and Realism

Diagrammatic Representation and Inference

Representation and Reality

Meaning and Mental Representation

Knowledge Representation and Reasoning

Lexical representation and process

Representation And Reality

Diagrammatic Representation and Inference

Urban Space And Representation

Idolatry and Representation

Sociology and Visual Representation

Culture, Capital and Representation

Representation and Expression

Lexical Representation and Process

Recommend Documents