Modality and Its Interaction with the Verbal System (Linguistik Aktuell Linguistics Today, LA 47)

Modality and its Interaction with the Verbal System

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 California at Berkeley University of Vienna

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 (University of Stuttgart) 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 47 Modality and its Interaction with the Verbal System Edited by Sjef Barbiers, Frits Beukema and Wim van der Wurff

Modality and its Interaction with the Verbal System Edited by

Sjef Barbiers Meertens Institute, Royal Dutch Academy of Sciences

Frits Beukema Wim van der Wurff University of Leiden

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 Modality and its Interaction with the Verbal System / edited by Sjef Barbiers, Frits Beukema and Wim van der Wurff. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166–0829 ; v. 47) Includes bibliographical references and index. 1. Modality (Linguistics) 2. P299.M6 M637 2002 415-dc21 isbn 90 272 2768 3 (Eur.) / 1 58811 167 9 (US) (Hb; alk. paper)

2002276250

© 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

Table of contents List of contributors Preface Current issues in modality: An introduction to modality and its interaction with the verbal system Sjef Barbiers

  

Modal verbs: Epistemics in German and English Werner Abraham



Modality and polarity Sjef Barbiers



Modals, objects and negation in late Middle English Frits Beukema and Wim van der Wurff



On the use and interpretation of root infinitives in early child Dutch Elma Blom



Modals and negation in English Annabel Cormack and Neil Smith



System interaction in the coding of modality Zygmunt Frajzyngier



Modality and theory of mind: Perspectives from language development and autism Anna Papafragou



Negative polarity and modality in Middle Dutch ghe-particle constructions Gertjan Postma



(Negative) Imperatives in Slovene Milena Milojevi´c Sheppard and Marija Golden



Modality and mood in Macedonian Olga Mišeska Tomi´c



Subject index



Name index



List of contributors Werner Abraham Maitschern 128, A-8942 Wörschach, Austria [email protected] Sjef Barbiers Meertens Instituut P.O. Box 94264 1090 GG Amsterdam, The Netherlands [email protected] Frits Beukema Department of English University of Leiden P.O. Box 9515 2300 RA Leiden, The Netherlands [email protected] Elma Blom Utrecht Institute of Linguistics/ OTS Trans 10, 3512 JK Utrecht, The Netherlands [email protected] Annabel Cormack Department of Phonetics & Linguistics University College London Gower Street, London WC1E 6BT, United Kingdom [email protected] Zygmunt Frajzyngier Linguistics Department Woodbury 407 (Campus Box 295) Boulder, CO 80309-0295 U.S.A. [email protected]

Marija Golden Kamniska 4a 2000 Maribor, Slovenia [email protected] Anna Papafragou Institute for Research in Cognitive Science University of Pennsylvania 3401 Walnut Str., Suite 400A Philadelphia, PA 19104-6228 [email protected] Gertjan Postma Department of Dutch/Centre of Linguistics University of Leiden P.O. Box 9515 2300 RA Leiden, The Netherlands [email protected] Milena Milojevi´c Sheppard Department of English and American Studies Faculty of Arts University of Ljubljana Ašker´ceva 2, 61000 Ljubljana, Slovenia [email protected] Neil V. Smith Department of Phonetics and Linguistics University College London Gower Street, London WC1E 6BT, Great Britain [email protected]

 List of contributors Olga Mišeska Tomi´c Spinoza Project Lexicon and Syntax University of Leiden P.O. Box 9515, 2300 RA Leiden, The Netherlands [email protected]

Wim van der Wurff Department of English University of Leiden P.O. Box 9515, 2300 RA Leiden, The Netherlands [email protected]

Preface

The majority of the papers included in this volume were first presented at the workshop on “Modality in Generative Grammar”, which was held during the 32nd annual meeting of the Societas Linguistica Europaea at the University of St Andrews (Scotland), 26–30 August 1998. We are very grateful to those contributors who responded to our call for papers and to our various deadlines for handing in versions of their papers: they deserve our warmest thanks. The contributions were subsequently refereed and revised in the light of reviewers’ comments. The volume also contains a number of invited papers, by Zygmunt Frajzyngier, Annabel Cormack and Neil Smith, and Anna Papafragou. Our debt to the linguists who assisted us in the reviewing process is enormous. Their help was freely and generously given; without it the volume could not have come about. The editing process was originally in the hands of Sjef Barbiers and Frits Beukema; in September 2000 Wim van der Wurff joined them, and helped to prepare the volume for the press. Sjef Barbiers Frits Beukema Wim van der Wurff

Current issues in modality An introduction to modality and its interaction with the verbal system Sjef Barbiers Royal Netherlands Academy of Sciences, Meertens Institute, Amsterdam

.

Introduction

The goal of this introduction is to provide an overview of the research questions concerning modality that have been addressed in the framework of generative grammar, of the answers that have been provided to these questions, and of the ways in which the papers in this volume relate to these questions. Modality is a semantic concept and covers, inter alia, notions such as possibility, necessity, probability, obligation, permission, ability and volition. Modality can be expressed syntactically by modal verbs, imperatives, verbal inflection, modal adverbs and modal particles.

. The epistemic–root distinction A great deal of work on modality in generative syntax has concentrated on the distinction between epistemic and root interpretations of modal verbs. Roughly, epistemic interpretations are a class of interpretations involving a speaker-oriented, or, in the case of embedded clauses, matrix-subject oriented qualification or modification of the truth of a proposition, while root interpretations involve the will, ability, permission or obligation to perform some action or bring about some state of affairs (cf. Lyons 1977 and Palmer 1986 for a fine-grained classification and description of modal interpretations). Sentences often are ambiguous between the two readings. In many Germanic and



Sjef Barbiers

Romance languages the same set of modals can have both the epistemic and the root interpretations. An example is given in (1): (1) John must be at home at six o’clock Epistemic: ‘(Given what I, the speaker, know), I predict that John will be at home at six o’clock.’ Root: ‘John is obliged to be at home at six o’clock.’

An important question is whether this systematic semantic ambiguity entails the existence of two different syntactic structures. One of the first answers to this question (Ross 1969) is that modals with an epistemic interpretation are one-place predicates, taking the entire proposition as their complement, whereas root modals are two-place predicates, i.e. they involve relations between the subject and the rest of the clause: (2) Epistemic: must (John be at home at six o’clock) Root: must (John, be at home at six o’clock)

We will take this idea as a starting point, as a great deal of work has been invested to translate it into syntactic terms. There are different ways to represent the alternation between monadic and dyadic modality syntactically. A first option would be to try to reduce the modal alternation to similar alternations with verbs like BREAK and SMOKE. However, it will be shown in Section 3 that these are alternations of a fundamentally different type. A second way to represent the alternation in terms of argument structure is to analyse root modals as control structures and epistemic modals as subject raising structures. This is the most common generative analysis of the root–epistemic ambiguity (Hofmann 1966; Ross 1969; Perlmutter 1970 and numerous studies since then). However, as shown in Section 4, there are a number of serious problems for this analysis. Thirdly, the modal alternation could be the result of generating modals in different base positions, a higher, possibly functional, position for epistemic modals and a lower, possibly lexical, position for root modals. This analysis is not entirely without problems either, as Sections 5.1–5.3 show. A fourth way to analyse the modal alternation is to assume that epistemic modals, but not root modals, undergo movement at LF. This possibility is discussed in Section 5.4. On this approach, the monadic nature of the modal verb in epistemic interpretations is not established until LF. An immediate question for this analysis is how the trace of the modal should be interpreted.

Current issues in modality

A final type of analysis, discussed in Section 6, is to attribute the difference in modal interpretation to the selection of different types of complements, a larger complement being selected in the case of epistemic interpretations, and a smaller one in the case of root interpretations. This amounts to providing a different perspective on the possibility of generating modals in different head positions. Obviously, if modals are generated in different positions depending on their interpretation, the type and size of their complements will vary accordingly.

. Transitive–intransitive alternations As is well known, English modals have a defective paradigm and a distribution differing from that of main verbs (cf. Section 5). Therefore, the question as to whether the dyadic– monadic alternation is reducible to other transitive–intransitive alternations does not seem to be relevant for English. It is a relevant question, however, for languages in which modals have many properties in common with main verbs, such as Dutch and German. This section shows that in such languages the modal alternation is not reducible to transitive–intransitive alternations either. Transitive–intransitive alternations are found cross-linguistically with main verbs like BREAK and SMOKE (3a–d). A common analysis of such alternations is that each of these verbs has only one lexical entry, which corresponds to the transitive variant. The intransitive variant is derived by a lexical operation that reduces transitive predicates to intransitive predicates by removing the external argument. (3) a. b. c. d.

John broke the window. The window broke. John smoked the eel. The chimney smoked.

In fact, BREAK and SMOKE represent two different types of alternations: BREAK is a transitive–unaccusative alternation and SMOKE a transitive–unergative alternation (cf. Abraham 2000). The two can be distinguished by using classic tests described in Burzio (1986) and Hoekstra (1984). In Dutch, the unaccusative alternant selects the auxiliary BE in the perfect tense and its adjectival participle can modify a noun that corresponds to the unaccusative’s subject. The unergative alternant selects the auxiliary HAVE and its participle cannot modify such a noun. This is illustrated in (4a–f).





Sjef Barbiers

(4) transitive–unaccusative a. Jan breekt de ruit. John breaks the window c. de gebroken ruit the broken window transitive–unergative d. Jan rookt een paling. John smokes an eel f. *de gerookte kachel the smoked stove

b. De ruit is/*heeft gebroken. the window is/has broken

e. De kachel heeft/*is gerookt.1 the stove has/is smoked

If the root–epistemic ambiguity of modals were to involve reduction of dyadic predicates to monadic ones, then these tests would show that modals belong to the same class as roken ‘smoke’, as (5a–c) illustrate. In the perfect tense HAVE is selected, and the participle cannot modify a noun corresponding to the surface subject of the intransitive alternant: (5) root–epistemic a. Jan mag eten. John may eat ‘John is allowed to eat.’ c. *het gemogen eten the allow- eating

b. Eten heeft/*is altijd gemogen. eat has/is always may- ‘It has always been allowed to eat.’

However, the monadic alternant of the modal (5b) does not have an epistemic interpretation, which shows that the epistemic–root ambiguity is not an instance of the transitive–unergative alternation either. Moreover, for the epistemic interpretation to be possible, the external argument does not need to disappear, as (1) shows. We can thus conclude that the epistemic–root ambiguity can neither be reduced to the transitive–unaccusative alternation nor to the transitive–unergative alternation.

. The raising–control analysis of the epistemic–root ambiguity A consequence of treating root modals as dyadic predicates is that they involve a control structure, if something like the theta-criterion holds. If a root modal assigns a theta-role to the surface subject DP and if a DP must receive exactly one theta role, it cannot be the case that root modals involve raising structures. If they did, the raised subject would receive a theta-role from both the embed-

Current issues in modality

ded predicate and the modal.2 Therefore, root modals can only be analysed as dyadic predicates if they involve control structures, which implies that PRO in the complement of the modal receives the external theta-role of the embedded verb, while the matrix subject receives its theta-role from the modal directly. (6) I. [[DP John] must [DP John] work from nine to five]raising & epistemic II. [[DP John]i must [PROi work from nine to five] control & root

The control–raising analysis rather elegantly captures the idea that root modals are dyadic predicates while epistemic modals are monadic predicates. However, many problems for this analysis have been noted in the literature. First, assuming a PRO-subject in the infinitival complement of the modal is problematical because infinitival complements without te ‘to’ usually cannot have PRO as their subject, as the Dutch example in (7a) illustrates.3 (7) a.

Jan voelde [zich/*PRO wegzakken in de modder] John felt himself/PRO sink in the mire ‘John felt that he was sinking in the mire.’ b. Jan voelde dat hij wegzonk in de modder ‘John felt that he was sinking in the mire.’ c. Jan at [SC zich/*PRO ziek] John ate himself/PRO sick ‘John ate so much that he got sick.’

The infinitival complement in (7a) requires a reflexive as its subject if it is to express the same meaning as (7b). If this infinitival complement can be analysed as a small clause (SC), the impossibility of PRO in (7a) is an instance of the broader generalization that SC-complements cannot have a PRO subject, as (7c) illustrates (cf. Stowell 1981). A possible way out would be to argue that modals do not take small clauses as their complements, but larger constituents, e.g. CP or TP. For the Germanic languages, there does not seem to be much evidence, however, that this is the case. C-elements such as complementisers are impossible in the complement of a modal, as is embedded WH. If infinitival to is a T-element, its obligatory absence in most modal contexts shows that the complement of a modal is not a TP either. Higher adjuncts such as evidential and modal adverbs cannot be part of the complement of a modal, which also suggests that complements of modals are smaller than CP. However, it cannot be concluded from the situation in Germanic that it is generally impossible for modals to have a CP or TP complement. As Tomi´c (this volume) argues, Macedonian has a set of modals that take subjunctive





Sjef Barbiers

CP-complements. Incidentally, these Macedonian constructions should not be confused with cases in which modals in Germanic seemingly take a CPcomplement, as in (8a). There is reason to think that CP in this construction is not a complement but an adjunct (cf. Bennis 1986). As with adjunct clauses, CP in this construction is opaque for extraction, as the Dutch example in (8b) illustrates. (8) a.

Het kan niet [CP dat Nederland de Portugese ploeg vanavond verslaat]. it can not that Netherland the Portuguese team tonight beats ‘It is impossible that the Dutch will beat the Portuguese team tonight.’ b. *Welke ploeg kan het dat Nederland welke ploeg vanavond verslaat? which team can it that Netherland which team tonight beats

A second problem for the analysis of root modals as dyadic control predicates is provided by sentences in which the matrix subject cannot be interpreted as the person being obliged or permitted to do something (Feldman 1986; Klooster 1986; Brennan 1993; Barbiers 1995): (9) a.

This letter must be in London before five o’clock. ‘It is necessary/obligatory that this letter be in London before five o’clock.’ b. Al deze kranten mogen weg. all these newspapers may away ‘It is permitted to throw away all of these newspapers.’

In these sentences, the modal clearly is a monadic predicate taking the entire clause as its argument, but (9a) can and (9b) must have a root interpretation. Obviously, it is the addressee of utterances like (9a, b) that has the obligation or permission to do something, but the crucial point is that this addressee cannot be expressed syntactically here. In addition to this, it has often been observed that root modals (in fact, modals in general) cannot be passivised.4 This is unexpected if they are dyadic verbs, since dyadic verbs which assign an external theta role satisfy an important condition for passivisability (Burzio 1986). However, it should be noted that modals share this inability to occur in the passive with other stative transitive verbs. It may be that the crucial factor blocking passivisation here is stativity. A third problem for the control analysis of root modals involves the behaviour of expletive subjects (Thráinsson and Vikner 1995; Warner 1993; McGinnis 1993). In some languages, e.g. English, expletive subjects occur with

Current issues in modality

raising verbs but not with control verbs (10a, b). If root interpretations always involved control structures, expletive subjects are predicted to be incompatible with root interpretations, contrary to fact, as (10c) illustrates. (10) a. There seems to be a man in the room. b. *There tried someone to call you. c. There must be a solution to this problem on my table this morning.

A fourth problem emerges when we look at languages with morphological case. Thráinsson and Vikner (1995: 60) show that in Icelandic modals with a root interpretation behave like raising predicates with respect to quirky case. When a verb requiring quirky case on its subject is embedded under a raising verb, the surface subject appears with quirky case. When such a verb is embedded under a control verb, the surface subject appears with nominative case. Regardless of the interpretation, the subject of modal clauses always occurs with quirky case when the embedded verb requires this. A fifth problem involves modals taking a non-verbal complement. Barbiers (1995, this volume) shows that Dutch modals can take a non-verbal complement, e.g. a PP or AP complement.The construction can also be found in Afrikaans, German, Norwegian, Yiddish and West Frisian.5 Modals with a PP or AP complement are subject raising constructions. Yet, they can only get non-epistemic interpretations. Finally, Wurmbrand and Bobaljik (1999) show that A-reconstruction for scope, i.e. the subject taking scope under the modal, is possible for raising verbs and for both root and epistemic modals, whereas this is impossible for control verbs. In sum, the observations presented in this section pose serious problems for a control analysis of modals with a root interpretation, but not for a raising analysis.

. The structural position of modals An alternative way to shed some light on the epistemic–root distinction is to investigate the structural position of modals. It turns out, however, that their position is rather difficult to establish; a simple and universal picture seems virtually impossible given the empirical facts. At least four distinctions have to be taken into account: i.

The distinction between modal auxiliaries and modal main verbs;





Sjef Barbiers

ii. The distinction between epistemic and root modality; iii. The distinction between necessity and possibility modals; iv. The distinction between the position where a modal is interpreted and the position where a modal is pronounced. The evidence currently available strongly suggests that cross-linguistic parametrisation involves all four distinctions. . Different positions for modal auxiliaries and modal main verbs The distinction between modals behaving like auxiliaries and modals behaving like main verbs has engendered a great deal of discussion. A well-known view is that modals in English are auxiliaries generated in Infl or Aux (cf. Roberts 1985) because they have a defective inflectional paradigm and behave like other auxiliaries in that they can undergo residual V2 and do not trigger DO-support in negative clauses. On this view, modals in languages like German and Dutch are main verbs projecting their own VPs, as their properties are largely (though not completely) identical to those of main verbs. Clearly, the distinction between main verb and auxiliary modals does not shed any light on the epistemic–root ambiguity. . Different positions for epistemic and root modals Although it is true that across and even within languages modals may differ with respect to their categorial status, the idea that this distinction alone determines the structural position of modals has turned out to be too simple. A complication is that there are languages such as Catalan that seem to have two positions for modals (Picallo 1990). There is a higher position above clausal negation that corresponds to epistemic interpretations, and there is a lower position below clausal negation corresponding to root interpretations. A similar view is defended in Cinque (1999), where it is claimed on the basis of the relative ordering of different adverb types that the universal order of modal heads is such that epistemic modals c-command root modals. Evidence potentially supporting this idea comes from languages that allow for double modals. In such languages a modal with a root interpretation cannot take scope over a modal with an epistemic interpretation (cf. Barbiers 1995 for Dutch; Thráinsson and Vikner 1995 for Icelandic; Roussou 1999 for Greek).

Current issues in modality

. Different positions for necessity and possibility modals Cormack and Smith (this volume) provide the most comprehensive analysis of the interaction between negation and modality that is currently available. They argue that English has two positions for modals too; a position higher than clausal negation and a position lower than clausal negation. However, in English this syntactic split does not correspond to the epistemic–root distinction, as it does in Catalan, but to a distinction between necessity modality (high) and possibility modality (low). They show that there are modals with a root interpretation occurring in the higher modal position and modals with an epistemic interpretation occurring in the lower modal position. Postma (this volume) argues on the basis of the distribution of the negative polarity prefix ghe- that Middle Dutch has the same distinction. If these claims are correct, Cinque’s hierarchy is not valid for English and Middle Dutch, and the impossibility of a root modal taking scope over an epistemic modal cannot have a syntactic explanation (see Cormack and Smith, this volume, for an account). . Different positions at PF and LF In establishing the structural position of modals, the position in which a modal is pronounced should be distinguished from the position in which it is interpreted. This can be done in various ways, depending on the set of assumptions adopted. Cormack and Smith (this volume) argue that a modal is inserted in the position where it is interpreted, while the position of pronunciation may be different. A second possibility is that scopal ambiguities as in They can frequently buy eggs are the result of overt raising of can across frequently (Ernst 1992). The narrow scope reading of can may then be obtained by reconstruction at LF or by computing the scope over the trace of can. It has also been argued that (some) modals are quantifiers that undergo LF-raising (McDowell 1987).6 An epistemic modal would differ from a root modal in that it raises to a clause initial position at LF where it takes scope over the entire proposition. McDowell claims that the incompatibility of epistemic interpretations and questions supports the LF-raising analysis. This becomes clear when we use a complement that forces an epistemic interpretation (see Section 6), as in (11a). A yes–no question cannot be based on (11a), as (11b) shows: (11) a.

Jan moet wel een moedertaalspreker van het Fins zijn. John must  a native speaker of the Finnish be





Sjef Barbiers

‘(John speaks Finnish fluently, so) John must be a native speaker of Finnish.’ b. *Moet Jan wel een moedertaalspreker van het Fins zijn? Must John  a native speaker of the Finnish be Intended meaning: ‘Is it necessarily true that John is a native speaker of Finnish?’

McDowell’s syntactic explanation for the fact that a clause with an epistemic modal cannot be interrogative is based on the doubly-filled Comp filter. This filter blocks structures in which Comp contains more than one element. Questions involve a Wh- or Q-operator; when such an operator fills Comp, it is no longer possible for an epistemic modal to move to Comp. This analysis does not automatically carry over to the current generative framework, in which there is a proliferation of left-peripheral functional positions that could potentially host operators. Moreover, McDowell’s observation requires further qualification, since the incompatibility of question formation with epistemic interpretations does not seem to hold for every type of modal. Modals expressing possibility do occur in epistemic questions, as (12) illustrates: (12) a. Could the keys be on John’s table? (Brennan 1993) b. Kan hij een moedertaalspreker van het Fins zijn? can he a native speaker of the Finnish be ‘Is it possible that he is a native speaker of Finnish?’

We may conclude from this that the interaction of epistemic and interrogative operators requires further research. . Developmental issues It has often been observed that first language learners acquire root modals earlier than epistemic modals. If it is possible to show that epistemic modals are generated in a higher syntactic position than root modals, this observation could be explained in a framework that assumes that language acquisition occurs from bottom-to-top or from lexical projection to functional projection (e.g. Radford 1992): children start out with a projection of the verb and its arguments, while functional projections come in later: the lower a functional projection, the earlier it occurs. However, if Cormack and Smith’s claim (this volume) is correct that there is no strict correlation between epistemic–root and height of insertion, such an explanation is problematical.

Current issues in modality

Papafragou (this volume) tries to explain the lagging behind of epistemic modality in extralinguistic terms, i.e. in terms of meta-cognition. In her view, the ability to use epistemic modality depends on theory-of-mind abilities. If such abilities do not develop, as in the case of autism, epistemic modality remains absent. Another interesting and much discussed developmental issue involves the occurrence of so-called root infinitives in many child languages. Children at a certain age use infinitival clauses as independent clauses, where in the adult grammar an independent clause would require a finite verb. Despite the absence of a finite verb or a modal, such infinitival clauses often carry modal meaning. The question then arises where this modal interpretation comes from. Does the infinitive itself or the infinitival suffix that it contains contribute to modal interpretation (Hoekstra and Hyams 1998), or is there hidden (functional) structure containing modal elements (Boser et al. 1992)? Blom’s paper (this volume) addresses these issues. Following Hyams and Hoekstra, she argues that it is indeed the infinitival morphology that contributes to modal interpretation. She explains the apparent problem that children also use non-modal root infinitives by assuming that children first have to learn that infinitives can be either nominal or verbal. As soon as they know that they can be verbal, modal interpretations arise. The idea that an infinitive or infinitival morphology contributes to modal interpretation does not seem far-fetched in the light of constructions such as (13), which are frequent cross-linguistically: (13) a. John has to obey. b. Dit probleem is op te lossen. this problem is up to solve ‘This problem is solvable.’ c. een op te lossen probleem an up to solve problem ‘a problem that must/can be solved’

Here the source of the modality could be to/te, the infinitive, the two in interaction, or just the absence of finite tense on the main verb. Many interesting issues arise, among which questions about cross-linguistic differences, such as the fact that not all languages can have modally interpreted infinitives in the three constructions in (13).





Sjef Barbiers

. The complement of modals The syntactic representation of the epistemic–root distinction can also be studied from the perspective of the modal’s complement. Relatively little attention has been paid in the literature to the influence of the modal’s complement on modal interpretation. This section provides an overview. . Non-disambiguating complements At first sight, it seems plausible to expect that modals with a root interpretation require verbs assigning an agent role as their complements. An agent may seem to be necessary to carry the permission, obligation, ability or will to perform the action denoted by the embedded verb. Passive, unaccusative, stative and complements with a perfect might then be expected to force epistemic interpretations. In passives, the agent role is demoted, in unaccusatives there is no agent role, stative verbs denote a state of affairs, not an action that can be performed by an agent, and in perfectives the event denoted by the verb has already been completed. However, these types of complements all allow root interpretations: (14) a.

This room must be cleaned immediately. passive ‘You must clean this room immediately’ b. Je mag niet weglopen. unaccusative You may not away-walk ‘You’re not allowed to walk away.’ c. You must know the answer before five o’clock. stative ‘You’re required to know the answer before five o’ clock.’ d. You must have cleaned your room before five o’clock. perfective ‘You’re obliged to have cleaned your room before five o’clock.’

. Complements that force an epistemic interpretation Barbiers (1995, this volume) argues that complements denoting a state or event that cannot change (anymore) force an epistemic interpretation. There are two types of complements for which this is the case: (15) i.

Stative complements which contain an individual-level predicate, provided that the subject has fixed reference.7 ii. Perfect complements, but only if the completion stage of the event has taken place in the past.

Current issues in modality

An example of both cases is given in (16): (16) a.

John must be a native speaker of Finnish. speaker of Finnish.’ b. They must have cleaned this room yesterday. #‘They must have had the obligation to clean this room yesterday.’ #‘John has the obligation to be a native

These two observations do not seem to have consequences for the syntactic analysis of modal ambiguity, as disambiguation is the result of semantic properties of the complements. Nevertheless, the two types of complements in (16) are useful diagnostics to establish whether an epistemic interpretation is possible in a certain construction. . Complements that force root interpretations For an epistemic interpretation to be available, it is necessary that the complement contain an infinitival verb. This can be seen in languages in which modals allow non-verbal complements, such as Dutch. Such constructions can only have root interpretations. In Barbiers (1995, this volume), I provide evidence that such complements are truly non-verbal, and not elliptical. Based on these observations I conclude that the availability of root and epistemic interpretations is determined by the complement of the modal. When the modal has VP, PP or AP as its complement, it receives a root interpretation; when it has vP (little vP, the abstract verbal projection dominating VP; cf. Larson 1988; Hale and Keyser 1993; Chomsky 1995) as its complement, it receives an epistemic interpretation. . Another disambiguating context Disambiguation also arises when the modal verb itself is in the present perfect. Modals in the present perfect can only have non-epistemic interpretations; a modal in the present perfect cannot co-occur with a complement that forces an epistemic interpretation (17a). For possible explanations, see Abraham (this volume), Diewald (1993), Zifonun et al. (1997) for German, and Barbiers (1995) for Dutch. (17) a. *Jan heeft een moedertaalspreker van het Fins moeten zijn. John has a native speaker of the Finnish must- be8 #‘John has had the obligation to be a native speaker of Finnish.’





Sjef Barbiers

b. Jan had een moedertaalspreker van het Fins moeten zijn. John had a native speaker of the Finnish must- be ‘John should have been a native speaker of Finnish (then he would have got the job).’

Interestingly, the disambiguating effect of the stative and individual-level complement disappears in the past perfect, as (17b) shows. The modal in (17b) has a root interpretation. The fact that the complement allows this in this construction is due to the irrealis interpretation forced by the past perfect: the sentence entails that John is not a native speaker of Finnish but could have been one in a different world. This makes it possible to interpret being a native speaker as a variable property of John, and this makes root interpretations available.

. Interaction of modals with other systems . Modality and negation The interaction between negation and modals is crucial for a proper understanding of modality. As we have seen above, many of the papers in this volume pay attention to it. In particular, the paper by Cormack and Smith provides a detailed description and novel analysis of the interaction between negation and modality in English. Two interesting observations can be found in the papers by Beukema and van der Wurff and by Postma. As Beukema and van der Wurff observe, the presence of modals turns out to make overt raising of negative constituents possible. They show that in fifteenth-century English, when English had changed from OV to VO, OV order could still occur, but only in restricted environments, one of which involves a negative object enclosed between a modal verb and a lexical verb. Postma argues that possibility modality is necessary to license verbal negative polarity items. . The coding of modality Modal verbs are just one way to encode modality in natural languages. Among the other means attested are modal particles and adverbs, imperatives and toinfinitives. Frajzyngier (this volume) shows how verbal inflection, word order, auxiliary verbs and modal particles participate in the coding of modality in Lele, an East-Chadic language. In Slovene (Sheppard and Golden, this vol-

Current issues in modality

ume) imperatives expressing deontic modality surprisingly may occur in finite embedded clauses.9

. Conclusion Generative syntactic research of the past forty years has brought to light a large number of properties of modals and modal constructions. Many of these properties are discussed in this volume, some of them for the first time. Questions about the argument structure of modals, the position of modals and the relation between syntactic properties and semantic ambiguity have been sharpened considerably. Many questions remain, however. A number of serious problems have been raised for the raising-control analysis of the epistemic-root ambiguity. Alternative analyses have been proposed, but so far there is no conclusive evidence to choose between them. Another important issue involves the tenability of the universal base hypothesis as exemplified by Cinque’s (1999) approach to modal positions, in the light of the observed cross-linguistic differences.

Notes . I am using an underlying object different from the one in (4d) to avoid complications caused by the fact that BE is also the perfect auxiliary of the passive in Dutch. . Thráinsson and Vikner (1995) discuss this problem for Scandinavian languages, where modals behave like raising verbs regardless of interpretation. They attempt to solve this by assuming that modals assign so-called adjunct theta roles. The theta criterion would dictate that a DP should not receive two theta roles, but receiving one theta role and one adjunct theta role would be allowed. . Note that this is true at the observational level. The theory-internal hypothesis that PRO cannot occur in governed positions is irrelevant for the problem under discussion. . The verb in the complement can be passivised if it has an external argument. The point at issue here is that the modal verb itself cannot be passivised. . The construction can also be found in Afrikaans, German, Norwegian, Yiddish and West Frisian. . See Brennan (1993, 1997) for a discussion of the quantificational properties of modals. . Names are DPs with fixed reference. In the normal case, the referent of a name is constant across time and across different possible worlds (but see Section 6.4 for irrealis contexts). . Note that this sentence contains an instance of the so-called Infinitivus pro Participio effect: an infinitive occurs where we would expect a past participle.





Sjef Barbiers . Cf. Platzack (2001) and references cited there for imperatives in embedded clauses in Old Icelandic and Old Swedish.

References Abraham, W. (2000). The aspect-case typology correlation. Perfectivity and Burzio’s generalization. In E.J. Reuland (Ed.), Arguments and Case. Explaining Burzio’s Generalization (pp. 131–193). Amsterdam: John Benjamins [Linguistik Aktuell/Linguistics Today 34]. Barbiers, S. (1995). The Syntax of Interpretation. PhD diss. Leiden. The Hague: HAG Publications. Bennis, H. (1986). Gaps and Dummies. Dordrecht: Foris. Boser, K., B. Lust, L. Santelmann, and J. Whitman (1992). The syntax of CP and V2 in early child language: The strong continuity hypothesis. In Proceedings of NELS 22. Amherst: GLSA. Brennan, V. (1993). Root and epistemic modal verbs. Unpublished PhD diss. UMass. Amherst: GLSA. Brennan, V. (1997). Quantificational modals. Linguistic Inquiry, 28(1), 165–169. Burzio, L. (1986). Italian Syntax: A Government – Binding approach. Dordrecht: Reidel. Chomsky, N. (1995). The Minimalist Program. Cambridge (Mass): MIT Press. Cinque, G. (1999). Adverbs and Functional heads. A Cross-Linguistic Perspective. Oxford Studies in Comparative Syntax. Oxford, New York: Oxford University Press. Diewald, G. (1993). Zur Grammatikalisierung der Modalverben im Deutschen. Zeitschrift für Sprachwissenschaft, 12, 218–234. Ernst, T. (1992). The phrase structure of English negation. The Linguistic Review, 9(2), 109–144. Feldman, F. (1986). Doing the best we can. Philosophical Studies 35. Dordrecht: Reidel. Hale, K. and S.J. Keyser (1993). On Argument Structure and the Lexical Expression of Syntactic Relations. In K. Hale and S.J. Keyser (Eds.), The View from Building 20. Essays in Linguistics in Honor of Sylvain Bromberger (pp. 53–110). Cambridge (Mass.): The MIT Press. Hoekstra, T. (1984). Transitivity. Dordrecht: Foris. Hoekstra, T. and N. Hyams (1998). Aspects of root infinitives. Lingua, 106, 81–112. Hofmann, T. (1966). Past tense replacement and the modal system. Reprinted in J. McCawley (Ed.), Syntax and Semantics, Vol. 7 (pp. 85–100). New York: Academic Press. Klooster, W. (1986). Problemen met complementen. TABU, 16, 112–132. Larson, R. (1988). On the double object construction. Linguistic Inquiry, 19, 335–391. Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press. McDowell, J. (1987). Assertion and modality. Unpublished PhD diss. USC, Los Angeles. McGinnis, M.J. (1993). The deontic/epistemic distinction in English modals: A compositional analysis. MA Thesis, University of Toronto. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press.

Current issues in modality

Perlmutter, D. (1970). The two verbs begin. In R. Jacobs and P. Rosenbaum (Eds.), Readings in English transformational grammar. Boston: Ginn. Picallo, C. (1990). Modal verbs in Catalan. Natural Language and Linguistic Theory, 8(2), 285–312. Platzack, C. (2001). Embedded imperatives. Unpublished manuscript, Lund University. Radford, A. (1992). The acquisition of the morphosyntax of finite verbs in English. In J. Meisel (Ed.), The Acquisition of Verb Placement: Functional Categories and V2 phenomena in language acquisition (pp. 23–62). Dordrecht: Kluwer. Roberts, I. (1985). Agreement patterns and the development of English auxiliaries. Natural Language and Linguistic Theory, 3, 21–58. Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in Philosophical Linguistics, Series 1 (pp. 77–102). Evanston: Great Expectations. Roussou, A. (1999). Modals and the subjunctive. In A. Alexiadou et al. (Eds.), Studies in Greek Syntax (pp. 169–183). Dordrecht: Kluwer. Stowell, T. (1981). Origins of Phrase Structure. Unpublished PhD diss., MIT. Thráinsson, H. and S. Vikner (1995). Modals and double modals in the Scandinavian languages. Working Papers in Scandinavian Syntax, 55, 51–88. Warner, A. (1993). English auxiliaries. Cambridge, Cambridge University Press. Wurmbrand S. and J. Bobaljik (1999). Modals, Raising and A-reconstruction. Lecture presented at the HIL Linguistics seminar. Leiden, October 1999. Zifonun, G., L. Hoffmann, B. Strecker, et al. (1997). Grammatik der deutschen Sprache. (Schriften des Instituts für Deutsche Sprache 7). Berlin and New York: Walter de Gruyter.



Modal verbs* Epistemics in German and English Werner Abraham University of Groningen

The distribution of modal verbs in German is considerably wider than that of their Modern English counterparts. It is generally claimed that this is partly due to the V-to-I movement in the history of Modern English. However, there is a considerable remainder of differences between Old English modals and their Modern English equivalents that cannot have anything to do with the position of the finite predicate. It will be argued in this paper that the specific, highly auxiliarised meanings of Modern English modals are due to the loss of aspectual or Aktionsart properties in the course of early Middle English.

.

Introduction: A solid set of differences between Modern English and Modern German

Diachronically, modal verbs (MVs) in Modern English and in Modern German derive from a common source and therefore demonstrate a solid equivalence, both syntactic and semantic, in their older stages (Old English, OE, and Old High German, OHG), but they differ considerably in their present-day distributional properties. Generally speaking, MVs in Modern German represent a far more homogeneous verbal class than their counterparts in Modern English. They display a clear morphological distinction from non-modal verbs, but, at the same time, they show a considerable syntactic similarity to them (as shown in great detail in Reis 2000). The morphological distinction is partly shared by Modern English modals while the syntactic similarity that German modals exhibit with full verbs in Modern German does not exist in Modern English. I will discuss these characteristics in detail, following and extending Abraham (1990, 1998). Let us first consider the frequency of occurrence of the Modern English modals in their root and epistemic uses. Tables 1 and 2 contain data on the



Werner Abraham

Table 1. Frequency of modal verbs in the Corpus of Spoken Professional American Modern English, Part 1 (93,472 words)

temporal root epistemic

will

may

1655 24 8

0 84

might

shall

must

144

0 0 0

0 0

Table 2. Frequency of modal expressions in the Corpus of Spoken Professional American Modern English, Part 1 (93,472 words) want to

wish to

be permitted

should

have to

is to

267

4

6

100

222

59

frequency of modal verbs and some of their equivalents, based on a corpus of spoken American English. A precise word count on Part 1 of the Corpus of Spoken Professional American Modern English (in CD-format) covering 93,472 words reveals a total of 918 modal verbs and modal expressions. This result confirms the impression one gathers when browsing through the 19 other Parts of the CD-ROM (the total Corpus of Spoken Professional American Modern English covers some 2 million words). It is striking that no volitional root use of will is found in the corpus, nor is there a single occurrence of postulative shall, which does not appear to occur anymore in Modern English. This is in stark contrast to Modern German wollen and sollen, which have been preserved as full deontic main verbs throughout their historical development. Also note that the notional equivalents of the root modals (want to, wish to, be permitted, have to, is to) are represented in sizeable numbers. Furthermore, the preterite forms might and should are rather frequent in the part of the corpus investigated here; most of the cases involve epistemic meanings, thus reinforcing the tendency for present-tense modals to have non-epistemic meanings. In particular, whenever will occurs, it has a predominantly temporal meaning. The remainder of the modals in Modern American English has a shaky status: with the exception of must and single idiomaticised root uses of other modals (e.g. May I?), there are only epistemic uses left. Moreover, will and shall had already attained a purely temporal function at some earlier stage of their development. While Coates (1995) estimates that Modern British English must has root meaning in half of all cases and epistemic meaning in the other half, this is not reflected in the American English data in Table 1. In the linguistic literature it is often claimed that may can show

Modal verbs in German and English

root meaning, but the Corpus of Spoken Professional American Modern English reveals that epistemic may is far more frequent.1 Recall that English past tense modals such as might and should were the first to develop epistemic meanings. This is also true for the development of the German modals (Abraham 1989, 1992, 1995). In conclusion, it can be said that, while modals verbs in German (and Dutch; see Barbiers 1995) have retained their polyfunctionality (i.e. functioning both as root-deontics and epistemics, sometimes adding ancillary temporal uses as Dutch zullen for the future), Modern American English is about to give up, or has lost already, the complete array of root modal meanings: with certain very specific provisions, all modals in American Modern English function as epistemics or, in the case of will, as a temporal auxiliary. The properties of Old English MVs were originally similar to those of Modern German. We also observe that the verbal lexicon in OE as well as in Modern German has a large number of verbs derived by means of prefixed particles which are based on adjectives, prepositions, and adverbs. These prefixes are separable from the verbal stem (in subordinate clauses, in topicalisation structures and in stylistic fronting etc.) and therefore have the syntactic status of secondary predicates (small clauses). This is still the unmarked and productive situation in Modern German. Lexically signalled aspect, or Aktionsart morphology, has disappeared in Modern British and American English. Moreover, OE as well as Modern German – but not Modern English – make a distinction between a stative passive and a dynamic passive with the help of auxiliary selection (OE beon vs. weorðan + past participle, Modern German sein vs. werden + past participle). In English, however, the stative passive be + past participle has additionally acquired a dynamic interpretation by extension of a general implicational relation in an earlier aspectual system (i.e. earlier ‘is being V-ed’ is implied by ‘is V-ed’, holding originally only in the case of telic verbs). An additional difference is that the ge-morpheme as used in the past participle still figures prominently in Modern German,2 while English lost it in early Middle English (ME). The participial morpheme was formally extended to the past participle from an original lexical perfectivising prefix, which also still survives in German verbs like geschehen ‘happen’. This reflects the historical situation in which the past participle had a perfective stative (i.e. resultative) function only (as documented for Gothic as well as Early Old High German; cf. Abraham 1992). The dynamic meaning was acquired from the resultative periphrasis. All continental West Germanic languages (German, Dutch, Flemish, West Frisian, Yiddish and their dialects) share these three properties: 1. paradigms of separable verbal prefixes with a perfective or resultative meaning; 2. the existence





Werner Abraham

of two types of passive with a distinct event reference; and 3. the participial morpheme ge-. None of these three properties are found in Modern English. I will argue that this threefold characterisation lies at the heart of the difference between the MV paradigms of Modern English and Modern German – exactly how and why will be discussed below.

. The organising facts and questions to be pursued The following linguistic facts are crucial: 1. MVs have retained, at least to some degree, their distinct paradigmatic status as formal preterite presents, as opposed to all other full verb paradigms. However, MVs in Modern English have done so in ways different from those in Modern German: generally speaking, Modern German modals share many syntactic properties with full verbs, while those of Modern English do not. See also 3 below. 2. MVs in Modern German can be shown to be sensitive to aspect (or Aktionsart), unlike their counterparts in Modern English. 3. Modern English MVs are far more grammaticalised than MVs in Modern German. The latter show many corresponding properties with lexical verbs. Modern English MVs do not show such properties. Rather, they behave syntactically like auxiliaries (e.g. as regards the syntax of do-support, a syntactic feature that Modern German does not share with Modern English). 4. As shown in Section 1 above, MVs in American Modern English have lost the root readings (with the exception of must and, to a certain extent, may). While the full range of MVs still exists, they are either used only epistemically (may, must, can) or exclusively as temporal auxiliaries ((shall), will). As to their root meanings, fully lexical paraphrases have replaced the modals (be permitted instead of root may, have to instead of root must with a personal subject, as well as their negations). Against the background of the radical restructuring of the lexical-aspectual verbal system in the historical development of English the following questions need to be answered: 1. What is the precise nature of the sensitivity to aspect or Aktionsart of German (and Old English) MVs? Why is it that MVs in Modern English do not share these characteristics?

Modal verbs in German and English

2. If the difference between epistemic versus root readings is the result of distributional aspect distinctions, as claimed here, does this carry over to the greatly deprived status of Modern English MVs? Is there a structural and functional explanation for this correlation? 3. Could certain distributional constraints (e.g. that epistemic modal verbs (EMVs) occur only as finite verbs) perhaps be accounted for by the general demise of contextual aspect dependencies? If true, this would bring the epistemic readings of Modern German in line with the epistemic and temporal readings of modal verbs in Modern English, and it would help explain the almost total absence of root readings in American Modern English. 4. Can a link be established between the development of the class of evidential modal verbs from perfect or perfective origins and the development of German EMVs from earlier (i.e. pre-OHG) preterite presents? 5. Is Modern English really without aspect or has a new type of aspectual system emerged? Here we could think of the notion ‘ ’: at some point this came to be expressed by the perfect (instead of the earlier present tense), which must have led to a restructuring of the aspectual system. In the rest of this article an attempt will be made to answer these questions (except for 5, which cannot be dealt with here). First, the distributional and morphological differences of the MVs in Modern English and Modern German are presented. The behaviour of German modal verbs in contexts of perfective and perfect embeddings will then be discussed, yielding the first important finding: i.e. that the deontic/root vs. epistemic/evidential distinction follows from the Aktionsart of the embedded lexical verb in Modern German, but not in Modern English. Next, the theta-role properties of modal subjects will be shown to collocate with distributional differences that set wollen and mögen apart from the other modals. The main conclusion here will be that MVs are, to all appearances, raising predicates and that control in the traditional sense is not involved. The overall conclusion will be the following: the fact that German, but not Modern English, allows infinitival forms of root modals, as well as the fact that epistemic readings are confined to finite MVs in both languages, can be explained on the grounds that non-finiteness presupposes lexical thetamarking (L-marking) and that it is only this type of theta-marking that goes hand in hand with aspectual oppositions of the kind displayed by root MVs. To the best of my knowledge, no such link has ever been established in the literature.





Werner Abraham

. The major distributional differences between English and German modal verbs The traditional view of the syntactic difference between Modern English and Modern German modal verbs is that, since Modern English modals cannot project non-finite forms, they project into the syntactic I(nflection) category (although this is contested, on a comparative-typological basis, by Rohrbacher 1999), while German root modals project into V (see Abraham 1995: 469ff.; 1998; Sprouse & Durbin 1997). Main verbs in Modern English surface in V, while modals and finite auxiliaries such as have and be surface in I. The phrase marker in (1) collapses the relevant s-structures of Modern English and Modern German. (1)

IP I’ IModern English

VP

VModern English VGerman MV Auxfinite Vlexical

V/MV/Aux

As shown in (1), the Modern English modal verb is represented under I, whereas the MV in German has the syntactic status of a full lexical, V. However, this cannot the whole story about German MVs since their deontic readings must be distinguished from their epistemic readings. In (2), a number of syntactic contrasts between Modern English and German are listed (following partly Durban & Sprouse 1998 and Reis 2000). (2) M E a. s/he sees/can(*s)/may(*s)/ must(*s) can, shall ( = )

b. *s/he musted/could/should c. *has could/*should/*would

 /

  

G sieht/kann(*t)/darf(*t)/muß(*t) kann/können, darf/dürfen, magmögen, muß/müssen; ihr seht/könnt/dürft/müßt/wollt 3 er mußte/konnte/sollte hat gekonnt/gemußt/gedurft

d. *interesting to can/may work (to+) interessant arbeiten zu können/dürfen

Modal verbs in German and English

e. *must can work f. *I can German *Work must, *Smoke may

-   

muß arbeiten können

g. *I work not I do/must not work h. *Works Mary? Must/Does Mary work? j. *Mary works hard, worksn’t she? Mary can work hard, can’t she? k. *that Claus that not sees/ can/may l. may be right/must be wrong m. *May that be right? *Shall he come?

 

Ich kann Englisch. *Arbeiten muß/ *Rauchen darf Er sieht/kann/darf das nicht

 

Sieht/Muß/Darf Klaus das?



–

p. *This may, I please, be right!



– - (-)  

daß Klaus das nicht sieht/ kann/darf mag stimmen/muß sich täuschen *Mag das stimmen? (*EMV!) Soll er kommen? (DMV/*EMV!) *Das mag gefälligst stimmen!

Note that a number of the distributional properties listed in (2) follow from the fact that the Modern English MV is base-generated in I, which does not accept items that have full lexical status. These inferable properties are given in (2c–k) above, while (2a, b, l) are independent features, which may, at least at first sight, require further explanation. I will argue presently that this is related to an ‘inherent aspect’ (i.e. Aktionsart4 ) property, present in German, but not in Modern English. Among the Germanic languages, Dutch is the most interesting language to compare with German, to the extent that it is even more conservative morphologically and has experienced greater syntactic change. Clearly, German and Dutch display the most conservative modals, whereas Modern English has evolved the most strongly grammaticalised forms and meanings, even beyond those in Scandinavian. I will briefly examine the formal and distributional criteria of Dutch modals (see in far greater detail Barbiers 1995). The table in (3) lists a few details in which Dutch and German differ, where Dutch appears to have undergone stronger grammaticalisation.





Werner Abraham

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

D Ik kan Engels   Dat kan die!  Werken moet/Roken mag. Piet moet haar *(niet). De wijn moet op. Koken dat Piet kan!

G ich kann/*darf/*muß/mag Englisch Das kann der! *Schaffen muß/*Rauchen darf. *Peter darf sie nicht Der Wein muß alle/*auf *Kochen daß P. kann/Kochen wie der Peter kann! g. Hij kan/mag/moet dit niet  Er kann/darf/*muß das nicht h. Wil/moet/mag hij dat?   Will/darf/muß er das? k. Dat Klaas dit niet ziet/kan   daß K. das nicht kann/darf/muß l. Mag/moet dat correct zijn? (-) Kann/muß das richtig sein? (DMV)  m. *Mag dat juist zijn?  *Mag/muß das stimmen? (*EMV) *Soll(te) er kommen? (*EMV) n. *Dit mag/moet heel graag  *Das mag/muß gefälligst stimmen! juist zijn! 

As illustrated in (3a–f), Dutch has gone further than German in using modal verbs as full transitive verbs. The diachronic processes behind this difference possibly instantiate cases of pragmatic ellipsis, which have become fully grammaticalised as full verb uses. The examples in (3l–n) demonstrate strong clause-type constraints on epistemic readings both in German and Dutch. More precisely, the epistemic uses of MVs in Dutch and German are restricted to declaratives. This is in itself a non-trivial generalisation. A tentative conclusion to be drawn from this is that the position of the EMV is so high in the clause structure that it does not permit the full range of clause type operations (i.e. it excludes questions and imperatives). At this juncture it should be pointed out once again that Modern American English has virtually lost the root uses of its MVs. What could be the reason for this deficiency? I have argued above that it can be attributed mainly to the absence of the aspectual/Aktionsart properties that the root-deontic MVs (see Abraham 2000) have retained in German and Dutch, together with their original main verb status. The next sections provide a closer look at aspect in relation to MVs.

Modal verbs in German and English

. Syntactic reflexes of the root-deontic vs. epistemic distinction in German In the following discussion of root vs. epistemic readings of the modal verbs knowledge of the massive literature on this topic is presupposed. No attempt will be made to motivate the terminology (root/deontic vs. epistemic/inferential/subjective/evidential, etc.) or define its use in this article.5 Rather, in line with modern syntactic methodology, the meaning distinctions will be taken to be inferable from the totality of the distributional characteristics provided above and below (see, however, Section 5.4 and Section 6 for semantic considerations concerning event structure, argumental theta assignment, and scope). Suffice it to say here that the syntactic reflexes of the root vs. epistemic distinction in German established above show a surprising dependency on the distributional criteria spelled out in (3a–g). Furthermore, aspect and lexical Aktionsart play a decisive role, as shown in (4)–(11) (repeating Abraham 1995, Ch. 6: 472f.; (7) has been taken from Sprouse & Durbin 1998). The symbols E and T in the characterisation in (6a–c) stand for the Reichenbach notation of event tensing (E = event, T = temporal point of the utterance; ‘E,T’ = present tense reference; ‘E_T’ = past tense reference; ‘T_E’ = future event reference; ‘Perf ’ = perfective, i.e. event-bounded). Where both readings are possible I provide glosses and translations. (4) MV-: a. - Er hat(te) viel Geld verdienen wollen/müssen/sollen He has/had much money earn will/must/shall DMV-reading: ‘He wanted/had to earn much money’ b. Das scheint so sein zu müssen That seems so be to must c.  Er wollte/mußte/sollte viel Geld verdienen He would/must/should much money earn DMV-reading: ‘He wanted/had to earn much money’ EMV-reading: ‘He pretended/was supposed to/ was said to earn much money’ d. Das muß so sein zu scheinen

DMV EMV +

*

(0)

(+)

+

+

(0)

(*)

The raising verb scheinen ‘seem’, as shown by (4b, d), patterns with EMV (but not DMV) in its non-finite use.





Werner Abraham

(5) A : a. Er will/soll/muß schlafen [–Perf] He will/shall/must sleep EMV-reading: ‘He pretends/is supposed to/is said to sleep’ DMV-reading: ‘He wants/must/has to sleep’ b. Er will/soll/muß einschlafen [+Perf] He will/shall/must in-sleep (= fall asleep) EMV-reading: ‘He pretends/is supposed to/is said to fall asleep’ c. Er will/soll/muß Geld verdienen [–Perf] He will/shall/must money earn DMV-reading: ‘He wants/must/has to earn money’ EMV-reading: ‘He pretends/is said to earn money/ It must be concluded that he earns money’

DMV EMV + +

+

*

+

+

(6) MV - : a. Es ist wichtig in der UB arbeiten zu können/müssen/dürfen [E,T] + It is important in the library work to can/may/must b. *Es ist dem Angeklagten sehr wichtig in der UB gearbeitet haben zu können/müssen/dürfen [E_T] * . . . worked have to can/must/may c. *Es ist dem Angeklagten sehr wichtig in der UB arbeiten können/müssen/dürfen zu werden [T_E] * . . . worked have can/must/may to become (7)

(8)

E ()  a. Von ihm muß/soll/will viel Geld verdient werden By him must/shall/will much money earned become DMV ‘It is necessary for him to earn much money’ EMV ‘It is said that he earns much money’ b. Von ihm muß/soll/will viel Geld verdient worden sein By him must/shall/will much money earned become be MV-  MV a. Er muß diesen Brief geschrieben haben mögen/dürfen He must this letter written have may ‘He must have written this letter’ b. Er müßte diesen Brief geschrieben haben mögen/dürfen He must-.. this letter written have may ‘To all appearances, he wrote this letter’

*

*

*

+

+

*

+

*

+

*

+

Modal verbs in German and English

DMV EMV c. Es muß/müßte sein, daß man sagt, er * + hat diesen Brief geschrieben It must/must-. be that one says he has this letter written ‘To all appearances it is said that he wrote this letter’ (9)

V : a. Er muß/müßte diesen Brief geschrieben haben können He must this letter written have can b. Er muß/müßte diesen Brief haben schreiben können/*gekonnt He must this letter have write can/could c. Er muß/müßte diesen Brief schreiben haben können/*gekonnt d. Weil ihn Peter belehren zu müssen glaubte Because him Peter teach to must believed

(10) MV =    a. Sie mag/kann/*darf/*muß/*soll Deutsch She may/can/must German b. Er tut Englisch können/mögen He does Modern English can/may c. Ich kann/mag(*darf/*muß) Englisch I can/may(may/must) English d. Das kann/darf/mag/muß der! This can/may/must he e. Der Wein muß weg/*auf. The wine must away/up f. Kochen wie Peter kann! Cook as Peter can (11) C     a. Mag/Kann/Darf/Muß das stimmen? May/can/must this be right b. Aber wer muß sich getäuscht haben? But who must himself erred have c. Ach, wenn er sich doch täuschen könnte! Oh, if he himself yet err could d. Das mag/kann gefälligst stimmen! This may/can be right

+

*

+

*

+ +

* *

(+)

(*)

(+)

(*)

+

*

+

*

+

*

+

*

+

(*)

+

(*)

+

*

+

*





Werner Abraham

A caveat is in order with respect to the data in (10a, c, d, e): none of the expressions (German Englisch können ‘to can English’; Dutch werken moet ‘to work must’, etc.) have a wide pragmatic extension, that is, their use is idiomatically restricted. Not much can therefore be concluded for the syntax. These idiom-like structures are not possible in Modern English, but they are in Dutch and in German (although to a lesser degree in German; compare Barbiers 1995: 170–173; 189; 192; 204). Further notice needs to be taken of the fact that in Modern English there are no root or epistemic readings for will and shall, but only temporal ones (to the extent that shall survives at all. Recall that in American English the modal root readings except for must are on their way to becoming obsolete (see Table 1 above). Except for idiomatically frozen uses (such as e.g. May I?), be permitted is preferred to root may, be able is preferred to root-alethic can, and have to is increasingly used instead of root must. What this means for the status of DMVs in Modern (American) English will be evaluated later.

. Deontic MVs, full verb status, and finite auxiliaries . The prenominal active participial construction If German DMVs behave syntactically like full lexical verbs, we expect a distributional behaviour in keeping with full lexical verbs, also in prenominal attributive position. This is fully borne out in German (also see Sprouse & Durbin 1998), while in Dutch this is only partly so.6 None of these constructions are possible in English. (12) a.

alle Deutsch sprechenden Studenten all German speaking students b. alle Deutsch (sprechen) könnenden/müssenden/dürfenden Studenten all German speak canning/musting/maying students c. alle Deutsch gesprochen habenden Studenten all German spoken having students d. alle gerade angekommen-(seiend-)en Studenten all just arrived being students e. alle Deutsch gelernt haben müssenden/könnenden/dürfenden Studenten all German learned have musting/canning/maying students f. ( ?) alle Deutsch lernen müssen/können/dürfen habenden Studenten all German learn must/can/may having students

Modal verbs in German and English

g. alle Deutsch lernen *gemußt/*gekonnt/*gedurft habenden Studenten all German learn musted/could/might having students

Despite this pervasive similarity to full lexical verbs, DMVs in Modern German do show some differences from full lexical verbs. In contrast to Dutch, German MVs are more restricted with respect to intransitive usage and infinitival embedding, respectively (cf. Dutch Dat kan/mag vs. German Das kann/darf *(sein)). . Non-standard embedding under (and periphrasis with) tun The periphrasis with tun is common in colloquial, spoken German, although it is not unrestricted (see (13b)–(14b) and (15b) below). Sprouse and Durbin (1998) have investigated the collocations of MVs with tun. However, it is doubtful whether relevant conclusions can be drawn from the examples below beyond the fact that there are – somewhat idiomatically restricted – full verb uses (see Abraham and Conradie 2001: 81). (13) a.

Er schreibt gut Englisch, aber er spricht es nicht so gut He writes well English, but he speaks it not so well b. Er tut gut Englisch schreiben, aber er tut nicht so gut sprechen He does well English write, but he does not so well speak

(14) a.

Er kann gut Englisch He can well English ? b. Er tut gut Englisch können He does well English can

(15) a.

Er hat Englisch gesprochen He has English spoken b. **Er tut gut Englisch gekonnt/gesprochen haben He does well English spoken have

From a minimalist point of view tun differs from finite auxiliaries in that it functions as a last-resort strategy. As a finite auxiliary, haben, the auxiliary of the perfect, thus takes priority over tun when raising to T (see 15b). (16) a.

Er muß Englisch gelernt haben He must English learned have ‘He must have learned English’ b. **Er tut Englisch gelernt haben müssen He does English learned have must

. . . only EMV





Werner Abraham

c. **Er tut Bulgarisch sprechen müssen, weil ich immer nur da verstehe he does Bulgarian speak must. because I always only da understand (17) *Er kann gut Englisch sprechen tun He can well Modern English speak do (18) *Englisch wird gut sprechen getan English becomes well speak done

These facts reconfirm the full verb status of MVs in the syntax of substandard German, in which discourse factors play a more prominent part than in standard German (cf. Abraham & Fischer 1998). . Passivisation The general conclusion about the full verb status of MVs reached above is further supported by passivisation facts. As (19) shows, DMV uses are grammatical under passivisation and their full verb status is retained. EMV-passives, however, are ungrammatical (also shown in more detail in Reis 2000: 5). Again, Modern English does not allow MVs in these constructions at all, and they are severely restricted in other Germanic languages. Dutch, however, is totally on a par with German (cf. Barbiers 1995). (19) DMV =    a. Hier wird/wurde alles versucht Here becomes/became all attempted ‘Here everything is/was attempted’ b. Hier wird/wurde alles gekonnt/gewollt/gedurft Here becomes/became all could/would/might ‘Here everything is/was possible/wanted/allowed’ c. Auch in der Schweiz wird/wurde Deutsch gekonnt Also in Switzerland becomes/became German can-part ‘Also in Switzerland people know/knew German’ d. Selbst in der Schweiz wurde dies gemußt (d.h. die Bankguthaben offen zu legen) Even in Switzerland became this must-part (i.e. the bank accounts open to lay) ‘Even in Switzerland this was required (i.e. to open up the bank accounts)’

Modal verbs in German and English

Summarizing, we may say that MVs in German and in Dutch, but not in English, behave like transitive verbs in all relevant syntactic respects. Our assumption has been so far that this is due to the fact that (American) English modals are on their way to losing their root meanings and retain only epistemic and finite uses. It remains to be seen, however, whether there are semantic correlates of this syntactic typological distinction. In the next section, I investigate the relation between the aspectual semantics of modal verbs and their syntactic structure. . The event structure of MV: Terminativity Since my discussion aims at providing a link between aspect and syntactic distinctions, it is necessary to present briefly, but succinctly, a formal account of aspectual structure. We will be concerned with the distinction of perfectivity (also called resultativity, terminativity, telicity7 ) vs. imperfectivity (or durativity), and logical notions such as ‘boundedness’ vs. ‘unboundedness’ will not suffice to make this distinction. In what follows, I will propose that (in)transitive verbs can be classified in terms of event structure, which I conceive of as distinguishing between resultativity and ongoing situations (statal or dynamic). This classification will be shown to carry over to the syntactic distinction between DMV and EMV (see for an early presentation Abraham 1990). The vocabulary adopted here accords with Reichenbach’s tense and event logic. Resultative events have a bi-phasic structure as in (20a), whereas imperfectives are characterised as in (20b). (20) a.

terminative verbs:

b. non-terminative verbs:

In (20), t1 , tm , tn are points on the temporal axis representing the event components (or phases) E1 and E2 . The event structure of terminative verbs is biphasic: it contains an approach phase as well as a resultative phase (∀tx:x,(1 – n) (t1 – tm )(E1 ) ⊂ (tm – tn )(E2 )), where tm is a referential point belonging to both event phases, E1 and E2 , simultaneously. On the other hand, the event structure of non-terminative verbs is mono-phasic: ∀tx:x,(1 – n) (t1 –tm )(E1 ) = (tm –tn )(E2 )). The representation in (20a) describes terminative events. The time-axial structure is bi-phasic, such that the first phase is bounded by tm . (20b), on





Werner Abraham

the other hand, describes unbounded situations, i.e. non-terminatives, statal as well as dynamic (mono-phasicness as genus proprium). (21) and (22) show how the verbal arguments are to be mapped on the event phases. The following abbreviations are used: tV = transitive verb; eV = ‘ergative’ verb (terminativeintransitive); iV = intransitive verb (non-terminative); EvP = event passive; StP = statal (adjectival) passive; PPP = perfect participle passive; PPA = perfect participle active; iA = internal argument (direct object); eA = external argument (subject). (21) a.

terminative transitive verb, tV: viz. einschläfern ‘make sleep’ eA: eA(t1) ∩ . . . ∩ eA(tm-1 ) t1 tm tn iA: iA(t1 ) ∩ . . . ∩ iA(tm ∩ . . . ∩ iA(tn )

b. terminative intransitive verb, iV: viz. einschlafen ‘fall asleep’ iA: iA(t1 ) ∩ . . . ∩ iA(tm ) . . . ∩ iA(tn ) t1

tm

tn

(22) non-terminative (i)tV: viz. schlafen ‘sleep’ eA: eA(t1)∩. . .∩eA(tn ) t1

tm

tn (iA: iA(t1 )∩. . .∩iA(tn ))

It should be noted that quasi-auxiliaries such as scheinen ‘seem’ and pflegen ‘used to’, which are raising verbs according to their derivational properties, are mono-phasic. This makes these verbs categorially distinct from terminatives with respect to their event structure. Their mono-phasic character accords well with the idea that EMVs as well as the epistemic readings of drohen ‘threaten’ and versprechen ‘promise’ are semantically bleached raising verbs (as in the comparable English construction The heat promises to break, which has a futural meaning; see Palmer 1986; Bybee 1985; Coates 1995). In their control readings, on the other hand, verbs like drohen and versprechen (+ complement verbs) are bi-phasic in the sense of (21a, b). The difference between the mono-phasic and bi-phasic readings is represented in (23a) vs. (23b). Note the similarity of the bi-phasic event structure for DMVs in (23b) with that in (20a)/(21a). (23) a.

EMV/raising-V:

b. DMV/control-V: c.

argument match: A(t1 ) ∩ . . . ∩ A(tm)

Modal verbs in German and English

d. event identification: ¬t1 (E) ∩ . . . ∩ ¬tm (E) ∩ tn (E); i.e. tm has the event characteristics of the approach phase as well as of the resultative phase.

Recall that an epistemic reading is not available for embedded, i.e. non-finite terminative predicates; see (5b). This is accounted for by (22) and (23a) in that the MV under epistemic interpretation, for lack of event-structural congruency (epistemic modals have a monophasic event structure, while root modals have a biphasic structure), cannot project any event structure of its own onto the structure of the embedded perfective.8 All an EMV does in a verbal complex is provide person and number agreement with the subject, taking over fully the theta grid of the embedded predicate. In this EMVs are similar to auxiliaries (raising predicates). As a consequence, the wide scope of an E-modal (raising modal) cannot ‘overwrite’ the event structure of, or de-terminativise, a terminative verb. This is the general yield of the matching mechanism between MV and the aktionsart feature [α term] of an embedded full predicate. The DMVreading, on the other hand, provides an biphasic event identification of its own as in (23b), overwriting the force of its argument grid onto the embedded, equally biphasic event structure of the perfective predicate. Now consider the similarity between the event structures for terminative full verbs (as in (21)) and DMVs (as in (23b)). Both lexical terminatives and modals in their DMV-readings are bi-phasic. As represented in (23b), event identification under the DMV-reading is not possible in the first phase (t1 –tm ); the relevant event E, as denoted by the full lexical predicate, does not hold until the inception of phase 2 (tm – tn ). As in the case of lexical terminative verbs, the medial temporal reference point, tm , is a border point satisfying both event characteristics, i.e. that of the DMV and that of the lexical predicate. This boils down to saying that the complete event characteristic under the DMV-reading satisfies the charaterisation of bi-phasics, by virtue of the special nature of the border point (tm ). In this way, the intensional link between approach phase and result phase for terminative predicates is established. Another parallel consists in the fact that the event references of DMVs as well as terminative transitive and intransitive verbs are satisfied in the first phase and also in the border point of the time axis. There is therefore a clear convergence of terminativity with the DMVreading, which is reflected by similar event structures. The data in (6) show beyond doubt that the non-terminative EMV is incompatible with an embedded terminative full verb. (24a, b) summarizes the links between syntactic be-





Werner Abraham

haviour (distribution as well as scope) and aspectual classification (Abraham 1995, Ch. 6; Abraham 2001a). (24) a.

the root MV, under a terminative reading, projects narrow, verbinternal, scope. b. the epistemic MV, under a non-terminative reading, projects wide clausal scope, which has no lexical effect and therefore leaves the lexical event characteristic (including theta grid) of the embedded predicate unaffected; in other words, the epistemic MV has no de-terminativizing effect on a terminative verb – there is no local ccommand relation between the high functional epistemic node and the theta-role bearing lexical aspect node inside VP.9

Having established a link between the structural positions of DMVs vs. EMVs and their aspectually distinct meanings, let us now turn to the semantic projections of their arguments. The main question to be asked is: do DMVs and EMVs have different characteristics in terms of semantic roles? Note that, if this turns out to be the case, this would yield a further correlation between deontic or epistemic reference, on the one hand, and distinct event identification, on the other hand.

. Argument structure and modal verbs . The status of the subject and semantic role assignment Vikner (1988: 12ff.) has suggested that DMVs but not EMVs in Danish assign a semantic role to their subject. This is a fundamental distinction, which is in principle open to empirical verification in other languages. Note that, in order to make this distinction more meaningful, we would like to relate it to the distributional distinctions seen to hold for German DMVs and EMVs. The assumption that a DMV assigns a thematic role to its subject raises a specific conceptual difficulty. If, as is commonly assumed (see, e.g., Evers 1975), modal verbs are raised into the lexical verb projection to form a verbal cluster, two thematic roles might collide, so that the complex verb cluster assigns two theta-roles, e.g. Agent and Patient, to a single argument. How is this problem to be solved? Vikner’s (1988) suggestion (following Zubizarreta 1982) is that, in the case of verbal clusters, the subject may have, in addition to its main and strong thematic role assigned by the full lexical verb, one, but not more than one, extra

Modal verbs in German and English

and weak thematic role. Some Danish examples are given in (25). Note that Danish, like Swedish and Norwegian, has two passives: a periphrastic one using blive ‘become’ as the auxiliary, and the synthetic s-passive (Vikner 1988: 13ff.). The crucial observation is that the two passives have different distributions under embedding under the two types of modals. (25) a.

Hun vil blive arresteret he will become arrested ‘He will be arrested.’ b. Hun vil arresteres he will arrested- ‘He wants to be arrested.’

. . . *DMV, EMV

. . . DMV, *EMV

Since Vikner assumes that the (auxiliary) uses of Danish blive, få and komme assign extra semantic roles, the subject hun in the deontic reading of (25a) would collect three thematic roles (one from vil, one from blive, and a third from arresteret), which is impossible, irrespective of whatever assumption is made with respect to the assignment of semantic roles. A root reading of (25a) is therefore ungrammatical, unlike the epistemic reading of (25a), in which the modal does not assign a semantic role of its own. (25a) will receive an epistemic interpretation if the auxiliary vil does not assign a third semantic role to the subject, hun. In the same vein, the subject hun in (25b) receives two semantic roles, but these are different from those in (25a): one semantic role is assigned by the participle of the main verb, arresteres, under lexical government, and the second (weaker) one is assigned by vil. There are, as far as I can see, two other ways of avoiding this violation of the theta-criterion (‘Each thematic role must be assigned to one and only one argument, and each argument bears one and only one theta-role’) when we have embeddings under DMVs vs. EMVs. One would be to block double thematic assignment (by DMV as well as the embedded main verb) unless the two semantic roles are identical. The second, more or less canonical, path to follow is to assume two different mechanisms that are responsible for the clustering of modal verbs with their embedded lexical verbs: raising for EMVs (where the problem of double assignment does not arise, since an EMV does not assign a semantic role to the subject in the first place) and control for DMVs, which assign their own subject theta-role.10 The control mechanism is spelled out in (26a, b) for drohen ‘threaten’ and versprechen ‘promise’ as well as the root meanings of the true modals wollen/müssen/sollen ‘will/must/shall’. [PRO = empty subject of the embedded infinitival clause.]





Werner Abraham

(26) a.

Eri drohte/versprach [PROi das zu tun] he threatened/promised that to do ‘he threatened/promised to do that.’ b. Eri will/muß/soll [PROi das tun] he will/must/shall  that do ‘He wants/must/has to do that.’

Drohen/versprechen are canonical subject-control verbs. In their full lexical meaning, they assign an -role to the subject. The same holds for the modal verbs in (26b) in the deontic readings given. In (27a, b), no root reading is possible. Instead, these two sentences receive epistemic readings, with the subject assigned the  or  role by the embedded lexical verb. (27) a.

Eri drohte/versprach [ti sichi zu verachten] he threatened/promised  to despise ‘unfortunately/fortunately, it looked as if he was going to despise himself.’ b. Eri will/muß/soll [ti sichi verachten] he will/must/shall  despise ‘He is going/cannot but/should despise himself.’

The selection of semantic roles by German DMVs is represented in the lefthand column of (28). The columns DMV and EMV represent the equivalents in (American) English. Note that American English uses modals to render the deontic readings of the German MVs only in two cases (i.e. must and shall). In American English, must, may, and can are the only members of the class of modals left. It is likely that such occurrences, which are unsupported by full paradigmatic status, are doomed to eventually vanish altogether. (28) S    M G

A E 

a. b. c. d. e.

DMV wollen mögen11 müssen sollen dürfen12

 [+hum]  (+hum)  ⁄  ⁄  ⁄

DMV wish be permitted to must (shall) be to

f.

können

 ⁄

be able

EMV be supposed to be supposed to be supposed to be supposed to may be permitted can (alethic)

Modal verbs in German and English

According to (28a, b), wollen and mögen are not on a par with the other MVs since, in their deontic uses, they project agentive subjects. This is supported by the finding that non-volitional MVs are demonstrably raising verbs also in their non-epistemic readings (Reis 2000: 5–6). Wurmbrand (1998, 1999) has argued that, in all their uses, irrespective of their root vs. epistemic function, modal verbs are raising verbs. Although different conceptually, Wurmbrand’s view is comparable to those of IJbema (1997) and IJbema & Abraham (2000), who argue that what are called ‘control’ complements do not require the concept of PRO. Rather, the infinitival preposition (German zu, Modern English to) and the infinitival suffix (German -(e)n) (as much as the circumfix of the passive participle) occupy the Spec of vP and Spec of VP respectively, which would block the embedded subject from surfacing in those Spec-positions. My claim that root readings, but not epistemic readings, go hand in hand with theta assignment may seem disputable. Consider the sentences in (29)– (30), which appear to be incompatible with this claim. (29) The key must always be kept on this hanger. (30) At least two guards must be present.

The argument against theta assignment in root MV-readings goes as follows: in these sentences, no single argument is projected that can be taken to be the bearer of the obligation expressed by must.13 Consequently, the deontic meaning cannot be linked to any specific theta role. Note, however, that, even under this argument, the subjects in (29) and (30) have a theta role; compare also (28c) above. More importantly, there is clearly an ‘obliger’ involved in (29) and (30), although it is not represented in the argument grid of the verbs. I have argued elsewhere (Abraham 1995: 480ff), following Zubizarreta (1982) and Vikner (1988), that root modals of this type are causatives involving covert agents responsible for the obligation that is part of the meaning of deontic / ¨. In a way, this assumption concerning the covert causative character of must is similar to Pesetsky’s assumption that psych, or preoccupare, verbs are covert causatives and, as a result, cannot be regarded as unaccusatives (Pesetsky 1995; contra Belletti & Rizzi 1988), since causatives have an argument structure and theta-grid totally different from those of unaccusatives. Whatever the solution to the problem of theta-assignment and DMVs may be, it is clear that EMVs do not assign a thematic role to their subjects.14 How does the difference between DMVs and EMVs in terms of semantic roles and arguments relate to the empirical generalizations established above?





Werner Abraham

Note that the claim that DMVs assign a subject theta role, while EMVs (like other auxiliaries) do not, has an interesting parallel in the development of modals in the history of English and the distinction between root/deontic and epistemic modals in German (see Sections 4 and 5.1–4 above). I am assuming here that verbs that have been reanalysed from lexical verbs to modals become stative as a consequence of the loss of certain components in their extended projection, that is ModP for modals and vP/Aktionsart – where resultative (perfective) Aktionsart is represented as a small clause predicate – for quasi-modal uses of lexical verbs. Stative verbs, which can never be agentive, are prone to dethematisation, i.e. to an exclusively functional (i.e. raising) interpretation (Roberts 2000: 5), as opposed to eventive verbs, which may involve thematic agentivity. Note that the modal verbs in (28) are parallel to the first class of ‘main verb modals’ in van Kemenade’s (1987) classification of Old English modals, which have subject theta-roles in their own right and, consequently, correspond to control verbs, as opposed to those modals which have no theta-role for the subject and therefore pattern like raising verbs (such as seem). This is nothing new, of course (see, for example, Abraham 1990). Van Kemenade’s ‘main verb modals’ can thus be brought into line with deontics, which are control verbs, whereas epistemics do not project an external theta-role and are raising verbs. EMVs can only occur as finites, which, in minimalistic terms, implies that they have to raise to AgrS0 . DMVs, on the other hand, also have non-finite forms and can therefore be embedded under EMVs. This yields the following picture with respect to clustering of modal verbs. (31) N-   EMVs a. *EMV embedded under DMV/EMV b. DMV embedded under EMV c. DMV embedded under DMV: not excluded on theoretical grounds (but may nevertheless be blocked on the stylistic grounds of horror aequi)

The theoretical option in (31c) appears to be supported empirically in languages that have ‘aspectual’ DMVs such as German and Dutch, but not in languages like Modern English, which cannot project non-finite modals. We have arrived at the following conclusions: while German (and Dutch, West Frisian, and Yiddish) allow all the options described in (31a–c), Modern English does not. This is due to the fact that Modern English has lost its DMV uses, retaining only EMV and purely temporal uses. I have argued that this development is a consequence of the loss of the aspectual properties of English

Modal verbs in German and English

MVs. This position, reached on empirical-comparative grounds, is in contrast to Lightfoot’s thesis that the emergence of the modern MVs in English is solely rooted in syntax (Lightfoot 1979). . On the interdependence of the thematic properties of MV and the Aktionsart nature of the embedded main verb From their inherent resultative character it follows that terminative transitive verbs (tVs) are agentive. Terminative intransitives (iVs), on the other hand, are typically non-agentives, in that their distributional properties identify them as passive-like, or ‘unaccusative’ (for the class of these phenomena, both lexical and syntactic, in German see Abraham 1990). If unaccusative, or ergative, verbs (henceforth ‘eVs’) are passive-like, their subjects are TH(emes), or PAT(ients), but never AG(ents). Note that in German eVs are taken to be terminative iVs (Abraham 2001a, b). The selection of semantic roles of DMVs in (28) as well as the lexical opposition based on the feature [+/– intention of the subject referent] is reflected in the properties of the two terminative classes of main verbs. Wollen ‘will’ is a member of the class of terminative tVs, while müssen ‘must’ and dürfen ‘may’ (and possibly also sollen ‘shall’) are classified as eVs (i.e. terminative iVs). See (32)–(33) below, in which EXP = Experiencer, Θ1 is the thematic role assigned to the matrix subject, Θ2 is the thematic role assigned to the embedded subject, and  is an empty category in the embedded subject position. (32) a.

Siei muß/soll i ihn ärgern/unterhalten she must/shall him annoy/entertain ‘She must/has to annoy/entertain him.’ Θ1 =  ⁄ (x) =  (x) =  (x) Θ2 = ⁄ (x) =  (x) or Θ2 = ⁄ (x)

DMV

However, the assignment of ⁄ to the embedded subject in an epistemic reading is impossible; see (32b). (32) b. Siei muß/soll i ihn ärgern/unterhalten ‘She must be annoying/entertaining him.’ Θ2 =  (x) (33) a.

Siei will i ihn ärgern/unterhalten she will him irritate/entertain ‘She wants to annoy/entertain him.’

EMV

DMV, ?EMV





Werner Abraham

Θ1 =  (x)=  (x) Θ2 = ⁄ (x) or Θ2 =  (x) b. Er will ein guter Syntaktiker sein DMV/EMV he will a good syntactician be DMV-reading: ‘He wants to be a good syntactician.’ EMV-reading: ‘He pretends/imagines to be/being a good syntactician.’ c. Er will den Marathon (gut) laufen DMV/EMV he will the marathon well run DMV-reading: ‘He wants to run the marathon well.’ EMV-reading: ‘He pretends to be a good marathon runner.’ It should be noted that the epistemic reading of the agentive verb wollen ‘will’ is either restricted to some property reading of the embedded infinitive, or else the reference points of the speech act and the event coincide. Then, and only then, is an EMV reading the correct interpretation. Let us try to generalise this observation further. See (34b) with alethic schreiben können ‘be able to write’, which supports this assumption. Furthermore, (34b) shows that only under some property reading do we arrive at an acceptable interpretation. This also holds for agentive MV wollen. However, in the case of the non-agentive MVs müssen ‘must’ and sollen ‘shall’, no root reading is possible, since it would be incompatible with predicates expressing a property reading. Thus, only epistemic interpretations remain. Compare (34a–c) with (33b–c). (34) a.

Er muß/soll ein (guter) Syntaktiker sein he must/shall a good syntactician be b. Er muß/soll gute Aufsätze schreiben können he must/shall good essays write can c. Er muß/soll den Marathon (gut) laufen können he must/shall the marathon well run can

??DMV, EMV

DMV, EMV ??DMV, EMV

(34b) is ambiguous. Under the DMV-reading, the reference of the embedded infinitival may be to an event in the future, whereas under the EMV-reading the reference of the infinitival must be to the time of the speech act. (34a), however, can only be epistemic, since stative sein-predications are incompatible with a deontic reading. These distributional facts allow for a number of conclusions, which will be taken up in what follows.

Modal verbs in German and English

. Comparative syntax If non-finite embedding is tied to theta-bearing subjects, an interesting consequence follows. Consider the restrictions on the embedding of MV in (31a–c), repeated in (35) below. (35) N-   EMVs a. *EMV embedded under DMV/EMV b. DMV embedded under EMV c. DMV embedded under DMV: not excluded on theoretical grounds (but may nevertheless be blocked on grounds of horror aequi)

The theoretical option of (35c), i.e. iteration of MVs, appears to be supported empirically in languages that have ‘aspectual’ or Aktionsart-supported DMVs, as in German and Dutch, but not in Modern English. It is usually assumed that the restriction in Modern English is due to the auxiliarisation of Modern English MVs. For German, Diewald (1999) claims that the deficiency of the inflectional paradigm of MVs has also contributed to a certain degree of auxiliarisation of the modals and to the restrictions on epistemic readings in particular. This argument appears to beg the question, however: what we would rather wish to say is that, in purely syntactic terms, the gradually emerging failure to license non-finite MV-forms is caused by the loss of AspP as well as the fact that MV raise to a higher position than VP (Roberts 2000: 3). Either condition can be reduced to the failure to assign a subject theta role, which holds for EMVs in German and Modern English, where the modal undergoes obligatory raising to at least TP/AgrP. It stands to reason that, if syntactic licensing is not possible, theta saturation of the subject as a lexical requirement needs to be provided. Such lexical saturation is not required for Modern English root modals given that VP-licensing of the external theta role is unnecessary. In German, as I argued, the lexical Aktionsart property saturates exactly this lexical theta criterion. Let us now return to (1) on page 24, my first attempt at the syntactic representation of MVs in Modern English vs. German and DMVs vs. EMVs, respectively. From what has emerged so far in terms of semantic distinctions and surface distributions, the representation in (1) cannot be correct. Consider Modern German (and Dutch) first. One way to give descriptive credit to the difference between DMV and EMV on the basis of their distinct behaviour under aspectual or Aktionsart conditions is that shown in (36) vs. (37).





Werner Abraham

(36) EMV German mag ‘may’: IP Spec

I’ I

vP Spec

v’ v

VP Spec

V’ V

Er

t

magi

ti dich

auslachen

Since aspectual terminativity, or Aktionsart, has turned out to be involved for DMVs, one may want to include an aspect phrase (=PerfP) close to the vP shell. (37) DMV German (nicht) mögen ‘(not) like to’ IP Spec

I’ PerfP

I Spec

PerfP’ vP

Perf Spec

v’ v

VP Spec

V’ V’ V

Ich magi t

ti

EC

dich

nicht

auslachen

Modal verbs in German and English

According to this structure, the root MV starts in the head of the PerfP and moves to I. This treatment of deontic modals has already been referred to in Abraham (1990) and is also widely supported for modals in other languages (see e.g. Picallo 1985 and, more recently, Barbiers 1995: 184–203 as well as Wurmbrand 1999). This accounts for the complementary distribution of EMVs and DMVs but it introduces an extra phrasal category to make the deontic modal into an aspect marker. Another account – and this would bring their syntax even closer to their semantics – can be based on the event-semantic status of perfectives as bounded verb meanings, i.e. their resultative readings. The idea would be that the syntax of DMVs, in contrast to EMVs, includes a small clause (see also Abraham 1990, 1995). But other syntactic characteristics do not appear to be reconcilable with this position. If it is true, as is commonly assumed, that epistemics are raising verbs, while deontics are controllers (PRO-syntax), it needs to be investigated how a PRO-syntax translates into a small clause syntax. Many linguists (among whom Roberts 2000) have remained non-committal on the exact syntactic status of epistemics as opposed to deontics. Wurmbrand (1999) opts for raising structures for all MVs except for will/wollen, which she takes to have full verb status (with the subject theta role as ; see (28) and (32) above). It should be clear now that DMVs, which are absent in (American) English, have a syntax different from that of EMV. Modern English, however, does not make any such distinction in a systematic, paradigm-supported way.

. Conclusion Our empirical findings were as follows. First, we established clear syntactic differences between the modals in German and English, the main distinction being that modals in German, but not in English, have full verb status and as such occur also non-finitely (both as infinitives and as participles) and in non-finite dependent position. The second and most important result of this investigation is that Aktionsart conditions triggering the split between DMVs and EMVs in Modern German do not to the same extent determine the behaviour of MVs in Modern English. English modals do not obey the same distributional criteria simply because they have no syntactic status as non-finites. Since German epistemic modal readings are sensitive to features of aspect/Aktionsart, there must be a link between their syntax and their aspect-event structure. Furthermore, if it is true that the specific behaviour and aspectual meaning of root modals is





Werner Abraham

based on a specific theta grid different from their epistemic counterparts, there would then be two sets of four contrasting properties: Set 1: perfectivity, root reading, lexical theta assignment, and control Set 2: imperfectivity, epistemic reading, absence of lexical theta assignment, and raising Such a view has not yet been proposed in work on the topic by scholars like Lightfoot, Palmer, Warner, Roberts, Giorgi & Pianesi, to name but a few. It has been shown elsewhere that true aspectual behaviour is relatively recent in Modern English (van Gelderen 1999). If we accept this, it cannot have had an impact on the possibilities of embedding under modal verbs. Since the older stages of English were on a par with German, a radical restructuring must have taken place (Abraham 2001b). I have tentatively proposed (but not really shown here; see Abraham 2001b) that the trigger for the far-reaching auxiliarisation of MVs in English, but not in German and Dutch, was the demise of verbal Aktionsart, and along with it lexical terminativity, in late Old English and Middle English – no doubt as part of a wider scenario involving all kinds of morphological and paradigmatic changes. I have suggested that the tendency in American English of replacing DMVs by other expressions, as well as the fact that will and shall serve no deontic or epistemic purposes any longer, is merely the final stage in this development.

Notes * The following people have commented on presentations of this paper: Geoffrey Pullum, Pieter Seuren, Marga Reis, Ian Roberts, Elly van Gelderen as well as anonymous reviewers. Geoffrey Pullum and Elly van Gelderen, the latter in particular, encouraged me to consider the relevance of corpus data. . I have disregarded can since it is often alethic and thus does not provide the same range of epistemic vs. root meanings. . See Old English (ge)niman ‘take’, gemacian ‘make equal’; cf. Brinton & Akimoto (1999: 10; 21). However, such verbs no longer appear in Early Modern English, as documented by Görlach (1991). . While integrating MVs into the overall syntax of verbs, German (and Dutch), modals in both languages constitute a clearly identifiable class. Not only does the 3rd person singular present tense lack the canonical ending in -t , but their singular and plural present tense forms also reflect the original 2nd and 3rd preterit stems of the original preterit presents. Compare the following German-Dutch modal pairs: kann-können/kan-kunnen, mag-mögen/mag-mogen, /zal-zullen, will-wollen/, muß-müssen. This can be related to the

Modal verbs in German and English

Middle High German /a/-/u/ distinction between 2nd and 3rd verbal root vowels in the 3rd class preterit paradigm. In Modern English, all such distinctions are absent. . The diathetic categories of Aspect and of Aktionsart are taken to be separate categories. The most prominent distinctions are as follows: Level of distinction Projection level of diathesis in question Repeated diathesis marking possible Analytic morphology (in terms of clause-structure and word-derivation) Perfectives are resultatives Perfectives provide present tense gap

Aspect

Aktionsart Vo ,

Above within VP + (as in Russian) +

Inside Vo – word syntax – (as in German) +/–

+ (cf. Russian) + (cf. Russian)

+ (cf. German) – (cf. German)

The chart shows both distinct and congruent features. I shall return to differences in the final section of this paper, showing that the distinctions clearly surface in Modern English too. . In fact, it seems to me that attempting this, without first establishing solid distributional and behavioural properties as provided in this article, would make us liable to methodological pitfalls such as the notorious discovery procedures. I will sidestep these issues here. For a summary of the distributional characteristics distinguishing root and epistemic modals see (24) below. . There is a strong dislike in Dutch for attributive extensions in DP such as in (12b, c, e, f, g). This might be a stylistic restriction rather than a grammatical one. . I will continue to use ‘terminativity’ as a term encompassing all aspects of resultativity, although I will limit it to the purely lexical component of aspect. This is in line with the traditional distinction between clausal aspect and lexical Aktionsart, both in the Slavic and the German literature on this subject (see, e.g., Tschirner 1991 for a recent discussion of the distinction between aspect and Aktionsart). Finer distinctions between these terms, which have been made in the literature, are of no importance here. In support of this view, see Zhang (1995: 114ff) for an identical position with respect to Chinese, in which aspect plays a prominent role. . For further identification of events and the projection of arguments for duratives, see Abraham (1990). I restrict my discussion to root modals, i.e. DMV. . This accords well with Cinque’s typological findings concerning the structurally high position of epistemic adverbs; see Cinque (1999). . This difference in derivation has been rejected in Wurmbrand (1999) and Barbiers (this volume); see, however, the final section of the present paper. . Cf. OE mugan, maeg ‘to be able’. . See OE ðearf, ðurfan (preterite present), later ðorfte, which has no reflex in Modern English, as opposed to all other Germanic languages. . Dutch equivalents of (29)–(30) were provided by one reviewer, who raised the objection that I am trying to rebut.





Werner Abraham . In Abraham (2001a) it is argued that, at least in German, the epistemic readings still faintly echo the corresponding root meanings: none of the six modals in (i) project identical epistemic readings, (i)

Sie will = muß = mag = soll Ärztin sein,

even though all of these would often be translated by one and the same English sentence, i.e. ‘She is supposed to be a physician’. It is as yet unclear how this empirical fact can be accommodated given the structural assumptions made in the present article. This position has been contested by one reviewer, who provides examples like Dutch (ii) and (iii). (ii) Het wil hier nog wel eens regenen. it will here yet WELL once rain ‘Rain is not unusual here’ (iii) De korenwolf mag dan uitgestorven zijn, Limburg heeft nog veel bijzondere dieren. the grain-wolf may then extinct be Limburg has yet many special animals ‘The grainwolf may be extinct now, but there are still many other species of animals left in Limburg’. Note, however, that my argument about the non-substitutability of MVs in these cases is reflected in examples like (iv) Het wil/*moet hier nog wel eens regenen. It will/must here yet WELL once rain If Dutch modals were indeed semantically empty, moet in (iv) should be possible. But it is not. Furthermore, the assumption that EMVs have scope over the whole sentence must not be taken to be an explanation, but as a formalised instantiation of the meaning of the modal in a clausal context. It should be clear that I do not oppose the view that epistemic modals occupy a high structural position as sentential operators; on the contrary, I fully endorse it. See (24) above and, in much greater detail, Abraham (1995, Ch. 6; 2001a, b).

References Abraham, W. (1989). Verbal substantives in German. In C. Bhatt (Ed.), Nominalizations [Linguistik Aktuell/Linguistics Today 6] (pp. 79–98). Amsterdam: John Benjamins. Abraham, W. (1990). Modalverben in der Germania. In M. Reis (Ed.), Akten des VIII. Internationalen Germanisten-Kongresses. Sektion Kontrastive Syntax (pp. 3–12). München: Iudicium. Abraham, W. (1992). The emergence of the periphrastic passive in Gothic. Leuvense Bijdragen, 81(1-3), 1–15. Abraham, W. (1995). Syntax des Deutschen im Sprachenvergleich: Grundlegung einer typologischen Syntax des Deutschen. Tübingen: Gunter Narr.

Modal verbs in German and English

Abraham, W. (1998). The aspectual source of the epistemic-root distinction of modal verbs in German. In W. Boeder, Chr. Schröder, K.H. Wagner and W. Wildgen (Eds.), Sprache in Raum und Zeit. In memoriam Johannes Bechert (pp. 231–250). Tübingen: G. Narr. Abraham, W. (2000). The morphological and semantic classification of ‘evidentials’ and modal verbs in German: The perfect(ive) catalyst. ZAS Papers in Linguistics, 15, 36–59. Abraham, W. (2001a). How far does semantic bleaching go? In J. Faarlund (Ed.), Grammatical Relations in Change (pp. 15–63). Amsterdam: John Benjamins. Abraham, W. (2001b). Modals: Toward explaining the epistemic ‘finiteness gap’. In R. Müller and M. Reis (Eds.), Modalität und Modalverben im Deutschen (pp. 7–36). Hamburg: H. Buske. Abraham, W. and C.J. Conradie (2001). Präteritumschwund und Diskursgrammatik. Amsterdam: John Benjamins. Abraham, W. and A. Fischer (1998). Das grammatische Optimalisierungsszenario von tun als Hilfsverb. In K. Donhauser and L. Eichinger (Eds.), Deutsche Grammatik – Thema und Variationen. Festschrift für Hans-Werner Eroms zum 60. Geburtstag (pp. 35–48). Heidelberg: Universitätsverlag C. Winter. Barbiers, S. (1995). The Syntax of Interpretation. Doctoral dissertation, University of Leiden. Belletti, A. and L. Rizzi (1988). Psych-verbs and theta theory. Natural Language and Linguistic Theory, 6, 291–352. Brinton, L. and M. Akimoto (1999). Collocational and Idiomatic Aspects of Composite Predicates in the History of English. [Studies in Language Companion Series 47]. Amsterdam: John Benjamins. Bybee, J. (1985). Morphology: A Study of the Relation between Form and Meaning. Amsterdam: John Benjamins. Cinque, G. (1999). Adverbs and Functional Heads: A Cross-linguistic Perspective. New York: Oxford University Press. Coates, J. (1995). The expression of root and epistemic possibility in Modern English. In J. Bybee and S. Fleischman (Eds.), Modality in Grammar and Discourse (pp. 55–66). Amsterdam: John Benjamins. Diewald, G. (1999). Die Modalverben im Deutschen. Grammatikalisierung und Polyfunktionalität. [Linguistische Arbeiten 208]. Tübingen: M. Niemeyer. Evers, A. (1975). The Transformational Cycle in Dutch and German. Doctoral dissertation, University of Utrecht. Geilfuß, J. (1992). Ist wollen ein Kontrollverb oder nicht? Arbeitspapiere des SFB 340 Stuttgart-Tübingen, Bericht Nr. 27, 29–51. Gelderen, E. van (1999). Aspect, modals, and infinitival endings in Germanic. Ms., Arizona State University. Giorgi, A. and F. Pianesi (1997). Tense and Aspect. Oxford: Oxford University Press. Görlach, M. (1991). Introduction to Early Modern Modern English. Cambridge: Cambridge University Press. IJbema, A. (1997). Zu+Infinitiv im Deutschen. GAGL, 40, 1–45. IJbema, A. and W. Abraham (2000). Die syntaktische Funktion des infinitivischen zu. In R. Thieroff, M. Tamrat, N. Fuhrhop and O. Teuber (Eds.), Deutsche Grammatik in Theorie und Praxis (pp. 123–137). Tübingen: M. Niemeyer.





Werner Abraham

Kemenade, A. van (1987). Syntactic Case and Morphological Case in the History of English. Dordrecht: Foris. Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press. Pesetsky, D. (1995). Zero Syntax: Experiencers and Cascades. Cambridge, Mass.: MIT Press. Picallo, C. (1985). Opaque Domains. Doctoral dissertation, CUNY. Reis, M. (2000). Bilden Modalverben im Deutschen eine syntaktische Klasse? (Nicht nur) Neues zu einem alten Thema. Paper read at the Modality workshop at the University of Tübingen, 16–17 March 2000. Roberts, I. (2000). The history of the modals yet again. Paper read at the Modality workshop at the University of Tübingen, 16–17 March, 2000. Rohrbacher, B.W. (1999). Morphology-driven Syntax: A Theory of V to I Raising and ProDrop. Amsterdam: John Benjamins. Sprouse, R. and J. Durbin (1998). The syntactic category of the preterit-present modal verbs in Modern Standard German. Paper read at the Berkeley Germanic Linguistics Round Table, 4 April 1998. Tschirner, E. (1991). Aktionalitätsklassen im Neuhochdeutschen. New York: Peter Lang. Vikner, S. (1988). Modals in Danish and event expression. Working Papers in Scandinavian Syntax, 39, 31–53. Vikner, S. (1995). Verb Movement and Expletive Subjects in the Germanic Languages. London/New York: Oxford University Press. Warner, A. (1993). Modern English Auxiliaries. Cambridge: Cambridge University Press. Wurmbrand, S. (1998). Infinitives. Doctoral dissertation, University of Massachusetts, Amherst. Wurmbrand, S. (1999). Modal verbs must be raising verbs. Paper read at WCCFL 1999. Zhang, L. (1995). A Contrastive Study of Aspectuality in German, English, and Chinese. New York: Peter Lang. Zubizarreta, M.-L. (1982). On the Relationship of the Lexicon to Syntax. Doctoral dissertation, MIT.

Modality and polarity* Sjef Barbiers Royal Netherlands Academy of Sciences, Meertens Instituut, Amsterdam

In this paper it is argued that the possible interpretations of a modal verb primarily depend on properties of its complement. On the basis of non-verbal complementation of modals in Dutch it is shown that modals always require a complement that denotes a bounded scale. The nature of this scale determines whether the modal gets a root or an epistemic interpretation.

.

Introduction

This paper investigates the relation between the interpretations of modal verbs and the nature of their complements, starting out with a cross-linguistically rare construction in which modals take non-verbal complements:1 (1) a.

Het licht moet aan. the light must on ‘The light must be switched on.’ b. De fles moet leeg. the bottle must empty ‘The bottle must be emptied.’

The ensuing analysis of this construction leads to the following claims: i.

The non-verbal complements in (1) cannot be analysed as involving a silent infinitive (Section 2; contra Geerts et al. 1984; Vanden Wyngaerd 1994). ii. The complement of a modal must denote a value on a bounded scale, regardless of the interpretation of the modal. If the complement does not denote such a value, the sentence is ungrammatical (Section 2). iii. The data in (1) are problematical for analyses according to which root interpretations always involve control structures while epistemic interpreta-



Sjef Barbiers

tions always involve subject raising structures (e.g. Hofmann 1966; Ross 1969; Perlmutter 1970).2 While the modals in (1) must be analysed as raising verbs, they can only have root interpretations, not epistemic interpretations. iv. In root interpretations, a modal can be subject-oriented (dyadic) or nonsubject-oriented (monadic). A modal is always monadic in epistemic interpretations (Section 3). v. Modals with a root interpretation always involve a potential change of the value on the scale introduced by the modal’s complement, i.e. a potential polarity transition. When such a change is impossible, e.g. when the complement of a modal denotes a invariable property of the subject, an epistemic interpretation is forced (Section 4). vi. The semantic ambiguity of modals is determined by two parameters: (1) the subject-orientation of the modal, and (2) the availability of a potential change of value on a bounded scale. This yields four classes of modal interpretations (Section 5). vii. The syntactic complement of the modal determines its interpretations. There are two types of complements that cannot be interpreted as involving a potential change of value. In the case of a definite nominal complement, this gives rise to sympathy/antipathy interpretations. In the case of a verbal complement, this yields an epistemic interpretation. These interpretations are possible only if negation or (abstract) affirmation is present to provide the required value on a bounded scale (Section 6). viii. Epistemic interpretations are non-subject-oriented. This follows from the assumption that the semantic relation between a verb and the subject is established by an (abstract) verbal head with D-features. In the epistemic interpretation, the projection of this verbal head including the subject is the single argument of the modal, making it impossible to establish a modal relation between subject and verbal complement (Section 7). ix. The negative polarity of moeten ‘must’ in (2a) and the bipolarity of mogen ‘may’ in (2b) follow directly from the general lexical argument structure of modals. There is no need to assume that this polarity is an accidental lexical idiosyncrasy or that polar and non-polar instances of moeten and mogen involve different lexical entries (Section 8). (2) a.

Marie moet die jongen *(niet). Marie must that boy not ‘Marie does not like that boy.’

Modality and polarity

b. Marie mag die jongen (wel)/niet. Marie may that boy /not ‘Marie likes/does not like that boy.’

. Modals with non-verbal complements Traditionally, Dutch modals with a non-verbal complement have been analysed as involving an infinitive that is phonologically silent or deleted at PF (Geerts et al. 1984; Vanden Wyngaerd 1994). An advantage of such an analysis is that it allows the crosslinguistic generalisation that modals always take a verbal complement, as in English. (3) a.

Jan moet weg (gaan). Jan must away go ‘Jan must go away.’ b. Jan wil een pizza (hebben). Jan want a pizza (have) ‘Jan wants to have a pizza.’

However, there are a number of strong counterarguments. The first argument against PF-deletion of the infinitive in the case of modals with a non-verbal complement is that certain cases would require deletion of a passive complement. This incorrectly predicts that such cases should allow a BY-phrase or an agent-oriented adverb. The crucial observation on which this argument rests is that it is not always sufficient to assume that GO, HAVE, BE or some other basic verb is deleted: other, more complex verbal expressions must be assumed to be deletable as well. Deze lampen moeten uit (= gaan/= zijn).3 these lights must out (go/be) ‘These lights must be switched off.’ b. Die boeken mogen weg (= gaan/= zijn) these books may away (go/be) ‘These books can be thrown away.’

(4) a.

The sentences in (4) with audible GO/BE yield an interpretation that is completely different from the interpretation with GO/BE absent. With GO/BE absent, the interpretation must be as given in the translation, whereas audible GO forces an interpretation that the lights go out by themselves and the books go away by themselves, while audible BE forces an interpretation that the lights





Sjef Barbiers

must be out and the books must be away at utterance time or some point in the future. As the translations in (4) indicate, the interpretation of (4a, b) without an infinitive is as with an embedded passive. We could assume that (4a, b) is the result of PF-deletion of the embedded passive: (5) a.

Deze lampen moeten uit worden gedaan. these lights must out (be done) ‘These lights must be switched off.’ b. Die boeken mogen weg worden gedaan. these books may away (be done) ‘These books can be thrown away.’

If (5a, b) were the correct analysis, it should be possible to add a BY-phrase or an agent-oriented adverb, since passives allow such an addition. However, BY-phrases and agent-oriented adverbs are impossible when there is no overt verbal complement (6a, b), even though deletion of infinitive and participle normally does not force deletion of BY-phrase or agent-oriented adverb (6c, d). (6) a.

Deze lampen moeten door jou uit *(worden gedaan). these lights must by you out (be done) ‘These lights must be switched off.’ b. Die lampen moeten zorgvuldig uit *(worden gedaan). these lights must carefully off (be done) ‘These lights must be switched off carefully.’ c. De lamp moet door P. aan worden gedaan en door J. uit worden gedaan. the light must by P. on be done and by J. off be done ‘The light must be switched on by Peter and switched off by Jan.’ d. De lamp moet zorgvuldig aan worden gedaan en uit worden gedaan. the light must carefully on be done and off be done ‘The light must be switched on and switched off carefully.’

The second argument against a PF-deletion analysis involves the reduced ambiguity of modals with a non-verbal complement. Since the alleged deletion of verbal material would take place at PF, this should not have any consequences for the interpretation possibilities. However, whereas a modal with a verbal complement can get an epistemic interpretation, a modal with a nonverbal complement can never be interpreted epistemically. This is illustrated in (7).4 (7) a.

Jan mag weggaan (, ooit zal hij terugkeren). Jan may away-go (someday will he return)

Modality and polarity

i. ‘Jan has permission to leave; (someday he will return).’ deontic ii. ‘It is allowed that Jan leaves; (someday he will return).’ ind.deontic iii. ‘It may be true that Jan is leaving, someday he will return’ epistemic b. Jan mag weg (, ooit zal hij terugkeren) Jan may away someday will he return i. ‘Jan has permission to leave; (someday he will return).’ deontic ii. ‘It is allowed that Jan leaves; (someday he will return).’ ind.deontic iii. *‘It may be true that Jan is leaving, someday he will return. epistemic

A third argument against the PF-deletion analysis involves cases in which there is no suitable candidate for deletion: (8) a.

Jan kan niet tegen katten. Jan cannot against cats ‘Jan is allergic to cats.’ b. Jan kan twee keer in die jas. Jan can twice in that coat ‘That coat is a size too big for John.’ c. Jan kan zijn werk niet aan. Jan can his work not on ‘Jan cannot cope with his work.’ d. Jan moet Marie niet. Jan must Marie not ‘Jan does not like Marie.’ e. Jan mag Marie wel. Jan may Marie  ‘Jan likes Marie.’

The final and perhaps strongest argument against the PF-deletion analysis is the following observation. Only constituents that denote a value on a bounded scale can be the non-verbal complement of a modal. The facts in (9) can be stated directly in terms of selectional restrictions imposed by the modal if we assume that there is no silent infinitive. If there were a silent infinitive, we would be forced to say that the modal imposes selectional restrictions on the complement of its complement, a rather unusual state of affairs.5 (Note by the way that the stage-level–individual-level distinction is irrelevant here, as the





Sjef Barbiers

examples (9d, e) show. It is also not sufficient for the complement to denote a scale that is unbounded, as (9f) shows.6,7 ) (9) a. b. c. d. e. f.

De fles moet leeg. the bottle must empty Het raam kan open.8 the window can open Het licht mag uit. the light may off *Het konijn kan ziek. the rabbit can sick *Jan moet intelligent. Jan must intelligent *De speler moet lang. the player must long

Scale: empty-full Scale: closed-open Scale: off-on

I will assume on the basis of these four arguments that Dutch modals can have a non-verbal complement, as long as this complement denotes a value on a bounded scale.

. Against a raising–control analysis of the epistemic– root distinction On the basis of the conclusion that modals in Dutch can have non-verbal complements, it is possible to show that the semantic epistemic–root distinction cannot always be expressed in terms of the syntactic distinction between raising and control verbs.9 An example of this traditional syntactic analysis of modal ambiguity (cf. Hofmann 1966; Ross 1969; Perlmutter 1970) is given in (10). (10) John must be kind. i. ‘It must be true that John is kind.’ Johni must be [ti kind] ii. ‘John has the obligation to be kind.’ Johni must be [PROi kind]

epistemic raising root control

In (10i), the modal is a monadic predicate, taking the proposition as its sole argument: must (John be kind). The subject does not receive a Theta-role from the modal. In (10ii), the modal is a dyadic predicate, a relation between the subject and a predicate: must (John, be kind). The subject John receives a Thetarole from the modal, expressing that the obligation is attributed to the subject in ii. but not in i.

Modality and polarity

Feldman (1986) and Brennan (1993) have already observed that sentences such as the ones in (11) clearly can have deontic interpretations but cannot be analysed as involving control: (11) a.

De brief moet van Jan voor vijf uur in Amsterdam zijn. the letter must of Jan before five hour in Amsterdam be ‘Jan requires that the letter be in Amsterdam before five o’clock.’ b. De nieuwe machine mag geen problemen geven. the new machine may no problems cause ‘It is unacceptable if the new machine causes problems.’

The sentences in (11) cannot be cases of control since the obligation and permission are not attributed to a subject; rather, the obligation/permission applies to an entire state of affairs here: the state of affairs ‘the letter be in Amsterdam before five o’clock’ is obligatory, and the state of affairs ‘the new machine causing problems’ is not permitted. The modals in (11) must therefore be analysed as monadic predicates. Modals with non-verbal complements provide an even stronger argument against the raising–control analysis of the epistemic–root distinction. As is well-known, small clause PP-complements in Dutch cannot be extraposed (12a; Hoekstra 1984), whereas all other types of PPs, such as PP-arguments (12b) and PP-adjuncts (12c), can: (12) a.

dat Jan gaat that Jan to home goes to home ‘that Jan is going home’ b. dat Jan luistert that Jan to Marie listens to Marie ‘that Jan is listening to Marie’ c. dat Jan zit te lezen that Jan in the garden sits to read in the garden ‘that Jan is reading in the garden’

PP-complements of modals behave like small clause complements in this respect: (13) a.

dat Jan morgen moet that Jan tomorrow to Amsterdam must to Amsterdam b. dat de brief morgen mag that the letter tomorrow into the trashcan may into the trashcan





Sjef Barbiers

If the PPs [naar Amsterdam] ‘to Amsterdam’ and [in de prullenbak] ‘into the trashcan’ are small clause complements, then the DPs [Jan] ‘Jan’ and [de brief ] ‘the letter’ must be their subjects. Their subjects cannot be PRO, since small clause complements cannot have PRO as their subject (14a). This in turn implies that the subjects in (13) must have raised from a position inside the small clause, as is illustrated in (14b). (14) a.

Jan drinkt [SC zich/* ziek] Jan drinks  sick b. Jani moet [PP ti naar Amsterdam] Jan must to Amsterdam

We have now established that (13a, b) must be analysed as subject raising constructions. If subject raising corresponded to an epistemic interpretation, the prediction would be that sentences like (13a, b) can only have an epistemic interpretation. But this prediction is wrong: sentences like (13a, b) cannot have an epistemic interpretation; they can only have root interpretations: (15) Jan moet naar Amsterdam. i. ‘Jan has the obligation to go to Amsterdam.’ ii. ‘It is required that Jan goes to Amsterdam.’ iii. ‘*It must be true that Jan is going to Amsterdam.’

direct deontic indirect deontic epistemic

We can conclude that the raising–control analysis of the epistemic–root distinction cannot be entirely correct.

. Potential polarity transition as a crucial property of root interpretations As discussed in the previous section, modals with a root interpretation can be subject-oriented and non-subject oriented (cf. Feldman 1986; Brennan 1993). Subject-orientation therefore cannot be used as a defining property of root interpretations. Non-subject-orientation, however, still counts as one of the defining properties of the epistemic interpretation. In this section it is shown that there is another factor playing a role in the distinction between the two types of interpretations: the availability of a potential polarity transition. One of the arguments mentioned in Section 3 in favour of the existence of constructions with a modal and a non-verbal complement is that this allows us to express selectional restrictions on the complement of the modal straightforwardly: the non-verbal complement of a modal must denote a value on a

Modality and polarity

bounded scale. If it does not, the sentence is ungrammatical. It is clear what role this value on a bounded scale is playing in the semantic interpretation of a sentence with a modal. For example, a sentence like (16) may be paraphrased as: ‘the value that the bottle has on the scale of empty (0) to full (1) is not 1 when (16) is uttered and should become 1 at some point in the future.’ Notice that the sentence does not say anything about the question whether this value will ever be reached. We may say that the interpretation of (16) involves a potential, not an actual polarity transition from [not 1] to 1. (16) Deze fles moet vol. this bottle must full

My claim is that the above paraphrase of (16) reveals two essential properties of root interpretations: (i) the complement of the modal must denote a value on a bounded scale, and (ii) this value is not the actual value at the moment T denoted by the modal. The latter requirement makes sense intuitively: (16) cannot be uttered in a situation in which the bottle is already full. It can be shown that a potential polarity transition is also crucial for the availability of root interpretations when modals have a verbal complement. When the complement of the modal denotes a fixed property of the subject, an epistemic interpretation is forced:10 (17) a.

De aarde moet een planeet zijn. the earth must a planet be i. ‘It must be true that the earth is a planet.’ epistemic 11 # ii. ‘It is required that the earth is a planet.’ root b. Jan moet een moedertaalspreker van het Fins zijn Jan must a native speaker of the Finnish be i. ‘It must be true that Jan is a native speaker of Finnish.’ epistemic ii. #‘Jan is required to be a native speaker of Finnish.’ root 12 c. Jan moet zijn kamer gisteren voor vijf uur hebben opgeruimd. Jan must his room yesterday before five hour have cleaned i. ‘It must be true that Jan has cleaned his room before five.’ o’clock yesterday’ epistemic # root ii. ‘Jan is required to have cleaned his room yesterday’

Non-verbal complements and infinitival complements of modals thus have in common that they only allow root interpretations if they denote a variable property. For non-verbal complements, this is not yet sufficient, as we have seen: the potential polarity transition applies to a value on a bounded scale provided by the complement. Many stage-level predicates cannot be the non-





Sjef Barbiers

verbal complement of a modal, since they do not denote a value on a bounded scale (cf. 9d, f). Now that it has been established that a potential polarity transition is also crucial for the availability of root interpretations with infinitival complements, the question must be asked whether infinitival complements involve a value on a bounded scale. The null hypothesis clearly is that they do. If we were to hypothesise that they do not, we would in fact claim that the requirement of a bounded scale in the case of non-verbal complements is completely accidental and that non-verbal and verbal complements are fundamentally different, despite the fact that they both must denote a variable property. The next question is which constituent provides the bounded scale in the case of infinitival complements. I would like to suggest that it is the embedded verb itself that does that. The behaviour of the adverbial modifier half ‘halfway’ supports this view. It typically modifies the end point of a bounded scale (18a) and therefore does not occur with constituents that do not denote an endpoint (18b). As expected, it can modify the infinitival complement of a modal (18c, d). (18) a.

De fles is half leeg. the bottle is halfway empty b. *Jan is half oud. Jan is halfway old c. Jan moest half huilen. Jan must- halfway cry ‘Jan was torn between laughing and crying.’ d. Ed mag maar half beseffen wat de bedoeling is. Ed may only half realise what the intention is ‘Ed should realise only partly what the intention is.’

The sentence in (18c) means that there was a half, not a full crying event. Aspect is not involved here, as (18c) is not about the completion of the crying event. What (18c) means is that Jan was halfway between not crying and crying. Similarly, (18d) means that Ed should be halfway between not understanding and understanding what the intention is. The scale involved is simply a numerical scale from 0 to 1, from no event to one event. As I argue in Section 6, this numerical scale should not be confused with the scale involved in epistemic interpretations. A striking difference between verbal and non-verbal complements of modals is that modals do not impose any further selectional restrictions on the embedded verb: any verb can be the complement of a modal. This can be

Modality and polarity

explained if we assume that a verb is always dominated by a functional projection, the head of which denotes the value 1 (cf. Section 6 and Barbiers 1995). The presence of the verb then guarantees the presence of this functional projection, and the functional projection satisfies the selectional restriction imposed by modals, making the choice of verb free.13 As the notion of potential polarity transition may easily give rise to misunderstandings, four qualifications of the observations in (17) are called for (cf. Barbiers 1995 and the Introduction to this volume): (19) i.

The aspectual class of the embedded verb is irrelevant for modal interpretation. More specifically, stative complements allow root interpretations. The crucial property of root interpretations is whether the situation denoted by the complement can switch from non-existent to existent; whether this situation is dynamic or stative internally is irrelevant. ii. Individual-level complements only block the root interpretations when the subject is a name, not when the reference of the subject may vary with different situations or worlds (20b–c). This is because the property denoted by the predicate within the complement can only be a permanent, invariable property relative to a particular referent. iii. Perfective complements force an epistemic interpretation only when they refer to the completion of an event at some point in the past, not when they refer to the completion of an event at some point in the future (20d). iv. It is possible to find contexts in which fixed properties such as the one in (17a) can be interpreted as variable properties. This actually confirms our point. In a possible world in which the earth is yet to be created, the property of being a planet or a star has not yet been fixed and is still variable. Therefore root interpretations are possible (20e).

(20) a.

Jan moet een auto hebben. Jan must a car have i. ‘It must be true that Jan has a car.’ ii. ‘Jan is required to have a car.’ b. The new professor must be a native speaker of Finnish. i. ‘It must be true that the new professor is a native speaker of Finnish.’ ii. ‘It is required that the new professor be a native speaker of Finnish’ c. Jan must be a native speaker of Finnish. i. ‘It must be true that Jan is a native speaker of F.’

stative epistemic root

epistemic root ind.level epistemic





Sjef Barbiers

ii. #‘It is required that Jan be a native speaker of Finnish’ root d. Jan moet zijn kamer voor morgen vijf uur hebben opgeruimd. perfective Jan must his room before tomorrow five o’clock have cleaned i. ‘It must be true that Jan will have cleaned his room tomorrow before five o’clock.’ epistemic ii. ‘Jan is required to have cleaned his room tomorrow.’ root before five o’clock’ e. God said: The earth must be a planet i. ‘God thought that the earth was a planet.’ epistemic ii. ‘God required that the earth be a planet (and not a star).’ root

. A new classification of modal interpretations The two parameters for modal interpretations isolated in the previous section, [± subject-oriented] and [± potential polarity transition], are independent, yielding the four different types of modal interpretation in Table 1. An example of each interpretation is given in (21). The most interesting result is that the two parameters correctly predict that modal interpretations of type d. should exist. This type of interpretation is discussed in Section 8. Table 1. A new classification of interpretations of modal verbs in Dutch

[+ potential polarity transition] [– potential polarity transition]

[+ subject-oriented]

[– subject-oriented]

a. Dispositional b. Direct deontic d. Sympathy/antipathy

c. Indirect deontic

Jan mag graag de baas spelen. Jan may eagerly the boss play ‘Jan likes to be the boss.’ b. Jij mag van mij de kamer uit. you may from me the room out ‘I allow you to leave the room.’ c. De kamer moet schoon zijn. the room must clean be ‘It is required that the room be clean.’

e. Epistemic

(21) a.

dispositional

direct deontic

indirect deontic

Modality and polarity

d. Jan moet Marie niet. Jan must Marie not ‘Jan does not like Marie.’ e. Jan moet z’n kamer gisteren hebben opgeruimd. Jan must his room yesterday have cleaned ‘Jan must have cleaned his room yesterday.’

antipathy

epistemic

As will be clear from Table 1, the two parameters cannot distinguish between dispositional and direct deontic interpretations. Perhaps this is simply a lexical ambiguity that cannot be reduced to the syntactic context of the modal. I leave this issue for future research.

. How the complement determines modal interpretation Table 1 captures the conclusion drawn in Section 3 that the epistemic–root distinction does not depend on the subject-orientation of the modal: a modal can be a monadic predicate and at the same time have a root interpretation. Under the assumption that the ambiguity of modals is not lexical, the complement of the modal is the only possible source of ambiguity. There are a number of differences between complements of epistemic modals and complements of root modals: Table 2. Differences between epistemic and root interpretations in Dutch Epistemic

Root

Potential polarity transition

not required

required

Scale of complement

negative – affirmative (or: no – yes)

0–1 (number of events)

Category of complement

verbal only

all categories

Definite complement

yes

no

The first three differences have been illustrated above. The fourth difference, the definiteness of the complement, is the key to understanding the influence of the complement on the interpretation of the modal. ‘Definite complement’ in this table means that the complement of an epistemic modal is a definite verbal argument, not a nominal argument, since it arguably is not.14 The complement of an epistemic modal behaves like a definite constituent on a number





Sjef Barbiers

of tests, whereas the infinitival complement of a modal in the root interpretations behaves like an indefinite constituent on the same tests.15 For example, indefinites but not definites can occur in the exclamative constructions in (22) and (23). These exclamative constructions disambiguate the modals. Modals can only have root interpretations here, so the infinitival complement of an epistemic modal patterns with definite constituents. (22) a.

Een auto dat Jan heeft! a car that Jan has ‘Jan has a very nice car!’ (exclamative) b. *Die auto dat Jan heeft! that car that Jan has c. Werken dat Jan kan! work that Jan can i. ‘Jan is able to work very hard.’ ii. #‘It is possible that Jan is working very hard.’

Wat heeft Jan een auto! what has Jan a car ‘Jan has a very nice car.’ b. *Wat heeft Jan die auto! what has Jan that car c. Wat kan Jan eten! what can Jan eat i. ‘Jan can eat very much!’ (exclamative) ii. *‘It is possible that Jan eats very much!’

root epistemic

(23) a.

root epistemic

I propose to characterise this difference between infinitival complements of epistemic and root modals syntactically as in (24). Root modals take an indefinite verbal complement IndP, whereas epistemic modals take a definite verbal complement Dv P. This Dv P complement must be specified by Negation/Affirmation, as will be argued below.16 (24) a. root: [DvP Dv [ModalP Modal [IndP [Ind ONE [VP V]]] b. epistemic: [ModalP Modal [DvP NEG/AFF [DvP Dv [IndP [Ind ONE [VP V]]]

The IndP (Individuator Phrase) with (abstract) head ONE is the verbal counterpart of an indefinite noun phrase (or NumP), and provides the value 1 on the numerical 0–1 scale. In root interpretations with an infinitival complement, modals take an IndP, thus satisfying the requirement that the complement of a modal denote a value on a bounded scale, leaving the choice of the verb free.

Modality and polarity

I take an (abstract) verbal Dv to establish the semantic relation between subject and event, identifying the subject as the source, starting point, agent, possessor of the event (cf. Barbiers 1995). In traditional terms, Dv is responsible for assigning an external Theta-role to the subject in its Spec when the verb is transitive or unergative. However, I take it to be present with unaccusative verbs as well. Dv is the equivalent of abstract or little v (Larson 1988; Hale and Keyser 1993; Chomsky 1995; see Collins 1997 for the D-like properties of little v). Due to the presence of Dv , the complement of the modal is definite in the case of the epistemic interpretation. If it is correct that Dv has to be present in the complement of an epistemic modal it follows that only verbal complements allow an epistemic interpretation. Verbs will always be dominated by verbal Dv . When Dv dominates a predicate of category A, P, N, the resulting constituent will be verbal too (cf. Hale and Keyser 1993). Only when Dv is absent can the resulting complement be non-verbal, and then epistemic interpretations are impossible. Maximally generalising the lexical argument structure of modals, I assume that also in the case of the epistemic interpretation the complement should denote a value on a bounded scale. The presence of Negation/Affirmation in the complement of an epistemic modal is necessary to provide such a value. Since a definite constituent denotes an entity or an event/state, not a scalar value, it cannot be the complement of a modal when negation/affirmation are absent. Epistemic and root interpretations involve different scales, then, because in epistemic interpretations Negation/Affirmation provides the scale and in root interpretations IndP. The claim that definite constituents without Negation/Affirmation cannot be the complement of a modal is supported by the behaviour of nominal complements of modals. Whereas indefinites and other constituents that provide a value on a bounded scale can be the complement of a modal, NPs with a definite determiner cannot.17 (25) a.

Jan kan alles/niets/iets/ ding. Jan can all/nothing/something/one thing ‘Jan is able to do everything/nothing/something/one thing.’ b. *Jan kan het werk. Jan can the work c. Jan moet een/*het vriendinnetje. Jan must a/the girl friend ‘Jan wants to find a girl friend.’





Sjef Barbiers

As in the case of epistemic interpretations, the modal can take an NP with a definite determiner when negation or affirmation provides the required value: (26) a.

Jan kan het werk *(niet/wel). Jan can the work not/ ‘Jan is (not) able to do the work.’ b. Jan moet het vriendinnetje *(niet). Jan must the girl friend not ‘Jan does not like the girl friend.’

In this view, then, epistemic modals and the modals in the construction in (26) have in common that they take a definite complement that cannot provide a value on a bounded scale. This forces the presence of negation/affirmation. The absence of a polarity transition both in the epistemic interpretation and in the definite nominal constructions in (26) (cf. Tables 1 and 2) can now be related to the nature of the complement. When the complement is definite, a polarity transition is impossible, whereas when the complement is indefinite, adjectival, prepositional or verbal, a polarity transition is possible. It is plausible that the impossibility of a transition is related to the fact that definite constituents have a fixed reference, while the possibility of a transition derives from (among other things) the variable reference of indefinite constituents.

. The syntactic representation of subject-orientation An epistemic modal is never subject-oriented (it is always monadic), whereas a root modal may or may not be subject-oriented. This follows straightforwardly from the structures proposed in (24). If it is correct that Dv establishes the semantic relation between a subject and a verb, identifying the subject as the source, starting point, possessor, agent of what is denoted by the verb, it depends entirely on the base position of the modal whether a semantic relation will be established between the subject and modal or not.18 (27) a.

root: [DvP Subject [Dv Ø [ModP [Mod Modal [IndP [Ind Ø [VP Verb]]]]]]] b. epistemic: [ModP [Mod Modal [DvP Subject [Dv Ø [IndP [Ind Ø [VP Verb]]]]]]]

According to standard compositional semantics (cf. Heim and Kratzer 1998), in (27a) Dv is a binary relation between subject and ModP, so the subject will be interpreted as the source of the modality (dispositional interpretation) or the

Modality and polarity

possessor of the modality (direct deontic interpretation). In (27b), Dv cannot be a binary relation between subject and ModP, because ModP dominates Dv P. The fact that an epistemic modal is never subject-oriented is thus a direct consequence of the assumption that the modal selects Dv P (including the subject) in such cases. Indirect root interpretations, in which the modal is a monadic predicate, may now arise in two ways: (i) Dv P is entirely absent and the modal selects NumP, PP or AP; (ii) Dv P is present, but the subject reconstructs at LF into a position within the scope of the modal, [SpecVP]. I will not discuss these possibilities any further here (cf. Barbiers 1995 for discussion). The argument structure of modals at LF is summarised in (28). The lexical relational information associated with modals, whatever its precise content, must fully match the semantic relations determined by syntactic structure. A modal can be dyadic or monadic. In both cases, one argument must provide a value on a bounded scale. Class V gives rise to polarity effects and is discussed in the next section. (28) I.

II.

Dispositional Relation: Example:

Direct deontic Relation: Example:

III. Indirect deontic Relation: Example:

IV. Epistemic Relation:

modal (entity, value on bounded scale) mag ([Jan], [IndP schaatsen]) may (Jan, skate) Jan mag graag schaatsen. Jan may eagerly skate ‘Jan likes skating.’ modal (entity, value on bounded scale) mag ([Jan], [IndP schaatsen]) may (Jan, skate) Jan mag van mij schaatsen. Jan may of me skate ‘I allow Jan to skate.’ modal (value on bounded scale) moet ([IndP de brief op tijd in Amsterdam zijn]) must (the letter be in Amsterdam on time) De brief moet op tijd in Amsterdam zijn. the letter must on time in Amsterdam be ‘The letter must be in Amsterdam on time.’ modal (value on bounded scale)





Sjef Barbiers

Example:

V.

mag ([NegP/AffP [DP Jan net zijn kamer opgeruimd hebben]]) may (Jan just his room cleaned have) Jan mag net zijn kamer opgeruimd hebben, het is Jan may just his room cleaned have, it is al weer een rommel. already again a mess ‘It may be true that Jan has cleaned his room, but it’s a mess again.’

Negative/Positive Relation: modal (entity, value on a bounded scale) Example: mag ([Marie], [Neg/Aff niet/wel]) may (Marie, not/) Jan mag Marie niet/wel Jan may Marie not/ ‘Jan likes Marie/Jan does not like Marie’

. Negative polar moeten and bipolar mogen As we have seen in the previous sections, the modals moeten ‘must’ en mogen ‘may’ are polarity items when they occur with definite nominal complements (cf. 29), while they are not polarity items in their other uses. In this section it is argued that this polarity is not an accidental lexical idiosyncracy, but follows from the general requirement that one of the arguments of the modal denote a value on a bounded scale.19 In the unmarked case, the scale is provided by the complement of the modal, and then moeten and mogen are not polarity items. A definite nominal complement cannot provide the required scale and therefore negation or affirmation must be present. Jan mag Marie *(niet/wel.)20 Jan may Marie not/ ‘Jan does not like/likes Marie.’ b. Jan moet Marie *(niet)/(*wel). Jan must Marie not ‘Jan does not like Marie.’

(29) a.

This explanation entails that Negation/Affirmation can be one of the arguments of a modal and that the modals in (29) should not be analysed as simple transitive verbs. Maintaining the hypothesis that the relation between a verb

Modality and polarity

and its ‘external’ argument is established by abstract Dv (cf. Section 6 and 7), I propose the structure in (30) for this construction.21 According to standard compositional semantics, the modal in (30) is a binary relation between Negation/Affirmation and DP2 Marie. (30) [DvP [DP1 Jan] [Dv Ø [ModP niet/wel [Mod mag/moet [DP2 Marie ]]]]]]] Jan not/ may/must Marie

The idea that Negation/Affirmation can be an argument of a verb is supported by the behavior of epistemic verbs (cf. Barbiers (2000) for a detailed analysis and more arguments). Note that the construction in (31a) cannot be analysed as a case of ellipsis since a full complement with a verbal core is impossible (31c). Put differently, van niet/wel and full CP-complements are in complementary distribution. (31) a. Jan denkt/hoopt/gelooft van niet/wel. b. Jan thinks/hopes/believes not/so. c. *Jan denkt/hoopt/gelooft van niet/wel dat Marie komt. Jan thinks/hopes/believes of not/so that Marie comes

Finally, the fact that moeten ‘must’ is negative polar while mogen ‘may’ is bipolar with definite nominal complements follows from an independent difference between the two modals: moeten ‘must’ is a universal quantifier, whereas mogen ‘may’ is an existential quantifier ([not may not] = must, just like [not one not] = all). It is a general property of universal quantifiers that they cannot be specified by wel (AFFIRMATIVE), whereas they can be specified by niet ‘not’. This may be a tautological effect: wel indicates that a given value is the highest on a scale, which is redundant in the case of universal quantifiers, which denote the highest value on a scale themselves. Existential quantifiers can be specified both by affirmative and negative morphemes, as the following contrasts show. (32) a.

Niet/*Wel allemaal gingen we naar het feest. not/ all went we to the party ‘We did not all go to the party.’ b. Niet/*Wel alles heeft Jan gedaan. not/ everything has Jan done ‘Jan did not do everything.’ c. Niet/*Wel altijd regende het not/ always rained it ‘It was not always raining.’ d. Niemand/Wel iemand heeft Jan gesproken (maar niet veel gasten).22 nobody/ someone has Jan spoken (but not many guests)





Sjef Barbiers

e.

f.

i. ‘Jan did not talk with anybody.’ ii. ‘Jan did talk with somebody, but not with many people.’ Niets/Wel iets heeft Jan gedaan (maar niet veel). nothing/ something has Jan done (but not much) i. ‘Jan did not do anything.’ ii. ‘Jan did do something, but not much.’ Nooit/Wel ooit is Jan daar geweest (maar niet vaak). never/ ever is Jan there been (but not often) i. ‘Jan has never been there.’ ii. ‘Jan has been there, but not often.’

Thus, moeten is unipolar because it is a universal quantifier and universal quantifiers cannot be specified by an affirmative morpheme. In sum, the negative polarity of moeten and the bipolarity of mogen are the result of the interaction between two properties: (i) the universal versus existential nature of moeten and mogen and (ii) the general property of modals that they require an argument providing a value on a bounded scale.

. Conclusion This investigation of the complements of modals leads to the conclusion that the ambiguity of modal verbs is primarily determined by syntactic and semantic properties of the complement of the modal.

Notes * I thank Jack Hoeksema and Virginia Brennan for comments on earlier versions of this paper. The usual disclaimers apply. . I discuss Dutch data only. The construction also exists in Norwegian, which has the same restrictions on the type of complement as Dutch. German and Afrikaans also have the construction, but with more restrictions: these languages allow only a subset of the complements that can occur with a modal in Dutch and Norwegian. . In the epistemic interpretation the modal qualifies the truth of a proposition. The root interpretations involve obligation, permission, ability and will. See Lyons (1977), Palmer (1986) and Barbiers (1995) for discussion and classification of modal interpretations. . The ‘=’ sign means that the interpretation of the clause without an infinitive is not equivalent to the interpretation of the clause with an infinitive.

Modality and polarity . For reasons that I do fully not understand, the epistemic interpretation of mogen ‘may’ is only felicitous in a concessive context, which requires the presence of a consecutive clause. . Cf. Barbiers (1995) for an elaboration of this argument and for other arguments. . Interestingly, complements that yield ungrammaticality because they do not denote a scale become grammatical in the comparative: (i)

a. *Deze wijn moet oud this wine must old b. Deze wijn moet (nog een jaar) ouder this wine must one more year older

In view of our generalisation, this must mean that the comparative denotes a value on a bounded scale. It is clear that a comparative introduces a scale with a lower bound, namely the degree of comparision. The upper bound must then be taken to be implicitly given, as in (i. b). The difference with the ungrammatical non-comparative cases is that here even implicit specification is impossible. . As is the case with selectional restrictions in other contexts, type-coercion arises when the complement does not satisfy the restriction that it denote a bounded scale. The effect of this is that the hearer will try to interpret as bounded any complement that denotes an unbounded scale in the unmarked case. Provided the proper context, this is possible for many unbounded predicates. Therefore, it is impossible to give a complete list of predicates that can be the complement of a modal. . open is an adjective here; a verb would have the infinitival form: openen. . For an overview of arguments against the raising – control analysis of the epistemic – root ambiguity, see the Introduction of this volume. . See below for a refutation of potential counterexamples. . The ‘#’ indicates that an interpretation is not available. . When the perfective complement denotes the completion of an event at some point in the past, this becomes a fixed property: once it has been established that John has cleaned his room at five o’clock, January 10, 1999, this remains true forever and cannot change anymore. . Contrary to what I claim in Barbiers (1995), it cannot be the infinitival form (the morpheme -en) that denotes the value 1. Finite verbs introduce the value 1 as well, if the diagnostic of modification by half is reliable: (i)

Jan lachte half John laughed half

(ii) Jan besefte maar half dat hij gekozen was John realised only half that he chosen was . For example, infinitival complements of modals may follow the finite verb in embedded clauses, unlike nominal complements. . The non-nominal nature of the complement may be determined by the verbal head or by functional elements.





Sjef Barbiers . Cf. Barbiers (1995) for an elaborated version of this idea within a compositional syntactic theory, and for more arguments. . Cases like Jan kan dat ‘John can that’ do not count as counterexamples because dat is not a definite determiner here but a propredicate, as in Jan is dokter en Piet is dat ook, lit. ‘John is doctor and Pete is that too’. . For reasons of space we have omitted Negation/Affirmation in this structure. . Obviously, it has to be specified in the lexicon that Dutch modals allow nominal complements. This may be considered an idiosyncracy given that in many languages modals cannot have a nominal complement. The claim in the main text, however, is restricted to the polarity of modals with a definite nominal complement. This is not a lexical idiosyncracy but follows from general properies of modals. . Since affirmative is the unmarked value of clauses, it may be implicit. . The structure abstracts away from linear word order. . The continuations in this and the following examples are necessary to make the variants with an affirmative morpheme acceptable. Such a continuation does not help in the negative polar cases.

References Barbiers, S. (1995). The Syntax of Interpretation. Doctoral dissertation, Leiden University. Barbiers, S. (2000). The right-periphery in SOV-languages: English and Dutch. In P. Svenonius (Ed.), The derivation of VO and OV (pp. 181–218). Linguistics Today 31. Amsterdam: John Benjamins. Brennan, V. (1993). Root and epistemic modal auxiliary verbs. Unpublished PhD diss. UMass Amherst: GLSA. Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris. Chomsky, N. (1995). The Minimalist Program. Cambridge, Mass.: MIT Press. Collins, C. (1997). Local Economy. Cambridge, Mass.: MIT Press. Feldman, F. (1986). Doing the best we can. Philosophical Studies 35. Dordrecht: Reidel. Geerts, G. et al. (1984). Algemene Nederlandse Spraakkunst. Groningen: Wolters-Noordhoff. Hale, K. and S.J. Keyser (1993). On argument structure and the lexical expression of syntactic relations. In K. Hale and S.J. Keyser (Eds.), The view from building 20: Essays in linguistics in honor of Sylvain Bromberger (pp. 53–109). Cambridge, Mass.: MIT Press. Heim, I. and A. Kratzer (1998). Semantics in Generative Grammar. Malden: Blackwell Publishers. Hoekstra, T. (1984). Transitivity. Grammatical relations in GB theory. Dordrecht: Foris. Hofmann, T. (1966). Past tense replacement and the modal system. Reprinted in J. McCawley (Ed.), Syntax and Semantics, Volume 7 (pp. 85–100). New York: Academic Press. Larson, R. (1988). On the double object construction. Linguistic Inquiry, 19, 335–391. Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press.

Modality and polarity

Perlmutter, D. (1970). The two verbs begin. In R. Jacobs and P. Rosenbaum (Eds.), Readings in English transformational grammar (pp. 107–120). Boston: Ginn. Pollock, J.-Y. (1989). Verb movement, universal grammar and the structure of IP. Linguistic Inquiry, 20, 365–424. Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in Philosophical Linguistics, Series 1. Wyngaerd, G.V. (1994). Pro-legomena, distribution and reference of infinitival subjects. Berlin/New York: Mouton de Gruyter.



Modals, objects and negation in late Middle English* Frits Beukema and Wim van der Wurff University of Leiden

In late Middle English, modal verbs occur in a construction with an at that time already exceptional object-verb order. The object in this type of construction is virtually always a negative constituent. This article investigates the nature of the modal verbs involved and their interaction with negation. It also proposes an articulated structure for the various manifestations of this construction and offers an account of its disappearance, focusing on the role played by language acquirers.

.

Introduction

Three of the many concerns in the study of modality have been the difference between deontic and epistemic modality, the interaction between modality and negation, and the coherence – or lack thereof – of the class of modal verbs. These areas have been studied from both the synchronic and the diachronic point of view. In this paper we take a close look at a specific construction with a modal verb in it as it is found in late Middle English texts, and ask ourselves the following questions: 1. Is there any difference in the behaviour of epistemic vs. deontic modals, either as a group or individually, in this construction? 2. Does this construction show any restrictions with respect to possible scope interactions of modality and negation? 3. More generally, what is the structure of this construction? 4. Can structural aspects of this construction shed light on its eventual disappearance? More broadly speaking, what role did acquisitional factors of this construction play in this?



Frits Beukema and Wim van der Wurff

A large part of this paper will be devoted to a discussion of word order in late Middle English, a necessary prerequisite for finding answers to the questions posed above. In doing so, we can draw on a large amount of earlier work, since in work on the historical development of the syntax of English, word order studies have always been prominent (see the review in Chapter 4 of Denison 1993). One of the reasons for the popularity of word order studies is no doubt the pervasive idea that English changed from a basic OV order in Old English (OE) to VO order in later stages of the language. Earlier studies on this development have sometimes presented this as a rather abrupt and sweeping change that took place in early Middle English (ME).1 But, as David Lightfoot points out, ‘One does not find one set of grammars being replaced by another set overnight; the textual record certainly does not suggest that English speakers had object-verb grammars replaced uniformly by verb-object grammars on some National Head-Change Day in the thirteenth century.’ (Lightfoot 1999: 107)

It is true that apart from sentences that clearly show movement of the object to clause-initial position, as in relative clauses or in topicalisation structures, the majority of sentences come to display VO order after 1200. However, the OV order is more tenacious and persistent than is often realised: as late as the middle of the fifteenth century, we still see that most prose texts contain examples of sentences with OV word order, while in poetry the OV order continues to be found until well into the nineteenth century. The late occurrence of these instances of OV order indicates that the OV-VO change is indeed a gradual phenomenon stretching over a much longer time span than previously assumed (for a description of the basic data, see Koopman and van der Wurff 2000). In this paper we will focus on the final stages of the change from OV to VO in prose, and consider the sentence patterns showing OV order most frequently in fifteenth-century prose texts. A first pair of examples is given in (1)–(2).2 (1) I wyll no thyng graunt ‘I will admit nothing’ (2) þey schuld no meyhir haue ‘They were not allowed to have a mayor’

(Paston Letters 59.15) (Capgrave’s Cronicles 199.6)

In Section 2 of this paper we discuss the OV order of OE and ME before 1400 and provide an analysis of this order in terms of the Minimalist Program of Chomsky (1995). We then go on in Section 3 to a discussion of OV order in fifteenth-century English, looking in particular at the contexts in which

Modals, objects and negation in late Middle English

OV order is found and illustrating these with examples. We will show that in fifteenth-century ME the OV order becomes restricted to a limited number of syntactic contexts. It will become clear that one particular type of OV pattern – where a negative object is enclosed between a modal verb and a lexical verb, as in (1) and (2) – is the most prominent among the OV patterns. In Section 4 we will undertake a closer examination of the OV patterns without a negative object, and try to determine their structure. In Section 5, we consider the OV pattern with a negative object in detail. In particular, we address the question of the scope interaction between the modal – epistemic and deontic – and the negative object, which will lead to the conclusion that fifteenth-century English must allow for two possible derivations of this sentence type. Basing ourselves on Kayne (1998), we will provide a sketch of these two structures. In Section 6 we consider the fifteenth-century OV facts from an acquisitional perspective and show that this can also shed light on the eventual disappearance of the construction in the sixteenth century.

. OV order in OE and ME until 1400 Let us begin our investigations with a brief discussion of the occurrence of OV in English texts written before 1400, providing an analysis of these data in conformity with the restrictive theory of phrase structure in Kayne (1994) and Chomsky (1995), as applied to the OE and ME data in Roberts (1997) and van der Wurff (1997a), respectively. First of all, we note that in embedded clauses in OE the predominant word order has the verb in final position, while in root clauses there is a V2 effect (see e.g. Denison 1993 for details). In showing this difference in the position of the finite verb, OE is similar to Modern Dutch and German. The sentences in (3) illustrate the OV order of OE embedded clauses: (3) a.

. . . þæt he his stefne up ahof (Bede 154.28) that he his voice up raised ‘that he raised up his voice’ b. . . . þæt ic þas boc of Ledenum gereorde to Engliscre spræce awende (Homilies of Ælfric, I, pref. 6) that I this book from Latin language to English tongue wouldtranslate ‘that I would translate this book from the Latin language into the English tongue’





Frits Beukema and Wim van der Wurff

As is clear from the data given by Koopman (1995), sentences with OV order are also found in root clauses, i.e. not all OE root clauses have V2. An example is given in (4): (4) Dryhten ðæt gecyððe God that showed ‘God showed that’

(Cura Pastoralis 113.8; Koopman 1995: 132)

These basic facts of OE word order have received a wide variety of treatments in the linguistic literature. Here we follow the analysis of Roberts (1997), who adopts the Minimalist Program of Chomsky (1995), supplemented with the proposals in Kayne (1994), and suggests that the underlying order of all OE clauses is VO. Case checking of the object takes place in AgrOP, which means that the object at some point in the derivation has to move to the specifier of AgrOP, with the verb moving to AgrO for reasons of equidistance (as expressed in Holmberg’s generalisation). In the sentence in (4), movement of the object ðæt has taken place in the overt syntax, yielding a structure as in (5). In line with common assumptions, we represent the subject Dryhten as having moved to the specifier of AgrSP. (5) [AgrSP Dryhten [TP [AgrOP ðæti [AgrO [V gecyððej]][VP tj ti ]]]]

A similar analysis can be given for the sentences in (3a) and (3b), although there the presence of material intervening between the object and the finite verb suggests that the object has moved further to a higher position in the clause; for specific proposals in this respect, see Roberts (1997). In van der Wurff (1997a) an analysis of OV order along the same lines is adopted to account for cases of OV order in fourteenth-century ME. It is suggested there that the lexical verb could still move to AgrO overtly at that period, and it is shown that a large number of sentence patterns can be analysed successfully in this way. Consider the sentences in (6) and (7): (6) hi ablent men zuo þet hi ham-zelue ne knawyþ ne yzeþ hire misdedes ne hire folies ne hire wyttes (Ayenbite of Inwyt 16.25f) it blinds men so that they themselves not know nor see their misdeeds nor their follies nor their minds ‘it [i.e. the sin of pride] blinds men so that they do not know themselves nor see their misdeeds or follies or know their own minds’ (7) Mykel lufe he schewes, þat never es irk to lufe, bot ay standand, sittand, gangand, or wirkand, es ay his lufe thynkand, and oftsyth þarof es dremande (Ego Dormio 3ff)

Modals, objects and negation in late Middle English

Much love he shows who never is weary to love but always standing, sitting, going or working is always his love thinking, and often thereof is dreaming ‘Much loves he shows that is never too weary to love, but always, standing, sitting, walking, or working, thinks of his love, and often dreams of it’

In (6) and (7), the objects (ham-zelue and his lufe) have moved to a position left of the lexical verbs (knawyþ and thynkand). If we take the VO order to be the underlying order, these OV patterns (and others like them, as discussed in van der Wurff 1997a) make it necessary to have specific mechanisms to account for the surface position of the object. An analysis in terms of movement of the object to SpecAgrOP, with concomitant movement of the verb to AgrO, therefore seems warranted for these fourteenth-century data as well. The relevant part of the structure of (6) is given in (8). (8) [CP þet [AgrSP hi [TP [AgrOP ham-zeluei [AgrO [V knawyþj ]] [VP tj ti ]]]]]

Of course, in fourteenth-century writings, VO order is very common as well. An example with several instances of VO order is given in (9). (9) Forþi, þof we seme in penance withowten, we sal have mykel joy within, if we ordayne us wysely to Goddes servyce, and sett in hym al owre thoghtes, and forsake al vanyte of þis worlde (Ego Dormio 77) ‘Therefore, although on the outside we seem to be doing penance, we will have great joy inside if we devote ourselves wisely to God’s service and direct all our thoughts to him and forsake all vanities of this world’

Following general minimalist guidelines, we assume that the VO order in (9) can be accommodated by allowing the object to move to SpecAgrOP at the level of Logical Form, i.e. after SPELL-OUT. Thus, the minimalist program can capture the two orders found in fourteenth century ME (OV and VO) by postulating the presence of an AgrO projection. Since surface OV and VO word orders are found side by side in a great many ME texts, it is safe to assume that the two options could coexist in the grammars of ME speakers. This suggests that the object could move to SpecAgrOP overtly as well as covertly, in this way accounting for the optionality. An analysis along these lines has been widely accepted to account for object shift phenomena in Modern Dutch, Modern Norwegian, Swedish, Danish, Icelandic and other languages. For further details of this analysis as it applies to ME, see van der Wurff (1997a). This analysis predicts that OV should be productive in all contexts, and that is exactly what we find until about 1400: OV order occurs with all types of NP objects, in all types





Frits Beukema and Wim van der Wurff

of clauses and with all types and combinations of verbs. In the fifteenth century, however, this is no longer the case, as we will see in the following section.

. OV in fifteenth-century English In fifteenth-century English prose texts, the use of OV order becomes restricted to a limited number of sentence patterns in a limited number of syntactic contexts.3 They are presented in (10): (10) a. b. c. d. e. f. g. h.

auxiliary + negative object + V auxiliary + quantified object + V object (+auxiliary) + V (in relative clause) object (+auxiliary) + V (in coordinate clause) object + Vinf object + V-ing object + imperative V fixed expressions

Examples illustrating these patterns, which are in italics, are provided in (11)–(18). (11) God [. . . ] withoute whom he perceyveþ that he may no gode þinge do (Imitation of Christ, 1.12.6) ‘God . . . , without whom he realises he can do no good thing’ (12) And if my feodaryes whiche lye in þe tye of my gret cofyr may ought wisse ther-in, lete them se it (Paston Letters, 55.6) ‘And if my feudal documents, which are in the compartment of my big chest, can shed any light on this, show them to these people’ (13) Besechyng al them that this litel werke shal see, here or rede to have me excused (Caxton’s Own Prose, 46a.29) ‘Requesting everyone who sees, hears, or reads this little work to forgive me’ (14) serteyn of the seid felechep have take fro John Wylton wythoute any cause, hese net, hese shep, and odyr catell, and summe ther-of have saltyd and eten, summe there-of have aloyned (Paston Letters, 40.28) ‘certain members of this company have without good reason taken from John Wylton his cows, his sheep and other cattle, and have salted and eaten some of them, and have taken away others of them’

Modals, objects and negation in late Middle English

(15) but I sall do hem examyn here-aftere as hastely as y may (Paston Letters, 190.48–49) ‘but I shall have them examined after this as soon as possible’4 (16) [. . . ] wheche they wolnot do in evel exaumple gevyng (Paston Letters, 40.78) ‘. . . which they will not do for fear of setting a bad example’ (17) Yive into my mouthe a trewe worde and a stable, and a fals wyly tonge make farre fro me (Imitation of Christ, 3.50.26) ‘Put a true and stable word into my mouth, and keep a false and wily tongue far away from me’ (18) for so God me helpe, I send yow a lettyr to london a-non aftyr Kandylmas (Paston Letters, 321.5) ‘for so God help me, I will send you a letter in London immediately after Candlemas’

As example (14) shows, some instances of pattern (10d) feature a negative or quantified object, making them somewhat similar to patterns (10a) and (10b). The same is true for pattern (10c). However, in both (10c) and (10d) the object does not need to be negative or quantified, thus suggesting that these are indeed distinct patterns. In pattern (10e) the object is usually pronominal, but full lexical noun phrases also occur (e.g. Paston Letters, 137.4). To obtain a rough and ready idea of the frequency of these patterns, we may consider the Paston Letters, a collection of letters and documents written between 1425 and 1500 by members of three successive generations of the Paston family. In them, each of the OV patterns of (10) is found, except for objects preceding imperative verbs. The percentages for the eight types are as follows (there are also a few cases of other OV patterns, especially in the more formal documents, which we disregard here; hence, the percentages in (19) do not total 100): (19) a. b. c. d. e. f. g. h.

auxiliary + negative object + V auxiliary + quantified object + V object (+aux) + V (in relative clause) object (+aux) + V (in coordinate clause) object + Vinf object + V-ing object + imperative fixed expressions

47% 7% 1% 14% 7% 9% 0% 1%





Frits Beukema and Wim van der Wurff

Clearly, the type with an auxiliary and a negative object stands out as the most frequent OV context. This is true not only for the Paston Letters, but also for most other fifteenth-century prose texts that we have investigated. These include original English works as well as translations, and as far as text type is concerned, range from beast fables (The History of Reynard the Fox) to religious prose (Lollard Sermons, Wycliffite texts, The Imitation of Christ); didactic texts (Jacob’s Well and Dives and Pauper); tales of the exotic (Life of Buddha, Mandeville’s Travels) and personal and business letters (Paston Letters, Cely Letters). Altogether, these texts contain over one million words. A first report of the data in these texts, with details on frequency, is provided in van der Wurff (1997b); detailed presentation of the data in the Paston Letters can be found in Moerenhout and van der Wurff (2000). It should be emphasised, however, that in all the texts in this period the prevalent word order is VO, also in all the contexts presented in (10). Interestingly, the first four patterns in (10) have a close parallel in Modern Icelandic. In (20), we give one Icelandic sentence for each pattern. (20) a.

Ég hef engan séð. I have nobody seen ‘I have seen nobody.’ b. Jón hefur fáar bækur lesið. John has few books read ‘John has read few books.’ c. þetta er maðurinn sem smásögunna skrifaði. this is the-man that the-short-story wrote ‘This is the man that wrote the short story.’ d. Suma hefur hann sært og suma drepið. Some has he wounded and some killed ‘He has wounded some and killed others.’

A full discussion of this parallel and the reasons for its existence can be found in van der Wurff (1999).

. OV patterns without a negative object Let us now turn to the analysis of the patterns summarised in (10). Before discussing the most frequent pattern (10a) in more detail, we will first devote some attention to the other types (for a fuller discussion, see van der Wurff 1999). For type (10b), which features a quantified preverbal object, an analysis

Modals, objects and negation in late Middle English

making use of the rule of Quantifier Raising (QR) would seem plausible. This would allow us to adopt the proposals made in Belletti (1990: 78) and Cinque (1995: 276–286), who deal with French sentences like the one in (21). (21) Il a tout compris. He has all understood ‘He has understood everything.’

Cinque (1995: 286) suggests that the position of the preverbal object in (21) is due to an application of A -movement ‘presumably anticipating in the syntax what must in any event happen in Logical Form, where quantifiers have to move to scope positions’. If this is correct, the same kind of analysis could be applied to fifteenth-century English quantified objects. In these cases too, we could say that the quantified object undergoes QR before SPELL-OUT, resulting in a structure as in (22) for sentence (12). (22) [AgrSP . . . my feodaryess . . . [AgrS maym ] [VP oughtx [VP tm [VP ts wisse tx . . . ]]]]

The subject my feodaryes has raised to the specifier of the AgrS projection, to check its N-feature, while the modal may has raised to AgrS. In the analysis provided in (22) the quantified object is adjoined to the highest VP, by QR, so that it has scope over all the elements in the clause, or their traces.5 An alternative analysis would be to assume that the quantified object has moved overtly to SpecAgrOP, to check its Case. This, however, would raise the question as to why it is only quantified objects that move to SpecAgrOP overtly. It seems more plausible to account for the empirical facts of fifteenth-century English by appealing to the well-established rule of QR, as is done in (22). These facts thus seem to undermine recent attempts in Minimalism to do away with the rule of QR altogether (cf. Hornstein 1994, 1995). For the patterns in (10c–g) we propose a unified analysis. What these cases have in common is that they have a non-lexical subject that receives the external theta-role from the verb assigning the internal theta-role to the preverbal object. This makes it possible to say that in the relevant sentences the object has been topicalised and actually precedes the entire clause, as shown schematically in (23). (23) Topicalised Objecti . . . null subject . . . V . . . ti

In Rizzi (1997), the left periphery of the clause receives a more articulated structure, in which CP is decomposed into a number of functional projections, among which a Topic Phrase (TopP) is postulated. If we follow Rizzi’s sugges-





Frits Beukema and Wim van der Wurff

tion, the topicalised object would be located in the specifier of this TopP, which dominates AgrSP, in the specifier of which the invisible subject is located. The null subject in (23) may either be the trace of a moved category, which we would postulate for pattern (10c), or a base-generated element, which seems plausible for patterns (10d–g). We would therefore suggest that, although these five cases look as if they display OV order, they are in fact instantiations of OSV order. This immediately explains why, in sentences of this type with an auxiliary, the object precedes the auxiliary, as in (13) and (14), while in patterns (10a) and (10b) the object is wedged between the auxiliary and the lexical verb.

. The OV pattern with a negative object We now turn to the analysis of the most frequent OV pattern, the one in (10a). First we note that, although some examples with the auxiliary have are attested, in the majority of cases the auxiliary in this pattern is a modal, and this is the type on which we are going to focus. It appears that the precursors of virtually the whole range of the present-day English modals and marginal modals (apart from must and need) are attested, as shown by the examples in (24)–(35). (24) I suppose he shall no cause have, ne his discyplis nothere, to avante of so short a remedy þer-of as ye wryght they sey now (Paston Letters 52.12) ‘I suppose that he and his vassals will have no reason to boast of such a quick remedy against it as you write that they are now saying’ (25) I [. . . ] warnyd yow and Richard that ye shuld nomore stuffe take in-to myn hows without ye paijid in hand (Paston Letters, 75.16) ‘I warned you and Richard that you should not bring more goods into my house without paying cash for them’ (26) I wyll no thyng graunt with-out the vnder-shreves assent (Paston Letters, 59.15) ‘I will assent to nothing without the deputy-sheriff’s consent’ (27) and sche seyd nay, be here feyth sche wuld no more days 3eve 3w þer-jn (Paston Letters, 128.13) ‘and she said no, by her faith, she would not allow you more time in this matter’ (28) And also I pray yow þat ye woll do bey ij gode hattis for yowr sonys, for I can none getyn in þis town (Paston Letters, 138.32–33) ‘And I also beg you that you have someone buy two good hats for your sons, since I cannot get any in this town’

Modals, objects and negation in late Middle English

(29) and told here þat 3e had sergyd to a fownd wrytyng þer-of and 3e kwd non fynd in non wyse (Paston Letters, 128.22–23) ‘and told her that you had searched in order to find written evidence of this, and could not find any at al’ (30) Item, that as for distreyn for rent or ferm, thow the Dewk had tytill, as he hath not, he may non ask til the next rent day after his entré, that is Michelmes (Paston Letters, 73.46) ‘Item, that concerning seizure of property for rent or lease payments, although the Duke was entitled to it, which he is not, he may not ask for any until the next rent day following his taking possession, which is Michaelmas’ (31) enfourmyng hym how that your lyffelod hath stond this ij yere in such trobill þat ye myght right nought haue of it (Paston Letters, 210.9–10) ‘informing him that your property has caused so much trouble these past two years that you have not been able to derive any income from it’ (32) whan men mon non witnesse han of trewþe (Dives and Pauper, 233.12) ‘when people are unable to have any witness of the truth’ (33) and the juges [. . . ] comaunded the scheryf to delyuer the seide Bone wythoute any fyne made, seyng that he out non to make (Paston Letters, 189.39–41) ‘and the judges ordered the sheriff to set the aforesaid Bon(d)e free without imposing a fine, because he ought to impose none’ (34) afor þe flood men vsyd no wyn to drynkyn ne to etyn flesch (Dives and Pauper, 305.38) ‘before the flood people used to drink no wine, nor eat meat’ (35) He þat is taught with the yifte of grac [. . . ] dar noþinge as[c]ryve to himselfe (Imitation of Christ, 2.10.15) ‘He who has been taught with the gift of grace . . . dare ascribe nothing to himself ’

The fact that nearly all modals and marginal modals are attested in this construction type can be construed as an indication that, as also argued in Warner (1993), in the fifteenth century these elements are becoming increasingly differentiated from the class of lexical verbs. We have found only one example in a fifteenth-century text of a negative object occurring in a clause with two modals, the object appearing after the first modal. The example is given in (36).





Frits Beukema and Wim van der Wurff

(36) [. . . ] so foul þat þou schalt nou3ht elles mowe se þerynne bot fylþehede and wrecchednesse (The Fyve Wyttes, p. 14 l.28f) ‘. . . so filthy that you will not be able to see anything else in it but foulness and wretchedness’

Notice that double-modal constructions are not uncommon in fifteenthcentury texts, but apart from the one in (36), we have not found any with a negative object. In discussions of the meanings of the modals, a distinction is often made between epistemic and deontic meanings. Epistemic modality involves ‘the truth, probability, possibility etc. of the whole proposition’, whereas deontic modality involves ‘permission given or obligation imposed performatively by the speaker/writer (or in a question, the hearer/reader)’ (Denison 1993: 293; see also Palmer 1986, 1990). If we apply this definition to the modals in (24)–(35), we see that they show both deontic and epistemic uses. In (24), shal refers to a (hypothetical) future, and is therefore most plausibly interpreted as epistemic; in (25), shuld expresses an obligation, and is therefore deontic. In (26), wil expresses willingness, hence is deontic; the same holds for the past tense wulde in (27); in (28) and (29), can and could both express ability, so they are deontic; in (30), may expresses permission, so it is deontic; in (31), might expresses ability, so again deontic; in (32), the modal is mon (a loanword from Scandinavian) and conveys inability, so it is also deontic; in (33), ought expresses imposition/obligation, so is deontic. The verbs used to in (34) and dare in (35) are somewhat difficult to classify, since it is not immediately clear how they fit into the epistemic–deontic distinction. The majority of the modals in these examples, then, are deontic, which in fact was generally the case for all modals in the older stages of the language (see Traugott 1989). However, there are also unequivocal examples of epistemic modals in this sentence pattern, as we have seen. As for the nature of the negative objects: they are usually quite simple, consisting of a single word (none or nought/nothing), or two or sometimes three words (no (+more/other/Adj) + noun). Longer structures are not found. Note the interesting pattern in (37), in which it appears as if part of the object (i.e. money) has been topicalised (triggering V2), with the rest (i.e. non) remaining downstairs. A similar effect has been observed by van Riemsdijk (1989) in Southern German sentences like Ein Buch hat er noch keins (lit. a book has he yet none).6 (37) but money can I non get. ‘but I can get no money’

(Paston Letters 338.59)

Modals, objects and negation in late Middle English

In most examples of the construction, the lexical verb following the negative object is dynamic, as in (25)–(28), (30), and (33)–(35). However, stative verbs also occur, notably the verb have, as in (24), (31), and (32), and see expressing direct sense perception, as in (36) and (38).7 (38) And whan he coude nowher none see he stopped the hole with sande and made hit euen and playn lyke to the other grounde by (Reynard the Fox, 37.18–20) ‘And when he could see none anywhere, he filled the hole with sand and made it level and flat as the other ground’

As is well known, in sentences with a modal and a negative, there is an interaction of the modal verb with the negative in terms of scope. To give a simple present-day English example of such scope interaction, consider the interaction of the elements POSSIBLE and NOT with a predicate such as TRUE in (39), which yields the readings given in (40). (39)

POSSIBLE NOT TRUE

(40) a.

it is not the case that this is possibly true [NOT [POSSIBLE [THIS TRUE]]] b. it is possibly the case that this is not true [POSSIBLE [NOT [THIS TRUE]]]

(40a) represents the reading in which NOT has scope over POSSIBLE (‘it is impossible for this to be true’/‘this cannot be true’), while (40b) represents the reading in which POSSIBLE takes scope over NOT (‘it is possible that this is untrue’/‘this may not be true’). We will assume here that such scope relations are reflected in the syntactic structure of sentences (see e.g. May 1985 and much subsequent work).8 With this in mind, let us go back to examples (24)–(35). A simple and attractive approach to the relative order of negative object and modal would be to assume that the OV order always signals wide scope of negation over the modal, and that VO order with a negative object, which is also frequent in fifteenth-century texts, signals narrow scope of negation. To show that negation and modality do indeed interact in terms of mutual scope, Kayne (1998: 171, ex. 196) draws attention to the following Icelandic sentence: (41) Pabbi hennar mun neyða hana til að giftast engum father her will force her till to marry no-one ‘Father will force her not to marry anyone’





Frits Beukema and Wim van der Wurff

In this sentence, the verb neyða, possibly in combination with the modal mun, has scope over the negative quantifier engum, so that engum has scope in the embedded clause only. Therefore, the meaning of this sentence can only be paraphrased as: ‘her father will force her to remain unmarried’ and there is no ambiguity. The Icelandic sentence in (41) cannot mean: ‘there is no particular person such that her father will force her to marry him’ (in which the negative would have wide scope over the verb force). However, as Kayne (1998: 171) notes, the English sentence in (42) is ambiguous: (42) I will force you to marry no-one

No-one in (42) may have narrow scope in the infinitival clause, which implies that the sentence means ‘I will force you to remain unmarried’ or no-one may have wide scope in the matrix clause, yielding the meaning ‘There is no-one such that I will force you to marry her/him’. Although, as we tentatively put it above, it might seem attractive at first sight to link the occurrence of narrow scope and wide scope in the fifteenthcentury sentences to the occurrence of OV versus VO order, this quickly turns out to be too simple an approach: inspection of (24)–(35) shows that the fifteenth-century OV construction could in principle have both wide and narrow-scope readings. To show this, we repeat the examples and discuss each of them in turn. (24) I suppose he shall no cause have, ne his discyplis nothere, to avante of so short a remedy þer-of as ye wryght they sey now

Here a narrow-scope reading of the negative object no cause vis-à-vis the modal shall is most likely, in that epistemic shall refers to the future, i.e. the sentence can only be rendered as ‘it will be the case that he has no cause . . . ’. (25) I [. . . ] warnyd yow and Richard that ye shuld nomore stuffe take in-to myn hows without ye paijid in hand

This sentence basically has the same scope interaction as (24): deontic shuld takes scope over nomore stuffe, since the embedded clause must be interpreted as ‘I (i.e. Richard) am obliging you to take no more stuff into my house . . . ’. (26) I wyll no thyng graunt with-out the vnder-shreves assent

In this sentence, the negative no thyng has scope over the deontic modal wyll since the meaning conveyed is ‘there is nothing such that I’d be willing to grant it’.

Modals, objects and negation in late Middle English

(27) and sche seyd nay, be here feyth sche wuld no more days 3eve 3w þer-jn

Here, the reading ‘she was willing [to give no more days]’, which would constitute an act of negative giving, seems less likely. It follows that the negative object no more days must have scope over the deontic modal wuld. (28) And also I pray yow þat ye woll do bey ij gode hattis for yowr sonys, for I can none getyn in þis town

The most natural interpretation of this sentence is something like ‘there isn’t any good hat such that I can buy it in this town’. The negative none therefore has scope over the deontic modal can. (29) and told here þat 3e had sergyd to a fownd wrytyng þer-of and 3e kwd non fynd in non wyse

This sentence is largely similar to the previous one, since the negative object non has scope over the deontic modal kwd, where the meaning is clearly not: ‘they had the ability of not finding the writing’. (30) Item, that as for distreyn for rent or ferm, thow the Dewk had tytill, as he hath not, he may non ask til the next rent day after his entré, that is Michelmes

Again, the negative object non has scope over the deontic modal may since the meaning of (30) is clearly not: ‘he has the right not to ask, i.e. to abstain from asking, rent.’ (31) enfourmyng hym how that your lyffelod hath stond this ij yere in such trobill þat ye myght right nought haue of it

The modal might is deontic here since it means ‘were not able’. Since the meaning of ye might right nought have cannot be ‘you had the ability to not have any livelihood’, the negative object clearly has scope over the modal. (32) whan men mon non witnesse han of trewþe

Analogous to previous sentences, the negative object in this sentence (i.e. non witnesse) must needs have wide scope, i.e. the sentence means ‘when people are unable to produce a witness’. (33) and the juges [. . . ] comaunded the scheryf to delyuer the seide Bone wythoute any fyne made, seyng that he out non to make





Frits Beukema and Wim van der Wurff

Non as the negative object is preceded by the deontic modal out (‘ought’), which must have scope over it since the meaning is ‘he ought to not make a fine (‘refrain from imposing a fine)’. (34) for why afor þe flood men vsyd no wyn to drynkyn ne to etyn flesch

There is not very much to choose between a wide-scope reading or a narrowscope reading of the negative no wyn vis-à-vis the modal usyd in this sentence. In many contexts ‘I am used to not drinking wine’ means the same as ‘I am not used to drinking wine’, which implies that there is no discernible interaction between the negative object and the modal. Note that used (to) in Modern English is frequently equivalent to a habitual past tense, and then has no modal meaning at all, as we have already observed above. (35) He þat is taught with the yifte of grac [. . . ] dar noþinge as[c]ryve to himselfe

Dare is equally difficult to classify along the epistemic/deontic parameter. It is obviously subject-oriented (Palmer 1990: 112), which suggests categorising it rather as a deontic than an epistemic modal. The scope of noþinge with respect to the (deontic) modal dar cannot but be wide: the sentence only means ‘there is nothing such that he dare ascribe it to himself ’. What we can conclude from the survey so far is firstly, that the modals in (24)–(35) nearly all have a deontic meaning: only one case, i.e. (24), contains an epistemic modal. Secondly, among the twelve examples there are only three in which the modal clearly has scope over the negative object, while the rest shows the negative object having wide scope over the modal. Nevertheless, both possibilities can be clearly evidenced, which implies that the grammar of fifteenth-century English should in principle make both options available, with the actual scope interaction between negation and modality being determined in large measure by the meaning of the modal, as is well documented for Modern English. We now turn to the syntactic representation of these sentences, using the ideas in Kayne (1998). We take over the clausal architecture sketched there, but modify it slightly in that we allow the presence in the same clause of a NegP dominating the modal projection on the one hand, and a NegP dominating the lexical verb projection on the other hand. We propose that the high NegP is accessed in wide-scope readings of negation, while the low NegP is activated in cases of narrow scope. The modal is assumed to head its own VP. The two general structures are given in (43) and (44).

Modals, objects and negation in late Middle English

(43) AgrSP TP NegP VP [modal] AgrOP VP (44) AgrsP TP VP [modal] NegP AgrOP VP

The essential idea that we borrow from Kayne (1998) is that a negative object must overtly move to SpecNegP, which may or may not be followed by preposing of the remnant VP. With this background, let us now consider the structure of sentences with a negative object and a modal in fifteenth-century English, i.e. sentences of the type he could nothing say. We sketch the possible derivations in (45), where in (45a) nothing would be in the lower SpecNegP and have narrow scope, while in (45b) nothing is in the higher SpecNegP and has wide scope. In both cases could has moved out of its VP into AgrS, and there is no preposing of the remnant VP [say ti ]. (45) a.

He could nothing say (‘he was able to not say anything, i.e. to remain silent’) AgrSP he

AgrS’ AgrS

TP

couldm

VP V tm

NegP nothingi Neg

Neg’ AgrOP VP say

ti





Frits Beukema and Wim van der Wurff

b. He could nothing say (‘he was not able to say anything’) AgrSP he

AgrS’ AgrS

TP

couldm NegP nothingi

Neg’

Neg

VP V

AgrOP VP

tm say

ti

Note that it is possible to extract part of the negative object from SpecNegP and move it to a higher position (presumably SpecCP), with concomitant V2 effects, as in the pattern in (46), which repeats the relevant part of example (37). (46) but [CP moneyi [C can] [AgrSP I [NegP [non ti ] get]]]

Of course, sentences with a postverbal negative object also existed in fifteenthcentury English. An example is given in (47). (47) he schal make no pes with þe kyng of Frauns (Cronicles 200.21) ‘he must not conclude peace with the king of France.’

This type of sentence can be analysed in the same way as in (45), again following the general guidelines of Kayne (1998). In a sentence like He could say nothing, nothing has moved to SpecNegP. This can be the low NegP, which would imply that remnant VP-preposing has taken place, moving say and the trace of the negative object to a functional projection immediately dominating the low NegP. This would result in narrow scope of the negative object vis-à-vis the modal. The structure is given in (48a).

Modals, objects and negation in late Middle English

(48) a.

He could say nothing (‘He was able to not say anything, i.e. to remain silent’) AgrSP he

AgrS’ AgrS

TP VP

couldm

FP

V tm

Spec [say to]vp

NegP nothingo Neg

Neg’ AgrOP VP tvp

Here, the modal could and its trace occupy positions higher than the position of the negative object. However, in the reading in which the negative object has wide scope with respect to the modal, the negative object is in the higher NegP, where it c-commands the trace of the modal. What has happened in such cases is that the negative object has moved out of the VP [could say nothing] and that the remnant VP [could say t] has been fronted to the specifier of the functional projection dominating NegP, with subsequent head movement of the modal to AgrS. This possibility is represented in (48b).





Frits Beukema and Wim van der Wurff

(48) b. He could say nothing (‘He was not able to say anything’) AgrSP he

AgrS’ AgrS

TP FP

couldm Spec [tm say to]vp

NegP nothingo Neg

Neg’ VP tvp

In both (48a) and (48b), the modal moves to AgrS for reasons of feature checking; this has no effect, however, on the scope relations holding between the negative object and the modal.9 Naturally, fifteenth-century English also features sentences with a negative object but no modal, i.e. sentences like the one in (49). (49) And syth þey louyn no pes ‘And since they love no peace’

(Dives and Pauper 198.22)

In such sentences the prevalent surface order is VO; hardly any cases with OV order (i.e. þey no pes louyen) are found. Within the framework of Kayne (1998), the negative object no pes must have moved to SpecNegP, with concomitant movement of the finite verb louyn to AgrS, possibly by means of remnant VPmovement as an intermediate step. Supporting evidence for a derivation along these lines can be found in late Middle English sentences of the type He spoke never the truth, which was the usual order until the end of the fifteenth century, when the modern order He never spoke the truth became usual (see also Roberts 1993). The structure of (49) is represented in (50):

Modals, objects and negation in late Middle English

(50)

AgrSP þey

AgrS’ TP

AgrS louynv

Spec

T¢ NegP

T tv

Neg’

no peso

AgrOP

Neg tv

Spec to

AgrO’ VP

AgrO tv

tv

to

What we have shown in this section is first of all that the pattern displaying a modal followed by a preverbal negative object is fully productive in the fifteenth century. Virtually all modals, both epistemic and deontic, occur in this pattern while the lexical verbs can either be stative or dynamic. We have also shown that the two types of scope interaction of the modal with the negative object are instantiated in the data and we have postulated two distinct structures corresponding to these different scope readings. We think that we have now answered the first three questions formulated in the introduction and we now turn to the fourth question, which involves the acquisition and disappearance of the construction.

. On acquisition and disappearance We have seen that, historically, three stages need to be distinguished: A. In the fourteenth century, all manner of objects can be found preverbally (see Foster and van der Wurff 1995). B. In the fifteenth century, preverbal objects become restricted to the types listed in (10) (see van der Wurff 1997b).





Frits Beukema and Wim van der Wurff

C. In the sixteenth century, preverbal objects of all types disappear (see van der Wurff and Foster 1997). As discussed in Section 2 of this paper, for stage A it can be assumed that all objects could in principle move to SpecAgrOP before SPELL OUT for reasons of case checking. As for the transition from stage A to stage B, it appears that around 1400, probably for reasons related to the loss of inflectional marking, overt movement of objects to SpecAgrOP was no longer available. As suggested in van der Wurff (1999), the effect was that surface OV order became impossible except in those cases that allowed the object to be reanalysed as occupying some other position than SpecAgrOP. This could be SpecNegP, as in cases like (10a), or a position adjoined to VP, as in (10b), or the specifier of a Topic Phrase (TopP), as in cases (10c–g). In other words, language acquirers could avail themselves of other possibilities licensed by UG when dealing with sentences with OV order. It discussing the transition from stage B to stage C, we restrict our attention to the most frequent pattern, the one with a modal and a negative object. In the acquisition of the structure of clauses with a negative object, children may reasonably be assumed to take their cues mainly from simple clauses, i.e. He said nothing, He spoke to no-one etc., rather than from more complex cases containing modals. In (50), we proposed a structure for these cases involving V-to-I movement. Roberts (1993: 246–273) relates the occurrence of V-to-I movement to the existence of a separate plural verbal inflection, as in the Middle English forms given in (51), in which the relevant inflections are italicised. (51) a. . . . þey louyn . . . b. The riche folk, that embraceden and oneden al hire herte to tresor of this world (Parson’s Tale 192) ‘The rich people who embraced and joined their whole heart to the treasure of this world’.

Roberts notes that by the end of the fifteenth century, the plural inflection had withered away (see also Lass 1992: 97f), and that as a consequence V-to-I movement disappeared.10 In terms of language acquisition, Roberts’ account would imply that at this point in time children would no longer hear forms like They louyn or They embraceden, but only forms like They love and They embraced. The morphological trigger for the syntactic verb movement rule had disappeared by that time and children would therefore no longer be exposed to the principal clue from which they could deduce that V-to-I was possible in their language. What

Modals, objects and negation in late Middle English

would happen when these V-to-I-less children heard sentences like They love no peace? According to the analysis proposed above, no peace must be in SpecNegP, so the lexical verb love must have undergone movement as well. V-to-I movement no longer being allowed, the verb must have reached its surface position by way of remnant VP-preposing to the Spec of the FP dominating the NegP. No further movement of V to a higher position is allowed. The resulting structure is shown in (52). (52)

AgrSP they

AgrS’ AgrS

TP FP Spec [love to]vp

NegP no peaceo Neg

Neg’ VP tvp

From the consistent occurrence of They love no peace rather than They no peace love, the child would conclude (perhaps through the mechanism of ‘indirect negative evidence’, cf. Atkinson 1992; Lasnik 1990) that the operation of remnant VP-preposing is obligatory. As for the real-time order of acquisition of negative objects and modal verbs by late Middle English children, the following can be observed. Klima and Bellugi (1966: 197) have pointed out that in Modern English, constructions with negative constituents appear in children’s speech only after they have reached the stage of mean utterance length (MLU) 4.0. We assume that in late Middle English such constructions were acquired somewhat earlier since the language at that stage lacked the competing negative strategy with not . . . any (see Tieken-Boon van Ostade 1997, who shows that what she and other scholars call ‘non-assertive any’ is virtually absent from late Middle English). Thus, the fifteenth-century language learner would be exposed to an input with rela-





Frits Beukema and Wim van der Wurff

tively many cases of negative constituents of the type nothing, none, nobody, no money rather than not . . . anything/anyone/anybody/any money, etc. Evidence pertaining to the acquisition of modal verbs in Stephany (1986: 387) reveals that modals become frequent in Modern English child language after MLU 3.5. There is no reason to suppose that this was any different in late Middle English, i.e. Middle English children could freely dispose of sentences containing modals and negative constituents at roughly the same time. What we have sketched as a conceivable Middle English acquisition scenario so far is that simple sentences without negative constituents are acquired before sentences containing negative constituents, and that sentences with modal expressions are acquired in tandem with sentences containing negative constituents. But did late Middle English children at this stage also acquire sentences containing a modal verb as well as a negative object (i.e. He could say nothing)? On the basis of arguments presented above, we contend that they did not. Recall that the disappearance of the order negative object-main verb (as in He could nothing say) crucially depends on remnant VP-preposing being obligatory. This obligatory nature could only be established by the language learner on the basis of the consistent use of the order main verb-negative object in clauses without modals (i.e. He said nothing). Hence, this type of sentence must have been acquired before the type containing a negative constituent as well as a modal. Our conclusion therefore is that for fifteenth-century children, the place of negated constituents like nothing is established on the basis of simplex clauses, i.e. clauses with a single finite verb like He said nothing. Sentences containing modals appear to have played no role in this process. The disappearance of the OV pattern with a modal and a negative object therefore suggests that the structure of clauses featuring both negation and modality is not acquired as such by children; rather, children figure out separately the structure of clauses with only negation, and the structure of clauses with only a modal. Sentences displaying the two phenomena together simply fall out from their interaction: they come for free. In all this, the absence of sentences such as They no peace love is crucial. If our analysis is correct, our account would therefore underpin the tenet that in language acquisition indirect negative evidence is also taken into account.

Notes * We thank the audience at the St. Andrews workshop for helpful suggestions, and two anonymous reviewers for detailed comments on an earlier version of this article.

Modals, objects and negation in late Middle English . It should be pointed out here that the OV underlying order of the earlier stages of the English language has not been undisputed: arguments have been put forward that Old English had an underlying VO order (Roberts 1997), or that both OV and VO order were simultaneously present in Old English and Middle English (e.g. Pintzuk 1996). . For the sources of all examples, see the list of Primary Sources at the end of this article. . For a full description of the empirical data, see van der Wurff (1997b). . Hem in this example must be the object of examyn rather than its subject. This and similar sentences are therefore not examples of the type of structures in which do is an ECM verb, as discussed in Roberts (1993: 282–292), an example of which is (i): (i)

thanne he dide the clerk of the council seek it ‘Then he made the clerk of the council seek it.’

(Paston Letters no. 432)

Rather, the sentence in (15) illustrates the type of structure that Roberts calls FP-do, in which the complement clause has a non-overt subject (small pro). . Our analysis in (22) makes crucial use of the Subject-in-VP hypothesis. The QR-moved object therefore c-commands the trace of the subject in SpecVP. . In these sentences, the fact that an X’ rather than an XP seems to be fronted (i.e. a case of ‘split topicalisation’, as van Riemsdijk calls it) might cause problems for a theory in which only heads of phrases or whole phrases are allowed to move. Van Riemsdijk proposes to account for these structures by means of a process he refers to as ‘regeneration’, which entails rebuilding the complete projection after movement, i.e. at S-structure. The form of the quantifying D (German keins/Middle English non, rather than the expected kein and no) van Riemsdijk (1989: 132, Note 5) explains by assuming that ‘the principles determining the form of the determiner, either by assignment or by checking, apply at or after S-structure, i.e. after the split has occurred’. An alternative to regeneration might be that these cases involve NP-fronting out of DP; this analysis has nothing to say about the form of the quantifying D, however. . Observe that in (38) the negative adjunct nowher as well as the negative object none appear to have moved to preverbal position. . Jackendoff (1972) appears to adopt different mechanisms for scope assignment by modals and by quantified DPs. This leaves the claims made in this paper unaffected. . An anonymous reviewer points out that Hornstein (1995) suggests that scope relations can also be defined on the positions occupied by traces of scope-bearing elements. We will leave the implications of this possibility aside here. . Lightfoot (1999: 163), basing himself on Warner (1997), suggests that V-to-I movement may have been operative in English as late as the eighteenth century. However, the precise analysis of the relevant data is a moot point. We observe that Lightfoot (1999: 176, Note 8) attempts to strengthen his point by adducing the Swedish sentence Om Jan inte köpte boken ‘If John did not buy the book’. To account for the place of inte ‘not’ to the left of the finite verb under a V-to-I account, Lightfoot has to assume that the sentential negator occurs to the left of I. However, the surface [Neg V] pattern has widely been taken as a diagnostic of the absence of V-to-I movement. See also Vikner (1995) and Holmberg and Platzack (1995).





Frits Beukema and Wim van der Wurff

Primary Sources Allen, H. (ed., 1931). English Writings of Richard Rolle, Hermit of Hampole. Oxford: Oxford University Press. Barnum, P. (ed., 1976). Dives and Pauper (EETS o.s. 275). Oxford: Oxford University Press. Benson, L. (ed., 1988). The Parson’s Tale. In The Riverside Chaucer. Oxford: Oxford University Press. Biggs, B. (ed., 1997). The Imitation of Christ: The First English Translation of the ‘Imitatio Christi’ (EETS o.s. 309). Oxford: Oxford University Press. Blake, N. (ed., 1970). The History of Reynard the Fox, Translated from the Dutch Original by William Caxton (EETS o.s. 263). London: Oxford University Press. Blake, N. (ed., 1973). Caxton’s Own Prose. London: André Deutsch. Brandeis, A. (ed., 1900). Jacob’s Well: An English Treatise on the Cleansing of Man’s Conscience (EETS o.s. 115). London: Kegan Paul, Trench, Trübner and Co. Bremmer, R. (ed., 1987). The Fyve Wyttes. Amsterdam: Rodopi. Davis, N. (ed., 1971). Paston Letters and Papers of the Fifteenth Century. Oxford: Clarendon. Hanham, A. (ed., 1975). The Cely Letters 1472–1488 (EETS o.s. 273). Oxford: Oxford University Press. Hirsh, J.C. (ed., 1986). Barlam and Iosaphat. A Middle English Life of Buddha (EETS o.s. 290). Oxford: Oxford University Press. Lucas, P.J. (ed., 1983). John Capgrave’s Abbreuiacion of Cronicles (EETS o.s. 285). Oxford: Oxford University Press. Miller, T. (ed., 1890–1898). The Old English Version of Bede’s Ecclesiastical History of the English People (EETS o.s. 95, 96, 110, 111). London: Kegan Paul, Trench, Trübner and Co. Morris, R. (ed., 1866). Dan Michel’s Ayenbite of Inwyt or Remorse of Conscience, Vol. 1 (EETS o.s. 23). London: Oxford University Press. Seymour, M.C. (ed., 1963). The Bodley Version of Mandeville’s Travels (EETS o.s. 253). Oxford: Oxford University Press. Sweet, H. (ed., 1871–1872). King Alfred’s West-Saxon Version of Gregory’s Pastoral Care (EETS o.s. 45, 50). London: Kegan Paul, Trench, Trübner and Co. Thorpe, B. (ed., 1846). The Homilies of the Anglo-Saxon Church. 2 vols. London: Ælfric Society.

References Atkinson, M. (1992). Children’s Syntax. Oxford: Blackwell. Belletti, A. (1990). Generalized Verb Movement. Turin: Rosenberg and Sellier. Chomsky, N. (1995). The Minimalist Program. Cambridge, Mass.: MIT Press. Cinque, G. (1995). Italian Syntax and Universal Grammar. Cambridge: Cambridge University Press. Denison, D. (1993). English Historical Syntax. London and New York: Longman. Foster, T. and W. van der Wurff (1995). The survival of object-verb order in Middle English: Some data. Neophilologus, 79, 309–327.

Modals, objects and negation in late Middle English 

Holmberg, A. and C. Platzack (1995). The Role of Inflection in Scandinavian Syntax. Oxford: Oxford University Press. Hornstein, N. (1994). An argument for Minimalism: The case of antecedent-contained deletion. Linguistic Inquiry, 25, 455–480. Hornstein, N. (1995). Logical Form: From GB to Minimalism. Oxford: Blackwell. Jackendoff, R.S. (1972). Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Kayne, R. (1994). The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kayne, R. (1998). Overt vs. covert movement. Syntax, 1, 128–191. Klima, E. and U. Bellugi (1966). Syntactic regularities in the speech of children. In J. Lyons and R. Wales (Eds.), Psycholinguistics Papers (pp. 183–208). Edinburgh: Edinburgh University Press. Koopman, W. (1995). Verb-final main clauses in Old English prose. Studia Neophilologica, 67, 129–144. Koopman, W. and W. van der Wurff (1999). Two word order patterns in the history of English: Stability, variation, and change. In R. Sornicola, E. Poppe and A. Shisha-Halevy (Eds.), Stability, Variation and Change of Word-Order Patterns over Time (pp. 259–283). Amsterdam: John Benjamins. Lasnik, H. (1990). On certain substitutes for negative data. In H. Lasnik, Essays on Restrictiveness and Learnability (pp. 184–197). Dordrecht: Kluwer. Lass, R. (1992). Phonology and morphology. In N. Blake (Ed.), The Cambridge History of the English Language, volume II 1066–1476 (pp. 23–155). Cambridge: Cambridge University Press. Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: Cambridge University Press. Lightfoot, D. (1991). How to Set Parameters: Arguments from Language Change. Cambridge, MA: MIT Press. Lightfoot, D. (1999). The Development of Language: Acquisition, Change, and Evolution. Oxford: Blackwell. May, R. (1985). Logical Form: Its Structure and Derivation. Cambridge, Mass.: MIT Press. Moerenhout, M. and W. van der Wurff (2000). Remnants of the old order: OV in the Paston Letters. English Studies, 81, 513–530. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press. Palmer, F. (1990). Modality and the English Modals. London: Longman. Pintzuk, S. (1996). Old English verb-complement word order and the change from OV to VO. York Papers in Linguistics, 17, 241–264. Riemsdijk, H. van (1989). Movement and regeneration. In P. Benincá (Ed.), Dialect Variation and the Theory of Grammar (pp. 105–136). Dordrecht: Foris. Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of Grammar: Handbook in Generative Syntax (pp. 281–337). Dordrecht: Kluwer. Roberts, I. (1993). Verbs and Diachronic Syntax. Dordrecht: Kluwer. Roberts, I. (1997). Directionality and word order change in the history of English. In A. van Kemenade and N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 397–426). Cambridge: Cambridge University Press.



Frits Beukema and Wim van der Wurff

Stephany, U. (1986). Modality. In P. Fletcher and M.Garman (Eds.), Language Acquisition: Studies in First Language Development, 2nd ed. (pp. 375–400). Cambridge: Cambridge University Press. Tieken-Boon van Ostade, I. (1997). Any or No: Functional Spread of Non-Assertive Any. In R. Hickey and S. Puppel (Eds.), Language History and Linguistic Modelling, Vol. 1 (pp. 1545–1554). Berlin: Mouton. Traugott, E. (1989). On the rise of epistemic meanings in English: An example of subjectification in semantic change. Language, 65, 31–55. Vikner, S. (1995). Verb Movement and Expletive Subjects in the Germanic Languages. Oxford: Oxford University Press. Warner, A. (1993). English Auxiliaries: Structure and History. Cambridge: Cambridge University Press. Warner, A. (1997). The structure of parametric change, and V-movement in the history of English. In A. van Kemenade and N. Vincent (Eds.), Parameters of Morphosyntactic Change (pp. 380–393). Cambridge: Cambridge University Press. Wurff, W. van der (1997a). Deriving object-verb order in late Middle English. Journal of Linguistics, 33, 485–509. Wurff, W. van der (1997b). OV-volgorde in vijftiende-eeuws Engels proza. In A. van Santen and M. van der Wal (Eds.), Taal in Tijd en Ruimte (pp. 73–84). Leiden: Stichting Neerlandistiek Leiden. Wurff, W. van der (1999). Objects and verbs in modern Icelandic and fifteenth-century English: A word order parallel and its causes. Lingua, 109, 237–265. Wurff, W. van der and T. Foster (1997). Object-verb order in 16th century English: A study of its frequency and status. In R. Hickey and S. Puppel (Eds.), Language History and Linguistic Modelling, Vol. 1 (pp. 439–453). Berlin: Mouton.

On the use and interpretation of root infinitives in early child Dutch* Elma Blom University of Utrecht

Spontaneous speech data from six Dutch children confirm the claim that root infinitives in Dutch are primarily used to express wishes, desires and needs (van Ginneken 1917; Ingram and Thompson 1996; Hoekstra and Hyams 1998). Strikingly, the association between modality and root infinitives does not appear from early on, but arises as children reach the optional infinitive stage. In this stage (as defined by Wijnen 1997) children begin to use finite verbs next to their non-finite counterparts: finiteness has become productive. In this article it is shown how children’s ability to use finite verb forms enables them to analyse the infinitive as a verb instead of a noun. For this reanalysis it is crucial that children recognise the infinitival suffix. The alternation with finite morphology functions as a cue for this recognition. Reanalysis of the infinitive accounts for the modal meaning of root infinitives as the availability of specific morphology makes it possible to attach a (modal) meaning to the infinitive.

.

Introduction

Early last century, van Ginneken (1917) observed that Dutch children use nonfinite utterances to refer to wishes, desires and needs. Recently, the observation that root infinitives (RIs) receive a modal meaning has been confirmed for Dutch child language (Wijnen 1997). The same generalisation seems to hold for the root infinitives of German children (Behrens 1993; Ingram and Thompson 1996; Lasser 1997). Based on these empirical studies, Hoekstra and Hyams (1998) formulated the Modal Reference Effect (MRE) for Dutch and German child language, which states that ‘with overwhelming frequency, RIs have modal interpretations’. In this paper I will discuss three proposals that account for the MRE: Boser, Lust, Santelmann and Whitman’s (1992) Null Auxiliary Hypothesis, Ingram and Thompson’s (1996) input-based hypothe-



Elma Blom

sis and Hoekstra and Hyams’ (1998) morphological marking hypothesis. The most recent account, proposed by Hoekstra and Hyams (henceforth: H&H), will be singled out for a more thorough discussion. At first sight, H&H’s account seems very promising: different cross-linguistic intuitions and observations are tied together in a model that covers a wide range of empirical data. In the second part of this paper, the MRE will be discussed from an empirical point of view. Data from six Dutch children show that the connection between modality and infinitive verb forms does not appear from early on. To connect modality to the infinitival form, children have to be able to make the finitenon-finite distinction. My claim is that Dutch children need this knowledge to categorise infinitives as verbs. The early frequent non-modal use of RIs by Dutch children and their shift to modal use reflects a categorial re-analysis of the infinitive by children.

. Root infinitives in Dutch RIs are sentences headed by a non-finite verb. The verb has an infinitival form, and in languages like Dutch and German it takes the final position. Null subjects are preferred (Weverink 1989; Krämer 1993; Haegeman 1995), and stative verbs hardly ever occur in RIs (Wijnen 1997). The meaning of RIs in Dutch and German is often described as ‘modal’ and their temporal reference is usually future (Behrens 1993; Lasser 1997; Wijnen 1997). Although RIs in Dutch and German child and adult language seem highly comparable (at least with respect to all the properties mentioned above), in this paper I focus on Dutch RIs only. See Lasser (1997) for details on German child and adult RIs. In adult Dutch, RIs are marginally acceptable. Their use is restricted to specific contexts: imperatives, interrogatives, exclamatives,1 announcements and assertions in narrative contexts (Blom 2000). An example of each usage is given in (1): (1) a.

Rennen! Run- ‘Run!’ b. Ook wat drinken? Also something drink- ‘Would you like something to drink too?’ c. Ajax dit jaar winnen? Kleine kans Ajax this year win-? Little chance ‘Ajax going to win this year? Slim chance!’

On the use and interpretation of root infinitives in early child Dutch 

d. Even washandje pakken Just washcloth get- ‘I am just going to get the washcloth’ e. Ik de hele tijd wakker liggen en jij gewoon slapen I the whole time awake lie-and you just sleep- ‘I am lying awake the whole time while you are sleeping’

In Dutch child language RIs are not restricted to the specific usages as in (1). Children’s use of assertive RIs, for example, is not restricted to the narrative genre. Examples of assertive child RIs are given in (2): (2) a.

in soel zitten in chair sit- ‘I want to sit in the chair’ b. Woef liggen Woef lie- ‘Woef [i.e. the dog] lies’

(Abel 1;11.26)

(Peter 1;09.20)

The use of assertive RIs is more widespread in child Dutch than in adult Dutch.2 On the other hand, exclamative RIs of the adult-like type in (1c) do not occur in child Dutch. The modal meaning of RIs in adult Dutch is often related to illocutionary force: deontic modality (obligation or the necessity to act) in imperatives, volitive modality3 (wishes) in interrogatives, epistemic modality (probability) in exclamatives and the modality-related category of intentionality in announcements. Assertions in narrative context do not receive a specific modal or modality-related reading. It is argued that they have an inchoative meaning (Lasser 1997), which is more aspectual than modal. Inchoative aspect, however, shares its future reference with deontic and volitive modality. The modal meaning of RIs in child Dutch shows some overlap with the adult RIs. Dutch child RIs can most often be paraphrased as ‘I want to INF’ or ‘You must INF’, leading to respectively a volitive and deontic modal interpretation.4 In contrast to adult RIs, the modal meaning of child RIs is not directly related to interrogative and imperative force.

. Modal RIs explained: Three proposals In this section three accounts explaining the modal use of RIs will be discussed. They fall within two general views on language acquisition: language acquisi-



Elma Blom

tion as an autonomous or as an input-driven process. Boser et al.’s (1992) Null Auxiliary Hypothesis and H&H’s (1998) recent proposal both belong to the former view, whereas Ingram and Thompson’s (1996) proposal is an example of the latter. . Null Auxiliary Hypothesis Boser et al.’s Null Auxiliary Hypothesis (1992) states that root infinitives contain a covert auxiliary. This auxiliary is not phonetically realised, but its presence is noticed through morphology, syntax and semantics. The covert auxiliary corresponds to a modal verb or dummy verb like do or go.5 In the former case the RI will have a modal meaning, in the latter a non-modal meaning. As the null auxiliary undergoes V-to-I raising, the infinitive main verb has to stay in its base position, which in Dutch is the final position. Although the Null Auxiliary Hypothesis accounts for the data in a fairly simple way, there are several empirical problems. I will briefly discuss some of these problems (see for a more extensive discussion H&H). Syntactically, the covert auxiliary behaves like an overt auxiliary. The constructions are comparable, hence it is expected that RIs behave as if they were ‘normal’ finite Vaux + Vinf constructions. This, however, is not the case. Topicalisation is not allowed in RIs whereas it is in finite sentences (Poeppel and Wexler 1993). The same asymmetry can be observed for wh-movement. Finite clauses allow wh-movement, whereas nonfinite clauses do not. Note that in this respect English child language differs from Dutch, German, Swedish and French child language (Phillips 1995 for English; Haegeman 1994 for Dutch; Kursawe 1994 for German; Santelmann 1994 for Swedish and Crisma 1992 for French). English child language does not show both of the asymmetries whereas the other child languages do. The final problem is concerned with licensing. As an empty category, the null auxiliary needs licensing. Boser et al. (1992) claim that this is done by an overt subject. One of the characteristics of RIs, however, is that the presence of a lexical subject is optional. . What they hear is what you get Ingram and Thompson (1996) claim that children learn modal RIs directly from the input they receive. Dutch and German children associate modality and non-finiteness (and conversely, non-modality and finiteness) because their caregivers very frequently use non-finite verb forms in modal contexts. Ingram and Thompson’s claim is not supported by empirical child-directed

On the use and interpretation of root infinitives in early child Dutch 

speech data. Moreover, Ingram and Thompson do not explain what is meant by ‘association’. Frequency counts have shown that input can indeed play a role in learning modal RIs. The connection between modality and non-finiteness in the children’s output is present in the direct language input they receive (Blom, Wijnen and Gillis 1998). This, however, does not tell us anything about the representation of modal infinitives in the child’s grammar. . Modal Reference Effect and the Eventivity Constraint H&H’s approach is based on a striking difference between Dutch and German on the one hand and English on the other. Ud Deen (1997) found that English children predominantly use RIs in non-modal contexts. H&H claim that this difference is caused by a semantic [–realised] feature that is marked by infinitival morphology. In languages like Dutch and German the infinitive has a suffix -en, which carries the [–realised] feature. This means that infinitives in these languages refer to an event that has not yet been realised. English bare verb forms, on the other hand, do not have any morphology. There is no potential carrier, hence they cannot be [–realised]. The bare verb form in English refers not only to the process part of an event; it also includes the completion of the event. This is shown by the contrast in (3): (3) a. *I see John cross the street b. I saw John cross the street

Following Giorgi and Pianesi (1997), H&H assume that the bare infinitive in English is [+perfective]. In this respect, the infinitive in Dutch/German is more like the English -ing form, which also refers to the process part of the event only: (4) I see John crossing the street

As infinitives in Dutch/German are inherently [–realised], H&H assume that the RIs of children learning these languages are [–realised]. By the same argument, i.e. the presence of an inherent feature, the RIs of English speaking children are [+perfective]. The modal interpretation of RI in Dutch and German follows from certain properties of volitive (or ‘bouletic’ as H&H call it) and deontic modality. The event expressed in a volitive or deontic modal utterance has a future temporal reference since obligation, permission (i.e. deontic modality) or wishes (i.e. volitive modality) have to precede the event itself. Because infinitives in Dutch and German share [–realised] with certain modalities, RIs are used by the chil-



Elma Blom

dren to express modality. H&H relate this modal meaning to the use of eventive verbs in RIs (Eventivity Constraint).6 In particular, deontic modality has a strong preference for eventive predication (while stative verbs yield an epistemic reading). Children under the age of four do not use epistemic modality, but they do use deontic modality very frequently. The overwhelming number of eventive RIs in child language neatly follows from this frequent usage. . Summary We have seen that the Null Auxiliary Hypothesis is successful in explaining various properties of RIs, such as the position of the verb and verbal morphology. Nevertheless, it runs into serious empirical problems. The strength of Ingram and Thompson’s proposal is that it focuses on an underexposed topic, namely the input children get. However, the scope of the ‘what they hear is what you get’ hypothesis is too limited to offer a satisfying solution. At first sight, H&H’s recent analysis has many advantages, because it ties non-finiteness, modality and eventivity together. In the next section their proposal will be discussed in more detail.

. A closer look at Hoekstra and Hyams (1998) H&H’s framework is based on a set of empirical observations and theoretical assumptions. In this section I single out three notions fundamental for H&H’s analysis. A discussion of these notions brings various problems to light. The relevant notions are, in order of appearance: the nature of [–realised], morphological marking of semantic features, and the connection between modality and verb type. . The nature of [–realised] A crucial notion in H&H’s analysis is the feature [–realised], proposed for Dutch and German infinitives. What [–realised] actually refers to is unclear, however, as no apparent definition is provided. It is not a feature which is embedded in a specific theoretical framework; there is no feature matrix to which [–realised] belongs. Theoretically, [–realised] can be interpreted in three ways: as an (i) aspectual, (ii) tense or (iii) modal feature. I discuss these three possibilities below.

On the use and interpretation of root infinitives in early child Dutch 

i. [–realised] as an aspectual feature The label seems to suggest this interpretation as the most obvious one. An aspectual reading is reinforced by the fact that [+perfective], which is a valid aspectual feature, functions as its opposite. [+perfective] includes the completion of the event while [–realised] excludes the completion. H&H also exclude a here-and-now interpretation of [–realised]: ‘In such languages [i.e. Dutch/German etc.], the unanchored RI does not receive the here-and-now interpretation of the English bare V [. . . ]’. Thus, when we try to translate H&H’s observation that ‘the infinitive denotes that the event is not yet realised’ into a known aspectual notion, this would result in inchoative aspect as the most appropriate candidate. ii. [–realised] as a tense feature Though [–realised] seems to be more aspectual than temporal, H&H do not exclude a temporal interpretation. Theoretically, there are two possibilities. When [–realised] refers to the non-completion of an event, it includes present tense as well as future tense. When it means that the event has not even started, it only refers to future tense. From the quotation above, it can be derived that H&H implicitly follow the second strategy: [+perfective] includes here-andnow events, and by implication [–realised] does not. iii. [–realised] as a modal feature The feature [–realised] restricts the number of possible modalities. Both deontic and volitive modality assume a temporal ordering: the permission, obligation or wish precedes the event for which one gets permission, for which one enters into an obligation or which is wished for. This has induced some linguists to not distinguish between future tense and modality. Chung and Timberlake (1985: 243) claim: ‘Future is thus not a semantic category where tense and mood merge. In practice, many languages do not distinguish morphologically between future tense and potential (irrealis) mood. Where a difference is made, the future tense is used for events that are presumed to be certain to occur, and the irrealis mood for events that are potentially possible but presumed to be certain.’ Example (5a), in which the infinitive refers to a future event, receives a deontic interpretation, while (5b), in which the predicate refers to an ongoing event, receives an epistemic interpretation: (5) a.

Hij moet afwassen ‘He must do the dishes’



Elma Blom

b. Hij moet aan het afwassen zijn He must on the the dishes do- be- ‘He must be doing the dishes’

As the event in deontic or volitive modal utterances still has to take place, or as H&H put it ‘is not yet realised’, the feature [–realised] delimits the modal interpretation to deontic and volitive modality. Epistemic modality is excluded. In this section we have seen that [–realised] can be described as an aspectual, tense or modal feature. H&H do not define their position and propose a feature that is neither one nor all. [–realised] is a purely empirically motivated7 feature that collapses inchoative aspect, future tense and deontic and volitive modality. . Morphological marking H&H connect [–realised] to an -en suffix, i.e. to morphology. Bare verb forms like the infinitive verb in English, which do not have any morphology, cannot be [–realised]. H&H follow Giorgi and Pianesi (1997) in assuming that English bare verb forms receive a perfective reading. This perfective reading yields a non-modal interpretation of RIs in child English. This gives rise to two questions. First, what morphology marks the [+perfective] feature? Second, as RIs in English cannot express modality, how do English-speaking children express modality? As to the question of what morphology marks [+perfective], H&H follow some sort of default strategy: unless morphology indicates otherwise, the infinitive is [+perfective]. In English, however, the element to could function as a modal marker. Or, in terms of H&H, to could function as a carrier of [–realised]. For example, Avrutin (1997) points to the use of to-infinitives for future events in headlines: (6) a. President to visit Russia b. Unions to go on strike c. McDonald’s to serve beer

Like the examples in (7), the ‘modal’ meaning of (6) may be a relict from the historical prepositional meaning of to: ‘direction towards, purpose or goal’ (Jespersen 1964). (7) a. To be or not to be b. To live and let die c. To look at you and never sleep

On the use and interpretation of root infinitives in early child Dutch

This brings us to the second question: how do English-speaking children express modality? The to-infinitive seems to possess the necessary properties to express modality. However, young English children do not yet use the toinfinitive. Forms that do occur from relatively early on – between the ages of two and three – are gonna, wanna, needta and hafta (Bloom, Tackeff and Lahey 1984; Gerhardt 1991). These forms express intentionality, volitive and deontic modality, which are exactly the same interpretations given to the RIs of Dutch children. Note that all these kinds of modality or modality-related categories are inherently [–realised]. In children’s speech, gonna, wanna, needta and hafta have in common that they all contain a ‘corrupted’ version of to spelled out as [6]. This suggests that in English child language the [6] suffix is like the -en suffix in Dutch and German, which is also pronounced as [6]: the carrier of [–realised]. This meaning is preserved in the adult English infinitival marker to. . Modality and verb type H&H observe a link between certain modalities and verb type. Young children only use deontic and volitive modality, i.e. semantic categories which Dutch and German children predominantly express with RIs. This pattern is formulated as the Modal Reference Effect (MRE). RIs in these languages are limited to eventive verbs, as summarised in the Eventivity Constraint (EC): ‘RIs are restricted to event-denoting predicates’. H&H show that the MRE and EC are closely connected: eventive verbs ‘force’ a deontic modal reading whereas stative verbs ‘force’ an epistemic reading. See the contrast in (8): (8) a.

Jan moet het antwoord weten Jan must the answer know- ‘John must know the answer’ b. Jan moet dit boek lezen Jan must this book read- ‘John must read this book’

The modal auxiliary moet in (8a) has a stative predicate; an epistemic reading is most natural (‘as far as I know, John has to know the answer’). In (8b) the same auxiliary has an eventive predicate. This yields a deontic interpetation (‘John is obliged to read this book’). H&H argue that the eventive character of RIs follows from the fact that children use deontic modality, not epistemic modality. At this point, some comments are in order. First, from the perspective of Dutch adult language, the correlation that H&H point out is not as clear as





Elma Blom

they present it. Depending on the context, a sentence like (8a) can receive not only an epistemic but also a deontic or volitive reading,8 as is shown by (9a–c): (9) a.

Het kan haast niet anders of Jan weet het antwoord It can hardly not different or John knows the answer ‘Based on what I know, it must be the case that John knows the answer’ b. Jan moet van de docent voor drie uur het antwoord weten John must of the teacher for three hour the answer know ‘The teacher obliged John to know the answer before three o’clock’ c. Jan moet en zal het antwoord weten John must and shall the answer know ‘John simply has to know the answer’

Barbiers (1995) observes that stative verbs do not force an epistemic reading. Furthermore, he argues that the relevant factor for a deontic or volitive reading is not eventivity, but the predicate’s allowance of a ‘polarity transition’ (p. 148). Secondly, H&H ignore the frequent use of volitive modality by children. Not only are the RIs of Dutch children very often volitive in meaning, but there is also the crucial fact that deontic and volitive modality cannot be collapsed. They differ from each other with respect to two structural properties: (i) volitive modality does not have the epistemic counterpart that deontic modality has, and (ii) volitive modality does not show a specific preference for eventive predication nor for stative predicates. Volitive modality in Dutch is prototypically expressed by the auxiliary willen (‘want to’). Barbiers (1995) points out that willen can have an epistemic reading, as exemplified by (10): (10) Er wil hier wel eens een ongeluk gebeuren There will here - sometimes an accident happen- ‘Accidents (may) happen here’

The sentence in (10), however, does not express epistemic modality. Epistemic modality involves the qualification of the potential truth of a proposition (Klinge 1998). Barbiers (1995) refers to this as probability. The sentence in (10), however, involves neither probability nor does it give a qualification of the potential truth of the proposition [accidents happen here]. It simply means that accidents sometimes happen here. In this sense, willen in (10) is not epistemic at all.9 Predication is the second argument for not treating volitive and deontic modality as one. Volitive modality does not show a preference for certain

On the use and interpretation of root infinitives in early child Dutch

predicates whereas deontic modality does. In (11a) the modal auxiliary wil selects the eventive complement trouwen, in (11b) wil selects the stative predicate getrouwd zijn. (11) a.

Hij wil trouwen He wants marry- ‘He wants to get married’ b. Hij wil getrouwd zijn He wants married be- ‘He wants to be married’

If we follow H&H’s analysis, (11b) should receive an epistemic reading. This prediction is not borne out because wil in both (11a) and (11b) refers to a wish or an internal need. As volitive modality can occur with both eventive and stative predicates, the kind of modality children use does not necessarily predict that their RIs have to be eventive. . Conclusion In this section I have discussed several aspects of H&H’s recent analysis of modal RIs. First, I have shown that [–realised] is a feature that can be translated into known aspectual, temporal and modal notions. Second, some implications of morphological marking have been explored. H&H assume that [+perfective] (I prefer to call it ‘here-and-now’) is the default reading that appears when there is no morphology. In English the bare verb form has this reading. To-infinitives, however, refer to events that are not yet realised. English to, reduced to [6] in child language, semantically functions as the -en suffix in Dutch and German: it is the infinitival marker and, as such, yields a [–realised] interpretation. Third, the relevance of the connection H&H establish between modality and verb type is discussed. Although such a relation exists, it does not provide a satisfactory solution for the Dutch and German child data. The main argument in support of this claim is that volitive and deontic modality have different properties with respect to the predicate they prefer. In general, H&H’s recent account inspires us to think about specific formfunction correlations in (child) language. Their framework raises interesting new questions, not only about language acquisition but also about languagespecific properties. These questions will be addressed in future research. In the second, empirical, part of this paper I will test the validity of the Modal Reference Effect for Dutch child language. Blom, Wijnen and Gillis (1998) found non-modal RIs in the language of two Dutch children. Behrens (1993) reports





Elma Blom

the same for German child language.10 Hoekstra and Hyams themselves use results from Ud Deen (1997) who found that 13% of the RIs of English- speaking children were modal. These modal RIs in English are problematical for Hoekstra and Hyams’ account as the infinitive in English should receive a [+perfective] interpretation. While modal RIs have been studied exhaustively, non-modal RIs have received less attention. My aim is to take a closer look at modal RIs including the role of non-modal RIs.

. The use of RIs in early child Dutch: A corpus study Let us consider two important empirical issues concerning modal RIs in Dutch child language. The first is the essential assumption underlying H&H’s proposal, which they formulate as the Modal Reference Effect (MRE). An analysis of corpus data from six Dutch-speaking children will show to what extent the RIs of Dutch children are modal. The second issue is concerned with the comparison of modal and non-modal RIs in Dutch child data. This comparison is guided by the hypothesis that non-modal RIs are categorised by the children as nominals and modal RIs as verbs. . Method Transcriptions of spontaneous speech data of six Dutch-speaking children will be used. All data are available via CHILDES (MacWhinney 1995). The tape recordings were made at home, in the presence of the mother or father as well as an investigator. In Laura’s case, the mother and the investigator are the same person. From each corpus, files have been selected representing the four stages which Dutch children go through in the acquisition of finiteness (Wijnen 1997): I. II. III. IV.

The first occurrences of RIs RI stage: almost no finite forms, only RIs Optional infinitive stage: RIs and finite forms co-occur ‘Adult’ finite stage: almost no RIs, only finite forms

The first (early two-word) stage was not available for every child. To begin with, the four stages have been collapsed, which has the advantage that there are enough data to analyse. Later, when developmental patterns are discussed, the data will be divided into the stages indicated above. The children’s age ranges are given in Table 1:

On the use and interpretation of root infinitives in early child Dutch

Table 1. Children’s age range and number of utterances Child

Age range

Number of utterances

Abel Daan Josse Laura Matthijs Peter

1;10.03 – 2;07.29 1;08.21 – 2;09.10 2;00.07 – 2;08.18 1;09.04 – 3;04.06 1;09.30 – 2;11.19 1;07.18 – 2;03.21

2890 4877 3323 4241 5260 2576

The selected files have been coded according to CHAT conventions.11 First, whether an utterance is modal or not is marked. As mentioned above, modality is the domain of possibility and necessity. Based on Palmer (1986), I divide modality into three subcategories: dynamic, deontic and epistemic modality. Dynamic modality implies a subject-internal necessity or possibility. This can be a wish, ability or capacity. The essence of deontic modality is a subjectexternal source that is responsible for the necessity or possibility. Obligation and permission are prototypically deontic modal notions. Epistemic modality refers to a speaker’s beliefs or judgements. This type of modality requires somewhat advanced cognitive skills and is therefore acquired after the age of four (Stephany 1986; Gonsalves 1998). (12) Modal → dynamic

→ → → deontic → → → epistemic →

wish (volition) ability, capacity obligation, requirement permission probability

Non-modal utterances are coded for aspect/tense. The event expressed can be completed, ongoing or prospective. (13) Non-modal → completed → ongoing → prospective

For the interpretation we made use of the information available in the transcription. For instance, utterances labelled ‘modal’ had to meet one of the following criteria: – –

preceding or subsequent parental utterances suggest a modal interpretation, or the contexts suggests a modal interpretation





Elma Blom

(14) a.

Peter bal pakken (Peter 2;01.27) Peter ball get- context: Peter wants to get the ball b. vrachtwagen emmer doen (Matthijs 2;04.24) truck basket do- context: Matthijs wants the investigator to put the truck in the basket c. op kist zitten (Josse 2;08.04) on box sit- context: Josse wants his mother to sit on the box

The label ‘non-modal’ was applied to utterances that refer to events in the hereand-now or the past, as in (15), for example: (15) a.

ah, mij bril vallen [= mijn bril valt] ah, my glasses fall- context: his glasses are falling b. poffie ginke [= koffie drinken]! coffee drink- context: people on television are drinking coffee c. boot svaje [= varen] boat sail- context: refers to a picture with a sailing boat

(Abel 2;05.27)

(Daan 2;01.21)

(Laura 2;04.01)

The above classification does not parallel H&H’s [–realised] vs. [+perfective] distinction. Based on H&H’s study, I have chosen to do the following: (16) Modal = [–realised] = modal + non-modal prospective (‘intentional’) Non-modal = [+perfective] = non-modal completed + non-modal ongoing

Non-modal prospective RIs are often intentional. I categorised these as nonmodal because formally they do not express necessity or possibility. The borderline between dynamic necessity (volition) and intentionality, however, can be very narrow. . Modal infinitives For German child language Ingram and Thompson (1996) have shown that there is a contrast between RIs and simple finite utterances with respect to modality: RIs are predominantly modal whereas simple finite utterances are used for non-modal purposes. Blom et al. (1998) have arrived at the same conclusion for Dutch. Using a more extensive database, containing six children

On the use and interpretation of root infinitives in early child Dutch

instead of two, the connection between modality and infinitives in Dutch child language will now be ‘rechecked’. To find out whether there is a connection between modality and infinitive verb forms, I have made two comparisons: 1. A comparison between RIs and sentences with simple lexical verbs shows if modality is can be connected to infinitival morphology. The prediction is that RIs occur in modal contexts, whereas simple finite verbs occur in non-modal contexts. Note that only finite lexical verbs are relevant here as modal verbs and copula do not occur in RIs (de Haan 1987; Jordens 1990; Wijnen 1997). 2. RIs as well as simple finite sentences are compared to a control group, which consists of sentences without a verb. The following scenario makes it clear that such a control group is necessary. Assume that the first prediction is borne out and RIs are modal as opposed to simple finite sentences. Now suppose that utterances without a verb pattern like RIs and are predominantly modal too. In this case, there may be a connection between finite morphology and non-modality, but there is no reason to assume a connection between modality and infinitival morphology on the verb. .. Results Table 2 gives the numbers and percentages of modal/non-modal RIs, whereas Table 3 gives the numbers and percentages of modal/non-modal simple finite utterances with lexical verbs. Non-interpretable utterances are excluded from the analysis. The tables show a reverse pattern: all children use RIs most frequently in modal contexts, whereas simple finite sentences are predominantly used in non-modal contexts.

Table 2. Modality in root infinitives Child Abel Daan Josse Laura Matthijs Peter

Number of RIs 118 160 192 306 252 203

Modal #

%

Non-modal # %

91 115 150 202 201 157

77 72 78 66 80 77

27 45 42 104 51 46

23 28 22 34 20 23





Elma Blom

Table 3. Modality in Simple Finite Sentences (SFS) with lexical verbs Child

Number of SFS

Abel Daan Josse Laura Matthijs Peter

129 289 141 160 150 439

Modal #

%

Non-modal # %

16 37 14 16 25 10

12 13 10 10 17 2

113 252 127 144 125 429

88 87 90 90 83 98

Table 4. Calculated χ2 (Modality in RIs/Utterances with simple finite lexical verbs) Child

χ2

Abel Daan Josse Laura Matthijs Peter

105.12 160.48 151.27 132.40 152.09 406.37

A statistical analysis confirms the patterns observed: for each child the calculated value of χ2 is greater than the 3.84 as well as 6.64 (which are, respectively, the critical values required for significance at the 5 and the 1 percent significance level for degrees of freedom = 1). See Table 4. Examples of modal and non-modal RIs have already been given in (14) and (15). The sentences in (17) and (18) give examples of modal and non-modal use of simple finite sentences. Note that a simple finite sentence with a modal reference often contains a verb like passen ‘to fit’ or gaan ‘to go’. Although they do not belong to the class of modal verbs, these verbs may be inherently modal as passen can express (im)possibility and gaan a future event: (17) a.

past niet ijsbeer in niet fits not polarbear in not ‘the polarbear does not fit in here’ b. ik ga ook naar de dok, of niet? I go also to the doctor, or not ‘I will go to the doctor too, or won’t I?’

(18) a.

ik hoor paatje niet I hear horse- not ‘I do not hear the little horse’

(Matthijs 2; 04.24)

(Laura 3; 04.06)

(Laura 2; 04.15)

On the use and interpretation of root infinitives in early child Dutch

b. Daan ligt in de wieg Daan lies in the crib ‘Daan is lying in the crib’

(Daan 2; 04.14)

The results and analysis in the Tables 2, 3 and 4 show that in order to express modality children strongly prefer using infinitive main verbs to simple finite lexical verbs. Table 5, which gives the usages of utterances without a verb, also shows a significant difference between RIs and verbless utterances in this respect. This confirms the connection between modality and RIs. Table 6 gives the relevant χ2 values showing this. Table 5. Modality in utterances without verbs Child Abel Daan Josse Laura Matthijs Peter

Number of utterances 275 557 635 484 298 210

Modal #

%

Non-modal # %

32 9 158 80 73 97

12 2 25 17 24 46

243 548 477 404 225 113

88 98 75 83 76 54

Table 6. Calculated χ2 (Modality in RIs/Utterances without verbs) Child

χ2

Abel Daan Josse Laura Matthijs Peter

164.66 428.97 178.81 199.99 166.81 42.30

All values are statistically significant, which implies that for all children it is the case that RIs are connected to modality as opposed to utterances without a verb. Note, however, that there is quite some variance between the children. Most striking is Peter’s use of utterances without a verb: the modal/non-modal distribution is almost at a 50% level. In his case it cannot be concluded that modality is exclusively associated with infinitive main verbs. The sentences in (19) give examples of the modal use of utterances without a verb, in (20) nonmodal verbless sentences are exemplified:





Elma Blom

(19) a.

mama schoen uit mama shoe off ‘mama has to take off her shoe’ b. ikke banaan I banana ‘I want a banana’

(20) a.

Bart ook thuis Bart also home ‘Bart is also at home’ b. ik ook broertje zusje I too brother- sister- ‘I get a little brother or sister too’

(Peter 1; 09.20)

(Josse 2; 03.28)

(Abel 2; 01.16)

(Josse 2; 03.28)

The above results support H&H’s Modal Reference Effect: Dutch children do prefer to use RIs to express modality. However, if infinitives in Dutch contain a [–realised] feature, which is the explanation for the MRE according to H&H, a percentage of 20–34 non-modal RIs is undesirable. In H&H’s framework, non-modal RIs should be excluded as all infinitives in Dutch show infinitival morphology in principle. . Change of Category One of the findings of the previous section is that Dutch children allow RIs in non-modal contexts. At first sight, this seems incompatible with the assumption that infinitives in Dutch are [–realised]. According to H&H’s definitions, as well as the definitions I used, [–realised] and [non-modal] exclude each other. This problem may be solved if modal and non-modal RIs have a different representation. My claim is that non-modal RIs are [+N], whereas modal RIs are [+V]. This is stated in the Change of Category Hypothesis: Change of Category Hypothesis Non-modal RIs are nominal, whereas modal RIs are categorised by the child as verbs. Non-modal RIs are like the verbless utterances of the previous section: like the non-modal RIs they can be analysed as non-verbal too. Note that for now the [+N] feature is rather arbitrary; instead, it could be [–V] or [+referential]. Importantly, I assume a basic distinction between these and the [+V] infinitives of Dutch children. The nominal RIs are deictic expressions. They are used to label an action, which is present in perceptible physical context.12 As deictic

On the use and interpretation of root infinitives in early child Dutch

RIs are labels of ongoing actions, it is expected that they appear as single elements, i.e. as a one-word utterance. Assuming that they are [+N], this could also be made to follow from their nominal character as we do not expect there to be an internal or external argument. Deixis plays a major role in early language use, therefore a developmental effect is predicted here: the number of deictic RIs will decrease over time. In contrast to nominal RIs, verbal RIs are not deictic expressions, which means that they are not used and interpreted as such. Verbal RIs reflect properties of the grammar: the infinitive in these utterances has infinitival morphology, which carries a [–realised] feature. These utterances can be assigned an aspectual/temporal reference by means of grammar. A VP is projected and argument structure may lead to utterances that are more complex. .. Results The predictions are summarised below: 1. Non-modal RIs refer to a here-and-now situation and a past reference is excluded. 2. Non-modal RIs are more frequently one-word utterances than modal RIs. 3. The number of non-modal RIs decreases over time. In Table 7 the interpretable RIs are divided into two aspectual categories: completed and ongoing. Cases in which it is impossible to distinguish between an infinitive and participle without prefix are left out. Examples of this are vallen ‘to fall’ or opeten ‘to eat’, with the particle forms gevallen and opgegeten. This is necessary, as children often omit the participial prefix (Jordens 1990). Examples of completed events are provided in (21), and (22) shows RIs that are used to refer to ongoing events:

Table 7. The interpretation of non-modal RIs Child Abel Daan Josse Laura Matthijs Peter

Number of RIs 27 45 42 104 51 46

Completed # %

Ongoing #

%

2 0 7 6 4 6

25 45 35 98 47 40

93 100 83 94 92 87

7 0 17 6 8 13





Elma Blom

(21) a.

Ieke spugen Ieke throw up- ‘Ieke has thrown up’ b. Peter koffer pakken Peter suitcase get- ‘Peter got his suitcase’

(22) a.

Ik een hand tekenen I a hand draw- ‘I am drawing a hand’ b. Pake foto maken Pake picture make- ‘Pake is taking a picture’

(Matthijs 1; 11.24)

(Peter 2; 00.28)

(Abel 2; 07.15)

(Daan 2; 04.14)

The first prediction is that completed RIs, or RIs with a past tense reference, do not occur. Though a small number of completed RIs do appear, nearly all (between 83–100%) non-modal RIs refer to ongoing, that is here-and-now, events. I note the following comments about the very few completed RIs that do occur. The cases in Table 7 in which the number of completed RIs are relatively high contain examples of repetitive use of RIs: the completed RIs of Josse are without exception formed with the verb hoesten ‘to cough’; Laura uses zwemmen ‘to swim’ twice, spugen ‘to throw up’ occurs twice in Matthijs’ completed RIs and Peter uses pakken ‘to get’ three times as a completed RI. If the repetitions are excluded from the analysis, there are hardly any completed RIs left. The second prediction is that non-modal RIs occur more frequently as oneword utterances than modal RIs do. All children except Peter show this pattern: 30–78% of the non-modal RIs represent one-word utterances whereas 16–43% of the modal RIs contain only one word. Figure 1 shows the distribution per child: per child the percentage of one-word non-modal and modal RIs relative to the total number of non-modal and modal RIs is given. This suggests that there may be a connection between being non-modal and appearing as a oneword utterance. Peter is an exception as he is the only child who does not show this difference: non-modal RIs are more often one-word utterances than modal RIs. However, this could be due to his repetitive volitive (i.e. modal) use of verbs like stappen ‘to walk’, and koken ‘to cook’. The third prediction focuses on longitudinal development. As modal RIs are supposed to occur as an effect of the less prominent use of deixis, which reveals grammatical properties,13 it is predicted that their number increases in time. The number of non-modal RIs, on the other hand, should decrease. As described in Paragraph 5.1, the data in Table 2 represent four stages. A

100 90 80 70 60 50 40 30 20 10 0

Pe ter

M att hi js

La ur a

Jo sse

Nonmodal RIs Modal RIs

Da an

Ab el

Percent

On the use and interpretation of root infinitives in early child Dutch

Child

Figure 1. Non-modal/modal distribution of one word RIs

reanalysis of the data in which the stages are taken into account shows that the proportion of modal and non-modal RIs varies over time. Strikingly, the connection between modality and RIs does not appear until stage three (when the number of RIs and finite constructions is equal, the ‘optional infinitive’ stage14 ). It is not until this stage that there is a significant difference between the number of non-modal and modal RIs. This holds for all children, except Peter again, whose data I will discuss later on. Behrens (1993) found the same pattern in German child language. In the early stage (age 1;9) there is no preference with respect to the temporal reference of RIs: 28% can be interpreted as ongoing and 28% as non-ongoing. In the last stage (2;7) there is a clear difference: 11% is ongoing whereas 85% is non-ongoing. Peter’s graph suggests a decrease of modal RIs, while the relative number of non-modal RIs seems to rise. This pattern, however, may be deceptive. A closer look at the data brings two relevant issues to light. As mentioned before, Peter tends to use certain lemmas repetitively (mostly in volitive contexts). He does this extremely frequently in the file representing the first stage. In this file stappen (‘to walk’) appears 18 times and koken (‘to cook’) appears 6 times, both only to express that Peter wants to walk or cook. As the total number of RIs in this file is 26, it seems reasonable to assume that the data are not widespread enough to draw any conclusions. Also, one of files representing the fourth stage contains nearly 50% (9 out of 19 RIs) highly elliptic RIs. These appear when Peter and his mother are reading a book. Every time they encounter a new picture, his mother asks Peter what’s X doing? Peter answers with a bare in-



Elma Blom 3. Daan's modal RIs

100 80 60 40 20 0

Percent

Percent

2. Abel's modal RIs

1

2

3

100 80 60 40 20 0 1

4

2

Stage

Percent

Percent 2

3

100 80 60 40 20 0

4

1

2

Stage

4

7. Peter's modal RIs

Percent

100 80 60 40 20 0 2

3 Stage

6. Matthijs' modal RIs

1

4

5. Laura's modal RIs

100 80 60 40 20 0 1

3 Stage

4. Josse's modal RIs

Percent



3 Stage

4

100 80 60 40 20 0 1

2

3

4

Stage

Figures 2–7. Developing modal RIs

finitive (which is, in fact, exactly the same as adult speakers of Dutch would do). Although sessions in which the caregiver and the child are reading a book together are frequent, this is the only file in which such highly elliptic RIs are disturbing the results. Figure 8 is corrected for this. In conclusion to this section I must point out that the second and third prediction may be correlated. The number of non-modal RIs decreases as an effect of development. Modal RIs are more frequent in a later developmental

Percent

On the use and interpretation of root infinitives in early child Dutch 100 80 60 40 20 0 1

2

3

4

Stage

Figures 8. Peter’s developing modal RIs (corrected version)

stage. So, the fact that modal RIs are more complex than non-modal RIs could be an effect of development. This has to be resolved in future research. . Summary and discussion In this section two claims have been tested empirically. First, the connection between modality and infinitive verb forms is checked by: 1. comparing utterances with verbs to utterances without a verb and 2. by comparing utterances with simple finite lexical verbs to RIs. Although children use verbless utterances in modal contexts and RIs in non-modal contexts, the two categories show a reverse pattern. Children tend to use utterances that lack a verb in non-modal contexts and RIs, as opposed to utterances with a simple finite verb form, in modal contexts. In general, these findings support H&H’s Modal Reference Effect, though non-modal use of RIs is not excluded. The second claim, formulated as the Change of Category Hypothesis, focuses on a basic distinction between modal RIs and non-modal RIs. The underlying assumption is that non-modal RIs are [+N] whereas modal RIs are categorised as [+V]. The first prediction, namely that non-modal RIs are restricted to here-and-now events, is borne out. The second prediction was that non-modal RIs predominantly occur as one-word utterances while modal RIs consist of more than one word. The results show a significant difference between modal and non-modal RIs, which could be interpreted as a reflection of a different representation of the two types of RIs. However, it could also be an effect of development. Moreover, it is a rather imprecise result: sometimes non-modal RIs contain more than one element and a more detailed study is required to establish whether there is a difference in context in which modal and non-modal infinitives are used. The categorial labels [+V] and [+N] may lead to rather specific predictions in this respect. Development brings us to the third prediction. With respect to the connection between modality and non-





Elma Blom

finite morphology, an analysis based on developmental stages leads to an interesting finding: the relation between modality and non-finite verbs does not exist ‘from the beginning’. Not until the optional infinitive stage do children use RIs significantly more frequently in modal than in non-modal contexts. If we follow H&H’s framework and assume that infinitives in Dutch are inherently [–realised], it must be explained why this feature is not always present or does not always receive an interpretation. In doing so, I return to the Change of Category Hypothesis. As the developmental graphs have shown, the OI stage plays a crucial role for the identification of infinitives as verbs. In the OI stage children acquire finiteness. In Dutch, finiteness is a specific property of verbs. This could mean that children have to learn finiteness before being able to recognise infinitives as verbs. Why should Dutch children initially categorise infinitives as nouns? This is an important question as an explanation based on reanalysis by children points to a position in the continuity debate: there are stages in which the child grammar differs from the adult grammar i.e. language acquisition is a discontinuous process. Discontinuity can easily occur if it concerns the categorisation of infinitives in Dutch. The ‘verbness’ of infinitives is not as straightforward as often is assumed. Infinitives in Dutch have only one unambiguous property that reveals their verbal character: they have finite (i.e. tense-marked/agreeing) counterparts. Morphology, for example, is not a sufficient cue, as the nominal plural suffix in Dutch could be -en like in boek– boeken (‘book– books’). Syntax is rather confusing too, as Dutch allows a construction in which the infinite verb is preceded by a definite determiner, as in (23): (23) Wat ben je aan het doen? Auto’s vernielen What are you  - do-? Cars smash- up ‘What are you doing? Smashing up cars’

Not only can infinitives in Dutch follow a determiner, they can also become nominal without changing appearance. This is shown in (24): (24) Baden in karnemelk deed Suzanna bij voorkeur in het weekend Bathe- in buttermilk did Suzanna by preference in the weekend Bathing in buttermilk was something Suzanna preferred to do in the weekend’

In Dutch finiteness is a property of verbs. As soon as children realise that the non-finite and finite form of a verb involve the same lexical item (formally distinguished by morphology), infinitives will be categorised as verbs. This happens in the OI stage. At this point, I want to turn to H&H’s hypothesis.

On the use and interpretation of root infinitives in early child Dutch 

H&H assume that infinitival morphology marks the [–realised] feature. This assumption implies that children recognise stem and suffix, because if there is no suffix, this means that there is nothing to connect [–realised] to. This situation describes the stage before the OI stage. Before the OI stage, Dutch children do not identify infinitival morphology as such. Therefore, there is no suffix that could mark [–realised]. This accounts for the finding that the association between modality and the infinite main verb does not appear from early on, but arises in the OI stage.

. Concluding remarks This article started with an overview of current theories that explain modal RIs in Dutch and German child language. Hoekstra and Hyams’ (1998) recent proposal was singled out and discussed in more detail. This led to many new questions about child language, form-function correlations and cross-linguistic differences. The second part focused on the empirical base of the theories which account for modal RIs. First of all, it was shown that the connection between modality and infinitive verb forms does exist, although there are also a relatively large number of non-modal RIs. In H&H’s view non-modal use of RIs should be excluded. If, however, modal and non-modal RIs are analysed differently, this problem may be solved. I proposed that modal RIs are [+V] whereas non-modal RIs, since they are deictic expressions, are [+N]. Data of six Dutch children seem to support this proposal. Though maybe not immediately obvious, the claim that infinitives are initially analysed by children as nouns seems feasible. In Dutch, morphology is the only unambiguous cue for children to classify infinitives as [+V]. My claim is that children have to be able to alternate between finite and non-finite morphology to recognise infinitival morphology. This ability is acquired in the OI stage. Strikingly, in the OI stage RIs become modal. If it is true that children do not recognise infinitival morphology before the OI stage, this finding supports H&H’s claim that [–realised] is connected to infinitival morphology.

Notes * I would like to thank Frank Wijnen, Nina Hyams, Evelien Krikhaar, Ingeborg Lasser, Brigit van der Pas, two anonymous reviewers and the editors for their careful reading of earlier versions of this paper. The research reported in this paper was supported by grants of the



Elma Blom

Dutch-Flemish Co-operative Programme on Language and Culture (VNC 200-41.031 and G.2201.96), sponsored by NFWO and the Netherlands Organisation for Scientific Research (NWO). . By using the term ‘exclamative’ for this type of RIs, I follow Grohman and Extepare (to appear). Taken as a whole, exclamative RIs are not (rhetorical) questions but rather descriptions of a ‘hypothetical event’ (Akmajian 1984). It is crucial is that exclamatives always consist of two parts: the first part describes the event (in the form of a rhetorical question) while the second part expresses that the truth of this event is very improbable. Lasser (1997: 40) states that ‘they [i.e. exclamatives] contain the specific presupposition that, from the point of view of the speaker, the proposition expressed is false or at least debatable’. . This observation is supported by Lasser’s (1997) findings for German child and adult RIs. Lasser divides her data into declaratives, interrogatives and imperatives. One of the differences she found between German adult and child RIs was that adults use RIs most frequently as imperatives while children use them most often as declaratives. Though declaratives and assertions do not refer to the same, the notions are closely related. As Clark (1996: 136) puts it: ‘To assert something, you choose a declarative’. . Opinions differ with respect to the treatment of volition as modality or as a modalityrelated category. For example, Givón (1984), Chung and Timberlake (1985), Palmer (1986) and Bybee and Fleischman (1995) include volition in their use of the term modality, while De Haan (1987) and Anderson (1986) exclude volition from it. . The two paraphrases, as well as the distinction between volitive and deontic modality, may suggest two different ‘core’ meanings here. However, I do not want to make this claim here. As described by Lyons (1997: 826), both child usages can have one source: ‘The origin of deontic modality, it has often been suggested, is to be sought in the desiderative and instrumental function of language: that is to say, in the use of language, on the one hand, to express or indicate wants and desires and, on the other, to get things done by imposing one’s will on other agents. It seems clear that these two functions are ontogenetically basic, in the sense that they are associated with language from the earliest stage of its development in the child. It is equally clear that they are very closely connected. It is a small step from a desiderative utterance meaning ‘I want the book’ to an instrumental utterance meaning ‘Give me the book’; and parents will commonly interpret the child’s early desiderative utterances as mands, thereby reinforcing, if not actually creating, the child’s developing awareness that he can use language in order to satisfy his wants and desires’. . Ferdinand (1996) analyses non-finite utterances in French child language along the same lines, though she explicitly assumes a covert modal or aspectual auxiliary. This modal or aspectual auxiliary accounts for the use of eventive verbs in these non-finite utterances. Ferdinand’s observation for French that certain modalities co-occur with eventive infinitives reappears as Hoekstra and Hyams (1998) explain the meaning of RIs in Dutch and German child language (see Section 2.3). . Note that Ferdinand (1996) makes this connection too. . It is motivated by its compact and efficient description of the predominant meaning of RIs in child Dutch and German. . Sjef Barbiers pointed this out to me.

On the use and interpretation of root infinitives in early child Dutch  . Among some generative linguists, however, a less strict definition of epistemicity is used. Here, modal verbs can have epistemic and ‘root’ meanings which differ structurally. Ross (1969), for example, claims that epistemic modals are one-place predicates that predicate over the whole sentence. It could be argued that this is what we see in (10). By contrast, the root modal willen in (i) is a two-place predicate expressing a wish of the subject. (i)

Marie wil een appel eten Mary wants to eat an apple

. Hoekstra and Hyams themselves use results from Ud Deen (1997) who found that 13% of the RIs of English-speaking children were modal. These modal RIs in English are problematical for Hoekstra and Hyams’ account as the infinitive in English should receive a [+perfective] interpretation. . I am grateful to Yvonne Steeneveld, who did most of the coding. . The proposed strategy as applied by the children may be comparable to Roeper and Pérez-Leloux’ (1997: 129) proposal: ‘Another possibility is that the child chooses a radically different structure and the correct interpretation is obtained pragmatically. For instance, a nominal compound is selected, and there is no grammatically marked temporal reference. For instance, if on walked into a nursery room and one said “Ah finger-painting”, one might refer to an ongoing activity, but the utterance itself contained no tense-marker. In principle, this says that if the grammar cannot generate a more explicit output which refers to Tense and is consistent with economy, then the child may choose an altogether different structure and rely on contextual inference to convey the remaining information.’ . See Blom and Wijnen (2000) for more discussion about the question whether [–realised] is an inherent or an acquired feature. . I am using this term only to refer to a certain point in the development in which the amount of infinite and finite constructions is approximately equal. There are no theoretical implications.

References Akmajian, A. (1984). Sentence types and the form-function fit. Natural Language and Linguistic Theory, 2, 1–23. Anderson, L. (1986). Evidentials, paths of change, and mental maps: Typology regular meets asymmetries. In W. Chafe and J. Nichols (Eds.), Evidentiality: The Linguistic Coding of Epistemology (pp. 227–274). Norwood: Ablex. Avrutin, S. (1997). EVENTS as units of discourse representation in root infinitives. MIT Occasional Papers in Linguistics, 12, 65–91. Barbiers, S. (1995). The Syntax of Interpretation. Doctoral dissertation, Leiden University. Behrens, H. (1993). Temporal Reference in German Child Language. Doctoral dissertation, University of Amsterdam.



Elma Blom

Blom, E. (2000). On the meaning of nonfiniteness in child and adult Dutch. Paper presented at the Workshop on Finiteness held at the Annual Conference of the Deutsche Gesellschaft für Sprachwissenschaft, Marburg March 1–3, 2000. Blom, E. and F. Wijnen (2000). How Dutch children’s root infinitives become modal. In S. Howell, S. Fish and T. Keith-Lucas (Eds.), Proceedings Boston University Conference on Language Development 24 (pp. 128–139). Blom, E., F. Wijnen, and S. Gillis. (1998). Modal infinitives in the speech of Dutch mothers and their children. In J. van der Auwera, F. Durieux and L. Lejeune (Eds.), English as a Human Language: To honour Louis Goossens (pp. 12–21). München: LINCOM Europa. Bloom, L., J. Tackeff, and M. Lahey (1984). Learning to in complement constructions. Journal of Child Language, 11, 391–406. Boser, K., B. Lust, L. Santelmann, and J. Whitman (1992). The syntax of CP and V2 in early child German (ECG): The strong continuity hypothesis. In K. Broderick (Ed.), Proceedings of NELS 23 (pp. 51–65). Amherst (Mass.): GSLA. Bybee, J. and S. Fleischman (1995). Modality in grammar and discourse: An introductory essay. In J. Bybee and S. Fleischman (Eds.), Modality in Grammar and Discourse (pp. 1– 14). Amsterdam: Benjamins. Chung, S. and A. Timberlake (1985). Tense, aspect and mood. In T. Shopen (Ed.), Language Typology and Syntactic Description, Vol. 3: Grammatical Categories and the Lexicon (pp. 202–258). Cambridge: Cambridge University Press. Clark, H. (1996). Using Language. Cambridge: Cambridge University Press. Crisma, P. (1992). On the acquisition of Wh-questions in French. In M. Starke (Ed.), GenGenP: Geneva Generative Papers (pp. 115–122). Geneva: Département de Linguistique Générale. Gerhardt, J. (1991). The meaning and use of the modals HAFTA, NEEDTA and WANNA in children’s speech. Journal of Pragmatics, 16, 531–590. Ginneken, J. van (1917). De Roman van een Kleuter. ’s Hertogenbosch/Antwerpen: L.C.G. Malmberg. Giorgi, A. and F. Pianesi (1997). Tense and Aspect: From Semantics to Morphosyntax. New York: Oxford University Press. Givón, T. (1984). Syntax: A Functional-typological Introduction. Vol. 1. Amsterdam: Benjamins. Gonsalves, J. (1998). Relations between Conceptual and Semantic Development: Preschoolers’ Understanding of Modality across Linguistic and Nonlinguistic domains. Doctoral dissertation, Clark University. Grohmann, K. and R. Extepare (to appear). Root infinitives: a comparative view. Ms, University of Maryland. Ferdinand, A. (1996). The Acquisition of the Subject in French. Doctoral dissertation, University of Leiden. Haan, F. de (1997). The Interaction of Modality and Negation: A Typological Study. New York: Garland. Haan, G. de (1987). A theory-bound approach to the acquisition of verb placement. In G. de Haan and W. Zonneveld (Eds.), Formal Parameters of Generative Grammar, Yearbook 1987 (pp. 15–30). Dordrecht: ICG.

On the use and interpretation of root infinitives in early child Dutch

Haegeman, L. (1994). Root infinitives, tense and truncated structures. Language Acquisition, 4, 205–255. Haegeman, L. (1995). Root null subjects and root infinitives in early Dutch. In C. Koster and F. Wijnen (Eds.), Proceedings of GALA 1995 (pp. 239–250). Groningen: Center for Language and Cognition. Hoekstra, T. and N. Hyams (1998). Aspects of root infinitives. Lingua, 106, 81–112. Ingram, D. and W. Thompson (1996). Early syntactic acquisition in German: Evidence for the modal hypothesis. Language, 72, 97–120. Jespersen, O. (1964). Essentials of English Grammar. Montgomery, Alabama: University of Alabama Press. Jordens, P. (1990). The acquistion of verb placement in Dutch and German. Linguistics, 28, 1407–1448. Klinge, A. (1998). Modal underspecification and problems of metalanguage. Paper presented at the Antwerp conference on Modal Verbs in Germanic and Romance Languages, 10–12 December 1998. Krämer, I. (1993). The licensing of subjects in early child language. MIT Working Papers in Linguistics, 19, 197–212. Kursawe, C. (1994). Fragesätze in der deutsche Kindersprache. MA dissertation, University of Düsseldorf. Lasser, I. (1997). Finiteness in Adult and Child German. Doctoral dissertation, University of Nijmegen. Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press. MacWhinney, B. (1995). The CHILDES Project: Tools for Analyzing Talk. (2nd edn). Hillsdale (NJ): Lawrence Erlbaum Associates. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press. Phillips, C. (1995). Syntax at age 2: Crosslinguistic differences. MIT Working Papers in Linguistics 26, 325–382. Poeppel, D. and K. Wexler (1993). The full competence hypothesis of clause structure in early German. Language, 69, 1–33. Roeper, T. and A. Pérez-Leloux (1997). The interpretation of bare nouns in semantics and syntax: Inherent possessors, pied-piping, and root infinitives. MIT Occasional Papers in Linguistics, 12, 114–133. Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in Philosophical Linguistics series 1 (pp. 77–102). Evanston: Great Expectations. Santelmann, L. (1995). The Acquisition of Verb Second in Child Swedish. Doctoral dissertation, Cornell University. Stephany, U. (1986). Modality. In P. Fletcher and M. Garman (Eds.), Language Acquisition (pp. 375–400). Cambridge: Cambridge University Press. Ud Deen, K. (1997). The interpretation of root infinitives in English: Is eventivity a factor? Ms, UCLA. Weverink, M. (1989). The subject in relation to inflection in child language. MA dissertation, University of Utrecht. Wijnen, F. (1997). Temporal reference and eventivity in root infinitives. MIT Occasional Papers in Linguistics, 12, 1–25.



Modals and negation in English* Annabel Cormack and Neil Smith University College London

We show in this paper than the English modals fall into two distinct classes on the basis of their semantic scope interpretation in relation to negation. Against a background of an austere minimalism, we argue that syntactically and semantically, there are three distinct negations available in English: Echo[NEG], Pol[NEG] and Adv[NEG]. These are associated with Metalinguistic negation, sentential negation, and VP negation respectively. The modals are uniformly interpreted as having scope lower than Echo[NEG], and higher than Adv[NEG], but there is systematic variation with respect to Pol[NEG]. We demonstrate the explanatory value of these hypotheses in English, with excursions into some other languages. Finally, we discuss and reinterpret as cognitive rather than syntactic, Cinque’s claims about the relative ordering of epistemic and deontic modals.

.

Background

We are concerned in this paper with the scope of modal auxiliaries with respect to negation. Although it would be possible to treat scope from a purely semantic point of view, we hope to demonstrate here that an approach which integrates syntactic, semantic, and pragmatic theories is more illuminating. In treatments of generative grammar based on the Principles and Parameters model, including the Minimalist model (Chomsky 1995), scope is read off or perhaps pragmatically derived from the level of LF. (1)



Annabel Cormack and Neil Smith

In the Minimalist approach, the pair {LF, PF}, giving the inputs to the processing by the auditory/articulatory and the conceptual/intentional processing systems, is generated as shown in (1) above. However, we have argued that at least with respect to heads, the system above is unnecessarily complicated. We propose that there is no movement of heads (and hence no Move on the diagram above for heads). Rather the LF-interpretable part of a head is merged in its proper scope position with respect to other heads, and the PFinterpretable part of a head is merged where it is heard relative to other heads. This   hypothesis is discussed in Cormack and (Smith 1997, 1998, 1999 and 2000). In this paper, then, we assume that the scope position for a head is the same as the merge position of the LF-interpretable part of that head, which we show in terms of a fairly standard syntactic representation. Morphological considerations, and other principles of UG, which we are not going to discuss here, determine where the PF-interpretable part of the head is merged – typically, in its LF-interpretable position.1 For most of this paper, the positions of the PFparts can be understood as derived by semantically vacuous head-movement. However, Sections 4 and 7 discuss cases not amenable to this interpretation. Readers rejecting these views of grammar may take the pieces of structure given in small capitals (e.g. as in (2) onwards) as independent representations of semantic scope. We will be concerned throughout with the modal auxiliaries of English, shall, should, must, can, could, may, might, ought + to, has + to, is + to, dare, need, will, and would, which we refer to simply as ‘modals’. What our Split Sign hypothesis gives us is the following. If in a particular interpretation of a sentence, we find that a modal Mx has scope over some item Y, then Mx is merged in a position c-commanding Y – and this is coded in the grammar by the usual requirement that Mx selects for Y(P), or that Y is an optional adjunct to the projection that Mx selects for. We rule out the possibility that although merged below Y, Mx has scope over Y because it has been moved to c-command Y at LF. Our assumption considerably simplifies the task of the child language learner, and the linguist, and turns out to have interesting consequences with respect to modals and negation.2 The issue of the syntactic scope of modals with respect to negation was raised by Picallo’s (1990) work on Catalan. She demonstrates that in Catalan, sentences such as (2) and (3) are monoclausal, and that some modals are merged below negation, and some above. If negation is present, modals with an epistemic interpretation are never within its scope, but those having a deontic (‘root’) interpretation always are. The examples in (2) and (3) are Picallo’s

Modals and negation in English

(1990: 287, (4) and (5)), and use the modal poder ‘may/can’.3 The scope/LF annotation, where small capitals represent the LF-interpretable part of a head, is ours. (2) En Jordi pot no haver sortit may not have left ‘It is possible that Jordi hasn’t left yet’

epistemic:  [

(3) En Jordi no ha pogut sortir not has could leave ‘Jordi hasn’t been able to leave’

deontic:  [

Here, and throughout, we use ‘epistemic’ as a cover term for all kinds of alethic and epistemic uses, including those based on logical or (folk-)scientific theories, and ‘deontic’ to cover all sorts of obligation, and also disposition, willingness, intention, and ability. The claim implicit in this is that if a distinction of this kind is made between two classes of modal meanings, then all natural languages draw the distinction in the way indicated. This claim may of course turn out to be false, but if it is true, it will simplify the task of the language learner (see Section 5). Consider by contrast the English modals in (4) to (7). Because negation in echoic contexts may give rise to a wider range of scope interpretations (see Section 6 below), we restrict ourselves for the moment to interpretations in non-echoic contexts. (4) Alfred shouldn’t eat nuts (deontic) ‘It is advisable for Alfred not to eat nuts’  [ *‘It is not the case that it is advisable for Alfred to eat nuts’ *NOT [SHOULD (5) Bob shouldn’t be late ‘It is predictable that Bob will not be late’ *‘It is not the case that it is predictable that Bob will be late’ (6) Edwin can’t climb trees *‘Edwin is able not to climb trees’ ‘It is not the case that Edwin can climb trees’ (7) Jean can’t have left *‘It is possible that Jean has not left’ ‘It is not the case that it is possible for Jean to have left’

(epistemic)  [ *NOT [SHOULD (deontic) *CAN [NOT  [ (epistemic) *CAN [NOT  [





Annabel Cormack and Neil Smith

Examples (4) and (5) with should are unambiguous: the modal should has scope over negation. Examples (6) and (7) with can’t are similarly unambiguous, but here the negation necessarily has scope over the modal can. Before drawing any conclusions, and investigating the rest of the modals, we need to clarify the possible types of negation in English. Given that the centre of interest here is the modals, we do this fairly briefly. More detailed discussion and argument may be found in Cormack and Smith (2000), which also discusses comparable scope data from Catalan, Italian and Basque. In Section 2, we distinguish two forms of negation, Pol[NEG] and Adv[NEG] (corresponding roughly to the familiar sentential vs. verb phrase negation). We then show in Section 3 that English modals vary in their relation to Pol[NEG], requiring that we divide them into those occurring pre-Pol (Modal1 ) and those occurring post-Pol (Modal2 ). The distribution of modals between the two classes is not however the same as in Catalan. Section 5 discusses the learnability problem this apparently causes. In Section 6 we show that in order to account for the scope facts exhibited in echoic contexts, our hypotheses require us to introduce a third negation/polarity position, which we call Echo, merged higher than Modal1 ; Section 7 shows further phenomena now accounted for. In Section 8, we discuss the relative scope of epistemics and deontics, particularly in relation to Cinque’s claims about their relative ordering (Cinque 1999), which we have to reject.

. Pol[NEG] and Adv[NEG] The permissible readings given in (4) to (7) are equally available if instead of -n’t we have not, as in (8) and (9). (8) Alfred should not eat nuts (9) Edwin can not climb trees a. ‘It is not permitted that Edwin climb trees’ b. ‘Edwin is permitted not to climb trees’

 [  [  [

However, other readings now appear as well. In particular, (9) allows both an interpretation with CAN in the scope of , as before, and an additional reading with  in the scope of CAN. This contrast is attributable to the difference between sentential negation, which we designate as Pol[NEG], and ‘VP’ or adverbial negation, which we give as Adv[NEG]. The contrast is well known (though frequently forgotten or ignored), and is marked by several syntactic

Modals and negation in English 

and inferential differences, as documented by Klima (1964). For example, in tag questions, the polarity of the normal tag is the opposite of the polarity of Pol, and is unaffected by Adv[NEG].4 That the extra reading in (9b), is due to the occurrence of Adv[NEG] can then be seen from the normal tag appropriate to each reading, shown in (10). (10) a.

Edwin can not climb trees, can he? ‘It is not permitted that Edwin climb trees’ b. Edwin can not climb trees, can’t he? ‘Edwin is permitted not to climb trees’

 [  [

Although (8) is logically unambiguous, similar alternations of tags arise, with Adv[NEG] in (11b), suggesting that the same structural alternation obtains here too. (11) a.

Alfred should not eat nuts, should he? ‘What Alfred should not do is eat nuts’ b. Alfred should not eat nuts, shouldn’t he. ‘What Alfred should do is not eat nuts’

 [  [

This distribution and the different scope interpretations can be explained as follows. Pol is a functional head, obligatorily present in all tensed clauses, and may have the value Pol[NEG] or Pol[POS]. Pol[NEG] has two instantiations, not and n’t. Adv[NEG] on the other hand is only optionally present, and is adjoined to some projection  than Pol.5 Adv[NEG] is only instantiated as not, as witness the absence of ambiguity in (6) and (7), as compared to (9). The clausal heads we are exploiting are then merged at LF as in (12). (12) C T

Pol(/) (Adv[]) . . .

Pol is also used by Haegeman (1995), but she places it above T. However, there are reasons in English for placing T above Pol. In John sometimes does not wear his glasses, the temporal adverb sometimes has scope over the negation Pol[NEG] not; but assuming that T is a temporal binder, sometimes must be in the scope of T. Hence Pol is in the scope of T.6

. Modal1 and Modal2 In addition to distinguishing two kinds of negation, we postulate that modals fall into two classes, Modal1 and Modal2 , and that each class is merged in a fixed position with respect to Pol. Should is an example of Modal1 , and is merged



Annabel Cormack and Neil Smith

higher than Pol; thus it has scope over both Pol[NEG] and Adv[NEG]. Can is an example of Modal2 , and is merged lower than Pol; thus it has scope lower than Pol[NEG], but higher than Adv[NEG], as required to give the alternation in (9). Plausibly, Modal1 and Modal2 in English and in Catalan are adjuncts, and optional (but see Section 5). The sequence in (12) can therefore be elaborated as in (13). (13) C T (Modal1 ) Pol(/) (Modal2 ) (Adv[]) . . .

Since the surface order in tensed clauses is never negation followed by a verbal element, as shown by the ungrammaticality of (14), some PF-displacement must take place. We return to this issue in Section 4. (14) a. *John not can eat pudding b. *John not eats pudding

Using -n’t as a probe for Pol, we are in a position to compare English with Catalan. As in Catalan, we find two classes of modal, given above as Modal1 and Modal2 . In the Catalan examples, the two classes divide along the epistemic vs. deontic dimension. In English, we find that the main division between Modal1 and Modal2 corresponds to the contrast between necessity and possibility, where obligation patterns with the former, and permission with the latter. However the epistemic vs. deontic division, also turns out to be relevant. The English modals which belong to the class Modal1 , like should, include almost all the ‘necessity’ operators under deontic readings: shall, must, ought + to, and is + to, as in (15) and (16): (15) Billy mustn’t/shan’t leave yet

 [

(16) Billy oughtn’t/isn’t/hasn’t to leave yet

 [

Epistemic readings are also possible for some of these, as shown in (17) to (19): (17) There isn’t to be another eclipse in my lifetime (18) There oughtn’t to be a problem finding the way (19) a. John must be at home b. *John must not be at home, must he c. *John mustn’t be at home

(normal tag)

Although (19a) has an epistemic reading, the impossibility of (19b, c) shows that neither of the two possible Pol[NEG] forms is available in standard British English. Because such sentences are available in other English dialects, including Tyneside English (McDonald and Beal (1987: 48), we suppose that this is

Modals and negation in English 

an accidental gap. We can code it by stipulating that epistemic must selects for Pol[POS]. It follows that epistemic must is necessarily also Modal1 . The status of the necessity modal has + to is unclear in Standard English, because it is obsolescent. However, it is current for example in Tyneside, where it unequivocally falls under Modal1 , as shown by the examples in (20), where (a) is accepted and (b) rejected or rephrased. (20) a.

(context: Mother to father over tea; Johnny upstairs, no previous conversation about TV) Johnny’s teacher says he hasn’t to watch any TV today: he’s got too much work to do.  [ # b. (context: Head to parents) In this school, a boy hasn’t to play games unless he wants to. * [

The only necessity modal falling under Modal2 is need (without to). This need is an NPI (negative polarity item), and in a declarative, has to be within the scope of a negative element.7 In (21), the only possible interpretations require that need is a Modal2 , in the scope of Pol[NEG]: (21) a. Hugh needn’t leave, need he b. Unicorns needn’t exist

(deontic)  [ (epistemic)  [

The assignment of a modal to necessity or possibility is not always obvious. For instance, it is not easy to tell whether volitional modal will and would, and the modal shall, are necessity or possibility operators. If we consider ‘wilfulness’, or ‘will’ in the sense of intention, then they should fall with necessity, but if we consider rather ‘willingness’, then they would appear to fall with possibility. Since ‘
¬’ and ‘¬♦’ are logically equivalent, scope facts with respect to Pol[NEG] are unhelpful. However, we saw in (10) and (11) that an occurrence of Adv[NEG] could be differentiated from one of Pol[NEG] by the use of tags. Consider (22), on a volitional reading. (22) John will not come home on time, won’t he

 [Adv 

If we treat volitional will as corresponding to intention, we get the correct scope interpretation (‘John intends not to come home on time’), but if we treat it as corresponding to willingness, we fail (#‘John is willing not to come home on time’). We conclude then that volitional will falls under the necessity modals. Accordingly, we now see that in (23), with Pol[NEG], we must assign will to Modal1 , along with the majority of the necessity modals. (23) John won’t come home on time, will he

 intention [Pol 



Annabel Cormack and Neil Smith

Similar arguments lead us to assign the related past form would, and volitional shall, to Modal1 , and it seems likely that futurate will (and the related would) are also Modal1 , unless they are under T itself, since finite Tense falls outside the scope of Pol. Of the ‘possibility’ operators, could behaves like can, in (6), (7) and (10), in systematically being in a post-Pol position, as does modal dare (which has only a root meaning, and is also generally an NPI). It seems then that for the majority of modals, the pre-Pol/post-Pol split in English corresponds to the ‘necessity’ vs. ‘possibility’ contrast, rather than to the epistemic vs. deontic contrast. We have not yet mentioned may and might. Surprisingly, their scope behaviour is like that of the Catalan modals. From (24a) and (25), we see that the interpretation for deontic may, is inside the scope of Pol[NEG] and outside it for epistemic may. For those who reject mayn’t, and allow only may not, (24b) shows that in addition to the possible Adv[NEG] interpretation in (i), there is also a Pol[NEG] interpretation, (ii). This is accounted for by assuming the LF order with deontic  inside the scope of Pol. In (25b), by contrast, no such interpretation arises, allowing us to infer that epistemic  is necessarily pre-Pol. (24) a.

Cyril mayn’t go to the party =‘It is permitted that Cyril not go to the party’ ‘It is not the case that Cyril is permitted to go to the party’ b. Cyril may not go to the party i. ‘It is permitted that Cyril not go to the party’ ii. ‘It is not the case that Cyril is permitted to go to the party’

(deontic) *MAY NOT   (deontic)    

(25) a.

David mayn’t be at home (epistemic) ‘It is possible that David is not at home’   =‘It is not the case that it is possible that David is at home’ *NOT MAY b. David may not be at home (epistemic) ‘It is possible that David is not at home’   =‘It is not the case that it is possible that David is at home’ *NOT MAY

Might is like may in changing its Modal category according to its interpretation as epistemic or deontic.8 If we assume that the position of a modal with respect to Pol[NEG] is fixed by selection, then the default assumption is that what is relevant for selection is simply Pol. We have referred to pre-Pol modals as Modal1 , and to the postPol modals as Modal2 . As far as we know, Modal2 and non-modal Aux show

Modals and negation in English

Table 1. LF positions for English modals Pre-Pol (Modal1 )

Post-Pol (Modal2 )

necessity

shall, should, must, will, would, ought + to, is + to, have +to

possibility

epistemic readings only: may, might

necessity

need

possibility

can, could, dare (only deontic) deontic readings only: may, might

no difference in their syntactic behaviour relating to Pol, so it might be that a single category label, with distinct selection features for individual items (i.e. selecting for Pol or not), is all that is needed. We leave the issue open. A summary of the LF positions for modals is given in Table 1. The data here conform, we believe, to the analyses given in Quirk et al. (1985 Section 10.67),9 in Palmer (1990) (with his class of “dynamic necessity” subsumed under our deontics), and to Brown (1991), with the proviso throughout that we have excluded readings which we take as obtainable only in echoic contexts, as well as factoring out readings due to Adv[NEG]. However, we disagree with Coates (1983) about some of the scope data, and in particular with her conclusion (Coates 1983: 237–239): “negation affects the modal predication if the modal has Root meaning” (i.e. deontics display Modal2 properties) “and affects the main predication if the modal has Epistemic meaning”, (i.e. epistemics display Modal1 properties).10 English also manifests Raising verbs with modal interpretations, recognisable by their requirement for do support, which include need+to and have+to. Because of their status as Verb rather than Modal, these are forced to have scope lower than Pol[NEG] and lower than any Modal/Aux. We will have little to say about them. We turn next to the PF positions of the modals, and to the problem raised by the irregular forms can’t and won’t.

. PF displacement: an alternative view of lexical entries In this section, we want to show that our analysis allows both for the displacement of a finitely inflected Modal2 to a higher position than Pol[NEG], and for a simple solution to a problem raised by can’t and won’t. Specifically, we argue that there are good reasons for treating these items as the PF part of a single lexical entry, although they are NOT unitary LF constituents.





Annabel Cormack and Neil Smith

We begin with the negative -n’t in can’t and won’t. One traditional treatment of -n’t postulates that it is a clitic, which attaches leftwards, to a tensed modal in T. An alternative has -n’t as a morphophonological operator, to allow for the irregularity when it cliticises for example to can or will, perhaps as a result of head-to-head movement. There are problems with both analyses. First, a clitic should not have irregular phonological effects, so that the first suggestion is implausible (Zwicky and Pullum (1983) provide systematic argumentation against the clitic hypothesis, in part on these grounds). Second, a morphophonological operator should affect only a head for which it selects, and we have argued that Modal1 (which includes will) selects for Pol, rather than vice versa, so the second suggestion is similarly suspect. In this situation, the appropriate and traditional move is to assume that items such as can’t and won’t, and indeed regular forms like mustn’t and hasn’t, are entered as such in the lexicon. We agree; but the form we postulate for the lexical entry is not the standard one. The standard solution is to assume that these irregular forms, and perhaps all the -n’t forms, are entered as wholes in the lexicon. But this is problematic, under standard assumptions about the architecture and the content of the lexicon. Consider the example in (26). (26) You can’t often bribe officials nowdays

 [ [

Example (26) contains the irregular form can’t. This must be given in the lexicon. However, as can be seen from the associated LF scope information,  +  is not a constituent at LF. Under these conditions, it is very hard to construct a meaning for can’t, and probably impossible without making the unjustified assumption that often is an optional argument of can’t. We claim that can’t has no unitary meaning, and hence no lexical entry of the standard form can be constructed.11 We claimed above that LF-parts of signs are merged in the positions at which they are interpreted. We further assumed – uncontroversially in a lexicalist framework – that the lexicon may contain morphologically complex phonological words. We see no reason to suppose that these are Merged anywhere other than in the place where they are heard. The PF-interpretable part of a head is merged in the position in which it is PF-interpreted. Since not every sign is interpreted at the LF interface in the position in which it is interpreted at the PF-interface (i.e. there may be ‘displacement’), it follows that a sign may be SPLIT into a PF-interpretable part and an LFinterpretable part before merger (see Cormack and Smith 1997, 1998). The operation ‘Split’ is not unique to our framework as it is required by UG,

Modals and negation in English 

since the structural representation of a sentence, phrase or word is interpreted independently at each of LF and PF. We take it that only the nodes and structure provided by LF may be used in determining the possible merge positions of the PF-interpretable part of a sign. The default case is that the PF-part is merged along with its LF-part on the same tree. In other cases, the PF-parts of signs are merged at nodes which are required by the LF-structure, rather than in an independently derived structure of the kind proposed by Jackendoff (1997).12 In (27), the different orderings at PF and LF show that either the Modal or the Negation must be split. (27) The dog could not reach the ball

PF order: could not LF (scope) order:  [  [ . . .

That it is the modal which is split in (27) can be seen from the interaction with adverbs (Ernst 1992), as in the examples shown in (28) with the scope readings given. (28) a. John can not come home late b. Sylvia can never climb that tree c. You could not often bribe officials, then

[Pol[NEG] [ [Pol[POS] Ø [ [ [Pol[NEG] [ [

In particular, in (28c) the LF scope ordering of not and can is disturbed at PF relative to the LF, but that of not and the temporal adverb is the same at PF. All the examples in (28) are consistent with the positioning of PF-can at some position preceding Pol, including Pol[POS] in the case of (28b). In Cormack and Smith (1997, 1998, 2000), we argue against the idea that could moves to, and relates to, T[]. Instead, we argue that it is a PF-form relating to two LF-items: a T-related inflectional head, Inflpast , and LF-CAN.13 It is merged at the position of LF-Inflpast . The T-related Infl has the identity function (ID) for its meaning. Hence for (28c) we obtain a structure with the parts distributed as in (29). (29) LF: [T  [InflPast  [Pol  [Adv  [Modal  PF: Ø could not often

The required lexical entry is not simply a pairing of PF-interpretable form and LF-interpretable meaning. In particular, if a PF form is associated with two LF forms A and B, checking of LF- and PF-parts will be needed for both A and B. We suppose that this depends on categories, so that the PF form will need to have as features both the category of A and that of B. For could, the entry relates one form to two meanings, as shown in (30):



Annabel Cormack and Neil Smith

(30) PF:
category {Inflpast , Modal}; form /kυd/  LF: (category Inflpast ; meaning ), (category Modal; meaning )

Similarly, the irregular form can’t is entered with PF- and LF-parts related as in (31), where the LF has three separate components: (31) PF: category {Inflpres , Modal, Pol[]}; form /ka:nt/
LF: (category Inflpres ; meaning ), (category Pol[]; meaning ),  (category Modal; meaning )

The PF-form can’t is simply   the three LF meanings  and  and the ID of LF-Inflpres by checking. How these turn up in the LF representation is not stipulated in the semantic part of the lexical entry, but arises only from selection possibilities. Since modals may or may not select for Pol, the relative positions of LF-modal and LF-Pol may vary according to the class of modal. All that is required is that a proper checking relation hold between each LF category and the related composite PF category. This account deals satisfactorily not only with the problem of can’t and won’t, but with the displacement seen in example (28), and the ungrammaticality of the examples in (14). For details of the checking requirements, see Cormack and Smith (1998).

. Categorisation of modals under Modal1 and Modal2 : implications We have used the labels ‘Modal1 ’ and ‘Modal2 ’ for convenience. However, it is probable that the modals have the same category, Aux, as the regular auxiliaries have, be, and do. Modal1 is then distinguished by selecting for (a projection of) the functional head Pol, and the modal auxiliaries generally are distinguished in standard English, but not in Scots English, by their lack of non-finite inflectional forms.14 One implication of the separation of Modal1 and Modal2 is that the claim (made for instance from Pollock (1989) to Lightfoot (1999: 185)), that all Modals are merged under T (or I), cannot be correct. While this analysis might be feasible for Modal1 , or at least for futurate will, it is not feasible for Modal2 .15 Picallo (1990) makes essentially this point with respect to Catalan (postulating that only epistemics are under I/T), as does Roberts (1998) for English. Further evidence can be provided from the double modal constructions of Scots English and other languages, as Brown (1991) shows for Hawick Scots in (32), and Picallo (1990) for Catalan in (33).

Modals and negation in English 

(32) He should can go tomorrow (33) En Pere deu poder tocar el piano must can play the piano ‘It must be the case that Peter is able/allowed to play the piano’

Since we have assumed that the modals are optional adjuncts, we should expect iteration of Modal1 or Modal2 to be possible in principle. This is excluded in Hawick Scots (see discussion in Section 8), if we assume that futurate will, would are merged under T. However, if we assume the same distribution of Modal1 and Modal2 as standard English, iteration can apparently occur in other varieties of Scottish English and in Jamaican Creole, as shown in (34) and (35). (34) a. They should ought to make the rules clear (Brown and Millar 1980) b. He might no should claim his expenses (Miller and Brown 1982) (35) de tif wuda mos hafi ron ‘the thief would have had to run’

(Bailey 1966)

However, more work on individual languages and dialects is needed before we draw this conclusion. For example, Brown and Millar (1980) state that in the Edinburgh dialect investigated ought to is a main verb, not a modal, so (34a) does not after all show Modal1 iteration. The example in (34b) is also not unequivocal, since might has some curious behaviours. It does not occur with negation except Adv[NEG], nor does it invert, in Scots. In Southern States English ‘he might could . . . ’ inverts if at all to give either ‘could he might . . . ’ or ‘might could he . . . ’ (Battistella 1991, 1995).16 In (35), the wuda is futurate ‘would’, and is possibly under T; mos ‘must’ is very likely a Modal1 , but hafi may be a main verb here, not a modal auxiliary.17 We have in fact no direct evidence that the categories Modal1 or Modal2 may iterate.18 We wish to raise two further issues. One, on relative scope, we defer to Section 8; the other, on learnability, we discuss now. The obvious problem which emerges from the English data is that the classification of a particular modal under Modal1 or Modal2 is going to have to be learned.19 If the Catalan pattern had been common to every language, we could have looked to UG or semantics for an explanation of the distribution, leaving the child with nothing to learn. As it is, however, the learner appears to have a formidable task, particularly given the erratic distribution of modals in English. However, there is evidence that deontics are always learned earlier than epistemics (see Papafragou 1998 for a survey), so we may assume that their classification is also learned first. If we take just the deontics from the table in



Annabel Cormack and Neil Smith

Table 2. LF positions for English deontic modals Pre-Pol (Modal1 )

necessity

shall, should, must, will, would, ought + to, is + to

Post-Pol (Modal2 )

possibility

can, could, dare, may, might

Table 1, and eliminate the obsolescent modals need and have + to, we obtain the sub-table shown in Table 2. This is much more helpful. We may suppose that the child knows that there is a  distinction between obligation vs. permission/ability modals (given as necessity vs. possibility in the table), and its task is to determine whether there is such a split in the ambient language: ‘No’, as in Catalan, where all the deontics fall under Modal2 , or ‘Yes’, as in English. The older child, learning the epistemics, can take the pattern for the deontics as the default. For English, this will entail learning that the exceptional epistemic may and might fall under Modal1 , and that the obsolescent necessity items need falls under Modal2 . For Catalan, it will entail learning that the totality of the epistemics fall under Modal1 . It is tempting to speculate further, but we need relevant data from more languages. In particular, we need to know whether all languages expressing these notions with modal auxiliaries have one major split, and whether the combinations epistemic+necessity and deontic+possibility always correlate with Modal1 and Modal2 , respectively.

. Negation in echoic contexts There are some exceptions to the generalisation that Modal1 takes negation in its scope (see Gazdar et al. 1982, fn.17; Palmer 1990; Brown 1991; Williams 1994; Menaugh 1995 among others).20 In all the following examples, negation may or must have wide scope over modal should. (36) A to B: Shouldn’t you be at school? Interpretation: Is it not the case that you (B) should be at school?

The same interpretation is available when the question is asked using just intonation: (37) A to B: You shouldn’t be at school?

In tags, wide scope negation is obligatory: (38) You should eat more vegetables, shouldn’t you?

Modals and negation in English 

It is also available in declaratives: (39) A: You should eat more vegetables B: No I shouldn’t Interpretation: No; it is not the case that I (B) should eat more vegetables.

It is clear that the natural interpretation of each of these requires the should to be in the scope of the negation. But there are robust data indicating that should is a Modal1 head, falling always outside the scope of negation in ordinary contexts. There are two problems then: what contexts require that should falls inside negation, and how is this accommodated within our framework. It is important to note that the surface position of the negative morpheme is insufficient to predict the scope order. For a negative element to have scope over should, it is neither necessary nor sufficient that it c-command should at surface. In (40), a negative element c-commands should, but the scope order is as usual, with the only interpretation being  .21 (40) Never should anyone have entered the room

In (41), there is an interpretation entirely parallel to that of (36), yet the negation is on the surface lower than should. (41) Should you not be at work? ‘Is it not the case that you should be at work?’

Borrowing from Relevance Theory (Sperber and Wilson 1995), we characterise the contexts exhibited in (36), (38) and (39) as  . The echoic nature of the reply in (39) is clear: the content ‘you should eat more vegetables’ is an echo of the original given by speaker A. Tags, as in (38), are always echoic. The idea with (36) is that the belief, or at least, expectation, ‘B should be at school’ is held by A, and the question asks whether that belief should be revised to its negation. We call the negation which is external to an echoed proposition echo-negation. By contrast, the example in (40), like those in previous sections, shows no echoic use. There is now a syntactic question to be answered. If Merge constructs LF, and if  is Merged above Pol, then how is echo-negation to be accounted for? If, as we have argued, Merge gives the scope position of a head, then there is only one possible answer: there must be a syntactic position, (call it Echo, for negation in an echoic context), somewhere above Modal1 .22 This in turn leads to questions about the status of the not in (41). We will simply assume here that although Echo[NEG] is high in the tree, there are reasons why its PF-part is merged at a position lower than the tense-inflected modal or auxiliary.23



Annabel Cormack and Neil Smith

The position of Echo is lower than that of the Q head indicating the ‘force’ of a question, but presumably above the Case-licensed position of the subject. The Echo head is normally confined to root environments, but can, as one would expect, appear embedded under heads which select for representations of thoughts or speech, such as (42), and it appears lower than the adjoined if -clause in (43). (42) Nora asked/wondered whether (on a weekday), Jo should not be at school (43) If a man owns a donkey, mustn’t he feed it?

We stipulate that the Echo head selects for C, and hence occurs freely with respect to other optional adjuncts such as Topic. The range of heads we are concerned with is now as in (44) at Merge/LF: (44) (Q) (Echo) C T

(Modal1 ) Pol (Modal2 ) (Adv[]) . . .

. Semantic implications of the three negation positions In this section we look a little more closely at some of the interactions between the modals and the different negation positions. We have couched our analysis of the data in accordance with our hypothesis that the LF-interpretable part of a head is merged at the position in which it is LF-interpreted. As we noted, this entails that the PF-part of Echo[NEG] is ‘lowered’ relative to the LF-part. A more standard analysis would probably assume that there is only one [NEG] head, subsuming Echo and Pol, and that this is merged at the PF-position. Subsequently, some process of LF-movement raises at least the LF-interpretable part of [NEG] in echoic contexts – perhaps to some head which we might still call Echo. Note however that given the data about the scope of adverbs (Section 4), we cannot similarly invoke movement of LF-NEG to explain the differing scope of NEG and the modals – nor, given the untidiness of the data in Table 1, is there any obvious motivation of a semantic kind for so doing. Discussions of the modals designed to explain their use and usages (such as that of Palmer 1990) are often concerned with how certain meanings may be expressed. If we look at the array of modals and possible negation positions in (44), it is clear that with respect to a fixed modal of the Modal1 class, there is a gap. It is not possible to express the semantic content given by ‘ [Modal1 . . . ’ directly except in echoic contexts.

Modals and negation in English 

In a language like English, with a variety of modals, there is a simple alternative: switch to a Modal2 or a main verb with the same meaning. Thus to express epistemic ‘ [ . . . ’, it is possible to use Pol[NEG] with Modal2 can; to express ‘ [ . . . ’ it is possible to use Pol[NEG] and Modal2 need or main verb has to. However, it is important to realise that the semantic content ‘ [Modal1 . . . ’ is available with Echo[NEG], and is appropriate for tags, and often for questions, as in (36) and (39) above. Palmer (1990: 76) notes the wide scope of negation in what he calls ‘verbal crossing out’, giving the example in (45): (45) He must come – Oh no he mustn’t.

Notice too that epistemic must, although it cannot occur with Pol[NEG] in standard British English, is available with Echo[NEG], as in (46), or in one interpretation of the rhetorical question in (47): (46) Mustn’t John have left? (47) In heaven, must the lion not lie down with the lamb?

We have looked so far at the direct scope relations between the modals and negation. Interestingly, the scope of the modals in relation to the various negation positions seems to have implications for the scope of negative elements of other kinds. Palmer (1990: 40) says: If any of the negative forms such as never, no one, nobody, nothing, nowhere, occur with a non-negative form of the modal, the same rules apply as for the negative forms of the modal. . . . You must never do that You need never do that I must tell no one about it I need tell no one about it.

As we would expect from this, with Modal2 can, the scope of a negative item such as nothing is ambiguous: (48) John can eat nothing ‘There is nothing that John is permitted to eat’ ‘That John eat nothing is permitted’

 [∃ [♦ ≡  [♦[∃ ♦ [ [∃

With the modal may, we get the expected difference between deontic and epistemic readings. (49) interpreted deontically is ambiguous in the same way as (48) is. However, interpreted epistemically, it can only have the ‘POSSIBLE



Annabel Cormack and Neil Smith

[NOT [∃. . . ’ interpretation, ‘That John may put nothing in the collection plate is possible’. (49) John may put nothing in the collection plate

The difference between the available interpretations with must and need is not, however, a simple matter of scope. The relevant scopes for non-echoic readings of Palmer’s last two examples above are as shown in (50) and (51). (50) I must tell no one about it

 [  [ ∃ . . .

(51) I need tell no one about it

 [ [ ∃ . . .

What is striking is the splitting of the negative existential of no-one around need.24 However, in all these cases, the  takes scope in relation to the modal in a position where either Adv[NEG] or Pol[NEG] could. This provides support for an analysis modelled on that of Haegeman (1995: 182–187), where the presence of an n-word/phrase induces a [NEG] feature on Pol.25 However, because of the varied scope positions of the existential part, an LF Neg-criterion, requiring the n-phrase, or part of it, to be in Spec[NEG], is not appropriate. Problems would equally arise for Kayne’s overt movement to Spec[NEG] (Kayne 1998). In our terms, we would stipulate in the lexicon that the quantifier ‘no’, for instance, has a PF no associated both with [X ¬] and with an existential quantifier [D ∃], where [X ¬] may be Pol[NEG] or Adv[NEG], or, as we shall see, Echo[NEG]. This solution is closely related to the neg-incorporation of Klima (1964: 267–280). We do not discuss here the means by which the existential gets to obtain the correct scope, nor do we give all the possible readings for the example sentences (in particular, other split readings are obtainable). An additional benefit of our analysis is that the scope of the negative element is correctly predicted for negative operators c-commanding must at LF: in a non-echoic context, (52) is unambiguous, and (53), with no scope alternation possible, unobtainable. (52) No-one must pass the barrier a. #‘There is no-one who must pass the barrier’ b. ‘That no-one pass the barrier is obligatory’

 [∃ [  [ [∃

(53) #They never must escape

 [

Further, only in Echoic contexts can a negative element be interpreted as having scope over a Modal1 . In such a context, the readings marked with ‘#’ in (52) and (53) are obtainable. Similar examples (shown in (54) below) are given by Palmer (1990: 117).

Modals and negation in English

(54) a. No one  go b. You never  do it (55) A: What must I eat? B: You  eat ! ‘There is nothing that you  eat’

Thus the scope patterns for the negation element of an English n-word is not predictable from the semantics of the modal, but precisely follow the vagaries of the scopes available for each individual modal with respect to negation, according to context.

. The relative scope of epistemics and deontics A related matter of some interest comes from various attempts to argue for a Modal Scope Constraint which takes the Catalan pattern to be operative as a matter of necessity. The argument has two versions: first, that epistemics must of necessity fall outside negation, and deontics inside; and second, that epistemic modality must have scope over deontic modality, and in consequence, that the two are ordered thus in syntax.26 Cinque (1999) is an extensive and elaborate cross-linguistic attempt to order functional heads, including modal adverbs and auxiliaries, but specifically excluding negation, in this fashion. His ordering of positions includes the array shown in (56) (abstracted and simplified from Cinque’s (90) of §3.7), but he casts the restrictions as generally belonging to UG, rather than to logical or conceptual necessity (§6.4).27 (56) Moodepistemic > T > Moodroot > . . . . . .

Given our analysis in (44), with Pol[NEG] as a functional head with a fixed position, and granting that English has Modal1 deontics (e.g. deontic must) and Modal2 epistemics (e.g. epistemic can), it is clear that we cannot agree that epistemics are merged above deontics (and we also argue in Cormack and Smith (2000) that epistemics are merged below T). Further, even in Italian, which provided the initial impetus for Cinque’s analysis, there is evidence that there are Modal1 deontics and Modal2 epistemics. In the deontic examples below, provided by Vieri Samek-Lodovici, the contrast between the scopes of potere and dovere with respect to the negation can be explained by assuming that we have Pol[NEG] here, and that deontic potere is a Modal2 and dovere a Modal1 .28





Annabel Cormack and Neil Smith

 [

(57) Gianni non può canTAre.

*CAN [NOT

*NOT [MUST  [

(58) Gianni non deve canTAre

Epistemic uses pattern the same way, as the translations for the clause in (59) indicate. (59) (If the forecast is right), it shouldn’t rain . . . , non può PIOvere not may rain . . . , non dovrebbe PIOvere not must rain

 [  [

The conclusion then is that in standard Italian, potere ‘can’ is a Modal2 , and dovere ‘must’ is a Modal1 , like their English equivalents. Cinque’s analysis, however, also makes some correct predictions for English. If in standard British English, we set up a clause with two modals or modal verbs, we indeed find that the interpretation never has a deontic with scope over an epistemic. (60) John might be able to get here in time Modal2 (epistemic) – verb (deontic) (61) John could have to leave before nine Modal2 (epistemic) – verb (deontic) (62) John must have to stay at home

Modal1 (epistemic) – verb (deontic)

(63) There must/might have to be three solutions Modal1 /Modal2 (epistemic) – verb (epistemic) (64) The cat can be allowed out at night Modal2 (deontic) – verb (deontic) ‘I permit you to permit the cat to go out at night’

The complexity of the last two examples is presumably in part a function of the fact that it is at least usual for two epistemic or two deontic modal terms used successively to have differing interpretations with respect to the source of the modal imperative or permit (Kratzer 1977), as in the sample gloss in (64) for instance. In Scots, where double modals are permitted, the same interpretive constraints hold. The examples below, and the paraphrases where given, are taken from Brown (1991). Futurate will/would is omitted from our scope annotation. (65) He should can go tomorrow Modal1 epistemic – Modal2 deontic (= ‘he ought to be able to go tomorrow’)

Modals and negation in English

(66) He would could do it if he tried Modal1 futurate – Modal2 deontic (= ‘he would be able to do it if he tried’) (67) He’ll might can come in the morn Modal1 epistemic – Modal2 deontic (It is possible that he will be able to . . .) (68) You’ll have to can do it whether you like it or not Modal1 epistemic – Modal2 deontic29 (= ‘you’ll have to be able to do it . . .’) (69) He might could have to go Modal1 epistemic – Modal2 deontic – Main-verb deontic

Brown (1991: 76–78) claims that with modal auxiliaries, the only possible modal combinations are drawn from a (future will)–(epistemic)–(deontic) sequence, with no two of the same class. This suggests that the modal sequence is restricted to one future (perhaps under T), one Modal1 , and one Modal2 . If this is correct, then it seems that Modal1 and Modal2 might be optional functional projections, rather than optional adjuncts which could iterate.30 Our hypothesis entails that there is another constraint on the ordering of two modals: Modal1 must precede Modal2 at LF (the ‘Modal1 >Modal2 ’ constraint). In a language like Catalan, the two constraints will coincide, but in Scottish English, they will not. Examination of the examples given in Brown (1991) reveals no counter evidence. Rather,  constraints must be met, so that the first modal must be both an epistemic and a Modal1 , and the second, both deontic, and a Modal2 . This accounts for the fact that in Hawick Scots, the only modal auxiliary that can stand in the deontic position in a double modal is can/could, since this is the only deontic Modal2 . Combinations such as must should are not permitted even when interpreted as epistemic followed by deontic.31 If both these independent constraints on scope ordering hold, then only one of them can be syntactic, that is, be coded by selection. In fact, there are several reasons for supposing that the Modal1 >Modal2 organisation is syntactic, and the epistemic–deontic Modal Scope Constraint is conceptual. A. As we have seen in English, the choice of Modal1 or Modal2 is not systematically related to meaning, so it must be lexical. The use of a fixed functional head Pol makes this a matter of selection, which is a normal part of a lexical entry. For Cinque, the situation would be harder, since the Modal1 >Modal2 constraint could not be one of selection, but (presumably) simply stipulation of scope with respect to Pol (where Pol may generally be inserted freely among the functional heads). In addition, it would also be necessary to en-





Annabel Cormack and Neil Smith

sure that Pol[POS] was inserted exactly once into or at the boundary of the relevant string of functional projections, where Pol[NEG] is not present.32 B. The epistemic–deontic ordering constraint does not hold just of modal auxiliaries (and adverbs), but holds also of modal main verbs. This can be seen in the English examples in (60) to (64), and holds of languages which probably only have main verb “modals”: e.g. Icelandic (Thráinsson and Vikner 1995), and Greek (Roussou 1999).33 C. The Epistemic>Deontic constraint is plausibly conceptual (where Modal1 > Modal2 cannot be). This last statement requires some elucidation. It has been suggested in the literature that epistemics refer to the speaker’s evaluation of a proposition (a statement about possible worlds) vis-à-vis the actual world, whereas deontics refer to situations rather than worlds. This is supposed to explain the scope constraint, though it is not clear exactly how. We suggest instead an explanation depending on the properties of deontics (as does Vikner 1988). Deontic and other root meanings are normally directed at some responsible person. Note that as the following examples demonstrate, the responsible person is not necessarily even the underlying subject of the modalised clause.34 (70) a. The baby need not eat before noon b. The room must be tidy before I come back

The responsible person is (or is not) permitted or required to undertake some action or realise some state: for obvious reasons, the requirement should be one that it is possible to undertake; and if its description were modified by epistemic or alethic modals, than that part of the undertaking is not within the powers of the responsible person to alter.35 Thus it cannot form part of a proper permission or request. However, in a few cases, the responsible person does not seem to exist, although the statement can clearly be interpreted as deontic, as in (71): (71) These things shouldn’t happen

In this sort of case, it does seem possible, after a plaint of “Why did it have to happen to me?” to have the deontically interpreted modal followed by an epistemically interpreted modal verb: (72) These things should not have to happen to anyone

Finally, we turn to the interaction of epistemic and deontic modals in more complex structures. The Epistemic>Deontic constraint cannot hold through-

Modals and negation in English

out an arbitrarily long sentence. We obviously would not expect it to hold across clause boundaries with an independent subject in the lower clause, as in (73a). It holds only for reasonably close modals. Thus deontic followed by epistemic interpretation is available for the examples in (73) but not those in (74) and (75): (73) a. b. c. d. e.

John can stipulate that the puzzle must have at least one solution John can claim it to have to be true that quarks exist John can want himself not to have to be mortal John can want PRO not to have to be mortal John can try PRO not to have to be mortal (by manipulating his genes)

(74) a. ?#John may begin t not to have to be mortal b. John may end up t not having to be mortal c. #John may have to t leave early (75) #John may seem t foolish

(epistemic seem)36

The data suggest that what is crucial is the presence of an intervening nonmodal V, irrespective of the kind of subject the intermediate verb licenses. It appears that the aspectuals begin and end up are not main verbs; or are “modal” in some relevant sense. The effect of the intervening verb is consistent with the Conceptual approach to the Epistemic>Deontic constraint: in all the cases above with the lower clause having the “same” subject as the matrix clause, the complement of the intervening verb is interpretive (in the sense of Relevance Theory), representing wishes, thoughts, etc., on the part of the subject, and so being freely attainable by the responsible person. As a conceptual constraint, we could interpret the Epistemic>Deontic constraint in two ways. One, which would rule out the suggested deontic>epistemic interpretation of (72), would hypothesise that the constraint held over representations in the Language of Thought (Fodor 1975). The combination of modal and epistemic is literally unthinkable. The weaker claim would be that there was a pragmatic processing constraint or strong preference, which determined that when two modal expressions occurred in the same modal domain (i.e. without an intervening lexical verb) then they would not be interpreted as a deontic followed by an epistemic, because this was never likely to be the correct interpretation. If the constraint that epistemics take scope over deontics really is conceptual, then the Catalan system would seem quite natural, though not a necessity. We know of no rationale for explaining the necessity/possibility split evidenced by English.





Annabel Cormack and Neil Smith

. Conclusion In undertaking this analysis of the behaviour of modals with respect to negation, we have made essential use of syntactic, semantic, and pragmatic theories. Our initial theoretical assumption, that heads are merged in their relative scope positions, drives the analysis. It led us to introduce a separate projection for Echo[NEG], in addition to Pol[NEG] and Adv[NEG], and to divide the modals into those merged in a pre-Pol position, and those merged post-Pol. We claim that distinguishing Modal1 and Modal2 , and postulating the three varieties of negation, makes sense of a large number of otherwise messy or unaccountable pieces of data relating to the modals in English. The analysis is incompatible with Cinque’s (1999) claim that epistemics are merged in a fixed syntactic position higher than that available for deontics; we reinterpret his data as a cognitive requirement that an epistemic cannot be in the local scope of a deontic. We showed also that there exist PF words such as can’t and nothing which may be associated with two non-adjacent LF-interpretable items. We show that if the relation between the LF-interpretable and the PF-interpretable parts of a sign is established by checking rather than movement, then there is a straightforward account of these possiblities. We expect that investigation of other languages along the same lines will confirm the explanatory value of our hypotheses.

Notes * We are grateful to Joan Beal, Virginia Brennan, Keith Brown, Sophie Cormack, Dorothy Disterheft, Jan Engh, John Harris, Larry Horn, Vieri Samek-Lodovici, Steve Nagle, Frank Palmer, Anna Papafragou, and two anonymous reviewers, for various forms of help with this paper. None of them is responsible for our conclusions. . Our position can be seen as the limiting case of a ‘merge chain’ architecture of the kind argued for by Brody (1995). . We assume that all the scopal elements with which we are concerned here are heads, and that the only specifier-like elements here are quantified noun phrases (see further in Cormack 1999). . The same interpretations are obtained when haber, which tends to bias the readings, is not present (Carme Picallo, pc to Frank Palmer). The negation of el lladre pogui entrar per la finestra will have the reading ‘the thief was not able to come in by the window’ if no is before the modal, and the reading ‘it is possible that the thief did not come in by the window’ if the negation is after the modal. However, la llet no pot bullir en aquestas condicions can have the

Modals and negation in English

reading ‘the milk may not boil in these conditions’. This is consistent with Picallo’s claims provided that the negation is Echo[NEG]. See further in Cormack and Smith (2000). . Abnormal tags may indicate irony, polite or threatening requests for confirmation, and other marked interrogative uses: “It was going to be sunny today, was it?”, “You’ve booked for Tuesday, have you?”, “You think I’m silly, do you?”; “You didn’t remember, didn’t you?”. Note that all tags are echoic (see Section 6), with either Echo[NEG] or Echo[POS]. . One of our reviewers is concerned that some instances of negation appear to be functional heads, and some to be adjuncts. However, we do not see this as a problem. Both are heads (we reject specifiers – see Cormack 1999); both have the same semantic content. The only real difference is optionality: functional heads are obligatorily projected; optionally projected heads are instantiated as adjuncts. . Alternatively, suppose the temporal adverb (or a default existential) introduces quantification over times, while the Tense node introduces a restriction on these. Such a restriction cannot have scope under the modal, and so here too T must have scope over Pol. Placing the Tense predicate under the modal would lead to a representation of the form ‘∃t
[(t) ∧ P(t)]’, where
is the necessity operator and P a predicate over times. But this entails that there are some times which are obligatorily in the present, which is presumably false. Further, with Modal1 have to, the  is intuitively within the scope of Tense. John often had to sneeze does not mean ‘there exist many times which necessarily are in the past and at which John sneezes’; nor even ‘there exist many time in the past at which necessarily John sneezes’. it means ‘During the past, there were many times at which necessarily John sneezed’. This corresponds to the scope order T Adv Modal, which is what we are claiming to be correct. . Auxiliary need nowadays seems to be licensed in some, but not all, the usual English NPI environments: *You need leave; You needn’t leave; Need I leave? but *Everyone who I need see must come to my office; *I will phone if I need leave late. . The deontic possibility judgements for could and might are less certain than the others. We refer here not to the past tense forms of can and may, but to the ‘conditional’ uses. We have not in this paper considered whether there should be a Mood head accounting for the can/could contrast where it is not one of tense (and similarly for should, would). There is evidence (pace Lightfoot 1999: 184) that the past tense versions are still current: (i)

They will keep nagging me/They would keep nagging me

(ii) She says I may leave/She said I might leave (strict Sequence of Tenses) . The authors however find the scope of what we call Pol[NEG] negation to be “neutralised” in the case of all senses of will. We disagree. . For Coates, only deontic must is claimed to be exceptional. Why do the discrepancies between our positions arise? Consider first the deontic readings of the necessity modals, which in effect she claims belong under Modal2 , where we claim they fall under Modal1 . First, understandably, there is some difficulty in ascertaining the scope facts for the modal readings of will and shall (and the related would and should), which Coates herself notes. Second, should and ought are misclassified owing to an unfortunate choice of paraphrase, though again Coates recognises a problem. Coates (op. cit. 239) gives for ‘I ought not to/should not





Annabel Cormack and Neil Smith

do x’ alternate paraphrases of ‘It is not advisable for me to do x’ and ‘it is advisable for me not to do x’. However, the first, although having a surface order ‘not . . . advisable’ is itself ambiguous as to its explicature (because of ‘neg raising’ effects). The paraphrases we supplied in (4) are unambiguous, and the wide-scope negation reading is clearly excluded. The other area of disagreement is over the epistemic readings of possibility modals, which for Coates are Modal1 , and for us, Modal2 . We disagree over can and could. Although Coates discusses epistemic can’t (op. cit. 20), she omits epistemic can/can’t from her table. Epistemic can is of course entirely possible: (i) Errors can occur at this stage of the process. This combination of errors and omissions is sufficient for her to have found English to be divided mainly on the epistemic/deontic line, rather than, as we have argued, mainly along the necessity/possibility line. . The meaning could be given as λp[(CAN p)], where p is a variable over predicate meanings (compare Gazdar et al. 1982, rule 15). However, for the case where the adverb is present, even if we could have the adverb selected by the modal, the latter would have to have a meaning λaλp[(a (CAN p)], where a is a variable over adverb meanings. The consequent disjunctive lexical entry is bizarre. . We do however assume that the internal structure of a complex PF-item is generated in an independent Morphophonological component, as has been proposed by Anderson (1992) and others. That such a component exists independently of syntax is supported by the dissociation exhibited by Christopher, a savant, whose second language learning exhibits poor syntax (little changed from his native English) along with strikingly good morphological acquisition (Smith and Tsimpli 1995). . We claim then that the situation with respect to Tense[], say, is the same as with Aux[] have: it checks for the occurrence of particular Infl associated with the following verb. The difference is simply that Tense[⁄-] itself, unlike Aux[] although bearing semantic content, is not phonologically overt. . A reviewer suggests that we consider the option of merging Pol either above or below a modal. There are three reasons for discarding this option. The first is that there is evidence (discussed in Cormack and Smith 2000) for Pol[POS]. Since this contributes nothing substantive to the semantics, its presence can be accounted for only if it is obligatory. We assume that this means that it is a functional head with a fixed position, as indicated in (12). Second, if Pol[NEG] were placed in relation to the modals by some sort of PPI/NPI distinction, this would have to be relativised to the kind of negation (i.e. Pol[NEG] but not Adv[NEG] or Echo[NEG]), unlike regular NPI/PPI items. Thirdly, we still need the Echo/Pol/Adv[NEG] distinction, since only the value of Pol is relevant to the polarity of the clause as a whole, as evidenced by the Klima tests (for use of a subset of the tests in a variety of languages, see de Haan 1997). . In fact we argue in Cormack and Smith (1998) that it is generally wrong for English. . There is the additional problem raised in Brown (1991: 97) of ‘tense concord’ between may/might and a following can/could. . Note that Bailey’s use of Mod1 and Mod2 refer to the PF positions of the modals in relation to negation, not the LF scope positions. Since there is evidence that Bailey’s Mod1 divide into our Modal1 and Modal2 (glosses of (shudn as ‘shouldn’t’ and kudn as ‘couldn’t’

Modals and negation in English 

(Bailey 1966: 141)), we take it that Bailey’s Mod1 are modal auxiliaries, and her Mod2 are either main verbs, or are those modals occurring in second position in a sequence of modals, as in the Hawick Scots examples. . A reviewer supplies a Dutch example hij zou moeten kunnen komen ‘he should be able to come’ (lit. ‘he should must can come’). If Dutch modals are Modal/Aux, then there must be iteration at Modal1 or Modal2 here. Note however that iteration is also possible in principle with Raising or Control verbs, and that the categories of Dutch modals are not established. See also the remarks on some Scandinavian languages in footnote 31. . Data in Palmer (1995, 1997) and de Haan (1997) suggest that many languages have modals whose classification is unpredictable on the basis of their semantics. . We are grateful to Keith Brown for insisting on our taking seriously the comparable scope readings in Brown (1991). . The scope order ‘ ₍ . . .’ would indicate for a deontic reading that there is never an obligation . . . Such a reading is not obtainable. . Linguistic evidence for echoic negation is also given by Espinal (1993) for Catalan and Zanuttini (1997a: 67–104, 1997b: 228) for Italian dialects. Zanuttini also argues that the two kinds of negation are merged in distinct positions. . For an explanation, see Cormack and Smith (1998, 2000), where we argue that the underlying reason is the same as that which drives raising of a finitely inflected modal or auxiliary over Pol[NEG]. . See Krifka (1998: 104) and Jacobs (1980) for analogous data in German, and for Dutch, “. . . waarin geen opgesplitst wordt in twee delen” (‘where geen is split up into two parts’) see Rullman (1995). Note that the separation of a determiner into two logical operators which take scope in distinct positions is required also for some instances of wh-phrase reconstruction (see for instance discussion of How many books does Chris want to buy? in Lechner 1998: 280). Note that no alternative form with ‘∀¬’ replacing ‘¬∃’ is possible here. Consequently, analyses of n-words postulating that they are universals such as that in Haegeman and Zanuttini (1996: 122) are probably incorrect. The authors recognise other problems with the analysis (p. 126 and footnote 20). . See also Rowlett (1998: 124–125). . The discussion usually includes modal adverbs, and generally makes more distinctions than just ‘epistemic’ vs. ‘deontic’. There is inconclusive discussion going back to Jackendoff (1972). . Cinque states (§6.4) that an epistemic modal must have scope over Tense, because it needs a tensed proposition for evaluation. However, if Tense is a binder of a tense-variable emanating from V, or is an operator on such a variable, as we assume (see Section 2 and footnote 6), then Cinque’s conclusion does not follow. . Capitalisation marks the main sentence stress. If the modal is stressed, Echo[NEG] readings will generally be obtained. If a transitive verb is used, and stress falls on the noun phrase complement, contrastive focus (also frequently associated with Echo[NEG]) may be induced. . If have to is a main verb, the can here will be in an embedded clause.



Annabel Cormack and Neil Smith . If the functional projection were obligatory, then there would have to be phonologically null and semantically trivial elements of these categories available when no overt modal is present. These would have to have ad hoc properties to ensure for example that a null Modal1 would not interfere with do support, so we assume this is not a viable alternative. . The data given in Vikner (1988) and Thráinsson and Vikner (1995) suggest that the same may be true of Danish, which in turn suggests that Danish modals may be modal auxiliaries, contra Vikner (1988). However, in Icelandic (Thráinsson & Vikner 1995), the combinations do not seem to be limited in this way. This is consistent with the authors’ analysis of these modals as main verbs. . Cormack and Smith (1998) provides independent evidence that Pol[POS] is present when Pol[NEG] is absent. Arguments include the need to account for the displacement shown in (28b). . Note that in many languages, it is hard to tell apart a modal auxiliary with possible nonfinite forms, and a raising verb. We would take as evidence for auxiliary status (i.e. that the modal is a minor category, not a lexical category) the existence of pre-Pol scope. There does not seem to be any good reason why a verb should not select for Pol, as it may for other functional heads such as C or non-finite T, perhaps. But such a verb would still be under the scope of the clause’s main Pol, on the reasonable assumption that Pol is selected by T, and V is selected by Pol. The existence of NPI modals is also suggestive, since they may be used to express an otherwise unobtainable ‘Pol[NEG] [Modal1 . . .’ scope. . It seems clear from such examples that deontic modals (with the possible exception of ‘ability’ can) do not project an external theta role, but rather that the apparent role is identified pragmatically. The basic compositional type for any modal is thus , allowing a status either as a semantic operator realised as an optional syntactic adjunct, or as a Raising head. . In the very rare cases where it might be, it seems that deontics with scope over epistemics may actually be possible. Consider a context where if the Knave of Hearts feeds Alice some of the liquid in the bottle, she will necessarily become small. The Queen of Hearts might stipulate, with the deontic addressed to the Knave of Hearts, “No, Alice may not have to be small today.” . The example in (75) has no normal interpretation with deontic may. If the sentence is interpretable, the seem, which is itself normally an epistemic predicate has to be reinterpreted as a causative ‘make it seem’, in the same way as be in John is being silly.

References Anderson, S.R. (1992). A-Morphous Morphology. Cambridge: Cambridge University Press. Bailey, B.L. (1966). Jamaican Creole Syntax. Cambridge: Cambridge University Press. Battistella, E.L. (1991). The treatment of negation in double modal constructions. Linguistic Analysis, 21, 49–65. Battistella, E.L. (1995). The Syntax of the Double Modal Construction in English. Linguistica Atlantica, 17, 19–44.

Modals and negation in English

Brody, M. (1995). Lexico-Logical Form. Cambridge MA: MIT Press. Brown, K. (1991). Double Modals in Hawick Scots. In P. Trudgill and J.K. Chambers (Eds.), Dialects of English (pp. 74–103). London: Longman. Brown, K. and M. Millar (1980). Auxiliary verbs in Edinburgh speech. Transactions of the Philological Society, 81–133. Chomsky, N. (1995). The Minimalist Program. Cambridge MA: MIT Press. Cinque, G. (1999). Adverbs and Functional Heads: a Cross-Linguistic Perspective. Oxford: Oxford University Press. Coates, J. (1983). The Semantics of the Modal Auxiliaries. London: Croom Helm. Cormack, A. (1999). Without Specifiers. In D. Adger, S. Pintzuk, B. Plunkett and G. Tsoulas (Eds.), Specifiers: Minimalist Approaches (pp. 46–68). Oxford: Oxford University Press. Cormack, A. and N. Smith (1997). Checking features and split signs. UCL Working Papers in Linguistics, 9, 223–252. Cormack, A. and N. Smith (1998). Negation, polarity, and verb movement. UCL Working Papers in Linguistics, 10, 285–322. Cormack, A. and N. Smith (1999). Where is a sign merged? Glot International, 4(6), 21. Cormack, A. and N. Smith (2000). Head movement and negation in English. Transactions of the Philological Society, 98, 49–85. Ernst, T. (1992). The phrase structure of English negation. The Linguistic Review, 9, 109–144. Espinal, M.T. (1993). The interpretation of no-pas in Catalan. Journal of Pragmatics, 19, 353–369. Fodor, J. (1975). The Language of Thought. New York: Crowell. Gazdar, G., G.K. Pullum, and I.A. Sag (1982). Auxiliaries and related phenomena in a restrictive theory of grammar. Language, 58, 591–638. Haan, F. de (1997). The interaction of modality and negation: a typological study. New York: Garland. Haegeman, L. (1995). The Syntax of Negation. Cambridge: Cambridge University Press. Haegeman, L. and R. Zanuttini (1996). Negative Concord in West Flemish. In A. Belletti and L. Rizzi (Eds.), Parameters and Functional Heads (pp. 117–179). Oxford: Oxford University Press. Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. Cambridge MA: MIT Press. Jackendoff, R. (1997). The Architecture of the Language Faculty. Cambridge MA: MIT Press. Jacobs, J. (1980). Lexical decomposition in Montague-Grammar. Theoretical Linguistics, 7, 121–136. Kayne, R.S. (1998). Overt vs. covert movement. Syntax, 1, 128–191. Klima, E.S. (1964). Negation in English. In J.A. Fodor and J.J. Katz (Eds.), The Structure of Language: Readings in the Philosophy of Language (pp. 246–323). Englewood Cliffs NJ: Prentice Hall. Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1, 337–355. Krifka, M. (1998). Scope inversion under the rise-fall contour in German. Linguistic Inquiry, 29, 75–112.





Annabel Cormack and Neil Smith

Lasnik, H. (1995). Verbal morphology: Syntactic Structures meets the Minimalist Program. In H. Campos and P. Kempchinsky (Eds.), Evolution and Revolution in Linguistic Theory: Essays in Honor of Carlos Otero (pp. 251–275). Washington DC: Georgetown U.P. Lechner, W. (1998). Two kinds of reconstruction. Studia Linguistica, 52, 276–310. Lightfoot, D. (1999). The Development of Language. Oxford: Basil Blackwell. McDonald, C. and J.C. Beal (1987). Modal verbs in Tyneside English. Journal of the Atlantic Provinces Linguistics Association, 9, 43–55. Menaugh, M. (1995). The English modals and established models of probability and possibility: a sign-based analysis. Studia Linguistica, 49, 196–227. Miller, J. and K. Brown. (1982). Aspects of Scottish English Syntax. English Worldwide, 3, 3–17. Nagle, S.J. (1994). The English double modal conspiracy. Diachronica, 11, 199–212. Palmer, F.R. (1990). Modality and the English Modals. London: Longman. Palmer, F.R. (1995). Negation and the modals of possibility and necessity. In J. Bybee and S. Fleischman (Eds.), Modality in Grammar and Discourse (pp. 453–471). Amsterdam: John Benjamins. Palmer, F.R. (1997). Negation and modality in the Germanic languages. In T. Swan and O.J. Westvik (Eds.), Modality in the Germanic Languages (pp. 134–149). Berlin: Mouton de Gruyter. Papafragou, A. (1998). The acquisition of modality: implications for theories of semantic representation. Mind and Language, 13, 370–399. Picallo, M.C. (1990). Modal verbs in Catalan. Natural Language and Linguistic Theory, 8, 285–312. Pollock, J.-Y. (1989). Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry, 20, 365–424. Pullum, G.K. and D. Wilson (1977). Autonomous Syntax and the Analysis of Auxiliaries. Language, 53, 741–788. Quirk, R., S. Greenbaum, G. Leech and J. Swartik (1985). A Comprehensive Grammar of the English language. London: Longman. Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of Grammar: Handbook in Generative Syntax (pp. 281–337). Dordrecht: Kluwer. Roberts, I. (1998). Have/be Raising, Move F, and Procrastinate. Linguistic Inquiry, 29, 113–125. Roussou, A. (1999). Modals and the subjunctive. In A. Alexiadou, G. Horrocks and M. Stavrou (Eds.), Studies in Greek Syntax (pp. 169–183). Dordrecht: Kluwer. Rowlett, P. (1998). Sentential Negation in French. Oxford: Oxford University Press. Rullman, H. (1995). Geen eenheid. Tabu, 25, 194–197. Smith, N. and I-M. Tsimpli (1995). The Mind of a Savant. Oxford: Basil Blackwell. Sperber, D. and D. Wilson (1995). Relevance: Communication and Cognition (second edition). Oxford: Basil Blackwell. Thráinsson, H. and S. Vikner (1995). Modals and double modals in the Scandinavian languages. Working Papers in Scandinavian Syntax 55 (pp. 51–88). Lund University. Vikner, S. (1988). Modals in Danish and event expressions. Working Papers in Scandinavian Syntax 39 (pp. 1–33). Lund University.

Modals and negation in English 

Warner, A.R. (1993). English Auxiliaries. Cambridge: Cambridge University Press. Williams, E. (1994). A reinterpretation of the evidence for verb movement in French. In D. Lightfoot and N. Hornstein (Eds.), Verb Movement (pp. 189–205). Cambridge: Cambridge University Press. Zanuttini, R. (1997a). Negation and Clausal Structure: a Comparative Study of Romance Languages. Oxford: Oxford University Press. Zanuttini, R. (1997b). Negation and verb movement. In L. Haegeman (Ed.), The new comparative syntax (pp. 214–245). London: Longman. Zwicky, A.M. and G.K. Pullum (1983). Cliticization vs. inflection: English n’t. Language, 59, 502–513.

System interaction in the coding of modality* Zygmunt Frajzyngier Department of Linguistics, University of Colorado at Boulder

The purpose of this paper is to show how the coding means available in a language interact in the coding of various modalities. It is shown in this paper that in Lele, an East-Chadic language, verbal inflection, word order, auxiliary verbs and modal particles all participate in the coding of modality. Verbal inflection codes the distinction between the epistemic and deontic modalities: all epistemic modalities are minimally coded by the indicative form of the verb, while all deontic modalities are coded by the imperative form of the verb. Auxiliary verbs and modal particles code various subdomains within epistemic modality, while differing word orders code the different types of deontic modalities.

.

Introduction

The purpose of this paper is to show how the coding means available in a language interact in the coding of various modalities. The terms ‘modality’ and ‘modalities’ are taken to mean the speaker’s attitude or attitudes toward his/her own current utterance (cf. Palmer 1986). Epistemic modality has to do with the speaker’s attitude toward the truth of the proposition. In a given language, submodalities within the epistemic domain may include indicative modality, expressing the speaker’s belief in the truth of the proposition;1 hypothetical modality, coding the speaker’s belief that the proposition may be true; dubitative modality, coding the speaker’s doubt in the truth of the proposition; interrogative modality, coding a question about the truth of the proposition; and negative modality, denying the truth of the proposition. Deontic modality, on the other hand, codes the speaker’s wishes with respect to the proposition; submodalities in this domain may include the imperative, the subjunctive, and the prohibitive modalities.



Zygmunt Frajzyngier

The systems interaction approach involves discovering the functional domains coded in a language and determining how each functional domain is coded. A functional domain is characterised by a functional feature that is common to all of its subdomains. This function may be in the realm of structure, e.g. distinguishing among the arguments of a verb; in the realm of semantics, e.g. coding some element of meaning; or in the realm of discourse, e.g., coding the role of a fragment in a text. Within a given functional domain there are various subdomains, functional categories characterised by formal coding means. Coding means include lexical items, configuration, inflection, auxiliary verbs, adpositions, phonological means or a combination of these; there may exist other, as yet unidentified, coding means. The coding means of the subdomains of a given functional domain interact in complementary fashion.2 The data for the present paper come from Lele, an East Chadic language of the Afroasiatic family, and the discussion is limited to the epistemic and deontic modalities.3 The epistemic modality comprises the following subdomains: indicative, hypothetical, dubitative, interrogative, and negative. The interrogative and negative modalities are not discussed in the present paper, but cf. Frajzyngier (2001). The domain of deontic modality comprises the imperative, subjunctive, and prohibitive modalities. It is shown in this paper that verbal inflection, word order, auxiliary verbs and modal particles all participate in the coding of modality. Verbal inflection codes the distinction between the epistemic and deontic modalities: all epistemic modalities are minimally coded by the indicative form of the verb, while all deontic modalities are coded by the imperative form of the verb. Auxiliary verbs and modal particles code various subdomains within epistemic modality, while differing word orders code the different types of deontic modalities. The final section of the paper shows that epistemic and deontic modality coding may occur in the same clause, providing evidence that the two domains are, in fact, separate.

. Inflectional modal categories In Lele, the indicative and imperative forms of the verb are derived directly from the root, which itself is not marked for modality, rather than from one another. In other words, neither the imperative nor the indicative may be considered basic. This is interesting in view of the fact that, in a great many languages, the unmarked value of the indicative clause is to code the speaker’s belief in the truth of the proposition; given this function, one might expect the

System interaction in the coding of modality 

unmarked, or basic, form of the verb to code the speaker’s belief in the truth of the proposition, yet this is not the case. Deriving the inflected verb form from the root in Lele involves complex phonological processes. I provide a few of the simpler derivations here, in order to illustrate the differences between the indicative and the imperative forms. The starting element in the derivation of the verb is the verbal root. A monosyllabic verbal root may consist of a vowel or a consonant followed by a vowel. If the root is polysyllabic, it consists of all vowels and consonants up to the last consonant. The underlying vowel is selected from the set consisting of a, u, or i. The root may have one of the following structures: Root (C)V(C)(C) ì ‘go’, pi ‘dig’, nì ‘make’4 (C)VC(C)VC(VC) 'ìs ‘pass liquid’ kirb ‘think’

The first step in forming the imperative from the root is to add a vowel suffix after the final consonant. If the root vowel is a or i, the imperative suffix is a; if the root vowel is u, the imperative suffix is u. The final vowels a and u are subsequently deleted from polysyllabic verbs if the last consonant of the verb can be word final. The formation of the imperative stem is summarised as follows: Step 1: Imperative stem = Root + a/u Step 2: (polysyllabic verbs) Final vowel deletion (except when disallowed)

The first step in deriving the indicative stem from the root is vowel lowering, which affects only roots having a high vowel. Next, the sequence C1 C1 is reduced to C. After these steps, the past tense suffix i is added to the root. If the root contains a high round vowel, this vowel is fronted. The past stem vowel is deleted from any verb whose root-final consonant may occur in word-final position. The indicative stem is the source from which the nominal and future forms of the verb are derived. The nominal form is derived by lowering the past stem final vowel i to e. The future tense form is derived by raising to high the first tone of the nominal form. Following are examples of surface forms derived from monosyllabic and bisyllabic roots with a high round vowel:



Zygmunt Frajzyngier

Monosyllabic roots Root Indicative/Past

Gloss

Future Nominal

Imperative

ì pi nì

‘go’ ‘dig’ ‘make’

éè péè né

à pa nà

è pe nè

è pe nè

Polysyllabic roots Root

Indicative/Past

Gloss

Future Nominal

Imperative

'ùs kurb gùrb dùgr kunjl wunjr

'ìsì kirbi gìrbì dìgrì kunjili winjiri

‘pass liquid’ ‘think’ ‘forget’ ‘kill’ ‘learn’ ‘rot’

'ísè kírbe gírbè dígrè kúnjíle wínjíre

'ùsù kurbu gùrbù dùgrù kunjulu wunjuru

'ìsè kirbe gìrbè dìgrè kunjile winjire

Following are examples of derived forms based on roots having the vowel a and a consonant or consonant cluster that may not occur in word-final position. These verbs have the suffix a in the imperative and the suffix i in the indicative/past form in word-final position: Root

Indicative/Past

Gloss

Future

Nominal

Imperative

sàg bàs das´ màgl´

sàgì bàsì dasí màglí

‘tear’ ‘fail, spend’ ‘husk’ ‘repair’

ságè básè dáasé máàglé

sàgè bàsè dasé màglé

sàgà bàsà dasá màglá

The suffixes of the indicative and imperative stems are deleted if the final consonant of the stem may occur in word-final position. All verbs affected by this rule end in l, m, n, or b, consonants that elsewhere are allowed to occur in word-final position. Past

Gloss

Future

Nominal

Imperative

gìl ìm im jìb wìl gul kil 'íl kìn

‘show’ ‘bury’ ‘rot’ ‘stick’ ‘glean’ ‘watch’ ‘sell’ ‘drag’ ‘return’

gílè ímè íme jíbè wílè gíle kíle 'íìlé kínè

gìlè ìmè ime jìbè wìlè gile kile 'ìlé kìnè

gùl ùm um jùb wùl gul kul 'ùl kùn

System interaction in the coding of modality 

wìl´ hím

‘warm up’ ‘gather, tr.’

wílè hímé

wìlé hìmé

wùl húm

Verbs that have a consonant cluster in the root may not delete the final vowel, hence the indicative/past and imperative forms of the roots kunjl ‘learn’ and wunjr ‘rot’ are kunjili and kunjulu; and winjiri and wunjuru. The proposed derivations account for all forms of the verbal paradigm. The vowel reduction rules in the indicative and the imperative are not motivated phonologically because the future and the nominative forms of the verb do not reduce their final vowels. Thus the vowel reduction is a coding means in addition to the suffixes i, u and a. The final vowel retention in the indicative and the imperative is, however, motivated by the phonological constraint on the types of consonants and consonant clusters that may occur in word final position. Note that stem reduction is one of the characteristics of deontic inflectional coding in Semitic languages, which constitute another branch of the Afroasiatic family. . Epistemic modality All constructions with epistemic modality are based on the indicative form of the verb. Some submodalities in this domain also are marked by auxiliary verbs or modal particles. . Indicative modality: Coding belief in truth The least-marked submodality of the epistemic domain is the indicative modality. A clause has indicative modality if it has no modal particles, adverbs, or auxiliary verbs, and if it contains an indicative form of the verb. In the Lele indicative clause, a nominal subject occurs before the verb: (1) kudo-ro-] jè 1e]lí kolo ká 'nyé kwáni ná [pause] yàá husband-3-  hear word certain: outside  say ísa] 1é nothing  ‘Her husband heard certain things outside, but he did not say anything.’

First- and second-person pronominal subjects also occur before the verb: (2) ] ómè-y 1 catch:-3 ‘I will catch him.’



Zygmunt Frajzyngier

(3) gi 'álè go òmé-y 2 suffice:  catch:-3 ‘You will be able to catch it.’

A third-person pronominal subject, if overtly coded, occurs after the verb:5 (4) càg dí kùbàrò kúsìyo] spread 3 blood ground ‘He spread blood on the ground.’ (5) tamá tolo láy dú woman  pretty 3 ‘That woman is pretty.’ (6) gol gé dú see 3 3 ‘They saw her.’ (7) gol dú gé see 3 3 ‘She saw them.’

A third-person subject need not be coded overtly if it is the same as the subject of the preceding clause. In the following sentence, the first instantiation of the third-person masculine subject is unmarked, since this subject is the same as that of the preceding clause (not shown here). The second instantiation of the third-person masculine subject is marked overtly because another subject, the third-person feminine, interferes between the two instantiations of the thirdperson masculine (cf. Frajzyngier 1997): (8) 1e-du ja] jè wèl dú kuná ná an dí è om leave-3 quietly  sleep 3 sleep  leave 3 go catch dùbàngú yó-ró ná mundu wàl sheep mother-3  silence slaughter ‘He left her quietly while she was asleep, left the room, silently caught an old sheep and slaughtered it.’

In Lele, as in many other languages, the unmarked form of the clause codes the speaker’s intention to convey belief in the truth of the proposition (cf. Frajzyngier 1985). One could argue, as Palmer (1987) does, that truth is not the unmarked value of the indicative modality, because there exist other means of overtly coding a speaker’s belief in the truth of the proposition. Lele itself has such means, e.g. the particle wa] ‘really, certainly’:

System interaction in the coding of modality

(9) kírbì hab kùlbá kò-rò wa] Kirbi found cow -3 really ‘Kirbi has certainly found her cow.’

Such means are only used, however, if someone has previously denied the truth of the proposition or expressed doubt about the truth of the proposition (cf. Frajzyngier 1987). Language-internal evidence for the modal function of indicative modality is provided by the fact that any other modal function, such as doubt as to truth, questions about the truth, negation of truth, and so on, must be overtly coded. Another piece of evidence for the function of the indicative clause is that an indicative clause whose subject is the speaker cannot be followed by a dubitative marker. The dubitative marker sá], freely used to code the speaker’s doubt as to the truth of another person’s statement, cannot be used in a clause whose subject is the speaker (cf. also Frajzyngier 1985); compare (10) with (11). (10) a. *] nè hómyà sá] 1  sick  Intended meaning: ‘I am sick [but I doubt it].’ b. *] yá na ] nè hómyà sá] 1 say  1 make sick  Intended meaning: ‘I said that I am sick [but it isn’t true].’ (11) a.

nè dí hómyà sá] make 3 sick  ‘He pretends to be sick.’ b. na-y hómyà sá] -3 sick  ‘He said that he is sick, but I doubt it.’

. Hedging on truth A number of auxiliary verbs allow the speaker to hedge about the truth of a statement. One means involves the verb tòb ‘want’, followed by the complementiser go and then by the main verb with its complements. The complementiser go is glossed REF for ‘referential’ in order to distinguish it from the de dicto complementiser na which is used as the hypothetical modality marker. The marker na is glossed as COMP if its function is to code the complement of verb of saying, and it is glossed as HYP if its function is to code hypothetical modality regardless of the verb of the matrix clause. The coding of comple-





Zygmunt Frajzyngier

ments of verbs of saying and of hypothetical modality are related functions, as argued for in Frajzyngier (1991). (12) kus-iy jè tob go né-y body-3  want  make-3 ‘He seems to be sick.’

Cf. the indicative clause: (13) kus-iy nè-y body-3 make-3 ‘He is sick.’ (lit. ‘his body makes him.’)

The speaker may hedge about the truth of his or her own statement by using the verb nom ‘to pretend, to court’, which forms a serial verb construction with the following verb. This means occurs in a few expressions: (14) kus-íy nom né-y body-3 court make-3 ‘He seems to be sick.’ (15) tùwa nom dóbe í (índì) màní sun pretend set.down with.him there ‘The sun seems to lie down there.’

Compare the use of nom as a main verb: (16) cànìgé nom kírbí Canige court Kirbi ‘Canige courts Kirbi.’

The adverb brùm ‘hardly’ expresses the speaker’s uncertainty regarding his or her perceptions: (17) kanya nomí-y brùm something pretend-3 hardly ‘There is something wrong with him, I believe.’ (18) ] gol kur brùm 1 see space hardly ‘I can hardly see. ’

Doubt about the truth of the speaker’s perceptions or evaluations may also be coded by the modal adverb wa] ‘really’ followed by the conditional gèrè ‘if ’: (19) kírbì hab kùlbá kò-rò wa] gèrè Kirbi found cow -3 really if ‘Kirbi has apparently found her cow.’

System interaction in the coding of modality 

. Hypothetical modality Hypothetical modality is coded by the hypothetical marker na preceding the predicate: (20) kiya na bòy kòjò kè-y Kiya  break hoe -3 ‘It is possible that Kiya broke his hoe.’

If the subject is a third-person pronoun, which follows the verb in the indicative clause (see above), the complementiser na precedes the verb: (21) na bòy dí kòjò kè-y  break 3 hoe -3 ‘It is possible that he broke his hoe.’

With a first- or second person pronominal subject, the hypothetical marker precedes the subject pronoun: (22) na ] bòy kòjò kò-nò]  1 break hoe -1 ‘It is possible that I broke my hoe.’ (23) na gi bòy kòjò kò-m  2 break hoe -2 ‘It is possible that you broke your hoe.’

Compare the following fragment from natural discourse with its counterpart from which the hypothetical marker is omitted. The third person subject is the topic, hence the presence of the pronoun dày at the beginning of the clause: (24) dày ná na nè tìbré mayna 3    ashes ceiba tree ‘That person would be like ashes of a ceiba tree.’ (25) dày ná nè tìbré mayna 3   ashes ceiba tree ‘That person is like ashes of a ceiba tree.’

Hypothetical modality may be coded in matrix and embedded clauses. The verb following the hypothetical marker may be in the future or the imperative form. The future form of the verb combined with the hypothetical marker codes potential events:



Zygmunt Frajzyngier

(26) dày kó] na-y kálè jénè dà lu]ba kùrmbàlò 3 certain -3 climb: sit:  horse chief kè-y ni na-y bórè sàríyà na ùs-ìy 1é -3  -3 cut: verdict  concern:-3  ‘The other [said] that if he could climb on the chief ’s horse and pronounce judgments, [they could kill him], he would not care.’ ` ni ná ge (27) Ore na wè-gé dà gìdìrè kòlòng yàá goat:  give birth-3  moon :    tell na karan-do gàn-di-gè na wèy já.  children:3 leg-.-3  become side ‘It is said that if goats give birth in that month, their kids will have crooked legs.’ (G-C & W 1981: 54–55)

Hypothetical modality codes an event as not being real. This may apply to the truth value of the clause: (28) kur wèl kay 1é 'ey ná tamá na du è sógú ni place pass finish  still  woman  3 go toilet  ‘Before the day ended [while it was still dark] the wife pretended that she was going to the toilet.’

Hypothetical modality may also indicate that an event was intended but did not occur: (29) kur go gúnyé ágì-y jè na-y è jéèé-y dà time  spider take:-3  -3 go throw-3  kama-] ni dày kàyo-] se an ná galmbo kíin-dì water-  3 squirrel-  leave  bag hole-3 bà se gìr   run ‘When Spider was about to take him to throw him into the water, Squirrel left through the hole in the bag and ran away.’ (30) bayndí go na ay tu ká] na lèé ná person   take goat   eat  ‘The person who would take this goat and would eat it.’ (31) kàyo se yàá na na-y sìnà máàní na-y 1e squirrel  say  -3 know:  -3 leave gúnyé na ge ímè-y ná làmndá yé-y 1é ba spider   bury-3  elephant mother-3  

System interaction in the coding of modality

na-y bòr gúnyé gèyl-ìy ba gúnyé na è jè tòb go -3 save spider health-3  spider  go  want  dígrè-y gà kill:-3  ‘Squirrel said that if he had known that, he would have left Spider to be buried together with the mother of elephant. Here I saved him, and doesn’t he want to kill me?’ (32) póì kúsíge-] se ma'y na-y kòy tèmlé-] 1é na-y Poi Kusige-  deny -3 steal corn-  -3 àn kwáni ná kúmno kam-di na 1é leave outside  God water-3   ‘Poi Kusige denied that he stole the corn. He claimed that he did not go out in the rain.’

The hypothetical marker na after a volitional verb also codes a hypothetical event. Here, the marker must be followed by a third-person subject pronoun: (33) wèlè pìnà ná gèy na-y gísè-dù day one  want (Ng.) -3 try:-3 ‘One day he wished to test her.’

If the complementiser is absent, the meaning of the clause is non-hypothetical: (34) wèlè pìnà ná gèy gísè-dù day one  want (Ng.) try:-3 ‘One day he wanted to test her.’

Natural texts clearly show that speakers try to avoid making categorical statements when explaining natural or cultural phenomena. Instead of using a straightforward indicative sentence, the speaker presents the statement as the indirect speech of an unspecified human subject, coded by ge ‘one’ or kara ‘people’. Here, the matrix clause has the form ge yàá or kara yàá ‘one says/people say’. In the texts gathered by Garrigues-Cresswell, all explanations of human or animal behavior or atmospheric phenomena have this form. The complement of the matrix clause is marked as being hypothetical by the particle na, followed by the indicative form of the verb: (35) Tùmò go kar di-nì past time  people .-1. nè 'òdu ba gèylì-nì kur  monkey  save-1. time

kínyé kè-gè ge yàá na  -3  say  go kamda ná kùlòngò.  women  pregnancy





Zygmunt Frajzyngier

‘At the time of our ancestors, they say, it was the monkey who saved us at the time when women were pregnant.’ (G-C & W 1981: 2–3)

The scope of the hypothetical marker na depends on its syntactic position. If the hypothetical marker precedes the subject, the whole proposition is in its scope, as illustrated in the next sentence from the same fragment: (36) Kumno yàá na, na láràdí ba è hab-i tùmò na yàá God said   chameleon  go find-3 first  tell bè-ì kolo kè-ì -3 word -3 ‘God said: if it is chameleon who comes first and tells him his word.’ (G-C & W 1981: 6–7)

. Dubitative modality Another subdomain of epistemic modality is dubitative modality, the grammatical coding of the speaker’s doubt concerning someone else’s claim. This is coded by the clause-final particles sá], glossed as DUB for ‘dubitative’, and gèrè ‘if ’. These particles may occur alone or in the sequence sá]gèrè. Dubitative markers occur with verba dicendi complements but they can also occur with other verbs. The dubitative marker denies the truth of someone else’s claims, as in (37) and (38), to be compared with (39). (37) nè dí hómyà sá] make 3 sick  ‘He claims to be sick.’ (38) na-y hómyà sá] -3 sick  ‘He said that he is sick, but I doubt it.’ (39) kus-iy nè-y body-3 make-3 ‘He is sick.’ (lit. ‘his body makes him’)

It may be recalled that the dubitative marker cannot be used with a first-person singular subject. The dubitative marker also cannot be used in equational sentences: (41) cànìgé nè lele Canige  Lele ‘Canige is Lele.’

System interaction in the coding of modality 

Cf. (42) *cànìgé nè lele sá] Canige  Lele  Intended meaning: ‘Canige is Lele, but I doubt it.’

However, if the equational clause is a complement of a verb of saying, the dubitative marker is acceptable: (43) cànìgé ná-y no go lele sá] Canige -3   Lele  ‘Canige said that he is a Lele man, but I have my doubts.’

. Deontic modality The domain of deontic modality has three subdomains, the imperative, the subjunctive, and the prohibitive, which is the negative imperative. These are grouped together because they all use the imperative form of the verb. The difference between the imperative and the subjunctive subdomains is coded syntactically, by the position of pronominal subjects with respect to the verb. . The imperative The syntax of giving orders differs from the syntax of indicative modality. In the imperative, the subject pronoun for the second-person plural follows the verb, while in the indicative, the second-person subject precedes the verb. Complements, if any, follow the second-person singular subject marker: (44) pàmà ngú search: 2 ‘Search!’ (45) kùlù ngú kùlbá buy: 2 cow ‘Buy a cow!’ (46) da ngù kasa ná já kolo] eat: 2 corn  side : ‘Eat the corn at the other side.’

A second-person singular subject in the imperative may be unmarked:



Zygmunt Frajzyngier

(47) kùl kùlbá buy: cow ‘Buy () a cow.’ (48) À-je ná bè-ng gúndù pìnà ná kama, ná gìlàdí go:-  -1 gourd one  water  stalk mìsé, à-je haba-ng ná dà sóòyó to-roò sorghum go:- find-1   tamarind bottom-3 ni.  ‘Bring me a gourd with water and stalks of sorghum. Come to find me under the tamarind tree.’ (G-C & W 1981: 2–3)

If the second-person singular subject is marked, it is marked by the independent form of the pronoun in the position before the verb: (49) Dàmè 1inglá 2 listen: ‘Listen!’

(G-C & W 1981: 2–3)

. The subjunctive The term ‘subjunctive’ refers to wishes with respect to the first and third person as well as non-imperative wishes with respect to the second person. The syntax of the subjunctive differs from that of both the indicative and the imperative. In a subjunctive clause with a first-person plural inclusive subject, the subject pronoun ngà follows the imperative verb. Any complements follow the subject: (50) yà-ngá kolo go1à] talk:-1. word another’ ‘Let us talk about something else.’ (51) ìrà-ngá go:-1. ‘Let us go!’ (52) kà[w] ngà túg kò-m ni gólè kùní kùlè-ndì go: 1. home -2  see: home interior-3 ‘Let’s go to your home [and] we will see the interior of the house.’

In a subjunctive clause with a third-person subject, the subject pronoun precedes the verb. In clauses with indicative modality, it may be recalled, the third-person pronoun follows the verb:

System interaction in the coding of modality 

(53) dí ìrà 3 go: ‘Let him go.’ (54) dú ìrà kàsúgù 3 go: market ‘Let her go to the market.’ (55) dí yàgà bé-] kúsó 3 plough: -1 field ‘Let him plough my field for me!’

If a subjunctive clause has a third-person nominal subject, this subject is followed by a third-person pronoun agreeing with the nominal subject in gender and number: (56) kiya dí à jè Kiya 3 go:  ‘Let Kiya come.’ (57) kiya dí ìrà Kiya 3 go: ‘Let Kiya go.’ (58) kiya dí yàgà bé-] kúsó Kiya 3 plough: -1 field ‘Let Kiya plough my field for me!’ (59) yàá bè-ì na gìdìrè na ma ná, ìrà ná kìrè tell -3  moon  die:  go:  road kíndi kara índùwé ma gé ná ana gé jè. truly people human: die: 3  leave: 3  ‘[he] told him that the moon should die and never return, and the people should die and return.’ (G-C & W 1981: 8–9)

. Prohibitive modality Prohibitive modality is formed through the placement of the negative particle 1é at the end of a clause containing the imperative form of the verb: (60) Tamá, ul 1é. woman, cry:  ‘Woman, do not cry!’

(G-C & W 1981: 2–3)



Zygmunt Frajzyngier

(61) ladá bìsí kò-nò] ná kà-m 1é touch: antelope -1  hand-2  ‘Do not touch my antelope with your hands!’ (62) ìrà 1é go:  ‘Don’t go!’ (63) à jé-ngú 1é go: :-2  ‘Don’t come!’

The prohibitive predicate may have two verbs in a sequence, both of them marked for imperative modality: (64) 1á wala kusu-m bàb 1é leave: swear: body-2 all  ‘Do not swear for nothing.’

. Coding hypothetical and deontic modalities in the same clause The hypothetical marker na, which codes the hypothetical subdomain of the epistemic modality, may co-occur with the imperative form of the verb, marker of the domain of deontic modality. This construction codes both hypothetical modality and the modality of obligation, providing evidence that epistemic and deontic modalities are two separate domains. The possible combinations of na with the future i.e. a form of indicative modality and imperative forms of the verb, are illustrated in the following fragment: (65) Kumno se tey láràdí dàgè ná gúló yàá bè kara God  call lizard 3  chameleon tell  people ‘God called Lizard and Chameleon and he told people.’ (66) na-y ódè wanyang wáyngulú ná láràdí na -3 leave: tomorrow morning  lizard  ‘that he would go tomorrow morning and that Lizard . . . ’

In the subsequent clause, the subject gìdìrè ‘moon’ is followed by the hypothetical marker na, followed by the verb in the imperative:

System interaction in the coding of modality

(67) nè kara go yà gé na gìdìrè na ma ìrà ná make: people  say 3  moon  die: go:  kìrè kíndi kègè, road truly :3 ‘would be [a spokesman] for people who say that the moon should die and go away forever.’ (68) gúló na nè kara go yà gé na kara chameleon  make: people  say 3  people índùwé ba na ma gé ìrà gé ná kìrè kíndiè human:   die: 3 go: 3  road truth kè-gè lay. -3 also ‘Chameleon would be a spokesman for people who say that humans should die and go away forever.’

. Conclusions Epistemic modality is coded by the indicative form of the verb. Subdomains within the epistemic modality domain are coded by auxiliary verbs and particles. Deontic modality is coded by the imperative form of the verb. Subdomains within the deontic modality domain are coded by the order of pronouns with respect to the verb. In the imperative modality, the second-person plural subject pronoun (obligatory when the addressee of the imperative is plural) follows the verb, while in all epistemic modalities, the second-person subject pronoun precedes the verb. Third-person pronouns in the subjunctive modality precede the verb, while first- and second-person pronouns follow the verb; this order is the reverse of that found in clauses with epistemic modality. Evidence that epistemic and deontic modalities are distinct domains is provided by the fact that they can be combined within the same clause.

Abbreviations 1 2 3 

First-person Second-person Third-person Allative





Zygmunt Frajzyngier

           Fr.   :          Ng.          

Anaphor Associative. Used for associative preposition ná and associative pronominal marker ín-. Collective Comment clause marker ba Copula Dative marker Definite Demonstrative Dubitative Exclusive Feminine French Future form of the verb Genitive marker kè (M) or tè (F) Genitive plural marker dí Unspecified human subject Hypothetical marker and complementiser na Imperative Imperfective Inceptive Inclusive Locative postposition Masculine Negative marker Ngambay (a Niger-Congo language) Noun Phrase Proximate Plural Preposition Pronoun Interrogative marker Remote Referential marker go, also masculine and plural relative clause marker Singular Verbal noun

System interaction in the coding of modality 

Notes * The work on Lele has been supported by grants from the National Endowment for the Humanities, the National Science Foundation, the American Philosophical Society, and the University of Colorado. The work on this paper was completed while I was a guest of the Max Planck Institute for Evolutionary Anthropology in Leipzig. I am most grateful to these institutions for their support. I am also grateful to Erin Shay and two anonymous referees whose comments on both the content and the form of the paper were most helpful. I alone am, however, responsible for any mistakes and infelicities in the analyses and conclusions. . Palmer (1986) does not recognise the modal function of the indicative clause. Frajzyngier (1985) was the first explicit claim and proof with respect to the function of the indicative sentence. While it is true that there are languages in which the speaker’s assertion of the truth of the proposition is a marked modality, this is not the case in Lele, the language on which this paper is based. . For a formal definition of the notion of functional domain, cf. Frajzyngier and Mycielski (1998). . The Lele data come from my own recordings and from texts published in GarriguesCresswell and Weibegué (1981). If the source of the data is not mentioned, this indicates that the data come from my own fieldnotes, which consisted in most cases of recording natural discourse. Data from Garrigues-Cresswell and Weibegué (1981) are identified by the abbreviation G-C & W 1981, and are cited as in the original. . The tonal marking adopted in the present paper is as follows: grave ` codes low tone, acute ´ codes high tone, and the absence of the tonal marking codes middle tone. . The rules concerning the function of the third person pronominal subject involve interaction with the nominal subject, and the pragmatic functions of the subject.

References Frajzyngier, Z. (1985). Truth and the indicative sentence. Studies in Language, 9, 243–254. Frajzyngier, Z. (1987). Truth and the compositionality principle: A reply to Palmer. Studies in Language, 11, 211–217. Frajzyngier, Z. (1991). The de dicto domain in language. In E. Traugott and B. Heine (Eds), Approaches to Grammaticalization, Vol. 1 (pp. 219–251). Amsterdam: Benjamins. Frajzyngier, Z. (1997). Pronouns and agreement: Systems interaction in the coding of reference. In H. Bennis, P. Pica, and J. Rooryck (Eds), Atomism and Binding (pp. 115– 140). Dordrecht: Foris. Frajzyngier, Z. (2001). A Grammar of Lele. [Stanford Monographs in African Linguistics]. Stanford: CSLI. Frajzyngier, Z. and J. Mycielski (1998). On some fundamental problems of mathematical linguistics. In C. Martin-Vide (Ed.), Mathematical and Computational Analysis of Natural Language (pp. 295–310). Amsterdam: Benjamins.



Zygmunt Frajzyngier

Garrigues-Cresswell, M. and C. Weibegué (1981). Livre de lecture lélé. Sarh: Centre d’études linguistiques. Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press. Palmer, F. (1987). Truth Indicative? Studies in Language, 11, 206–210.

Modality and theory of mind Perspectives from language development and autism* Anna Papafragou University of Pennsylvania It is widely assumed in the developmental literature that certain classes of modal readings appear later in language acquisition than others; specifically, epistemic interpretations lag behind non-epistemic (or root) interpretations. An explanation for these findings is proposed in terms of the child’s developing theory of mind, i.e. the ability to attribute to oneself and others mental representations, and to reason inferentially about them. It is hypothesised that epistemic modality crucially implicates theory of mind abilities and is therefore expected to depend on prior developments in the child’s ability to handle representations of mental representations. In support of this hypothesis, it is shown that autistic individuals (who arguably possess a deficient theory-of-mind mechanism) have difficulty with epistemics.

.

Introduction

It has long been recognised that modal items in natural language are contextdependent expressions. A sample of possible interpretations for some English modal verbs is provided by the glosses below:1 (1) He must be back before dark. a. ‘He is obliged to be back before dark’. b. ‘He will certainly be back before dark’. (2) The test should not take longer than 30 minutes. a. ‘It is recommended that the test does not take longer than 30 minutes’. b. ‘The test is not likely to take longer than 30 minutes’. (3) Students may use the sports facilities. a. ‘Students are allowed to use the sports facilities’. b. ‘It is possible that students will use the sports facilities’.



Anna Papafragou

(4) John has to have the right solution; he’s the expert. a. ‘John is obliged to have the right solution’. b. ‘It is certain that John has the right solution’. (5) The bank will give you a new credit card. a. ‘The bank intends/is willing to give you a new credit card’. b. ‘I predict that the bank is going to give you a new credit card’. (6) I can ride a bicycle. a. ‘The circumstances make it possible for me to ride a bicycle’. b. ‘I am able to ride a bicycle’.

Such differences in interpretation are standardly characterised in the literature in terms of the distinction between root and epistemic modality. Root modality includes obligation, permission and related notions (traditionally known as deontic modality), as well as a family of concepts dealing with intention/willingness and ability (traditionally known as dynamic modality): examples (1a), (2a), (3a), (4a), (5a) and (6a, b) above are typical cases of root modality. Epistemic modality involves inference from known premises: the examples in (1b), (2b), (3b), (4b), and (arguably) (5b) fall under the category of epistemic modality. Root and epistemic modality are based on different premises. As Angelika Kratzer puts it, ‘if we use an epistemic modal, we are interested in what else may or may not be the case, given everything we know already’; on the other hand, with non-epistemics, ‘we are interested in what can or must happen, given circumstances of a certain kind’ (Kratzer 1981: 52). An interesting fact, which has been repeatedly pointed out in the psycholinguistic literature, is that root meanings precede epistemic ones in language acquisition. Since in many languages the same class of modal expressions is used to convey both root and epistemic meanings (as is the case with the English modals above), the claim is that modal items initially appear in child language with root interpretations only and their epistemic interpretations are acquired markedly later. In this paper, I survey the evidence (both longitudinal and experimental) standardly adduced to support the acquisitional priority of root over epistemic modal meanings. My main aim is to motivate a specific proposal about the emergence of epistemic modality in language development. According to this proposal, the appearance of genuine epistemic modal uses is closely tied to the development of the child’s theory of mind, that is, the ability to understand mental phenomena and to attribute mental states to oneself and to others (see also Papafragou 1998, 2000). I explore the explanatory potential of this view with regard to the existing cross-linguistic data on the ac-

Modality and theory of mind 

quisition of modality and relate these results to developmental findings from mental verbs such as think, or believe. Finally, I show how certain data from autism provide support for the proposed analysis of epistemic modality.

. The acquisition data The set of modal data for which we have the most extensive longitudinal studies is probably the class of English modal verbs. According to Shatz and Wilcox (1991), the first uses of the modals in English appear between the end of the second and the middle of the third year, initially in restricted syntactic environments (mainly declaratives) and often with a single negative form such as can’t. Gradually, both the range of constructions in which modality appears and the type of modal item used become more varied. According to available production data, English-speaking children quite early on express intention/desire and ability with will and can, respectively. In Wells’ (1979) sample (which consisted of sixty children time-sampled along with their mothers every three months from one year and three months to three years and six months), it was found that by two and a half years more than half of the sample used can to convey both ability and permission. By the same time, children used will to communicate intention. In her studies of a single child, Shepherd (1982) has found that will extends from volition to prediction between two years and five months and three years. Will refers to more distant future or events which lie beyond the child’s control, while gonna takes up the space of events in the immediate future which the child controls (cf. Gee and Savasir 1985). Similarly, Gerhardt (1991) reports on the use of hafta, needta and wanna in the speech of two three-year-olds and finds that the three modal-like verbs have root uses: hafta is used to convey compulsion based on a norm/external source, needta communicates compulsion based on an internal source, and wanna communicates volition. Later on, a fuller repertoire of root uses comes into play. Between two years nine months and three years, the children in Wells’ (1985) sample used must, have (got) to and should to communicate obligation and (root) necessity, although these uses stabilised later in development. As Wells (1985) reports, by the beginning of the third year all categories of root modality were attested in his sample. In contrast, the first statements with epistemic modals (must and may) occur in the second half of the third year, i.e. about six months later than root meanings. Furthermore, epistemically modalised utterances are extremely rare until the middle of the fourth year (Stephany 1986, 1993 but cf. O’Neill and



Anna Papafragou

Atance 2000). In Wells’ (1985) population, may and might were used with an epistemic meaning only by three years and three months of age. Use of modals to convey certainty emerged even later, since by five years only 25% of the sample gave evidence of it. Certain epistemic uses of will (as in That will be John, uttered on hearing the doorbell) appear even later than epistemic uses of must. Other research suggests that the development of the epistemic modal system, especially as far as adverbs and adjectives are concerned, continues even between six and twelve years (Perkins 1983). Young children’s comprehension of modal expressions has also been tested experimentally. Hirst and Weil (1982) gave children between three and six two different modal propositions of varying ‘strength’ (e.g. with a possibility vs. a necessity marker). In the epistemic cases, the propositions concerned the location of a peanut. In the root cases, they were commands by two teachers about the room a puppet was to go to. The child was to indicate in the first case where the peanut was, and in the second where the puppet would go. The strength ordering assumed was is > must > should > may. The general result was that children appreciate the relative strength of epistemic modal propositions about a year earlier than root modal propositions (five vs. six years approximately); moreover, the greater the distance between the modals, the earlier the distinction is appreciated. It should be pointed out that these results cannot be interpreted as evidence against previous research documenting the developmental priority of root over epistemic interpretations of modals. Children’s performance in the deontic tasks was probably influenced by factors other than the relative strength of the modals (for instance, the evaluation of the authority of the persons issuing the command, or the puppet’s compliance with the rules). Still, this paradigm shows that children have a good grasp of differences between modals within each major modal class (root or epistemic) by the age of five or six. Hirst and Weil’s original tests were replicated by Noveck, Ho and Sera (1996) with five-year-olds and Byrnes and Duff (1989) with three- to five-year-olds. There is some evidence that the precedence of non-epistemic meanings of modal expressions in acquisition is attested in other languages as well. Stephany (1993) reports findings which show that, in German child language, wollen (‘will’) and können (‘can’) are the first modals to appear. Moreover, she shows that, when müssen (‘must’) and sollen (‘should’) appear, they have root rather than epistemic interpretations. In Polish, expressions of root modality appear before two and are used with increasing frequency during the third year to communicate obligation/root possibility and prohibition/permission. No epistemic uses of modals are attested for this period, although during the second

Modality and theory of mind 

year certain epistemic modal particles emerge – e.g. chyba (‘probably’), na peuno (‘for sure’) (Smoczynska 1993). Data from Mandarin Chinese also follow the developmental ordering of dynamic, deontic and epistemic modality (Guo 1994). Modal data from Modern Greek, Finnish and Turkish seem to pattern in similar ways: Stephany (1986) notes that, in these languages, epistemic interpretations appear well after the third birthday. She also points out that a variety of grammatical means for marking epistemic modality other than modal verbs (e.g. the conditional in Finnish, the aorist inflection in Turkish) coincide around the third year. Even though observational data are not by themselves sufficient to obtain an accurate picture of development, they are a useful point of departure in discussing the acquisition of modality. In the next section, I present the beginning of an explanation for the acquisition facts that draws on the connection between epistemicity and the child’s developing ability to reason about the mind.

. Theory of mind: An overview One of the most characteristic properties of (adult) human beings is the ability to attribute mental states to oneself and to others, and to use information about human cognition and motivation in understanding and predicting human action. In other words, adults generally assume that people have a mental life of interconnected beliefs, desires, intentions, ideas; they also assume that people’s observable behaviour stems from and is best understood in terms of the mental states which lie behind overt behaviour. The ability to impute mental states to oneself and others leads humans to try and use the mind and increase its powers, to share inner experiences, to distinguish between imaginary and real events, and to interact with other persons by searching for and reaching out to their underlying mentalities (Butterworth, Harris, Leslie and Wellman 1991a). This everyday conception of the mind (or ‘folk psychology’) is commonly referred to within recent cognitive science as theory of mind (see the contributions in Astington, Harris and Olson 1988; Whiten 1991; Butterworth, Harris, Leslie and Wellman 1991b; Frye and Moore 1991; Hirschfeld and Gelman 1994; Carruthers and Smith 1996). Although knowledge of ‘other minds’ has been one of the traditional preoccupations of philosophy, current interest in folk psychology resurfaced after some researchers raised the question of whether certain primates might also possess and use (some form of) mindreading (Premack and Woodruff 1978). This question forced philosophers and psychologists to think more clearly



Anna Papafragou

about what is involved in having a theory of mind and to consider cases in which the workings of theory of mind are either absent or very different from the normal adult case (including the cognition of animals, children and atypical individuals). Recent years have seen a flood of interdisciplinary research on theory of mind, mainly centred around two main issues: a. What is the nature, and what are the characteristic properties, of normal adult folk psychology? b. How do folk-psychological abilities develop in children? Both issues are surrounded by considerable debate. For instance, adult folk psychology has been analysed as a module underwritten by an innate mechanism (Leslie 1988), a truly theory-like object constructed through general-purpose learning capacities (Gopnik and Wellman 1995), or a device for projecting oneself onto other people, thereby simulating their mental states (Gordon 1986). These differences in perspective with regard to the final form and deployment of theory of mind carry over to developmental work: studying what the intermediate stages in folk-psychological development look like typically presupposes a stance on the structure and function of adult folk psychology. In what follows, I present certain core facts about children’s performance in theoryof-mind tasks without adjudicating between competing theoretical proposals. The description is meant to highlight some relatively uncontroversial data concerning children’s performance. It is also meant to give some background for discussing the data from the acquisition of modality in later sections. There is some evidence for early understanding of mental life by two years of age. Very young children are able to pretend and to engage in pretence alongside others (Leslie 1988). They also have an understanding of the causal links between desire or perception and the world: for instance, they know that desires can be used in predicting actions (Wellman and Woolley 1990). Two-yearolds also know that they can produce perceptions in others by showing things to them, and by three they can also deprive others of perception by hiding things (Yaniv and Shatz 1988). By the age of three, children can distinguish between real and mental entities. For instance, they know that, when someone is thinking about a cookie, that cookie cannot be eaten, shared, and so on – unlike the situation in which someone actually has a cookie (Wellman and Estes 1986). They also develop a better understanding of the connection between perception and belief fixation. They realise, for instance, that if an object is hidden in a box, someone who has looked in the box knows what is in there, and someone who has not looked does not know (Pratt and Bryant 1990).

Modality and theory of mind

Between three and four or five years children acquire the ability to pass the so-called false belief task. The essence of the task is to check whether children realise that other people may have beliefs which differ from their own. This is important in theory-of-mind development, since it is a reliable way of testing whether children actually attribute beliefs to others or simply assume that others share their own beliefs. In the classic version of the false belief task (Wimmer and Perner 1983), one character puts an object in some place and goes away. In his absence, the object is moved to another place. The first character comes back and wants to retrieve the object. The question is where the actor will look for the hidden object. In order to pass the test, children have to attribute a false belief to the agent, even though they themselves hold a different (and true) belief. It turns out that children under four wrongly predict that the character will look at the correct (new) place, even though the person has had no actual access to the information that the object was moved.2 The results from this test generalise across a broad range of related experimental tasks. If one shows a three-year-old a familiar candy box and lets her find out that it contains pencils, not candy, then puts the pencils back and asks her what her friend, who has not seen what is in the box, will think about its contents, three-year-olds typically say the friend will think there are pencils in the box (Perner, Leekham and Wimmer 1987). In the same task, when three-year-olds are asked what they initially thought was in the box, they reply ‘pencils’, not ‘candy’, that is, they do not remember their previous false beliefs (Gopnik and Slaughter 1991). The performance of children over four in such tasks is considerably better. Other important developments during this period include children’s ability to distinguish between appearance and reality (Flavell 1986), and to remember the source of their beliefs (e.g. a sensory modality – Wimmer, Hogrefe and Sodian 1988; O’Neill and Gopnik 1991). In sum, it seems that, around the age of four or five, children become capable of successfully handling second-order representations (metarepresentations), that is, representations about their mental states. Moreover, children of this age begin to understand that these representations may vary in reliability, may change over time and may differ across individuals. After five, development related to mindreading continues in a number of directions. For instance, children appreciate that mental state attributions can be multiply embedded (‘John thinks that Mary fears that Beatrice suspects . . .’), and acquire social concepts such as ‘duty’ and ‘commitment’ which rely on such complex reasoning.





Anna Papafragou

. Metarepresentation and modal interpretations . A proposal I now want to consider the development of epistemic modal interpretations in the context of the child’s theory of mind. Consider the familiar case of the English modal verbs. When using a verb such as must, may, might, should, will, or ought to on an epistemic interpretation, the speaker relativises possibility or necessity with respect to her set of beliefs. In (7a), for instance, the speaker communicates that it follows from her beliefs that there is some mistake; similarly, in (7b) the speaker conveys that it is compatible with her beliefs that it will rain tomorrow: (7) a. There must be some mistake. b. It may rain tomorrow.

It follows, then, that genuine epistemic interpretations of modals presuppose the ability to reason about mental representations – and possibly also the capacity to assess their accuracy in representing the real world and to understand that they may be revised with time. All of these abilities fall in the domain of theory of mind. In this sense, epistemic items closely parallel other so-called mental terms such as know, think, forget, remember, etc. As Shatz, Wellman and Silber (1983) point out, the first uses of mental verbs do not make genuine reference to mental states but are rather conversational devices; for instance, the verb know initially occurs in standardised phrases such as You know what?, I don’t know. By the end of the third year, however, their subjects use mental terms with true mental reference (e.g. Before I thought this was a crocodile; now I know it’s an alligator – Shatz et al. 1983: 309). According to other studies, by the age of four children recognise the greater reliability of know over think and similar predicates (Moore and Davidge 1989; see the reviews in Olson and Astington 1986; Moore and Furrow 1991). Recall that epistemic modals also appear by the end of the third year but their first uses are probably incompletely understood: it is not until the age of four or five that children recognise that epistemic must is stronger than may or should. A strong piece of support for the hypothesis that epistemic modals and mental terms implicate metarepresentational abilities comes from experiments carried out by Moore, Pure and Furrow (1990). These researchers showed that there is a correlation between the comprehension of the relative reliability of know vs. think/guess and the ‘strength’ of epistemic must vs. might on the one

Modality and theory of mind 

hand, and performance on false belief, belief change and appearance-reality tasks on the other hand. This finding suggests that there is a connection between the ability to distinguish epistemic strength encoded by natural language words and the ability to reason about false beliefs in non-linguistic tasks. The proposal that epistemic modality relies on theory-of-mind abilities opens up the possibility of studying the acquisition of modal vocabulary in parallel with the development of the child’s cognitive resources. This is especially valuable since, as we saw, most of what is known about the early semantics and pragmatics of modal expressions is based on observational data (and is therefore open to interpretation problems). A promising direction is to use independent cognitive tasks tapping theory of mind as a way to gain insight into the concepts underlying the use of modal and mental vocabulary. For instance, it would be interesting to compare children’s use and comprehension of expressions of epistemic possibility such as may or might to their understanding of uncertainty or undecidability in non-linguistic tasks. One area in which this approach can be particularly illuminating is the early modal lexicon. Even though epistemic modal items are largely absent from the speech of two-year-olds learning English, there is some evidence that in languages other than English some epistemic modal expressions occur before the age of three. Based on longitudinal data, Choi (1995) has argued that in Korean a set of sentence-ending evidential suffixes is acquired between one year eight months and three years, long before the set of epistemic and root modal verbs appears. These suffixes indicate the status of the speaker’s beliefs, in terms of either the degree of certainty/commitment to the belief, or its source (e.g. communication, perception, hearsay). Such data have led some researchers (e.g. Shatz and Wilcox 1991) to propose a cognitive (theory-ofmind) constraint which would block the acquisition of epistemic predicates before two years and six months in languages such as English but could be overcome by language-specific input in languages such as Korean. Even though it is not clear what semantic knowledge underlies the use of epistemic expressions in the Korean two-year-olds’ vocabulary (or whether these occurrences are truly mentalistic – Papafragou 1998, 2001), these data raise interesting questions about the relation between linguistic and cognitive development. Recall, for instance, that English-speaking children still experience difficulties with non-linguistic tasks involving sources of beliefs until around age four (O’Neill and Gopnik 1991). One would therefore want to know whether these early occurrences of evidentials in child Korean are truly metarepresentational and, if so, whether their appearance in the productive vocabulary of child language correlates with performance in theory-of-mind



Anna Papafragou

tasks (e.g. false belief tasks, or simpler tasks investigating understanding of belief sources). Some work in this direction is already under way (Papafragou and Li, in press). What about root modal interpretations? Clearly, at least some of these interpretations do not presuppose sophisticated understanding of the mind. It is perhaps for this reason that ability can and volition will appear earlier than epistemic (or other root) interpretations. Some of the other early uses of root modals, such as hafta/have (got) to which appears around three, are predominantly used to state an obligation and do not particularly involve mental-state understanding. However, it is worth pointing out that the development of theory of mind definitely affects more complex root uses. For instance, in order to use a modal like must to impose (rather than simply state) an obligation, some consensus is required among the interlocutors as to social relations, issues of power, authority, duty, commitment and other social concepts which rely heavily on mental state attribution. Furthermore, in order for speakers to use must felicitously on this interpretation, they must have the ability to calculate the addressee’s desires, preferences and goals. It is probably no coincidence that some of these more sophisticated aspects of root interpretations are not acquired by children until the age of six or later. . The mapping problem I have argued above that theory of mind can provide a central part of an account of the acquisition of epistemic modality. Naturally, one needs to consider several other factors which also play a role in the acquisition of modal vocabulary. One major consideration relates to input. There is evidence that adults talking to very young children predominantly use root modals. Wells (1979) reports that can and will were the only modals which were used by all of the mothers in his sample of very young children; they were also the modals with the highest frequency. Shatz, Grimm, Wilcox and Niemeier-Wind (1990) present evidence from American English and German maternal speech to children around three years of age: They note that fewer than 10% of the modals used were interpreted epistemically (cf. similar findings in Shatz, Grimm, Wilcox and Niemeier-Wind 1989). These results are hardly surprising given the circumstances of production of modal terms. Since parents are likely to tailor their speech to the interests, activities and goals of their children, it is easy to see why desire, ability, obligation and permission interpretations of modals outnumber those of epistemic necessity or possibility in early speech to children. Furthermore, epistemic modals

Modality and theory of mind 

in English, as in other languages, are more formal expressions than closely related vocabulary items such as semi-auxiliaries (have to), or mental terms (think, know). Their markedness contributes to their restricted distribution in the adult speech to children. Interestingly, for the very same reasons, adult English speakers in general use root modals much more frequently than epistemic modals (see the corpus data in Coates 1983). In sum, then, it seems that children learning English for a variety of reasons simply do not hear enough modal verbs with epistemic interpretations in their parents’ speech – hence they do not have sufficient opportunities to acquire them. More generally, it is plausible that epistemic interpretations of modals are harder to acquire than root interpretations because the mapping between the word and the epistemic concept may be harder to construct. That is, regardless of the conceptual difficulty posed by epistemic terms, the task of figuring out that a certain vocabulary item has such an abstract meaning may be hard for the young learner (Gleitman 1990). Consider the similar case of mental verbs: one of the reasons that verbs like want, know and think pose greater challenges to young learners than verbs such as jump, eat or catch, is that the referents of the former class are not straightforwardly observable. Therefore, the information supplied in the extra-linguistic context is a less helpful cue in figuring out the meaning of mental verbs (compared to more ‘concrete’ verbs). Recent experimental evidence lends support to this conclusion (see Gillette, Gleitman, Gleitman and Lederer 1999). A full account of how modal interpretations are learned would have to disentangle the precise contribution of conceptual development and mapping processes to the acquisition of the meanings of epistemic predicates.3

. A test case: Autism . The mindblindness hypothesis Autism is a severe childhood psychiatric condition which is characterised by a number of social and communicative impairments. Autistic children lack the usual cognitive flexibility, imagination and pretence, and their behaviour is marked by a restricted range of interests and activities. A number of authors have suggested that (at least certain aspects of) autism can be explained as resulting from lack or delay in theory of mind (Baron-Cohen, Leslie and Frith 1985; Leslie and Roth 1993; Leslie and Thaiss 1992; Frith 1989; Baron-Cohen 1995). On this hypothesis, autism is a form of mindblindness with a number of



Anna Papafragou

implications for cognitive development. Children who suffer from autism are unable to understand (and attribute) false beliefs; they cannot appreciate the mind/brain as an organ with mental functions; they are unable to realise that seeing leads to knowing; they experience difficulty with the mental-physical and appearance-reality distinctions. As I mentioned earlier in this paper, these aspects of the human theory of mind are normally acquired around the age of three or four. Autistic children differ from both normal children and children suffering from various other disorders (e.g. Down’s syndrome) in that their mentalizing abilities are selectively impaired. The mindblindness hypothesis about autism offers a solid testbed for the link which I have sought to establish between epistemicity and mindreading. If, as I have suggested, the development of the metarepresentational machinery responsible for sophisticated mentalizing is a prerequisite for the full-blown and correct production and comprehension of epistemic modal markers, then there should be a severe difficulty with such terms in the language of autism. In the following pages, I want to discuss the findings of a study by de Roeck and Nuyts (1994) which appears to disconfirm this dissociation between autistic language and epistemicity and thus to pose a challenge for my proposal. . Modal language in autism De Roeck and Nuyts (1994) set out to investigate the use of three markers of epistemic modality in Dutch by four high-functioning autistic adults. The subjects covered an age range between nineteen and twenty-nine years. Their total and verbal IQ ratings were 113t (116v), 96.5t (121v), 88t (77v) and 68t (77v) (where t = total and v = verbal IQ). Three modal items were studied on the basis of a corpus of spontaneous speech data: waarschijnlijk ‘probable/probably’ (the item has both adjectival and adverbial uses), denken ‘think’ and kunnen ‘can/may’. The results of the study can be summarised as follows. The adjectival use of waarschijnlijk is completely absent from the data, a fact which is not particularly surprising given the low rates of occurrence of this use even in normal speech. As for the adverbial use, it occurs rather normally in the speech of two of the four subjects. Denken is used by all four subjects, and there is evidence of both its complement-taking and its parenthetical uses. Finally, kunnen is used as an epistemic modal with higher frequency than in normal speech (where it is more often used to express simple root/deontic modality). The authors conclude that all four subjects demonstrably use epistemic expressions in ways much similar to normal subjects and therefore appear to engage in metarep-

Modality and theory of mind 

resentation quite unproblematically. Consequently, the performance of these high-functioning autistic adults is taken to provide some evidence against the theory-of-mind hypothesis for autism. As far as I can see, there are various alternative conclusions one might draw from de Roeck and Nuyts’ observations (and the authors themselves explore some of them). For instance, one might question the initial assumption that epistemic modality involves metarepresentation. I want to argue for two alternative explanations of their findings, which do not threaten the premise that epistemic predicates involve metarepresentation. Firstly, one could argue that, in fact, the specific items used by the four autistic adults in the study were not always used to communicate epistemic modality. Even though it is not possible to find compelling evidence simply by inspecting the subjects’ spontaneous speech data, there are certain examples which suggest that something less than a fully metarepresentational reading might be available to the speakers. Specifically, one of the subjects seemed to use the same modal construction ‘t Kan (ook) zijn dat (‘it may/can (also) be that’) as a set phrase (and often in inappropriate circumstances) in the majority of on-line epistemic judgements she produced. This subject was both the youngest (nineteen years old) and the one with the lowest IQ scores of the group (68t, 77v). An actual example is (8), where the subject (T) is talking about her not liking a particular boy (translation by de Roeck and Nuyts): (8) A: With Bart you don’t really like it, do you? T: No. A: Why? Do you know? T: I mean but now it is better already, [. . . ] but you cannot do anything about it, do you, when you don’t like it? A: No. T: ‘t Kan ook zijn da je daar helemaal niets aan kunt doen. ‘It may also be that you cannot do anything about it.’

Apparently genuine cases of epistemic modality in T’s speech might have been the product of imitation, or simply an incompletely understood place-holder, as the authors acknowledge: (9) ‘t Kan zijn dat er in de middelste kast daar, plaklint zit. ‘It may be that in the cupboard in the middle there, there is some tape.’

Secondly, and more importantly, the autistic subjects whose linguistic production was studied might belong together with that talented minority of autistic individuals (around 20–30%) who pass the (simple versions of) false-belief



Anna Papafragou

tasks, and who have been shown to perform well in a variety of tasks involving metacognition.4 Such exceptional competence is usually explained in two ways: i. Autism might cause a delay in the operation of human mindreading capacities, rather than a permanent and absolute inability to metarepresent (Eisenmajer and Prior 1991; Baron-Cohen 1995). Depending on the different components which may be taken to form the human mentalizing capacity, delay may affect one or several of them at different time-points. ii. In high-functioning cases of autism, subjects may be capable of performing some sort of mentalizing through the usage of alternative strategies. As Leslie and Roth (1993: 103) remark, ‘such strategies are arrived at by exercising general reasoning abilities, and hence require a quite high level of such abilities, together with extensive practice and general knowledge, and hence do not appear before adolescence’. Although the exact nature of these strategies is at present not fully known, it is possible that they rely on an analogy to the pictorial format for the representation of thought. Pictures as a representational medium, unlike mental constructs, are rather well-understood by autistic subjects; evidence for this is provided by good performance of autistics in variations of the ‘false belief ’ task which involve ‘false (dated) photographs’ and ‘false drawings’ (see Leekam and Perner 1991; Leslie and Thaiss 1992; and Charman and Baron-Cohen 1992, respectively). Moreover, there is some tentative evidence which suggests that, whereas normal subjects report their inner experience in terms of inner speech, pictures or ‘pure thought’, very able autistic subjects describe their mental contents entirely in terms of pictures (Hurlburt, Happé and Frith 1994; cf. Hurlburt 1990). In the small group of autistic people studied by Hurlburt et al., ability to report inner experience in terms of pictures correlated closely with performance on standard theory-ofmind tasks, independent of IQ. Overall, then, this evidence suggests that highfunctioning autistic people may use their understanding of external representations such as pictures to achieve an understanding of mental representations such as thoughts and beliefs. Both the age and the high verbal and non-verbal IQ of the subjects in de Roeck and Nuyts’ study corroborate an explanation for their performance in terms of compensatory mechanisms for theory-of-mind abilities. It is worth noting here that T., the youngest subject and also the one with the lowest verbal and general intelligence, had markedly greater difficulty in using modal expressions than the rest of the group. In sum, then, de Roeck and Nuyts’ (1994) data offer an impressive manifestation of the success of very able autistic subjects to overcome the difficulties in their condition and to engage in normal verbal communication. However, they do not offer a counterexample to the theory-

Modality and theory of mind 

of-mind hypothesis for autism, or an argument for dissociating epistemicity from metarepresentational (more specifically, metacognitive) abilities. In order to present a truly critical test for both hypotheses, one has to turn to linguistic data from young autistic people (between four and twelve years) where the presence of compensatory mechanisms is less likely. The prediction then is that early autistic child language should be quite impoverished in the use of mental terms. This prediction is borne out. Tager-Flusberg (1993) reports the results of a longitudinal study, in which six children with autism between three and six were followed for between one and two years. Their spontaneous productions were compared to those of a Down’s syndrome group of six children of the same productive language level and age. Her findings suggest that, while both groups of children talk about perception and mental states of desire and emotion, autistic children make significantly fewer references to cognitive mental states than children suffering from Down’s syndrome. Elsewhere, TagerFlusberg specifically comments on the use of modal verbs in the language of the two sample groups of the above study (see Tager-Flusberg 1997). It appears that can and will make up 95% of all modals used by both groups of children, while epistemic uses are rare overall. What is more important for present purposes, though, is that even within the non-epistemic range those uses of can and will which require some mentalistic understanding (e.g. the volitional/intentional uses of will) are very rare in the language of autism, whereas they occur more frequently in the speech of the children with Down’s syndrome. It appears, therefore, that impairments in theory of mind are reflected in the way language is used by children with autism. Moreover, it seems that this is done in a way consistent with the analysis of specific linguistic items (e.g. epistemics or volition markers) in mentalistic terms.

. Concluding remarks In this paper, I presented some preliminary arguments to support the hypothesis that the acquisition of epistemic modality presupposes the ability to metarepresent mental representations (e.g. beliefs). I also argued that this metarepresentation hypothesis yields a number of interesting and testable predictions about the connection between early semantic/pragmatic abilities and cognitive development. Finally, I showed that an impairment in the ability to construct metarepresentations of mental representations – attested in autism – affects the ability to learn and use epistemic predicates.

 Anna Papafragou

As it stands, the metarepresentation hypothesis is not committed to a specific model of folk psychology. Furthermore, regardless of how theories of theory of mind will turn out, they will have to account for the fact that complex metacognitive abilities are the prerequisite for several aspects of linguistic development. Nevertheless, it might be interesting to examine whether linguistic data of the sort I have discussed may be used to tease apart different theoretical proposals about the structure of the human metarepresentational mechanism. This possibility remains open for future research.

Notes * I wish to thank Annabel Cormack, Helen Tager-Flusberg, Lila Gleitman, Alison Gopnik, Francesca Happé, Nina Hyams, Alan Leslie, Marilyn Shatz, Dan Slobin, and especially Neil Smith and Deirdre Wilson. I am also grateful to the editors and to the series editor, Werner Abraham, for comments and suggestions. . I will focus on British English throughout. . For some interesting manipulations of the standard method, see Cassidy (1998). . I have not considered here the development of the syntax of auxiliaries, which is also implicated in any account of the acquisition of modality. For instance, in Italian root potere (‘can’) and dovere (‘must’) appear some time between one year and ten months and two years and nine months, i.e. earlier than their English counterparts. This fact can be explained on syntactic grounds (Hyams 1986: 117–118). . I am referring to first-order theory-of-mind tasks. Second-order theory-of-mind tasks, which involve multiple attribution of belief (Jane thinks that her mother believes that Bob is at home) and present no difficulty for normal children around six or seven are hard for most of the members of the talented minority of autistic children.

References Astington, J. and A. Gopnik (1991). Theoretical explanations of children’s understanding of the mind. In G. Butterworth, P. Harris, A. Leslie and H. Wellman (Eds.), Perspectives on the Child’s Theory of Mind (pp. 7–31). Oxford: Oxford University Press/The British Psychological Society. Astington, J., P. Harris and Olson, D. (Eds.) (1988). Developing Theories of Mind. New York: Cambridge University Press. Baron-Cohen, S. (1995). Mindblindness: An Essay on Autism and Theory of Mind. Cambridge, Mass.: MIT Press. Baron-Cohen, S., A. Leslie and U. Frith (1985). Does the autistic child have a ‘theory of mind’? Cognition, 21, 37–46.

Modality and theory of mind 

Butterworth, G., P. Harris, A. Leslie and H. Wellman (1991a). Editorial preface. In G. Butterworth, P. Harris, A. Leslie and H. Wellman (Eds.), Perspectives on the Child’s Theory of Mind (pp. 1–4). Oxford: Oxford University Press/The British Psychological Society. Butterworth, G., P. Harris, A. Leslie and H. Wellman (Eds.) (1991b). Perspectives on the Child’s Theory of Mind. Oxford: Oxford University Press/The British Psychological Society. Byrnes, J. and M. Duff (1989). Young children’s comprehension of modal expressions. Cognitive Development, 4, 369–387. Carruthers, P. and P. Smith (Eds.) (1996). Theories of Theories of Mind. Cambridge: Cambridge University Press. Cassidy, K. (1998). Three- and four-year-old children’s ability to use desire- and belief-based reasoning. Cognition, 66, B1–B11. Charman, T. and S. Baron-Cohen (1992). Understanding beliefs and drawings: A further test of the metarepresentation theory of autism. Journal of Child Psychology and Psychiatry, 33, 1105–1112. Choi, S. (1995). The development of epistemic sentence-ending modal forms and functions in Korean children. In J. Bybee and S. Fleischman (Eds.), Modality in Grammar and Discourse (pp. 165–204). Amsterdam: Benjamins. Coates, J. (1983). The Semantics of the Modal Auxiliaries. London: Croom Helm. De Roeck, A. and J. Nuyts (1994). Epistemic modal expressions by high-functioning autistic adults: A test case for the ‘theory of mind’ hypothesis. Antwerp Papers in Linguistics 82. Also published as J. Nuyts and A. de Roeck (1997). Autism and metarepresentation: The case of epistemic modality. European Journal of Disorders of Communication, 32, 113–137. Eisenmajer, J. and M. Prior (1991). Cognitive linguistic correlates of ‘theory of mind’ ability in autistic children. British Journal of Developmental Psychology, 9, 351–364. Flavell, J. (1986). The development of children’s knowledge about the appearance-reality distinction. American Psychologist, 41, 418–425. Frith, U. (1989). Autism: Explaining the Enigma. Oxford: Blackwell. Frye, D. and C. Moore (Eds.) (1991). Children’s Theories of Mind. Hillsdale, NJ: Erlbaum. Gee, J. and I. Savasir (1985). On the use of will and gonna: Toward a description of activitytypes for child language. Discourse Processes, 8, 143–175. Gerhardt, J. (1991). The meaning and use of the modals HAFTA, NEEDTA and WANNA in children’s speech. Journal of Pragmatics, 16, 531–590. Gillette, J., H. Gleitman, L. Gleitman and A. Lederer (1999). Human Simulation of vocabulary learning. Cognition, 73, 135–176. Gleitman, L. (1990). The structural sources of verb meanings’. Language Acquisition, 1, 1–55. Gopnik, A. and V. Slaughter (1991). Young children’s understanding of changes in their mental states. Child Development, 62, 98–110. Gopnik, A. and H. Wellman (1995). Why the child’s theory of mind really is a theory’. In M. Davies and T. Stone (Eds.), Folk Psychology: The Theory of Mind Debate (pp. 232–258). Oxford: Blackwell. Gordon, P. (1986). Folk psychology as simulation’. Mind and Language, 1, 158–171.



Anna Papafragou

Guo, J. (1994). Social Interaction, Meaning, and Grammatical Form: Children’s Development and Use of Modal Auxiliaries in Mandarin Chinese. Doctoral dissertation, University of California at Berkeley. Hirschfeld, L. and S. Gelman (Eds.) (1994). Mapping the Mind: Domain-Specificity in Cognition and Culture. Cambridge: Cambridge University Press. Hirst, W. and J. Weil (1982). Acquisition of epistemic and deontic meaning of modals. Journal of Child Language, 9, 659–666. Hurlburt, R. (1990). Sampling Normal and Schizophrenic Inner Experience. New York: Plenum Press. Hurlburt, R., F. Happé and U. Frith (1994). Sampling the inner experience of autism: A preliminary report. Psychological Medicine, 24, 385–395. Hyams, N. (1986). Language Acquisition and the Theory of Parameters. Dordrecht: Reidel. Kratzer, A. (1981). The notional category of modality. In H. Eikmeyer and H. Rieser (Eds.), Words, Worlds and Contexts (pp. 38–74). Berlin: De Gruyter. Leekam, S. and J. Perner (1991). Does the autistic child have a metarepresentational deficit? Cognition, 40, 203–218. Leslie, A. and D. Roth (1993). What autism teaches us about metarepresentation. In S. Baron-Cohen, H. Tager-Flusberg and D. Cohen (Eds.), Understanding Other Minds: Perspectives from Autism (pp. 83–111). Oxford: Oxford University Press. Leslie, A. and L. Thaiss (1992). Domain specificity in conceptual development: Evidence from autism. Cognition, 43, 225–251. Leslie, A. (1988). Some implications of pretence for mechanisms underlying the child’s theory of mind. In J. Astington, P. Harris and D. Olson (Eds.), Developing Theories of Mind (pp. 19–46). Cambridge: Cambridge University Press. Moore, C. and J. Davidge (1989). The development of mental terms: Pragmatics or semantics? Journal of Child Language, 16, 633–641. Moore, C. and D. Furrow (1991). The development of the language of belief: The expression of relative certainty. In D. Frye and C. Moore (Eds.), Children’s Theories of Mind (pp. 173–193). Hillsdale, NJ: Erlbaum. Moore, C., K. Pure and D. Furrow (1990). Children’s understanding of the modal expression of speaker certainty and uncertainty and its relation to the development of a representational theory of mind. Child Development, 61, 722–730. Noveck, I., S. Ho and M. Sera (1996). Children’s understanding of epistemic modals. Journal of Child Language, 23, 621–643. O’Neill, D. and C. Atance (2000). “Maybe my daddy give me a big piano”: The development of children’s use of modals to express uncertainty. First Language, 20, 29–52. O’Neill, D. and A. Gopnik (1991). Young children’s ability to identify the sources of their beliefs. Developmental Psychology, 27, 390–397. Olson, D. and J. Astington (1986). Children’s acquisition of metalinguistic and metacognitive verbs. In W. Demopoulos and A. Marras (Eds.), Language Learning and Concept Acquisition (pp. 184–199). Norwood, NJ: Ablex. Papafragou, A. (1998). The acquisition of modality: Implications for theories of semantic representation. Mind and Language, 13, 370–399. Papafragou, A. (2000). Modality: Issues in the Semantics-Pragmatics Interface. Amsterdam & New York: Elsevier Science.

Modality and theory of mind 

Papafragou, A. (2001). Linking early linguistic and conceptual capacities: The role of theory of mind. In A. Cienki, B. Luka and M. Smith (Eds.), Conceptual and Discourse Factors in Linguistic Structure (pp. 169–184). Stanford: CSLI Publications. Papafragou, A. and P. Li (in press). Evidential morphology and theory of mind. To appear in Proceedings from the 26th Annual Boston University Conference on Language Development. Boston: Cascadilla Press. Perkins, M. (1983). Modal Expressions in English. London: Frances Pinter. Perner, J., S. Leekam and H. Wimmer (1987). Three-year-olds’ difficulty with false belief: The case for a conceptual deficit. British Journal of Developmental Psychology, 5, 125– 137. Pratt, C. and P. Bryant (1990). Young children’s understanding that seeing leads to knowing (so long as they are looking into a single barrel). Child Development, 61, 973–982. Premack, D. and D. Woodruff (1978). Does the chimpanzee have a theory of mind? Behavioural and Brain Sciences, 4, 515–526. Shatz, M., H. Grimm, S. Wilcox and K. Niemeier-Wind (1989). The uses of modal expressions in conversations between German and American mothers and their twoyear-olds. Paper presented at the biennial meeting of the Society for Research in Child Development, Kansas City, MO. Shatz, M., H. Grimm, S. Wilcox and K. Niemeier-Wind (1990). Modal expressions in German and American mother-child conversations: implications for input theories of language acquisition. Unpublished ms., University of Michigan, Ann Arbor. Shatz, M., H. Wellman and S. Silber (1983). The acquisition of mental verbs: A systematic investigation of the first reference to mental state. Cognition, 14, 301–321. Shatz, M. and S. Wilcox (1991). Constraints on the acquisition of English modals. In S. Gelman and J. Byrnes (Eds.), Perspectives on Language and Thought (pp. 319–353). Cambridge: Cambridge University Press. Shepherd, S. (1982). From deontic to epistemic: An analysis of modals in the history of English, creoles, and language acquisition. In A. Ahlqvist (Ed.), Papers from the 5th International Conference on Historical Linguistics (pp. 316–123). Amsterdam: Benjamins. Smoczynska, M. (1993). The acquisition of Polish modal verbs. In N. Dittmar and A. Reich (Eds.), Modality in Language Acquisition (pp. 145–169). Berlin: De Gruyter. Stephany, U. (1986). Modality. In P. Fletcher and M. Garman (Eds.), Language Acquisition (pp. 375–400). Cambridge: Cambridge University Press. (1st ed. 1979). Stephany, U. (1993). Modality in first language acquisition: The state of the art. In N. Dittmar and A. Reich (Eds.), Modality in Language Acquisition (pp. 133–144). Berlin and New York: De Gruyter. Tager-Flusberg, H. (1993). What language reveals about the understanding of minds in children with autism. In S. Baron-Cohen, H. Tager-Flusberg and D. Cohen (Eds.), Understanding Other Minds: Perspectives from Autism (pp. 138–157). Oxford: Oxford University Press. Tager-Flusberg, H. (1997). Language acquisition and theory of mind: Contributions from the study of autism. In L. Adamson and M. Romski (Eds.), Research on Communication and Language Development (pp. 133–158). Baltimore, MD: Paul Brookes Publishing.

 Anna Papafragou

Wellman, H. and D. Estes (1986). Early understanding of mental entities: A reexamination of childhood realism. Child Development, 57, 910–923. Wellman, H. and J. Woolley (1990). From simple desires to ordinary beliefs: The early development of everyday psychology. Cognition, 35, 245–275. Wells, G. (1979). Learning and using the auxiliary verb in English. In V. Lee (Ed.), Cognitive Development: Language and Thinking from Birth to Adolescence (pp. 250–270). London: Croom Helm. Wells, G. (1985). Language Development in the Pre-school Years. Cambridge: Cambridge University Press. Whiten, A. (Ed.) (1991). Natural Theories of Mind. Oxford: Blackwell. Wimmer, H. and J. Perner (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13, 103–128. Wimmer, H., J. Hogrefe and B. Sodian (1988). A second stage in children’s conception of mental life: Understanding informational accesses as origins of knowledge and belief. In J. Astington, P. Harris and D. Olson (Eds.), Developing Theories of Mind (pp. 173–192). Cambridge: Cambridge University Press. Yaniv, I. and M. Shatz (1988). Children’s understanding of perceptibility. In J. Astington, P. Harris and D. Olson (Eds.), Developing Theories of Mind (pp. 93–108). Cambridge: Cambridge University Press.

Negative polarity and modality in Middle Dutch ghe-particle constructions* Gertjan Postma University of Leiden

One of the uses of the Middle Dutch prefix ghe- is sensitive to the negative polarity of its syntactic environment. This use as a negative polarity item (NPI) can be clearly identified in a verb that does not allow other uses of ghe-: the copular verb sijn ‘to be’. The negative polarity use can, however, also be traced in other verbs, although in a less pure form. Negative polarity ghehas all the properties of Modern Dutch verbal negative polarity items, such as kunnen uitstaan ‘can stand’. Crucially, these involve locality conditions, anti-additivity, and low-scope modality. By extending the Neg-criterion to NPIs with negative force, various restrictions on the ghe-construction can be derived, such as 1. the preterite restriction, 2. the non-V2 restriction (‘freezing’), and 3. the low-scope modality restriction.

.

Introduction

From Aristotle’s Organon onward, it has been a well-known fact in semantics and logic that negation and modality are closely related. Multiple-world semantics has made it possible to express the relation between necessity and possibility with the help of negation. From this perspective, it does not come as a surprise that, cross-linguistically, negative polarity and modality are closely related as well: many negative polarity verbs and verbal expressions, such as English can stand, are stative or modal in nature. Some modal verbs, such as English must, have a negative polarity counterpart (need). The nature of this relation, however, has remained unclear. Usually, a lexical approach is adopted: the lexeme can stand, together with its modality, is taken to be a negative polarity item. Its modal nature is thus stored in its lexical specification, rather than being derived from its NPI nature, or from a condition on its proper use as a NPI.

 Gertjan Postma

Middle Dutch is interesting in this connection as it exhibits a productive strategy of forming negative polarity verbs. The Middle Dutch verbal prefix ghe- turns a verb into a negative polarity verb. Hence, Middle Dutch is an excellent tool to study the relation between negative polarity and modality, and we will see that various interactions with modality show up. After a discussion of the general properties of Middle Dutch ghe-, it will be shown to be a NPI (Sections 1–6). I then turn to its interactions with modality (Sections 7–8), while some conclusions are provided in Section 9. . Properties of the Middle Dutch prefix gheTo understand the function of Middle Dutch ghe-, it is instructive to compare it with its Modern Dutch counterpart. The Modern Dutch preverbal ge- particle functions in two ways: 1. As a participial prefix in verbs without a weak particle, such as geslapen ‘slept’, the past participle of slapen ‘sleep’; 2. As an inherent verbal prefix, as in gebeuren ‘happen’. I will characterise these two uses as morphological ge- and as lexical/semantic ge-, respectively. Examples are given in (1). (1) a.

Jan heeft geslapen ‘John has slept.’ b. Dat gebeurt wel vaker ‘That happens more often.’

morphological gelexical/semantic ge-

In (1a), the form geslapen functions as the past participle of the verb slapen ‘sleep’. In (1b), the prefix functions as part of the lexical verb gebeuren ‘happen’, which has ge- in all forms of the paradigm. In Middle Dutch, the ghe- particle has a slightly wider use. Besides the uses existing in Modern Dutch, i.e. participial ghe- and lexical/semantic ghe- (cf. 2a, b), Middle Dutch shows a third use that is dependent on the syntactic context (cf. 2c). (2) a.

type Ontfaerme u miere scade, die mi Reynaert heeft ghedaen morphological pity you my- harm, that me- Reynaert has done ‘Take note of my damage, which Reynaert has done to me.’

Negative polarity and modality in Middle Dutch ghe-particle constructions 

b. Niet mijn wille, maer dijn wille ghescie not my wil, but thy will happen-subj ‘Not my will, but thy will be done.’ c. Van eenen goeden boem en kan niet dan Of a good tree  can not than goede vruchten ghecomen good fruits ghe-come ‘From a good tree only good fruits can come.’

lexical

syntactic

In (2a) the use of participial ghe- is parallel to that in (1a) in Modern Dutch. I will call this ghe-1 . In (2b) we encounter a use that can be compared with the inherent lexical/semantic ge- in Modern Dutch (1b). I will refer to it as ghe-2 . It occurs in all forms of the verbal paradigm, as shown in (3). (3) a.

Misdaden die den wisen ghescien present Sins that the wise- ghe-happen ‘Crimes that befall the wise’ b. Niet mijn wille, maar dijn wille ghescie subjunctive Not my will, but thy will ghe-happen ‘Not my will but thy will be done.’ c. Binnen dier tijd so gheschiede een dootslach preterite In that time it ghe-happened a manslaughter ‘A case of manslaughter was done in that time.’ d. Twivel ende vrese van allen dingen die ghescien moghen infinitive Doubt and fear of all things that ghe-happen may ‘Doubt and fear of all things that can happen’ e. Met alle dien goeden werken die ghesciet waren With all those good works that ghe-happened were ende noch ghescien souden participle and yet ghe-happen would ‘With all the good works that had happened and were yet to happen.’ f. Ju staat te ghesciene grote ramp You stand to ghe-happen great disaster ende swaer verlies (Wat. 8012) gerund and heavy loss ‘Now a great disaster and heavy loss will take place.’

The use of ghe-2 can be compared with other aspectual prefixes, such as be-, ver-, ont- and the Middle Dutch prefix et-, which show up in the entire verbal paradigm as well. It forms an essential part of the verb itself. In those cases in which the verb has such a weak prefix, no extra participial prefix ghe- is added,



Gertjan Postma

as (3e) shows. In this respect, Middle Dutch behaves in essentially the same way as Modern Dutch. In addition to these uses, Middle Dutch exhibits a third use of ghe-, which has received ample attention in grammars of Middle Dutch. It is usually characterised as an aspectual particle. I will call it ghe-3 . It is typically attached to infinitives and preterites, as is illustrated in (4a, b). (4) a.

Nu en can ic langher niet gheswighen Now  can I any-longer ghe-keep-silent ‘Now I cannot keep silent anymore.’ b. Alle den smaec ende bliscap, die alle hoverdighe menschen ye ghehadden All the taste and joy all haughty humans ever ghe-had. ‘All the taste and joy that all haughty people ever had.’

In no other forms of the verbal paradigm does this use of ghe- occur.1 This restricted use makes ghe-3 in a certain sense comparable to participial ghe-2 . However, there is something odd in the description of this third use in the literature. Whereas the Middle Dutch dictionary of Verwijs & Verdam lists the participial forms with ghe-1 under the entry of the verb without ghe- (this is the situation in modern dictionaries as well), syntactic ghe-, as in (4), is listed under the prefixed verb: ghehadden in (4b) is listed under the putative verb ghehebben. Strictly speaking, this is not correct: only verbs with lexical ghe-2 , which occurs in the entire paradigm, deserve an entry of their own. The differential treatment of ghe-1 en ghe-3 must be understood in the light of two considerations. Firstly, there is the fact that ghe-3 occurs as a specific paradigmatic form of the infinitive, which is the base form normally used in western lexicographical tradition. This creates a trap for dictionary makers, which does not exist for ghe-1 . Secondly, Verwijs & Verdam did not have the theoretical tools to sharply distinguish cases of ghe-3 from the inherent prefix ghe-2 . However understandable Verwijs & Verdam’s classification may be, it would be the same as listing the participle of dragen ‘carry’, i.e. gedragen ‘carried’ under the verb gedragen ‘GE-carry’ instead of dragen. (5) a.

Ik heb de piano naar boven gedragen from dragen ‘to carry’ ‘I have carried the piano upstairs.’ b. Hij moet zich beter gedragen from (zich) gedragen ‘to behave’ ‘He must behave better.’

Negative polarity and modality in Middle Dutch ghe-particle constructions 

(6) a.

Ghi laedt de menschen met bordenen die You burden the humans with burdens that si nit ghedragen en connen from draghen ‘carry’ they not ghe-carry  can. ‘You burden the people with loads that they cannot carry.’ b. Ene arme scone vrouwe die A poor beautiful lady that een kind bi hem gedrouch from ghedraghen ‘be pregnant’ a child by him ghe-carried ‘A poor beautiful woman who was with child by him.’

In listing both occurrences of ghe-draghen in (6) under the same entry, Verwijs & Verdam increase the number of entries for which there is no lexical/semantic [ghe-2 +V] combination. This practice also disguises the difference between the lexical/semantic ghe-particle, which usually makes a quite distinct idiosyncratic semantic contribution, and the syntactic ghe-prefix, which makes hardly any or no semantic contribution at all. Instead of treating ghe-2 en ghe-3 together, as Verwijs & Verdam do, I will make an attempt at unravelling the distinct uses of ghe-, using modern linguistic tools. I will start by showing that the three particles are independent from each other. . The independence of ghe-3 and ghe-1 The Middle Dutch verb comen ‘to come’ can be used to demonstrate the independence of ghe-3 and ghe-1 . This verb is special in that it lacks ghe-prefixing in the participle.2 This is usually explained on the basis of the inherent perfectivity of the verb comen: no additional perfective ghe- is necessary in the participle. On this traditional view, participial ghe- coincides with perfective ghe-. It should be noted, however, that ghe- does occur as a prefix to comen in the infinitive and preterite, where it is found in only a limited number of specific syntactic contexts. In (7), I provide some instances of ghe-comen in Verwijs & Verdam’s dictionary (11 out of 27 in total), classified according to whether they occur in a negative clause, in a clause introduced by the conjunction (al)eer ‘before’, or in another context. (7) a.

Tot ghenen goeden leven en connen ghecomen (who) to no good life  could ghe-reach ‘Who could not reach a good life.’

 before other +



Gertjan Postma

 before other b. Van eenen goeden boom en kan niet dan goede vruchten ghecomen + Of a good tree  can not but good fruits ghe-come ‘Only good fruits can come from a good tree.’ c. Also dat hi met crachte noch met conste wten boeme ghecomen en konde + So that he with force nor with trick out the tree ghe-come could ‘In such a way that he could not leave the tree neither by force nor by a trick.’ d. Aleer sy te Steenvoerde ghequamen + Before they at Steenvoerde ghe-came ‘Before they reached Steenvoerde.’ e. By desen lande ende by der sonnen sweer ic u, + On this land and on the sun swear I to-you mocht so ghecommen (dat waren hoghe- saken), might it that-way ghe-come (that were very good), ic souder mynen eet by maken dat . . . I would there my oath to make that . . . ‘By this land and by the sun, I swear to you that, if it happened that way (which would be excellent), I would swear by these that. . .’ f. Eer sire gecomen conden . . . + Before they there ghe-come could ‘Before they could go there, . . .’ g. Alle dEnghelsce (. . . )/so verre als si (. . . ) in der heeren handen ghecomen conden + all English, (. . . ) so far as they (. . .) in the lords’s hands ghe-come could ‘all the English people that would fall into the lord’s hands.’ h. Daer die vianden niet wel by en conden gecommen + there the enemies not well at  could ghe-come ‘(. . . ), which the enemies could not reach very well.’ i. Eer si toe conden gecomen, + before they to (it) could ghe-come, so heeft die coninc de vlucht genomen so has the king flight taken ‘Before they could reach the place, the king had escaped.’

Negative polarity and modality in Middle Dutch ghe-particle constructions

j.

Hoe ic daer ghecomen sal, . . . How I there ghe-come will, . . . ‘In whatever way I will come there, . . .’

 before other +

These examples make it abundantly clear that a block on participial ghe- in perfective verbs does not imply a block on ghe-3 . Some additional regularities stand out. In the first place, all instances in (7) involve infinitives and preterites, indicating that we are not dealing with a lexical verb ghecomen but with a specific paradigmatic form of the verb comen. Secondly, we observe that this use of ghe- predominantly occurs in negative contexts: only 10 of the 27 cases are not negative. Most grammars and dictionaries of Middle Dutch mention this tendency (see Stoett [1923] 1977: 205; Overdiep 1914: 6–8; Van Swaay 1899), but they do not make any attempt at explaining it. Four of the 10 non-negative cases involve a context with eer ‘before’. This is often called the perfective use (see for example Stoett [1923] 1977: 205). Finally, there are six cases classified as ‘other’, to which we will return in Section 4. It is a striking fact that [ghe-3 + infinitive] always occurs with a modal verb (Streitberg 1891: 144; Van Swaay 1899: 55) in the deontic reading. In most cases the modal verb is connen ‘can’, but [ghe- + infinitive] also occurs in the complement of moghen (in its possibility reading), sullen ‘will’, dorren ‘dare’, willen ‘wish’ (very rare), and also with the verb laten ‘let’ (in the sense of ‘permit’; rare). (8) a.

Meer dan ic can ghenoemen more than I can ghe-mention ‘More than I can mention’ b. By desen lande ende by der sonnen sweer ic u, mocht so ghecommen By this land and by the sun swear I thee, might-it so GHE-happen ‘In the name of this land and the sun, I swear that if it might happen so . . . ’ c. Hoe ic daer ghecomen sal How I there GHE-come shall ‘How I will ever get there.’ d. Dat hijt wale doerste ghedoen That he-it well dared GHE-do ‘That he did dare to do it.’

connen

moghen

sullen

dorren





Gertjan Postma

e.

f.

Dander en wilde niet hem geantwerden The-other  wanted not him GHE-answer ‘The other one did not want to answer him.’ (not attested)3

willen

laten

What is crucial here is that not a single case has been found where ghe3 +inf occurs in the complement of moeten ‘must’, or in the complement of non-modal ECM verbs such as sien ‘see’ en horen ‘hear’.4 This shows that a specific type of modality is necessary to license the occurrence of ghe-3 . Only CAN modality allows for [ghe+V inf]. I will return to this modality restriction in Section 7.4.1.

. The independence of ghe-3 and ghe-2 The independence of syntactic ghe-3 and lexical/semantic ghe-2 can be demonstrated best with the help of the verb sijn ‘be’. As shown in (9), the Modern Dutch copular verb zijn is one of the few verbs that does not allow for a lexical prefix at all: the prefixes be-, ver-, ge- or ont- are inadmissible with zijn. The same is true for the Middle Dutch prefixes be-, ver-, ont-, ghe- en te-. (9) *ic be-ben, *ic ont-ben, *ic ver-ben, *ic ge-ben, *ic te-ben I be-am, I ont-am, . . .

Nevertheless, the Middle Dutch Dictionary (Verwijs & Verdam 1885–1952) contains the entry ghesijn.5 I provide a number of representative examples in (10). (10) a.

Hy en can niet ver gezijn negation He  can not far GE-be ‘He cannot be far.’ b. Alle die mensen die ye gewaren restrictive set ∀ All the people that ever GE-were ‘All people who ever existed’ c. Hoe soude sy allen ghesijn, die soe vele boeke by haer hadde? question (rhetorical?) How would they all GE-be, who so many books with them had ‘How would they be who had so many books with them.’ d. Hoe groot, hoe wijs, hoe heylich, sy oock connen gezijn concessive How great, how wise, how holy they ever could GE-be ‘However great, however wise, however holy they could be’

Negative polarity and modality in Middle Dutch ghe-particle constructions 

e.

f.

Dat onder dat volc nu meer heyligher liede sijn dant hier voermaels ye ghewaren comparative That under that people now more holy  men are than-it before ever GE-were ‘That there are more holy men among those people than there were ever before.’ En konden sy van den elfsten niet eens gesijn, so sullen die thiene den elfsten kiesen conditional  could they of the eleven not unanimous GE-be, then shall the ten the eleventh choose ‘If they cannot agree upon the eleventh person, then the ten must choose the eleventh’

Parallel to the result in (8), ghe- is only found in infinitives and preterites, indicating that we are dealing with ghe-3 .6 Apparently, ghe-3 is possible in contexts where lexical ghe-2 is blocked. It is striking that negation shows up again as a licensing factor of ghe-3 . Moreover, [ghe + infinitive] only occurs in modality contexts, in this case only in the complement of connen ‘can’. Apart from negative contexts, ghesijn occurs in the restrictive set of the universal quantifier (10b), in rhetorical questions (10c), in concessive contexts (10d), in the scope of the comparative (10e), and in conditional clauses (10f). In the following section, I will go into the question of what these contexts have in common. For the moment, it suffices to conclude that an independent ghe-3 exists. This ghe-3 prefix has specific properties that make it distinct from participial gheon the one hand, and from lexical/semantic ghe- on the other hand. I list its properties in (11). (11) Properties of ghe-3 1. It is completely productive, i.e. it occurs with all verbs: lexical verbs, copular verbs and auxiliaries. 2. It is only prefixed to the infinitive and the preterite.7 3. It imposes specific constraints on the syntactic context: for example, it requires the presence of a. negation (+ modal verb) b. comparative, eer, etc. c. other specific contexts.

I will make a detailed study of these contexts in the following section.



Gertjan Postma

. Negative polarity contexts It is a well-known fact that certain lexemes or collocations only occur in the scope of negation. A typical example is the Dutch adverbial phrase ook maar, which may be compared with English at all. Whereas ook maar cannot be used in an affirmative sentence such as (12a), it is perfectly acceptable in a negative clause such as (12b). (12) a. *Jan had ook maar één boek gelezen John had at all one book read b. Geen enkele leerling had ook maar één boek gelezen. no single pupil had at all one book read ‘Not a single pupil had read one single book.’

Items such as ook maar are called negative polarity items. Following Klima (1964), the grammatical properties of negative polarity have been studied intensively. One of the points at issue has been why the set of contexts allowing the presence of items like ook maar is wider than merely the predicative scope of ordinary negation. The restrictive set of the universal quantifier, conditional sentences, comparatives, rhetorical questions, and concessive clauses all provide a good context for the use of ook maar. A number of instances are given in (13). (13) a.

Niemand had ook maar iets gedaan. NEG nobody had at all anything done ‘Nobody had done anything at all.’ b. Alle toeristen die ook maar iets hadden gekocht, all tourists who at all anything had bought voelden zich bedrogen. restrict. set ∀ felt  cheated ‘All tourists who had bought anything at all felt cheated.’ c. Geen toerist die ook maar iets begrepen had no tourist who at all anything understood had vond de gids redelijk. restr. set ¬ found the guide reasonable ‘No tourist who had understood anything at all considered the guide to be reasonable.’ d. Hij ging vaker naar de bioscoop he went more often to the cinema dan ook maar iemand vermoed had. comparative

Negative polarity and modality in Middle Dutch ghe-particle constructions 

than at all anybody suspected had ‘He went to the cinema more often than anybody had ever suspected.’ e. Wie zou daar ook maar iets van kunnen begrijpen? rhetorical question who would there at all anything of be-able understand? ‘Who would understand anything at all of this?’ f. Mocht je er ook maar iets van begrepen hebben, . . . concessive might you of it at all anything understood have ‘If you should have understood anything at all of this, . . . ’ g. Als je er ook maar iets van begrepen hebt, . . . conditional if you of it at all anything understood have ‘If you have understood anything at all of this, . . . ’ h. Voordat Jan ook maar een woord kon uitbrengen, . . . before Jan at all one word could utter conjunctions like ‘Before Jan could utter one single word, . . . ’ voor(dat) ‘before’, zonder ‘without’, tot ‘until’

It has remained unclear for a long time what these contexts have in common. Seuren (1975) assumes that they contain a covert negative operator. Ladusaw (1980) proposes that they share the logical property of allowing for downward entailments. An entailment F(X) → F(Y) is called ‘downward’ if X ⊂ Y. In (14a) for instance, the restrictive set ‘left-handed Frisian’ is properly included in ‘Frisian’. Similarly, the nuclear scope of ‘sing loudly’ is contained in that of ‘sing’. Hence, the inferences in (14a) and (14b) are both downward. (14) a. Every Frisian sings → Every left-handed Frisian sings b. Every Frisian sings *→ Every Frisian sings loudly.

A downward inference is licit in (14a) but not in (14b): the quantifier every is evidently monotone-decreasing in its restrictive set, but not in its nuclear scope. We thus predict that the negative polarity item ever is allowed in (15a) but not in (15b). (15) a. Every Frisian who has ever been in Paris sings. b. *Every Frisian ever sings a song.

We can set up a similar line of reasoning for negative quantifiers such as Dutch geen ‘no’ and English no. Once again, we see the two downward entailments in (16), which are both licit.



Gertjan Postma

(16) a. No Frisian sings → No left-handed Frisian sings b. No Frisian sings → No Frisian sings loudly

Correspondingly, the Dutch negative quantifier geen and English no allow for the presence of negative polarity items like Dutch ook maar or English ever, in the restrictive set as well as in the nuclear scope, as shown in (17). (17) a.

No Frisian who has ever been in Paris would say anything bad about France. b. No Frisian will ever sing a song.

Hoeksema (1983) shows that comparatives create monotone-decreasing contexts, and hence allow for negative polarity items. Sanchez, van der Wouden & Zwarts (1993) argue that contexts with voor/eer ‘before’ are also monotonedecreasing. Van der Wouden (1994) also argues for subjunctive contexts being monotone-decreasing. Although our understanding of negative polarity is still fragmentary, it has become clear that negative polarity is a linguistic phenomenon showing up in a wide range of languages, with a by-and-large similar pattern, i.e. the same type of contexts seem to license negative polarity items across languages. Some of them are listed in (18). (18) Negative polarity contexts 1. nuclear scope of negation; 2. restrictive set of universal quantifiers, e.g. all, no . . . ; 3. comparatives; 4. conditional clauses; 5. rhetorical questions; 6. concessive clauses and subjunctives; 7. clauses introduced by words meaning ‘without’, ‘before’, ‘until’ etc. 8. . . .

In some languages a negative polarity item alternates with an element with almost the same meaning, but without polarity sensitivity: English some and any both indicate existential quantification (∃). The essential difference is that any is a negative polarity item, but some is not. Let us now return to the use of syntactic ghe-3 . We now see that the set of contexts in which it occurs, shown in (10), closely corresponds to the set in (18). We can therefore in general terms characterise ghe-3 as a negative polarity item: the function of ghe-3 is to turn a verb into negative polarity verb. We capture this in (19).

Negative polarity and modality in Middle Dutch ghe-particle constructions 

(19) Hypothesis Middle Dutch ghe-3 is a negative polarity item

Modern Dutch no longer has this or any other productive strategy for forming negative polarity verbs.8 However, Dutch has verbs and verbal expressions that are inherent negative polarity items. I give some in (20). (20) a.

Niemand/*Jan hoeft zo vroeg weg te gaan. nobody/Jan needs so early away to go ‘Nobody/*Jan needs to leave so early.’ b. Niemand/*Jan kon de nieuwe direkteur uitstaan. nobody/Jan could the new manager stand ‘Nobody/*Jan could stand the new manager.’ c. Niemand/*Jan kon het iets schelen. nobody/Jan could it something mind ‘Nobody/*Jan cared at all.’ d. Niemand/*Jan taalde daarnaar. nobody/Jan wanted it ‘Nobody/*Jan even asked for it.’ e. Dat dondert *(niet) that thunders (not) ‘It doesn’t matter a damn.’

There is a special ring to these negative polarity verbs/verbal expressions, however: they are modal in nature. This shows up quite clearly in those expressions that convey modality explicitly, such as kunnen uitstaan ‘can stand’ in (20b) and kunnen schelen ‘care’ in (20c). The modality, however, can also be implicit, as in donderen ‘matter’ in (20e). That there is an underlying modality in donderen ‘matter’ and talen ‘want/ask for’ is shown conclusively by the impossibility of embedding these verbs under kunnen ‘can’.9 (21) a. *Niemand kon daar naar talen nobody could there for want b. *Dat kan niet donderen that could not thunder

Apparently, negative polarity in the verbal domain goes hand in hand with modality. Whatever the reason for this (see Section 6 and 7.5.1), the fact that the Middle Dutch form [ghe-3 +infinitive] only occurs in modal contexts supports the negative polarity hypothesis.



Gertjan Postma

. The syntactic nature of ghe-3 . Clitic raising In Sections 2 and 3, syntactic ghe- has been set off against lexical/semantic ghe(in verbs like ghescien ‘happen’) and morphological ghe- (as in ic bem ghevallen ‘I am fallen’). Up to this point I have characterised Middle Dutch ghe-3 as ‘syntactic ghe-’ without any scruple. The reason for calling ghe-3 syntactic is that it is completely productive, and is only found in specific syntactic environments like negative polarity and modality. Moreover, it is virtually impossible to attach any precise semantic content to ghe-3 . Ghe-3 clearly encodes something syntactic. It would be desirable, however, to be able to point to a real syntactic process in which ghe-3 is involved. Consider the fact that ghe- can move from the embedded infinitive to the dominating modal verb. Some instances of this are given in (22). (22) a.

Dat negeen broeder negeen knecht, die den huse omme sout of in karitaten dient ummer ghedorre slaan. that no brother no servant who works in the house for money or in caritas serves ever GHE-dares hit ‘That no brother should ever dare to hit a servant who works in the house for money or out of charity’. b. Hoe gesal ic dit emermeere ghebeteren mogen binnen minen live? how GHE-will I this ever improve be-able within my life? ‘How will I ever be able to improve this in my life?’ c. Ende hi niet geconste castien coninc Arolt. and he not GHE-could punish king Arolt ‘And he could not punish king Harold’. d. Men zoudze niet gheconnen craken. One would-them not GHE-be-able damage ‘One would not be able to damage them’. e. Ic waende men niet en soude hebben gheconnen vinden man. I thought one not  would have GHE-be-able find man ‘I thought one would not be able to find anyone’. f. Sodat gi die liede gelaten niet en cont neven u geliden voren noch achter no besiden so-that you those people let not  can beside you walk before nor behind nor beside ‘So that you can not let people walk next to you, nor in front of you, nor behind you nor at your side’.

Negative polarity and modality in Middle Dutch ghe-particle constructions 

g. Dat ic se niet en ghemach sien (tot ien dat se mi orlof ghevet) that I her not  can see (until that she me permission gives) ‘That I cannot see her until she gives me permission’

Verwijs & Verdam’s comment regarding the example in (21g) is that ghe-mach sien is the same as mach ghe-sien, the form one would expect. Stoett (1977: 206) remarks that modal verbs sometimes ‘adopt the prefix by anticipation’. This seems to be the correct intuition. There is some evidence for this intuition: the occurrence of ghe- with a present tense modal, as in (22b) and in (22g) implies that the modal itself cannot be the underlying source of ghe-, as we have seen that ghe- only occurs with infinitives and preterites. In a generative framework, this would mean that Stoett’s ‘anticipation’ can be analysed as involving syntactic movement of ghe- from the main verb to the modal verb, as shown schematically in (23), which represents the first part of (22g). (23) dat ic se niet en ghei -mach [ PRO ti -sien . . . ]

If this analysis is correct, this movement can be compared to clitic movement in Italian verb constructions with potere ‘kunnen’, volere ‘want’ and fare ‘do/make’, as shown in (24) (Rizzi 1982; Burzio 1986: 322). (24) Mario loi vuole [ leggere-ti ] Mario it wants read ‘Mario wants to read it.’

Movement of the clitic into the matrix clause is obviously syntactic here. Note that movement of ghe- seems to be obligatory in those cases in which the embedded verb has a lexical prefix, such as in ghebeteren in (22b).10 . Locality A second indication that ghe-3 is not lexical-semantic but syntactic can be obtained from locality effects. The licensing of the ghe-prefix is blocked across an opaque (tense) domain: no barrier may be present between ghe- and its licensing factor. This can be inferred from the absence of ghe- in gerunds, i.e. the form of the verb that occurs after the infinitival particle te, ‘to’ as in English to go. In other words, there are no examples like (25a, b), not even in a modal reading of the construction. I have represented this formally in (25c). (25) a.

Doe was daer niemen sonder die das, hine hadde te claghene/*gheclaghene over Reynaerde Then was there nobody but the badger, he  had to complain about Reinaerd



Gertjan Postma

‘Then there was nobody but the badger who had no complaints about Reinaerd’ b. dat hij jeghen tvoorbod van den stede meer souts zood dan hij vermochte te ziedene/*gheziedene naar dordinanche that he against the prohibition of the city, more salt produced than he -could to produce- after the-rules ‘That he, against the prohibition of the city, produced more salt than he was allowed to produce according to the rules.’ c. *ne  [ te [ ghe-Vgerund . . . ]

Locality effects also show up in constructions with multiple embeddings without an infinitival particle. In these cases, ghe- appears on the infinitive that is hierarchically closest to the modal verb.11 We can represent this abstractly as in (27a, b). (26) a.

Sine mochten . . . den muer niet ghedoen vallen They could the wall not make fall ‘They could not cause the wall to come down.’ b. *Sine mochten . . . den muer niet doen ghevallen

(27) a. ne  ghedoen V b. *ne  doen ghe-V

If we accept this argument ex silentio, we conclude that once again syntactic locality appears to be necessary for the licensing of ghe-3 , on peril of ungrammaticality. Yet, we do encounter constructions of the type in (28), where there is no local relation between the negation and ghe but where a local relation does exist between modality and ghe: (28) Ic waende, men niet en soude hebben ghe-i +connen ti +vinden man I thought one not  would have GHE-can find anyone ‘I thought that one would not have been able to find anyone’

This indicates that the distance between ghe- and the modal verb is crucial (strict locality), not the distance between ghe- and the negation niet en. The negative lexeme only needs to c-command the polarity item. This is entirely analogous to the requirements on negative polarity items in Modern Dutch. This is illustrated in (29). (29) a.

Niemand had gedacht [dat Marie hem zou kunnen uitstaan] Nobody had thought that Marie him would can- stand ‘Nobody thought that Marie would be able to stand him.’

Negative polarity and modality in Middle Dutch ghe-particle constructions 

b. *Niemand kon denken [dat Marie hem uitstond] Nobody could think that Marie him stood c. *Marie zou kunnen denken [dat niemand hem uitstond] Mary would can- think that nobody him stood

The relative well-formedness of (29a) with respect to (29b, c) indicates that, for the licensing of the polarity item uitstaan, the negation may reside in a higher clause than the polarity item, and that the relation between modal and polarity item is bound to strict locality. We conclude that ghe- behaves in all respects as a negative polarity item. We formulate the locality requirement on ghe-3 as in (30): (30) ghe-3 and the modal verb are locally construed with a lexical predicate: [ ghe-V]

We will return to this in Section 7.

. Negative polarity items and modality . Two main types of negative polarity items and the modality requirement Negative polarity items can be subdivided into two major classes. The first type includes those expressions that can occur in any monotone-decreasing environment. Examples are Dutch hoeven ‘need’, kunnen schelen, ‘care’ en kunnen uitstaan ‘can stand’. Monotone-decreasing contexts fulfill two formal requirements, i.e. L1 en L2 (for details, see Zwarts 1986: 357).12 The second type of NPI imposes an additional requirement on its environment in that it can only occur in a specific type of monotone-decreasing contexts, namely those that are anti-additive. These contexts thus show an additional property L3 (for details, see Zwarts 1986: 351).13 I have represented the two types in the table in (31). (31) formal characterization

in the scope of

examples of NPI

a. monotone decreasing nauwelijks ‘hardly’ hoeven ‘need’, kunnen {L1,L2} niet elke ‘not every’ uitstaan ‘can stand’, slechts/niet dan ‘only’ kunnen schelen ‘matter’, . . . niemand ‘nobody’ niet ‘not’ ...



Gertjan Postma

b. monotone decreasing *nauwelijks ‘hardly’ ook maar ‘at all’, enig [e:nıx] of the anti-additive *niet elke ‘not every’ ‘any’, ooit ‘ever’, . . . subtype {L1,L2,L3} *slechts/niet dan ‘only’ niemand ‘nobody’ niet ‘not’

Evidently, the verbal NPIs fall in the first category: they impose the fewest requirements on their environment. We should note, however, that in addition to their monotone decreasing characterisation, they impose a requirement of modality. The other type of additional requirement is anti-additivity. We now obtain the scheme in (32). (32) NPIs can occur in contexts that are monotone decreasing {L1,L2} + additional requirement additional requirement in the scope of a. modality

b. anti-additivity

nauwelijks ‘hardly’ niet elke ‘not every’ slechts/niet dan ‘only’ niemand ‘nobody’ niet ‘not’ ... *nauwelijks ‘hardly’ *niet elke ‘not every’ *slechts/niet dan ‘only’ niemand ‘nobody’ niet ‘not’

examples of NPI hoeven ‘need’, kunnen uitstaan ‘can stand’, kunnen schelen ‘matter’, . . .

ook maar ‘at all’, enig [e:nıx] ‘any’, ooit ‘ever’, . . .

I leave for further research the question as to whether these two additional requirements reduced to a single formal condition. In the case of verbal NPIs, the additional requirement can be satisfied by modality alone. In the case of non-verbal NPIs, modality licensing is excluded and anti-additivity is required. Let us now address the question to which type ghe-3 belongs: is it antiadditive or not? If ghe-3 is a prefix that turns a verb into an NPI, we expect that it belongs to the first type since other verbal NPIs of Modern Dutch belong to this class as well. If we find a context that has ghe-3 in the scope of the Middle Dutch equivalent of nauwelijks ‘hardly’ or niet elke ‘not every’, we could conclude that ghe- belongs to the monotone-decreasing subtype that does not impose a further requirement on its environment. Unfortunately, I have not found a single construction of the type: niet elc . . . ghe-V. However, this seems

Negative polarity and modality in Middle Dutch ghe-particle constructions 

to follow from the total absence of niet elc in the Middle Dutch corpus.14 We are therefore left with contexts with nauwelijks ‘hardly’ and slechts ‘only’. In (33) I give two sentences with ghe-3 that provide conclusive evidence. (33) a.

Cume mach hij ghespreken van moede ‘Hardly can he speak of tiredness.’ b. Van eenen goeden boom en kan niet dan goede vruchten ghecomen from a good tree  can not than good fruits GHE-come ‘From a good tree only good fruits can come.’

These sentences show that ghe-3 can occur in the scope of cume ‘hardly’ and niet dan ‘only’, proving that ghe-3 indeed belongs to the NPI-type that requires monotone-decreasing contexts where the additional requirement is satisfied by modality. We are therefore justified in concluding that Middle Dutch ghe+V behaves like Modern Dutch negative polarity verbal expressions. . Summary In the previous sections, we have seen that one of the uses of the Middle Dutch verbal prefix ghe- is that of a negative polarity item. In this use, it behaves like Modern Dutch NPI verbal expressions. We have found two additional restrictions: 1. if ghe- is prefixed to infinitives, it must be embedded under a modality of the CAN type (cf. Section 2); 2. if it is prefixed to finite verbs, the verb may not be a present tense (cf. Section 2). In Section 7.5.3, we will identify a third restriction: the verbal form with ghe- never undergoes V-second. These highly specific restrictions seem to be of a syntactic nature. In order to account for these restrictions, we need to accommodate the behaviour of ghe- in a syntactic format. As yet, the only syntactic condition on NPIs reported in the literature is that they should be c-commanded by a licenser. Fortunately, for our analysis of ghe- we can make use of a syntactic condition that has been formulated for negative items, especially in the Romance languages, where negative items can pile up and behave very much the same as NPIs in the Germanic languages: they must comply with the Neg-criterion. In the following section, I will make an attempt to generalise the Neg-criterion to a certain type of NPIs.



Gertjan Postma

. Negative polarity items and the NEG-criterion . Negation in Middle Dutch Middle Dutch is like French in that it exhibits negative concord and negative spread. N-words in Middle Dutch behave as negative polarity items: when they are in the scope of an n-item they have an indefinite reading, not a negative reading. The negative clitic seems to function as a scope marker, as is illustrated in (34). (34) a.

Personne n’a vu rien nobody not has seen nothing ‘Nobody has seen anything.’ b. Niemen en hevet ghesien dinc negheen nobody not has seen thing none ‘Nobody has seen anything.’

When n-items occupy a scope position, they clearly have negative force, such as personne and niemen in (34a, b). Apart from these n-items, Middle Dutch has true NPIs, such as een quade bone ‘a bad bean’, which must always be in the scope of an n-item. (35) ic en weet daer af niet/*ø ene quade bone I  know there of not/ø a bad bean ‘I don’t know anything about it’

The NPI een quade bone can never move to a scope position and be the head of a negative chain. In this respect it is similar to the expression a red cent in English. Apart from n-items and NPIs, Middle Dutch has a clitic negative en, which is never found in isolation (Postma 2002) and can be compared to the French scope marker ne. There is a limited number of Middle Dutch NPIs, however, that have a dual nature: in most cases these behave like NPIs, but occasionally they carry negative force. The sentential negator niet can be absent although the NPI itself never occupies a higher position than the verbal neclitic (We may assume that they obtain their operator force in a lower NegP position). A typical example is een twint ‘nothing at all’, an expression without any real meaning, which can best be compared to English squat or fuck all or other taboo items (Postma 2000b). Een twint can be used without niet if it has a preverbal position (see Postma 2002 for an inventory of the Middle Dutch items that behave like een twint).15

Negative polarity and modality in Middle Dutch ghe-particle constructions 

(36) a.

Gine sout macht een twint hebben te anegroetene hare You- would power nothing have to greet her ‘You wouldn’t have any authority to greet her.’ b. *Gine sout hebben macht een twint te anegroetene hare c. . . . , sine connen minnen niet een twint . . . , they  could love not a bit ‘they could not make love a bit’

preverbal

postverbal

Items like een twint can thus take over the function of the negator niet, the normal carrier of the negation in Middle Dutch. We can now see that ghebehaves like een twint: there are occurrences where ghe+V can be used in a negative sense without the strong negator niet being present, as in (37). (37) Mine bliscap es so groot, dat ict gesegghen en mach My happiness is so great that I-it ge-say  may ‘My happiness is so great that I cannot utter it.’

We thus see that ghe+V can act like a reinforcing n-item. Reinforcing n-items need to meet the Neg-criterion (Van Kemenade 2000). In what follows I discuss the NEG-criterion and investigate to what extent we can apply it to ghe-. . The Neg-criterion in the CP domain Negative and interrogative sentences have many things in common, not only logically, but also syntactically: interrogative and negative constituents may move to an operator position; both may cause inversion; both may license negative polarity items, etc. This has led syntacticians to treat these operators and/or quantifiers as syntactically equal. The proposal by Haegeman & Zanuttini (1991), to the effect that, parallel to the WH-criterion there is a NEG-criterion, has been very influential. According to the WH-criterion (Rizzi 1991), a WH-feature must be involved in a specifier-head agreement relation. A WH-word such as where must move to a sentence-peripheral position and induces movement of the finite verb to the second position (V2). In terms of X-bar theory (Jackendoff 1977; Chomsky 1982), movement of where is XP movement to SpecCP whereas movement of the verb constitutes head-movement to C. According to Rizzi (1991), the verb has an abstract WH-feature in such cases. (38) a.

[Where has [he t gone ]] [+WH ] [+WH ]



Gertjan Postma

b. [Neither has [he t gone]] [+ NEG ] [+NEG ] c.

CP SPEC

C’ C0

d.

IP

Where has [+WH] [+WH] CP SPEC

...

C’ C0

Neither has [+NEG] [+NEG]

IP ...

Negative phrases behave in a parallel fashion, cf. (39a, b). In those cases, abstract agreement takes place in CP with respect to the neg-feature [+NEG ] (Haegeman & Zanuttini 1991). (39) The -criterion a. A -operator must be in a spec-head configuration with a [+NEG ] head X0 . b. A [+NEG ] head X0 must be in a spec-head configuration with a operator.

Notice that the NEG-criterion does not specify the projection in which negagreement takes place. . The Neg-criterion in the IP domain In lower domains of the sentence, for instance in the IP domain, neg-agreement phenomena can be observed as well. In (40a, b), a Middle Dutch and a French sentence, negation is not only realised in the subject but also on the verb as a clitic (en/ne). (40) a.

niemen en ginc te dier stede nobody neg went to that place ‘Nobody went that way’.

Negative polarity and modality in Middle Dutch ghe-particle constructions 

b. personne ne vient à la maison nobody neg comes to the house ‘Nobody goes home’

The clitic is frequently taken to be a kind of a verbal inflection. Two reasons are provided in the literature for this. According to Weerman (1988) and van Gestel et al. (1992), the demise of the clitic negation goes hand in hand with the gradual decline of the verbal inflection with regard to person and number features (‘deflection’). These authors argue that dropping the negative clitic is an instance of deflection. Haegeman (1995: 146–149) reports that in Middle Dutch the absence of person and number features, as in infinitival subclauses, prevents negative inflection by means of the negative clitic en. Since in the IP domain Neg-agreement is not realised abstractly as verb movement, but overtly as inflection, there needs to be agreement with respect to number/person specifier features and the head of the projection that has the [+] feature. According to Pollock (1989), the IP domain is split into AgrP, NegP and TP, but a compelling reason for the existence of a separate NegP is still to be found (Klooster 1993). For simplicity, I will assume that projections like CP and IP can be used as domains for negative agreement: nothing in the Neg-criterion forces us to assume a specific negative head. . The Neg-Criterion in the VP domain Recently, van Kemenade (2000) has argued that there is a negation projection just above the predicative domain, i.e. just above VP in verbal clauses.16 She derives this from the distribution of not/naught in early Middle English. Basically, however, her claim applies to all languages. The reinforcing negator would then be realised in the lower negation projection. Taboo items like (didley) squat and fuck-all can only be in (an underlying) object position (Paul Postal and John Robert Ross, pers. comm; McCloskey 1993; Postma 2000b). They can occasionally take over the function of the negator not (Postma 2002): they have negative potentiality. (41) He understands fuck-all/bugger-all/diddly squat of it

I will assume that these items acquire negative force by moving to a scope position lower in the tree, say specFP, where they enter into NEG-agreement with a head F. Let us label this projection NegP2 (presumably it can be identified as AgrOP). Applying the Neg-criterion, we conclude that in sentences like (41), agreement with respect to [+] must be realised in NegP2. Given that this



Gertjan Postma

takes place outside the inflectional domain (the IP domain), I will assume that this is realised by an abstract operation, i.e. by verb movement, just as in CP (see Section 7.2). If V-movement in negative sentences takes place overtly (cf. Kayne 1999), we obtain adjacency between the negative constituent and V. In observational terms, we then see a Neg-V order, i.e. the negator surfaces in front of the verb, provided V2 or other verb movements do not blur this picture. In this way, van der Wurff (1999) and Beukema & Van der Wurff (this volume) explain the preverbal position of negative objects in late Middle English. The made-up sentences in (42) illustrate the difference between non-negative and negative objects: (42) a. John will visit Peter b. John will nobody visit

The structure of NegP2 (or AgrOP) in (42b) will be as in (43). (43)

NegP Spec nobody

Neg’ Neg

VP

visit

...

It will be clear that this theory makes predictions with respect to the location and behaviour of elements playing the role of the reinforcing negator in Middle Dutch, such as een twint. As we have seen, this NPI is realised preverbally (see Section 7.1), i.e. in front of the non-finite verb, and may take over the role of the reinforcing negator niet. It may therefore be necessary to allow for not only reinforcing elements to move to specNegP, but also (some) NPIs. If these do, they acquire the force of a negative item. Laka (1994) also suggests that in the Romance languages there is no difference between negative elements and NPIs. Movement of an NPI to the specifier of a functional position, i.e. to a scope-position (Laka’s ΣP in Romance and Basque) makes it behave like a negative operator. While Laka’s claim merely involves movement of maximal projections and their interpretation, the NEG-criterion imposes the restriction that there must be a correlation between XP-movement and head-movement. Because of the word order phenomena observed in Middle Dutch and Middle English, and because of theoretical considerations of symmetry, we need to apply the Neg-criterion to Van Kemenade’s NegP2 as well.

Negative polarity and modality in Middle Dutch ghe-particle constructions 

. The Neg-criterion and verbal NPIs Let us now consider what happens if a verbal NPI takes over the role of the reinforcing negator, as ghe+V in (37) shows. We now arrive at the following situation: 1. the element, or a relevant projection of it, must move to SpecNegP2; 2. the verbal head must undergo head movement to the head of NegP2 to establish formal NEG-agreement. Such a verbal NPI must, as it were, split itself into two to play this dual role. There are two ways in which this can be realised. .. The modal restriction The first possibility of complying with the Neg-criterion is that the verbal projection itself consists of two lexical heads, the lower of which carries the NPI property. These are cases with an auxiliary, as in (44). In (44a) the VP in which the NPI ghecomen ‘ghe-come’ is included moves to SpecNegP and the auxiliary can moves to the head of NegP2. In (44b), the finite verb can moves on to the V2 position. (44) a.

dat daer niemen ghecomen en can that there nobody ghe+come neg can ‘that nobody can come there’ b. niemen en can daer ghecomen nobody neg can there ghe+come

In order to be able to move to the head of NegP2, the auxiliary must at some stage be below NegP2. Modals appear to take two types of scope with respect to negation: high scope and low scope. This can be seen in (45), in which Dutch moeten and English must are typical high-scope modals. The combinations moet . . . niet and must not can be paraphrased as ‘having the obligation to not X’.17 (45) a. Jan moet dat niet doen b. John must not do that  >  ‘There is the obligation/John has the obligation not to do that’.

The auxiliary can in (46), on the other hand, is a low-scope modal.18 (46) a. Jan kan dat niet doen b. John cannot do that  >  ‘There is no possibility/John has no possibility to do that’.



Gertjan Postma

Finally, a low-scope variant of must exists: Dutch hoeven and English need. (47) a. Jan hoeft dat niet te doen b. John does not need to do that  >  ‘There is no obligation/John has no obligation to do that’.

All this would predict that CAN modality but not MUST modality can be combined with the NPI ghe-particle. This is indeed the case, as we have seen in Section 2. What is more, we now predict that a Middle Dutch counterpart of English need, Dutch hoeven might also be combinable with ghe-. Unfortunately, Middle Dutch does not have a separate lexeme for low-scope MUST. The sentence in (48), though, might serve as an indication that the correct generalisation is not to be formulated in terms of a restriction to CAN modality, but rather in terms of a restriction to low-scope modality. (48) Noyt ne moeste van hem varen ridder no seriant negheen aen anderen here te soekene leen Never neg must- by him go knight nor sergeant no-one to other lord to seek fief ‘As for him, never did knight or sergeant need to go and ask for a fief of another lord’.

What seems to be the correct formalisation is given in (49): (49) a.

Definition A wide/narrow scope modal is a modal that has wide/narrow scope with respect to negation. b. Generalisation The NPI ghe- requires licensing by a narrow-scope modal.

For verbal NPIs I derived this effect from the Neg-criterion as it applied to NegP2, which implies that this only holds for NPIs that are 1. verbal and 2. may take over the function of negation. At first sight, (49b) seems to be a special case of a broader interaction of NPIs and modality. Not only the Middle Dutch ghe- + infinitive is sensitive to scope effects, negative polarity verbs like English need and Dutch hoeven realise a low-scope modality as well. There are many verbal expressions in Dutch with explicit or implicit CAN-modality (kunnen uitstaan, kunnen velen ‘can stand’, kunnen schelen ‘care’, kunnen maken, etc.), but not a single NPI with MUSTmodality. The only MUST NPI is, as expected, the low-scope realisation hoeven ‘need’.19 Similar data can be provided for the implicit modality in the Dutch NPI-verb talen naar, in the context niet talen naar (this expression does not

Negative polarity and modality in Middle Dutch ghe-particle constructions 

have an exact English equivalent but its semantics is roughly NOT > WANT > HAVE). Talen naar is the low-scope counterpart of the high-scope ‘modal’ expression niet lusten ‘not like’ in the sense of ‘dislike’ (its semantics roughly as WANT > NOT > HAVE). Whereas talen naar is an NPI, the high-scope modal lusten is polarity-insensitive. Such scope effects in verbal expressions in modern Dutch and English might leave us with the suspicion that our generalisation is too weak. Can we not make verbal NPIs such as need and hoeven sensitive to the Neg-Criterion and derive their low-scope nature from this? Despite the initial attractiveness of such a move, I do not think that this is correct. To begin with, it is theoretically undesirable. The Neg-criterion is applicable to expressions that have negative potential, in the sense that they can carry negation. Dutch hoeven and English need have a different status. Secondly, such a generalisation would not be maintainable in view of the freezing effects of ghe-preterites discussed in Section 7.5.3. As we will see there, the Neg-criterion applied to verbal NPIs causes freezing effects with respect to Vsecond. No such freezing effects of verbal NPIs can be found with hoeven or talen naar. Thirdly, it should be noted that the low-scope modal requirement is absolute in Middle Dutch ghe-infinitives but there are systematic exceptions in the case of the modern NPI verbal expressions. It is not immediately clear whether there is any modality in verbal NPIs such as donderen in dat dondert niet ‘that does not matter’. These verbs seem to be stative in nature rather than modal. The scope condition in these cases is more like a conditional statement: if modal in nature, modern Dutch NPI verbal expressions realise low-scope modality. Put differently, there appears to be a block on high-scope modality. Fourthly, and most importantly, the move is not necessary. This can be seen if we extend our modality considerations to all NPIs, whether verbal or not. This point can best be illustrated in a table, in which NPIs are classified with respect to the two dimensions under discussion, ±verbal and ±negative potentiality. NPIs

+verbal

–verbal

+ negative potential

ghe-V

twint

– negative potential

kunnen uitstaan

ene bal

The spurious generalisation would be extending the modality requirement of ghe-V to (kunnen) uitstaan and explain it by the same mechanism, i.e. movement to NegP2. If we do so, negative potentiality is dissociated from movement to SpecNegP. However, we should then do the same in the nominal domain and extend our proposal for twint, i.e. movement to SpecNegP2 because of its neg-



Gertjan Postma

ative potentiality, to NPIs such as ene bal. It is significant that these nominal NPIs also have a modality requirement, although of a slightly different type. This can be seen from the paradigm in (50), which shows the behaviour of negative polarity taboo items such as Dutch ene bal or English shit, fuck-all (cf. Postma 2000b). (50) a.

Niemand begrijpt er ene bal van nobody understand there one ball of ‘Nobody understands the first thing about it’. b. *Niemand moest er ene bal van begrijpen20 nobody must- there one ball of understand ‘Nobody had to understand a single thing of it’. c. Niemand kon er ene bal van begrijpen nobody could there one ball of understand ‘Nobody could understand the first thing about it’

In contrast to NPI verbal expressions, ene bal does not need modality, as can be seen from (50a). Significantly, this NPI noun phrase is blocked by MUST modality (50b), although it does tolerate CAN modality (50c). This behaviour seems to be systematic. If we now return to the verbal NPI expressions, we can subsume these restrictions under the general incompatibility of negative polarity and high-scope modality. In the verbal domain, this shows up in a relatively strong tendency for low-scope modality. The restriction on NPI complementation can be captured in the following two generalizations. 1. NPI verbal expressions with negative potentiality require low-scope root modality 2. NPI complements (verbal as well as nominal) exclude MUST modality. Further investigation of the nature of the general incompatibility of highscope root modality and NPI complements is outside the scope of this paper. What should be stressed here is that the low-scope modality of Middle Dutch ghe + infinitive is by no means a fundamental semantic property; rather, it seems to be an accidental syntactic consequence of the Neg-criterion which, felicitously but accidentally, coincides with the general block on high-scope modality in NPIs. .. The preterite restriction Let us now turn to the preterite restriction of ghe-prefixation. The nature of ghe- as an NPI requires it to move to the specifier of NegP. A corollary of the NEG-criterion is that the verbal head to which ghe- is prefixed must move to

Negative polarity and modality in Middle Dutch ghe-particle constructions 

the head of NegP. This puts a double burden on ghe+V. It follows that simplex tenses cannot undergo ghe-prefixation unless they are complex at some level. A first consequence is the modality restriction discussed in the previous section. A second consequence is the preterite restriction. As we have seen in Section 2, the ghe-prefix does not occur in the present tense, but only in the preterite.21 (51) a.

alle menschen die daer oyt ghewaren all people who there ever ghe+were ‘all people who were ever there’. b. *alle menschen die daer oyt ghesijn all people who there ever ghe+are

It is therefore natural to assume that the present tense is a simplex tense but the preterite is a complex tense, which may have incorporated a kind of aspectual projection. This would explain why in finite tenses, the NPI ghe- only attaches to preterites. The NPI ghe-V must not only undergo XP movement to SpecNegP, but also head movement to Neg, because of the NEG-criterion. Only complex tenses can play this double role. The complex nature of the preterite is at present no more than a stipulation, but it is a reasonable one, with attractive consequences. These will be discussed in the next section. .. Freezing effects If we adopt the line of reasoning sketched above, we could say that syntactically the preterite consists of two heads, which spell out as one lexeme. Obviously, at spell-out, these two heads must be adjacent. There are two possibilities, the first of which is that head-to-head incorporation takes place. The complex is once again a head, which can then freely undergo head movement, e.g. to C. This is the case with a regular preterite. The second possibility is instantiated by the ghe-preterite. The lexical part, ghe+V, which carries the NPI-feature, must move to SpecNegP, while the V-head marked with PAST undergoes head movement to Neg, as required by the NEG-criterion. The structure is illustrated in (52). (52) a.

[NegP [VP . . . ghe-V] [Neg HAVE] [. . .



Gertjan Postma

b.

NegP Spec

Neg’ Neg

VP ghe+V .....ti

HAVEi

[+NPI]

If this configuration does not change, ghe+V[PAST] will be spelled out as a lexical preterite. It cannot move to a higher functional projection as it is a spellout of a hybrid syntactic entity: the ghe+V as the remaining part of VP and the abstract PAST tense as the adjacent head. We therefore predict that the ghepreterite displays ‘freezing’ in NegP and undergoes no further movement to C, i.e.V2, in keeping with the facts. As we have seen in Section 3, the NPI ghe- can be isolated in its purest form in the verb sijn ‘to be’. In (53), I have listed all occurrences in the Middle Dutch Dictionary with the preterite of ghesijn (i.e. ghewas/ghewaren).22 (53) a.

Vor al dat es of ye ghewas blivic in trauwen haer eighijn. For all that is or ever GHE-was, remain-I in faithfulness her own ‘Because of everything that exists or ever existed, I faithfully remain her property.’ b. Noch blivic di ghestade vor al dat es of ye ghewas. Yet remain-I to-you firm for everything that is or ever GHE-was ‘Nevertheless, I remain firm to you for eveyrthing that is and ever was.’ d. Wat ye ghewas off noch sel wesen dat heeft sijn wijsheit al voersien; what ever GHE-was or yet will be, that has its wisdom already foreseen ‘His (i.e. God’s) wisdom has already foreseen everything that ever was or will be.’ e. Overmits des vaders cracht, die al hevet in sijnre macht, wat ye ghewas off sel gheschien; because the father- force that all has in his power what ever GHEwas or will happen ‘Because the father’s, who has everything in his power that ever was or will happen . . . ’ f. God, die is ende ye ghewas ende ewich blijft, God who is and ever GHE-was and for-eternity remains ‘God, who is and always was and will remain forever’

Negative polarity and modality in Middle Dutch ghe-particle constructions 

g. . . . thooste bot (. . . ) dats minne in caritas, die ewich blijft ende ye ghewas . . . the-highest commandment (. . . ) that-is love in charity that forever remains and ever GHE-was ‘the highest commandment that is love in charity which remains forever and ever was’ h. . . . dat dat volc van ertricke also goed es gemeynlike ia dleeke volc alst ye ghewas . . . that the people of the earth as good is common or-rather lay people as-there ever GHE-was ‘ that the people of the earth are as good as the common, or rather lay, people that ever were’ i. . . . dat na hem dat eerste beelde was ghemaect, dat ie ghewas. that after him the first image was made that ever GHE-was. ‘that the first image that ever was was made after him’ j. Wy sijn alle ghemenentlike van conincs gheslechte comen . . . , van enen troncke . . . , die edelste die ye ghewas. We are all equally of king’s descent come . . . from one stem the noblest that ever GHE-was ‘We are all equally of royal descent, from one stem, the noblest that ever was.’ k. . . . dat onder dat volc nu meer heyligher liede sijn . . . , dant hier voermaels ye ghewaren. . . . that under that people now more holier men are (. . . ) than-there here before ever GHE-were ‘that there are now more holier men among that people than there ever were before.’ l. Haer onbesceidenheit is soe groet datse (. . . ) meer versmaden worden twaren, dan si te voren ye ghewaren Their immodesty is so great that-they . . . more despised become were than they before ever GHE-were ‘Their (i.e. the people’s) immodesty is so great that they have been despised more than they ever were before.’ m. Die sconste vrouwe die ye gewas. The most-beautiful lady that ever GHE-was ‘The most beautiful lady that ever existed’ n. Den makere van al dat ye gewas. The maker of everything that ever GHE-was ‘The maker of everything that ever existed’



Gertjan Postma

In all cases the ghe-prefixed verb is in clause-final position, never in V2position. Sentences like (54a), where was has undergone V2, are quite common, but sentences like (54b), in which ghewas has undergone V2, do not occur. (54) a.

Niemen en was daer soo groot, hine vruchte dagelijcs na sine doot Nobody  was there so great he- feared daily for his death ‘Nobody was there so great that he did not fear for his death every day.’ b. *Niemen en ghewas daer soo groot, hine vruchte dagelijcs na sine doot

It will now be abundantly clear that ‘freezing’ is found with ghe-preterites (see Appendix for more details on freezing and V2). The explanation that was given above crucially depends on the nature of ghe- as an NPI which can take over the function of the reinforcing negator niet ‘not’.23

. Conclusions In one of its uses, the Middle Dutch prefix ghe- is sensitive to the negative polarity of the context. This use as an NPI is most clearly attested in a verb that does not allow other uses of ghe: the copular verb sijn ‘to be’. This NPI use can also be seen in other verbs, although in a less pure form. Negative polarity ghe- has all the properties of verbal negative polarity items in Modern Dutch, such as kunnen uitstaan ‘can stand’, most clearly shown in the infinitival verbal form, which requires low-scope root modality (can and need). The low-scope modality restriction, the preterite restriction and the non-V2 restriction were derived by applying the Neg-criterion to those NPIs that can take over the role of reinforcing negation.

Appendix Freezing is not restricted to NegP: TP can also be a locus of freezing. Koopman (1995) reports freezing in Dutch double-particle constructions with [her (i.e. English re-)+ heavy particle] prefixed to the verb. (i)

a. *Jan heropvoedde zijn dochter John re-up-brought his daughter ‘John re-educated his daughter’

(main clause with V2)

Negative polarity and modality in Middle Dutch ghe-particle constructions 

b. . . . dat Jan zijn dochter heropvoedde (subordinate clause with V-final) . . . that John his daughter re-up-brought ‘that John re-educated his daughter’

Koopman attributes the degraded grammaticality of (i a) to a conflict in stranding/pied-piping properties of her- and the following particle (here: op-). Koopman’s data are not exhaustive, however: similar effects can be found in double-particle constructions in which both particles have identical piedpiping properties separately, like for example her- and be-. (ii) a. *Gisteren herbegonnen ze de procedure yesterday re-be-gun they the procedure ‘Yesterday they restarted the procedure’ b. Toen ze de procedure gisteren herbegonnen, . . . when they the procedure yesterday re-be-gun ‘When they restarted the procedure yesterday, . . . ’

Freezing occurs with her+weak particle, such as be-, if the verb is strong (i.e. shows Ablaut: begin vs. begon). If the verb is weak (i.e. shows no Ablaut), there is no freezing effect. (iii) Gisteren herverdeelden ze de buit yesterday re--divided they the loot ‘They redivided the loot yesterday’.

No ablaut effect is present in her+strong-particle constructions in V2 constructions: they have a degraded grammaticality already. We thus have the patterns in (iv): (iv) *her + heavy Prt +V iff in C (Koopman’s effect) *her+light Prt + strong V iff in C

Similar freezing effects are found in compound verbs consisting of a stressed adverbial or noun as the first element, and a strong verb as the second element: (v) a. *Jan HARDliep twee uur Jan fast-walked two hours ‘Jan ran for two hours.’ b. . . . dat Jan twee uur HARDliep . . . that Jan ran for two hours.’ c. Jan SNELwandelde twee uur Jan fast-strolled two hours ‘Jan speed-walked for two hours.



Gertjan Postma

d. . . . dat Jan twee uur SNELwandelde ‘that Jan speed-walked for two hours.’

Since Ablaut can only occur under main stress, the correct generalisation is that there is a block on double stress if the verbal complex moves to C, but not if it stays in clause-final position. We suggest that it is stressed her- in (i a) and (vi b), in contrast with unstressed her- in (vi a), which causes freezing in (i a) and (vi b), but not in (vi a). (vi) a.

Gisteren herRIEP de politie de uitspraak (unstressed her) yesterday revoked the police the statement ‘Ýesterday the police revoked the statement.’ b. *Gisteren HERbeGONnen ze de procedure (stressed her) yesterday re-began they the procedure Ýesterday they restarted the procedure.’

In some cases, the switch to a non-ablauting paradigm can make a complex verb well-behaved. (vii) a.

. . . omdat Jan toen STOF-ZOOG because Jan then dust-sucked ‘because Jan was hoovering then.’ b. *Jan STOF-ZOOG toen c. Jan STOFzuigde toen

and hence, by analogy: d. omdat Jan toen stofzuigde

Preferably, Koopman’s freezing effects should get an account of the kind discussed in this paper. Since Ablaut expresses tense-oppositions, we suggest that grammatical stress ([+s]), licensing Ablaut, is available in TP. The particle herdoes not have inherent stress but acquires it from T by the two mechanisms available: 1. head adjunction to T (after which the complex can move further to C) or 2. under adjacency to T in SpecTP at spell-out. In both cases, stress conditions are met in the phonology-syntax interface. Adjacency, in fact, doubles the stress assignment: [TP X [T Y]], where X and Y are both adjacent to T. In the presence of her+stressed particle/ablaut, only the second strategy is available, which results in the freezing configuration. This account predicts that triple-stressed verbal complexes should exhibit degraded grammaticality, even when they stay in the freezing position. This prediction is borne out by (viii):

Negative polarity and modality in Middle Dutch ghe-particle constructions 

(viii) a.

omdat deze uitgeverij zulke boeken vandaag de dag niet heruitgeeft (Koopman 1995, 5a) because this publisher such books today the day not re-publishes ‘because this publisher nowadays does not reissue such books.’ b. *. . . omdat deze uitgeverij zulke boeken vroeger niet heruitgaf . . . because this publisher such book earlier not reissued ‘because this publisher did not reissue such books in the past.’ c. . . . omdat ze zulke programma’s regelmatig heruitzenden (Koopman 1995, 5c) . . . because they such programmes regularly re-broadcast ‘because they re-broadcast such programmes regularly.’ d. *. . . omdat ze zulke programma’s toen regelmatig heruitzonden . . . because they such programmes then regularly re-broadcast ‘because they re-broadcast such programmes regularly then.’

Koopman does not give example sentences with the past tense of an Ablauting verb, but their degraded grammaticality follows from the account given here.

Notes * I would like to thank Jack Hoeksema, Teun Hoekstra, Jan Kooij, Marijke van der Wal, two anonymous reviewers, as well as the audience of OOO Leiden for their criticism and comments. . This generalisation applies to the Middle Dutch corpus used by Stoett (1977) and Verwijs & Verdam (1885–1952). This corpus contains mainly narrative texts. However, the nonoccurrence of ghe- in the present tense is not an artefact of this corpus: statistical investigations of the corpus Gysseling, which includes charters and administrative texts, show the same restrictions. . Other verbs lacking ghe- in the participle are brengen ‘bring’, bliven ‘remain’, comen ‘come’, liden, ‘lead’, vinden ‘find’, and worden ‘become’. . Cf. Stoett (1977). See also (22f). . Ghe-3 does occur with moeten, willen, zien en horen if these verbs are embedded under connen or sullen as licensing modals, as in (i)–(ii). (i)

Dat men die stede ghesien en can aldaar hi es That one the place GHE-see  van -there he is ‘That one can see the place where he is.’

(ii) Dat hi anders nyet en sal connen gewillen dan God en wil that he differently  shal can ghe-will than God  will ‘That he will not be able to want anything that God does not want.’

 Gertjan Postma . Participial ghe- does occur with sijn, both in the suppletive form geweest/gewesen and with the root /zij-/: hy heeft ghesijn. The latter is the general form (outside rhyme contexts) in the work of the Middle Dutch author Maerlant For the use of gheweest versus ghesijn, see Postma (1993). . As Verwijs & Verdam note, the verb ghesijn ‘only occurs in the infinitival and the preterite forms’. The same cannot be said for the other ghe-verbs since lexical ghe- en syntactic gheare subsumed under the same lexical item. In the case of sijn, the form ghesijn can only be interpreted as containing ghe-3 . . Exceptions are modals allowing for ghe-3 in the present tense (see Section 5). . Modern Dutch does exhibit productive strategies of negative polarity but not in the verbal domain (see Postma 2000a). . The verb roeken (past tense rochte) ‘care’ is a negative polarity verb in Middle Dutch (cf. also Hoeksema 1997). (i)

Mine roekt wire om belge me-- cares who-there with become-angry ‘I don’t care who gets angry about that’

The root roek- still exists in Modern Dutch roekeloos ‘reckless’, where roek- satisfies its negative polarity by combining with the suffix -loos ‘-less’. . This can also be seen in the rhetorical questions in (i) and (ii): (i)

Hoe gemochtic Gode dat gebreken? How GHE-might I God- that fail ‘How could I withhold that from God?’

(ii) Hoe gescal ic dit emmermeere ghebeteren moge? How GHE-shall I this ever improve may ‘How will I ever be able to expiate this’ . A parallel phenomenon is found in Modern Dutch in the so-called IPP (Infinitivus Pro Participio) effect, where an underlying participle (i.e. a form containing the equivalent of Middle Dutch ghe-2 ) shows up as an infinitive: (i)

omdat ik hem dat heb *gezien/zien doen because I him that have seen/see do ‘because I have seen him do that.’

. Monotone-decreasing contexts can be formulated as functions that satisfy particular formal equations. The four formal properties of contexts defining a classification of quantifiers are listed in (i). (i)

a. b. c. d.

f(X)∪f(Y)⊆ f(X∩Y) f(X∪Y)⊆ f(X)∩f(Y) f(X)∩f(Y)⊆ f(X∪Y) f(X∩Y)⊆ f(X)∪f(Y)

L1 L2 L3 L4

{L1 and L2} can be considered the core properties of negative polarity.

Negative polarity and modality in Middle Dutch ghe-particle constructions  . Van der Wouden (1994: 37) adds L4 as a property relevant to the typology of NPIs, and thus distinguishes two additional types of NPIs: those that impose {L1,L2,L4} on their contexts, and those that impose {L1,L2,L3,L4} on their contexts. The set of NPIs of the first type is completely or virtually empty (Van der Wouden 1996:59ff), while the set of NPIs of the second type is strongly collocational (e.g. niet pluis ‘fishy’, niet mals ‘trenchant’). It cannot be ruled out that these are collocations which merely happen to contain the lexeme niet ‘not’. . This absence might well be due to structural reasons. The impossibility of [not every N] can also be seen in Modern French: (i)

a. *Pas tout Hollandais est un avare! b. Tout Hollandais n’est pas un avare! Not every Dutch person is a penny-pincher!

Middle Dutch and Modern French show similar behaviour, and contrast with the other Germanic and Romance languages. This might be related to the obligatory negative doubling in subject position in French en Middle Dutch (see (34)). The French negative clitic ne and the Middle Dutch negative clitic en must be licensed under c-command by a negative operator at LF. It follows that the operator may not be embedded as in [niet elke]. . I abstract away from V2 effects here. It should also be noted that Middle Dutch is like Yiddish and Middle German in that it allows for both OV and VO surface orders. Here I assume an underlying VO order, as argued in Zwart (1986). . Further investigation will be necessary to show whether this lower NegP is the syntactic realisation of the bounded scale present in modal constructions (Barbiers 1995, Chapter 5; Barbiers (this volume), Section 8). Presumably, negative polarity is a special case of scalar complementation. If this is the case, the striking interaction of (low scope) modality and negative polarity (as discussed in this paper) would be identical to the interaction of (low scope) modality and scalarity studied by Barbiers. . Both in the subject-directed reading and the non-subject-directed reading, the scope relations with respect to negation are as indicated. . To function as a high-scope modal, a strong stress on not appears to be necessary: (i)

John can, of course, also not do it

. At first sight, it may be tempting to derive the low-scope modality from the fact that the modal is part of the verbal expression. But this turns out to be unattractive, for three reasons. First, it does not provide a deep explanation for the modality of verbal expression at all. Second, it is uitstaan rather than kunnen uitstaan that is listed as an NPI, as can be seen from the paradigm in (i). (i)

a.

b.

Niemand kan hem uitstaan nobody can him out-stand ‘Nobody can stand him’ Hij is niet om uit te staan He is not  out to stand ‘He is unbearable’



Gertjan Postma

c.

Hij is onuitstaanbaar he is un-out-standable ‘He is unbearable’

The absence of lexical modality in (i b, c) shows that uitstaan is the NPI rather than kunnen uitstaan, which does not require kunnen lexically, but selects modality on a more abstract level. Thirdly, in the case of Middle Dutch ghe-, which behaves in all respects as a NPI verbal expression, CAN modality is not really required when it is attached to preterites. . In the non-root, i.e. epistemic, interpretation, this sentence is correct. Non-root interpretations are strongly favoured when there is a perfect tense in the embedded clause, as in (i). (i)

Niemand moet er ene bal van begrepen hebben nobody must there one ball of understood have ‘This must have been unclear to everybody’.

If we adopt the standard view that only epistemic readings involve a raising construction (see for example Klooster 1986), these judgements follow from the fact that A-movement allows for scopal reconstruction (Fox 1999; Bobaljik & Wurmbrand 1999). (ii) a. b.

niemand moet [ t er ene bal van begrepen hebben] SS – moet [ niemand er ene bal van begrepen hebben] LF

At LF, the relevant IP with negation does not contain a modal. For a different view on the structural opposition between root and non-root modals, see Barbiers (1995, Chapter 5). . The pattern is largely similar to what is found in modern Dutch, as shown in (i)–(iii): (i) *Alle mensen die er ooit lopen all people who there ever walk (ii) Alle mensen die er ooit zullen lopen all people who there ever will walk (iii) Alle mensen die er ooit liepen all people who there ever walked A parallel account can be given if we assume that Modern Dutch ooit ‘ever’ has absorbed gheat the time that the NPI ghe- disappeared in the late 15th century. While absorbing the verbal ghe- particle, the newly created NPI ooit ‘ever’ came to mark the VP as expressing negative polarity. As a result, the VP must move to SpecNegP, and the NEG-criterion requires that there be a head to move to Neg. (The phonological shape of ooit is noteworthy: all Dutch words ending in -ooit are verbal and composite, i.e. -ooi- + -t, e.g. gooit ‘(he) throws’, dooit ‘thaws’, rooit ‘digs’, strooit ‘strews’, looit ‘tans’, etc. Moreover, words ending in -ooit cannot be simplex: no new noun can exist such as *mijn plooit “my PLOOIT”. This may be a further argument for the composite nature of ooit.) . I omitted one example with a strong Middle High-German colouring, although it fully complies with the generalisation.

Negative polarity and modality in Middle Dutch ghe-particle constructions  . The Middle Dutch verb roeken ‘care’ is an NPI which can take over the role of reinforcing negator just like ghe-V. Nevertheless, it can undergo V2, as shown in (i) (i)

Mi en roeket (niet) ware si gaen Me neg care (not) where they go ‘I do not care where they go’

It can be shown, however, that it is the WH-clausal complement that functions as the NPI moving to SpecNegP. No double role is played by the verb and, hence, no freezing effects show up. See Postma (2002) for a full treatment of verbs with single clitic negation.

References Barbiers, S. (1995). The Syntax of Interpretation. Doctoral dissertation, University of Leiden. Beukema, F. & W. van der Wurff (this volume). Modals, Objects and Negation in Late Middle-English. Bobaljik, J. & S. Wurmbrand (1999). Modals, raising, and A-reconstruction. Paper presented at the University of Leiden, October 1999. Burzio, L. (1986). Italian Syntax, Dordrecht: Reidel. Chomsky, N. (1982). Some Concepts and Consequences of the Theory of Government and Binding, Cambridge (MA): MIT Press. Fox, D. (1999). Reconstruction, binding theory, and the interpretation of chains. Linguistic Inquiry, 30, 157–196. Gestel, F. van, J. Nijen Twilhaar, T. Rinkel and F. Weerman (1992). Oude zinnen: Grammaticale Analyse van het Nederlands tussen 1200–1700. Leiden: Nijhoff. Haegeman, L. & R. Zanuttini (1991). Negative heads and the Neg criterion. The Linguistic Review, 8, 233–251. Haegeman, L. (1995). The Syntax of Negation. Cambridge: Cambridge University Press. Hoeksema, J. (1983). Negative polarity and the comparative. Natural Language and Linguistic Theory, 1, 403–434. Hoeksema, J. (1997). Negation and negative concord in Middle Dutch. In D. Forget, P. Hirschbühler, F. Martineau and M.-L. Rivero (Eds.), Negation and Polarity: Syntax and Semantics (pp. 139–156). Amsterdam: Benjamins. Jackendoff, R. (1977). X-bar Syntax: A Study of Phrase Structure. Cambridge, Mass: MIT Press. Kayne, R. (1999). Overt vs. covert movement. Syntax, 1, 128–191. Kemenade, A. van (2000). Jespersen’s cycle revisited: Formal properties of grammaticalization. Ms., University of Nijmegen. Klima, E. (1964). Negation in English. In J. Fodor and J. Katz (Eds.), The Structure of Language (pp. 246–323). Englewood Cliffs (NJ): Prentice-Hall. Klooster, W. (1986). Problemen met complementen. TABU, 16, 112–132. Klooster, W. (1993). Negatieve zinnen. Tijdschrift voor Taalwetenschap, 2, 119–143.



Gertjan Postma

Koopman, H. (1995). On verbs that fail to undergo V-second. Linguistic Inquiry, 26, 137–163. Ladusaw, W. (1980). Polarity Sensitivity as Inherent Scope Relations. Doctoral dissertation, University of Texas. Laka, I. (1994). On the Syntax of Negation. New York: Garland. McCloskey, J. (1993). A crude test for unaccusativity in English. In G. Pullum and E. Potsdam (Eds.), Syntax at Santa Cruz. Santa Cruz: Linguistic Research Center/UCSC. Overdiep, G.S. (1914). De Vormen van het Aoristisch Praeteritum in de Middelnederlandsche Epische Poezie. Rotterdam: Brusse. Pollock, J.-Y. (1989). Verb Movement, Universal Grammar and the Structure of IP. Linguistic Inquiry, 20, 365–424. Postma, G. (1993). The sytax of the morphological defectivity of BE. HIL Manuscripts, 3, 31–67. Postma, G. (2000a). Negative polarity and the syntax of taboo. In J. Hoeksema, H. Rullmann, V. Sanchez-Valencia and T. van der Wouden (Eds.), Perspectives on Negation and Polarity Items (pp. 285–322). Amsterdam: Benjamins. Postma, G. (2000b). Anthropological taboo, grammatical mismatches, and the dynamics of language. Paper read at Leiden University, HIL seminar, 5 April 2000. Postma, G. (2002). De enkelvoudige clitische negatie in het Middelnederlands en de Jespersen-cyclus. Nederlandse Taalkunde, 7, 44–82. Rizzi, L. (1982). Issues in Italian Syntax. Dordrecht: Foris Publications. Rizzi, L. (1991). Residual verb second and the WH-criterion. University of Geneva. Technical Reports in Formal and Computational Linguistics, 2. Sanchez, V., T. van der Wouden and F. Zwarts (1993). Polarity and the flow of time. Paper read at the Amsterdam Colloquium 9. Seuren, P. (1975). Tussen Taal en Denken. Utrecht: Oosthoek. Stoett, F.A. (1977). Middelnederlandse Spraakkunst – Syntaxis. (3rd ed.). Den Haag: Martinus Nijhoff. Streitberg, W. (1891). Perfective und imperfective Aktionsart im Germanischen. Beiträge zur Geschichte der Deutschen Sprache und Literatur, 15, 70–177. Swaay, H.A.J. van (1899). Het Prefix ga-, gi-, ge-, en de Aktionsart. Doctoral dissertation, University of Utrecht. Swaay, H.A.J. van (1909). De perfectiva simplicia. Tijdschrift voor Nederlandse Taal- en Letterkunde, 28, 39ff. Verwijs, E. and J. Verdam (1885–1952). Middelnederlandsch Woordenboek. Den Haag: Martinus Nijhoff. Weerman, F. (1988). Moet kunnen: Middelnederlandse zinnen zonder subject. Nieuwe Taalgids, 81, 289–310. Wouden, T. van der (1994). Negative Contexts. Doctoral dissertation, University of Groningen. Wurff, W. van der (1999). Objects and verbs in modern Icelandic and fifteenth-century English: A word order parallel and its causes. Lingua, 109, 237–265. Zwarts, F. (1986). Categoriale Grammatica en Algebraische Semantiek: Een Onderzoek naar Negatie en Polariteit in het Nederlands. Doctoral dissertation, University of Groningen.

(Negative) Imperatives in Slovene* Milena Milojevi´c Sheppard and Marija Golden University of Ljubljana

This paper provides a syntactic account of Slovene imperative sentences. Two typologically relevant aspects of imperatives are considered: (i) negation; (ii) pronominal clitic placement. We attempt to account for the Slovene data in terms of strength of the imperative feature located in Co . We show that in this approach the unacceptability of sentence-initial Clitic– Imperative Verb order is unexpected since Slovene clitics can be either enclitic or proclitic.

.

Introduction

In this paper we investigate Slovene imperative sentences with verbs inflected for imperative mood. Our principal aim is to present the relevant Slovene data against the background of recent generative approaches to cross-linguistic variation in imperative systems, and to call attention to two theoretically interesting properties of Slovene: (i) while clitics are otherwise allowed to open a sentence, in imperatives this possibility vanishes – the obligatory order is VerbClitic; (ii) contrary to standard assumptions that the imperative is strictly a main-clause item, imperatives in Slovene can be embedded. For languages whose mood systems provide their verbs with an imperative paradigm, Rivero and Terzi (1995) propose a typology with respect to whether the imperative verb exhibits a syntax distinct from the indicative. They discuss in detail two such distinguishing properties: (i) sentential negation, and (ii) clitic placement. Thus, Modern Greek, Spanish and Italian belong to Class I languages, where the syntax of imperative verbs differs from the syntax of indicative verbs. With respect to the first property, sentential negation, these languages allow only affirmative imperatives, as in (1a); prohibitions cannot be expressed by negating the imperative verb, as shown in (1b). With respect to the second property, clitic placement, Class I languages allow the clitic to precede the indicative verb as in (2a), but they do not allow the clitic to precede the



Milena Milojevi´c Sheppard and Marija Golden

imperative verb, as shown in (2b). In the French example (2b), the pronominal clitic le must follow the verb, when it is imperative, although it precedes it when it is indicative or subjunctive. (1) a.

Léelo! read..2-it ‘Read it!’ (Spanish; Rivero & Terzi 1995: (4b)) b. *No lee!  read..2 Intended meaning: ‘Don’t read!’ (Spanish; Rivero & Terzi 1995: (2b))

(2) a.

Lo leiste/*Leístelo it read-.2 ‘You read it.’ b. Faites le!/*Le faites! Do..2 it ‘Do it.’

(Spanish; Rivero & Terzi 1995: (4d))

(French)

On the other hand, in Class II languages, among them Bulgarian and SerboCroatian (SC), the syntax of imperative verbs does not differ from the syntax of indicative verbs. Imperative verbs can be negated to express prohibition, as in (3b); clitics are placed according to the second-position constraint, irrespective of verbal morphology, as shown in (4). Ne cˇ itate.  read..2 ‘You (.) are not reading.’ b. Ne cˇ itajte!  read..2 ‘Do not read.’ ˇ (4) a. Citate je. Read..2 it ‘You (.) are reading it.’ ˇ b. Citajte je! read..2 it ‘You () read it!’

(3) a.

(SC; Rivero & Terzi 1995: (11b))

(SC; Rivero & Terzi 1995: (11a))

(SC; Rivero & Terzi 1995: (12b))

(SC; Rivero & Terzi 1995: (12a))

Our account of Slovene imperatives begins with a brief overview of the relevant parts of Rivero and Terzi’s (1995) syntactic analysis and Han’s (1998) semantic proposal for the (un)availability of negative imperatives (Section 2). In the discussion that follows we shall argue that both approaches fail to account for the full range of Slovene data. Section 3 outlines the basic descriptive properties

(Negative) Imperatives in Slovene 

of Slovene sentential negation and imperatives and examines the syntax of the Slovene imperative verb with respect to pronominal clitic placement. We show that root imperatives are different from indicatives in that at the beginning of the sentence, clitics cannot precede the imperative verb. In Section 4 we focus on negation and clitic placement. With respect to these two properties, Slovene looks like a mixed language. It belongs together with Serbo-Croatian and Bulgarian (Class II) in having negative imperatives and Wackernagel clitics, yet it appears to have one property of Class I languages, namely, clitic placement in imperatives is not identical to the clitic placement in indicatives. In view of the fact that Slovene clitics are prosodically either enclitic or proclitic, the crucial question is what makes the order sentence-initial Clitic-Verb unacceptable in root imperatives.

. (Un)availability of negative imperatives . A syntactic account: Rivero and Terzi (1995) Rivero and Terzi (1995) attribute the unavailability in Class I languages of negative imperatives and the unique position of imperative verbs with respect to pronominal clitics to the location of the imperative operator in syntax. In Class I, a strong V-feature in Co encodes the logical mood (illocutionary force) of imperatives, and the verb, inflected for the imperative morphology must raise overtly to Co , obeying Greed (Chomsky 1994: 14). The absence of negative imperatives is attributed by Rivero and Terzi to the Head Movement Constraint. The Neg head acts as a relativized minimality barrier, past which Vo cannot raise, and a strong feature thus remains unchecked. For Class II languages, Rivero and Terzi propose not only that the root Co is featureless, but that all V-features, including the imperative V-feature, are in Io . They reserve Co as a last resort landing site of the verb (both indicative and imperative) to provide the PF-licensing of second-position (2P) clitics. If no phrase fronts, and the clitic is stranded sentence-initially, then all verbs move to Co . This is an instance of last resort movement, obeying Lasnik’s (1993: 12–13) Enlightened Self-Interest Principle. In both groups of languages, the imperative verb may end up in Co . However, in Class I languages, the imperative Vo is in Co as a result of featurechecking (Greed); in Class II, the imperative verb ends in Co as a last resort to provide the first-position (1P) constituent for the Wackernagel pronominal clitics.



Milena Milojevi´c Sheppard and Marija Golden

Class I: [CP Co [FP Clo [NegP Nego [IP V] ] ]] [imperative] b. Class II: [CP [Co Vi ] [WP Clo [IP ti ] ] ]] or [CP [SpecCP X” [C Ø]] [WP Clo [NegP Neg [IP V] ] ]]

(5) a.

. A semantic account: Han (1998) Han (1998) derives the unavailability of negative imperatives in Class I languages from their ill-formed semantic representation, with respect to two sentential operators, the negative operator and the imperative operator. According to her analysis negative imperatives are unavailable in languages where the negative operator ends up having syntactic scope over the imperative operator. In such structures, the directive force contributed by the imperative operator would be negated, which would not be the intended meaning. Negating the imperative does not remove the directive illocutionary force. CP

(6)

C’ Co

IP

Co

I

Imp-Op

Neg

I

In the LF structure suggested in (6), negation asymmetrically scopes over the imperative operator.1 This structure occurs when the negation marker cliticises onto the imperative verb, and both raise overtly to Co for feature checking. In languages without negative imperatives, the preverbal negation marker and the verb form a unit in syntax. If negation is present, it moves with the verb as a unit, due to its clitic-like nature. If the Neg-V complex moves to Co for imperative mood feature checking, the imperative in Co ends up within the scope of Neg.

(Negative) Imperatives in Slovene 

. Negation, imperatives and clitics in Slovene . Sentential negation Slovene expresses sentential negation by a negation particle (ne) in preverbal position. In discourse-neutral sentences, the negation particle appears early in the clause and is adjacent to the tensed verb. In the embedded clause (7a), the negation particle occurs after the complementiser, the clitics, and the subject DP. (7) a.

da mu ga Janez še ne zaupa. that him.. him/it.. Janez yet not trust..3 ‘That Janez doesn’t yet entrust him/it to him.’ b. *da mu ga Janez ne še zaupa.

As (7b) shows, the negation particle ne cannot be separated from the finite verb even by clitic-like adverbials (še). The negation marker is not part of the clitic cluster; in (7a), it is separated from the pronominal clitic cluster by the subject and an adverbial. In the spirit of Pollock (1989) and much subsequent work, we propose to treat the sentential negation marker as a functional head, projecting to NegP. The position of the negation marker is determined by the position of the NegP, which is a complement of AgrS and dominates TP. The structure CP > AgrSP > NegP > TP for the Slovene clause is supported by the unmarked order in indicative interrogative sentences such as (8), in which the subject DP precedes the negation and the tensed verb. (8) [C Ali ga [AgrSP [Spec AgrSP Špela] [NegP ne [TP [VP bere . . . ? Špela not whether it.. read ‘Doesn’t Špela read it?’

The negated tensed verb functions as a 1P constituent with respect to clitic placement, as shown in (9). When the verb moves to Co – as in (9) below, where it hosts 2P pronominal clitics – the negation marker moves along with it. (9) is thus evidence that the complex verbal head [Neg+V ne dam] counts as a single syntactic constituent with respect to the 2P constraint on Slovene clitic placement. (9) [Ne dam] =ti =ga. [CP [ C’ [Co ne+dami [Co ti ga [ AgrSP ti [ NegP ti [TP ti [VP ti ]]]]]]]] ‘I won’t give it to you.’



Milena Milojevi´c Sheppard and Marija Golden

The placement of the negation particle in negative verbal questions is shown in (10). In a discourse-neutral verbal question like (10b), the auxiliary clitic si comes first in the clitic cluster (occupying the position of Aux1 in clitic cluster (10a)). In a discourse-neutral negative verbal question (10c), the auxiliary clitic si no longer comes first in the clitic cluster. As the unacceptability of (10d) shows, the tensed auxiliary clitic cannot appear left of the negation marker ne. (10) a. Aux1 >Reflexive>Dative>Accusative>Genitive>Aux2 b. Ali si mu ga whether are. him.. it.. pokazal? (* . . . mu ga si . . .) shown. ‘Have you shown it to him?’ c. Ali mu ga nisi pokazal? ‘Haven’t you shown it to him?’ d. *Ali si mu ga ne pokazal?

Pronominal clitics never intervene between the negation marker and the tensed verbal form. The adjacency requirement between the negation marker and the verb is shared by indicative (11) and imperative verb alike (12). (11) a.

Ne opraviˇcuje se mi veˇc. not excuse..3  me.. more ‘He doesn’t apologise to me any more.’ b. *Ne se mi veˇc opraviˇcuje.

(12) a.

Ne opraviˇcuj se mi veˇc! not excuse..2  me.. more ‘Do not apologise to me any more!’ b. *Ne se mi veˇc opraviˇcuj!

. Imperatives In Slovene, the imperative is one of the three morphosyntactic mood categories expressed in the inflectional paradigm of the verb. The paradigm in (13) shows that the imperative forms of the Slovene verb are not morphologically identical to its indicative forms. Furthermore, there are selectional restrictions on the imperative verb such that it can have only 2nd person singular and 1st and 2nd person plural and dual subjects.

(Negative) Imperatives in Slovene 

(13)

Indicative and Imperative Forms 2-Sg

1-Pl

2-Pl

1-Du

present indicative delaš delamo delate delava imperative delaj delajmo delajte delajva ‘you.work’ ‘let.us.work’ ‘you.work’ ‘let.us.two. work’

2-Du delata delajta ‘you.two. work’

In (14) we show that in Slovene imperatives can be embedded. They are not restricted to appearing only under reporting verbs, as in (14a), but occur in such unambiguous structures of embedding as argument clauses (14b), restrictive relative clauses (14c) and adnominal complement clauses (14d). (14) a.

Rekel je, da delaj bolje. said is that work..2 better ‘He said that you must work better.’ b. Peter vztraja, da pridi jutri. Peter insists that come..2 tomorrow ‘Peter insists that you must come tomorrow.’ c. To je film, ki si ga oglej this is film which yourself.. it.. see..2 cˇ imprej. as.soon ‘This is a film which you must see as soon as you can.’ d. Zakaj te moj nasvet, da bodi pameten, why you.. my advice that be..2 sensible tako jezi? so angers ‘Why does my advice that you must be sensible make you so angry?’

The imperative feature differs from the V-features of other morphological moods in that it signals logical mood. It is the formal ingredient that contributes to the determination of illocutionary force, i.e. an illocutionary (logical mood) operator (Rivero and Terzi 1995). However, contrary to Rivero and Terzi’s claim that imperative verbs can only be used in imperative sentences, as their morphology carries intrinsic logical mood, Slovene imperatives are not restricted to sentences with imperative illocutionary force, as shown in (15). (15) a.

Potem pa še komu zaupaj! then   whom trust.



Milena Milojevi´c Sheppard and Marija Golden

‘And then trust anybody.’ (Intended meaning: ‘It is impossible to trust anybody.’) b. Reci bedaku, da je pameten, pa ti tell. fool. that is clever  you. bo verjel. will believe ‘Tell a fool that he is clever and he’ll believe you.’ (Intended meaning: ‘If you tell a fool . . . .’)

These data suggest that the presence of an imperative verb does not necessarily correlate with imperative illocutionary force. We identify the root Co as the natural syntactic location in the clause for the logical mood. . Clitic placement As argued in Golden and Milojevi´c Sheppard (2000), while both SerboCroatian and Slovene are 2P clitic languages, only Serbo-Croatian clitics need to be lexically specified as prosodically inherently enclitic (Boškovi´c 2000). Slovene clitics may also be proclitic, and in this respect resemble clitics in Czech (Toman 1996). We also suggested that the 2P of Slovene pronominal clitics is to be syntactically identified as right adjunction to Co , and concluded that whereas the relevant domain for Serbo-Croatian 2P clitics is the intonation phrase, for Slovene clitics, it is CP. (16) Mu ga boš pokazal? him.. it.. you.will. show. ‘Will you show it to him?’

(16) is a verbal question beginning with a pronominal clitic cluster. The fact that the clitic cluster in (16) is sentence-initial cannot be attributed to pragmatically controlled omission of the overt interrogative complementiser (ali ‘whether’: Ali mu ga boš pokazal?) since Slovene clitics are frequently sentenceinitial in declaratives, as well. Thus (17a) can be a response to a question such as Kje je pismo? ‘Where is the letter?’, and (17b) is a threat. (17) a.

Sem ga že oddal. am.. it.. already mailed ‘I have already mailed it.’ b. Ti bom že pokazal! you.. I.will. already show. ‘I will teach you a lesson!’

(Negative) Imperatives in Slovene 

Slovene clitic pronouns can be prosodically either enclitic or proclitic, yet we do not find them in initial position in root imperatives. The imperative verb cannot follow a sentence-initial pronominal clitic (18a, b) whereas the indicative verb can ((16), (17b)). Note that in embedded clauses, there is no difference in clitic placement, irrespective of whether the verb is indicative or imperative (18c). (18) a. *Ga poslušaj, cˇ e hoˇceš! him.. you.listen. if you.want. b. Poslušaj ga, cˇ e hoˇceš. ‘Listen to him, if you want.’ c. Vztrajal je, da ga pokliˇci/ insisted. is...3 that him.. call..2 pokliˇce sam. call..3 himself ‘He insisted that you must call him yourself/he calls him himself.’

The proclisis in (16) and (17) might be assumed to be brought about by (pragmatically controlled) ellipsis of 1P material: the complementiser ali in (16) and the subject NP in (17). However, a pragmatic approach cannot account for the unacceptability of (18a). While it is true that Slovene has no lexical complementiser in root imperative clauses, the subject of root imperatives is, as in the case of non-imperatives, normally unexpressed, so the question remains unanswered: why can the imperative response in (19b) not have the ordering of the indicative sentence (19a)? (19)

Kje so moja oˇcala? ‘Where are my glasses?’ a. Jih že išˇcem. them..  I.look.for ‘I am already looking for them.’ b. *Jih išˇci! them.. look.for..2 ‘Look for them!’

. Analysis Intuitively, Slovene looks like a candidate for a mixed language with respect to the properties investigated in Rivero and Terzi’s (1995) typology. The fact that it has negative imperatives puts Slovene in Class II languages. Slovene is also



Milena Milojevi´c Sheppard and Marija Golden

a Wackernagel language, yet it sides with Class I languages in that clitic placement in imperatives is not identical to the clitic placement in indicatives. Crucially, clitics can appear initially in indicative root sentences but they cannot be initial in root imperatives. . Alternative 1 Let us first try to account for the Slovene data in terms of Rivero and Terzi’s analysis of imperatives in Class II languages. Their approach turns out to be problematic for several reasons. i. In Rivero and Terzi’s analysis of Class II imperatives, Vo -to-Co raising is considered to be a last-resort, phonologically-triggered operation. Furthermore, they crucially assume the mutually exclusive application of Vo -to-Co raising and Topicalisation. In Slovene, however, the rule that raises Vo to Co cannot be a last-resort movement, since the raised verb can co-occur with a topicalised phrase, as (20a, b) shows. (20) a.

Vrni mu knjigo še return..2 him.. book.  ‘Return the book to him today!’ b. Knjigo vrni mu še book. return..2 him..  [Spec CP [Knjigo] [Co vrni [C mu [AgrSP . . .

danes. today danes. today

ii. Vo -to-Co raising is argued to be a last-resort movement, affecting all verbs, regardless of their morphology. In Slovene, however, the Verb-Clitic order is obligatory only with imperative verbs, as shown in (18a, b). iii. Vo -to-Co is conceived of as a last-resort movement with a phonological trigger. Vo moves to Co in order to provide phonological support for clitics. As has already been noted, Slovene clitics may, in contrast to SC clitics, be prosodically either enclitic or proclitic. As a last-resort operation, Vo -to-Co is obligatory; if it does not apply, the derivation crashes. In Slovene, as shown in (16) and (17), this is not always the case: Vo -to-Co raising may be optional. iv. As a last-resort movement, Vo -to-Co raising should apply only when necessary; unnecessary applications are ruled out by the principle of economy of derivation. If Vo moves to Co , with SpecCP being filled, the resulting sentence is predicted to be ungrammatical. In Slovene, this prediction is not borne out, witness the grammaticality of (20b).

(Negative) Imperatives in Slovene 

v. Rivero and Terzi propose that in Class I languages, Co is the locus of imperative force, whereas in Class II languages, Co is empty, reserved for last-resort V-movement to prevent clitics from being initial in the sentence. For Slovene, with the dual nature of its clitics, the motivation for an empty Co is absent. Furthermore, it must be noted, that Co is commonly assumed to be the locus of the question-force operator, also in Class I languages. In the simplest theory, one would expect that operators signalling the illocutionary force of a sentence should have a single location in the CP clausal structure, with root Co being the obvious candidate. . Alternative 2 Proceeding from the assumption that root Co is the locus of the imperative mood operator irrespective of language type, let us now examine Han’s (1998) approach in the light of Slovene data. Han’s analysis predicts that negative imperatives will be absent in languages in which the negation marker and the imperative verb form a unit in syntax, and raise as a unit to Co . Contrary to expectation, the two Slavic languages that she discusses, Bulgarian and Serbo-Croatian, do have negative imperatives, in spite of the fact that the negation marker affixes onto the imperative verb. Han suggests that in these two languages the imperative verb remains in Io in syntax and then moves alone to Co at LF. Since the negation thus stays low in the CP structure, it does not scope over the imperative operator at LF, rendering negative imperatives possible. To account for the obligatory Imperative Verb-Clitic order in Bulgarian and Serbo-Croatian, Han draws upon Embick and Izvorski’s (1997) analysis of Verb-Clitic orders in Slavic. They propose that when the clitic has been stranded sentence-initially, Morphological Merger (Marantz 1989) applies on the branch to PF as a last-resort post-syntactic mechanism to invert the clitic with an adjacent element, thus resolving the clitic’s leftward phonological dependency. Following Embick and Izvorski, Han argues that in Bulgarian and Serbo-Croatian ((21a) and (22a) respectively) the clitic has affixed onto the imperative verb in Io at a post-syntactic level. The presence of the negation marker in (21b) makes Morphological Merger unnecessary. The obligatory encliticisation of the clitic in negative imperatives in Serbo-Croatian (22b) is to be attributed, according to Han, to an independent constraint of the language.



Milena Milojevi´c Sheppard and Marija Golden

ˇ Ceti ja! read.2. it ‘Read it!’ b. Ne ja cˇ eti!  it read.2. ‘Don’t read it!’ ˇ (22) a. Citaj je! read.2. it ‘Read it!’ b. Ne cˇ itaj je!  read.2. it ‘Don’t read it!’ (21) a.

(Bulgarian; Han 1998: (22))

(SC; Han 1998: (23))

In Slovene, the distribution of imperative verbs with respect to clitic placement and negation is identical to that in Serbo-Croatian (cf. (23) and (24)). (23) a.

b. c.

d. (24) a.

Beri jo! read.2. it ‘Read it!’ *Jo beri! Ne beri je!  read.2. it ‘Don’t read it!’ *Ne jo beri! ˇ Citaj je!

(Slovene)

(SC)

read.2. it ‘Read it!’ b. *Je cˇ itaj! c. Ne cˇ itaj je!  read.2. it ‘Don’t read it!’ d. *Ne je cˇ itaj!

Adopting Han’s idea of LF imperative verb movement to Co and applying the standard feature-checking theory (Chomsky 1995), let us now consider a possible derivation of Slovene (negative) imperatives. Given the clause structure CP > AgrSP > Neg > TP, (negative) imperatives would be derived as follows: the imperative verb, having raised overtly out of VP to T (for T-feature checking), attaches (in negative sentences) to Neg, the two forming a single complex head, which subsequently raises to AgrSP (for phi-feature checking). All the movements are overt since all the features of

(Negative) Imperatives in Slovene 

functional heads within AgrsP are assumed to be strong. The weak imperative feature of Co is then checked at LF, by the formal V-feature raising to Co .2 The availability of negative imperatives is correctly predicted: as LF movement involves formal features only, the V-feature moves to Co alone, leaving the negation marker behind, hence Neg does not scope over the imperative operator in Co . Negative imperatives are thus not ruled out. The presence of a weak imperative feature in Co predicts that imperative verbs distribute like indicative verbs with respect to negation and clitic placement. This prediction is borne out in the case of negation (cf. 3.1) as well as clitic placement when the clitic is not sentence-intial (cf. 3.3). However, with sentence-inital clitics, the Clitic-Verb order in imperatives is unacceptable, but is restored to full acceptability when an item is inserted in 1P (cf. (23b), repeated below as (25a), and (25b)). Since Slovene clitics can be prosodically either enclitic or proclitic, this cannot be explained by appealing to a phonological constraint, i.e. the leftward phonological dependency of the clitic, to be resolved by a post-syntactic mechanism of Morphological Merger, as proposed by Han (1998) for Bulgarian and Serbo-Croatian. (25) a. *Jo beri! it read.2. ‘Read it!’ b. Sam jo beri! yourself it read.2. ‘Read it yourself!’

In sum, neither of the two approaches discussed above (Rivero and Terzi 1995; and Han 1998) is able to account for the full range of Slovene data. Specifically, neither manages to get the crucial datum – the unacceptability of sentenceinitial Clitic-Imperative Verb order – under control.

. Conclusion In this paper, we have examined the syntax of Slovene imperatives. We have focused on negative imperatives and clitic placement in imperatives, negation and clitic placement being the two aspects of imperatives in which typological differences in the syntax of imperatives have been observed. An analysis of Slovene imperatives in terms of standard feature-checking theory has been shown to be able to account for the relevant data concerning negative imperatives as well as clitic placement when clitics do not occur in sentence-



Milena Milojevi´c Sheppard and Marija Golden

intial position. However, the question of what makes the sentence-inital CliticImperative Verb order unacceptable remains unsolved and merits further research.

Notes * We would like to thank Frits Beukema and two anonymous reviewers for valuable comments and suggestions. . In (6), under Kayne’s (1994) definition of c-command, Neg c-commands Co but Co does not c-command Neg because Co does not exclude Neg (a segment of Co dominates Neg). . The motivation for positing strong V-features of functional projections within AgrsP might be sought in the rich morphology of the Slovene verb. However, the same motivation for the strength of the imperative feature in Co is absent. The feature involved is a logical mood feature, different from ‘ordinary’ morphological features that are related, for example, to Tense and Agr. It would seem that it is more naturally expected to be strong in languages where morphological mood always correlates one-to-one with the logical mood (cf. Rivero and Terzi’s Class I languages). Slovene is not such a language; verbs with imperative morphology are not restricted to sentences with imperative logical mood (cf. (15)).

References Boškovi´c, Ž. (2000). Second position cliticisation: Syntax and/or phonology? In F. Beukema and M. den Dikken (Eds.), Clitic Phenomena in European Languages (pp. 71–119). Amsterdam: John Benjamins. Chomsky, N. (1994). Bare phrase structure. MIT Occasional Papers in Linguistics, 5. Chomsky, N. (1995). The Minimalist Program. Cambridge, Mass.: MIT Press. Embick, D. and R. Izvorski (1997). Participle-auxiliary word orders in Slavic. In W. Browne, E. Dornisch, N. Kondrashova and D. Zec (Eds.), Annual Workshop on Formal Approaches to Slavic Linguistics: The Cornell Meeting 1995 (pp. 210–239). Ann Arbor: Michigan Slavic Publications. Golden, M. and M. Milojevi´c Sheppard (2000). Slovene pronominal clitics. In F. Beukema and M. den Dikken (Eds.), Clitic Phenomena in European Languages (pp. 191–207). Amsterdam: John Benjamins. Han, C.-H. (1998). Cross-linguistic variation in the compatibility of negation and imperatives. In S. Blake, E.-S. Kim and K. Shahin (Eds.), West Coast Conference on Formal Linguistics Proceedings 17 (pp. 1–17). Stanford: CSLI Publications. Kayne, R.S. (1994). The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Lasnik, H. (1993). Lectures on Minimalist syntax. University of Connecticut Working Papers in Linguistics, Occasional Papers, 1. Marantz, A. (1989). Clitics and phrase structure. In M. Baltin and A. Kroch (Eds.), Alternative Conceptions of Phrase Structure (pp. 99–116). Chicago: Chicago Press.

(Negative) Imperatives in Slovene 

Pollock, J.-Y. (1989). Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry, 20, 365–424. Rivero, M.L. and A. Terzi (1995). Imperatives, V-movement and logical mood. Journal of Linguistics, 31, 301–323. Toman, J. (1996). A Note on clitics and prosody. In A. Halpern and A.M. Zwicky (Eds.), Approaching Second: Second Position Clitics and Related Phenomena (pp. 505–510). Stanford: CSLI Publications.

Modality and mood in Macedonian Olga Mišeska Tomi´c University of Novi Sad/University of Leiden

The paper focuses on the syntactic behaviour of the Macedonian lexical and auxiliary modals and the interaction of negation with mood and modality. While lexical modals take as their complements Mood Phrases headed by the subjunctive marker da, auxiliary modals, having become tenseless and having lost their t’-features, project Modality Phrases subcategorising for the TenseP of the verb which in previous stages of the language had appeared in the subjunctive clause in their complement. The Mood node projected by the da-complementiser is also the node where imperative mood is checked. On the basis of the syntactic and phonological behaviour of pronominal clitics, the paper argues that the imperative verb raises overtly to this node. In negative imperative clauses with pronominal clitics, there is a dialectal distinction in the positioning of the pronominal clitics relative to the verb. This distinction follows from a difference in the strength of the negation operator. In Standard Macedonian, Neg is weak and occupies the head position of NegP; at PF, it forms a single phonological word with the pronominal clitics and the verb. In North-Western Macedonian, on the other hand, Neg is strong and occupies the specifier position of NegP; at PF the clitics encliticise to it.

.

Introduction

Recent generative work has dealt with an increasing number of languages. With each new language analysed, new insights concerning the structures and processes that are part of Universal Grammar emerge. This paper provides an account of the relationship of modals to mood and modality phrases, emerging from the syntactic behaviour of Macedonian lexical and auxiliary modals. It will be argued that Mood is the node where imperative verbs check their features overtly. The relationship of mood and modality to negation will also be analysed, with special emphasis on negative imperatives.



Olga Mišeska Tomi´c

In Section 2 the syntactic behaviour of the Macedonian lexical and auxiliary modals will be discussed; evidence will be given for the existence of Modality and Mood Phrases. In Section 3 it will be argued that, in Macedonian positive imperative clauses, the verb raises to MoodP. In Section 4 the specific phonological behaviour of clitics in the environment of the raised imperative verb will be described. In Section 5 the relationship of mood and modality to negation will be dealt with: it will be shown that the differences between the behaviour of pronominal clitics in Standard Macedonian and NorthWestern Macedonian negative imperative clauses follow from differences in the strength of the negation operator, rather than from the strength of the modality operator. In Section 6 the results of the analysis will be summed up.

. The Macedonian modals Macedonian has five basic lexical modal verbs – saka1 ‘want’, može ‘can/may,’ mora ‘must’, treba ‘should’ and ima ‘have (to),’ and two invariant modal auxiliaries – кÜe ‘will/shall’ and bi ‘would’. The lexical modals are tensed verbs which take as complements tensed phrases introduced by the subjunctive marker (.) da. Saka always inflects for person and number, its person/number markers being independent of the person/number markers of the clause in its complement: (1) Sakam da dojdam/(dojdeš). (Mac) want.1 . come.1/2. ‘I want (you) to come.’

Može and mora can, but need not, carry person and number markers (which, if present, have to agree with the person/number markers of their complement clauses): (2) a.

Možat da dojdat. (Mac) can.3 . come.3. ‘They can come.’ b. Može da dojdete. (Mac) may. . come.2. ‘You may come.’ c. Moravme da dojdeme. (Mac) must.1. . come.1. ‘We had to come.’

Modality and mood in Macedonian 

d. Moraše da go storime toa. (Mac) must.. . it. do.1. that ‘It had to be done.’

Treba and ima are regularly impersonal: (3) a.

Treba(*m/*š) da odam/odiš. (Mac) should (1/2) . go.1/2. ‘I/you should go.’ b. Ima(*š/*at) da dojdeš/dojdat. (Mac) have (2/3) . come.2/3. ‘You/they have to come.’

The modal auxiliaries кÜe and bi are non-inflecting clausal clitics with a fixed position in the Macedonian clausal clitic cluster; they are always accommodated to the right of the negation operator and to the left of the be-auxiliary and pronominal clitics.2 While bi occurs in clauses in which V is instantiated by an l-participle and cooccurs with an auxiliary clitic, кÜe occurs in clauses in which V is instantiated by any of the items that can be instantiated in V. Examples of the use of bi and кÜe are given in (4): (4) a.

Ne bi sum mu go not would.. am.. him.. it.. dal proektot. (Mac) given.l-.. project-the ‘I would be unwilling to give him the project.’ b. Ne кÜe mu e izpraznet not will.. him.. is. emptied.... stanot. (Mac) apartment-the ‘His apartment will not be vacated.’

As argued in Tomi´c (1996), the clausal clitics are heads of functional projections. Thus, the modal clitic in (4a) is a head projecting a modality phrase, ModP. It occurs in the complement position of a negation phrase, NegP, and has as its complement a Tense/AgrSP,3 projected by the first person be-auxiliary clitic sum. The relevant structure of (4a) is given in (5):



Olga Mišeska Tomi´c

(5)

The relevant structure of (4b), where the head of the VP is instantiated by a passive participle, and the modal clitic is preceded by the negation operator and followed by the pronominal clitic mu and the third person be-auxiliary clitic e, is represented in (6):4 (6)

NegP Neg’ Neg

ModP

ne

Mod e

Tense/AgrSP Tense/AgrS AgrIOP AgrIO

AuxP

mu

Aux

VP

e

V izpraznet

XP

Modality and mood in Macedonian 

The subcategorisation properties of lexical modals differ from those of auxiliary (clitic) modals: rather than Tense/AgrSPs, the lexical modals take as complements tensed subjunctive phrases introduced by subjunctive markers. These complements, however, differ from ‘canonical’ CPs. In her analyses of the Balkan clause structure, Rivero (1994) places the Balkan subjunctive markers in the same position as modal auxiliaries. However, in Aromanian – a typically Balkan Romance language, spoken in Macedonia, Albania and Greece – subjunctive complements occur in complement positions of auxiliary modals, and so they do in dialectal Macedonian, though in the latter case the auxiliary modal is a non-inflecting clitic: (7) a.

Va s-yin will.3.. .-come.1. s-ti ved mãne. (Arom) .-you.. see.1. tomorrow. ‘I will come to see you tomorrow.’ b. КÜ e da dojde nekoj. (Dialectal Mac) will.. . came.3. somebody ‘Somebody seems to have come.’5

The subcategorisation properties of the Balkan subjunctive markers, and those of the Macedonian subjunctive markers in particular, differ from the subcategorisation properties of that-complementisers: while that-complementisers subcategorise for IPs with subjects in specifier positions, the subjunctive markers never occur to the immediate left of subjects, though subjects of subordinate clauses can occur to the immediate left of the subjunctive markers. This is illustrated in (8): (8) a.

Reˇce (ti) deka (ti) si mu said.3 you that. you are.2.. him.. go dal. (Mac) him.. given.. ‘(S)he said that you (yourself) have given it to him.’ b. Saka (ti) da *ti mu go wants you . you him.. him.. dadeš. (Mac) give.2. ‘(S)he wants you (yourself) to give it to him.’

The non-occurrence of subjects between auxiliaries and subjunctive complements in Romanian led Dobrovie-Sorin (1994: 15–47) to analyse Romanian

 Olga Mišeska Tomi´c

IPs as V-initial and Romanian structures with Aux followed by a subjunctive marker as CP/IPs. Thus, for (9a) she proposes the structure (9b): (9) a.

b.

Are s˘a vin˘a Ion. (Rom) has . come.3. John ‘John has to come.’ CP/IP Auxi

CP/IP Ci

IP

Cli

IP V-Infli

VP NP

V¢ tv

NP

The verb of the higher clause – the tensed auxiliary – is here adjoined to the CP/IP,6 while Infl is projected by the tensed verb of the lower clause. The main argument for the analysis is the fact that in interrogative clauses the auxiliary cannot raise by itself, which Dobrovie-Sorin illustrates by the ungrammaticality of sentences such as (10): (10) *Are Ion s˘a cînte la pian. (Rom) has John . play... on piano ‘Does John have to play the piano?’

The Macedonian lexical modals, however, can and do raise to (their own) CP, as witnessed by the grammaticality of (11), where the modal saka ‘want’ occurs to the left of the interrogative clitic li, which, in all the Slavic languages that have it, is typically generated in C: (11) Saka li (toj) ti da mu go wants . he you . him.. him.. dadeš? (Mac) give.2. ‘Does he want you (yourself) to give it to him?’

Moreover, in Macedonian (and in the Balkan languages in general), structures with post-verbal subjects are marked.

Modality and mood in Macedonian 

In view of the above facts, the Macedonian subjunctive markers are best analysed as heads of Mood Phrases to the left of TenseP. The structure of (12a) would be (12b): (12) a.

Saka ti da dojdeš. (Mac) wants you . come.2. ‘(S)he wants you (yourself) to come.’

b. CP Spec C’ C AgrS/TenseP Spec

AgrS/Tense’

AgrS/Tense Spec

vP v’ v

VP

Spec

V’

pro

V saka Spec

MoodP Mood’ Mood

TenseP

da

Tense’ Tense

vP v’ v

VP

Spec

V’

ti

V dojdeš



Olga Mišeska Tomi´c

In sentences in which the subjunctive structure is preceded by a modal clitic, MoodP would appear in the complement position of a modality phrase and the upper IP would be absent. The structure of (7b), for convenience repeated as (13a), is given in (13b): (13) a.

КÜ e da dojde nekoj. (Dialectal Mac) will.. . came.3. somebody.. ‘Somebody seems to have come.’ ModP

b. Spec

Mod’ Mod

MoodP

e

Mood’ Mood

TenseP

da

Tense’ Tense

vP v’ v

VP

Spec

V’

nekoj

V dojde

At an earlier stage, the Balkan Slavic modal clitics were tensed modal auxiliaries, subcategorizing for subjunctive complements introduced by the subjunctive marker da. A vestige of this subcategorisation has remained in the Bulgarian modal šte ‘will’, which can have past tense morphology and, when it does, inflects for person and number and takes da-complements:7 (14) a.

Štjax da dojda. (Bulg) would.1 . come.1. ‘I would have come.’

Modality and mood in Macedonian 

b. Šteše da dojde. (Bulg) would.3 . come.3. ‘(S)he would have come.’

Having become tenseless and having lost their φ-features, the Macedonian auxiliary modals turned into modal clitics, clinging to the verb which had previously appeared in their da-complement. The latter verb remained as the only verb in the sentence and the da-marker disappeared. Accordingly, the loss of the subjunctive complements came as a consequence of the loss of tense and φ-features.

. Imperatives The Mood node projected by the da-marker is also the node where imperative mood is checked. As illustrated in (15), the position of pronominal clitics in imperative clauses differs from their position in clauses with modal clitics: in clauses with modal clitics the pronominal clitics are wedged between these clitics and the verb, whereas in imperative clauses (in which no modal clitics occur), they appear to the right of the verb: КÜ e mu go dadeš will.. him.. it.. give.2. podarokot. (Mac) present-the ‘You will give him the present.’ b. Daj mu go podarokot! (Mac) give.2. him.. it.. present-the ‘Give him the present!’

(15) a.

At first glance, one might assume that the Macedonian imperative verb behaves like predicate nouns in that language, which, unlike the tensed verbs and l-participles, do not form local domains with the clitics and, when in clauseinitial position, host the clitics to their right. Nevertheless, the imperative verb always appears to the left of the pronominal clitics, whereas the predicate nominal does so only when there is no other constituent in this position. The position of the imperative verb relative to the pronominal clitics is invariant, whether the verb occurs in clause-initial position, as in (15b) or is preceded by one or more other constituents as in (16a) and (16b), respectively:



Olga Mišeska Tomi´c

(16) a.

Ti daj mu go podarokot! (Mac) you give.2. him.. it.. present-the ‘You give him the present!’ b. Ti utre ako sakaš daj mu go you tomorrow if want.2 give.2. him. it.. podarokot. (Mac) present-the ‘You give him the present tomorrow, if you please/want.’

Evoking Rivero’s (1991) long-head movement, I envisage the following scenario: The imperative verb is a [+V,+N] head and as such it does not form extended local domain with the clitics to its left and is free to long-head move to Mood, attracted by its strong features. Subsequently, in PF, the moved imperative verb and the pronominal clitics, which in syntax sit in their base-derived positions, form a single phonological word with the antepenultimate stress pattern characteristic for the language.

. Specific phonological behaviour The strength of Mood has a distinct phonological reflex. Macedonian clausal clitics have been treated as prototypical examples of proclitic verbal clitics (cf. Spencer 1991). Tomi´c (1999, 2000) and Franks (2000), however, observe that Macedonian clausal clitics are inherently neutral with respect to directionality of cliticisation. As argued in Tomi´c (1997, 2000, 2001), the directionality of cliticisation in Macedonian depends on the values for the verbal features, [±V], and the nominal features, [±N], of the head of the clause: in clauses with [+V,–N] heads, they are proclitic, in clauses with [–V,+N] heads, they are enclitic, while in clauses with [+V,+N] heads we find dual behaviour.8 Nevertheless, in clauses in which V is instantiated by imperative verbs, there is neither procliticisation nor encliticisation. In such clauses, the verb and the clitics form a single phonological word with the antepenultimate stress pattern characteristic for the language:9 (17) a.

DAJ – mu – give.2. him.. ‘Give it to him!’ b. ZeMI – mu – take.2. him.. ‘Take it from him!’

go! (Mac) it.. go! (Mac) it..

Modality and mood in Macedonian 

c.

PokažiTE – mu – go! (Mac) show.2. him.. it.. ‘Show it to him!’ d. Na – MI – ti – go! (Mac) ‘here you are’. me.. you.2.. it.. ‘Take it (my dear little one)!’

In (17a) the stress falls on the first and only syllable of the verb; in (17b) it falls on the second syllable of the two-syllabic verb; while in (17c) it falls on the last syllable of the four-syllabic verb. In each case, the stress falls on the antepenultimate syllable of the verb + clitics complex. The same stress-strategy is applied in (17d), where a monosyllabic verb is followed by three pronominal clitics and the stress falls on the clitic to the immediate left of the verb, which constitutes the antepenultimate syllable of the verb + clitics complex. As argued in Tomi´c (2001), the clitics in the scope of an operator form a single phonological word with their host. The distinct phonological behaviour of the clitics in the environment of the imperative verb follows from the fact that the imperative verb has moved to an operator in whose scope the clitics are.

. The relation between mood and negation In Macedonian, the phonological behavior of the verb and the clausal clitics in the scope of the negation and wh- operators is the same as that of the verb and the clausal clitics in the scope of the imperative mood operator. As illustrated in (18) and (19), in clauses with [+V,–N] heads, the operators and the verb, or the operators, the verb and the pronominal clitics wedged between them, form single antepenultimately stressed phonological words: (18) a.

ŠTO – sakaš? what want.2 ‘What do you want?’ b. Što – MU – dade? (Mac) what him.. gave.2/3. ‘What did you/(s)he give to him?’

(19) a.

NE – baram NIšto. (Mac) not seek.1 nothing ‘I am not asking for anything.’



Olga Mišeska Tomi´c

b. Ne – mu – GO – davam. (Mac) not him.. it.. give.1 ‘I am not giving it to him.’

In (18a) and (19a), where the operators immediately precede the disyllabic verbs, the stress falls on the operators. In (18b) and (19b), on the other hand, where pronominal clitics occur between the operator and the disyllabic verb, the stress falls on the clitic to the immediate left of the verb. When the negation operator cooccurs with a modality operator projected by a modal (non-inflecting) clitic, we have analogous phonological behavior. In clauses with [+V,–N] heads both the negation operator and the modal clitic form a single antepenultimately stressed word with the pronominal clitics and the verb:10 Ne – КÜ E – dojdam. (Mac) not will.. come.1. ‘I won’t come.’ b. Ne – кÜe – mu – GO – dadam. (Mac) not will.. him.. it.. give.1. ‘I won’t give it to him.’

(20) a.

In both (20a) and (20b) the stress falls on the syllable to the immediate left of the disyllabic verb: While in (20a) this syllable represents the modal clitic, in (20b) it represents the accusative pronominal clitic. In negated imperative clauses, the negation operator also forms an antepenultimately stressed single phonological word with the verb and the clitics to its right. In this case, however, the verb occurs to the immediate right of the negation operator, while the clitics, if any, follow the verb. Examine the following examples: (21) a.

NE – davaj NIšto! (Mac) not give.2. nothing ‘Don’t give (away) anything!’ b. Ne – DAvaj – mu NIšto! (Mac) not give.2. him.. nothing ‘Don’t give him anything!’ c. Ne – daVAJ – mu – go! (Mac) not give.2. him.. it.. ‘Don’t give it to him!’

In (21a) the disyllabic verb is not followed by any clitics and the stress falls on the negation operator, which actually constitutes the antepenultimate syllable

Modality and mood in Macedonian 

of the phonological word made up of the negation operator and the verb. In (21b), the disyllabic verb is followed by one pronominal clitic, and the stress falls on the first syllable of the verb, which occupies the antepenultimate syllable of the phonological word. In (21c), where the disyllabic verb is followed by two pronominal clitics, the stress falls on the second syllable of the verb, which also represents the antepenultimate syllable of the phonological word. This pattern does not obtain in all the Macedonian dialects. Thus, in North-Western Macedonian the pronominal clitics occur between the negation operator and the verb; they actually encliticise to the negation operator, which is stressed independently of the head of the clause. The North-Western Macedonian counterparts of (21b) and (21c) are given in (22a) and (22b), respectively: NE ← mu DAvaj NIšto! (NWMac) not him.. give.2. nothing ‘Don’t give him anything!’ b. NE ← mu ← go DAvaj! (NWMac) not him.. it.. give.2. ‘Don’t give it to him!’

(22) a.

Franks (1998: 55) sees two ‘realistic’ options for the cliticisation strategy in Macedonian imperative clauses: ‘Either (i) there is some kind of strong imperative features to be checked that would force raising of V past the clitics or (ii) the clitics for some reason lower onto the imperative.’ Arguing that nonfinite verb forms should move further up the tree than finite ones, Franks settles for option (ii), assuming that ‘the functional impoverishment of nonfinite forms could cause them not even to make the minimal moves required by the clitics, which would then somehow lead to the necessity for the clitics to take action themselves’ (p. 57). He also argues that the negation marker ne must in some way be attracting the verb and the relevant feature could be either strong or weak. In the former case the imperative verb raises to ne, picking up the clitics as it goes; in the latter, the imperative verb does not raise and the clitics lower onto it instead. Franks’ analysis is in many respects unsatisfactory. First of all, the distinction finite: nonfinite verb forms has no significant import in Macedonian: with respect to cliticisation the l-participles behave like tensed verbs; the past and passive participles have dual behavior – like tensed verbs or like predicate nouns; and imperative verbs in positive clauses occur to the left of the pronominal clitics, while in negative clauses we have different behaviours in different dialects. Franks does not explain why the clitics encliticise rather than procliti-



Olga Mišeska Tomi´c

cise in either Standard or North-Western Macedonian, and, why in the latter case, illustrated in (22), they encliticise to ne, leaving the verb, which ne should have attracted, by itself. Neither does he explain why clitics in Standard Macedonian negative imperative clauses, exemplified in (21), do not procliticise to the verb (if they cannot encliticise to a weak ne) and have to lower. But above all, Franks neglects the fact that the cliticisation strategies illustrated in (21) and (22) belong to two different varieties. The difference between the positions of the clitics in Standard Macedonian and North-Western Macedonian negated imperative constructions does follow from a difference in the strength of the negation operator. As argued in Tomi´c (2001), in Standard Macedonian, Neg is weak and occupies the head position of NegP. At PF, the negation operator forms a single phonological word with the pronominal clitics and the verb, which, due to the strength of Mood, is pronounced to the left of the clitics. In North-Western Macedonian, on the other hand, Neg is strong and occupies the specifier position of NegP.11 The pronominal clitics move to Mood along with the functional features of the verb. At PF the clitics procliticise to the strong Neg.

. Summing up While lexical modals take as their complements Mood Phrases headed by the subjunctive marker da, auxiliary modals project a Modality Phrase and subcategorise for TenseP. At an earlier stage in the history of the language, all the Balkan Slavic modals were tensed and subcategorised for subjunctive complements introduced by subjunctive markers. Having become tenseless and having lost their φ-features, the Macedonian auxiliary modals stopped taking subjunctive complements. The Mood node projected by the subjunctive marker da is the node where imperative mood is overtly checked; material evidence for this comes from the distinct syntactic and phonological behaviour of the clitics in the environment of the imperative verb. In negative imperative clauses with pronominal clitics, there is a dialectal distinction in the positioning of the pronominal clitics relative to the verb. This distinction follows from a difference in the strength of the negation operator. In Standard Macedonian, Neg is weak and occupies the head position of NegP; at PF, it forms a single phonological word with the pronominal clitics and the verb. In North-Western Macedonian, on the other hand, Neg is strong and occupies the specifier position of NegP; at PF the clitics encliticise to it.

Modality and mood in Macedonian 

Notes . Macedonian has no infinitives. The third person singular present tense forms are used as citation forms. . The Macedonian clausal clitic cluster has six slots: (i)

1 2 3 4 5 6   1/2. ⁄. . . 3. ⁄.

The third and the fifth slot are never filled in the same clause. The clitics in the other slots can occur simultaneously and combine with either the clitics in slot 3, or with those in slot 5. The order is always fixed. . A joint Tense and Subject Agreement Phrase is argued for by the portmanteau morph which in all Balkan Slavic languages represents tense, person and number. . Whereas the first and second person auxiliary clitics (singular and plural) are analysed as heads of Tense/AgrSP, the third person auxiliary clitics (also singular and plural) are analysed as heads of AuxP. Crucial evidence for the distinct analysis of the third person auxiliary clitics is the fact that, in the clitic cluster, they occur to the right of the pronominal clitics, while the first and second person clitics occur to the left of the pronominal clitics. Moreover, third person auxiliary clitics occur in clauses in which V is instantiated by passive participles, such as (4b), but not in clauses in which V is instantiated by l-participles, such as (i). (i)

Ne bi mu go *e dal not would.. him.. it.. is. given.l-.. proektot. (Mac) project-the ‘(S)he wouldn’t give him the project.’

. Macedonian sentences in which a subjunctive expression occurs in the complement position of кÜ e do not have future reference. . Dobrovie-Sorin (1994: 17) argues that the adjunction analysis of the auxiliary is possible if we assume the folowing conventions: i. ii.

Functional Coindexation: Coindex adjacent functional X0 categories. Functional Adjunction: Adjoin X0 categories to the YP functional projection with which they are coindexed.

. In the present tense, this modal does not inflect for person and number and, consequently, cannot take a da-complement: (i)

Šte (*da) dojda. (Bulg) will . come.1. ‘I will come.’

(ii) Šte (*da) dojde. will . come.3. ‘(S)he will come.’



Olga Mišeska Tomi´c . Tensed verbs and l-participles are [+V,–N] categories; nominal predicates are [–V,+N] categories; past and passive participles as well as adjectives are [+V,+N] categories. The behavior of clitics in clauses headed by l-participles, nouns and past participles is illustrated in (i), (ii) and (iii), respectively: (i)

Ste → mu → go → SKInale PALtoto. (Mac) are.2. him.. him.. torn.l-. coat-the ‘You have, reportedly, torn his coat.’

(ii) BRAкÜ a ← ste ← mu. (Mac) brothers are.2. him.. ‘You are his brothers.’ (iii) Mu → e → SKInato PALtoto. (Mac) him.. is. torn.... coat-the ‘His coat is torn.’ (iii ) SKInato ← mu ← e PALtoto. (Mac) torn.... him.. is. coat-the ‘His coat is torn.’ (Arrows indicate directionality of cliticisation; capitals mark stressed syllables.) . The dashes connect the elements that form single phonological words. . In clauses with [+N] heads, the head of the clause is stressed distinctly from the operators and the pronominal and/or auxiliary clitics. This distinction follows from the nature of the head of the clause: as argued in Tomi´c (2001), the head of the clause forms or does not form a local domain with the clausal clitics depending on whether this head is a [–N] or [+N] category. In any case, the cliticisation strategy in clauses with operators is distinct from the cliticisation strategy in clauses without operators. In the former case the operator(s) and the auxiliary and pronominal clitics to the right form a single phonological word (in which the head of the clause is or is not included), while in the latter case the pronominal clitics procliticise to the head of the clause or encliticise to an element to their left, the stress pattern of the host remaining as it had been originally. . The structural differences between Standard Macedonian and North-West Macedonian negated imperative structures can be observed by comparing (i) and (ii): (i)

[NegP

[ Neg’ [Neg ne]

(ii) [NegP [Spec ne] [NegP’ [Neg ]

Tense/AgrSP]] (StMac) Tense/AgrSP]] (NWMac)

I am grateful to Lisa Cheng for pointing out that my analysis of the structural difference between Standard Macedonian and North-Western Macedonian negated clauses is reminiscent of Zanuttini’s (1991, 1997) analysis of the differences between a variety of Italian dialects, thus turning my attention to the possibility of analyzing the strong NWMac negation operator as an XP in the Spec of NegP.

Modality and mood in Macedonian 

References Chomsky, N. (1995). The Minimalist Program. Cambridge, Mass.: MIT Press. Dobrovie-Sorin, C. (1994). The Syntax of Romanian [Studies in Generative Grammar 40]. Berlin/New York: Mouton de Gruyter. Franks, S. (1998). Clitics in Slavic. Paper presented at the Comparative Slavic Morphosyntax Workshop. Bloomington: Indiana University, June 1998. (downloadable at http://www.indiana.edu/∼slavconf/linguistics/index.html) Franks, S. (2000). Clitics at the interface. In F. Beukema and M. den Dikken (Eds.), Clitic Phenomena in European Languages (pp. 1–46). Amsterdam: Benjamins. Rivero, M.-L. (1991). Long head movement and negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review, 8, 319–351. Rivero, M.-L. (1994). Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory, 12, 63–120. Spencer, A. (1991). Morphological Theory. Oxford: Blackwell. Tomi´c, O.M. (1996). The Balkan Slavic clausal clitics. Natural Language and Linguistic Theory, 14, 811–872. Tomi´c, O.M. (1997). Non-first as a Default Clitic Position. Journal of Slavic Linguistics, 5(2), 1–23. Tomi´c, O.M. (1999). On Clitichood. In I. Kenesei (Ed.), Crossing Boundaries (pp. 9–32). Amsterdam: Benjamins. Tomi´c, O.M. (2000). On clitic sites. In F. Beukema and M. den Dikken (Eds.), Clitic Phenomena in European Languages (pp. 293–317). Amsterdam: Benjamins. Tomi´c, O.M. (2001). The Macedonian negation operator and cliticization. Natural Language and Linguistic Theory, 19, 647–682. Zanuttini, R. (1991). Syntactic Properties of Sentential Negation. A Comparative Study of Romance Languages. Ph.D. dissertation, University of Pennsylvania. Zanuttini, R. (1997). Negation and Clausal Structure: A Comparative Study of Romance Languages [Oxford Studies in Comparative Syntax]. New York: Oxford University Press.

Subject index

A acquisition 10, 95–98, 105, 106, 113, 114, 126, 158, 186–190, 193–195, 199, 200 adjectives 21, 188, 276 adjunct 5, 6, 15, 57, 99, 134, 138, 145, 148, 153, 157, 160 adjunction 238, 252, 275 adjuncts 5, 57, 138, 145, 148, 153, 157 adverb 1, 5, 8, 14, 21, 47, 53, 54, 143, 148, 151, 154, 158, 159, 169, 172, 188 adverbs 1, 5, 14, 21, 47, 54, 143, 148, 151, 154, 159, 169, 188 affirmation 52, 64–66, 68, 69, 72 agreement 35, 225–227, 229, 275 AgrOP 78, 79, 91, 227, 228 AgrSP 78, 79, 83, 84, 91, 92, 249, 254, 256–258, 263, 265, 275, 276 Aktionsart 21–23, 25–27, 35, 40, 41, 43–47 alethic 30, 38, 42, 46, 135, 154 anti-additive 221, 222 Aromanian 265 aspect 21–23, 25–27, 33, 36, 45, 47, 60, 105, 109, 110, 115 aspect phrase 44 aspectual class 61 aspectuals 155 autism 11, 187, 195–199 autistic 195–200 Aux 8, 81, 140, 141, 144, 158, 159, 245, 250, 261, 263, 265, 266, 268, 275 auxiliary modal 8, 261, 262, 265, 269, 274

B Balkan clause structure 265 Balkan Romance 265 Balkan Slavic 268, 274, 275 Basque 136, 228 belief in truth 169 bipolar 68, 69 bipolarity 52, 70 bouletic 107 bounded scale 51, 52, 55, 56, 59, 60, 64–68, 70, 71, 241 British English 20, 138, 149, 152, 200 Bulgarian 32, 246, 247, 255–257, 268

C Catalan 8, 9, 134, 136, 138, 140, 144–146, 151, 153, 155, 159 change of category hypothesis 120, 125, 126 chat conventions 115 checking 78, 94, 96, 99, 143, 144, 156, 247, 248, 256, 257 child language 11, 98, 103, 105, 106, 108, 111, 113, 114, 116, 117, 123, 127, 128, 134, 186, 188, 193, 199 CHILDES 114 clausal clitic 263, 270, 271, 275, 276 clause boundaries 155 clause type 26 clitic 142, 219, 224, 226, 227, 241, 243, 245–250, 252–255, 257, 258, 262–266, 269–276 clitic cluster 249, 250, 252, 275

 Subject index

clitic placement 245–247, 249, 252–254, 256, 257 clitic raising 218 cliticisation 270, 273, 274, 276 complement 2, 3, 5–7, 9, 12–15, 34, 39, 51–61, 63–66, 68–72, 99, 113, 155, 159, 171, 172, 175–178, 196, 211–213, 232, 243, 249, 251, 262, 263, 265, 268, 269, 274, 275 complement of modals 5, 12, 57, 59, 65, 70, 71 complementiser 5, 171, 173, 175, 182, 249, 252, 253, 265 conditional 152, 157, 172, 189, 213–216, 231 control 2, 4–7, 15, 23, 34, 37–40, 46, 51, 56–58, 71, 117, 159, 187, 257 D Danish 36, 37, 79, 160 definiteness 63 deontic 15, 20, 21, 23, 24, 26, 27, 30, 36–40, 42, 45, 46, 55, 57, 58, 62, 63, 67, 75, 77, 86, 88–90, 95, 105, 107–113, 115, 128, 134–136, 138–141, 145, 146, 149, 151–160, 165, 166, 169, 177, 180, 181, 186, 188, 189, 196, 211 displacement 138, 141, 142, 144, 160 dispositional 62, 63, 66, 67 disyllabic 272, 273 double-modal construction 86 downward entailment 215 dubitative 165, 166, 171, 176, 177, 182 Dutch 3, 5–8, 13, 15, 21, 25, 26, 30–33, 40, 43, 46–48, 53, 56, 57, 62, 63, 70, 72, 77, 79, 103–109, 111–114, 116, 117, 120, 124, 126–128, 159, 196, 206–208, 212, 214–217, 220–223, 229–232, 236, 240–242 dyadic 2–6, 52, 56, 67 dynamic 21, 33, 34, 61, 87, 95, 115, 116, 141, 186, 189

E Echo 48, 136, 147–150, 156–159 echoic 147, 150, 157, 159 echoic contexts 135, 136, 141, 146, 148, 150 embedding 23, 25, 28, 31, 37, 43, 46, 217, 220, 251 encliticisation 255, 270 encliticise 273, 274, 276 English 3, 6, 8, 9, 14, 19–26, 29–34, 38–41, 43, 45–48, 53, 76, 77, 80, 82–84, 87, 88, 90–92, 94, 97–99, 106, 107, 109–111, 113, 114, 129, 134–141, 144–146, 149, 151–158, 185–187, 192–195, 200, 205, 210, 214–216, 219, 224, 229–232, 236 epistemic 1–5, 7–13, 15, 19–21, 23, 24, 26, 27, 30, 33–43, 45–48, 51, 52, 54–67, 69–71, 75, 77, 86, 88, 90, 95, 105, 108–113, 115, 129, 134–136, 138–141, 144–146, 149, 151–156, 158–160, 165, 166, 169, 176, 180, 181, 186–189, 192–197, 199, 242 event structure 27, 33–35, 45 eventivity 107, 108, 111, 112 evidential 5, 23, 27, 193 existential 69, 70, 150, 157 existential quantification 216 expletive 6, 7 extended now 23

F finite 11, 15, 23–25, 27, 28, 30, 31, 33, 35, 40, 43, 45, 71, 77, 78, 94, 98, 99, 103, 104, 106, 114, 116–119, 123, 125–129, 140, 144, 160, 223, 225, 228, 229, 233, 249, 273 finiteness 23, 27, 106–108, 114, 126 freezing 231, 233, 234, 236–238, 243 frequency of modal verbs 20 full lexical verb 30–32, 36

Subject index 

functional 2, 10, 11, 23, 36, 40, 71, 166, 183, 228, 273–275 functional head 137, 144, 151, 153, 157, 158, 160, 249, 257 functional projection 10, 61, 83, 92, 93, 153, 154, 160, 234, 258, 263, 275

infinitival morphology 11, 107, 117, 120, 121, 127 input-based hypothesis 104 intransitive 3, 4, 31, 34, 35, 41 irony 157 Italian 136, 151, 152, 159, 200, 219, 245, 276

G German 3, 7, 8, 13, 15, 19–27, 29–33, 36, 38–41, 43–48, 70, 77, 86, 99, 103, 104, 106–109, 111, 113, 114, 116, 123, 127, 128, 159, 188, 194, 241, 242 Greek 8, 154, 189, 245

J Jamaican creole

H head-movement 134, 225, 228 head-to-head 142, 233 hedging on truth 171 historical 20–22, 76, 110 hypothetical 86, 128, 166, 173, 175, 176, 180, 182 hypothetical modality 165, 171–174, 180

I Icelandic 7, 8, 16, 79, 82, 87, 88, 154, 160 identity function 143 imperative 1, 14–16, 25, 26, 80, 81, 104, 105, 128, 152, 165–169, 173, 177–182, 245–258, 261, 262, 269–274, 276 imperative feature 251, 257, 258, 273 imperative operator 247, 248, 255, 257 independent full verb 29 indicative 165–173, 175, 177, 178, 180, 181, 183, 245–247, 249–251, 253, 254, 257 individual-level 12, 14, 55, 61

145

L Language of Thought 155 learnability 136, 145 left periphery 83 Lele 14, 166, 167, 169, 170, 176, 177, 183 lexical modal 262, 265, 266, 274 lexicalist 142 LF 2, 9, 67, 133–135, 137, 140–144, 146–148, 150, 153, 156, 158, 241, 242, 248, 255–257 LF-movement 148 licensing 43, 106, 213, 219–222, 230, 238, 239, 247 local domain 269, 270, 276 locality 219–221 M Macedonian 5, 6, 261–263, 265–267, 269–271, 273–276 marginal modal 84, 85 Merge 109, 134, 143, 147, 148, 156 Middle Dutch 9, 206–209, 211, 212, 217, 218, 222–228, 230–232, 234, 236, 239–243 Middle English 21, 46, 75, 76, 94, 96–99, 227, 228 Minimalism 83 Minimalist 31, 76, 78, 79, 133, 134 modal auxiliaries 7, 8, 133, 134, 144, 146, 153, 154, 159, 160, 262, 263, 265, 268 modal clitic 263, 264, 268, 269, 272

 Subject index

modal main verbs 7, 8, 154 modal reference effect 103, 107, 111, 113, 114, 120, 125 modal verbs 1, 14, 19–21, 23, 24, 26, 27, 33, 36–40, 46, 51, 62, 70, 75, 97, 98, 117, 118, 129, 152, 185, 187, 189, 192, 193, 195, 199, 205, 219 modality operator 262, 272 modality phrase 261, 263, 268, 274 monadic 2–6, 52, 56, 57, 63, 66, 67 monotone-decreasing context 216, 221, 223, 240 Mood 109, 151, 157, 245, 247, 248, 250–252, 255, 258, 261, 262, 269–271, 274 mood phrase 262, 267, 274 MoodP 262, 268 morphological marking hypothesis 104 morphophonological 158 morphophonological operator 142 movement 2, 76, 78, 79, 83, 93, 94, 96, 97, 99, 106, 134, 142, 148, 150, 156, 219, 225, 227–229, 231, 233, 234, 242, 247, 254–257, 270

N necessity 1, 8, 9, 105, 115, 116, 138–141, 146, 151, 155, 157, 158, 187, 188, 192, 194, 205, 273 necessity modals 139, 157 neg-agreement 226, 227, 229 Neg-criterion 150, 223–233, 236, 242 negation 8, 9, 14, 22, 26, 52, 64–66, 68, 69, 72, 75, 87, 90, 98, 133–138, 141, 143, 145–149, 151, 156–159, 171, 205, 212–214, 216, 220, 221, 224–227, 229–231, 236, 241–243, 245, 247–250, 255–257, 262, 263, 271, 273

negation operator 262–264, 272–274, 276 negative imperatives 246–248, 253, 255, 257, 262 negative operator 150, 215, 228, 241, 248 negative polar 68, 69, 72 negative polarity 9, 14, 52, 70, 139, 205, 206, 214–218, 220, 221, 223–225, 230, 232, 236, 240–242 NegP 68, 90–93, 97, 224, 227, 232–234, 236, 241, 248, 249, 263, 274, 276 nominal 11, 52, 63, 65, 66, 68, 69, 71, 72, 114, 120, 121, 126, 129, 167–169, 179, 183, 231, 232, 269, 270, 276 non-disambiguating complements 12 non-finite embedding 43 non-subject-oriented 52 non-verbal complement 7, 13, 51, 53–60 nonfinite 273 North-Western Macedonian 262, 273, 274, 276 NPI 139, 140, 157, 158, 160, 205, 206, 221–224, 228–234, 236, 241–243 null auxiliary hypothesis 103, 106, 108

O obligation 1, 6, 12, 13, 39, 56–58, 70, 86, 105, 107, 109, 115, 135, 138, 146, 159, 180, 186–188, 194, 229, 230 OE 19, 21, 47, 76–78 OHG 19, 23 Old English 19, 21, 22, 40, 46, 76, 99 Old High German 19, 21 OV 14, 76–82, 84, 87, 88, 94, 96, 98, 99, 241

Subject index 

P participial construction 30 participle 3, 4, 15, 21, 34, 37, 45, 54, 121, 206–209, 239, 240, 263, 269, 273, 275, 276 passive participle 39, 264, 273, 275, 276 passivisation 6, 32 perfective 12, 21, 23, 27, 35, 40, 45, 47, 61, 62, 71, 107, 109, 110, 113, 114, 116, 129, 158, 209, 211 periphrasis 21, 31 PF 9, 53–55, 134, 138, 141–144, 147, 148, 150, 156, 158, 247, 255, 270, 274 PF-deletion analysis 54, 55 phonological word 142, 270–274, 276 polarity 52, 63, 67, 68, 72, 136, 137, 158, 216, 220, 221, 231 polarity transition 52, 58–62, 66, 112 portmanteau morph 275 possibility 1–3, 9, 10, 14, 66, 86, 93, 99, 115, 116, 118, 129, 134, 138–141, 146, 155, 157, 158, 188, 192–194, 200, 205, 211, 229, 233, 245, 276 possibility modals 8, 9, 158 possible worlds 15, 154 post-verbal subjects 266 potential 52, 58–63, 71, 107, 109, 112, 173, 186, 231 pragmatics 193 predicate noun 269, 273 predication 42, 108, 112, 141 present perfect 13 preterite 20, 22, 23, 47, 207–209, 211, 213, 219, 231–234, 236, 240, 242 Principles and Parameters 133 proclitic 247, 252–254, 257, 270 procliticise 274, 276 prohibitive 165, 166, 177, 179, 180

pronominal clitic 246, 247, 249, 250, 252, 253, 262–264, 269–276 Q QR 83, 99 quirky case 7 R raising 2, 4, 5, 7, 9, 14, 15, 23, 27, 31, 34, 35, 37, 39, 40, 43, 45, 46, 52, 56–58, 71, 83, 106, 141, 158–160, 167, 242, 254, 257, 273 Relevance Theory 147, 155 remnant VP-preposing 92, 97, 98 Romanian 265, 266 root 1, 2, 4–10, 12, 13, 15, 19–24, 26, 27, 30, 33, 35–40, 42, 43, 45–48, 56–59, 61–64, 66, 71, 77, 78, 129, 134, 140, 141, 148, 151, 154, 166–169, 186–188, 193–196, 200, 232, 236, 240, 242, 247, 252–255 root imperatives 247, 253, 254 root infinitive 11, 103, 104, 106, 117 root interpretation 1–3, 6–9, 12–14, 51, 52, 58–61, 63–65, 67, 70, 186, 194, 195, 242 S scope 7–9, 27, 36, 48, 67, 75, 77, 83, 87–92, 94, 95, 99, 108, 133–143, 145–153, 155–160, 176, 213–216, 221–224, 227–232, 236, 241, 248, 255, 257, 271 scope constraint 151, 153, 154 Scots 144, 145, 152, 153, 159 Scots English 144 select 3, 67, 113, 134, 139, 142, 144, 148, 160, 242 selection 3, 21, 38, 41, 140, 141, 144, 153 semantic role 36–39, 41 sequence of tenses 157 Slavic 47, 255, 266 Slovene 14, 245–247, 249–258

 Subject index

Southern States English 145 Specifier 78, 83, 84, 93, 96, 156, 157, 225, 227, 228, 232, 265, 274 speech act 29, 42 Spell-out 79, 83, 233, 234, 238 Split Sign 134 splitting 150 stage-level 55, 59 stress 159, 238, 241, 270–273, 276 structural position 7–9, 36, 48 subcategorisation 265, 268 subject-orientation 52, 58, 63, 66 subject-oriented 52, 58, 62, 66, 67, 90 subjunctive 5, 165, 166, 177–179, 181, 207, 216, 246, 262, 265–269, 274, 275 sympathy 52, 62 systems interaction 166 T taboo 224, 227, 232 tag 25, 137–139, 146, 147, 149, 157 temporal adverb 137, 143, 157 tense 3, 4, 11, 20, 21, 23, 24, 27, 28, 33, 46, 47, 86, 90, 108–110, 115, 122, 126, 129, 140, 147, 157–159, 167, 219, 223, 233, 234, 238–240, 242, 258, 263, 265, 268, 269, 275, 276 tense concord 158 tensed verb 249, 262, 266, 269, 273, 276 thematic role 36, 37, 39, 41

theory of mind 186, 189, 190, 192–196, 199, 200 theta-role 4–6, 15, 23, 36, 37, 39, 40, 43, 45, 56, 65, 83, 160 transitive 3, 4, 6, 26, 33–35, 41, 65, 68, 159 truth 1, 70, 85, 86, 94, 112, 128, 165–167, 170–172, 174, 176, 181, 183 Tyneside English 138

U UG 96, 134, 142, 145, 151

V V-to-I 96, 97, 99, 106 V2 8, 77, 78, 86, 92, 225, 228, 229, 234, 236, 237, 241, 243 verbal NPI 222, 229–232 VO 14, 76, 78, 79, 82, 87, 88, 94, 99, 241

W Wackernagel 247, 248, 254 West Frisian 7, 15, 21, 40 wide scope 35, 87–91, 93, 146, 149 word order 14, 72, 76–79, 82, 166, 228

Y Yiddish

7, 15, 21, 40, 241

Name index

A Abraham, W. 3, 13, 19, 21, 24, 26, 27, 31–33, 200 Akmajian, A. 128 Anderson, L. 128 Anderson, S. R. 158 Astington, J. 189, 192 Atance, C. 188 Atkinson, M. 97 Avrutin, S. 110

B Bailey, B. L. 145, 158, 159 Barbiers, S. vii, 1, 6–8, 12, 13, 21, 25, 30, 32, 51, 61, 65, 67, 69–72, 112, 128 Baron-Cohen, S. 195, 198 Battistella, E. L. 145 Beal, J. C. 138, 156 Behrens, H. 103, 104, 113, 123 Belletti, A. 83 Bellugi, U. 97 Bennis, H. 6 Beukema, F. vii, 14, 75 Blom, E. 11, 103, 104, 107, 113, 116, 129 Bloom, L. 111 Bobaljik, J. 7 Boser, K. 11, 103, 106 Boškovi´c, Ž. 252 Brennan, V. 6, 10, 15, 57, 58, 70, 156 Brinton, L. 46 Brody, M. 156 Brown, K. 141, 144–146, 152, 153, 156, 158, 159 Bryant, P. 190

Burzio, L. 3, 6 Butterworth, G. 189 Bybee, J. 128 Byrnes, J. 188 C Carruthers, P. 189 Cassidy, K. 200 Charman, T. 198 Choi, S. 193 Chomsky, N. 13, 65, 76–78, 133, 247, 256 Chung, S. 109, 128 Cinque, G. 8, 9, 15, 83, 133, 136, 151–153, 156, 159 Clark, H. 128 Coates, J. 20, 141, 157, 158, 195 Collins, C. 65 Conradie, J. 31 Cormack, A. vii, 9, 10, 14, 133, 134, 136, 142–144, 151, 156–160, 200 D Davidge, J. 192 De Roeck, A. 196 Denison, D. 76, 77, 86 Diewald, G. 13 Dobrovie-Sorin, C. 265, 266, 275 Duff, M. 188 Durbin, J. 24, 27, 30, 31 E Eisenmajer, J. 198 Embick, D. 255 Ernst, T. 9, 143 Espinal, M. T. 159

 Name index

Estes, D. 190 Extepare, R. 128

F Feldman, F. 6, 57, 58 Ferdinand, A. 128 Fischer, A. 32 Flavell, J. 191 Fleischman, S. 128 Fodor, J. 155 Foster, T. 95, 96 Fox, D. 82, 87, 100 Frajzyngier, Z. vii, 14, 165, 166, 170–172, 183 Franks, S. 270, 273, 274 Frith, U. 195, 198 Frye, D. 189 Furrow, D. 192

G Garrigues-Cresswell, M. 175, 183 Gazdar, G. 146, 158 Gee, J. 51, 53, 187, 214 Geerts, G. 51, 53 Gelman, S. 189 Gelderen, E. van 46 Gerhardt, J. 111, 187 Gillette, J. 195 Gillis, S. 107, 113 Ginneken, J. van 103 Giorgi, A. 107, 110 Givón, T. 128 Gleitman, H. 201 Gleitman, L. 195, 200 Golden, M. 14, 245, 252 Gonsalves, J. 115 Gopnik, A. 190, 191, 193, 200 Gordon, P. 190 Grimm, H. 194 Guo, J. 189

H Haan, F. de

117, 128, 158, 159

Haegeman, L. 104, 106, 137, 150, 159 Hale, K. 13, 65 Han, C.-H. 246, 248, 255–257 Happé, F. 198, 200 Harris, P. 156, 189 Heim, I. 66 Hirschfeld, L. 189 Hirst, W. 188 Ho, S. 188 Hoeksema, J. 70 Hoekstra, T. 3, 11, 57, 103, 104, 108, 114, 127–129 Hofmann, T. 2, 52, 56 Hogrefe, J. 191 Holmberg, A. 78, 99 Hornstein, N. 83, 99 Hurlburt, R. 198 Hyams, N. 11, 103, 104, 108, 114, 127–129, 200 I Ingram, D. 103, 106–108, 116 Izvorski, R. 255 J Jackendoff, R. 99, 143, 159 Jacobs, J. 159 Jespersen, O. 110 Jordens, P. 117, 121 K Kayne, R. 77, 78, 87, 88, 90–92, 94, 150, 258 Keyser, S. J. 13, 65 Klima, E. 97, 137, 150, 158, 214 Klinge, A. 112 Klooster, W. 6 Koopman, H. 236–239 Koopman, W. 76, 78 Krämer, I. 104 Kratzer, A. 66, 152, 186 Krifka, M. 159 Kursawe, C. 106

Name index 

L Ladusaw, W. 215 Lahey, M. 111 Larson, R. 13, 65 Lasnik, H. 97, 247 Lass, R. 96, 103–105, 127, 128 Lasser, I. 103–105, 127, 128 Lechner, W. 159 Lederer, A. 195 Leekam, S. 198 Leslie, A. 189, 190, 195, 198, 200 Li, P. 194 Lightfoot, D. 76, 99, 144, 157 Lust, B. 103 Lyons, J. 1, 70, 128 M MacWhinney, B. 114 Marantz, A. 255 May, R. 87 McDowell, J. 9, 10 McGinnis, M. J. 6 Menaugh, M. 146 Millar, M. 145 Miller, J. 100, 145 Moerenhout, M. 82 Moore, C. 189, 192 Mycielski, J. 183 N Nagle, S. J. 156 Niemeier-Wind, K. 194 Noveck, I. 188 Nuyts, J. 196–198 O O’Neill, D. 187, 191, 193 Olson, D. 189, 192 Overdiep, G. S. 211 P Palmer, F. 1, 70, 86, 90, 115, 128, 141, 146, 148–150, 156, 159, 165, 170, 183

Papafragou, A. vii, 11, 145, 156, 185, 186, 193, 194 Pérez-Leloux, A. 129 Perkins, M. 188 Perlmutter, D. 2, 52, 56 Perner, J. 191, 198 Pianesi, F. 107, 110 Picallo, C. 8, 134, 144, 156, 157 Pintzuk, S. 99 Platzack, C. 16, 99 Poeppel, D. 106 Pollock, J.-Y. 144, 249 Postma, G. 9, 14, 205 Pratt, C. 190 Premack, D. 189 Prior, M. 198 Pullum, G. K. 142 Pure, K. 192

Q Quirk, R.

141

R Radford, A. 10 Reis, M. 19, 24, 32 Riemsdijk, H. van 86, 99 Rivero, M. L. 245–247, 251, 253–255, 257, 258, 265, 270 Rizzi, L. 83 Roberts, I. 8, 77, 78, 94, 96, 99, 144 Roeper, T. 129 Ross, J. R. 2, 52, 56, 129 Roth, D. 195, 198 Roussou, A. 8, 154 Rowlett, P. 159 Rullman, H. 159

S Santelmann, L. 103, 106 Savasir, I. 187 Sera, M. 188 Seuren, P. 215 Shatz, M. 187, 190, 192–194, 200

 Name index

Shepherd, S. 187 Sheppard, M. M. 14, 245, 252 Silber, S. 192 Slaughter, V. 191 Smith, N. vii, 9, 10, 14, 133, 134, 136, 142–144, 151, 157–160, 200 Smith, P. 189 Smoczynska, M. 189 Sodian, B. 191 Spencer, A. 270 Sperber, D. 147 Sprouse, R. 24, 27, 30, 31 Stephany, U. 98, 115, 187–189 Stoett, F. A. 211 Stowell, T. 5 Streitberg, W. 211 Swaay, H. A. J. van 211 T Tackeff, J. 111 Tager-Flusberg, H. 199, 200 Terzi, A. 245–247, 251, 253–255, 257, 258 Thaiss, L. 195, 198 Thompson, W. 103, 106–108, 116 Thráinsson, H. 6–8, 15, 154, 160 Tieken-Boon van Ostade, I. 97 Timberlake, A. 109, 128 Toman, J. 252 Tomi´c, O. M. 5, 261, 263, 270, 271, 274, 276 Traugott, E. 86 Tsimpli, I.-M. 158 U Ud Deen, K.

107, 114, 129

V Verdam, J. 208, 209, 212 Verwijs, E. 208, 209, 212 Vikner, S. 6–8, 15, 99, 154, 160 W Warner, A. 6, 85, 99 Weibegué, C. 183 Weil, J. 29, 188 Wellman, H. 189, 190, 192 Wells, G. 187, 188, 194 Weverink, M. 104 Wexler, K. 106 Whiten, A. 189 Whitman, J. 103 Wijnen, F. 103, 104, 107, 113, 114, 117, 127, 129 Wilcox, S. 187, 193, 194 Williams, E. 146 Wilson, D. 147, 200 Wimmer, H. 191 Woodruff, D. 189 Woolley, J. 190 Wurff, W. van der vii, 14, 75–79, 82, 95, 96, 99 Wurmbrand, S. 7, 242 Wyngaerd, G. vanden 51, 53 Y Yaniv, I.

190

Z Zanuttini, R. 159, 276 Zifonun, G. 13 Zwicky, A. M. 142

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 Hans Martin GAERTNER (eds.): Dimensions of Movement. From features to remnants. n.y.p. 49. ALEXIADOU, Artemis (ed.): Theoretical Approaches to Universals. n.y.p. 50. STEINBACH, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. n.y.p. 51. GERLACH, Birgit: Clitics between Syntax and Lexicon. n.y.p. 52. SIMON, Horst J. and Heike WIESE (eds.): Pronouns. Grammar and representation. n.y.p. 53. ZWART, C. Jan-Wouter and Werner ABRAHAM (ed.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26-27, 2000)(Workshop). n.y.p.