barbiers,et.al.2002.modality and its interaction with the verbal system

<DOCINFO

AUTHOR ""

TITLE "Modality and its Interaction with the Verbal System"

SUBJECT "Linguistik Aktuell/Linguistics Today, Volume 47"

KEYWORDS ""

SIZE HEIGHT "220"

WIDTH "150"

VOFFSET "4">

Modality and its Interaction with the Verbal System

Linguistik Aktuell/Linguistics Today

Linguistik Aktuell/Linguistics Today (LA) provides a platform for originalmonograph studies into synchronic and diachronic linguistics. Studies in LAconfront empirical and theoretical problems as these are currently discussed insyntax, semantics, morphology, phonology, and systematic pragmatics with theaim to establish robust empirical generalizations within a universalisticperspective.

Series Editor

Werner Abraham

University of California at BerkeleyUniversity 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 SystemEdited by Sjef Barbiers, Frits Beukema and Wim van der Wurff

Modality and its Interactionwith the Verbal System

Edited by

Sjef BarbiersMeertens Institute, Royal Dutch Academy of Sciences

Frits BeukemaWim van der WurffUniversity of Leiden

John Benjamins Publishing CompanyAmsterdam�/�Philadelphia

The paper used in this publication meets the minimum requirements of American8 TM

National Standard for Information Sciences – Permanence of Paper for PrintedLibrary 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 Beukemaand 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 2002415-dc21 2002276250isbn 90 272 2768 3 (Eur.) / 1 58811 167 9 (US) (Hb; alk. paper)

© 2002 – John Benjamins B.V.No part of this book may be reproduced in any form, by print, photoprint, microfilm, or anyother means, without written permission from the publisher.

John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The NetherlandsJohn 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 languagedevelopment and autism

Anna Papafragou

Negative polarity and modality in Middle Dutch ghe-particleconstructions

Gertjan Postma

(Negative) Imperatives in Slovene

Milena Milojevic Sheppard and Marija Golden

Modality and mood in Macedonian

Olga Mišeska Tomic

Subject index

Name index


Werner AbrahamMaitschern 128,A-8942 Wörschach, [email protected]

Sjef BarbiersMeertens InstituutP.O. Box 942641090 GG Amsterdam,The [email protected]

Frits BeukemaDepartment of EnglishUniversity of LeidenP.O. Box 95152300 RA Leiden,The [email protected]

Elma BlomUtrecht Institute of Linguistics/OTSTrans 10, 3512 JK Utrecht,The [email protected]

Annabel CormackDepartment of Phonetics & LinguisticsUniversity College LondonGower Street, London WC1E 6BT,United [email protected]

Zygmunt FrajzyngierLinguistics DepartmentWoodbury 407 (Campus Box 295)Boulder, CO 80309-0295 [email protected]

Marija GoldenKamniska 4a 2000 Maribor,[email protected]

Anna PapafragouInstitute for Research inCognitive ScienceUniversity of Pennsylvania3401 Walnut Str., Suite 400APhiladelphia, PA [email protected]

Gertjan PostmaDepartment of Dutch/Centre ofLinguisticsUniversity of LeidenP.O. Box 9515 2300 RA Leiden,The [email protected]

Milena Milojevic SheppardDepartment of English andAmerican StudiesFaculty of ArtsUniversity of LjubljanaAškerceva 2, 61000 Ljubljana,[email protected]

Neil V. SmithDepartment of Phonetics andLinguisticsUniversity College LondonGower Street, London WC1E 6BT,Great [email protected]


Olga Mišeska TomicSpinoza Project Lexicon and SyntaxUniversity of LeidenP.O. Box 9515, 2300 RA Leiden,The [email protected]

Wim van der WurffDepartment of EnglishUniversity of LeidenP.O. Box 9515, 2300 RA Leiden,The [email protected]

Preface

The majority of the papers included in this volume were first presented at theworkshop on “Modality in Generative Grammar”, which was held during the32nd annual meeting of the Societas Linguistica Europaea at the Universityof St Andrews (Scotland), 26–30 August 1998. We are very grateful to thosecontributors who responded to our call for papers and to our various deadlinesfor handing in versions of their papers: they deserve our warmest thanks. Thecontributions were subsequently refereed and revised in the light of reviewers’comments. The volume also contains a number of invited papers, by ZygmuntFrajzyngier, Annabel Cormack and Neil Smith, and Anna Papafragou.

Our debt to the linguists who assisted us in the reviewing process is enor-mous. Their help was freely and generously given; without it the volume couldnot have come about.

The editing process was originally in the hands of Sjef Barbiers and FritsBeukema; in September 2000 Wim van der Wurff joined them, and helped toprepare the volume for the press.

Sjef BarbiersFrits BeukemaWim van der Wurff

Current issues in modality

An introduction to modality and its interactionwith the verbal system

Sjef BarbiersRoyal Netherlands Academy of Sciences, Meertens Institute, Amsterdam

. Introduction

The goal of this introduction is to provide an overview of the research questionsconcerning modality that have been addressed in the framework of generativegrammar, of the answers that have been provided to these questions, and ofthe ways in which the papers in this volume relate to these questions. Modalityis a semantic concept and covers, inter alia, notions such as possibility, neces-sity, probability, obligation, permission, ability and volition. Modality can beexpressed syntactically by modal verbs, imperatives, verbal inflection, modaladverbs and modal particles.

. The epistemic–root distinction

A great deal of work on modality in generative syntax has concentrated onthe distinction between epistemic and root interpretations of modal verbs.Roughly, epistemic interpretations are a class of interpretations involving aspeaker-oriented, or, in the case of embedded clauses, matrix-subject orientedqualification or modification of the truth of a proposition, while root inter-pretations involve the will, ability, permission or obligation to perform someaction or bring about some state of affairs (cf. Lyons 1977 and Palmer 1986 fora fine-grained classification and description of modal interpretations). Sen-tences 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 andthe 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 beat 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 entailsthe existence of two different syntactic structures. One of the first answersto this question (Ross 1969) is that modals with an epistemic interpretationare one-place predicates, taking the entire proposition as their complement,whereas root modals are two-place predicates, i.e. they involve relations be-tween 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 beeninvested to translate it into syntactic terms.

There are different ways to represent the alternation between monadic anddyadic modality syntactically. A first option would be to try to reduce themodal alternation to similar alternations with verbs like BREAK and SMOKE.However, it will be shown in Section 3 that these are alternations of a funda-mentally different type.

A second way to represent the alternation in terms of argument structureis to analyse root modals as control structures and epistemic modals as sub-ject raising structures. This is the most common generative analysis of theroot–epistemic ambiguity (Hofmann 1966; Ross 1969; Perlmutter 1970 andnumerous studies since then). However, as shown in Section 4, there are anumber of serious problems for this analysis.

Thirdly, the modal alternation could be the result of generating modals indifferent base positions, a higher, possibly functional, position for epistemicmodals and a lower, possibly lexical, position for root modals. This analysis isnot entirely without problems either, as Sections 5.1–5.3 show.

A fourth way to analyse the modal alternation is to assume that epistemicmodals, but not root modals, undergo movement at LF. This possibility is dis-cussed in Section 5.4. On this approach, the monadic nature of the modal verbin epistemic interpretations is not established until LF. An immediate questionfor this analysis is how the trace of the modal should be interpreted.


A final type of analysis, discussed in Section 6, is to attribute the differencein modal interpretation to the selection of different types of complements, alarger complement being selected in the case of epistemic interpretations, anda smaller one in the case of root interpretations. This amounts to providing adifferent perspective on the possibility of generating modals in different headpositions. Obviously, if modals are generated in different positions depend-ing on their interpretation, the type and size of their complements will varyaccordingly.

. Transitive–intransitive alternations

As is well known, English modals have a defective paradigm and a distribu-tion differing from that of main verbs (cf. Section 5). Therefore, the ques-tion as to whether the dyadic–monadic alternation is reducible to othertransitive– intransitive alternations does not seem to be relevant for English.It is a relevant question, however, for languages in which modals have manyproperties in common with main verbs, such as Dutch and German. This sec-tion shows that in such languages the modal alternation is not reducible totransitive– intransitive alternations either.

Transitive– intransitive alternations are found cross-linguistically withmain verbs like BREAK and SMOKE (3a–d). A common analysis of such alter-nations is that each of these verbs has only one lexical entry, which correspondsto the transitive variant. The intransitive variant is derived by a lexical opera-tion that reduces transitive predicates to intransitive predicates by removingthe external argument.

(3) a. John broke the window.b. The window broke.c. John smoked the eel.d. The chimney smoked.

In fact, BREAK and SMOKE represent two different types of alternations:BREAK is a transitive–unaccusative alternation and SMOKE a transitive–un-ergative alternation (cf. Abraham 2000). The two can be distinguished by us-ing classic tests described in Burzio (1986) and Hoekstra (1984). In Dutch, theunaccusative alternant selects the auxiliary BE in the perfect tense and its ad-jectival participle can modify a noun that corresponds to the unaccusative’ssubject. The unergative alternant selects the auxiliary HAVE and its participlecannot modify such a noun. This is illustrated in (4a–f).

Sjef Barbiers

(4) transitive–unaccusativea. Jan breekt de ruit. b. De ruit is/*heeft gebroken.

John breaks the window the window is/has brokenc. de gebroken ruit

the broken window

transitive–unergatived. Jan rookt een paling. e. De kachel heeft/*is gerookt.1

John smokes an eel the stove has/is smokedf. *de gerookte kachel

the smoked stove

If the root–epistemic ambiguity of modals were to involve reduction of dyadicpredicates to monadic ones, then these tests would show that modals belong tothe same class as roken ‘smoke’, as (5a–c) illustrate. In the perfect tense HAVE isselected, and the participle cannot modify a noun corresponding to the surfacesubject of the intransitive alternant:

(5) root–epistemica. Jan mag eten. b. Eten heeft/*is altijd gemogen.

John may eat eat has/is always may-

‘John is allowed to eat.’ ‘It has always been allowed to eat.’c. *het gemogen eten

the allow- eating

However, the monadic alternant of the modal (5b) does not have an epistemicinterpretation, which shows that the epistemic–root ambiguity is not an in-stance of the transitive–unergative alternation either. Moreover, for the epis-temic interpretation to be possible, the external argument does not need todisappear, as (1) shows. We can thus conclude that the epistemic–root ambi-guity can neither be reduced to the transitive–unaccusative alternation nor tothe transitive–unergative alternation.

. The raising–control analysis of the epistemic–root ambiguity

A consequence of treating root modals as dyadic predicates is that they involvea control structure, if something like the theta-criterion holds. If a root modalassigns a theta-role to the surface subject DP and if a DP must receive exactlyone 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-


ded predicate and the modal.2 Therefore, root modals can only be analysed asdyadic predicates if they involve control structures, which implies that PRO inthe complement of the modal receives the external theta-role of the embeddedverb, 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 & epistemicII. [[DP John]i must [PROi work from nine to five] control & root

The control–raising analysis rather elegantly captures the idea that root modalsare 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 modalis problematical because infinitival complements without te ‘to’ usually cannothave 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 toexpress the same meaning as (7b). If this infinitival complement can be anal-ysed as a small clause (SC), the impossibility of PRO in (7a) is an instance ofthe 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 clausesas their complements, but larger constituents, e.g. CP or TP. For the Germaniclanguages, there does not seem to be much evidence, however, that this is thecase. C-elements such as complementisers are impossible in the complementof a modal, as is embedded WH. If infinitival to is a T-element, its obligatoryabsence in most modal contexts shows that the complement of a modal is nota TP either. Higher adjuncts such as evidential and modal adverbs cannot bepart of the complement of a modal, which also suggests that complements ofmodals are smaller than CP.

However, it cannot be concluded from the situation in Germanic that itis generally impossible for modals to have a CP or TP complement. As Tomic(this volume) argues, Macedonian has a set of modals that take subjunctive

Sjef Barbiers

CP-complements. Incidentally, these Macedonian constructions should notbe confused with cases in which modals in Germanic seemingly take a CP-complement, as in (8a). There is reason to think that CP in this constructionis 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 ver-slaat].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 isprovided by sentences in which the matrix subject cannot be interpreted as theperson being obliged or permitted to do something (Feldman 1986; Klooster1986; 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 fiveo’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 entireclause 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 orpermission to do something, but the crucial point is that this addressee cannotbe 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 dyadicverbs, since dyadic verbs which assign an external theta role satisfy an impor-tant condition for passivisability (Burzio 1986). However, it should be notedthat modals share this inability to occur in the passive with other stative tran-sitive verbs. It may be that the crucial factor blocking passivisation here isstativity.

A third problem for the control analysis of root modals involves the be-haviour of expletive subjects (Thráinsson and Vikner 1995; Warner 1993;McGinnis 1993). In some languages, e.g. English, expletive subjects occur with


raising verbs but not with control verbs (10a, b). If root interpretations alwaysinvolved control structures, expletive subjects are predicted to be incompatiblewith 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 rootinterpretation behave like raising predicates with respect to quirky case. Whena verb requiring quirky case on its subject is embedded under a raising verb, thesurface subject appears with quirky case. When such a verb is embedded undera control verb, the surface subject appears with nominative case. Regardless ofthe interpretation, the subject of modal clauses always occurs with quirky casewhen 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 com-plement, e.g. a PP or AP complement.The construction can also be foundin Afrikaans, German, Norwegian, Yiddish and West Frisian.5 Modals with aPP or AP complement are subject raising constructions. Yet, they can only getnon-epistemic interpretations.

Finally, Wurmbrand and Bobaljik (1999) show that A-reconstruction forscope, i.e. the subject taking scope under the modal, is possible for raisingverbs and for both root and epistemic modals, whereas this is impossible forcontrol verbs.

In sum, the observations presented in this section pose serious problemsfor a control analysis of modals with a root interpretation, but not for a raisinganalysis.

. The structural position of modals

An alternative way to shed some light on the epistemic–root distinction is toinvestigate the structural position of modals. It turns out, however, that theirposition is rather difficult to establish; a simple and universal picture seemsvirtually impossible given the empirical facts. At least four distinctions have tobe 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 parametri-sation involves all four distinctions.

. Different positions for modal auxiliaries and modal main verbs

The distinction between modals behaving like auxiliaries and modals behav-ing like main verbs has engendered a great deal of discussion. A well-knownview is that modals in English are auxiliaries generated in Infl or Aux (cf.Roberts 1985) because they have a defective inflectional paradigm and behavelike other auxiliaries in that they can undergo residual V2 and do not triggerDO-support in negative clauses. On this view, modals in languages like Ger-man and Dutch are main verbs projecting their own VPs, as their properties arelargely (though not completely) identical to those of main verbs. Clearly, thedistinction between main verb and auxiliary modals does not shed any light onthe epistemic–root ambiguity.

. Different positions for epistemic and root modals

Although it is true that across and even within languages modals may differwith respect to their categorial status, the idea that this distinction alone de-termines the structural position of modals has turned out to be too simple. Acomplication is that there are languages such as Catalan that seem to have twopositions for modals (Picallo 1990). There is a higher position above clausalnegation that corresponds to epistemic interpretations, and there is a lowerposition below clausal negation corresponding to root interpretations.

A similar view is defended in Cinque (1999), where it is claimed on thebasis of the relative ordering of different adverb types that the universal order ofmodal heads is such that epistemic modals c-command root modals. Evidencepotentially supporting this idea comes from languages that allow for doublemodals. In such languages a modal with a root interpretation cannot take scopeover a modal with an epistemic interpretation (cf. Barbiers 1995 for Dutch;Thráinsson and Vikner 1995 for Icelandic; Roussou 1999 for Greek).


. Different positions for necessity and possibility modals

Cormack and Smith (this volume) provide the most comprehensive analysisof the interaction between negation and modality that is currently available.They argue that English has two positions for modals too; a position higherthan clausal negation and a position lower than clausal negation. However,in English this syntactic split does not correspond to the epistemic–root dis-tinction, as it does in Catalan, but to a distinction between necessity modality(high) and possibility modality (low). They show that there are modals with aroot interpretation occurring in the higher modal position and modals with anepistemic interpretation occurring in the lower modal position. Postma (thisvolume) argues on the basis of the distribution of the negative polarity prefixghe- 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 impos-sibility of a root modal taking scope over an epistemic modal cannot have asyntactic 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 modalis pronounced should be distinguished from the position in which it is inter-preted. This can be done in various ways, depending on the set of assumptionsadopted. Cormack and Smith (this volume) argue that a modal is inserted inthe position where it is interpreted, while the position of pronunciation maybe different. A second possibility is that scopal ambiguities as in They can fre-quently buy eggs are the result of overt raising of can across frequently (Ernst1992). The narrow scope reading of can may then be obtained by reconstruc-tion at LF or by computing the scope over the trace of can.

It has also been argued that (some) modals are quantifiers that undergoLF-raising (McDowell 1987).6 An epistemic modal would differ from a rootmodal in that it raises to a clause initial position at LF where it takes scopeover the entire proposition. McDowell claims that the incompatibility of epis-temic interpretations and questions supports the LF-raising analysis. This be-comes 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 ofFinnish.’

b. *Moet Jan wel een moedertaalspreker van het Fins zijn?Must John a native speaker of the Finnish beIntended meaning: ‘Is it necessarily true that John is a native speakerof Finnish?’

McDowell’s syntactic explanation for the fact that a clause with an epistemicmodal cannot be interrogative is based on the doubly-filled Comp filter. Thisfilter blocks structures in which Comp contains more than one element. Ques-tions involve a Wh- or Q-operator; when such an operator fills Comp, it is nolonger possible for an epistemic modal to move to Comp.

This analysis does not automatically carry over to the current generativeframework, in which there is a proliferation of left-peripheral functional po-sitions that could potentially host operators. Moreover, McDowell’s observa-tion requires further qualification, since the incompatibility of question for-mation with epistemic interpretations does not seem to hold for every type ofmodal. 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 interrogativeoperators requires further research.

. Developmental issues

It has often been observed that first language learners acquire root modalsearlier than epistemic modals. If it is possible to show that epistemic modalsare generated in a higher syntactic position than root modals, this observationcould be explained in a framework that assumes that language acquisition oc-curs 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 itsarguments, while functional projections come in later: the lower a functionalprojection, the earlier it occurs. However, if Cormack and Smith’s claim (thisvolume) is correct that there is no strict correlation between epistemic–rootand height of insertion, such an explanation is problematical.


Papafragou (this volume) tries to explain the lagging behind of epistemicmodality 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. Ifsuch abilities do not develop, as in the case of autism, epistemic modalityremains absent.

Another interesting and much discussed developmental issue involves theoccurrence of so-called root infinitives in many child languages. Children ata certain age use infinitival clauses as independent clauses, where in the adultgrammar an independent clause would require a finite verb. Despite the ab-sence of a finite verb or a modal, such infinitival clauses often carry modalmeaning. The question then arises where this modal interpretation comesfrom. Does the infinitive itself or the infinitival suffix that it contains con-tribute 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 andHoekstra, she argues that it is indeed the infinitival morphology that con-tributes to modal interpretation. She explains the apparent problem that chil-dren also use non-modal root infinitives by assuming that children first haveto learn that infinitives can be either nominal or verbal. As soon as they knowthat they can be verbal, modal interpretations arise.

The idea that an infinitive or infinitival morphology contributes to modalinterpretation 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 probleeman 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 inter-action, or just the absence of finite tense on the main verb. Many interestingissues arise, among which questions about cross-linguistic differences, such asthe fact that not all languages can have modally interpreted infinitives in thethree constructions in (13).

Sjef Barbiers

. The complement of modals

The syntactic representation of the epistemic–root distinction can also bestudied from the perspective of the modal’s complement. Relatively little atten-tion has been paid in the literature to the influence of the modal’s complementon modal interpretation. This section provides an overview.

. Non-disambiguating complements

At first sight, it seems plausible to expect that modals with a root interpreta-tion require verbs assigning an agent role as their complements. An agent mayseem to be necessary to carry the permission, obligation, ability or will to per-form the action denoted by the embedded verb. Passive, unaccusative, stativeand complements with a perfect might then be expected to force epistemic in-terpretations. In passives, the agent role is demoted, in unaccusatives there isno agent role, stative verbs denote a state of affairs, not an action that can beperformed by an agent, and in perfectives the event denoted by the verb hasalready been completed. However, these types of complements all allow rootinterpretations:

(14) a. This room must be cleaned immediately. passive‘You must clean this room immediately’

b. Je mag niet weglopen. unaccusativeYou 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 eventthat cannot change (anymore) force an epistemic interpretation. There are twotypes of complements for which this is the case:

(15) i. Stative complements which contain an individual-levelpredicate, provided that the subject has fixed reference.7

ii. Perfect complements, but only if the completion stage of the event hastaken place in the past.


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

(16) a. John must be a native speaker of Finnish.#‘John has the obligation to be a native speaker of Finnish.’

b. They must have cleaned this room yesterday.#‘They must have had the obligation to clean this room yesterday.’

These two observations do not seem to have consequences for the syntac-tic analysis of modal ambiguity, as disambiguation is the result of semanticproperties of the complements. Nevertheless, the two types of complements in(16) are useful diagnostics to establish whether an epistemic interpretation ispossible in a certain construction.

. Complements that force root interpretations

For an epistemic interpretation to be available, it is necessary that the comple-ment contain an infinitival verb. This can be seen in languages in which modalsallow non-verbal complements, such as Dutch. Such constructions can onlyhave root interpretations. In Barbiers (1995, this volume), I provide evidencethat such complements are truly non-verbal, and not elliptical.

Based on these observations I conclude that the availability of root andepistemic interpretations is determined by the complement of the modal.When the modal has VP, PP or AP as its complement, it receives a root inter-pretation; when it has vP (little vP, the abstract verbal projection dominatingVP; 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 per-fect. Modals in the present perfect can only have non-epistemic interpreta-tions; a modal in the present perfect cannot co-occur with a complement thatforces an epistemic interpretation (17a). For possible explanations, see Abra-ham (this volume), Diewald (1993), Zifonun et al. (1997) for German, andBarbiers (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 mustbe‘John should have been a native speaker of Finnish (then he wouldhave got the job).’

Interestingly, the disambiguating effect of the stative and individual-level com-plement disappears in the past perfect, as (17b) shows. The modal in (17b) hasa root interpretation. The fact that the complement allows this in this construc-tion is due to the irrealis interpretation forced by the past perfect: the sentenceentails that John is not a native speaker of Finnish but could have been one ina different world. This makes it possible to interpret being a native speaker as avariable 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 under-standing of modality. As we have seen above, many of the papers in this volumepay attention to it. In particular, the paper by Cormack and Smith providesa detailed description and novel analysis of the interaction between negationand modality in English. Two interesting observations can be found in the pa-pers by Beukema and van der Wurff and by Postma. As Beukema and van derWurff observe, the presence of modals turns out to make overt raising of neg-ative constituents possible. They show that in fifteenth-century English, whenEnglish had changed from OV to VO, OV order could still occur, but only inrestricted environments, one of which involves a negative object enclosed be-tween a modal verb and a lexical verb. Postma argues that possibility modalityis necessary to license verbal negative polarity items.

. The coding of modality

Modal verbs are just one way to encode modality in natural languages. Amongthe other means attested are modal particles and adverbs, imperatives and to-infinitives. Frajzyngier (this volume) shows how verbal inflection, word order,auxiliary verbs and modal particles participate in the coding of modality inLele, an East-Chadic language. In Slovene (Sheppard and Golden, this vol-


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

. Conclusion

Generative syntactic research of the past forty years has brought to light a largenumber of properties of modals and modal constructions. Many of these prop-erties are discussed in this volume, some of them for the first time. Questionsabout the argument structure of modals, the position of modals and the rela-tion between syntactic properties and semantic ambiguity have been sharpenedconsiderably. Many questions remain, however. A number of serious problemshave been raised for the raising-control analysis of the epistemic-root ambi-guity. Alternative analyses have been proposed, but so far there is no conclu-sive evidence to choose between them. Another important issue involves thetenability of the universal base hypothesis as exemplified by Cinque’s (1999)approach to modal positions, in the light of the observed cross-linguistic dif-ferences.

Notes

. I am using an underlying object different from the one in (4d) to avoid complicationscaused 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, wheremodals behave like raising verbs regardless of interpretation. They attempt to solve this byassuming that modals assign so-called adjunct theta roles. The theta criterion would dictatethat a DP should not receive two theta roles, but receiving one theta role and one adjuncttheta role would be allowed.

. Note that this is true at the observational level. The theory-internal hypothesis that PROcannot 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 pointat issue here is that the modal verb itself cannot be passivised.

. The construction can also be found in Afrikaans, German, Norwegian, Yiddish and WestFrisian.

. 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 constantacross 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 Participioeffect: 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 OldIcelandic and Old Swedish.

References

Abraham, W. (2000). The aspect-case typology correlation. Perfectivity and Burzio’sgeneralization. In E.J. Reuland (Ed.), Arguments and Case. Explaining Burzio’s General-ization (pp. 131–193). Amsterdam: John Benjamins [Linguistik Aktuell/LinguisticsToday 34].

Barbiers, S. (1995). The Syntax of Interpretation. PhD diss. Leiden. The Hague: HAGPublications.

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. Essaysin Linguistics in Honor of Sylvain Bromberger (pp. 53–110). Cambridge (Mass.): TheMIT 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: AcademicPress.

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 composi-

tional analysis. MA Thesis, University of Toronto.Palmer, F. (1986). Mood and Modality. Cambridge: Cambridge University Press.


Perlmutter, D. (1970). The two verbs begin. In R. Jacobs and P. Rosenbaum (Eds.), Readingsin 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 V2phenomena in language acquisition (pp. 23–62). Dordrecht: Kluwer.

Roberts, I. (1985). Agreement patterns and the development of English auxiliaries. NaturalLanguage and Linguistic Theory, 3, 21–58.

Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in PhilosophicalLinguistics, Series 1 (pp. 77–102). Evanston: Great Expectations.

Roussou, A. (1999). Modals and the subjunctive. In A. Alexiadou et al. (Eds.), Studies inGreek 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 deGruyter.

Modal verbs*Epistemics in German and English

Werner AbrahamUniversity of Groningen

The distribution of modal verbs in German is considerably wider than that oftheir Modern English counterparts. It is generally claimed that this is partlydue to the V-to-I movement in the history of Modern English. However, thereis a considerable remainder of differences between Old English modals andtheir Modern English equivalents that cannot have anything to do with theposition 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 ofaspectual or Aktionsart properties in the course of early Middle English.

. Introduction: A solid set of differences between Modern English andModern German

Diachronically, modal verbs (MVs) in Modern English and in Modern Germanderive from a common source and therefore demonstrate a solid equivalence,both syntactic and semantic, in their older stages (Old English, OE, and OldHigh German, OHG), but they differ considerably in their present-day distri-butional properties. Generally speaking, MVs in Modern German represent afar more homogeneous verbal class than their counterparts in Modern English.They display a clear morphological distinction from non-modal verbs, but, atthe same time, they show a considerable syntactic similarity to them (as shownin great detail in Reis 2000). The morphological distinction is partly sharedby Modern English modals while the syntactic similarity that German modalsexhibit 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 Englishmodals 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 ProfessionalAmerican Modern English, Part 1 (93,472 words)

will may might shall must

temporal 1655 0root 24 0 0 0epistemic 8 84 144 0 0

Table 2. Frequency of modal expressions in the Corpus of Spoken Professional AmericanModern 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 ofspoken American English.

A precise word count on Part 1 of the Corpus of Spoken Professional Ameri-can Modern English (in CD-format) covering 93,472 words reveals a total of 918modal verbs and modal expressions. This result confirms the impression onegathers when browsing through the 19 other Parts of the CD-ROM (the totalCorpus of Spoken Professional American Modern English covers some 2 millionwords). 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 tooccur anymore in Modern English. This is in stark contrast to Modern Ger-man wollen and sollen, which have been preserved as full deontic main verbsthroughout their historical development. Also note that the notional equiva-lents of the root modals (want to, wish to, be permitted, have to, is to) are repre-sented in sizeable numbers. Furthermore, the preterite forms might and shouldare rather frequent in the part of the corpus investigated here; most of the casesinvolve epistemic meanings, thus reinforcing the tendency for present-tensemodals to have non-epistemic meanings. In particular, whenever will occurs, ithas a predominantly temporal meaning. The remainder of the modals in Mod-ern American English has a shaky status: with the exception of must and singleidiomaticised root uses of other modals (e.g. May I?), there are only epistemicuses left. Moreover, will and shall had already attained a purely temporal func-tion at some earlier stage of their development. While Coates (1995) estimatesthat Modern British English must has root meaning in half of all cases and epis-temic meaning in the other half, this is not reflected in the American Englishdata 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 Englishreveals that epistemic may is far more frequent.1 Recall that English past tensemodals 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 (andDutch; see Barbiers 1995) have retained their polyfunctionality (i.e. function-ing both as root-deontics and epistemics, sometimes adding ancillary temporaluses as Dutch zullen for the future), Modern American English is about to giveup, or has lost already, the complete array of root modal meanings: with cer-tain very specific provisions, all modals in American Modern English functionas epistemics or, in the case of will, as a temporal auxiliary.

The properties of Old English MVs were originally similar to those of Mod-ern German. We also observe that the verbal lexicon in OE as well as in Mod-ern German has a large number of verbs derived by means of prefixed particleswhich are based on adjectives, prepositions, and adverbs. These prefixes areseparable from the verbal stem (in subordinate clauses, in topicalisation struc-tures and in stylistic fronting etc.) and therefore have the syntactic status ofsecondary predicates (small clauses). This is still the unmarked and productivesituation in Modern German. Lexically signalled aspect, or Aktionsart mor-phology, has disappeared in Modern British and American English. Moreover,OE as well as Modern German – but not Modern English – make a distinctionbetween a stative passive and a dynamic passive with the help of auxiliary se-lection (OE beon vs. weorðan + past participle, Modern German sein vs. werden+ past participle). In English, however, the stative passive be + past participlehas additionally acquired a dynamic interpretation by extension of a generalimplicational 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). Anadditional difference is that the ge-morpheme as used in the past participle stillfigures prominently in Modern German,2 while English lost it in early Mid-dle English (ME). The participial morpheme was formally extended to the pastparticiple from an original lexical perfectivising prefix, which also still survivesin German verbs like geschehen ‘happen’. This reflects the historical situation inwhich the past participle had a perfective stative (i.e. resultative) function only(as documented for Gothic as well as Early Old High German; cf. Abraham1992). The dynamic meaning was acquired from the resultative periphrasis. Allcontinental West Germanic languages (German, Dutch, Flemish, West Frisian,Yiddish and their dialects) share these three properties: 1. paradigms of sepa-rable 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 participialmorpheme ge-. None of these three properties are found in Modern English. Iwill argue that this threefold characterisation lies at the heart of the differencebetween the MV paradigms of Modern English and Modern German – exactlyhow 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 sta-tus as formal preterite presents, as opposed to all other full verb paradigms.However, MVs in Modern English have done so in ways different fromthose in Modern German: generally speaking, Modern German modalsshare many syntactic properties with full verbs, while those of ModernEnglish do not. See also 3 below.

2. MVs in Modern German can be shown to be sensitive to aspect (or Ak-tionsart), unlike their counterparts in Modern English.

3. Modern English MVs are far more grammaticalised than MVs in Mod-ern German. The latter show many corresponding properties with lexicalverbs. Modern English MVs do not show such properties. Rather, they be-have syntactically like auxiliaries (e.g. as regards the syntax of do-support,a syntactic feature that Modern German does not share with ModernEnglish).

4. As shown in Section 1 above, MVs in American Modern English have lostthe 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 epis-temically (may, must, can) or exclusively as temporal auxiliaries ((shall),will). As to their root meanings, fully lexical paraphrases have replaced themodals (be permitted instead of root may, have to instead of root must witha personal subject, as well as their negations).

Against the background of the radical restructuring of the lexical-aspectual ver-bal system in the historical development of English the following questionsneed to be answered:

1. What is the precise nature of the sensitivity to aspect or Aktionsart of Ger-man (and Old English) MVs? Why is it that MVs in Modern English do notshare these characteristics?


2. If the difference between epistemic versus root readings is the result ofdistributional aspect distinctions, as claimed here, does this carry over to thegreatly deprived status of Modern English MVs? Is there a structural and func-tional 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 generaldemise of contextual aspect dependencies? If true, this would bring the epis-temic readings of Modern German in line with the epistemic and temporalreadings of modal verbs in Modern English, and it would help explain thealmost total absence of root readings in American Modern English.4. Can a link be established between the development of the class of evidentialmodal verbs from perfect or perfective origins and the development of GermanEMVs from earlier (i.e. pre-OHG) preterite presents?5. Is Modern English really without aspect or has a new type of aspectualsystem emerged? Here we could think of the notion ‘ ’: at somepoint this came to be expressed by the perfect (instead of the earlier presenttense), 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 andmorphological differences of the MVs in Modern English and Modern Germanare presented. The behaviour of German modal verbs in contexts of perfectiveand perfect embeddings will then be discussed, yielding the first importantfinding: i.e. that the deontic/root vs. epistemic/evidential distinction followsfrom the Aktionsart of the embedded lexical verb in Modern German, but notin Modern English. Next, the theta-role properties of modal subjects will beshown to collocate with distributional differences that set wollen and mögenapart from the other modals. The main conclusion here will be that MVs are,to all appearances, raising predicates and that control in the traditional senseis not involved. The overall conclusion will be the following: the fact that Ger-man, but not Modern English, allows infinitival forms of root modals, as wellas the fact that epistemic readings are confined to finite MVs in both languages,can be explained on the grounds that non-finiteness presupposes lexical theta-marking (L-marking) and that it is only this type of theta-marking that goeshand 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 theliterature.

Werner Abraham

. The major distributional differences between English and Germanmodal verbs

The traditional view of the syntactic difference between Modern English andModern German modal verbs is that, since Modern English modals cannotproject non-finite forms, they project into the syntactic I(nflection) category(although this is contested, on a comparative-typological basis, by Rohrbacher1999), while German root modals project into V (see Abraham 1995:469ff.;1998; Sprouse & Durbin 1997). Main verbs in Modern English surface inV, while modals and finite auxiliaries such as have and be surface in I. Thephrase marker in (1) collapses the relevant s-structures of Modern English andModern German.

(1) IP

I’

IModern English VP

MVAuxfinite

VModern English

Vlexical

VGerman

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 readingsmust be distinguished from their epistemic readings.

In (2), a number of syntactic contrasts between Modern English and Ger-man are listed (following partly Durban & Sprouse 1998 and Reis 2000).

(2) M E G

a. s/he sees/can(*s)/may(*s)/ sieht/kann(*t)/darf(*t)/muß(*t)must(*s)can, shall ( = ) / kann/können, darf/dürfen, mag-

mögen, muß/müssen;ihr seht/könnt/dürft/müßt/wollt3

b. *s/he musted/could/should er mußte/konnte/solltec. *has could/*should/*would hat gekonnt/gemußt/gedurft

d. *interesting to can/may work (to+) interessant arbeitenzu können/dürfen


e. *must can work - muß arbeiten können

f. *I can German Ich kann Englisch.*Work must, *Smoke may *Arbeiten muß/

*Rauchen darfg. *I work not Er sieht/kann/darf das nicht

I do/must not workh. *Works Mary? Sieht/Muß/Darf Klaus das?

Must/Does Mary work?j. *Mary works hard, –

worksn’t she?Mary can work hard, –can’t she?

k. *that Claus that not sees/ - daß Klaus das nicht sieht/can/may kann/darf

l. may be right/must be (-) mag stimmen/muß sichwrong täuschen

m. *May that be right? *Mag das stimmen? (*EMV!)*Shall he come? Soll er kommen?

(DMV/*EMV!)p. *This may, I please, *Das mag gefälligst stimmen!

be right!

Note that a number of the distributional properties listed in (2) follow fromthe fact that the Modern English MV is base-generated in I, which does notaccept 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 atfirst sight, require further explanation. I will argue presently that this is relatedto an ‘inherent aspect’ (i.e. Aktionsart4) property, present in German, but notin Modern English.

Among the Germanic languages, Dutch is the most interesting language tocompare with German, to the extent that it is even more conservative mor-phologically and has experienced greater syntactic change. Clearly, Germanand Dutch display the most conservative modals, whereas Modern English hasevolved the most strongly grammaticalised forms and meanings, even beyondthose in Scandinavian. I will briefly examine the formal and distributional cri-teria 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 tohave undergone stronger grammaticalisation.

Werner Abraham

(3) D G

a. Ik kan Engels ich kann/*darf/*muß/mag Englischb. Dat kan die! Das kann der!c. Werken moet/Roken mag. *Schaffen muß/*Rauchen darf.d. Piet moet haar *(niet). *Peter darf sie nichte. De wijn moet op. Der Wein muß alle/*auff. Koken dat Piet kan! *Kochen daß P. kann/Kochen wie

der Peter kann!g. Hij kan/mag/moet dit niet Er kann/darf/*muß das nichth. Wil/moet/mag hij dat? Will/darf/muß er das?k. Dat Klaas dit niet ziet/kan daß K. das nicht kann/darf/mußl. 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 modalverbs as full transitive verbs. The diachronic processes behind this differ-ence possibly instantiate cases of pragmatic ellipsis, which have become fullygrammaticalised as full verb uses. The examples in (3l–n) demonstrate strongclause-type constraints on epistemic readings both in German and Dutch.More precisely, the epistemic uses of MVs in Dutch and German are restrictedto declaratives. This is in itself a non-trivial generalisation. A tentative conclu-sion to be drawn from this is that the position of the EMV is so high in theclause 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 Ameri-can English has virtually lost the root uses of its MVs. What could be the rea-son for this deficiency? I have argued above that it can be attributed mainly tothe absence of the aspectual/Aktionsart properties that the root-deontic MVs(see Abraham 2000) have retained in German and Dutch, together with theiroriginal main verb status. The next sections provide a closer look at aspect inrelation to MVs.


. Syntactic reflexes of the root-deontic vs. epistemic distinctionin German

In the following discussion of root vs. epistemic readings of the modalverbs knowledge of the massive literature on this topic is presupposed. Noattempt will be made to motivate the terminology (root/deontic vs. epis-temic/inferential/subjective/evidential, etc.) or define its use in this article.5

Rather, in line with modern syntactic methodology, the meaning distinctionswill be taken to be inferable from the totality of the distributional character-istics provided above and below (see, however, Section 5.4 and Section 6 forsemantic considerations concerning event structure, argumental theta assign-ment, and scope).

Suffice it to say here that the syntactic reflexes of the root vs. epistemic dis-tinction in German established above show a surprising dependency on thedistributional criteria spelled out in (3a–g). Furthermore, aspect and lexicalAktionsart 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 Eand T in the characterisation in (6a–c) stand for the Reichenbach notation ofevent tensing (E = event, T = temporal point of the utterance; ‘E,T’ = presenttense reference; ‘E_T’ = past tense reference; ‘T_E’ = future event reference;‘Perf ’ = perfective, i.e. event-bounded). Where both readings are possible Iprovide glosses and translations.

(4) MV-: DMV EMVa. -

Er hat(te) viel Geld verdienen wollen/müssen/sollen + *He has/had much money earn will/must/shallDMV-reading: ‘He wanted/had to earn much money’

b. Das scheint so sein zu müssen (0) (+)That seems so be to must

c.

Er wollte/mußte/sollte viel Geld verdienen + +He would/must/should much money earnDMV-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 (0) (*)

The raising verb scheinen ‘seem’, as shown by (4b, d), patterns with EMV (butnot DMV) in its non-finite use.

Werner Abraham

(5) A : DMV EMVa. Er will/soll/muß schlafen [–Perf] + +

He will/shall/must sleepEMV-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 fallasleep’

c. Er will/soll/muß Geld verdienen [–Perf] + +He will/shall/must money earnDMV-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’

(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 inder UB gearbeitet haben zu können/müssen/dürfen [E_T] * *. . . worked have to can/must/may

c. *Es ist dem Angeklagten sehr wichtig inder UB arbeiten können/müssen/dürfen zu werden [T_E] * *. . . worked have can/must/may to become

(7) E ()

a. Von ihm muß/soll/will viel Geld verdient werden + +By him must/shall/will much money earned becomeDMV ‘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

(8) MV- MVa. 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’


DMV EMVc. Es muß/müßte sein, daß man sagt, er * +

hat diesen Brief geschriebenIt must/must-. be that one says he has this letterwritten‘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 canb. Er muß/müßte diesen Brief haben

schreiben können/*gekonnt + *He must this letter have write can/could

c. Er muß/müßte diesen Brief schreiben habenkönnen/*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 ex-pressions (German Englisch können ‘to can English’; Dutch werken moet ‘towork must’, etc.) have a wide pragmatic extension, that is, their use is id-iomatically restricted. Not much can therefore be concluded for the syntax.These idiom-like structures are not possible in Modern English, but they arein Dutch and in German (although to a lesser degree in German; compareBarbiers 1995:170–173; 189; 192; 204).

Further notice needs to be taken of the fact that in Modern English thereare no root or epistemic readings for will and shall, but only temporal ones (tothe extent that shall survives at all. Recall that in American English the modalroot readings except for must are on their way to becoming obsolete (see Table 1above). Except for idiomatically frozen uses (such as e.g. May I?), be permittedis preferred to root may, be able is preferred to root-alethic can, and have to isincreasingly used instead of root must. What this means for the status of DMVsin 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 dis-tributional behaviour in keeping with full lexical verbs, also in prenominalattributive 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 construc-tions are possible in English.

(12) a. alle Deutsch sprechenden Studentenall German speaking students

b. alle Deutsch (sprechen) könnenden/müssenden/dürfenden Studentenall German speak canning/musting/maying students

c. alle Deutsch gesprochen habenden Studentenall German spoken having students

d. alle gerade angekommen-(seiend-)en Studentenall just arrived being students

e. alle Deutsch gelernt habenmüssenden/könnenden/dürfenden Studentenall German learned have musting/canning/maying students

f. (?)alle Deutsch lernen müssen/können/dürfen habenden Studentenall German learn must/can/may having students


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

Despite this pervasive similarity to full lexical verbs, DMVs in Modern Ger-man do show some differences from full lexical verbs. In contrast to Dutch,German MVs are more restricted with respect to intransitive usage and infiniti-val 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, althoughit is not unrestricted (see (13b)–(14b) and (15b) below). Sprouse and Durbin(1998) have investigated the collocations of MVs with tun. However, it isdoubtful whether relevant conclusions can be drawn from the examples be-low beyond the fact that there are – somewhat idiomatically restricted – fullverb uses (see Abraham and Conradie 2001:81).

(13) a. Er schreibt gut Englisch, aber er spricht es nicht so gutHe writes well English, but he speaks it not so well

b. Er tut gut Englisch schreiben, aber er tut nicht so gut sprechenHe does well English write, but he does not so well speak

(14) a. Er kann gut EnglischHe can well English

b. ?Er tut gut Englisch könnenHe does well English can

(15) a. Er hat Englisch gesprochenHe has English spoken

b.**Er tut gut Englisch gekonnt/gesprochen habenHe does well English spoken have

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

(16) a. Er muß Englisch gelernt haben . . . only EMVHe must English learned have‘He must have learned English’

b.**Er tut Englisch gelernt haben müssenHe does English learned have must

Werner Abraham

c. **Er tut Bulgarisch sprechen müssen, weil ich immer nur da verstehehe does Bulgarian speak must. because I always only da under-stand

(17) *Er kann gut Englisch sprechen tunHe can well Modern English speak do

(18) *Englisch wird gut sprechen getanEnglish becomes well speak done

These facts reconfirm the full verb status of MVs in the syntax of substan-dard German, in which discourse factors play a more prominent part than instandard German (cf. Abraham & Fischer 1998).

. Passivisation

The general conclusion about the full verb status of MVs reached above is fur-ther supported by passivisation facts. As (19) shows, DMV uses are grammat-ical 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 areseverely restricted in other Germanic languages. Dutch, however, is totally ona par with German (cf. Barbiers 1995).

(19) DMV =

a. Hier wird/wurde alles versuchtHere becomes/became all attempted‘Here everything is/was attempted’

b. Hier wird/wurde alles gekonnt/gewollt/gedurftHere becomes/became all could/would/might‘Here everything is/was possible/wanted/allowed’

c. Auch in der Schweiz wird/wurde Deutsch gekonntAlso 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 offenzu legen)Even in Switzerland became this must-part (i.e. the bank accountsopen to lay)‘Even in Switzerland this was required (i.e. to open up the bank ac-counts)’


Summarizing, we may say that MVs in German and in Dutch, but not in En-glish, behave like transitive verbs in all relevant syntactic respects. Our assump-tion has been so far that this is due to the fact that (American) English modalsare on their way to losing their root meanings and retain only epistemic andfinite uses. It remains to be seen, however, whether there are semantic corre-lates of this syntactic typological distinction. In the next section, I investigatethe relation between the aspectual semantics of modal verbs and their syntacticstructure.

. The event structure of MV: Terminativity

Since my discussion aims at providing a link between aspect and syntactic dis-tinctions, it is necessary to present briefly, but succinctly, a formal accountof aspectual structure. We will be concerned with the distinction of perfec-tivity (also called resultativity, terminativity, telicity7) vs. imperfectivity (ordurativity), 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 con-ceive of as distinguishing between resultativity and ongoing situations (statalor dynamic). This classification will be shown to carry over to the syntac-tic distinction between DMV and EMV (see for an early presentation Abra-ham 1990). The vocabulary adopted here accords with Reichenbach’s tense andevent logic. Resultative events have a bi-phasic structure as in (20a), whereasimperfectives 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 com-ponents (or phases) E1 and E2. The event structure of terminative verbs is bi-phasic: 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 bothevent phases, E1 and E2, simultaneously. On the other hand, the event structureof 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-axialstructure 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, statalas well as dynamic (mono-phasicness as genus proprium). (21) and (22) showhow the verbal arguments are to be mapped on the event phases. The followingabbreviations are used: tV = transitive verb; eV = ‘ergative’ verb (terminative-intransitive); iV = intransitive verb (non-terminative); EvP = event passive; StP= statal (adjectival) passive; PPP = perfect participle passive; PPA = perfect par-ticiple active; iA = internal argument (direct object); eA = external argument(subject).

(21) a. terminative transitive verb, tV: viz. einschläfern ‘make sleep’

t1 tntm

eA: eA(t1) ∩ . . . ∩ eA(tm-1)iA: iA(t1) ∩ . . . ∩ iA(tm ∩ . . . ∩ iA(tn)

b. terminative intransitive verb, iV: viz. einschlafen ‘fall asleep’

t1 tntm

iA: iA(t1) ∩ . . . ∩ iA(tm) . . . ∩ iA(tn)

(22) non-terminative (i)tV: viz. schlafen ‘sleep’

t1 tntm

eA: eA(t1)∩. . .∩eA(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 termina-tives with respect to their event structure. Their mono-phasic character ac-cords well with the idea that EMVs as well as the epistemic readings of drohen‘threaten’ and versprechen ‘promise’ are semantically bleached raising verbs (asin the comparable English construction The heat promises to break, which hasa futural meaning; see Palmer 1986; Bybee 1985; Coates 1995). In their con-trol readings, on the other hand, verbs like drohen and versprechen (+ com-plement verbs) are bi-phasic in the sense of (21a, b). The difference betweenthe mono-phasic and bi-phasic readings is represented in (23a) vs. (23b). Notethe 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)


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

Recall that an epistemic reading is not available for embedded, i.e. non-finiteterminative predicates; see (5b). This is accounted for by (22) and (23a) in thatthe MV under epistemic interpretation, for lack of event-structural congruency(epistemic modals have a monophasic event structure, while root modals havea biphasic structure), cannot project any event structure of its own onto thestructure of the embedded perfective.8 All an EMV does in a verbal complex isprovide person and number agreement with the subject, taking over fully thetheta grid of the embedded predicate. In this EMVs are similar to auxiliaries(raising predicates). As a consequence, the wide scope of an E-modal (raisingmodal) cannot ‘overwrite’ the event structure of, or de-terminativise, a termi-native verb. This is the general yield of the matching mechanism between MVand the aktionsart feature [α term] of an embedded full predicate. The DMV-reading, on the other hand, provides an biphasic event identification of its ownas 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 terminativefull verbs (as in (21)) and DMVs (as in (23b)). Both lexical terminatives andmodals in their DMV-readings are bi-phasic. As represented in (23b), eventidentification 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 untilthe inception of phase 2 (tm – tn). As in the case of lexical terminative verbs, themedial temporal reference point, tm, is a border point satisfying both eventcharacteristics, i.e. that of the DMV and that of the lexical predicate. This boilsdown to saying that the complete event characteristic under the DMV-readingsatisfies the charaterisation of bi-phasics, by virtue of the special nature of theborder point (tm). In this way, the intensional link between approach phase andresult phase for terminative predicates is established. Another parallel consistsin the fact that the event references of DMVs as well as terminative transitiveand intransitive verbs are satisfied in the first phase and also in the border pointof the time axis.

There is therefore a clear convergence of terminativity with the DMV-reading, which is reflected by similar event structures. The data in (6) showbeyond doubt that the non-terminative EMV is incompatible with an embed-ded terminative full verb. (24a, b) summarizes the links between syntactic be-

Werner Abraham

haviour (distribution as well as scope) and aspectual classification (Abraham1995, Ch. 6; Abraham 2001a).

(24) a. the root MV, under a terminative reading, projects narrow, verb-internal, scope.

b. the epistemic MV, under a non-terminative reading, projects wideclausal scope, which has no lexical effect and therefore leaves thelexical event characteristic (including theta grid) of the embed-ded predicate unaffected; in other words, the epistemic MV has node-terminativizing effect on a terminative verb – there is no local c-command relation between the high functional epistemic node andthe theta-role bearing lexical aspect node inside VP.9

Having established a link between the structural positions of DMVs vs. EMVsand their aspectually distinct meanings, let us now turn to the semantic pro-jections of their arguments. The main question to be asked is: do DMVs andEMVs have different characteristics in terms of semantic roles? Note that, if thisturns out to be the case, this would yield a further correlation between deonticor epistemic reference, on the one hand, and distinct event identification, onthe 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 assigna semantic role to their subject. This is a fundamental distinction, which isin principle open to empirical verification in other languages. Note that, inorder to make this distinction more meaningful, we would like to relate it tothe distributional distinctions seen to hold for German DMVs and EMVs.

The assumption that a DMV assigns a thematic role to its subject raises aspecific 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 twotheta-roles, e.g. Agent and Patient, to a single argument. How is this problemto be solved?

Vikner’s (1988) suggestion (following Zubizarreta 1982) is that, in the caseof verbal clusters, the subject may have, in addition to its main and strongthematic role assigned by the full lexical verb, one, but not more than one, extra


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

(25) a. Hun vil blive arresteret . . . *DMV, EMVhe will become arrested‘He will be arrested.’

b. Hun vil arresteres . . . DMV, *EMVhe will arrested-

‘He wants to be arrested.’

Since Vikner assumes that the (auxiliary) uses of Danish blive, få and kommeassign 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 thirdfrom arresteret), which is impossible, irrespective of whatever assumption ismade with respect to the assignment of semantic roles. A root reading of (25a)is therefore ungrammatical, unlike the epistemic reading of (25a), in which themodal does not assign a semantic role of its own. (25a) will receive an epistemicinterpretation if the auxiliary vil does not assign a third semantic role to thesubject, hun. In the same vein, the subject hun in (25b) receives two semanticroles, but these are different from those in (25a): one semantic role is assignedby the participle of the main verb, arresteres, under lexical government, and thesecond (weaker) one is assigned by vil.

There are, as far as I can see, two other ways of avoiding this violation ofthe theta-criterion (‘Each thematic role must be assigned to one and only oneargument, and each argument bears one and only one theta-role’) when wehave embeddings under DMVs vs. EMVs. One would be to block double the-matic assignment (by DMV as well as the embedded main verb) unless the twosemantic roles are identical. The second, more or less canonical, path to followis to assume two different mechanisms that are responsible for the clusteringof modal verbs with their embedded lexical verbs: raising for EMVs (wherethe problem of double assignment does not arise, since an EMV does not as-sign 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 spelledout in (26a, b) for drohen ‘threaten’ and versprechen ‘promise’ as well as theroot 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 lexicalmeaning, they assign an -role to the subject. The same holds for themodal verbs in (26b) in the deontic readings given.

In (27a, b), no root reading is possible. Instead, these two sentences receiveepistemic readings, with the subject assigned the or roleby 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 him-self.’

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 left-hand column of (28). The columns DMV and EMV represent the equivalentsin (American) English. Note that American English uses modals to render thedeontic 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 ofmodals left. It is likely that such occurrences, which are unsupported by fullparadigmatic status, are doomed to eventually vanish altogether.

(28) S

M G A E

DMV DMV EMVa. wollen [+hum] wish be supposed tob. mögen11 (+hum) be permitted to be supposed toc. müssen ⁄ must be supposed tod. sollen ⁄ (shall) be supposed toe. dürfen12 ⁄ be to may

be permittedf. können ⁄ be able can (alethic)


According to (28a, b), wollen and mögen are not on a par with the other MVssince, in their deontic uses, they project agentive subjects. This is supported bythe finding that non-volitional MVs are demonstrably raising verbs also in theirnon-epistemic readings (Reis 2000:5–6). Wurmbrand (1998, 1999) has arguedthat, in all their uses, irrespective of their root vs. epistemic function, modalverbs are raising verbs. Although different conceptually, Wurmbrand’s view iscomparable to those of IJbema (1997) and IJbema & Abraham (2000), whoargue that what are called ‘control’ complements do not require the concept ofPRO. Rather, the infinitival preposition (German zu, Modern English to) andthe infinitival suffix (German -(e)n) (as much as the circumfix of the passiveparticiple) occupy the Spec of vP and Spec of VP respectively, which wouldblock the embedded subject from surfacing in those Spec-positions.

My claim that root readings, but not epistemic readings, go hand in handwith 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 thebearer of the obligation expressed by must.13 Consequently, the deontic mean-ing cannot be linked to any specific theta role. Note, however, that, even underthis 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. Ihave argued elsewhere (Abraham 1995:480ff), following Zubizarreta (1982)and Vikner (1988), that root modals of this type are causatives involving covertagents responsible for the obligation that is part of the meaning of deontic/. In a way, this assumption concerning the covert causative char-acter 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 argumentstructure and theta-grid totally different from those of unaccusatives.

Whatever the solution to the problem of theta-assignment and DMVs maybe, it is clear that EMVs do not assign a thematic role to their subjects.14 Howdoes the difference between DMVs and EMVs in terms of semantic roles andarguments 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 developmentof modals in the history of English and the distinction between root/deonticand epistemic modals in German (see Sections 4 and 5.1–4 above). I am as-suming here that verbs that have been reanalysed from lexical verbs to modalsbecome stative as a consequence of the loss of certain components in their ex-tended projection, that is ModP for modals and vP/Aktionsart – where re-sultative (perfective) Aktionsart is represented as a small clause predicate – forquasi-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) in-terpretation (Roberts 2000:5), as opposed to eventive verbs, which may involvethematic agentivity.

Note that the modal verbs in (28) are parallel to the first class of ‘main verbmodals’ in van Kemenade’s (1987) classification of Old English modals, whichhave subject theta-roles in their own right and, consequently, correspond tocontrol verbs, as opposed to those modals which have no theta-role for thesubject and therefore pattern like raising verbs (such as seem). This is nothingnew, of course (see, for example, Abraham 1990). Van Kemenade’s ‘main verbmodals’ 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 theyhave to raise to AgrS0. DMVs, on the other hand, also have non-finite formsand can therefore be embedded under EMVs. This yields the following picturewith respect to clustering of modal verbs.

(31) N- EMVs

a. *EMV embedded under DMV/EMVb. DMV embedded under EMVc. DMV embedded under DMV: not excluded on theoretical grounds

(but may nevertheless be blocked on the stylistic grounds of horroraequi)

The theoretical option in (31c) appears to be supported empirically in lan-guages that have ‘aspectual’ DMVs such as German and Dutch, but not inlanguages 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), ModernEnglish does not. This is due to the fact that Modern English has lost its DMVuses, retaining only EMV and purely temporal uses. I have argued that thisdevelopment is a consequence of the loss of the aspectual properties of English


MVs. This position, reached on empirical-comparative grounds, is in contrastto Lightfoot’s thesis that the emergence of the modern MVs in English is solelyrooted in syntax (Lightfoot 1979).

. On the interdependence of the thematic properties of MV and theAktionsart nature of the embedded main verb

From their inherent resultative character it follows that terminative transitiveverbs (tVs) are agentive. Terminative intransitives (iVs), on the other hand, aretypically non-agentives, in that their distributional properties identify themas passive-like, or ‘unaccusative’ (for the class of these phenomena, both lexicaland 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 op-position based on the feature [+/– intention of the subject referent] is reflectedin the properties of the two terminative classes of main verbs. Wollen ‘will’ is amember 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 assignedto 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 DMVshe must/shall him annoy/entertain‘She must/has to annoy/entertain him.’Θ1 = ⁄ (x) = (x) = (x)Θ2 = ⁄ (x) = (x) orΘ2 = ⁄ (x)

However, the assignment of ⁄ to the embedded subject inan epistemic reading is impossible; see (32b).

(32) b. Siei muß/soll i ihn ärgern/unterhalten EMV‘She must be annoying/entertaining him.’Θ2 = (x)

(33) a. Siei will i ihn ärgern/unterhalten DMV, ?EMVshe will him irritate/entertain‘She wants to annoy/entertain him.’

Werner Abraham

Θ1 = (x)= (x)Θ2 = ⁄ (x) orΘ2 = (x)

b. Er will ein guter Syntaktiker sein DMV/EMVhe will a good syntactician beDMV-reading: ‘He wants to be a good syntactician.’EMV-reading: ‘He pretends/imagines to be/being a good syntacti-cian.’

c. Er will den Marathon (gut) laufen DMV/EMVhe will the marathon well runDMV-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 elsethe reference points of the speech act and the event coincide. Then, and onlythen, is an EMV reading the correct interpretation. Let us try to generalise thisobservation further. See (34b) with alethic schreiben können ‘be able to write’,which supports this assumption. Furthermore, (34b) shows that only undersome property reading do we arrive at an acceptable interpretation. This alsoholds for agentive MV wollen. However, in the case of the non-agentive MVsmüssen ‘must’ and sollen ‘shall’, no root reading is possible, since it would be in-compatible with predicates expressing a property reading. Thus, only epistemicinterpretations remain. Compare (34a–c) with (33b–c).

(34) a. Er muß/soll ein (guter) Syntaktiker sein ??DMV, EMVhe must/shall a good syntactician be

b. Er muß/soll gute Aufsätze schreiben können DMV, EMVhe must/shall good essays write can

c. Er muß/soll den Marathon (gut) laufen können ??DMV, EMVhe must/shall the marathon well run can

(34b) is ambiguous. Under the DMV-reading, the reference of the embeddedinfinitival may be to an event in the future, whereas under the EMV-reading thereference 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 adeontic reading.

These distributional facts allow for a number of conclusions, which will betaken up in what follows.


. Comparative syntax

If non-finite embedding is tied to theta-bearing subjects, an interesting conse-quence follows. Consider the restrictions on the embedding of MV in (31a–c),repeated in (35) below.

(35) N- EMVs

a. *EMV embedded under DMV/EMVb. DMV embedded under EMVc. 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 supportedempirically in languages that have ‘aspectual’ or Aktionsart-supported DMVs,as in German and Dutch, but not in Modern English. It is usually assumedthat the restriction in Modern English is due to the auxiliarisation of ModernEnglish MVs. For German, Diewald (1999) claims that the deficiency of theinflectional paradigm of MVs has also contributed to a certain degree of aux-iliarisation of the modals and to the restrictions on epistemic readings in par-ticular. This argument appears to beg the question, however: what we wouldrather wish to say is that, in purely syntactic terms, the gradually emerging fail-ure to license non-finite MV-forms is caused by the loss of AspP as well as thefact that MV raise to a higher position than VP (Roberts 2000:3). Either con-dition can be reduced to the failure to assign a subject theta role, which holdsfor EMVs in German and Modern English, where the modal undergoes oblig-atory raising to at least TP/AgrP. It stands to reason that, if syntactic licensingis not possible, theta saturation of the subject as a lexical requirement needsto be provided. Such lexical saturation is not required for Modern English rootmodals given that VP-licensing of the external theta role is unnecessary. In Ger-man, as I argued, the lexical Aktionsart property saturates exactly this lexicaltheta criterion.

Let us now return to (1) on page 24, my first attempt at the syntactic rep-resentation of MVs in Modern English vs. German and DMVs vs. EMVs, re-spectively. From what has emerged so far in terms of semantic distinctionsand surface distributions, the representation in (1) cannot be correct. Con-sider Modern German (and Dutch) first. One way to give descriptive credit tothe difference between DMV and EMV on the basis of their distinct behaviourunder aspectual or Aktionsart conditions is that shown in (36) vs. (37).

Werner Abraham

(36) EMV German mag ‘may’:

IP

Spec I’

I vP

v’

v

Spec

VP

Spec V’

V

Er magi ti dich auslachent

Since aspectual terminativity, or Aktionsart, has turned out to be involved forDMVs, 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’

I PerfP

Spec PerfP’

Perf vP

Spec v’

v VP

Spec V’

V’

V

Ich magi t ti EC dich nicht auslachen


According to this structure, the root MV starts in the head of the PerfP andmoves to I. This treatment of deontic modals has already been referred toin Abraham (1990) and is also widely supported for modals in other lan-guages (see e.g. Picallo 1985 and, more recently, Barbiers 1995:184–203 aswell as Wurmbrand 1999). This accounts for the complementary distributionof EMVs and DMVs but it introduces an extra phrasal category to make thedeontic modal into an aspect marker.

Another account – and this would bring their syntax even closer to their se-mantics – can be based on the event-semantic status of perfectives as boundedverb meanings, i.e. their resultative readings. The idea would be that the syntaxof DMVs, in contrast to EMVs, includes a small clause (see also Abraham 1990,1995). But other syntactic characteristics do not appear to be reconcilable withthis position. If it is true, as is commonly assumed, that epistemics are raisingverbs, while deontics are controllers (PRO-syntax), it needs to be investigatedhow a PRO-syntax translates into a small clause syntax. Many linguists (amongwhom Roberts 2000) have remained non-committal on the exact syntactic sta-tus of epistemics as opposed to deontics. Wurmbrand (1999) opts for raisingstructures for all MVs except for will/wollen, which she takes to have full verbstatus (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 notmake any such distinction in a systematic, paradigm-supported way.

. Conclusion

Our empirical findings were as follows. First, we established clear syntactic dif-ferences between the modals in German and English, the main distinction be-ing that modals in German, but not in English, have full verb status and as suchoccur also non-finitely (both as infinitives and as participles) and in non-finitedependent position. The second and most important result of this investigationis that Aktionsart conditions triggering the split between DMVs and EMVs inModern German do not to the same extent determine the behaviour of MVs inModern English. English modals do not obey the same distributional criteriasimply because they have no syntactic status as non-finites. Since German epis-temic modal readings are sensitive to features of aspect/Aktionsart, there mustbe 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, therewould 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 assign-ment, and raising

Such a view has not yet been proposed in work on the topic by scholars likeLightfoot, Palmer, Warner, Roberts, Giorgi & Pianesi, to name but a few. It hasbeen shown elsewhere that true aspectual behaviour is relatively recent in Mod-ern English (van Gelderen 1999). If we accept this, it cannot have had an impacton the possibilities of embedding under modal verbs. Since the older stages ofEnglish were on a par with German, a radical restructuring must have takenplace (Abraham 2001b). I have tentatively proposed (but not really shown here;see Abraham 2001b) that the trigger for the far-reaching auxiliarisation of MVsin 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 andparadigmatic changes. I have suggested that the tendency in American Englishof replacing DMVs by other expressions, as well as the fact that will and shallserve no deontic or epistemic purposes any longer, is merely the final stage inthis 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 considerthe relevance of corpus data.

. I have disregarded can since it is often alethic and thus does not provide the same rangeof 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 docu-mented by Görlach (1991).

. While integrating MVs into the overall syntax of verbs, German (and Dutch), modalsin both languages constitute a clearly identifiable class. Not only does the 3rd person sin-gular present tense lack the canonical ending in -t , but their singular and plural presenttense forms also reflect the original 2nd and 3rd preterit stems of the original preteritpresents. 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


Middle High German /a/-/u/ distinction between 2nd and 3rd verbal root vowels in the 3rdclass 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 Aspect Aktionsart

Projection level of diathesis in question Above Vo, within VP Inside Vo – word syntaxRepeated diathesis marking possible + (as in Russian) – (as in German)Analytic morphology (in terms of + +/–clause-structure and word-derivation)Perfectives are resultatives + (cf. Russian) + (cf. German)Perfectives provide present tense gap + (cf. Russian) – (cf. German)

The chart shows both distinct and congruent features. I shall return to differences in the finalsection 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 distributionaland behavioural properties as provided in this article, would make us liable to methodolog-ical pitfalls such as the notorious discovery procedures. I will sidestep these issues here. For asummary 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 thetraditional distinction between clausal aspect and lexical Aktionsart, both in the Slavic andthe German literature on this subject (see, e.g., Tschirner 1991 for a recent discussion of thedistinction between aspect and Aktionsart). Finer distinctions between these terms, whichhave been made in the literature, are of no importance here. In support of this view, seeZhang (1995:114ff) for an identical position with respect to Chinese, in which aspect playsa prominent role.

. For further identification of events and the projection of arguments for duratives, seeAbraham (1990). I restrict my discussion to root modals, i.e. DMV.

. This accords well with Cinque’s typological findings concerning the structurally highposition of epistemic adverbs; see Cinque (1999).

. This difference in derivation has been rejected in Wurmbrand (1999) and Barbiers (thisvolume); 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 ModernEnglish, as opposed to all other Germanic languages.

. Dutch equivalents of (29)–(30) were provided by one reviewer, who raised the objectionthat I am trying to rebut.

Werner Abraham

. In Abraham (2001a) it is argued that, at least in German, the epistemic readings stillfaintly echo the corresponding root meanings: none of the six modals in (i) project identicalepistemic 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 beaccommodated given the structural assumptions made in the present article. This positionhas 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 animalsleft in Limburg’.

Note, however, that my argument about the non-substitutability of MVs in these cases isreflected 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 isnot. Furthermore, the assumption that EMVs have scope over the whole sentence must notbe taken to be an explanation, but as a formalised instantiation of the meaning of the modalin a clausal context. It should be clear that I do not oppose the view that epistemic modalsoccupy 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 desVIII. Internationalen Germanisten-Kongresses. Sektion Kontrastive Syntax (pp. 3–12).München: Iudicium.

Abraham, W. (1992). The emergence of the periphrastic passive in Gothic. LeuvenseBijdragen, 81(1-3), 1–15.

Abraham, W. (1995). Syntax des Deutschen im Sprachenvergleich: Grundlegung einertypologischen Syntax des Deutschen. Tübingen: Gunter Narr.


Abraham, W. (1998). The aspectual source of the epistemic-root distinction of modal verbsin German. In W. Boeder, Chr. Schröder, K.H. Wagner and W. Wildgen (Eds.), Sprachein Raum und Zeit. In memoriam Johannes Bechert (pp. 231–250). Tübingen: G. Narr.

Abraham, W. (2000). The morphological and semantic classification of ‘evidentials’ andmodal 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. Müllerand 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 tunals Hilfsverb. In K. Donhauser and L. Eichinger (Eds.), Deutsche Grammatik – Themaund 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 ofLeiden.

Belletti, A. and L. Rizzi (1988). Psych-verbs and theta theory. Natural Language andLinguistic Theory, 6, 291–352.

Brinton, L. and M. Akimoto (1999). Collocational and Idiomatic Aspects of CompositePredicates 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. Ams-terdam: 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. InJ. 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 Polyfunk-tionalität. [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 340Stuttgart-Tübingen, Bericht Nr. 27, 29–51.

Gelderen, E. van (1999). Aspect, modals, and infinitival endings in Germanic. Ms., ArizonaState 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 inTheorie 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 UniversityPress.

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 ofTübingen, 16–17 March 2000.

Roberts, I. (2000). The history of the modals yet again. Paper read at the Modality workshopat the University of Tübingen, 16–17 March, 2000.

Rohrbacher, B.W. (1999). Morphology-driven Syntax: A Theory of V to I Raising and Pro-Drop. Amsterdam: John Benjamins.

Sprouse, R. and J. Durbin (1998). The syntactic category of the preterit-present modal verbsin Modern Standard German. Paper read at the Berkeley Germanic Linguistics RoundTable, 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 disser-

tation, MIT.

Modality and polarity*

Sjef BarbiersRoyal Netherlands Academy of Sciences, Meertens Instituut, Amsterdam

In this paper it is argued that the possible interpretations of a modal verbprimarily depend on properties of its complement. On the basis ofnon-verbal complementation of modals in Dutch it is shown that modalsalways require a complement that denotes a bounded scale. The nature of thisscale determines whether the modal gets a root or an epistemicinterpretation.

. Introduction

This paper investigates the relation between the interpretations of modal verbsand the nature of their complements, starting out with a cross-linguisticallyrare 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 asilent infinitive (Section 2; contra Geerts et al. 1984; Vanden Wyngaerd1994).

ii. The complement of a modal must denote a value on a bounded scale, re-gardless of the interpretation of the modal. If the complement does notdenote such a value, the sentence is ungrammatical (Section 2).

iii. The data in (1) are problematical for analyses according to which root in-terpretations always involve control structures while epistemic interpreta-

Sjef Barbiers

tions always involve subject raising structures (e.g. Hofmann 1966; Ross1969; Perlmutter 1970).2 While the modals in (1) must be analysed as rais-ing verbs, they can only have root interpretations, not epistemic interpre-tations.

iv. In root interpretations, a modal can be subject-oriented (dyadic) or non-subject-oriented (monadic). A modal is always monadic in epistemic in-terpretations (Section 3).

v. Modals with a root interpretation always involve a potential change of thevalue on the scale introduced by the modal’s complement, i.e. a poten-tial polarity transition. When such a change is impossible, e.g. when thecomplement of a modal denotes a invariable property of the subject, anepistemic 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 poten-tial change of value on a bounded scale. This yields four classes of modalinterpretations (Section 5).

vii. The syntactic complement of the modal determines its interpretations.There are two types of complements that cannot be interpreted as involvinga potential change of value. In the case of a definite nominal complement,this gives rise to sympathy/antipathy interpretations. In the case of a verbalcomplement, this yields an epistemic interpretation. These interpretationsare possible only if negation or (abstract) affirmation is present to providethe required value on a bounded scale (Section 6).

viii. Epistemic interpretations are non-subject-oriented. This follows from theassumption that the semantic relation between a verb and the subject is es-tablished by an (abstract) verbal head with D-features. In the epistemic in-terpretation, the projection of this verbal head including the subject is thesingle argument of the modal, making it impossible to establish a modalrelation 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 structureof modals. There is no need to assume that this polarity is an accidentallexical idiosyncrasy or that polar and non-polar instances of moeten andmogen involve different lexical entries (Section 8).

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


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 anal-ysed as involving an infinitive that is phonologically silent or deleted at PF(Geerts et al. 1984; Vanden Wyngaerd 1994). An advantage of such an anal-ysis is that it allows the crosslinguistic generalisation that modals always take averbal 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 argumentagainst PF-deletion of the infinitive in the case of modals with a non-verbalcomplement is that certain cases would require deletion of a passive comple-ment. This incorrectly predicts that such cases should allow a BY-phrase or anagent-oriented adverb. The crucial observation on which this argument restsis that it is not always sufficient to assume that GO, HAVE, BE or some otherbasic verb is deleted: other, more complex verbal expressions must be assumedto be deletable as well.

(4) a. 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.’

The sentences in (4) with audible GO/BE yield an interpretation that is com-pletely different from the interpretation with GO/BE absent. With GO/BE ab-sent, the interpretation must be as given in the translation, whereas audibleGO forces an interpretation that the lights go out by themselves and the booksgo 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 thefuture. As the translations in (4) indicate, the interpretation of (4a, b) withoutan infinitive is as with an embedded passive. We could assume that (4a, b) isthe 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-phraseor an agent-oriented adverb, since passives allow such an addition. However,BY-phrases and agent-oriented adverbs are impossible when there is no overtverbal complement (6a, b), even though deletion of infinitive and participlenormally 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 wordengedaan.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 am-biguity of modals with a non-verbal complement. Since the alleged deletionof verbal material would take place at PF, this should not have any conse-quences for the interpretation possibilities. However, whereas a modal witha verbal complement can get an epistemic interpretation, a modal with a non-verbal complement can never be interpreted epistemically. This is illustratedin (7).4

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


i. ‘Jan has permission to leave; (someday he will return).’ deonticii. ‘It is allowed that Jan leaves; (someday he will return).’

ind.deonticiii. ‘It may be true that Jan is leaving, someday he will return’

epistemicb. Jan mag weg (, ooit zal hij terugkeren)

Jan may away someday will he returni. ‘Jan has permission to leave; (someday he will return).’ deonticii. ‘It is allowed that Jan leaves; (someday he will return).’

ind.deonticiii. *‘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 thereis 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 isthe following observation. Only constituents that denote a value on a boundedscale can be the non-verbal complement of a modal. The facts in (9) can bestated directly in terms of selectional restrictions imposed by the modal if weassume that there is no silent infinitive. If there were a silent infinitive, wewould be forced to say that the modal imposes selectional restrictions on thecomplement of its complement, a rather unusual state of affairs.5 (Note by theway 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 ascale that is unbounded, as (9f) shows.6,7)

(9) a. De fles moet leeg. Scale: empty-fullthe bottle must empty

b. Het raam kan open.8 Scale: closed-openthe window can open

c. Het licht mag uit. Scale: off-onthe light may off

d. *Het konijn kan ziek.the rabbit can sick

e. *Jan moet intelligent.Jan must intelligent

f. *De speler moet lang.the player must long

I will assume on the basis of these four arguments that Dutch modals can havea non-verbal complement, as long as this complement denotes a value on abounded 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 com-plements, it is possible to show that the semantic epistemic–root distinctioncannot always be expressed in terms of the syntactic distinction between raisingand control verbs.9 An example of this traditional syntactic analysis of modalambiguity (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.’ epistemicJohni must be [ti kind] raising

ii. ‘John has the obligation to be kind.’ rootJohni must be [PROi kind] control

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


Feldman (1986) and Brennan (1993) have already observed that sentencessuch as the ones in (11) clearly can have deontic interpretations but cannot beanalysed 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 andpermission are not attributed to a subject; rather, the obligation/permissionapplies to an entire state of affairs here: the state of affairs ‘the letter be in Am-sterdam before five o’clock’ is obligatory, and the state of affairs ‘the new ma-chine causing problems’ is not permitted. The modals in (11) must thereforebe analysed as monadic predicates.

Modals with non-verbal complements provide an even stronger argumentagainst the raising–control analysis of the epistemic–root distinction. As iswell-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 <naar huis> gaat <*naar huis>that Jan to home goes to home‘that Jan is going home’

b. dat Jan <naar Marie> luistert <naar Marie>that Jan to Marie listens to Marie‘that Jan is listening to Marie’

c. dat Jan <in de tuin> zit te lezen <in de tuin>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 thisrespect:

(13) a. dat Jan morgen <naar Amsterdam> moet <*naar Amsterdam>that Jan tomorrow to Amsterdam must to Amsterdam

b. dat de brief morgen <in de prullenbak> mag <*in de prullenbak>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 thetrashcan’ are small clause complements, then the DPs [Jan] ‘Jan’ and [de brief ]‘the letter’ must be their subjects. Their subjects cannot be PRO, since smallclause complements cannot have PRO as their subject (14a). This in turn im-plies that the subjects in (13) must have raised from a position inside the smallclause, 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 con-structions. If subject raising corresponded to an epistemic interpretation, theprediction would be that sentences like (13a, b) can only have an epistemic in-terpretation. But this prediction is wrong: sentences like (13a, b) cannot havean epistemic interpretation; they can only have root interpretations:

(15) Jan moet naar Amsterdam.

i. ‘Jan has the obligation to go to Amsterdam.’ direct deonticii. ‘It is required that Jan goes to Amsterdam.’ indirect deonticiii. ‘*It must be true that Jan is going to Amsterdam.’ epistemic

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

. Potential polarity transition as a crucial propertyof root interpretations

As discussed in the previous section, modals with a root interpretation can besubject-oriented and non-subject oriented (cf. Feldman 1986; Brennan 1993).Subject-orientation therefore cannot be used as a defining property of rootinterpretations. Non-subject-orientation, however, still counts as one of thedefining properties of the epistemic interpretation. In this section it is shownthat there is another factor playing a role in the distinction between the twotypes of interpretations: the availability of a potential polarity transition.

One of the arguments mentioned in Section 3 in favour of the existence ofconstructions with a modal and a non-verbal complement is that this allowsus to express selectional restrictions on the complement of the modal straight-forwardly: the non-verbal complement of a modal must denote a value on a


bounded scale. If it does not, the sentence is ungrammatical. It is clear whatrole this value on a bounded scale is playing in the semantic interpretation of asentence with a modal. For example, a sentence like (16) may be paraphrasedas: ‘the value that the bottle has on the scale of empty (0) to full (1) is not 1when (16) is uttered and should become 1 at some point in the future.’ No-tice that the sentence does not say anything about the question whether thisvalue will ever be reached. We may say that the interpretation of (16) involvesa 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 propertiesof root interpretations: (i) the complement of the modal must denote a valueon a bounded scale, and (ii) this value is not the actual value at the momentT 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 theavailability 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 bei. ‘It must be true that the earth is a planet.’ epistemicii. #‘It is required that the earth is a planet.’11 root

b. Jan moet een moedertaalspreker van het Fins zijnJan must a native speaker of the Finnish bei. ‘It must be true that Jan is a native speaker of Finnish.’ epistemicii. #‘Jan is required to be a native speaker of Finnish.’ root

c. Jan moet zijn kamer gisteren voor vijf uur hebben opgeruimd.12

Jan must his room yesterday before five hour have cleanedi. ‘It must be true that Jan has cleaned his room before five.’

o’clock yesterday’ epistemicii. #‘Jan is required to have cleaned his room yesterday’ root

Non-verbal complements and infinitival complements of modals thus have incommon that they only allow root interpretations if they denote a variableproperty. For non-verbal complements, this is not yet sufficient, as we haveseen: the potential polarity transition applies to a value on a bounded scaleprovided 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 boundedscale (cf. 9d, f).

Now that it has been established that a potential polarity transition is alsocrucial for the availability of root interpretations with infinitival complements,the question must be asked whether infinitival complements involve a valueon a bounded scale. The null hypothesis clearly is that they do. If we were tohypothesise that they do not, we would in fact claim that the requirement ofa bounded scale in the case of non-verbal complements is completely acciden-tal 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 thecase of infinitival complements. I would like to suggest that it is the embeddedverb 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 end-point (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. Aspectis 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. Sim-ilarly, (18d) means that Ed should be halfway between not understanding andunderstanding what the intention is. The scale involved is simply a numeri-cal scale from 0 to 1, from no event to one event. As I argue in Section 6, thisnumerical scale should not be confused with the scale involved in epistemicinterpretations.

A striking difference between verbal and non-verbal complements ofmodals is that modals do not impose any further selectional restrictions onthe embedded verb: any verb can be the complement of a modal. This can be


explained if we assume that a verb is always dominated by a functional projec-tion, 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 projec-tion, and the functional projection satisfies the selectional restriction imposedby modals, making the choice of verb free.13

As the notion of potential polarity transition may easily give rise to misun-derstandings, 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 in-terpretation. More specifically, stative complements allow root inter-pretations. The crucial property of root interpretations is whether thesituation denoted by the complement can switch from non-existentto existent; whether this situation is dynamic or stative internally isirrelevant.

ii. Individual-level complements only block the root interpretationswhen the subject is a name, not when the reference of the subject mayvary with different situations or worlds (20b–c). This is because theproperty denoted by the predicate within the complement can onlybe a permanent, invariable property relative to a particular referent.

iii. Perfective complements force an epistemic interpretation only whenthey refer to the completion of an event at some point in the past, notwhen they refer to the completion of an event at some point in thefuture (20d).

iv. It is possible to find contexts in which fixed properties such as theone in (17a) can be interpreted as variable properties. This actuallyconfirms our point. In a possible world in which the earth is yet to becreated, the property of being a planet or a star has not yet been fixedand is still variable. Therefore root interpretations are possible (20e).

(20) a. Jan moet een auto hebben. stativeJan must a car havei. ‘It must be true that Jan has a car.’ epistemicii. ‘Jan is required to have a car.’ root

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.’ epistemicii. ‘It is required that the new professor be a native

speaker of Finnish’ rootc. Jan must be a native speaker of Finnish. ind.level

i. ‘It must be true that Jan is a native speaker of F.’ epistemic

Sjef Barbiers

ii. #‘It is required that Jan be a native speaker of Finnish’ rootd. Jan moet zijn kamer voor morgen vijf uur hebben opgeruimd.

perfectiveJan must his room before tomorrow five o’clock have cleanedi. ‘It must be true that Jan will have cleaned his room

tomorrow before five o’clock.’ epistemicii. ‘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.’ epistemicii. ‘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 exampleof each interpretation is given in (21). The most interesting result is that thetwo parameters correctly predict that modal interpretations of type d. shouldexist. This type of interpretation is discussed in Section 8.

Table 1. A new classification of interpretations of modal verbs in Dutch

[+ subject-oriented] [– subject-oriented]

[+ potential polarity transition] a. Dispositional c. Indirect deonticb. Direct deontic

[– potential polarity transition] d. Sympathy/antipathy e. Epistemic

(21) a. Jan mag graag de baas spelen.Jan may eagerly the boss play‘Jan likes to be the boss.’ dispositional

b. Jij mag van mij de kamer uit.you may from me the room out‘I allow you to leave the room.’ direct deontic

c. De kamer moet schoon zijn.the room must clean be‘It is required that the room be clean.’ indirect deontic


d. Jan moet Marie niet.Jan must Marie not‘Jan does not like Marie.’ antipathy

e. Jan moet z’n kamer gisteren hebben opgeruimd.Jan must his room yesterday have cleaned‘Jan must have cleaned his room yesterday.’ epistemic

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

. How the complement determines modal interpretation

Table 1 captures the conclusion drawn in Section 3 that the epistemic–rootdistinction does not depend on the subject-orientation of the modal: a modalcan 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 comple-ment of the modal is the only possible source of ambiguity. There are a numberof differences between complements of epistemic modals and complements ofroot modals:

Table 2. Differences between epistemic and root interpretations in Dutch

Epistemic Root

Potential polarity not required requiredtransition

Scale of complement negative– affirmative 0–1(or: no–yes) (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 influenceof the complement on the interpretation of the modal. ‘Definite complement’in this table means that the complement of an epistemic modal is a definiteverbal argument, not a nominal argument, since it arguably is not.14 The com-plement 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 interpreta-tions 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. Modalscan only have root interpretations here, so the infinitival complement of anepistemic 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 cani. ‘Jan is able to work very hard.’ rootii. #‘It is possible that Jan is working very hard.’ epistemic

(23) a. 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 eati. ‘Jan can eat very much!’ (exclamative) rootii. *‘It is possible that Jan eats very much!’ epistemic

I propose to characterise this difference between infinitival complements ofepistemic and root modals syntactically as in (24). Root modals take an in-definite verbal complement IndP, whereas epistemic modals take a definiteverbal complement DvP. This DvP complement must be specified by Nega-tion/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 [IndONE [VPV]]]

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


I take an (abstract) verbal Dv to establish the semantic relation betweensubject and event, identifying the subject as the source, starting point, agent,possessor of the event (cf. Barbiers 1995). In traditional terms, Dv is responsiblefor assigning an external Theta-role to the subject in its Spec when the verb istransitive or unergative. However, I take it to be present with unaccusative verbsas well. Dv is the equivalent of abstract or little v (Larson 1988; Hale and Keyser1993; 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 thecase of the epistemic interpretation. If it is correct that Dv has to be present inthe complement of an epistemic modal it follows that only verbal complementsallow an epistemic interpretation. Verbs will always be dominated by verbal Dv.When Dv dominates a predicate of category A, P, N, the resulting constituentwill be verbal too (cf. Hale and Keyser 1993). Only when Dv is absent can theresulting complement be non-verbal, and then epistemic interpretations areimpossible.

Maximally generalising the lexical argument structure of modals, I assumethat also in the case of the epistemic interpretation the complement shoulddenote a value on a bounded scale. The presence of Negation/Affirmation inthe 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 scalarvalue, it cannot be the complement of a modal when negation/affirmation areabsent. Epistemic and root interpretations involve different scales, then, be-cause in epistemic interpretations Negation/Affirmation provides the scale andin root interpretations IndP.

The claim that definite constituents without Negation/Affirmation cannotbe the complement of a modal is supported by the behaviour of nominal com-plements of modals. Whereas indefinites and other constituents that providea value on a bounded scale can be the complement of a modal, NPs with adefinite 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 adefinite 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 avalue on a bounded scale. This forces the presence of negation/affirmation.

The absence of a polarity transition both in the epistemic interpretationand in the definite nominal constructions in (26) (cf. Tables 1 and 2) can nowbe 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 plau-sible that the impossibility of a transition is related to the fact that definiteconstituents have a fixed reference, while the possibility of a transition derivesfrom (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), whereasa root modal may or may not be subject-oriented. This follows straightfor-wardly from the structures proposed in (24). If it is correct that Dv establishesthe semantic relation between a subject and a verb, identifying the subject asthe source, starting point, possessor, agent of what is denoted by the verb, itdepends entirely on the base position of the modal whether a semantic relationwill 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 beinterpreted as the source of the modality (dispositional interpretation) or the


possessor of the modality (direct deontic interpretation). In (27b), Dv cannotbe a binary relation between subject and ModP, because ModP dominates DvP.The fact that an epistemic modal is never subject-oriented is thus a direct con-sequence of the assumption that the modal selects DvP (including the subject)in such cases.

Indirect root interpretations, in which the modal is a monadic predicate,may now arise in two ways: (i) DvP is entirely absent and the modal selectsNumP, PP or AP; (ii) DvP is present, but the subject reconstructs at LF intoa position within the scope of the modal, [SpecVP]. I will not discuss thesepossibilities any further here (cf. Barbiers 1995 for discussion).

The argument structure of modals at LF is summarised in (28). The lexicalrelational information associated with modals, whatever its precise content,must fully match the semantic relations determined by syntactic structure. Amodal can be dyadic or monadic. In both cases, one argument must provide avalue on a bounded scale. Class V gives rise to polarity effects and is discussedin the next section.

(28) I. DispositionalRelation: modal (entity, value on bounded scale)Example: mag ([Jan], [IndP schaatsen])

may (Jan, skate)Jan mag graag schaatsen.Jan may eagerly skate‘Jan likes skating.’

II. Direct deonticRelation: modal (entity, value on bounded scale)Example: mag ([Jan], [IndP schaatsen])

may (Jan, skate)Jan mag van mij schaatsen.Jan may of me skate‘I allow Jan to skate.’

III. Indirect deonticRelation: modal (value on bounded scale)Example: 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.’

IV. EpistemicRelation: modal (value on bounded scale)

Sjef Barbiers

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

V. Negative/PositiveRelation: modal (entity, value on a bounded scale)Example: mag ([Marie], [Neg/Aff niet/wel])

may (Marie, not/)Jan mag Marie niet/welJan 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 itis argued that this polarity is not an accidental lexical idiosyncracy, but followsfrom the general requirement that one of the arguments of the modal denotea value on a bounded scale.19 In the unmarked case, the scale is provided bythe complement of the modal, and then moeten and mogen are not polarityitems. A definite nominal complement cannot provide the required scale andtherefore negation or affirmation must be present.

(29) a. 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.’

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


and its ‘external’ argument is established by abstract Dv (cf. Section 6 and7), I propose the structure in (30) for this construction.21 According to stan-dard compositional semantics, the modal in (30) is a binary relation betweenNegation/Affirmation and DP2 Marie.

(30) [DvP [DP1 Jan] [Dv Ø [ModP

Janniet/wel [Mod

not/

mag/moet [DP2 Marie ]]]]]]]may/must Marie

The idea that Negation/Affirmation can be an argument of a verb is supportedby the behavior of epistemic verbs (cf. Barbiers (2000) for a detailed analysisand more arguments). Note that the construction in (31a) cannot be anal-ysed as a case of ellipsis since a full complement with a verbal core is im-possible (31c). Put differently, van niet/wel and full CP-complements are incomplementary 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 bipo-lar with definite nominal complements follows from an independent differencebetween the two modals: moeten ‘must’ is a universal quantifier, whereas mo-gen ‘may’ is an existential quantifier ([not may not] = must, just like [not onenot] = all). It is a general property of universal quantifiers that they cannot bespecified 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 higheston a scale, which is redundant in the case of universal quantifiers, which denotethe highest value on a scale themselves. Existential quantifiers can be specifiedboth 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 hetnot/ 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

i. ‘Jan did not talk with anybody.’ii. ‘Jan did talk with somebody, but not with many people.’

e. 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.’

f. 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 universalquantifiers cannot be specified by an affirmative morpheme.

In sum, the negative polarity of moeten and the bipolarity of mogen arethe result of the interaction between two properties: (i) the universal versusexistential nature of moeten and mogen and (ii) the general property of modalsthat they require an argument providing a value on a bounded scale.

. Conclusion

This investigation of the complements of modals leads to the conclusion thatthe ambiguity of modal verbs is primarily determined by syntactic and seman-tic properties of the complement of the modal.

Notes

* I thank Jack Hoeksema and Virginia Brennan for comments on earlier versions of thispaper. The usual disclaimers apply.

. I discuss Dutch data only. The construction also exists in Norwegian, which has thesame restrictions on the type of complement as Dutch. German and Afrikaans also havethe construction, but with more restrictions: these languages allow only a subset of thecomplements that can occur with a modal in Dutch and Norwegian.

. In the epistemic interpretation the modal qualifies the truth of a proposition. The rootinterpretations 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 notequivalent to the interpretation of the clause with an infinitive.


. For reasons that I do fully not understand, the epistemic interpretation of mogen ‘may’ isonly 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 ascale become grammatical in the comparative:

(i) a. *Deze wijn moet oudthis wine must old

b. Deze wijn moet (nog een jaar) ouderthis wine must one more year older

In view of our generalisation, this must mean that the comparative denotes a value on abounded scale. It is clear that a comparative introduces a scale with a lower bound, namelythe degree of comparision. The upper bound must then be taken to be implicitly given, asin (i. b). The difference with the ungrammatical non-comparative cases is that here evenimplicit specification is impossible.

. As is the case with selectional restrictions in other contexts, type-coercion arises whenthe complement does not satisfy the restriction that it denote a bounded scale. The effectof this is that the hearer will try to interpret as bounded any complement that denotes anunbounded scale in the unmarked case. Provided the proper context, this is possible formany unbounded predicates. Therefore, it is impossible to give a complete list of predicatesthat 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– rootambiguity, 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 inthe past, this becomes a fixed property: once it has been established that John has cleaned hisroom 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 (themorpheme -en) that denotes the value 1. Finite verbs introduce the value 1 as well, if thediagnostic of modification by half is reliable:

(i) Jan lachte halfJohn laughed half

(ii) Jan besefte maar half dat hij gekozen wasJohn realised only half that he chosen was

. For example, infinitival complements of modals may follow the finite verb in embeddedclauses, unlike nominal complements.

. The non-nominal nature of the complement may be determined by the verbal head orby functional elements.

Sjef Barbiers

. Cf. Barbiers (1995) for an elaborated version of this idea within a compositional syntac-tic theory, and for more arguments.

. Cases like Jan kan dat ‘John can that’ do not count as counterexamples because dat isnot 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 com-plements. This may be considered an idiosyncracy given that in many languages modalscannot have a nominal complement. The claim in the main text, however, is restricted to thepolarity of modals with a definite nominal complement. This is not a lexical idiosyncracybut 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 variantswith an affirmative morpheme acceptable. Such a continuation does not help in the negativepolar 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. Sve-

nonius (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 inlinguistics in honor of Sylvain Bromberger (pp. 53–109). Cambridge, Mass.: MIT Press.

Heim, I. and A. Kratzer (1998). Semantics in Generative Grammar. Malden: BlackwellPublishers.

Hoekstra, T. (1984). Transitivity. Grammatical relations in GB theory. Dordrecht: Foris.Hofmann, T. (1966). Past tense replacement and the modal system. Reprinted in J. Mc-

Cawley (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.


Perlmutter, D. (1970). The two verbs begin. In R. Jacobs and P. Rosenbaum (Eds.), Readingsin English transformational grammar (pp. 107–120). Boston: Ginn.

Pollock, J.-Y. (1989). Verb movement, universal grammar and the structure of IP. LinguisticInquiry, 20, 365–424.

Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in PhilosophicalLinguistics, 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 lateMiddle English*

Frits Beukema and Wim van der WurffUniversity of Leiden

In late Middle English, modal verbs occur in a construction with an at thattime already exceptional object-verb order. The object in this type ofconstruction is virtually always a negative constituent. This articleinvestigates the nature of the modal verbs involved and their interaction withnegation. It also proposes an articulated structure for the variousmanifestations of this construction and offers an account of itsdisappearance, focusing on the role played by language acquirers.

. Introduction

Three of the many concerns in the study of modality have been the differencebetween deontic and epistemic modality, the interaction between modality andnegation, and the coherence – or lack thereof – of the class of modal verbs.These areas have been studied from both the synchronic and the diachronicpoint of view. In this paper we take a close look at a specific construction witha modal verb in it as it is found in late Middle English texts, and ask ourselvesthe 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 scopeinteractions 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 dis-

appearance? More broadly speaking, what role did acquisitional factors ofthis construction play in this?


A large part of this paper will be devoted to a discussion of word order in lateMiddle English, a necessary prerequisite for finding answers to the questionsposed above. In doing so, we can draw on a large amount of earlier work, sincein work on the historical development of the syntax of English, word orderstudies have always been prominent (see the review in Chapter 4 of Deni-son 1993). One of the reasons for the popularity of word order studies is nodoubt the pervasive idea that English changed from a basic OV order in OldEnglish (OE) to VO order in later stages of the language. Earlier studies on thisdevelopment have sometimes presented this as a rather abrupt and sweepingchange that took place in early Middle English (ME).1 But, as David Lightfootpoints out,

‘One does not find one set of grammars being replaced by another setovernight; the textual record certainly does not suggest that English speak-ers had object-verb grammars replaced uniformly by verb-object grammarson 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 toclause-initial position, as in relative clauses or in topicalisation structures, themajority of sentences come to display VO order after 1200. However, the OVorder is more tenacious and persistent than is often realised: as late as the mid-dle of the fifteenth century, we still see that most prose texts contain examplesof sentences with OV word order, while in poetry the OV order continues to befound until well into the nineteenth century. The late occurrence of these in-stances of OV order indicates that the OV-VO change is indeed a gradual phe-nomenon stretching over a much longer time span than previously assumed(for a description of the basic data, see Koopman and van der Wurff 2000). Inthis paper we will focus on the final stages of the change from OV to VO inprose, and consider the sentence patterns showing OV order most frequentlyin fifteenth-century prose texts. A first pair of examples is given in (1)–(2).2

(1) I wyll no thyng graunt (Paston Letters 59.15)‘I will admit nothing’

(2) þey schuld no meyhir haue (Capgrave’s Cronicles 199.6)‘They were not allowed to have a mayor’

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


OV order is found and illustrating these with examples. We will show that infifteenth-century ME the OV order becomes restricted to a limited number ofsyntactic 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 4we will undertake a closer examination of the OV patterns without a negativeobject, and try to determine their structure. In Section 5, we consider the OVpattern with a negative object in detail. In particular, we address the questionof the scope interaction between the modal – epistemic and deontic – and thenegative object, which will lead to the conclusion that fifteenth-century Englishmust allow for two possible derivations of this sentence type. Basing ourselveson Kayne (1998), we will provide a sketch of these two structures. In Section 6we consider the fifteenth-century OV facts from an acquisitional perspectiveand show that this can also shed light on the eventual disappearance of theconstruction in the sixteenth century.

. OV order in OE and ME until 1400

Let us begin our investigations with a brief discussion of the occurrence ofOV in English texts written before 1400, providing an analysis of these data inconformity with the restrictive theory of phrase structure in Kayne (1994) andChomsky (1995), as applied to the OE and ME data in Roberts (1997) and vander Wurff (1997a), respectively.

First of all, we note that in embedded clauses in OE the predominant wordorder 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 ofthe 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 would-translate‘that I would translate this book from the Latin language into theEnglish tongue’


As is clear from the data given by Koopman (1995), sentences with OV orderare also found in root clauses, i.e. not all OE root clauses have V2. An exampleis given in (4):

(4) Dryhten ðæt gecyððe (Cura Pastoralis 113.8; Koopman 1995:132)God that showed‘God showed that’

These basic facts of OE word order have received a wide variety of treatmentsin the linguistic literature. Here we follow the analysis of Roberts (1997), whoadopts the Minimalist Program of Chomsky (1995), supplemented with theproposals in Kayne (1994), and suggests that the underlying order of all OEclauses is VO. Case checking of the object takes place in AgrOP, which meansthat the object at some point in the derivation has to move to the specifier ofAgrOP, with the verb moving to AgrO for reasons of equidistance (as expressedin Holmberg’s generalisation). In the sentence in (4), movement of the objectðæt has taken place in the overt syntax, yielding a structure as in (5). In linewith common assumptions, we represent the subject Dryhten as having movedto 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), althoughthere the presence of material intervening between the object and the finiteverb suggests that the object has moved further to a higher position in theclause; for specific proposals in this respect, see Roberts (1997).

In van der Wurff (1997a) an analysis of OV order along the same linesis adopted to account for cases of OV order in fourteenth-century ME. It issuggested there that the lexical verb could still move to AgrO overtly at thatperiod, and it is shown that a large number of sentence patterns can be analysedsuccessfully 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 nehire folies ne hire wyttes (Ayenbite of Inwyt 16.25f)it blinds men so that they themselves not know nor see their misdeeds northeir follies nor their minds‘it [i.e. the sin of pride] blinds men so that they do not know themselvesnor see their misdeeds or follies or know their own minds’

(7) Mykel lufe he schewes, þat never es irk to lufe, bot ay standand, sit-tand, gangand, or wirkand, es ay his lufe thynkand, and oftsyth þarof esdremande (Ego Dormio 3ff)


Much love he shows who never is weary to love but always standing, sit-ting, going or working is always his love thinking, and often thereof isdreaming‘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 leftof the lexical verbs (knawyþ and thynkand). If we take the VO order to be theunderlying order, these OV patterns (and others like them, as discussed in vander Wurff 1997a) make it necessary to have specific mechanisms to account forthe surface position of the object. An analysis in terms of movement of the ob-ject to SpecAgrOP, with concomitant movement of the verb to AgrO, thereforeseems warranted for these fourteenth-century data as well. The relevant part ofthe 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, ifwe 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 willhave great joy inside if we devote ourselves wisely to God’s service anddirect 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 thelevel of Logical Form, i.e. after SPELL-OUT. Thus, the minimalist program cancapture the two orders found in fourteenth century ME (OV and VO) by pos-tulating the presence of an AgrO projection. Since surface OV and VO wordorders are found side by side in a great many ME texts, it is safe to assume thatthe two options could coexist in the grammars of ME speakers. This suggeststhat the object could move to SpecAgrOP overtly as well as covertly, in this wayaccounting for the optionality. An analysis along these lines has been widely ac-cepted to account for object shift phenomena in Modern Dutch, Modern Nor-wegian, Swedish, Danish, Icelandic and other languages. For further details ofthis analysis as it applies to ME, see van der Wurff (1997a). This analysis pre-dicts that OV should be productive in all contexts, and that is exactly what wefind until about 1400: OV order occurs with all types of NP objects, in all types


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 restrictedto a limited number of sentence patterns in a limited number of syntacticcontexts.3 They are presented in (10):

(10) a. auxiliary + negative object + Vb. auxiliary + quantified object + Vc. object (+auxiliary) + V (in relative clause)d. object (+auxiliary) + V (in coordinate clause)e. object + Vinf

f. object + V-ing

g. object + imperative Vh. 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 wissether-in, lete them se it (Paston Letters, 55.6)‘And if my feudal documents, which are in the compartment of my bigchest, 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 ex-cused (Caxton’s Own Prose, 46a.29)‘Requesting everyone who sees, hears, or reads this little work to for-give 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 andeten, summe there-of have aloyned (Paston Letters, 40.28)‘certain members of this company have without good reason taken fromJohn Wylton his cows, his sheep and other cattle, and have salted and eatensome of them, and have taken away others of them’


(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 makefarre fro me (Imitation of Christ, 3.50.26)‘Put a true and stable word into my mouth, and keep a false and wilytongue 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 afterCandlemas’

As example (14) shows, some instances of pattern (10d) feature a negative orquantified object, making them somewhat similar to patterns (10a) and (10b).The same is true for pattern (10c). However, in both (10c) and (10d) the objectdoes not need to be negative or quantified, thus suggesting that these are indeeddistinct patterns. In pattern (10e) the object is usually pronominal, but fulllexical noun phrases also occur (e.g. Paston Letters, 137.4).

To obtain a rough and ready idea of the frequency of these patterns, wemay consider the Paston Letters, a collection of letters and documents writ-ten between 1425 and 1500 by members of three successive generations of thePaston family. In them, each of the OV patterns of (10) is found, except forobjects preceding imperative verbs. The percentages for the eight types are asfollows (there are also a few cases of other OV patterns, especially in the moreformal documents, which we disregard here; hence, the percentages in (19) donot total 100):

(19) a. auxiliary + negative object + V 47%b. auxiliary + quantified object + V 7%c. object (+aux) + V (in relative clause) 1%d. object (+aux) + V (in coordinate clause) 14%e. object + Vinf 7%f. object + V-ing 9%

g. object + imperative 0%h. fixed expressions 1%


Clearly, the type with an auxiliary and a negative object stands out as the mostfrequent OV context. This is true not only for the Paston Letters, but also formost other fifteenth-century prose texts that we have investigated. These in-clude original English works as well as translations, and as far as text type isconcerned, range from beast fables (The History of Reynard the Fox) to religiousprose (Lollard Sermons, Wycliffite texts, The Imitation of Christ); didactic texts(Jacob’s Well and Dives and Pauper); tales of the exotic (Life of Buddha, Man-deville’s Travels) and personal and business letters (Paston Letters, Cely Letters).Altogether, these texts contain over one million words. A first report of the datain 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 Moeren-hout and van der Wurff (2000). It should be emphasised, however, that in allthe texts in this period the prevalent word order is VO, also in all the contextspresented in (10).

Interestingly, the first four patterns in (10) have a close parallel in ModernIcelandic. 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. JónJohn

hefurhas

fáarfew

bækurbooks

lesið.read

‘John has read few books.’c. þetta

thiseris

maðurinnthe-man

semthat

smásögunnathe-short-story

skrifaði.wrote

‘This is the man that wrote the short story.’d. Suma

Somehefurhas

hannhe

særtwounded

ogand

sumasome

drepið.killed

‘He has wounded some and killed others.’

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

. OV patterns without a negative object

Let us now turn to the analysis of the patterns summarised in (10). Beforediscussing the most frequent pattern (10a) in more detail, we will first devotesome attention to the other types (for a fuller discussion, see van der Wurff1999). For type (10b), which features a quantified preverbal object, an analysis


making use of the rule of Quantifier Raising (QR) would seem plausible. Thiswould 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 syn-tax what must in any event happen in Logical Form, where quantifiers haveto move to scope positions’. If this is correct, the same kind of analysis couldbe 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 wissetx . . . ]]]]

The subject my feodaryes has raised to the specifier of the AgrS projection, tocheck its N-feature, while the modal may has raised to AgrS. In the analysisprovided 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 Analternative analysis would be to assume that the quantified object has movedovertly to SpecAgrOP, to check its Case. This, however, would raise the questionas to why it is only quantified objects that move to SpecAgrOP overtly. It seemsmore plausible to account for the empirical facts of fifteenth-century Englishby appealing to the well-established rule of QR, as is done in (22). These factsthus seem to undermine recent attempts in Minimalism to do away with therule of QR altogether (cf. Hornstein 1994, 1995).

For the patterns in (10c–g) we propose a unified analysis. What these caseshave in common is that they have a non-lexical subject that receives the ex-ternal theta-role from the verb assigning the internal theta-role to the pre-verbal object. This makes it possible to say that in the relevant sentences theobject has been topicalised and actually precedes the entire clause, as shownschematically in (23).

(23) Topicalised Objecti . . . null subject . . . V . . . ti

In Rizzi (1997), the left periphery of the clause receives a more articulatedstructure, 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-


tion, the topicalised object would be located in the specifier of this TopP, whichdominates AgrSP, in the specifier of which the invisible subject is located. Thenull subject in (23) may either be the trace of a moved category, which wewould postulate for pattern (10c), or a base-generated element, which seemsplausible for patterns (10d–g). We would therefore suggest that, although thesefive cases look as if they display OV order, they are in fact instantiations of OSVorder. This immediately explains why, in sentences of this type with an auxil-iary, 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 thetype on which we are going to focus. It appears that the precursors of virtuallythe whole range of the present-day English modals and marginal modals (apartfrom 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 soshort 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 aquick remedy against it as you write that they are now saying’

(25) I [. . . ] warnyd yow and Richard that ye shuld nomore stuffe take in-tomyn hows without ye paijid in hand (Paston Letters, 75.16)‘I warned you and Richard that you should not bring more goods into myhouse 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 thismatter’

(28) And also I pray yow þat ye woll do bey ij gode hattis for yowr sonys, for Ican none getyn in þis town (Paston Letters, 138.32–33)‘And I also beg you that you have someone buy two good hats for yoursons, since I cannot get any in this town’


(29) and told here þat 3e had sergyd to a fownd wrytyng þer-of and 3e kwdnon fynd in non wyse (Paston Letters, 128.22–23)‘and told her that you had searched in order to find written evidence ofthis, and could not find any at al’

(30) Item, that as for distreyn for rent or ferm, thow the Dewk had tytill, ashe hath not, he may non ask til the next rent day after his entré, that isMichelmes (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 askfor any until the next rent day following his taking possession, which isMichaelmas’

(31) enfourmyng hym how that your lyffelod hath stond this ij yere in suchtrobill þat ye myght right nought haue of it (Paston Letters, 210.9–10)‘informing him that your property has caused so much trouble these pasttwo 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 Bonewythoute 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 with-out 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 him-selfe (Imitation of Christ, 2.10.15)‘He who has been taught with the gift of grace . . . dare ascribe nothing tohimself ’

The fact that nearly all modals and marginal modals are attested in this con-struction type can be construed as an indication that, as also argued in Warner(1993), in the fifteenth century these elements are becoming increasingly dif-ferentiated from the class of lexical verbs.

We have found only one example in a fifteenth-century text of a negativeobject occurring in a clause with two modals, the object appearing after thefirst modal. The example is given in (36).


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

Notice that double-modal constructions are not uncommon in fifteenth-century texts, but apart from the one in (36), we have not found any with anegative object.

In discussions of the meanings of the modals, a distinction is often madebetween epistemic and deontic meanings. Epistemic modality involves ‘thetruth, probability, possibility etc. of the whole proposition’, whereas deonticmodality involves ‘permission given or obligation imposed performatively bythe speaker/writer (or in a question, the hearer/reader)’ (Denison 1993:293; seealso 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 ex-presses willingness, hence is deontic; the same holds for the past tense wuldein (27); in (28) and (29), can and could both express ability, so they are deon-tic; in (30), may expresses permission, so it is deontic; in (31), might expressesability, so again deontic; in (32), the modal is mon (a loanword from Scan-dinavian) and conveys inability, so it is also deontic; in (33), ought expressesimposition/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 fitinto the epistemic–deontic distinction.

The majority of the modals in these examples, then, are deontic, whichin 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 epistemicmodals in this sentence pattern, as we have seen.

As for the nature of the negative objects: they are usually quite simple, con-sisting of a single word (none or nought/nothing), or two or sometimes threewords (no (+more/other/Adj) + noun). Longer structures are not found. Notethe 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) remain-ing downstairs. A similar effect has been observed by van Riemsdijk (1989) inSouthern German sentences like Ein Buch hat er noch keins (lit. a book has heyet none).6

(37) but money can I non get. (Paston Letters 338.59)‘but I can get no money’


In most examples of the construction, the lexical verb following the negativeobject is dynamic, as in (25)–(28), (30), and (33)–(35). However, stative verbsalso occur, notably the verb have, as in (24), (31), and (32), and see expressingdirect sense perception, as in (36) and (38).7

(38) And whan he coude nowher none see he stopped the hole with sande andmade 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 andmade it level and flat as the other ground’

As is well known, in sentences with a modal and a negative, there is an inter-action of the modal verb with the negative in terms of scope. To give a simplepresent-day English example of such scope interaction, consider the interac-tion 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 isimpossible for this to be true’/‘this cannot be true’), while (40b) represents thereading in which POSSIBLE takes scope over NOT (‘it is possible that this isuntrue’/‘this may not be true’). We will assume here that such scope relationsare reflected in the syntactic structure of sentences (see e.g. May 1985 and muchsubsequent work).8

With this in mind, let us go back to examples (24)–(35). A simple andattractive approach to the relative order of negative object and modal wouldbe to assume that the OV order always signals wide scope of negation overthe modal, and that VO order with a negative object, which is also frequent infifteenth-century texts, signals narrow scope of negation. To show that negationand 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 engumfather her will force her till to marry no-one‘Father will force her not to marry anyone’


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 theembedded clause only. Therefore, the meaning of this sentence can only beparaphrased as: ‘her father will force her to remain unmarried’ and there is noambiguity. The Icelandic sentence in (41) cannot mean: ‘there is no particularperson such that her father will force her to marry him’ (in which the nega-tive 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 impliesthat the sentence means ‘I will force you to remain unmarried’ or no-one mayhave wide scope in the matrix clause, yielding the meaning ‘There is no-onesuch that I will force you to marry her/him’.

Although, as we tentatively put it above, it might seem attractive at firstsight to link the occurrence of narrow scope and wide scope in the fifteenth-century sentences to the occurrence of OV versus VO order, this quicklyturns out to be too simple an approach: inspection of (24)–(35) shows thatthe fifteenth-century OV construction could in principle have both wide andnarrow-scope readings. To show this, we repeat the examples and discuss eachof them in turn.

(24) I suppose he shall no cause have, ne his discyplis nothere, to avante of soshort 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 modalshall is most likely, in that epistemic shall refers to the future, i.e. the sentencecan 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-tomyn hows without ye paijid in hand

This sentence basically has the same scope interaction as (24): deontic shuldtakes scope over nomore stuffe, since the embedded clause must be interpretedas ‘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 wyllsince the meaning conveyed is ‘there is nothing such that I’d be willing togrant it’.


(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 con-stitute an act of negative giving, seems less likely. It follows that the negativeobject 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 Ican none getyn in þis town

The most natural interpretation of this sentence is something like ‘there isn’tany good hat such that I can buy it in this town’. The negative none thereforehas scope over the deontic modal can.

(29) and told here þat 3e had sergyd to a fownd wrytyng þer-of and 3e kwdnon fynd in non wyse

This sentence is largely similar to the previous one, since the negative objectnon 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, ashe hath not, he may non ask til the next rent day after his entré, that isMichelmes

Again, the negative object non has scope over the deontic modal may since themeaning of (30) is clearly not: ‘he has the right not to ask, i.e. to abstain fromasking, rent.’

(31) enfourmyng hym how that your lyffelod hath stond this ij yere in suchtrobill þat ye myght right nought haue of it

The modal might is deontic here since it means ‘were not able’. Since the mean-ing of ye might right nought have cannot be ‘you had the ability to not have anylivelihood’, 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. nonwitnesse) must needs have wide scope, i.e. the sentence means ‘when people areunable to produce a witness’.

(33) and the juges [. . . ] comaunded the scheryf to delyuer the seide Bonewythoute any fyne made, seyng that he out non to make


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 afine (‘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 narrow-scope 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 notused to drinking wine’, which implies that there is no discernible interactionbetween the negative object and the modal. Note that used (to) in Modern En-glish is frequently equivalent to a habitual past tense, and then has no modalmeaning at all, as we have already observed above.

(35) He þat is taught with the yifte of grac [. . . ] dar noþinge as[c]ryve tohimselfe

Dare is equally difficult to classify along the epistemic/deontic parameter. It isobviously subject-oriented (Palmer 1990:112), which suggests categorising itrather as a deontic than an epistemic modal. The scope of noþinge with respectto the (deontic) modal dar cannot but be wide: the sentence only means ‘thereis nothing such that he dare ascribe it to himself ’.

What we can conclude from the survey so far is firstly, that the modalsin (24)–(35) nearly all have a deontic meaning: only one case, i.e. (24), con-tains an epistemic modal. Secondly, among the twelve examples there are onlythree in which the modal clearly has scope over the negative object, while therest shows the negative object having wide scope over the modal. Nevertheless,both possibilities can be clearly evidenced, which implies that the grammarof fifteenth-century English should in principle make both options available,with the actual scope interaction between negation and modality being deter-mined in large measure by the meaning of the modal, as is well documentedfor Modern English.

We now turn to the syntactic representation of these sentences, using theideas in Kayne (1998). We take over the clausal architecture sketched there, butmodify it slightly in that we allow the presence in the same clause of a NegPdominating the modal projection on the one hand, and a NegP dominatingthe lexical verb projection on the other hand. We propose that the high NegPis accessed in wide-scope readings of negation, while the low NegP is activatedin cases of narrow scope. The modal is assumed to head its own VP. The twogeneral structures are given in (43) and (44).


(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 ob-ject must overtly move to SpecNegP, which may or may not be followed bypreposing of the remnant VP.

With this background, let us now consider the structure of sentences witha negative object and a modal in fifteenth-century English, i.e. sentences of thetype 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 couldhas moved out of its VP into AgrS, and there is no preposing of the remnantVP [say ti].

(45) a. He could nothing say (‘he was able to not say anything, i.e. to remainsilent’)

AgrSP

he AgrS’

AgrS TP

VPcouldm

V NegP

tm nothingi Neg’

Neg AgrOP

VP

tisay


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

AgrSP

he AgrS’

AgrS TP

couldmNegP

nothingi Neg’

Neg VP

V

tm

AgrOP

VP

say ti

Note that it is possible to extract part of the negative object from SpecNegP andmove it to a higher position (presumably SpecCP), with concomitant V2 ef-fects, 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 fifteenth-century 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 follow-ing the general guidelines of Kayne (1998). In a sentence like He could say noth-ing, nothing has moved to SpecNegP. This can be the low NegP, which wouldimply that remnant VP-preposing has taken place, moving say and the traceof the negative object to a functional projection immediately dominating thelow NegP. This would result in narrow scope of the negative object vis-à-vis themodal. The structure is given in (48a).


(48) a. He could say nothing (‘He was able to not say anything, i.e. to remainsilent’)

AgrSP

he AgrS’

AgrS TP

couldm VP

V

tm

FP

Spec

[say t ]o vp

NegP

nothingo Neg’

Neg AgrOP

VP

tvp

Here, the modal could and its trace occupy positions higher than the positionof the negative object. However, in the reading in which the negative object haswide 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 casesis that the negative object has moved out of the VP [could say nothing] and thatthe remnant VP [could say t] has been fronted to the specifier of the functionalprojection dominating NegP, with subsequent head movement of the modal toAgrS. This possibility is represented in (48b).


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

AgrSP

he AgrS’

AgrS

couldm

TP

FP

Spec

[t say t ]m o vp NegP

nothingo Neg’

Neg VP

tvp

In both (48a) and (48b), the modal moves to AgrS for reasons of feature check-ing; this has no effect, however, on the scope relations holding between thenegative object and the modal.9

Naturally, fifteenth-century English also features sentences with a negativeobject but no modal, i.e. sentences like the one in (49).

(49) And syth þey louyn no pes (Dives and Pauper 198.22)‘And since they love no peace’

In such sentences the prevalent surface order is VO; hardly any cases with OVorder (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 concomitantmovement of the finite verb louyn to AgrS, possibly by means of remnant VP-movement as an intermediate step. Supporting evidence for a derivation alongthese lines can be found in late Middle English sentences of the type He spokenever 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 Roberts1993). The structure of (49) is represented in (50):


(50) AgrSP

þey AgrS’

AgrS

louynv

TP

Spec T�

T

tv

NegP

no peso Neg’

Neg

tv

AgrOP

Spec AgrO’

AgrO

tv

to VP

tv to

What we have shown in this section is first of all that the pattern displaying amodal followed by a preverbal negative object is fully productive in the fifteenthcentury. Virtually all modals, both epistemic and deontic, occur in this patternwhile the lexical verbs can either be stative or dynamic. We have also shownthat the two types of scope interaction of the modal with the negative objectare instantiated in the data and we have postulated two distinct structures cor-responding to these different scope readings. We think that we have now an-swered the first three questions formulated in the introduction and we nowturn to the fourth question, which involves the acquisition and disappearanceof 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 typeslisted in (10) (see van der Wurff 1997b).


C. In the sixteenth century, preverbal objects of all types disappear (see vander Wurff and Foster 1997).

As discussed in Section 2 of this paper, for stage A it can be assumed that allobjects could in principle move to SpecAgrOP before SPELL OUT for reasonsof case checking. As for the transition from stage A to stage B, it appears thataround 1400, probably for reasons related to the loss of inflectional marking,overt movement of objects to SpecAgrOP was no longer available. As suggestedin van der Wurff (1999), the effect was that surface OV order became impos-sible except in those cases that allowed the object to be reanalysed as occu-pying some other position than SpecAgrOP. This could be SpecNegP, as incases like (10a), or a position adjoined to VP, as in (10b), or the specifier ofa Topic Phrase (TopP), as in cases (10c–g). In other words, language acquirerscould avail themselves of other possibilities licensed by UG when dealing withsentences with OV order.

It discussing the transition from stage B to stage C, we restrict our attentionto the most frequent pattern, the one with a modal and a negative object. In theacquisition of the structure of clauses with a negative object, children may rea-sonably be assumed to take their cues mainly from simple clauses, i.e. He saidnothing, He spoke to no-one etc., rather than from more complex cases contain-ing modals. In (50), we proposed a structure for these cases involving V-to-Imovement. Roberts (1993:246–273) relates the occurrence of V-to-I move-ment to the existence of a separate plural verbal inflection, as in the MiddleEnglish 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 thetreasure of this world’.

Roberts notes that by the end of the fifteenth century, the plural inflectionhad withered away (see also Lass 1992:97f), and that as a consequence V-to-Imovement disappeared.10

In terms of language acquisition, Roberts’ account would imply that at thispoint in time children would no longer hear forms like They louyn or Theyembraceden, but only forms like They love and They embraced. The morpho-logical trigger for the syntactic verb movement rule had disappeared by thattime and children would therefore no longer be exposed to the principal cluefrom which they could deduce that V-to-I was possible in their language. What


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

(52) AgrSP

they AgrS’

AgrS TP

FP

Spec

[love t ]o vp NegP

no peaceo Neg’

Neg VP

tvp

From the consistent occurrence of They love no peace rather than They nopeace love, the child would conclude (perhaps through the mechanism of ‘in-direct negative evidence’, cf. Atkinson 1992; Lasnik 1990) that the operation ofremnant VP-preposing is obligatory.

As for the real-time order of acquisition of negative objects and modalverbs by late Middle English children, the following can be observed. Klimaand Bellugi (1966:197) have pointed out that in Modern English, construc-tions with negative constituents appear in children’s speech only after they havereached the stage of mean utterance length (MLU) 4.0. We assume that in lateMiddle English such constructions were acquired somewhat earlier since thelanguage at that stage lacked the competing negative strategy with not . . . any(see Tieken-Boon van Ostade 1997, who shows that what she and other schol-ars 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-


tively many cases of negative constituents of the type nothing, none, nobody, nomoney 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 languageafter MLU 3.5. There is no reason to suppose that this was any different in lateMiddle English, i.e. Middle English children could freely dispose of sentencescontaining modals and negative constituents at roughly the same time.

What we have sketched as a conceivable Middle English acquisition sce-nario so far is that simple sentences without negative constituents are ac-quired before sentences containing negative constituents, and that sentenceswith modal expressions are acquired in tandem with sentences containing neg-ative constituents. But did late Middle English children at this stage also acquiresentences containing a modal verb as well as a negative object (i.e. He could saynothing)? On the basis of arguments presented above, we contend that they didnot. Recall that the disappearance of the order negative object-main verb (as inHe could nothing say) crucially depends on remnant VP-preposing being oblig-atory. This obligatory nature could only be established by the language learneron the basis of the consistent use of the order main verb-negative object inclauses without modals (i.e. He said nothing). Hence, this type of sentence musthave been acquired before the type containing a negative constituent as well asa modal. Our conclusion therefore is that for fifteenth-century children, theplace of negated constituents like nothing is established on the basis of simplexclauses, i.e. clauses with a single finite verb like He said nothing. Sentences con-taining modals appear to have played no role in this process. The disappearanceof the OV pattern with a modal and a negative object therefore suggests thatthe structure of clauses featuring both negation and modality is not acquired assuch by children; rather, children figure out separately the structure of clauseswith only negation, and the structure of clauses with only a modal. Sentencesdisplaying 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 peacelove is crucial. If our analysis is correct, our account would therefore underpinthe tenet that in language acquisition indirect negative evidence is also takeninto account.

Notes

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


. It should be pointed out here that the OV underlying order of the earlier stages of theEnglish language has not been undisputed: arguments have been put forward that Old En-glish had an underlying VO order (Roberts 1997), or that both OV and VO order weresimultaneously 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 sim-ilar sentences are therefore not examples of the type of structures in which do is an ECMverb, as discussed in Roberts (1993:282–292), an example of which is (i):

(i) thanne he dide the clerk of the council seek it (Paston Letters no. 432)‘Then he made the clerk of the council seek it.’

Rather, the sentence in (15) illustrates the type of structure that Roberts calls FP-do, in whichthe complement clause has a non-overt subject (small pro).

. Our analysis in (22) makes crucial use of the Subject-in-VP hypothesis. The QR-movedobject 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 caseof ‘split topicalisation’, as van Riemsdijk calls it) might cause problems for a theory in whichonly heads of phrases or whole phrases are allowed to move. Van Riemsdijk proposes toaccount for these structures by means of a process he refers to as ‘regeneration’, which en-tails rebuilding the complete projection after movement, i.e. at S-structure. The form of thequantifying 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 theform 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 involveNP-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 noneappear to have moved to preverbal position.

. Jackendoff (1972) appears to adopt different mechanisms for scope assignment bymodals and by quantified DPs. This leaves the claims made in this paper unaffected.

. An anonymous reviewer points out that Hornstein (1995) suggests that scope relationscan also be defined on the positions occupied by traces of scope-bearing elements. We willleave the implications of this possibility aside here.

. Lightfoot (1999:163), basing himself on Warner (1997), suggests that V-to-I movementmay have been operative in English as late as the eighteenth century. However, the preciseanalysis 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 finiteverb under a V-to-I account, Lightfoot has to assume that the sentential negator occurs tothe left of I. However, the surface [Neg V] pattern has widely been taken as a diagnostic ofthe absence of V-to-I movement. See also Vikner (1995) and Holmberg and Platzack (1995).


Primary Sources

Allen, H. (ed., 1931). English Writings of Richard Rolle, Hermit of Hampole. Oxford: OxfordUniversity 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 Univer-

sity 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 Uni-

versity 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 En-

glish 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.


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-containeddeletion. 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: EdinburghUniversity 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 ofEnglish: 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 onRestrictiveness and Learnability (pp. 184–197). Dordrecht: Kluwer.

Lass, R. (1992). Phonology and morphology. In N. Blake (Ed.), The Cambridge Historyof the English Language, volume II 1066–1476 (pp. 23–155). Cambridge: CambridgeUniversity Press.

Lightfoot, D. (1979). Principles of Diachronic Syntax. Cambridge: Cambridge UniversityPress.

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.


Stephany, U. (1986). Modality. In P. Fletcher and M.Garman (Eds.), Language Acquisition:Studies in First Language Development, 2nd ed. (pp. 375–400). Cambridge: CambridgeUniversity 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 ofsubjectification 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 Uni-versity Press.

Warner, A. (1997). The structure of parametric change, and V-movement in the historyof English. In A. van Kemenade and N. Vincent (Eds.), Parameters of MorphosyntacticChange (pp. 380–393). Cambridge: Cambridge University Press.

Wurff, W. van der (1997a). Deriving object-verb order in late Middle English. Journal ofLinguistics, 33, 485–509.

Wurff, W. van der (1997b). OV-volgorde in vijftiende-eeuws Engels proza. In A. van Santenand M. van der Wal (Eds.), Taal in Tijd en Ruimte (pp. 73–84). Leiden: Stichting Neer-landistiek Leiden.

Wurff, W. van der (1999). Objects and verbs in modern Icelandic and fifteenth-centuryEnglish: 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 studyof its frequency and status. In R. Hickey and S. Puppel (Eds.), Language History andLinguistic Modelling, Vol. 1 (pp. 439–453). Berlin: Mouton.

On the use and interpretation of rootinfinitives in early child Dutch*

Elma BlomUniversity of Utrecht

Spontaneous speech data from six Dutch children confirm the claim that rootinfinitives in Dutch are primarily used to express wishes, desires and needs(van Ginneken 1917; Ingram and Thompson 1996; Hoekstra and Hyams1998). Strikingly, the association between modality and root infinitives doesnot appear from early on, but arises as children reach the optional infinitivestage. In this stage (as defined by Wijnen 1997) children begin to use finiteverbs next to their non-finite counterparts: finiteness has become productive.In this article it is shown how children’s ability to use finite verb formsenables them to analyse the infinitive as a verb instead of a noun. For thisreanalysis it is crucial that children recognise the infinitival suffix. Thealternation with finite morphology functions as a cue for this recognition.Reanalysis of the infinitive accounts for the modal meaning of root infinitivesas 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 non-finite utterances to refer to wishes, desires and needs. Recently, the observa-tion that root infinitives (RIs) receive a modal meaning has been confirmedfor Dutch child language (Wijnen 1997). The same generalisation seems tohold for the root infinitives of German children (Behrens 1993; Ingram andThompson 1996; Lasser 1997). Based on these empirical studies, Hoekstra andHyams (1998) formulated the Modal Reference Effect (MRE) for Dutch andGerman child language, which states that ‘with overwhelming frequency, RIshave modal interpretations’. In this paper I will discuss three proposals thataccount for the MRE: Boser, Lust, Santelmann and Whitman’s (1992) NullAuxiliary Hypothesis, Ingram and Thompson’s (1996) input-based hypothe-

Elma Blom

sis and Hoekstra and Hyams’ (1998) morphological marking hypothesis. Themost 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 ac-count seems very promising: different cross-linguistic intuitions and observa-tions 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 empiricalpoint of view. Data from six Dutch children show that the connection betweenmodality and infinitive verb forms does not appear from early on. To connectmodality to the infinitival form, children have to be able to make the finite-non-finite distinction. My claim is that Dutch children need this knowledgeto categorise infinitives as verbs. The early frequent non-modal use of RIs byDutch children and their shift to modal use reflects a categorial re-analysis ofthe 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 sub-jects are preferred (Weverink 1989; Krämer 1993; Haegeman 1995), and stativeverbs hardly ever occur in RIs (Wijnen 1997). The meaning of RIs in Dutch andGerman is often described as ‘modal’ and their temporal reference is usuallyfuture (Behrens 1993; Lasser 1997; Wijnen 1997). Although RIs in Dutch andGerman child and adult language seem highly comparable (at least with respectto 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, RIsare marginally acceptable. Their use is restricted to specific contexts: impera-tives, interrogatives, exclamatives,1 announcements and assertions in narrativecontexts (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!’


d. Even washandje pakkenJust 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 narrativegenre. Examples of assertive child RIs are given in (2):

(2) a. in soel zitten (Abel 1;11.26)in chair sit-

‘I want to sit in the chair’b. Woef liggen (Peter 1;09.20)

Woef lie-

‘Woef [i.e. the dog] lies’

The use of assertive RIs is more widespread in child Dutch than in adultDutch.2 On the other hand, exclamative RIs of the adult-like type in (1c) donot occur in child Dutch.

The modal meaning of RIs in adult Dutch is often related to illocution-ary force: deontic modality (obligation or the necessity to act) in impera-tives, volitive modality3 (wishes) in interrogatives, epistemic modality (prob-ability) in exclamatives and the modality-related category of intentionality inannouncements. Assertions in narrative context do not receive a specific modalor 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 mean-ing of RIs in child Dutch shows some overlap with the adult RIs. Dutch childRIs can most often be paraphrased as ‘I want to INF’ or ‘You must INF’, leadingto respectively a volitive and deontic modal interpretation.4 In contrast to adultRIs, the modal meaning of child RIs is not directly related to interrogative andimperative 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) NullAuxiliary Hypothesis and H&H’s (1998) recent proposal both belong to theformer view, whereas Ingram and Thompson’s (1996) proposal is an exampleof the latter.

. Null Auxiliary Hypothesis

Boser et al.’s Null Auxiliary Hypothesis (1992) states that root infinitives con-tain a covert auxiliary. This auxiliary is not phonetically realised, but its pres-ence is noticed through morphology, syntax and semantics. The covert auxil-iary corresponds to a modal verb or dummy verb like do or go.5 In the formercase the RI will have a modal meaning, in the latter a non-modal meaning. Asthe null auxiliary undergoes V-to-I raising, the infinitive main verb has to stayin its base position, which in Dutch is the final position. Although the NullAuxiliary Hypothesis accounts for the data in a fairly simple way, there are sev-eral empirical problems. I will briefly discuss some of these problems (see fora more extensive discussion H&H). Syntactically, the covert auxiliary behaveslike an overt auxiliary. The constructions are comparable, hence it is expectedthat 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 infinite sentences (Poeppel and Wexler 1993). The same asymmetry can be ob-served for wh-movement. Finite clauses allow wh-movement, whereas non-finite clauses do not. Note that in this respect English child language differsfrom Dutch, German, Swedish and French child language (Phillips 1995 forEnglish; Haegeman 1994 for Dutch; Kursawe 1994 for German; Santelmann1994 for Swedish and Crisma 1992 for French). English child language doesnot show both of the asymmetries whereas the other child languages do. Thefinal problem is concerned with licensing. As an empty category, the null aux-iliary needs licensing. Boser et al. (1992) claim that this is done by an overtsubject. One of the characteristics of RIs, however, is that the presence of alexical subject is optional.

. What they hear is what you get

Ingram and Thompson (1996) claim that children learn modal RIs directlyfrom the input they receive. Dutch and German children associate modal-ity and non-finiteness (and conversely, non-modality and finiteness) becausetheir caregivers very frequently use non-finite verb forms in modal contexts.Ingram and Thompson’s claim is not supported by empirical child-directed


speech data. Moreover, Ingram and Thompson do not explain what is meantby ‘association’. Frequency counts have shown that input can indeed play a rolein learning modal RIs. The connection between modality and non-finiteness inthe 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 therepresentation 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 Germanon the one hand and English on the other. Ud Deen (1997) found that Englishchildren predominantly use RIs in non-modal contexts. H&H claim that thisdifference is caused by a semantic [–realised] feature that is marked by infiniti-val morphology. In languages like Dutch and German the infinitive has a suffix-en, which carries the [–realised] feature. This means that infinitives in theselanguages refer to an event that has not yet been realised. English bare verbforms, on the other hand, do not have any morphology. There is no potentialcarrier, hence they cannot be [–realised]. The bare verb form in English refersnot only to the process part of an event; it also includes the completion of theevent. This is shown by the contrast in (3):

(3) a. *I see John cross the streetb. I saw John cross the street

Following Giorgi and Pianesi (1997), H&H assume that the bare infinitivein English is [+perfective]. In this respect, the infinitive in Dutch/German ismore like the English -ing form, which also refers to the process part of theevent only:

(4) I see John crossing the street

As infinitives in Dutch/German are inherently [–realised], H&H assume thatthe RIs of children learning these languages are [–realised]. By the same ar-gument, i.e. the presence of an inherent feature, the RIs of English speakingchildren are [+perfective].

The modal interpretation of RI in Dutch and German follows from certainproperties of volitive (or ‘bouletic’ as H&H call it) and deontic modality. Theevent expressed in a volitive or deontic modal utterance has a future tempo-ral reference since obligation, permission (i.e. deontic modality) or wishes (i.e.volitive modality) have to precede the event itself. Because infinitives in Dutchand 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 even-tive verbs in RIs (Eventivity Constraint).6 In particular, deontic modality hasa strong preference for eventive predication (while stative verbs yield an epis-temic reading). Children under the age of four do not use epistemic modality,but they do use deontic modality very frequently. The overwhelming numberof eventive RIs in child language neatly follows from this frequent usage.

. Summary

We have seen that the Null Auxiliary Hypothesis is successful in explaining var-ious properties of RIs, such as the position of the verb and verbal morphology.Nevertheless, it runs into serious empirical problems. The strength of Ingramand Thompson’s proposal is that it focuses on an underexposed topic, namelythe input children get. However, the scope of the ‘what they hear is what youget’ hypothesis is too limited to offer a satisfying solution. At first sight, H&H’srecent analysis has many advantages, because it ties non-finiteness, modalityand eventivity together. In the next section their proposal will be discussed inmore detail.

. A closer look at Hoekstra and Hyams (1998)

H&H’s framework is based on a set of empirical observations and theoreticalassumptions. In this section I single out three notions fundamental for H&H’sanalysis. A discussion of these notions brings various problems to light. Therelevant notions are, in order of appearance: the nature of [–realised], mor-phological marking of semantic features, and the connection between modalityand verb type.

. The nature of [–realised]

A crucial notion in H&H’s analysis is the feature [–realised], proposed forDutch and German infinitives. What [–realised] actually refers to is unclear,however, as no apparent definition is provided. It is not a feature which isembedded in a specific theoretical framework; there is no feature matrix towhich [–realised] belongs. Theoretically, [–realised] can be interpreted in threeways: as an (i) aspectual, (ii) tense or (iii) modal feature. I discuss these threepossibilities below.


i. [–realised] as an aspectual featureThe label seems to suggest this interpretation as the most obvious one. An as-pectual reading is reinforced by the fact that [+perfective], which is a validaspectual feature, functions as its opposite. [+perfective] includes the com-pletion of the event while [–realised] excludes the completion. H&H also ex-clude 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 in-terpretation of the English bare V [. . . ]’. Thus, when we try to translate H&H’sobservation that ‘the infinitive denotes that the event is not yet realised’ intoa known aspectual notion, this would result in inchoative aspect as the mostappropriate candidate.

ii. [–realised] as a tense featureThough [–realised] seems to be more aspectual than temporal, H&H do not ex-clude a temporal interpretation. Theoretically, there are two possibilities. When[–realised] refers to the non-completion of an event, it includes present tenseas well as future tense. When it means that the event has not even started, itonly refers to future tense. From the quotation above, it can be derived thatH&H implicitly follow the second strategy: [+perfective] includes here-and-now events, and by implication [–realised] does not.

iii. [–realised] as a modal featureThe feature [–realised] restricts the number of possible modalities. Both deon-tic and volitive modality assume a temporal ordering: the permission, obliga-tion or wish precedes the event for which one gets permission, for which oneenters into an obligation or which is wished for. This has induced some lin-guists to not distinguish between future tense and modality. Chung and Tim-berlake (1985:243) claim: ‘Future is thus not a semantic category where tenseand mood merge. In practice, many languages do not distinguish morpholog-ically between future tense and potential (irrealis) mood. Where a difference ismade, the future tense is used for events that are presumed to be certain to oc-cur, and the irrealis mood for events that are potentially possible but presumedto be certain.’ Example (5a), in which the infinitive refers to a future event, re-ceives a deontic interpretation, while (5b), in which the predicate refers to anongoing event, receives an epistemic interpretation:

(5) a. Hij moet afwassen‘He must do the dishes’

Elma Blom

b. Hij moet aan het afwassen zijnHe 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, oras H&H put it ‘is not yet realised’, the feature [–realised] delimits the modalinterpretation to deontic and volitive modality. Epistemic modality is excluded.

In this section we have seen that [–realised] can be described as an aspec-tual, tense or modal feature. H&H do not define their position and propose afeature that is neither one nor all. [–realised] is a purely empirically motivated7

feature that collapses inchoative aspect, future tense and deontic and volitivemodality.

. Morphological marking

H&H connect [–realised] to an -en suffix, i.e. to morphology. Bare verb formslike the infinitive verb in English, which do not have any morphology, cannotbe [–realised]. H&H follow Giorgi and Pianesi (1997) in assuming that Englishbare verb forms receive a perfective reading. This perfective reading yields anon-modal interpretation of RIs in child English. This gives rise to two ques-tions. First, what morphology marks the [+perfective] feature? Second, as RIsin English cannot express modality, how do English-speaking children expressmodality?

As to the question of what morphology marks [+perfective], H&H fol-low some sort of default strategy: unless morphology indicates otherwise, theinfinitive is [+perfective]. In English, however, the element to could functionas 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 forfuture events in headlines:

(6) a. President to visit Russiab. Unions to go on strikec. McDonald’s to serve beer

Like the examples in (7), the ‘modal’ meaning of (6) may be a relict fromthe historical prepositional meaning of to: ‘direction towards, purpose or goal’(Jespersen 1964).

(7) a. To be or not to beb. To live and let diec. To look at you and never sleep


This brings us to the second question: how do English-speaking children ex-press modality? The to-infinitive seems to possess the necessary propertiesto express modality. However, young English children do not yet use the to-infinitive. Forms that do occur from relatively early on – between the ages oftwo and three – are gonna, wanna, needta and hafta (Bloom, Tackeff and Lahey1984; Gerhardt 1991). These forms express intentionality, volitive and deonticmodality, which are exactly the same interpretations given to the RIs of Dutchchildren. Note that all these kinds of modality or modality-related categoriesare inherently [–realised]. In children’s speech, gonna, wanna, needta and haftahave in common that they all contain a ‘corrupted’ version of to spelled outas [6]. This suggests that in English child language the [6] suffix is like the -ensuffix 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 childrenonly use deontic and volitive modality, i.e. semantic categories which Dutchand German children predominantly express with RIs. This pattern is formu-lated as the Modal Reference Effect (MRE). RIs in these languages are lim-ited to eventive verbs, as summarised in the Eventivity Constraint (EC): ‘RIsare restricted to event-denoting predicates’. H&H show that the MRE and ECare closely connected: eventive verbs ‘force’ a deontic modal reading whereasstative verbs ‘force’ an epistemic reading. See the contrast in (8):

(8) a. Jan moet het antwoord wetenJan 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 readingis most natural (‘as far as I know, John has to know the answer’). In (8b) thesame auxiliary has an eventive predicate. This yields a deontic interpetation(‘John is obliged to read this book’). H&H argue that the eventive characterof RIs follows from the fact that children use deontic modality, not epistemicmodality. At this point, some comments are in order. First, from the perspectiveof 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 notonly 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 antwoordIt can hardly not different or John knows the answer‘Based on what I know, it must be the case that John knows theanswer’

b. Jan moet van de docent voor drie uur het antwoord wetenJohn 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 wetenJohn 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 read-ing 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 thereis also the crucial fact that deontic and volitive modality cannot be collapsed.They differ from each other with respect to two structural properties: (i) voli-tive modality does not have the epistemic counterpart that deontic modalityhas, and (ii) volitive modality does not show a specific preference for eventivepredication nor for stative predicates.

Volitive modality in Dutch is prototypically expressed by the auxiliarywillen (‘want to’). Barbiers (1995) points out that willen can have an epistemicreading, as exemplified by (10):

(10) Er wil hier wel eens een ongeluk gebeurenThere will here - sometimes an accident happen-

‘Accidents (may) happen here’

The sentence in (10), however, does not express epistemic modality. Epis-temic modality involves the qualification of the potential truth of a proposi-tion (Klinge 1998). Barbiers (1995) refers to this as probability. The sentencein (10), however, involves neither probability nor does it give a qualificationof the potential truth of the proposition [accidents happen here]. It simplymeans that accidents sometimes happen here. In this sense, willen in (10) isnot epistemic at all.9

Predication is the second argument for not treating volitive and deon-tic modality as one. Volitive modality does not show a preference for certain


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

(11) a. Hij wil trouwenHe 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. Thisprediction is not borne out because wil in both (11a) and (11b) refers to awish or an internal need. As volitive modality can occur with both eventiveand stative predicates, the kind of modality children use does not necessarilypredict that their RIs have to be eventive.

. Conclusion

In this section I have discussed several aspects of H&H’s recent analysis ofmodal RIs. First, I have shown that [–realised] is a feature that can be trans-lated into known aspectual, temporal and modal notions. Second, some im-plications of morphological marking have been explored. H&H assume that[+perfective] (I prefer to call it ‘here-and-now’) is the default reading that ap-pears when there is no morphology. In English the bare verb form has thisreading. To-infinitives, however, refer to events that are not yet realised. Englishto, reduced to [6] in child language, semantically functions as the -en suffix inDutch and German: it is the infinitival marker and, as such, yields a [–realised]interpretation. Third, the relevance of the connection H&H establish betweenmodality and verb type is discussed. Although such a relation exists, it does notprovide a satisfactory solution for the Dutch and German child data. The mainargument in support of this claim is that volitive and deontic modality havedifferent properties with respect to the predicate they prefer.

In general, H&H’s recent account inspires us to think about specific form-function correlations in (child) language. Their framework raises interestingnew questions, not only about language acquisition but also about language-specific properties. These questions will be addressed in future research. In thesecond, empirical, part of this paper I will test the validity of the Modal Ref-erence Effect for Dutch child language. Blom, Wijnen and Gillis (1998) foundnon-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 useresults from Ud Deen (1997) who found that 13% of the RIs of English- speak-ing children were modal. These modal RIs in English are problematical forHoekstra 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 atmodal 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 Dutchchild language. The first is the essential assumption underlying H&H’s pro-posal, which they formulate as the Modal Reference Effect (MRE). An analysisof corpus data from six Dutch-speaking children will show to what extent theRIs of Dutch children are modal. The second issue is concerned with the com-parison of modal and non-modal RIs in Dutch child data. This comparison isguided by the hypothesis that non-modal RIs are categorised by the children asnominals and modal RIs as verbs.

. Method

Transcriptions of spontaneous speech data of six Dutch-speaking children willbe used. All data are available via CHILDES (MacWhinney 1995). The taperecordings were made at home, in the presence of the mother or father aswell as an investigator. In Laura’s case, the mother and the investigator arethe same person. From each corpus, files have been selected representing thefour stages which Dutch children go through in the acquisition of finiteness(Wijnen 1997):

I. The first occurrences of RIsII. RI stage: almost no finite forms, only RIsIII. Optional infinitive stage: RIs and finite forms co-occurIV. ‘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 areenough data to analyse. Later, when developmental patterns are discussed, thedata will be divided into the stages indicated above. The children’s age rangesare given in Table 1:


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

Child Age range Number of utterances

Abel 1;10.03 – 2;07.29 2890Daan 1;08.21 – 2;09.10 4877Josse 2;00.07 – 2;08.18 3323Laura 1;09.04 – 3;04.06 4241Matthijs 1;09.30 – 2;11.19 5260Peter 1;07.18 – 2;03.21 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, modal-ity is the domain of possibility and necessity. Based on Palmer (1986), I dividemodality into three subcategories: dynamic, deontic and epistemic modality.Dynamic modality implies a subject-internal necessity or possibility. This canbe a wish, ability or capacity. The essence of deontic modality is a subject-external source that is responsible for the necessity or possibility. Obligationand permission are prototypically deontic modal notions. Epistemic modal-ity refers to a speaker’s beliefs or judgements. This type of modality requiressomewhat advanced cognitive skills and is therefore acquired after the age offour (Stephany 1986; Gonsalves 1998).

(12) Modal → dynamic → wish (volition)→ ability, capacity

→ deontic → obligation, requirement→ permission

→ epistemic → probability

Non-modal utterances are coded for aspect/tense. The event expressed can becompleted, ongoing or prospective.

(13) Non-modal → completed→ ongoing→ prospective

For the interpretation we made use of the information available in the tran-scription. For instance, utterances labelled ‘modal’ had to meet one of thefollowing criteria:

– preceding or subsequent parental utterances suggest a modal interpreta-tion, 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 ballb. vrachtwagen emmer doen (Matthijs 2;04.24)

truck basket do-

context: Matthijs wants the investigator to put the truck in the basketc. 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 here-and-now or the past, as in (15), for example:

(15) a. ah, mij bril vallen [= mijn bril valt] (Abel 2;05.27)ah, my glasses fall-

context: his glasses are fallingb. poffie ginke [= koffie drinken]! (Daan 2;01.21)

coffee drink-

context: people on television are drinking coffeec. boot svaje [= varen] (Laura 2;04.01)

boat sail-

context: refers to a picture with a sailing boat

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-modalongoing

Non-modal prospective RIs are often intentional. I categorised these as non-modal because formally they do not express necessity or possibility. The bor-derline between dynamic necessity (volition) and intentionality, however, canbe very narrow.

. Modal infinitives

For German child language Ingram and Thompson (1996) have shown thatthere is a contrast between RIs and simple finite utterances with respect tomodality: RIs are predominantly modal whereas simple finite utterances areused for non-modal purposes. Blom et al. (1998) have arrived at the same con-clusion for Dutch. Using a more extensive database, containing six children


instead of two, the connection between modality and infinitives in Dutch childlanguage will now be ‘rechecked’.

To find out whether there is a connection between modality and infinitiveverb forms, I have made two comparisons:

1. A comparison between RIs and sentences with simple lexical verbs showsif modality is can be connected to infinitival morphology. The predictionis that RIs occur in modal contexts, whereas simple finite verbs occur innon-modal contexts. Note that only finite lexical verbs are relevant here asmodal 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 makesit clear that such a control group is necessary. Assume that the first predic-tion 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 pre-dominantly modal too. In this case, there may be a connection betweenfinite morphology and non-modality, but there is no reason to assume aconnection between modality and infinitival morphology on the verb.

.. ResultsTable 2 gives the numbers and percentages of modal/non-modal RIs, whereasTable 3 gives the numbers and percentages of modal/non-modal simple finiteutterances with lexical verbs. Non-interpretable utterances are excluded fromthe analysis. The tables show a reverse pattern: all children use RIs most fre-quently in modal contexts, whereas simple finite sentences are predominantlyused in non-modal contexts.

Table 2. Modality in root infinitives

Child Number of RIs Modal Non-modal# % # %

Abel 118 91 77 27 23Daan 160 115 72 45 28Josse 192 150 78 42 22Laura 306 202 66 104 34Matthijs 252 201 80 51 20Peter 203 157 77 46 23

Elma Blom

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

Child Number of SFS Modal Non-modal# % # %


Table 4. Calculated χ2 (Modality in RIs/Utterances withsimple finite lexical verbs)

Child χ2

Abel 105.12Daan 160.48Josse 151.27Laura 132.40Matthijs 152.09Peter 406.37

A statistical analysis confirms the patterns observed: for each child the cal-culated value of χ2 is greater than the 3.84 as well as 6.64 (which are, respec-tively, the critical values required for significance at the 5 and the 1 percentsignificance 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-modaluse of simple finite sentences. Note that a simple finite sentence with a modalreference often contains a verb like passen ‘to fit’ or gaan ‘to go’. Although theydo not belong to the class of modal verbs, these verbs may be inherently modalas passen can express (im)possibility and gaan a future event:

(17) a. past niet ijsbeer in niet (Matthijs 2; 04.24)fits not polarbear in not‘the polarbear does not fit in here’

b. ik ga ook naar de dok, of niet? (Laura 3; 04.06)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 (Laura 2; 04.15)I hear horse- not‘I do not hear the little horse’


b. Daan ligt in de wieg (Daan 2; 04.14)Daan lies in the crib‘Daan is lying in the crib’

The results and analysis in the Tables 2, 3 and 4 show that in order to expressmodality children strongly prefer using infinitive main verbs to simple finitelexical verbs. Table 5, which gives the usages of utterances without a verb, alsoshows a significant difference between RIs and verbless utterances in this re-spect. This confirms the connection between modality and RIs. Table 6 givesthe relevant χ2 values showing this.

Table 5. Modality in utterances without verbs

Child Number of utterances Modal Non-modal# % # %


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

Child χ2

Abel 164.66Daan 428.97Josse 178.81Laura 199.99Matthijs 166.81Peter 42.30

All values are statistically significant, which implies that for all children it isthe case that RIs are connected to modality as opposed to utterances withouta 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-modaldistribution is almost at a 50% level. In his case it cannot be concluded thatmodality is exclusively associated with infinitive main verbs. The sentences in(19) give examples of the modal use of utterances without a verb, in (20) non-modal verbless sentences are exemplified:

Elma Blom

(19) a. mama schoen uit (Peter 1; 09.20)mama shoe off‘mama has to take off her shoe’

b. ikke banaan (Josse 2; 03.28)I banana‘I want a banana’

(20) a. Bart ook thuis (Abel 2; 01.16)Bart also home‘Bart is also at home’

b. ik ook broertje zusje (Josse 2; 03.28)I too brother- sister-

‘I get a little brother or sister too’

The above results support H&H’s Modal Reference Effect: Dutch children doprefer to use RIs to express modality. However, if infinitives in Dutch containa [–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 infinitivalmorphology in principle.

. Change of Category

One of the findings of the previous section is that Dutch children allow RIs innon-modal contexts. At first sight, this seems incompatible with the assump-tion that infinitives in Dutch are [–realised]. According to H&H’s definitions,as well as the definitions I used, [–realised] and [non-modal] exclude eachother. This problem may be solved if modal and non-modal RIs have a differ-ent representation. My claim is that non-modal RIs are [+N], whereas modalRIs are [+V]. This is stated in the Change of Category Hypothesis:

Change of Category HypothesisNon-modal RIs are nominal, whereas modal RIs are categorised by thechild as verbs.

Non-modal RIs are like the verbless utterances of the previous section: like thenon-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]. Im-portantly, I assume a basic distinction between these and the [+V] infinitivesof Dutch children. The nominal RIs are deictic expressions. They are used tolabel an action, which is present in perceptible physical context.12 As deictic


RIs are labels of ongoing actions, it is expected that they appear as single ele-ments, i.e. as a one-word utterance. Assuming that they are [+N], this couldalso be made to follow from their nominal character as we do not expect thereto be an internal or external argument. Deixis plays a major role in early lan-guage use, therefore a developmental effect is predicted here: the number ofdeictic RIs will decrease over time. In contrast to nominal RIs, verbal RIs arenot deictic expressions, which means that they are not used and interpreted assuch. Verbal RIs reflect properties of the grammar: the infinitive in these ut-terances has infinitival morphology, which carries a [–realised] feature. Theseutterances can be assigned an aspectual/temporal reference by means of gram-mar. A VP is projected and argument structure may lead to utterances that aremore complex.

.. ResultsThe predictions are summarised below:

1. Non-modal RIs refer to a here-and-now situation and a past reference isexcluded.

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: com-pleted and ongoing. Cases in which it is impossible to distinguish between aninfinitive 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 isnecessary, as children often omit the participial prefix (Jordens 1990). Exam-ples of completed events are provided in (21), and (22) shows RIs that are usedto refer to ongoing events:

Table 7. The interpretation of non-modal RIs

Child Number of RIs Completed Ongoing# % # %


Elma Blom

(21) a. Ieke spugen (Matthijs 1; 11.24)Ieke throw up-

‘Ieke has thrown up’b. Peter koffer pakken (Peter 2; 00.28)

Peter suitcase get-

‘Peter got his suitcase’

(22) a. Ik een hand tekenen (Abel 2; 07.15)I a hand draw-

‘I am drawing a hand’b. Pake foto maken (Daan 2; 04.14)

Pake picture make-

‘Pake is taking a picture’

The first prediction is that completed RIs, or RIs with a past tense reference, donot occur. Though a small number of completed RIs do appear, nearly all (be-tween 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 highcontain examples of repetitive use of RIs: the completed RIs of Josse are with-out exception formed with the verb hoesten ‘to cough’; Laura uses zwemmen ‘toswim’ twice, spugen ‘to throw up’ occurs twice in Matthijs’ completed RIs andPeter uses pakken ‘to get’ three times as a completed RI. If the repetitions areexcluded from the analysis, there are hardly any completed RIs left.

The second prediction is that non-modal RIs occur more frequently as one-word 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 perchild: per child the percentage of one-word non-modal and modal RIs relativeto the total number of non-modal and modal RIs is given. This suggests thatthere may be a connection between being non-modal and appearing as a one-word utterance. Peter is an exception as he is the only child who does not showthis difference: non-modal RIs are more often one-word utterances than modalRIs. However, this could be due to his repetitive volitive (i.e. modal) use ofverbs like stappen ‘to walk’, and koken ‘to cook’.

The third prediction focuses on longitudinal development. As modal RIsare supposed to occur as an effect of the less prominent use of deixis, whichreveals grammatical properties,13 it is predicted that their number increasesin 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


0102030405060708090

100

Abel

Daan

Josse

Laura

Matt

hijsPete

r

Child

Per

cent Nonmodal RIs

Modal RIs

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

reanalysis of the data in which the stages are taken into account shows thatthe proportion of modal and non-modal RIs varies over time. Strikingly, theconnection between modality and RIs does not appear until stage three (whenthe number of RIs and finite constructions is equal, the ‘optional infinitive’stage14). It is not until this stage that there is a significant difference betweenthe number of non-modal and modal RIs. This holds for all children, exceptPeter again, whose data I will discuss later on. Behrens (1993) found the samepattern in German child language. In the early stage (age 1;9) there is no pref-erence with respect to the temporal reference of RIs: 28% can be interpretedas ongoing and 28% as non-ongoing. In the last stage (2;7) there is a cleardifference: 11% is ongoing whereas 85% is non-ongoing.

Peter’s graph suggests a decrease of modal RIs, while the relative numberof non-modal RIs seems to rise. This pattern, however, may be deceptive. Acloser look at the data brings two relevant issues to light. As mentioned before,Peter tends to use certain lemmas repetitively (mostly in volitive contexts). Hedoes this extremely frequently in the file representing the first stage. In this filestappen (‘to walk’) appears 18 times and koken (‘to cook’) appears 6 times, bothonly to express that Peter wants to walk or cook. As the total number of RIs inthis file is 26, it seems reasonable to assume that the data are not widespreadenough to draw any conclusions. Also, one of files representing the fourth stagecontains nearly 50% (9 out of 19 RIs) highly elliptic RIs. These appear whenPeter and his mother are reading a book. Every time they encounter a newpicture, his mother asks Peter what’s X doing? Peter answers with a bare in-

Elma Blom

2. Abel's modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

3. Daan's modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

4. Josse's modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

5. Laura's modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

6. Matthijs' modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

7. Peter's modal RIs

020406080

100

1 2 3 4

Stage

Per

cent

Figures 2–7. Developing modal RIs

finitive (which is, in fact, exactly the same as adult speakers of Dutch woulddo). Although sessions in which the caregiver and the child are reading a booktogether are frequent, this is the only file in which such highly elliptic RIs aredisturbing the results. Figure 8 is corrected for this.

In conclusion to this section I must point out that the second and thirdprediction may be correlated. The number of non-modal RIs decreases as aneffect of development. Modal RIs are more frequent in a later developmental


020406080

100

1 2 3 4

StageP

erce

nt

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

stage. So, the fact that modal RIs are more complex than non-modal RIs couldbe 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 connectionbetween modality and infinitive verb forms is checked by: 1. comparing utter-ances with verbs to utterances without a verb and 2. by comparing utteranceswith simple finite lexical verbs to RIs. Although children use verbless utterancesin modal contexts and RIs in non-modal contexts, the two categories show areverse pattern. Children tend to use utterances that lack a verb in non-modalcontexts and RIs, as opposed to utterances with a simple finite verb form, inmodal contexts. In general, these findings support H&H’s Modal ReferenceEffect, though non-modal use of RIs is not excluded.

The second claim, formulated as the Change of Category Hypothesis, fo-cuses on a basic distinction between modal RIs and non-modal RIs. The un-derlying assumption is that non-modal RIs are [+N] whereas modal RIs arecategorised as [+V]. The first prediction, namely that non-modal RIs are re-stricted to here-and-now events, is borne out. The second prediction was thatnon-modal RIs predominantly occur as one-word utterances while modal RIsconsist of more than one word. The results show a significant difference be-tween modal and non-modal RIs, which could be interpreted as a reflectionof a different representation of the two types of RIs. However, it could also bean effect of development. Moreover, it is a rather imprecise result: sometimesnon-modal RIs contain more than one element and a more detailed study isrequired to establish whether there is a difference in context in which modaland non-modal infinitives are used. The categorial labels [+V] and [+N] maylead to rather specific predictions in this respect. Development brings us to thethird prediction. With respect to the connection between modality and non-

Elma Blom

finite morphology, an analysis based on developmental stages leads to an in-teresting finding: the relation between modality and non-finite verbs does notexist ‘from the beginning’. Not until the optional infinitive stage do childrenuse 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 in-herently [–realised], it must be explained why this feature is not always presentor does not always receive an interpretation. In doing so, I return to the Changeof Category Hypothesis. As the developmental graphs have shown, the OI stageplays a crucial role for the identification of infinitives as verbs. In the OI stagechildren acquire finiteness. In Dutch, finiteness is a specific property of verbs.This could mean that children have to learn finiteness before being able torecognise infinitives as verbs. Why should Dutch children initially categoriseinfinitives as nouns? This is an important question as an explanation basedon reanalysis by children points to a position in the continuity debate: thereare stages in which the child grammar differs from the adult grammar i.e. lan-guage acquisition is a discontinuous process. Discontinuity can easily occur ifit concerns the categorisation of infinitives in Dutch. The ‘verbness’ of infini-tives is not as straightforward as often is assumed. Infinitives in Dutch haveonly one unambiguous property that reveals their verbal character: they havefinite (i.e. tense-marked/agreeing) counterparts. Morphology, for example, isnot a sufficient cue, as the nominal plural suffix in Dutch could be -en like inboek–boeken (‘book–books’). Syntax is rather confusing too, as Dutch allowsa construction in which the infinite verb is preceded by a definite determiner,as in (23):

(23) Wat ben je aan het doen? Auto’s vernielenWhat 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 becomenominal without changing appearance. This is shown in (24):

(24) Baden in karnemelk deed Suzanna bij voorkeur in het weekendBathe- in buttermilk did Suzanna by preference in the weekendBathing in buttermilk was something Suzanna preferred to do in theweekend’

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


H&H assume that infinitival morphology marks the [–realised] feature. Thisassumption implies that children recognise stem and suffix, because if there isno suffix, this means that there is nothing to connect [–realised] to. This situa-tion describes the stage before the OI stage. Before the OI stage, Dutch childrendo not identify infinitival morphology as such. Therefore, there is no suffix thatcould mark [–realised]. This accounts for the finding that the association be-tween modality and the infinite main verb does not appear from early on, butarises in the OI stage.

. Concluding remarks

This article started with an overview of current theories that explain modal RIsin Dutch and German child language. Hoekstra and Hyams’ (1998) recent pro-posal was singled out and discussed in more detail. This led to many new ques-tions about child language, form-function correlations and cross-linguistic dif-ferences. The second part focused on the empirical base of the theories whichaccount for modal RIs. First of all, it was shown that the connection betweenmodality and infinitive verb forms does exist, although there are also a rela-tively large number of non-modal RIs. In H&H’s view non-modal use of RIsshould be excluded. If, however, modal and non-modal RIs are analysed differ-ently, this problem may be solved. I proposed that modal RIs are [+V] whereasnon-modal RIs, since they are deictic expressions, are [+N]. Data of six Dutchchildren seem to support this proposal. Though maybe not immediately obvi-ous, the claim that infinitives are initially analysed by children as nouns seemsfeasible. In Dutch, morphology is the only unambiguous cue for children toclassify infinitives as [+V]. My claim is that children have to be able to alternatebetween finite and non-finite morphology to recognise infinitival morphology.This ability is acquired in the OI stage. Strikingly, in the OI stage RIs becomemodal. If it is true that children do not recognise infinitival morphology beforethe OI stage, this finding supports H&H’s claim that [–realised] is connectedto infinitival morphology.

Notes

* I would like to thank Frank Wijnen, Nina Hyams, Evelien Krikhaar, Ingeborg Lasser, Brigitvan der Pas, two anonymous reviewers and the editors for their careful reading of earlierversions 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 andG.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 (toappear). Taken as a whole, exclamative RIs are not (rhetorical) questions but rather de-scriptions of a ‘hypothetical event’ (Akmajian 1984). It is crucial is that exclamatives alwaysconsist 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 thepoint 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 differ-ences she found between German adult and child RIs was that adults use RIs most frequentlyas imperatives while children use them most often as declaratives. Though declaratives andassertions do not refer to the same, the notions are closely related. As Clark (1996:136) putsit: ‘To assert something, you choose a declarative’.

. Opinions differ with respect to the treatment of volition as modality or as a modality-related 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, whileDe 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 claimhere. As described by Lyons (1997:826), both child usages can have one source: ‘The originof deontic modality, it has often been suggested, is to be sought in the desiderative and in-strumental function of language: that is to say, in the use of language, on the one hand, toexpress or indicate wants and desires and, on the other, to get things done by imposing one’swill on other agents. It seems clear that these two functions are ontogenetically basic, in thesense that they are associated with language from the earliest stage of its development in thechild. It is equally clear that they are very closely connected. It is a small step from a desider-ative utterance meaning ‘I want the book’ to an instrumental utterance meaning ‘Give methe book’; and parents will commonly interpret the child’s early desiderative utterances asmands, thereby reinforcing, if not actually creating, the child’s developing awareness that hecan use language in order to satisfy his wants and desires’.

. Ferdinand (1996) analyses non-finite utterances in French child language along the samelines, though she explicitly assumes a covert modal or aspectual auxiliary. This modal oraspectual auxiliary accounts for the use of eventive verbs in these non-finite utterances. Fer-dinand’s observation for French that certain modalities co-occur with eventive infinitivesreappears as Hoekstra and Hyams (1998) explain the meaning of RIs in Dutch and Germanchild 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 ofRIs in child Dutch and German.

. Sjef Barbiers pointed this out to me.


. 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 predicateover the whole sentence. It could be argued that this is what we see in (10). By contrast, theroot modal willen in (i) is a two-place predicate expressing a wish of the subject.

(i) Marie wil een appel etenMary 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 prob-lematical 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 andPérez-Leloux’ (1997:129) proposal: ‘Another possibility is that the child chooses a radicallydifferent structure and the correct interpretation is obtained pragmatically. For instance, anominal 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 mightrefer 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 Tenseand is consistent with economy, then the child may choose an altogether different structureand 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 theamount of infinite and finite constructions is approximately equal. There are no theoreticalimplications.

References

Akmajian, A. (1984). Sentence types and the form-function fit. Natural Language andLinguistic Theory, 2, 1–23.

Anderson, L. (1986). Evidentials, paths of change, and mental maps: Typology regular meetsasymmetries. In W. Chafe and J. Nichols (Eds.), Evidentiality: The Linguistic Coding ofEpistemology (pp. 227–274). Norwood: Ablex.

Avrutin, S. (1997). EVENTS as units of discourse representation in root infinitives. MITOccasional 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 pres-ented at the Workshop on Finiteness held at the Annual Conference of the DeutscheGesellschaft für Sprachwissenschaft, Marburg March 1–3, 2000.

Blom, E. and F. Wijnen (2000). How Dutch children’s root infinitives become modal. InS. Howell, S. Fish and T. Keith-Lucas (Eds.), Proceedings Boston University Conferenceon Language Development 24 (pp. 128–139).

Blom, E., F. Wijnen, and S. Gillis. (1998). Modal infinitives in the speech of Dutch mothersand their children. In J. van der Auwera, F. Durieux and L. Lejeune (Eds.), English as aHuman 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 inearly 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 introductoryessay. 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.), LanguageTypology 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 deLinguistique Générale.

Gerhardt, J. (1991). The meaning and use of the modals HAFTA, NEEDTA and WANNA inchildren’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. NewYork: 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. deHaan and W. Zonneveld (Eds.), Formal Parameters of Generative Grammar, Yearbook1987 (pp. 15–30). Dordrecht: ICG.


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. Kosterand F. Wijnen (Eds.), Proceedings of GALA 1995 (pp. 239–250). Groningen: Center forLanguage 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 pres-

ented at the Antwerp conference on Modal Verbs in Germanic and RomanceLanguages, 10–12 December 1998.

Krämer, I. (1993). The licensing of subjects in early child language. MIT Working Papers inLinguistics, 19, 197–212.

Kursawe, C. (1994). Fragesätze in der deutsche Kindersprache. MA dissertation, Universityof Düsseldorf.

Lasser, I. (1997). Finiteness in Adult and Child German. Doctoral dissertation, University ofNijmegen.

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 Lin-

guistics 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 inLinguistics, 12, 114–133.

Ross, J.R. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in PhilosophicalLinguistics series 1 (pp. 77–102). Evanston: Great Expectations.

Santelmann, L. (1995). The Acquisition of Verb Second in Child Swedish. Doctoraldissertation, 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 OccasionalPapers in Linguistics, 12, 1–25.

Modals and negation in English*

Annabel Cormack and Neil SmithUniversity College London

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

. Background

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

(1)


In the Minimalist approach, the pair {LF, PF}, giving the inputs to the pro-cessing by the auditory/articulatory and the conceptual/intentional process-ing systems, is generated as shown in (1) above. However, we have arguedthat at least with respect to heads, the system above is unnecessarily compli-cated. We propose that there is no movement of heads (and hence no Moveon the diagram above for heads). Rather the LF-interpretable part of a head ismerged in its proper scope position with respect to other heads, and the PF-interpretable 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 sameas the merge position of the LF-interpretable part of that head, which we showin terms of a fairly standard syntactic representation. Morphological consid-erations, 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, inits LF-interpretable position.1 For most of this paper, the positions of the PF-parts 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 structuregiven in small capitals (e.g. as in (2) onwards) as independent representationsof 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 SplitSign hypothesis gives us is the following. If in a particular interpretation ofa sentence, we find that a modal Mx has scope over some item Y, then Mx ismerged in a position c-commanding Y – and this is coded in the grammar bythe usual requirement that Mx selects for Y(P), or that Y is an optional adjunctto the projection that Mx selects for. We rule out the possibility that althoughmerged below Y, Mx has scope over Y because it has been moved to c-commandY at LF. Our assumption considerably simplifies the task of the child languagelearner, and the linguist, and turns out to have interesting consequences withrespect to modals and negation.2

The issue of the syntactic scope of modals with respect to negation wasraised by Picallo’s (1990) work on Catalan. She demonstrates that in Cata-lan, sentences such as (2) and (3) are monoclausal, and that some modals aremerged below negation, and some above. If negation is present, modals withan epistemic interpretation are never within its scope, but those having a deon-tic (‘root’) interpretation always are. The examples in (2) and (3) are Picallo’s


(1990:287, (4) and (5)), and use the modal poder ‘may/can’.3 The scope/LF an-notation, where small capitals represent the LF-interpretable part of a head,is ours.

(2) En Jordi potmay

nonot

haverhave

sortitleft

‘It is possible that Jordi hasn’t left yet’ epistemic: [

(3) En Jordi nonot

hahas

pogutcould

sortirleave

‘Jordi hasn’t been able to leave’ deontic: [

Here, and throughout, we use ‘epistemic’ as a cover term for all kinds of alethicand epistemic uses, including those based on logical or (folk-)scientific theo-ries, and ‘deontic’ to cover all sorts of obligation, and also disposition, willing-ness, intention, and ability. The claim implicit in this is that if a distinction ofthis kind is made between two classes of modal meanings, then all natural lan-guages draw the distinction in the way indicated. This claim may of course turnout 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 negationin echoic contexts may give rise to a wider range of scope interpretations (seeSection 6 below), we restrict ourselves for the moment to interpretations innon-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 (epistemic)‘It is predictable that Bob will not be late’ [

*‘It is not the case that it is predictable that Bobwill be late’ *NOT [SHOULD

(6) Edwin can’t climb trees (deontic)*‘Edwin is able not to climb trees’ *CAN [NOT

‘It is not the case that Edwin can climb trees’ [

(7) Jean can’t have left (epistemic)*‘It is possible that Jean has not left’ *CAN [NOT

‘It is not the case that it is possible for Jean to have left’ [


Examples (4) and (5) with should are unambiguous: the modal should has scopeover negation. Examples (6) and (7) with can’t are similarly unambiguous, buthere 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 thecentre of interest here is the modals, we do this fairly briefly. More detaileddiscussion and argument may be found in Cormack and Smith (2000), whichalso discusses comparable scope data from Catalan, Italian and Basque. In Sec-tion 2, we distinguish two forms of negation, Pol[NEG] and Adv[NEG] (corre-sponding roughly to the familiar sentential vs. verb phrase negation). We thenshow in Section 3 that English modals vary in their relation to Pol[NEG], re-quiring that we divide them into those occurring pre-Pol (Modal1) and thoseoccurring post-Pol (Modal2). The distribution of modals between the twoclasses is not however the same as in Catalan. Section 5 discusses the learn-ability problem this apparently causes. In Section 6 we show that in order toaccount for the scope facts exhibited in echoic contexts, our hypotheses requireus to introduce a third negation/polarity position, which we call Echo, mergedhigher than Modal1; Section 7 shows further phenomena now accounted for.In Section 8, we discuss the relative scope of epistemics and deontics, particu-larly 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 aninterpretation with CAN in the scope of , as before, and an additional read-ing with in the scope of CAN. This contrast is attributable to the differ-ence 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


and inferential differences, as documented by Klima (1964). For example, intag questions, the polarity of the normal tag is the opposite of the polarity ofPol, and is unaffected by Adv[NEG].4 That the extra reading in (9b), is due tothe occurrence of Adv[NEG] can then be seen from the normal tag appropriateto 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, withAdv[NEG] in (11b), suggesting that the same structural alternation obtainshere 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 asfollows. Pol is a functional head, obligatorily present in all tensed clauses, andmay 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 isadjoined to some projection than Pol.5 Adv[NEG] is only instantiatedas 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, thereare reasons in English for placing T above Pol. In John sometimes does notwear his glasses, the temporal adverb sometimes has scope over the negationPol[NEG] not; but assuming that T is a temporal binder, sometimes must be inthe 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 modalsfall into two classes, Modal1 and Modal2, and that each class is merged in a fixedposition with respect to Pol. Should is an example of Modal1, and is merged


higher than Pol; thus it has scope over both Pol[NEG] and Adv[NEG]. Can isan example of Modal2, and is merged lower than Pol; thus it has scope lowerthan Pol[NEG], but higher than Adv[NEG], as required to give the alterna-tion in (9). Plausibly, Modal1 and Modal2 in English and in Catalan are ad-juncts, and optional (but see Section 5). The sequence in (12) can therefore beelaborated as in (13).

(13) C T (Modal1) Pol( /) (Modal2) (Adv[]) . . .

Since the surface order in tensed clauses is never negation followed by a ver-bal element, as shown by the ungrammaticality of (14), some PF-displacementmust take place. We return to this issue in Section 4.

(14) a. *John not can eat puddingb. *John not eats pudding

Using -n’t as a probe for Pol, we are in a position to compare English withCatalan. As in Catalan, we find two classes of modal, given above as Modal1and Modal2. In the Catalan examples, the two classes divide along the epis-temic vs. deontic dimension. In English, we find that the main division be-tween Modal1 and Modal2 corresponds to the contrast between necessity andpossibility, where obligation patterns with the former, and permission with thelatter. However the epistemic vs. deontic division, also turns out to be relevant.

The English modals which belong to the class Modal1, like should, includealmost 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 homeb. *John must not be at home, must he (normal tag)c. *John mustn’t be at home

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


an accidental gap. We can code it by stipulating that epistemic must selects forPol[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, whereit 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 con-versation about TV) Johnny’s teacher says he hasn’t to watch any TVtoday: he’s got too much work to do. [

b. (context: Head to parents) #In this school, a boy hasn’t to play gamesunless he wants to. * [

The only necessity modal falling under Modal2 is need (without to). This needis an NPI (negative polarity item), and in a declarative, has to be within thescope of a negative element.7 In (21), the only possible interpretations requirethat need is a Modal2, in the scope of Pol[NEG]:

(21) a. Hugh needn’t leave, need he (deontic) [

b. Unicorns needn’t exist (epistemic) [

The assignment of a modal to necessity or possibility is not always obvious. Forinstance, it is not easy to tell whether volitional modal will and would, and themodal shall, are necessity or possibility operators. If we consider ‘wilfulness’,or ‘will’ in the sense of intention, then they should fall with necessity, but ifwe consider rather ‘willingness’, then they would appear to fall with possibil-ity. Since ‘�¬’ and ‘¬♦’ are logically equivalent, scope facts with respect toPol[NEG] are unhelpful. However, we saw in (10) and (11) that an occurrenceof 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 scopeinterpretation (‘John intends not to come home on time’), but if we treat it ascorresponding to willingness, we fail (#‘John is willing not to come home ontime’). 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 toModal1, along with the majority of the necessity modals.

(23) John won’t come home on time, will he intention [Pol


Similar arguments lead us to assign the related past form would, and volitionalshall, 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 outsidethe 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 hasonly a root meaning, and is also generally an NPI). It seems then that for themajority of modals, the pre-Pol/post-Pol split in English corresponds to the‘necessity’ vs. ‘possibility’ contrast, rather than to the epistemic vs. deonticcontrast.

We have not yet mentioned may and might. Surprisingly, their scope be-haviour is like that of the Catalan modals. From (24a) and (25), we see thatthe interpretation for deontic may, is inside the scope of Pol[NEG] and out-side it for epistemic may. For those who reject mayn’t, and allow only maynot, (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 as-suming the LF order with deontic inside the scope of Pol. In (25b), bycontrast, no such interpretation arises, allowing us to infer that epistemic

is necessarily pre-Pol.

(24) a. Cyril mayn’t go to the party (deontic)�=‘It is permitted that Cyril not go to the party’ *MAY NOT

‘It is not the case that Cyril is permitted to go to the party’

b. Cyril may not go to the party (deontic)i. ‘It is permitted that Cyril not go to the party’

ii. ‘It is not the case that Cyril is permitted to go tothe party’

(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 interpretationas epistemic or deontic.8

If we assume that the position of a modal with respect to Pol[NEG] is fixedby selection, then the default assumption is that what is relevant for selectionis simply Pol. We have referred to pre-Pol modals as Modal1, and to the post-Pol modals as Modal2. As far as we know, Modal2 and non-modal Aux show


Table 1. LF positions for English modals

Pre-Pol (Modal1) necessity shall, should, must, will, would, ought + to, is + to,have +to

possibility epistemic readings only: may, might

Post-Pol (Modal2) 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 asingle 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 hereconform, 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 ourdeontics), and to Brown (1991), with the proviso throughout that we have ex-cluded readings which we take as obtainable only in echoic contexts, as wellas 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 modalhas Root meaning” (i.e. deontics display Modal2 properties) “and affects themain predication if the modal has Epistemic meaning”, (i.e. epistemics displayModal1 properties).10

English also manifests Raising verbs with modal interpretations, recognis-able 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 scopelower than Pol[NEG] and lower than any Modal/Aux. We will have little to sayabout them.

We turn next to the PF positions of the modals, and to the problem raisedby 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 displace-ment of a finitely inflected Modal2 to a higher position than Pol[NEG], and fora simple solution to a problem raised by can’t and won’t. Specifically, we arguethat there are good reasons for treating these items as the PF part of a singlelexical entry, although they are NOT unitary LF constituents.


We begin with the negative -n’t in can’t and won’t. One traditional treat-ment of -n’t postulates that it is a clitic, which attaches leftwards, to a tensedmodal in T. An alternative has -n’t as a morphophonological operator, to al-low for the irregularity when it cliticises for example to can or will, perhapsas 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 firstsuggestion is implausible (Zwicky and Pullum (1983) provide systematic argu-mentation against the clitic hypothesis, in part on these grounds). Second, amorphophonological operator should affect only a head for which it selects,and we have argued that Modal1 (which includes will) selects for Pol, ratherthan vice versa, so the second suggestion is similarly suspect.

In this situation, the appropriate and traditional move is to assume thatitems 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 forthe lexical entry is not the standard one. The standard solution is to assumethat these irregular forms, and perhaps all the -n’t forms, are entered as wholesin the lexicon. But this is problematic, under standard assumptions about thearchitecture 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 lex-icon. However, as can be seen from the associated LF scope information,

+ is not a constituent at LF. Under these conditions, it is very hard toconstruct a meaning for can’t, and probably impossible without making theunjustified assumption that often is an optional argument of can’t. We claimthat can’t has no unitary meaning, and hence no lexical entry of the standardform can be constructed.11

We claimed above that LF-parts of signs are merged in the positions atwhich they are interpreted. We further assumed – uncontroversially in a lex-icalist framework – that the lexicon may contain morphologically complexphonological words. We see no reason to suppose that these are Merged any-where other than in the place where they are heard. The PF-interpretable partof 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 inwhich 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 LF-interpretable part before merger (see Cormack and Smith 1997, 1998). Theoperation ‘Split’ is not unique to our framework as it is required by UG,


since the structural representation of a sentence, phrase or word is interpretedindependently at each of LF and PF.

We take it that only the nodes and structure provided by LF may be usedin determining the possible merge positions of the PF-interpretable part of asign. The default case is that the PF-part is merged along with its LF-part on thesame tree. In other cases, the PF-parts of signs are merged at nodes which arerequired by the LF-structure, rather than in an independently derived structureof the kind proposed by Jackendoff (1997).12

In (27), the different orderings at PF and LF show that either the Modal orthe Negation must be split.

(27) The dog could not reach the ball PF order: could notLF (scope) order: [ [ . . .

That it is the modal which is split in (27) can be seen from the interaction withadverbs (Ernst 1992), as in the examples shown in (28) with the scope readingsgiven.

(28) a. John can not come home late [Pol[NEG] [

b. Sylvia can never climb that tree [Pol[POS]Ø [ [

c. You could not often bribe officials, then [Pol[NEG] [ [

In particular, in (28c) the LF scope ordering of not and can is disturbed at PFrelative 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 someposition preceding Pol, including Pol[POS] in the case of (28b). In Cormackand 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 twoLF-items: a T-related inflectional head, Inflpast, and LF-CAN.13 It is merged atthe position of LF-Inflpast. The T-related Infl has the identity function (ID) forits meaning. Hence for (28c) we obtain a structure with the parts distributedas in (29).

(29) LF:PF:

[T

Ø[InflPast

could[Pol

not[Adv

often[Modal

The required lexical entry is not simply a pairing of PF-interpretable form andLF-interpretable meaning. In particular, if a PF form is associated with two LFforms 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 tohave as features both the category of A and that of B. For could, the entry relatesone form to two meanings, as shown in (30):


(30) PF:LF:

category {Inflpast, Modal}; form /kυd/{(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:LF:

category {Inflpres, Modal, Pol[]}; form /ka:nt/{(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 repre-sentation is not stipulated in the semantic part of the lexical entry, but arisesonly 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 classof modal. All that is required is that a proper checking relation hold betweeneach LF category and the related composite PF category. This account dealssatisfactorily not only with the problem of can’t and won’t, but with the dis-placement 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, itis probable that the modals have the same category, Aux, as the regular auxil-iaries have, be, and do. Modal1 is then distinguished by selecting for (a projec-tion of) the functional head Pol, and the modal auxiliaries generally are distin-guished in standard English, but not in Scots English, by their lack of non-finiteinflectional 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 allModals are merged under T (or I), cannot be correct. While this analysis mightbe 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 (postulatingthat only epistemics are under I/T), as does Roberts (1998) for English. Fur-ther evidence can be provided from the double modal constructions of ScotsEnglish and other languages, as Brown (1991) shows for Hawick Scots in (32),and Picallo (1990) for Catalan in (33).


(32) He should can go tomorrow

(33) En Pere deumust

podercan

tocarplay

elthe

pianopiano

‘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 ex-pect iteration of Modal1 or Modal2 to be possible in principle. This is excludedin 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 ofModal1 and Modal2 as standard English, iteration can apparently occur in othervarieties 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 wedraw this conclusion. For example, Brown and Millar (1980) state that in theEdinburgh 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 un-equivocal, since might has some curious behaviours. It does not occur withnegation except Adv[NEG], nor does it invert, in Scots. In Southern StatesEnglish ‘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 maybe a main verb here, not a modal auxiliary.17 We have in fact no direct evidencethat the categories Modal1 or Modal2 may iterate.18

We wish to raise two further issues. One, on relative scope, we defer to Sec-tion 8; the other, on learnability, we discuss now. The obvious problem whichemerges from the English data is that the classification of a particular modalunder Modal1 or Modal2 is going to have to be learned.19 If the Catalan patternhad been common to every language, we could have looked to UG or semanticsfor an explanation of the distribution, leaving the child with nothing to learn.As it is, however, the learner appears to have a formidable task, particularlygiven the erratic distribution of modals in English.

However, there is evidence that deontics are always learned earlier thanepistemics (see Papafragou 1998 for a survey), so we may assume that theirclassification is also learned first. If we take just the deontics from the table in


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 obtainthe sub-table shown in Table 2.

This is much more helpful. We may suppose that the child knows that thereis a distinction between obligation vs. permission/ability modals(given as necessity vs. possibility in the table), and its task is to determinewhether there is such a split in the ambient language: ‘No’, as in Catalan, whereall the deontics fall under Modal2, or ‘Yes’, as in English. The older child, learn-ing the epistemics, can take the pattern for the deontics as the default. For En-glish, this will entail learning that the exceptional epistemic may and mightfall under Modal1, and that the obsolescent necessity items need falls underModal2. For Catalan, it will entail learning that the totality of the epistemicsfall under Modal1.

It is tempting to speculate further, but we need relevant data from morelanguages. In particular, we need to know whether all languages express-ing these notions with modal auxiliaries have one major split, and whetherthe combinations epistemic+necessity and deontic+possibility always correlatewith Modal1 and Modal2, respectively.

. Negation in echoic contexts

There are some exceptions to the generalisation that Modal1 takes negation inits scope (see Gazdar et al. 1982, fn.17; Palmer 1990; Brown 1991; Williams1994; Menaugh 1995 among others).20 In all the following examples, negationmay 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 justintonation:

(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?


It is also available in declaratives:

(39) A: You should eat more vegetablesB: No I shouldn’tInterpretation: 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 shouldto be in the scope of the negation. But there are robust data indicating thatshould is a Modal1 head, falling always outside the scope of negation in ordinarycontexts. There are two problems then: what contexts require that should fallsinside negation, and how is this accommodated within our framework.

It is important to note that the surface position of the negative morphemeis insufficient to predict the scope order. For a negative element to have scopeover should, it is neither necessary nor sufficient that it c-command should atsurface. In (40), a negative element c-commands should, but the scope order isas 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 nega-tion 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 characterisethe contexts exhibited in (36), (38) and (39) as . The echoic natureof the reply in (39) is clear: the content ‘you should eat more vegetables’ is anecho of the original given by speaker A. Tags, as in (38), are always echoic. Theidea 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 toits negation. We call the negation which is external to an echoed propositionecho-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 accountedfor? If, as we have argued, Merge gives the scope position of a head, then thereis only one possible answer: there must be a syntactic position, (call it Echo, fornegation in an echoic context), somewhere above Modal1.22 This in turn leadsto questions about the status of the not in (41). We will simply assume herethat although Echo[NEG] is high in the tree, there are reasons why its PF-partis merged at a position lower than the tense-inflected modal or auxiliary.23


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 onewould expect, appear embedded under heads which select for representationsof thoughts or speech, such as (42), and it appears lower than the adjoinedif -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 withrespect to other optional adjuncts such as Topic. The range of heads we areconcerned 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 betweenthe modals and the different negation positions.

We have couched our analysis of the data in accordance with our hypoth-esis that the LF-interpretable part of a head is merged at the position in whichit 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 probablyassume that there is only one [NEG] head, subsuming Echo and Pol, and thatthis is merged at the PF-position. Subsequently, some process of LF-movementraises at least the LF-interpretable part of [NEG] in echoic contexts – per-haps to some head which we might still call Echo. Note however that giventhe data about the scope of adverbs (Section 4), we cannot similarly invokemovement 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 motivationof a semantic kind for so doing.

Discussions of the modals designed to explain their use and usages (suchas that of Palmer 1990) are often concerned with how certain meanings may beexpressed. 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 isa gap. It is not possible to express the semantic content given by ‘ [Modal1

. . . ’ directly except in echoic contexts.


In a language like English, with a variety of modals, there is a simple al-ternative: switch to a Modal2 or a main verb with the same meaning. Thus toexpress epistemic ‘ [ . . . ’, it is possible to use Pol[NEG] with Modal2

can; to express ‘ [ . . . ’ it is possible to use Pol[NEG] and Modal2 needor main verb has to. However, it is important to realise that the semantic con-tent ‘ [Modal1 . . . ’ is available with Echo[NEG], and is appropriate for tags,and often for questions, as in (36) and (39) above. Palmer (1990:76) notesthe wide scope of negation in what he calls ‘verbal crossing out’, giving theexample in (45):

(45) He must come – Oh no he mustn’t.

Notice too that epistemic must, although it cannot occur with Pol[NEG] instandard British English, is available with Echo[NEG], as in (46), or in oneinterpretation 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 andnegation. Interestingly, the scope of the modals in relation to the various nega-tion positions seems to have implications for the scope of negative elements ofother 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 thenegative forms of the modal. . . .

You must never do thatYou need never do thatI must tell no one about itI need tell no one about it.

As we would expect from this, with Modal2 can, the scope of a negative itemsuch 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 epis-temic readings. (49) interpreted deontically is ambiguous in the same way as(48) is. However, interpreted epistemically, it can only have the ‘POSSIBLE


[NOT [∃. . . ’ interpretation, ‘That John may put nothing in the collection plateis 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 readingsof 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 aroundneed.24 However, in all these cases, the takes scope in relation to the modalin a position where either Adv[NEG] or Pol[NEG] could. This provides sup-port for an analysis modelled on that of Haegeman (1995:182–187), where thepresence of an n-word/phrase induces a [NEG] feature on Pol.25 However, be-cause 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 appropri-ate. 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 quanti-fier ‘no’, for instance, has a PF no associated both with [X¬] and with an exis-tential quantifier [D∃], where [X¬] may be Pol[NEG] or Adv[NEG], or, as weshall see, Echo[NEG]. This solution is closely related to the neg-incorporationof Klima (1964:267–280). We do not discuss here the means by which the exis-tential gets to obtain the correct scope, nor do we give all the possible readingsfor the example sentences (in particular, other split readings are obtainable).

An additional benefit of our analysis is that the scope of the negative ele-ment is correctly predicted for negative operators c-commanding must at LF: ina non-echoic context, (52) is unambiguous, and (53), with no scope alternationpossible, 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 havingscope 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 byPalmer (1990:117).


(54) a. No one gob. 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 notpredictable from the semantics of the modal, but precisely follow the vagariesof 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 aModal Scope Constraint which takes the Catalan pattern to be operative as amatter of necessity. The argument has two versions: first, that epistemics mustof necessity fall outside negation, and deontics inside; and second, that epis-temic modality must have scope over deontic modality, and in consequence,that the two are ordered thus in syntax.26 Cinque (1999) is an extensive andelaborate cross-linguistic attempt to order functional heads, including modaladverbs and auxiliaries, but specifically excluding negation, in this fashion. Hisordering of positions includes the array shown in (56) (abstracted and sim-plified from Cinque’s (90) of §3.7), but he casts the restrictions as generallybelonging 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 fixedposition, and granting that English has Modal1 deontics (e.g. deontic must)and Modal2 epistemics (e.g. epistemic can), it is clear that we cannot agreethat epistemics are merged above deontics (and we also argue in Cormack andSmith (2000) that epistemics are merged below T).

Further, even in Italian, which provided the initial impetus for Cinque’sanalysis, there is evidence that there are Modal1 deontics and Modal2 epis-temics. In the deontic examples below, provided by Vieri Samek-Lodovici, thecontrast between the scopes of potere and dovere with respect to the negationcan be explained by assuming that we have Pol[NEG] here, and that deonticpotere is a Modal2 and dovere a Modal1.28


(57) Gianni non può canTAre. [ *CAN [NOT

(58) Gianni non deve canTAre *NOT [MUST [

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

. . . , nonnot

puòmay

PIOvererain

[

. . . , nonnot

dovrebbemust

PIOvererain

[

The conclusion then is that in standard Italian, potere ‘can’ is a Modal2, anddovere ‘must’ is a Modal1, like their English equivalents.

Cinque’s analysis, however, also makes some correct predictions for En-glish. If in standard British English, we set up a clause with two modals ormodal verbs, we indeed find that the interpretation never has a deontic withscope over an epistemic.

(60) John might be able to get here in timeModal2 (epistemic) – verb (deontic)

(61) John could have to leave before nineModal2 (epistemic) – verb (deontic)

(62) John must have to stay at home Modal1 (epistemic) – verb (deontic)

(63) There must/might have to be three solutionsModal1/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 thefact that it is at least usual for two epistemic or two deontic modal terms usedsuccessively to have differing interpretations with respect to the source of themodal imperative or permit (Kratzer 1977), as in the sample gloss in (64) forinstance.

In Scots, where double modals are permitted, the same interpretive con-straints hold. The examples below, and the paraphrases where given, are takenfrom Brown (1991). Futurate will/would is omitted from our scope annotation.

(65) He should can go tomorrow Modal1epistemic – Modal2deontic(= ‘he ought to be able to go tomorrow’)


(66) He would could do it if he tried Modal1futurate – Modal2deontic(= ‘he would be able to do it if he tried’)

(67) He’ll might can come in the morn Modal1epistemic – Modal2deontic(It is possible that he will be able to . . .)

(68) You’ll have to can do it whether you like it or notModal1epistemic – Modal2deontic29

(= ‘you’ll have to be able to do it . . .’)

(69) He might could have to goModal1epistemic – Modal2deontic – Main-verb deontic

Brown (1991:76–78) claims that with modal auxiliaries, the only possiblemodal combinations are drawn from a (future will)–(epistemic)–(deontic) se-quence, with no two of the same class. This suggests that the modal sequence isrestricted to one future (perhaps under T), one Modal1, and one Modal2. If thisis correct, then it seems that Modal1 and Modal2 might be optional functionalprojections, rather than optional adjuncts which could iterate.30

Our hypothesis entails that there is another constraint on the ordering oftwo modals: Modal1 must precede Modal2 at LF (the ‘Modal1>Modal2’ con-straint). In a language like Catalan, the two constraints will coincide, but inScottish English, they will not. Examination of the examples given in Brown(1991) reveals no counter evidence. Rather, constraints must be met, sothat 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 doublemodal is can/could, since this is the only deontic Modal2. Combinations suchas must should are not permitted even when interpreted as epistemic followedby deontic.31

If both these independent constraints on scope ordering hold, then onlyone of them can be syntactic, that is, be coded by selection. In fact, there areseveral reasons for supposing that the Modal1>Modal2 organisation is syntac-tic, and the epistemic–deontic Modal Scope Constraint is conceptual.

A. As we have seen in English, the choice of Modal1 or Modal2 is not systemat-ically related to meaning, so it must be lexical. The use of a fixed functionalhead Pol makes this a matter of selection, which is a normal part of a lexicalentry. For Cinque, the situation would be harder, since the Modal1>Modal2

constraint could not be one of selection, but (presumably) simply stipula-tion of scope with respect to Pol (where Pol may generally be inserted freelyamong the functional heads). In addition, it would also be necessary to en-


sure that Pol[POS] was inserted exactly once into or at the boundary of therelevant string of functional projections, where Pol[NEG] is not present.32

B. The epistemic–deontic ordering constraint does not hold just of modalauxiliaries (and adverbs), but holds also of modal main verbs. This canbe seen in the English examples in (60) to (64), and holds of languageswhich probably only have main verb “modals”: e.g. Icelandic (Thráinssonand 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 liter-ature that epistemics refer to the speaker’s evaluation of a proposition (a state-ment about possible worlds) vis-à-vis the actual world, whereas deontics referto situations rather than worlds. This is supposed to explain the scope con-straint, though it is not clear exactly how. We suggest instead an explanation de-pending on the properties of deontics (as does Vikner 1988). Deontic and otherroot meanings are normally directed at some responsible person. Note that asthe following examples demonstrate, the responsible person is not necessarilyeven the underlying subject of the modalised clause.34

(70) a. The baby need not eat before noonb. The room must be tidy before I come back

The responsible person is (or is not) permitted or required to undertake someaction or realise some state: for obvious reasons, the requirement should beone that it is possible to undertake; and if its description were modified byepistemic or alethic modals, than that part of the undertaking is not withinthe powers of the responsible person to alter.35 Thus it cannot form part of aproper 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 haveto happen to me?” to have the deontically interpreted modal followed by anepistemically 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 morecomplex structures. The Epistemic>Deontic constraint cannot hold through-


out an arbitrarily long sentence. We obviously would not expect it to holdacross clause boundaries with an independent subject in the lower clause, asin (73a). It holds only for reasonably close modals. Thus deontic followed byepistemic interpretation is available for the examples in (73) but not those in(74) and (75):

(73) a. John can stipulate that the puzzle must have at least one solutionb. John can claim it to have to be true that quarks existc. John can want himself not to have to be mortald. John can want PRO not to have to be mortale. 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 mortalb. John may end up t not having to be mortalc. #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 non-modal V, irrespective of the kind of subject the intermediate verb licenses. Itappears 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 withthe Conceptual approach to the Epistemic>Deontic constraint: in all the casesabove with the lower clause having the “same” subject as the matrix clause, thecomplement of the intervening verb is interpretive (in the sense of RelevanceTheory), representing wishes, thoughts, etc., on the part of the subject, and sobeing freely attainable by the responsible person.

As a conceptual constraint, we could interpret the Epistemic>Deontic con-straint in two ways. One, which would rule out the suggested deontic>epistemicinterpretation of (72), would hypothesise that the constraint held over rep-resentations in the Language of Thought (Fodor 1975). The combination ofmodal and epistemic is literally unthinkable. The weaker claim would be thatthere was a pragmatic processing constraint or strong preference, which deter-mined that when two modal expressions occurred in the same modal domain(i.e. without an intervening lexical verb) then they would not be interpretedas a deontic followed by an epistemic, because this was never likely to be thecorrect interpretation.

If the constraint that epistemics take scope over deontics really is concep-tual, then the Catalan system would seem quite natural, though not a necessity.We know of no rationale for explaining the necessity/possibility split evidencedby English.


. Conclusion

In undertaking this analysis of the behaviour of modals with respect to nega-tion, we have made essential use of syntactic, semantic, and pragmatic theo-ries. Our initial theoretical assumption, that heads are merged in their relativescope positions, drives the analysis. It led us to introduce a separate projectionfor Echo[NEG], in addition to Pol[NEG] and Adv[NEG], and to divide themodals into those merged in a pre-Pol position, and those merged post-Pol.We claim that distinguishing Modal1 and Modal2, and postulating the threevarieties of negation, makes sense of a large number of otherwise messy or un-accountable pieces of data relating to the modals in English. The analysis isincompatible with Cinque’s (1999) claim that epistemics are merged in a fixedsyntactic position higher than that available for deontics; we reinterpret hisdata as a cognitive requirement that an epistemic cannot be in the local scopeof a deontic. We showed also that there exist PF words such as can’t and noth-ing which may be associated with two non-adjacent LF-interpretable items. Weshow that if the relation between the LF-interpretable and the PF-interpretableparts of a sign is established by checking rather than movement, then there isa straightforward account of these possiblities. We expect that investigation ofother languages along the same lines will confirm the explanatory value of ourhypotheses.

Notes

* We are grateful to Joan Beal, Virginia Brennan, Keith Brown, Sophie Cormack, DorothyDisterheft, Jan Engh, John Harris, Larry Horn, Vieri Samek-Lodovici, Steve Nagle, FrankPalmer, Anna Papafragou, and two anonymous reviewers, for various forms of help withthis 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 kindargued 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 inCormack 1999).

. The same interpretations are obtained when haber, which tends to bias the readings, isnot present (Carme Picallo, pc to Frank Palmer). The negation of el lladre pogui entrar per lafinestra will have the reading ‘the thief was not able to come in by the window’ if no is beforethe modal, and the reading ‘it is possible that the thief did not come in by the window’ if thenegation is after the modal. However, la llet no pot bullir en aquestas condicions can have the


reading ‘the milk may not boil in these conditions’. This is consistent with Picallo’s claimsprovided that the negation is Echo[NEG]. See further in Cormack and Smith (2000).

. Abnormal tags may indicate irony, polite or threatening requests for confirmation, andother marked interrogative uses: “It was going to be sunny today, was it?”, “You’ve bookedfor 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 functionalheads, and some to be adjuncts. However, we do not see this as a problem. Both are heads (wereject specifiers – see Cormack 1999); both have the same semantic content. The only realdifference is optionality: functional heads are obligatorily projected; optionally projectedheads are instantiated as adjuncts.

. Alternatively, suppose the temporal adverb (or a default existential) introduces quantifi-cation over times, while the Tense node introduces a restriction on these. Such a restric-tion 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 ispresumably false. Further, with Modal1 have to, the is intuitively within the scope ofTense. John often had to sneeze does not mean ‘there exist many times which necessarily arein the past and at which John sneezes’; nor even ‘there exist many time in the past at whichnecessarily John sneezes’. it means ‘During the past, there were many times at which neces-sarily John sneezed’. This corresponds to the scope order T Adv Modal, which is what we areclaiming to be correct.

. Auxiliary need nowadays seems to be licensed in some, but not all, the usual English NPIenvironments: *You need leave; You needn’t leave; Need I leave? but *Everyone who I need seemust 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. Wehave not in this paper considered whether there should be a Mood head accounting for thecan/could contrast where it is not one of tense (and similarly for should, would). There isevidence (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 “neu-tralised” in the case of all senses of will. We disagree.

. For Coates, only deontic must is claimed to be exceptional. Why do the discrepancies be-tween our positions arise? Consider first the deontic readings of the necessity modals, whichin effect she claims belong under Modal2, where we claim they fall under Modal1. First, un-derstandably, there is some difficulty in ascertaining the scope facts for the modal readingsof 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, thoughagain Coates recognises a problem. Coates (op. cit. 239) gives for ‘I ought not to/should not


do x’ alternate paraphrases of ‘It is not advisable for me to do x’ and ‘it is advisable for menot to do x’. However, the first, although having a surface order ‘not . . . advisable’ is itselfambiguous as to its explicature (because of ‘neg raising’ effects). The paraphrases we sup-plied in (4) are unambiguous, and the wide-scope negation reading is clearly excluded. Theother area of disagreement is over the epistemic readings of possibility modals, which forCoates are Modal1, and for us, Modal2. We disagree over can and could. Although Coatesdiscusses epistemic can’t (op. cit. 20), she omits epistemic can/can’t from her table. Epis-temic can is of course entirely possible: (i) Errors can occur at this stage of the process. Thiscombination of errors and omissions is sufficient for her to have found English to be di-vided mainly on the epistemic/deontic line, rather than, as we have argued, mainly alongthe necessity/possibility line.

. The meaning could be given as λp[(CAN p)], where p is a variable over predicatemeanings (compare Gazdar et al. 1982, rule 15). However, for the case where the adverb ispresent, even if we could have the adverb selected by the modal, the latter would have tohave a meaning λaλp[(a (CAN p)], where a is a variable over adverb meanings. Theconsequent disjunctive lexical entry is bizarre.

. We do however assume that the internal structure of a complex PF-item is generatedin an independent Morphophonological component, as has been proposed by Anderson(1992) and others. That such a component exists independently of syntax is supportedby the dissociation exhibited by Christopher, a savant, whose second language learningexhibits poor syntax (little changed from his native English) along with strikingly goodmorphological acquisition (Smith and Tsimpli 1995).

. We claim then that the situation with respect to Tense[], say, is the same aswith Aux[] have: it checks for the occurrence of particular Infl associatedwith the following verb. The difference is simply that Tense[ ⁄-] itself, unlikeAux[] although bearing semantic content, is not phonologically overt.

. A reviewer suggests that we consider the option of merging Pol either above or below amodal. 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 sub-stantive to the semantics, its presence can be accounted for only if it is obligatory. We assumethat 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, thiswould have to be relativised to the kind of negation (i.e. Pol[NEG] but not Adv[NEG] orEcho[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, asevidenced by the Klima tests (for use of a subset of the tests in a variety of languages, see deHaan 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’ betweenmay/might and a following can/could.

. Note that Bailey’s use of Mod1 and Mod2 refer to the PF positions of the modals inrelation 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’


(Bailey 1966:141)), we take it that Bailey’s Mod1 are modal auxiliaries, and her Mod2 areeither 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 tocome’ (lit. ‘he should must can come’). If Dutch modals are Modal/Aux, then there must beiteration at Modal1 or Modal2 here. Note however that iteration is also possible in principlewith 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 havemodals whose classification is unpredictable on the basis of their semantics.

. We are grateful to Keith Brown for insisting on our taking seriously the comparablescope readings in Brown (1991).

. The scope order ‘ ₍ . . .’ would indicate for a deontic reading that there isnever an obligation . . . Such a reading is not obtainable.

. Linguistic evidence for echoic negation is also given by Espinal (1993) for Catalan andZanuttini (1997a:67–104, 1997b:228) for Italian dialects. Zanuttini also argues that the twokinds of negation are merged in distinct positions.

. For an explanation, see Cormack and Smith (1998, 2000), where we argue that theunderlying reason is the same as that which drives raising of a finitely inflected modal orauxiliary 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’) seeRullman (1995). Note that the separation of a determiner into two logical operators whichtake scope in distinct positions is required also for some instances of wh-phrase reconstruc-tion (see for instance discussion of How many books does Chris want to buy? in Lechner1998:280). Note that no alternative form with ‘∀¬’ replacing ‘¬∃’ is possible here. Conse-quently, analyses of n-words postulating that they are universals such as that in Haegemanand Zanuttini (1996:122) are probably incorrect. The authors recognise other problemswith the analysis (p. 126 and footnote 20).

. See also Rowlett (1998:124–125).

. The discussion usually includes modal adverbs, and generally makes more distinctionsthan 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 itneeds a tensed proposition for evaluation. However, if Tense is a binder of a tense-variableemanating from V, or is an operator on such a variable, as we assume (see Section 2 andfootnote 6), then Cinque’s conclusion does not follow.

. Capitalisation marks the main sentence stress. If the modal is stressed, Echo[NEG] read-ings will generally be obtained. If a transitive verb is used, and stress falls on the nounphrase complement, contrastive focus (also frequently associated with Echo[NEG]) maybe induced.

. If have to is a main verb, the can here will be in an embedded clause.


. If the functional projection were obligatory, then there would have to be phonologicallynull and semantically trivial elements of these categories available when no overt modalis present. These would have to have ad hoc properties to ensure for example that a nullModal1 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 samemay 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 combinationsdo not seem to be limited in this way. This is consistent with the authors’ analysis of thesemodals as main verbs.

. Cormack and Smith (1998) provides independent evidence that Pol[POS] is presentwhen Pol[NEG] is absent. Arguments include the need to account for the displacementshown in (28b).

. Note that in many languages, it is hard to tell apart a modal auxiliary with possible non-finite forms, and a raising verb. We would take as evidence for auxiliary status (i.e. that themodal is a minor category, not a lexical category) the existence of pre-Pol scope. There doesnot seem to be any good reason why a verb should not select for Pol, as it may for otherfunctional heads such as C or non-finite T, perhaps. But such a verb would still be under thescope of the clause’s main Pol, on the reasonable assumption that Pol is selected by T, and Vis selected by Pol. The existence of NPI modals is also suggestive, since they may be used toexpress an otherwise unobtainable ‘Pol[NEG] [Modal1 . . .’ scope.

. It seems clear from such examples that deontic modals (with the possible exceptionof ‘ability’ can) do not project an external theta role, but rather that the apparent role isidentified pragmatically. The basic compositional type for any modal is thus <t,t>, allowinga status either as a semantic operator realised as an optional syntactic adjunct, or as a Raisinghead.

. In the very rare cases where it might be, it seems that deontics with scope over epistemicsmay actually be possible. Consider a context where if the Knave of Hearts feeds Alice someof the liquid in the bottle, she will necessarily become small. The Queen of Hearts mightstipulate, with the deontic addressed to the Knave of Hearts, “No, Alice may not have to besmall today.”

. The example in (75) has no normal interpretation with deontic may. If the sentenceis interpretable, the seem, which is itself normally an epistemic predicate has to be re-interpreted 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.


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: OxfordUniversity Press.

Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. Cambridge MA: MITPress.

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.


Lasnik, H. (1995). Verbal morphology: Syntactic Structures meets the Minimalist Program.In H. Campos and P. Kempchinsky (Eds.), Evolution and Revolution in LinguisticTheory: Essays in Honor of Carlos Otero (pp. 251–275). Washington DC: GeorgetownU.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 pos-

sibility: 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 andO.J. Westvik (Eds.), Modality in the Germanic Languages (pp. 134–149). Berlin: Moutonde Gruyter.

Papafragou, A. (1998). The acquisition of modality: implications for theories of semanticrepresentation. 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. LinguisticInquiry, 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 theEnglish language. London: Longman.

Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements ofGrammar: 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. Stav-rou (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.


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 RomanceLanguages. Oxford: Oxford University Press.

Zanuttini, R. (1997b). Negation and verb movement. In L. Haegeman (Ed.), The newcomparative 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 ofmodality*

Zygmunt FrajzyngierDepartment of Linguistics, University of Colorado at Boulder

The purpose of this paper is to show how the coding means available in alanguage interact in the coding of various modalities. It is shown in thispaper 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 deonticmodalities: all epistemic modalities are minimally coded by the indicativeform of the verb, while all deontic modalities are coded by the imperativeform of the verb. Auxiliary verbs and modal particles code varioussubdomains within epistemic modality, while differing word orders code thedifferent types of deontic modalities.

. Introduction

The purpose of this paper is to show how the coding means available in a lan-guage interact in the coding of various modalities. The terms ‘modality’ and‘modalities’ are taken to mean the speaker’s attitude or attitudes toward his/herown current utterance (cf. Palmer 1986). Epistemic modality has to do withthe 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 hypotheticalmodality, coding the speaker’s belief that the proposition may be true; dubita-tive modality, coding the speaker’s doubt in the truth of the proposition; inter-rogative modality, coding a question about the truth of the proposition; andnegative modality, denying the truth of the proposition. Deontic modality, onthe other hand, codes the speaker’s wishes with respect to the proposition; sub-modalities in this domain may include the imperative, the subjunctive, and theprohibitive modalities.

Zygmunt Frajzyngier

The systems interaction approach involves discovering the functional do-mains coded in a language and determining how each functional domain iscoded. A functional domain is characterised by a functional feature that is com-mon 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., codingthe role of a fragment in a text. Within a given functional domain there are var-ious 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 existother, as yet unidentified, coding means. The coding means of the subdomainsof a given functional domain interact in complementary fashion.2

The data for the present paper come from Lele, an East Chadic language ofthe Afroasiatic family, and the discussion is limited to the epistemic and deon-tic modalities.3 The epistemic modality comprises the following subdomains:indicative, hypothetical, dubitative, interrogative, and negative. The interrog-ative 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 verbsand modal particles all participate in the coding of modality. Verbal inflectioncodes the distinction between the epistemic and deontic modalities: all epis-temic modalities are minimally coded by the indicative form of the verb, whileall deontic modalities are coded by the imperative form of the verb. Auxiliaryverbs 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 modalitycoding may occur in the same clause, providing evidence that the two domainsare, in fact, separate.

. Inflectional modal categories

In Lele, the indicative and imperative forms of the verb are derived directlyfrom the root, which itself is not marked for modality, rather than from oneanother. In other words, neither the imperative nor the indicative may be con-sidered basic. This is interesting in view of the fact that, in a great many lan-guages, the unmarked value of the indicative clause is to code the speaker’sbelief in the truth of the proposition; given this function, one might expect the


unmarked, or basic, form of the verb to code the speaker’s belief in the truth ofthe proposition, yet this is not the case.

Deriving the inflected verb form from the root in Lele involves complexphonological processes. I provide a few of the simpler derivations here, in orderto illustrate the differences between the indicative and the imperative forms.

The starting element in the derivation of the verb is the verbal root. Amonosyllabic verbal root may consist of a vowel or a consonant followed bya vowel. If the root is polysyllabic, it consists of all vowels and consonants upto the last consonant. The underlying vowel is selected from the set consistingof 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)

'ìskirb

‘pass liquid’‘think’

The first step in forming the imperative from the root is to add a vowel suffixafter the final consonant. If the root vowel is a or i, the imperative suffix isa; if the root vowel is u, the imperative suffix is u. The final vowels a and uare subsequently deleted from polysyllabic verbs if the last consonant of theverb can be word final. The formation of the imperative stem is summarised asfollows:

Step 1:Step 2:

Imperative stem(polysyllabic verbs)

= Root + a/uFinal 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 C1C1 is re-duced to C. After these steps, the past tense suffix i is added to the root. If theroot contains a high round vowel, this vowel is fronted. The past stem vowelis deleted from any verb whose root-final consonant may occur in word-finalposition. The indicative stem is the source from which the nominal and futureforms of the verb are derived. The nominal form is derived by lowering the paststem final vowel i to e. The future tense form is derived by raising to high thefirst tone of the nominal form.

Following are examples of surface forms derived from monosyllabic andbisyllabic roots with a high round vowel:

Zygmunt Frajzyngier

Monosyllabic roots

Root Indicative/Past Gloss Future Nominal Imperative

ì è ‘go’ éè è àpi pe ‘dig’ péè pe panì nè ‘make’ né nè nà

Polysyllabic roots


'ùs 'ìsì ‘pass liquid’ 'ísè 'ìsè 'ùsùkurb kirbi ‘think’ kírbe kirbe kurbugùrb gìrbì ‘forget’ gírbè gìrbè gùrbùdùgr dìgrì ‘kill’ dígrè dìgrè dùgrùkunjl kunjili ‘learn’ kúnjíle kunjile kunjuluwunjr winjiri ‘rot’ wínjíre winjire wunjuru

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


sàg sàgì ‘tear’ ságè sàgè sàgàbàs bàsì ‘fail, spend’ básè bàsè bàsàdas´ dasí ‘husk’ dáasé dasé dasámàgl´ màglí ‘repair’ máàglé màglé màglá

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

Past Gloss Future Nominal Imperative

gìl ‘show’ gílè gìlè gùlìm ‘bury’ ímè ìmè ùmim ‘rot’ íme ime umjìb ‘stick’ jíbè jìbè jùbwìl ‘glean’ wílè wìlè wùlgul ‘watch’ gíle gile gulkil ‘sell’ kíle kile kul'íl ‘drag’ 'íìlé 'ìlé 'ùlkìn ‘return’ kínè kìnè kùn


wìl´ ‘warm up’ wílè wìlé wùlhím ‘gather, tr.’ hímé hìmé 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’ andwunjr ‘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 moti-vated phonologically because the future and the nominative forms of the verbdo not reduce their final vowels. Thus the vowel reduction is a coding meansin addition to the suffixes i, u and a. The final vowel retention in the indicativeand the imperative is, however, motivated by the phonological constraint onthe types of consonants and consonant clusters that may occur in word finalposition. Note that stem reduction is one of the characteristics of deontic in-flectional coding in Semitic languages, which constitute another branch of theAfroasiatic family.

. Epistemic modality

All constructions with epistemic modality are based on the indicative form ofthe verb. Some submodalities in this domain also are marked by auxiliary verbsor modal particles.

. Indicative modality: Coding belief in truth

The least-marked submodality of the epistemic domain is the indicativemodality. 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 Leleindicative clause, a nominal subject occurs before the verb:

(1) kudo-ro-]husband-3-

jè

1e]líhear

koloword

ká 'nyécertain:

kwánioutside

ná [pause]

yàásay

ísa]nothing

1é

‘Her husband heard certain things outside, but he did not say anything.’

First- and second-person pronominal subjects also occur before the verb:

(2) ]1

ómè-ycatch:-3

‘I will catch him.’

Zygmunt Frajzyngier

(3) gi2

'álèsuffice:

go

òmé-ycatch:-3

‘You will be able to catch it.’

A third-person pronominal subject, if overtly coded, occurs after the verb:5

(4) càgspread

dí3

kùbàròblood

kúsìyo]ground

‘He spread blood on the ground.’

(5) tamáwoman

tolo

láypretty

dú3

‘That woman is pretty.’

(6) golsee

gé3

dú3

‘They saw her.’

(7) golsee

dú3

gé3

‘She saw them.’

A third-person subject need not be coded overtly if it is the same as the subjectof the preceding clause. In the following sentence, the first instantiation of thethird-person masculine subject is unmarked, since this subject is the same asthat of the preceding clause (not shown here). The second instantiation of thethird-person masculine subject is marked overtly because another subject, thethird-person feminine, interferes between the two instantiations of the third-person masculine (cf. Frajzyngier 1997):

(8) 1e-duleave-3

ja]quietly

jè

wèlsleep

dú3

kunásleep

ná

anleave

dí3

ègo

omcatch

dùbàngúsheep

yó-rómother-3

ná

mundusilence

wàlslaughter

‘He left her quietly while she was asleep, left the room, silently caught anold sheep and slaughtered it.’

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


(9) kírbìKirbi

habfound

kùlbácow

kò-rò-3

wa]really

‘Kirbi has certainly found her cow.’

Such means are only used, however, if someone has previously denied the truthof the proposition or expressed doubt about the truth of the proposition (cf.Frajzyngier 1987).

Language-internal evidence for the modal function of indicative modalityis 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 anindicative clause whose subject is the speaker cannot be followed by a dubita-tive marker. The dubitative marker sá], freely used to code the speaker’s doubtas to the truth of another person’s statement, cannot be used in a clause whosesubject is the speaker (cf. also Frajzyngier 1985); compare (10) with (11).

(10) a. *]1

nè

hómyàsick

sá]

Intended meaning: ‘I am sick [but I doubt it].’b. *]

1

yásay

na

]1

nèmake

hómyàsick

sá]

Intended meaning: ‘I said that I am sick [but it isn’t true].’

(11) a. nèmake

dí3

hómyàsick

sá]

‘He pretends to be sick.’b. na-y

-3

hómyàsick

sá]

‘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 astatement. One means involves the verb tòb ‘want’, followed by the comple-mentiser go and then by the main verb with its complements. The complemen-tiser go is glossed REF for ‘referential’ in order to distinguish it from the dedicto complementiser na which is used as the hypothetical modality marker.The marker na is glossed as COMP if its function is to code the complementof verb of saying, and it is glossed as HYP if its function is to code hypotheticalmodality 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, asargued for in Frajzyngier (1991).

(12) kus-iybody-3

jè

tobwant

go

né-ymake-3

‘He seems to be sick.’

Cf. the indicative clause:

(13) kus-iybody-3

nè-ymake-3

‘He is sick.’ (lit. ‘his body makes him.’)

The speaker may hedge about the truth of his or her own statement by usingthe verb nom ‘to pretend, to court’, which forms a serial verb construction withthe following verb. This means occurs in a few expressions:

(14) kus-íybody-3

nomcourt

né-ymake-3

‘He seems to be sick.’

(15) tùwasun

nompretend

dóbeset.down

í (índì)with.him

màníthere

‘The sun seems to lie down there.’

Compare the use of nom as a main verb:

(16) cànìgéCanige

nomcourt

kírbíKirbi

‘Canige courts Kirbi.’

The adverb brùm ‘hardly’ expresses the speaker’s uncertainty regarding his orher perceptions:

(17) kanyasomething

nomí-ypretend-3

brùmhardly

‘There is something wrong with him, I believe.’

(18) ]1

golsee

kurspace

brùmhardly

‘I can hardly see. ’

Doubt about the truth of the speaker’s perceptions or evaluations may also becoded by the modal adverb wa] ‘really’ followed by the conditional gèrè ‘if ’:

(19) kírbìKirbi

habfound

kùlbácow

kò-rò-3

wa]really

gèrèif

‘Kirbi has apparently found her cow.’


. Hypothetical modality

Hypothetical modality is coded by the hypothetical marker na preceding thepredicate:

(20) kiyaKiya

na

bòybreak

kòjòhoe

kè-y-3

‘It is possible that Kiya broke his hoe.’

If the subject is a third-person pronoun, which follows the verb in the indicativeclause (see above), the complementiser na precedes the verb:

(21) na

bòybreak

dí3

kòjòhoe

kè-y-3

‘It is possible that he broke his hoe.’

With a first- or second person pronominal subject, the hypothetical markerprecedes the subject pronoun:

(22) na

]1

bòybreak

kòjòhoe

kò-nò]-1

‘It is possible that I broke my hoe.’

(23) na

gi2

bòybreak

kòjòhoe

kò-m-2

‘It is possible that you broke your hoe.’

Compare the following fragment from natural discourse with its counterpartfrom which the hypothetical marker is omitted. The third person subject is thetopic, hence the presence of the pronoun dày at the beginning of the clause:

(24) dày3

ná

na

nè

tìbréashes

maynaceiba tree

‘That person would be like ashes of a ceiba tree.’

(25) dày3

ná

nè

tìbréashes

maynaceiba tree

‘That person is like ashes of a ceiba tree.’

Hypothetical modality may be coded in matrix and embedded clauses. Theverb following the hypothetical marker may be in the future or the impera-tive form. The future form of the verb combined with the hypothetical markercodes potential events:

Zygmunt Frajzyngier

(26) dày3

kó]certain

na-y-3

kálèclimb:

jénèsit:

dà

lu]bahorse

kùrmbàlòchief

kè-y-3

ni

na-y-3

bórècut:

sàríyàverdict

na

ùs-ìyconcern:-3

1é

‘The other [said] that if he could climb on the chief ’s horse and pronouncejudgments, [they could kill him], he would not care.’

(27) Oregoat:

na

wè-gégive birth-3

dà

gìdìrèmoon

kòlòng:

ni

ná

ge

yàátell

na

karan-dochildren:3

gàn-di-gèleg-.-3

na

wèybecome

já.side

‘It is said that if goats give birth in that month, their kids will havecrooked legs.’ (G-C & W 1981:54–55)

Hypothetical modality codes an event as not being real. This may apply to thetruth value of the clause:

(28) kurplace

wèlpass

kayfinish

1é

'eystill

ná

tamáwoman

na

du3

ègo

sógútoilet

ni

‘Before the day ended [while it was still dark] the wife pretended that shewas going to the toilet.’

Hypothetical modality may also indicate that an event was intended but didnot occur:

(29) kurtime

go

gúnyéspider

ágì-ytake:-3

jè

na-y-3

ègo

jéèé-ythrow-3

dà

kama-]water-

ni3

dày kàyo-]squirrel-

se

anleave

ná

galmbobag

kíin-dìhole-3

bà

se

gìrrun

‘When Spider was about to take him to throw him into the water, Squirrelleft through the hole in the bag and ran away.’

(30) bayndíperson

go

na

aytake

tugoat

ká]

na

lèéeat

ná

‘The person who would take this goat and would eat it.’

(31) kàyosquirrel

se

yàásay

na

na-y-3

sìnàknow:

máàní

na-y-3

1eleave

gúnyéspider

na

ge

ímè-ybury-3

ná

làmndáelephant

yé-ymother-3

1é

ba


na-y-3

bòrsave

gúnyéspider

gèyl-ìyhealth-3

ba

gúnyéspider

na

ègo

jè

tòbwant

go

dígrè-ykill:-3

gà

‘Squirrel said that if he had known that, he would have left Spider tobe buried together with the mother of elephant. Here I saved him, anddoesn’t he want to kill me?’

(32) póìPoi

kúsíge-]Kusige-

se

ma'ydeny

na-y-3

kòysteal

tèmlé-]corn-

1é

na-y-3

ànleave

kwánioutside

ná

kúmnoGod

kam-diwater-3

na

1é

‘Poi Kusige denied that he stole the corn. He claimed that he did not goout in the rain.’

The hypothetical marker na after a volitional verb also codes a hypotheticalevent. Here, the marker must be followed by a third-person subject pronoun:

(33) wèlèday

pìnàone

ná

gèywant (Ng.)

na-y-3

gísè-dùtry:-3

‘One day he wished to test her.’

If the complementiser is absent, the meaning of the clause is non-hypothetical:

(34) wèlèday

pìnàone

ná

gèywant (Ng.)

gísè-dùtry:-3

‘One day he wanted to test her.’

Natural texts clearly show that speakers try to avoid making categorical state-ments when explaining natural or cultural phenomena. Instead of using astraightforward indicative sentence, the speaker presents the statement as theindirect speech of an unspecified human subject, coded by ge ‘one’ or kara ‘peo-ple’. Here, the matrix clause has the form ge yàá or kara yàá ‘one says/peoplesay’. In the texts gathered by Garrigues-Cresswell, all explanations of human oranimal behavior or atmospheric phenomena have this form. The complementof the matrix clause is marked as being hypothetical by the particle na, followedby the indicative form of the verb:

(35) Tùmòpast time

go

karpeople

di-nì.-1.

kínyé

kè-gè-3

ge

yàásay

na

nè

'òdumonkey

ba

gèylì-nìsave-1.

kurtime

go

kamdawomen

ná

kùlòngò.pregnancy

Zygmunt Frajzyngier

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

The scope of the hypothetical marker na depends on its syntactic position. Ifthe hypothetical marker precedes the subject, the whole proposition is in itsscope, as illustrated in the next sentence from the same fragment:

(36) KumnoGod

yàásaid

na,

na

láràdíchameleon

ba

ègo

hab-ifind-3

tùmòfirst

na

yàátell

bè-ì-3

koloword

kè-ì-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 gram-matical coding of the speaker’s doubt concerning someone else’s claim. Thisis coded by the clause-final particles sá], glossed as DUB for ‘dubitative’, andgèrè ‘if ’. These particles may occur alone or in the sequence sá]gèrè. Dubitativemarkers occur with verba dicendi complements but they can also occur withother verbs. The dubitative marker denies the truth of someone else’s claims,as in (37) and (38), to be compared with (39).

(37) nèmake

dí3

hómyàsick

sá]

‘He claims to be sick.’

(38) na-y-3

hómyàsick

sá]

‘He said that he is sick, but I doubt it.’

(39) kus-iybody-3

nè-ymake-3

‘He is sick.’ (lit. ‘his body makes him’)

It may be recalled that the dubitative marker cannot be used with a first-personsingular subject. The dubitative marker also cannot be used in equational sen-tences:


nè

leleLele

‘Canige is Lele.’


Cf.

(42) *cànìgéCanige

nè

leleLele

sá]

Intended meaning: ‘Canige is Lele, but I doubt it.’

However, if the equational clause is a complement of a verb of saying, thedubitative marker is acceptable:


ná-y-3

no

go

leleLele

sá]

‘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, thesubjunctive, and the prohibitive, which is the negative imperative. These aregrouped together because they all use the imperative form of the verb. Thedifference between the imperative and the subjunctive subdomains is codedsyntactically, 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 followsthe verb, while in the indicative, the second-person subject precedes the verb.Complements, if any, follow the second-person singular subject marker:

(44) pàmàsearch:

ngú2

‘Search!’

(45) kùlùbuy:

ngú2

kùlbácow

‘Buy a cow!’

(46) daeat:

ngù2

kasacorn

ná

jáside

kolo]:

‘Eat the corn at the other side.’

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

Zygmunt Frajzyngier

(47) kùlbuy:

kùlbácow

‘Buy () a cow.’

(48) À-jego:-

ná

bè-ng-1

gúndùgourd

pìnàone

ná

kama,water

ná

gìlàdístalk

mìsé,sorghum

à-jego:-

haba-ngfind-1

ná

dà

sóòyótamarind

to-roòbottom-3

ni.

‘Bring me a gourd with water and stalks of sorghum. Come to find meunder the tamarind tree.’ (G-C & W 1981:2–3)

If the second-person singular subject is marked, it is marked by the indepen-dent form of the pronoun in the position before the verb:

(49) Dàmè2

1inglálisten:

‘Listen!’ (G-C & W 1981:2–3)

. The subjunctive

The term ‘subjunctive’ refers to wishes with respect to the first and third personas well as non-imperative wishes with respect to the second person. The syntaxof the subjunctive differs from that of both the indicative and the imperative.In a subjunctive clause with a first-person plural inclusive subject, the subjectpronoun ngà follows the imperative verb. Any complements follow the subject:

(50) yà-ngátalk:-1.

koloword

go1à]another’

‘Let us talk about something else.’

(51) ìrà-ngágo:-1.

‘Let us go!’

(52) kà[w]go:

ngà1.

túghome

kò-m-2

ni

gólèsee:

kùníhome

kùlè-ndìinterior-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 pre-cedes the verb. In clauses with indicative modality, it may be recalled, thethird-person pronoun follows the verb:


(53) dí3

ìràgo:

‘Let him go.’

(54) dú3

ìràgo:

kàsúgùmarket

‘Let her go to the market.’

(55) dí3

yàgàplough:

bé-]-1

kúsófield

‘Let him plough my field for me!’

If a subjunctive clause has a third-person nominal subject, this subject is fol-lowed by a third-person pronoun agreeing with the nominal subject in genderand number:

(56) kiyaKiya

dí3

àgo:

jè

‘Let Kiya come.’

(57) kiyaKiya

dí3

ìràgo:

‘Let Kiya go.’

(58) kiyaKiya

dí3

yàgàplough:

bé-]-1

kúsófield

‘Let Kiya plough my field for me!’

(59) yàátell

bè-ì-3

na

gìdìrèmoon

na

madie:

ná,

ìràgo:

ná

kìrèroad

kínditruly

karapeople

índùwéhuman:

madie:

gé3

ná

analeave:

gé3

jè.

‘[he] told him that the moon should die and never return, and the peopleshould die and return.’ (G-C & W 1981:8–9)

. Prohibitive modality

Prohibitive modality is formed through the placement of the negative particle1é at the end of a clause containing the imperative form of the verb:

(60) Tamá,woman,

ulcry:

1é.

‘Woman, do not cry!’ (G-C & W 1981:2–3)

Zygmunt Frajzyngier

(61) ladátouch:

bìsíantelope

kò-nò]-1

ná

kà-mhand-2

1é

‘Do not touch my antelope with your hands!’

(62) ìràgo:

1é

‘Don’t go!’

(63) àgo:

jé-ngú:-2

1é

‘Don’t come!’

The prohibitive predicate may have two verbs in a sequence, both of themmarked for imperative modality:

(64) 1áleave:

walaswear:

kusu-mbody-2

bàball

1é

‘Do not swear for nothing.’

. Coding hypothetical and deontic modalities in the same clause

The hypothetical marker na, which codes the hypothetical subdomain of theepistemic modality, may co-occur with the imperative form of the verb, markerof the domain of deontic modality. This construction codes both hypotheticalmodality and the modality of obligation, providing evidence that epistemic anddeontic modalities are two separate domains. The possible combinations of nawith the future i.e. a form of indicative modality and imperative forms of theverb, are illustrated in the following fragment:

(65) KumnoGod

se

teycall

láràdílizard

dàgè3

ná

gúlóchameleon

yàátell

bè

karapeople

‘God called Lizard and Chameleon and he told people.’

(66) na-y-3

ódèleave:

wanyangtomorrow

wáyngulúmorning

ná

láràdílizard

na

‘that he would go tomorrow morning and that Lizard . . . ’

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


(67) nèmake:

karapeople

go

yàsay

gé3

na

gìdìrèmoon

na

madie:

ìràgo:

ná

kìrèroad

kínditruly

kègè,:3

‘would be [a spokesman] for people who say that the moon should dieand go away forever.’

(68) gúlóchameleon

na

nèmake:

karapeople

go

yàsay

gé3

na

karapeople

índùwéhuman:

ba

na

madie:

gé3

ìràgo:

gé3

ná

kìrèroad

kíndiètruth

kè-gè-3

lay.also

‘Chameleon would be a spokesman for people who say that humansshould die and go away forever.’

. Conclusions

Epistemic modality is coded by the indicative form of the verb. Subdomainswithin the epistemic modality domain are coded by auxiliary verbs and parti-cles.

Deontic modality is coded by the imperative form of the verb. Subdomainswithin the deontic modality domain are coded by the order of pronouns withrespect to the verb. In the imperative modality, the second-person plural sub-ject pronoun (obligatory when the addressee of the imperative is plural) fol-lows the verb, while in all epistemic modalities, the second-person subject pro-noun precedes the verb. Third-person pronouns in the subjunctive modalityprecede the verb, while first- and second-person pronouns follow the verb; thisorder is the reverse of that found in clauses with epistemic modality.

Evidence that epistemic and deontic modalities are distinct domains isprovided by the fact that they can be combined within the same clause.

Abbreviations

1 First-person2 Second-person3 Third-person Allative

Zygmunt Frajzyngier

Anaphor Associative. Used for associative preposition ná

and associative pronominal marker ín-. Collective Comment clause marker ba Copula Dative marker Definite Demonstrative Dubitative Exclusive FeminineFr. 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 markerNg. 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


Notes

* The work on Lele has been supported by grants from the National Endowment for theHumanities, the National Science Foundation, the American Philosophical Society, and theUniversity of Colorado. The work on this paper was completed while I was a guest of theMax Planck Institute for Evolutionary Anthropology in Leipzig. I am most grateful to theseinstitutions for their support. I am also grateful to Erin Shay and two anonymous refereeswhose comments on both the content and the form of the paper were most helpful. I aloneam, 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 indicativesentence. While it is true that there are languages in which the speaker’s assertion of thetruth of the proposition is a marked modality, this is not the case in Lele, the language onwhich 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 Garrigues-Cresswell and Weibegué (1981). If the source of the data is not mentioned, this indicatesthat the data come from my own fieldnotes, which consisted in most cases of recordingnatural discourse. Data from Garrigues-Cresswell and Weibegué (1981) are identified by theabbreviation 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 inter-action 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 mathematicallinguistics. In C. Martin-Vide (Ed.), Mathematical and Computational Analysis ofNatural Language (pp. 295–310). Amsterdam: Benjamins.

Zygmunt Frajzyngier

Garrigues-Cresswell, M. and C. Weibegué (1981). Livre de lecture lélé. Sarh: Centre d’étudeslinguistiques.

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 developmentand autism*

Anna PapafragouUniversity of Pennsylvania

It is widely assumed in the developmental literature that certain classes ofmodal 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’sdeveloping theory of mind, i.e. the ability to attribute to oneself and othersmental representations, and to reason inferentially about them. It ishypothesised that epistemic modality crucially implicates theory of mindabilities and is therefore expected to depend on prior developments in thechild’s ability to handle representations of mental representations. In supportof this hypothesis, it is shown that autistic individuals (who arguably possessa deficient theory-of-mind mechanism) have difficulty with epistemics.

. Introduction

It has long been recognised that modal items in natural language are context-dependent expressions. A sample of possible interpretations for some Englishmodal 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 litera-ture in terms of the distinction between root and epistemic modality. Rootmodality includes obligation, permission and related notions (traditionallyknown as deontic modality), as well as a family of concepts dealing with inten-tion/willingness and ability (traditionally known as dynamic modality): exam-ples (1a), (2a), (3a), (4a), (5a) and (6a, b) above are typical cases of root modal-ity. Epistemic modality involves inference from known premises: the examplesin (1b), (2b), (3b), (4b), and (arguably) (5b) fall under the category of epis-temic modality. Root and epistemic modality are based on different premises.As Angelika Kratzer puts it, ‘if we use an epistemic modal, we are interested inwhat else may or may not be the case, given everything we know already’; onthe other hand, with non-epistemics, ‘we are interested in what can or musthappen, given circumstances of a certain kind’ (Kratzer 1981:52).

An interesting fact, which has been repeatedly pointed out in the psy-cholinguistic literature, is that root meanings precede epistemic ones in lan-guage acquisition. Since in many languages the same class of modal expressionsis used to convey both root and epistemic meanings (as is the case with theEnglish modals above), the claim is that modal items initially appear in childlanguage with root interpretations only and their epistemic interpretations areacquired markedly later. In this paper, I survey the evidence (both longitudinaland experimental) standardly adduced to support the acquisitional priority ofroot over epistemic modal meanings. My main aim is to motivate a specificproposal about the emergence of epistemic modality in language development.According to this proposal, the appearance of genuine epistemic modal uses isclosely tied to the development of the child’s theory of mind, that is, the abil-ity to understand mental phenomena and to attribute mental states to oneselfand to others (see also Papafragou 1998, 2000). I explore the explanatory po-tential of this view with regard to the existing cross-linguistic data on the ac-


quisition of modality and relate these results to developmental findings frommental verbs such as think, or believe. Finally, I show how certain data fromautism 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 stud-ies 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 thesecond and the middle of the third year, initially in restricted syntactic environ-ments (mainly declaratives) and often with a single negative form such as can’t.Gradually, both the range of constructions in which modality appears and thetype of modal item used become more varied.

According to available production data, English-speaking children quiteearly on express intention/desire and ability with will and can, respectively. InWells’ (1979) sample (which consisted of sixty children time-sampled alongwith their mothers every three months from one year and three months tothree years and six months), it was found that by two and a half years morethan half of the sample used can to convey both ability and permission. By thesame time, children used will to communicate intention. In her studies of asingle child, Shepherd (1982) has found that will extends from volition to pre-diction between two years and five months and three years. Will refers to moredistant future or events which lie beyond the child’s control, while gonna takesup 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 threemodal-like verbs have root uses: hafta is used to convey compulsion based ona norm/external source, needta communicates compulsion based on an inter-nal source, and wanna communicates volition. Later on, a fuller repertoire ofroot uses comes into play. Between two years nine months and three years, thechildren in Wells’ (1985) sample used must, have (got) to and should to com-municate obligation and (root) necessity, although these uses stabilised laterin development. As Wells (1985) reports, by the beginning of the third year allcategories of root modality were attested in his sample.

In contrast, the first statements with epistemic modals (must and may) oc-cur in the second half of the third year, i.e. about six months later than rootmeanings. Furthermore, epistemically modalised utterances are extremely rareuntil 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 anepistemic meaning only by three years and three months of age. Use of modalsto convey certainty emerged even later, since by five years only 25% of thesample gave evidence of it. Certain epistemic uses of will (as in That will beJohn, uttered on hearing the doorbell) appear even later than epistemic uses ofmust. Other research suggests that the development of the epistemic modal sys-tem, especially as far as adverbs and adjectives are concerned, continues evenbetween six and twelve years (Perkins 1983).

Young children’s comprehension of modal expressions has also been testedexperimentally. Hirst and Weil (1982) gave children between three and six twodifferent modal propositions of varying ‘strength’ (e.g. with a possibility vs. anecessity marker). In the epistemic cases, the propositions concerned the lo-cation of a peanut. In the root cases, they were commands by two teachersabout the room a puppet was to go to. The child was to indicate in the firstcase where the peanut was, and in the second where the puppet would go. Thestrength ordering assumed was is > must > should > may. The general result wasthat children appreciate the relative strength of epistemic modal propositionsabout a year earlier than root modal propositions (five vs. six years approxi-mately); moreover, the greater the distance between the modals, the earlier thedistinction is appreciated. It should be pointed out that these results cannotbe interpreted as evidence against previous research documenting the devel-opmental priority of root over epistemic interpretations of modals. Children’sperformance in the deontic tasks was probably influenced by factors other thanthe relative strength of the modals (for instance, the evaluation of the author-ity of the persons issuing the command, or the puppet’s compliance with therules). Still, this paradigm shows that children have a good grasp of differencesbetween modals within each major modal class (root or epistemic) by the ageof five or six. Hirst and Weil’s original tests were replicated by Noveck, Hoand Sera (1996) with five-year-olds and Byrnes and Duff (1989) with three- tofive-year-olds.

There is some evidence that the precedence of non-epistemic meanings ofmodal 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 showsthat, when müssen (‘must’) and sollen (‘should’) appear, they have root ratherthan epistemic interpretations. In Polish, expressions of root modality appearbefore two and are used with increasing frequency during the third year tocommunicate obligation/root possibility and prohibition/permission. No epis-temic uses of modals are attested for this period, although during the second


year certain epistemic modal particles emerge – e.g. chyba (‘probably’), na pe-uno (‘for sure’) (Smoczynska 1993). Data from Mandarin Chinese also followthe developmental ordering of dynamic, deontic and epistemic modality (Guo1994). Modal data from Modern Greek, Finnish and Turkish seem to patternin similar ways: Stephany (1986) notes that, in these languages, epistemic in-terpretations appear well after the third birthday. She also points out that a va-riety of grammatical means for marking epistemic modality other than modalverbs (e.g. the conditional in Finnish, the aorist inflection in Turkish) coincidearound the third year.

Even though observational data are not by themselves sufficient to obtainan accurate picture of development, they are a useful point of departure in dis-cussing the acquisition of modality. In the next section, I present the beginningof an explanation for the acquisition facts that draws on the connection be-tween 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 abilityto attribute mental states to oneself and to others, and to use information abouthuman cognition and motivation in understanding and predicting human ac-tion. In other words, adults generally assume that people have a mental life ofinterconnected beliefs, desires, intentions, ideas; they also assume that people’sobservable behaviour stems from and is best understood in terms of the men-tal states which lie behind overt behaviour. The ability to impute mental statesto oneself and others leads humans to try and use the mind and increase itspowers, to share inner experiences, to distinguish between imaginary and realevents, and to interact with other persons by searching for and reaching out totheir underlying mentalities (Butterworth, Harris, Leslie and Wellman 1991a).This everyday conception of the mind (or ‘folk psychology’) is commonly re-ferred to within recent cognitive science as theory of mind (see the contribu-tions 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 pre-occupations of philosophy, current interest in folk psychology resurfaced aftersome researchers raised the question of whether certain primates might alsopossess 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 inwhich the workings of theory of mind are either absent or very different fromthe normal adult case (including the cognition of animals, children and atypi-cal individuals). Recent years have seen a flood of interdisciplinary research ontheory of mind, mainly centred around two main issues:

a. What is the nature, and what are the characteristic properties, of normaladult folk psychology?

b. How do folk-psychological abilities develop in children?

Both issues are surrounded by considerable debate. For instance, adult folk psy-chology has been analysed as a module underwritten by an innate mechanism(Leslie 1988), a truly theory-like object constructed through general-purposelearning capacities (Gopnik and Wellman 1995), or a device for projecting one-self onto other people, thereby simulating their mental states (Gordon 1986).These differences in perspective with regard to the final form and deploymentof theory of mind carry over to developmental work: studying what the inter-mediate stages in folk-psychological development look like typically presup-poses a stance on the structure and function of adult folk psychology. In whatfollows, I present certain core facts about children’s performance in theory-of-mind tasks without adjudicating between competing theoretical proposals.The description is meant to highlight some relatively uncontroversial data con-cerning children’s performance. It is also meant to give some background fordiscussing the data from the acquisition of modality in later sections.

There is some evidence for early understanding of mental life by two yearsof age. Very young children are able to pretend and to engage in pretence along-side others (Leslie 1988). They also have an understanding of the causal linksbetween desire or perception and the world: for instance, they know that de-sires can be used in predicting actions (Wellman and Woolley 1990). Two-year-olds also know that they can produce perceptions in others by showing thingsto them, and by three they can also deprive others of perception by hidingthings (Yaniv and Shatz 1988). By the age of three, children can distinguish be-tween real and mental entities. For instance, they know that, when someone isthinking about a cookie, that cookie cannot be eaten, shared, and so on – un-like the situation in which someone actually has a cookie (Wellman and Estes1986). They also develop a better understanding of the connection betweenperception and belief fixation. They realise, for instance, that if an object ishidden 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).


Between three and four or five years children acquire the ability to passthe so-called false belief task. The essence of the task is to check whether chil-dren 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 oftesting whether children actually attribute beliefs to others or simply assumethat 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 andgoes away. In his absence, the object is moved to another place. The first char-acter comes back and wants to retrieve the object. The question is where theactor will look for the hidden object. In order to pass the test, children have toattribute a false belief to the agent, even though they themselves hold a differ-ent (and true) belief. It turns out that children under four wrongly predict thatthe character will look at the correct (new) place, even though the person hashad no actual access to the information that the object was moved.2

The results from this test generalise across a broad range of related exper-imental tasks. If one shows a three-year-old a familiar candy box and lets herfind out that it contains pencils, not candy, then puts the pencils back and asksher what her friend, who has not seen what is in the box, will think aboutits contents, three-year-olds typically say the friend will think there are pen-cils in the box (Perner, Leekham and Wimmer 1987). In the same task, whenthree-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 suchtasks is considerably better. Other important developments during this periodinclude children’s ability to distinguish between appearance and reality (Flavell1986), 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 ca-pable of successfully handling second-order representations (metarepresenta-tions), that is, representations about their mental states. Moreover, childrenof this age begin to understand that these representations may vary in relia-bility, may change over time and may differ across individuals. After five, de-velopment related to mindreading continues in a number of directions. Forinstance, children appreciate that mental state attributions can be multiply em-bedded (‘John thinks that Mary fears that Beatrice suspects . . .’), and acquiresocial concepts such as ‘duty’ and ‘commitment’ which rely on such complexreasoning.

Anna Papafragou

. Metarepresentation and modal interpretations

. A proposal

I now want to consider the development of epistemic modal interpretationsin the context of the child’s theory of mind. Consider the familiar case of theEnglish modal verbs. When using a verb such as must, may, might, should, will,or ought to on an epistemic interpretation, the speaker relativises possibilityor necessity with respect to her set of beliefs. In (7a), for instance, the speakercommunicates that it follows from her beliefs that there is some mistake; sim-ilarly, in (7b) the speaker conveys that it is compatible with her beliefs that itwill rain tomorrow:

(7) a. There must be some mistake.b. It may rain tomorrow.

It follows, then, that genuine epistemic interpretations of modals presupposethe ability to reason about mental representations – and possibly also the ca-pacity to assess their accuracy in representing the real world and to understandthat they may be revised with time. All of these abilities fall in the domain oftheory of mind.

In this sense, epistemic items closely parallel other so-called mental termssuch 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 men-tal states but are rather conversational devices; for instance, the verb know ini-tially occurs in standardised phrases such as You know what?, I don’t know. Bythe end of the third year, however, their subjects use mental terms with truemental reference (e.g. Before I thought this was a crocodile; now I know it’s analligator – Shatz et al. 1983:309). According to other studies, by the age of fourchildren recognise the greater reliability of know over think and similar predi-cates (Moore and Davidge 1989; see the reviews in Olson and Astington 1986;Moore and Furrow 1991). Recall that epistemic modals also appear by the endof the third year but their first uses are probably incompletely understood: it isnot until the age of four or five that children recognise that epistemic must isstronger than may or should.

A strong piece of support for the hypothesis that epistemic modals andmental terms implicate metarepresentational abilities comes from experimentscarried out by Moore, Pure and Furrow (1990). These researchers showed thatthere is a correlation between the comprehension of the relative reliability ofknow vs. think/guess and the ‘strength’ of epistemic must vs. might on the one


hand, and performance on false belief, belief change and appearance-realitytasks on the other hand. This finding suggests that there is a connection be-tween the ability to distinguish epistemic strength encoded by natural languagewords and the ability to reason about false beliefs in non-linguistic tasks.

The proposal that epistemic modality relies on theory-of-mind abilitiesopens up the possibility of studying the acquisition of modal vocabulary inparallel with the development of the child’s cognitive resources. This is espe-cially valuable since, as we saw, most of what is known about the early seman-tics and pragmatics of modal expressions is based on observational data (and istherefore open to interpretation problems). A promising direction is to use in-dependent cognitive tasks tapping theory of mind as a way to gain insight intothe concepts underlying the use of modal and mental vocabulary. For instance,it would be interesting to compare children’s use and comprehension of ex-pressions of epistemic possibility such as may or might to their understandingof uncertainty or undecidability in non-linguistic tasks.

One area in which this approach can be particularly illuminating is theearly modal lexicon. Even though epistemic modal items are largely absentfrom the speech of two-year-olds learning English, there is some evidence thatin languages other than English some epistemic modal expressions occur be-fore the age of three. Based on longitudinal data, Choi (1995) has argued thatin Korean a set of sentence-ending evidential suffixes is acquired between oneyear eight months and three years, long before the set of epistemic and rootmodal verbs appears. These suffixes indicate the status of the speaker’s be-liefs, in terms of either the degree of certainty/commitment to the belief, orits source (e.g. communication, perception, hearsay). Such data have led someresearchers (e.g. Shatz and Wilcox 1991) to propose a cognitive (theory-of-mind) constraint which would block the acquisition of epistemic predicatesbefore two years and six months in languages such as English but could beovercome by language-specific input in languages such as Korean.

Even though it is not clear what semantic knowledge underlies the useof epistemic expressions in the Korean two-year-olds’ vocabulary (or whetherthese occurrences are truly mentalistic – Papafragou 1998, 2001), these dataraise interesting questions about the relation between linguistic and cogni-tive development. Recall, for instance, that English-speaking children still ex-perience difficulties with non-linguistic tasks involving sources of beliefs un-til around age four (O’Neill and Gopnik 1991). One would therefore want toknow whether these early occurrences of evidentials in child Korean are trulymetarepresentational and, if so, whether their appearance in the productivevocabulary of child language correlates with performance in theory-of-mind

Anna Papafragou

tasks (e.g. false belief tasks, or simpler tasks investigating understanding of be-lief sources). Some work in this direction is already under way (Papafragou andLi, in press).

What about root modal interpretations? Clearly, at least some of these in-terpretations do not presuppose sophisticated understanding of the mind. Itis perhaps for this reason that ability can and volition will appear earlier thanepistemic (or other root) interpretations. Some of the other early uses of rootmodals, such as hafta/have (got) to which appears around three, are predomi-nantly used to state an obligation and do not particularly involve mental-stateunderstanding. However, it is worth pointing out that the development of the-ory of mind definitely affects more complex root uses. For instance, in orderto use a modal like must to impose (rather than simply state) an obligation,some consensus is required among the interlocutors as to social relations, is-sues of power, authority, duty, commitment and other social concepts whichrely heavily on mental state attribution. Furthermore, in order for speakers touse must felicitously on this interpretation, they must have the ability to calcu-late the addressee’s desires, preferences and goals. It is probably no coincidencethat some of these more sophisticated aspects of root interpretations are notacquired 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 ac-count of the acquisition of epistemic modality. Naturally, one needs to con-sider several other factors which also play a role in the acquisition of modalvocabulary. One major consideration relates to input. There is evidence thatadults talking to very young children predominantly use root modals. Wells(1979) reports that can and will were the only modals which were used byall of the mothers in his sample of very young children; they were also themodals with the highest frequency. Shatz, Grimm, Wilcox and Niemeier-Wind(1990) present evidence from American English and German maternal speechto children around three years of age: They note that fewer than 10% ofthe 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 productionof modal terms. Since parents are likely to tailor their speech to the interests, ac-tivities and goals of their children, it is easy to see why desire, ability, obligationand permission interpretations of modals outnumber those of epistemic neces-sity or possibility in early speech to children. Furthermore, epistemic modals


in English, as in other languages, are more formal expressions than closelyrelated vocabulary items such as semi-auxiliaries (have to), or mental terms(think, know). Their markedness contributes to their restricted distribution inthe adult speech to children. Interestingly, for the very same reasons, adult En-glish speakers in general use root modals much more frequently than epistemicmodals (see the corpus data in Coates 1983). In sum, then, it seems that chil-dren learning English for a variety of reasons simply do not hear enough modalverbs with epistemic interpretations in their parents’ speech – hence they donot have sufficient opportunities to acquire them.

More generally, it is plausible that epistemic interpretations of modals areharder to acquire than root interpretations because the mapping between theword and the epistemic concept may be harder to construct. That is, regardlessof the conceptual difficulty posed by epistemic terms, the task of figuring outthat a certain vocabulary item has such an abstract meaning may be hard forthe young learner (Gleitman 1990). Consider the similar case of mental verbs:one of the reasons that verbs like want, know and think pose greater challengesto young learners than verbs such as jump, eat or catch, is that the referentsof the former class are not straightforwardly observable. Therefore, the infor-mation supplied in the extra-linguistic context is a less helpful cue in figuringout the meaning of mental verbs (compared to more ‘concrete’ verbs). Recentexperimental evidence lends support to this conclusion (see Gillette, Gleitman,Gleitman and Lederer 1999). A full account of how modal interpretations arelearned would have to disentangle the precise contribution of conceptual devel-opment and mapping processes to the acquisition of the meanings of epistemicpredicates.3

. A test case: Autism

. The mindblindness hypothesis

Autism is a severe childhood psychiatric condition which is characterised bya number of social and communicative impairments. Autistic children lackthe usual cognitive flexibility, imagination and pretence, and their behaviouris marked by a restricted range of interests and activities. A number of authorshave suggested that (at least certain aspects of) autism can be explained as re-sulting from lack or delay in theory of mind (Baron-Cohen, Leslie and Frith1985; Leslie and Roth 1993; Leslie and Thaiss 1992; Frith 1989; Baron-Cohen1995). On this hypothesis, autism is a form of mindblindness with a number of

Anna Papafragou

implications for cognitive development. Children who suffer from autism areunable to understand (and attribute) false beliefs; they cannot appreciate themind/brain as an organ with mental functions; they are unable to realise thatseeing leads to knowing; they experience difficulty with the mental-physicaland appearance-reality distinctions. As I mentioned earlier in this paper, theseaspects of the human theory of mind are normally acquired around the age ofthree or four. Autistic children differ from both normal children and childrensuffering from various other disorders (e.g. Down’s syndrome) in that theirmentalizing abilities are selectively impaired.

The mindblindness hypothesis about autism offers a solid testbed for thelink which I have sought to establish between epistemicity and mindreading.If, as I have suggested, the development of the metarepresentational machineryresponsible for sophisticated mentalizing is a prerequisite for the full-blownand correct production and comprehension of epistemic modal markers, thenthere should be a severe difficulty with such terms in the language of autism. Inthe following pages, I want to discuss the findings of a study by de Roeck andNuyts (1994) which appears to disconfirm this dissociation between autisticlanguage 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 ofepistemic modality in Dutch by four high-functioning autistic adults. The sub-jects covered an age range between nineteen and twenty-nine years. Their totaland 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 ba-sis 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 useof waarschijnlijk is completely absent from the data, a fact which is not partic-ularly surprising given the low rates of occurrence of this use even in normalspeech. As for the adverbial use, it occurs rather normally in the speech of twoof the four subjects. Denken is used by all four subjects, and there is evidence ofboth its complement-taking and its parenthetical uses. Finally, kunnen is usedas an epistemic modal with higher frequency than in normal speech (whereit is more often used to express simple root/deontic modality). The authorsconclude that all four subjects demonstrably use epistemic expressions in waysmuch similar to normal subjects and therefore appear to engage in metarep-


resentation quite unproblematically. Consequently, the performance of thesehigh-functioning autistic adults is taken to provide some evidence against thetheory-of-mind hypothesis for autism.

As far as I can see, there are various alternative conclusions one might drawfrom de Roeck and Nuyts’ observations (and the authors themselves exploresome of them). For instance, one might question the initial assumption thatepistemic modality involves metarepresentation. I want to argue for two alter-native explanations of their findings, which do not threaten the premise thatepistemic predicates involve metarepresentation.

Firstly, one could argue that, in fact, the specific items used by the fourautistic adults in the study were not always used to communicate epistemicmodality. Even though it is not possible to find compelling evidence simply byinspecting the subjects’ spontaneous speech data, there are certain exampleswhich suggest that something less than a fully metarepresentational readingmight be available to the speakers. Specifically, one of the subjects seemed touse the same modal construction ‘t Kan (ook) zijn dat (‘it may/can (also) bethat’) as a set phrase (and often in inappropriate circumstances) in the ma-jority of on-line epistemic judgements she produced. This subject was boththe youngest (nineteen years old) and the one with the lowest IQ scores of thegroup (68t, 77v). An actual example is (8), where the subject (T) is talkingabout 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 anythingabout 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 beenthe 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 produc-tion was studied might belong together with that talented minority of autisticindividuals (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 involvingmetacognition.4 Such exceptional competence is usually explained in two ways:

i. Autism might cause a delay in the operation of human mindreadingcapacities, rather than a permanent and absolute inability to metarepresent(Eisenmajer and Prior 1991; Baron-Cohen 1995). Depending on the differ-ent 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 per-forming some sort of mentalizing through the usage of alternative strategies.As Leslie and Roth (1993:103) remark, ‘such strategies are arrived at by exer-cising general reasoning abilities, and hence require a quite high level of suchabilities, together with extensive practice and general knowledge, and hencedo not appear before adolescence’. Although the exact nature of these strate-gies is at present not fully known, it is possible that they rely on an analogyto the pictorial format for the representation of thought. Pictures as a repre-sentational medium, unlike mental constructs, are rather well-understood byautistic subjects; evidence for this is provided by good performance of autisticsin variations of the ‘false belief ’ task which involve ‘false (dated) photographs’and ‘false drawings’ (see Leekam and Perner 1991; Leslie and Thaiss 1992; andCharman and Baron-Cohen 1992, respectively). Moreover, there is some ten-tative evidence which suggests that, whereas normal subjects report their in-ner experience in terms of inner speech, pictures or ‘pure thought’, very ableautistic subjects describe their mental contents entirely in terms of pictures(Hurlburt, Happé and Frith 1994; cf. Hurlburt 1990). In the small group ofautistic people studied by Hurlburt et al., ability to report inner experience interms of pictures correlated closely with performance on standard theory-of-mind tasks, independent of IQ. Overall, then, this evidence suggests that high-functioning autistic people may use their understanding of external represen-tations such as pictures to achieve an understanding of mental representationssuch as thoughts and beliefs.

Both the age and the high verbal and non-verbal IQ of the subjects in deRoeck and Nuyts’ study corroborate an explanation for their performance interms of compensatory mechanisms for theory-of-mind abilities. It is worthnoting here that T., the youngest subject and also the one with the lowest verbaland general intelligence, had markedly greater difficulty in using modal expres-sions than the rest of the group. In sum, then, de Roeck and Nuyts’ (1994) dataoffer an impressive manifestation of the success of very able autistic subjectsto overcome the difficulties in their condition and to engage in normal verbalcommunication. However, they do not offer a counterexample to the theory-


of-mind hypothesis for autism, or an argument for dissociating epistemicityfrom metarepresentational (more specifically, metacognitive) abilities. In or-der to present a truly critical test for both hypotheses, one has to turn to lin-guistic data from young autistic people (between four and twelve years) wherethe presence of compensatory mechanisms is less likely. The prediction thenis that early autistic child language should be quite impoverished in the use ofmental terms.

This prediction is borne out. Tager-Flusberg (1993) reports the results ofa longitudinal study, in which six children with autism between three and sixwere followed for between one and two years. Their spontaneous productionswere compared to those of a Down’s syndrome group of six children of thesame productive language level and age. Her findings suggest that, while bothgroups of children talk about perception and mental states of desire and emo-tion, autistic children make significantly fewer references to cognitive men-tal states than children suffering from Down’s syndrome. Elsewhere, Tager-Flusberg specifically comments on the use of modal verbs in the language of thetwo sample groups of the above study (see Tager-Flusberg 1997). It appears thatcan and will make up 95% of all modals used by both groups of children, whileepistemic 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 willwhich require some mentalistic understanding (e.g. the volitional/intentionaluses of will) are very rare in the language of autism, whereas they occur morefrequently in the speech of the children with Down’s syndrome. It appears,therefore, that impairments in theory of mind are reflected in the way lan-guage is used by children with autism. Moreover, it seems that this is done ina way consistent with the analysis of specific linguistic items (e.g. epistemics orvolition markers) in mentalistic terms.

. Concluding remarks

In this paper, I presented some preliminary arguments to support the hy-pothesis that the acquisition of epistemic modality presupposes the abilityto metarepresent mental representations (e.g. beliefs). I also argued that thismetarepresentation hypothesis yields a number of interesting and testable pre-dictions about the connection between early semantic/pragmatic abilities andcognitive development. Finally, I showed that an impairment in the ability toconstruct 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 spe-cific model of folk psychology. Furthermore, regardless of how theories of the-ory of mind will turn out, they will have to account for the fact that complexmetacognitive abilities are the prerequisite for several aspects of linguistic de-velopment. Nevertheless, it might be interesting to examine whether linguisticdata of the sort I have discussed may be used to tease apart different theoreticalproposals 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 NeilSmith and Deirdre Wilson. I am also grateful to the editors and to the series editor, WernerAbraham, 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 alsoimplicated 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 twoyears and nine months, i.e. earlier than their English counterparts. This fact can be explainedon 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 isat home) and present no difficulty for normal children around six or seven are hard for mostof the members of the talented minority of autistic children.

References

Astington, J. and A. Gopnik (1991). Theoretical explanations of children’s understanding ofthe mind. In G. Butterworth, P. Harris, A. Leslie and H. Wellman (Eds.), Perspectiveson the Child’s Theory of Mind (pp. 7–31). Oxford: Oxford University Press/The BritishPsychological 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 ofmind’? Cognition, 21, 37–46.


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’sTheory of Mind (pp. 1–4). Oxford: Oxford University Press/The British PsychologicalSociety.

Butterworth, G., P. Harris, A. Leslie and H. Wellman (Eds.) (1991b). Perspectives on theChild’s Theory of Mind. Oxford: Oxford University Press/The British PsychologicalSociety.

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-basedreasoning. Cognition, 66, B1–B11.

Charman, T. and S. Baron-Cohen (1992). Understanding beliefs and drawings: A further testof 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 functionsin Korean children. In J. Bybee and S. Fleischman (Eds.), Modality in Grammar andDiscourse (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’ abilityin autistic children. British Journal of Developmental Psychology, 9, 351–364.

Flavell, J. (1986). The development of children’s knowledge about the appearance-realitydistinction. 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 activity-

types 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 Devel-opment 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 inCognition 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: Apreliminary 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: Evidencefrom autism. Cognition, 43, 225–251.

Leslie, A. (1988). Some implications of pretence for mechanisms underlying the child’stheory of mind. In J. Astington, P. Harris and D. Olson (Eds.), Developing Theoriesof Mind (pp. 19–46). Cambridge: Cambridge University Press.

Moore, C. and J. Davidge (1989). The development of mental terms: Pragmatics orsemantics? Journal of Child Language, 16, 633–641.

Moore, C. and D. Furrow (1991). The development of the language of belief: The expressionof 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 modalexpression of speaker certainty and uncertainty and its relation to the developmentof a representational theory of mind. Child Development, 61, 722–730.

Noveck, I., S. Ho and M. Sera (1996). Children’s understanding of epistemic modals. Journalof Child Language, 23, 621–643.

O’Neill, D. and C. Atance (2000). “Maybe my daddy give me a big piano”: The developmentof 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 theirbeliefs. Developmental Psychology, 27, 390–397.

Olson, D. and J. Astington (1986). Children’s acquisition of metalinguistic andmetacognitive verbs. In W. Demopoulos and A. Marras (Eds.), Language Learning andConcept Acquisition (pp. 184–199). Norwood, NJ: Ablex.

Papafragou, A. (1998). The acquisition of modality: Implications for theories of semanticrepresentation. Mind and Language, 13, 370–399.

Papafragou, A. (2000). Modality: Issues in the Semantics-Pragmatics Interface. Amsterdam &New York: Elsevier Science.


Papafragou, A. (2001). Linking early linguistic and conceptual capacities: The role of theoryof mind. In A. Cienki, B. Luka and M. Smith (Eds.), Conceptual and Discourse Factorsin Linguistic Structure (pp. 169–184). Stanford: CSLI Publications.

Papafragou, A. and P. Li (in press). Evidential morphology and theory of mind. Toappear in Proceedings from the 26th Annual Boston University Conference on LanguageDevelopment. 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 modalexpressions in conversations between German and American mothers and their two-year-olds. Paper presented at the biennial meeting of the Society for Research in ChildDevelopment, Kansas City, MO.

Shatz, M., H. Grimm, S. Wilcox and K. Niemeier-Wind (1990). Modal expressions inGerman and American mother-child conversations: implications for input theories oflanguage acquisition. Unpublished ms., University of Michigan, Ann Arbor.

Shatz, M., H. Wellman and S. Silber (1983). The acquisition of mental verbs: A systematicinvestigation 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 historyof English, creoles, and language acquisition. In A. Ahlqvist (Ed.), Papers fromthe 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). Berlinand New York: De Gruyter.

Tager-Flusberg, H. (1993). What language reveals about the understanding of minds inchildren with autism. In S. Baron-Cohen, H. Tager-Flusberg and D. Cohen (Eds.),Understanding Other Minds: Perspectives from Autism (pp. 138–157). Oxford: OxfordUniversity Press.

Tager-Flusberg, H. (1997). Language acquisition and theory of mind: Contributions fromthe study of autism. In L. Adamson and M. Romski (Eds.), Research on Communicationand Language Development (pp. 133–158). Baltimore, MD: Paul Brookes Publishing.

Anna Papafragou

Wellman, H. and D. Estes (1986). Early understanding of mental entities: A reexaminationof childhood realism. Child Development, 57, 910–923.

Wellman, H. and J. Woolley (1990). From simple desires to ordinary beliefs: The earlydevelopment of everyday psychology. Cognition, 35, 245–275.

Wells, G. (1979). Learning and using the auxiliary verb in English. In V. Lee (Ed.), CognitiveDevelopment: Language and Thinking from Birth to Adolescence (pp. 250–270). London:Croom Helm.

Wells, G. (1985). Language Development in the Pre-school Years. Cambridge: CambridgeUniversity 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 ofmental life: Understanding informational accesses as origins of knowledge and belief. InJ. 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 MiddleDutch ghe-particle constructions*

Gertjan PostmaUniversity of Leiden

One of the uses of the Middle Dutch prefix ghe- is sensitive to the negativepolarity 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 ofghe-: the copular verb sijn ‘to be’. The negative polarity use can, however, alsobe traced in other verbs, although in a less pure form. Negative polarity ghe-has all the properties of Modern Dutch verbal negative polarity items, such askunnen uitstaan ‘can stand’. Crucially, these involve locality conditions,anti-additivity, and low-scope modality. By extending the Neg-criterion toNPIs with negative force, various restrictions on the ghe-construction can bederived, 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 semanticsand logic that negation and modality are closely related. Multiple-world se-mantics has made it possible to express the relation between necessity and pos-sibility with the help of negation. From this perspective, it does not come asa surprise that, cross-linguistically, negative polarity and modality are closelyrelated as well: many negative polarity verbs and verbal expressions, such asEnglish can stand, are stative or modal in nature. Some modal verbs, such asEnglish must, have a negative polarity counterpart (need). The nature of thisrelation, however, has remained unclear. Usually, a lexical approach is adopted:the lexeme can stand, together with its modality, is taken to be a negative po-larity item. Its modal nature is thus stored in its lexical specification, ratherthan being derived from its NPI nature, or from a condition on its proper useas a NPI.

Gertjan Postma

Middle Dutch is interesting in this connection as it exhibits a productivestrategy of forming negative polarity verbs. The Middle Dutch verbal prefixghe- turns a verb into a negative polarity verb. Hence, Middle Dutch is an ex-cellent tool to study the relation between negative polarity and modality, andwe will see that various interactions with modality show up. After a discussionof 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 ghe-

To understand the function of Middle Dutch ghe-, it is instructive to compare itwith its Modern Dutch counterpart. The Modern Dutch preverbal ge- particlefunctions 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/semanticge-, respectively. Examples are given in (1).

(1) a. Jan heeft geslapen morphological ge-‘John has slept.’

b. Dat gebeurt wel vaker lexical/semantic ge-‘That happens more often.’

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 usesexisting 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) type

a. Ontfaerme u miere scade, die mi Reynaert heeft ghedaenmorphological

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 lexicalnot my wil, but thy will happen-subj‘Not my will, but thy will be done.’

c. Van eenen goeden boem en kan niet danOf a good tree can not thangoede vruchten ghecomen syntacticgood fruits ghe-come‘From a good tree only good fruits can come.’

In (2a) the use of participial ghe- is parallel to that in (1a) in Modern Dutch. Iwill call this ghe-1. In (2b) we encounter a use that can be compared with theinherent 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 presentSins that the wise- ghe-happen‘Crimes that befall the wise’

b. Niet mijn wille, maar dijn wille ghescie subjunctiveNot my will, but thy will ghe-happen‘Not my will but thy will be done.’

c. Binnen dier tijd so gheschiede een dootslach preteriteIn 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 infinitiveDoubt 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 warenWith all those good works that ghe-happened wereende noch ghescien souden participleand yet ghe-happen would‘With all the good works that had happened and were yet to happen.’

f. Ju staat te ghesciene grote rampYou stand to ghe-happen great disasterende swaer verlies (Wat. 8012) gerundand 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 verbalparadigm as well. It forms an essential part of the verb itself. In those cases inwhich 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 wayas Modern Dutch.

In addition to these uses, Middle Dutch exhibits a third use of ghe-, whichhas received ample attention in grammars of Middle Dutch. It is usually char-acterised as an aspectual particle. I will call it ghe-3. It is typically attached toinfinitives and preterites, as is illustrated in (4a, b).

(4) a. Nu en can ic langher niet gheswighenNow can I any-longer ghe-keep-silent‘Now I cannot keep silent anymore.’

b. Alle den smaec ende bliscap, die alle hoverdighe menschen ye ghe-haddenAll 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 Thisrestricted 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 theliterature. Whereas the Middle Dutch dictionary of Verwijs & Verdam lists theparticipial forms with ghe-1 under the entry of the verb without ghe- (this isthe situation in modern dictionaries as well), syntactic ghe-, as in (4), is listedunder the prefixed verb: ghehadden in (4b) is listed under the putative verbghehebben. 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 thelight of two considerations. Firstly, there is the fact that ghe-3 occurs as a spe-cific paradigmatic form of the infinitive, which is the base form normally usedin western lexicographical tradition. This creates a trap for dictionary mak-ers, which does not exist for ghe-1. Secondly, Verwijs & Verdam did not havethe theoretical tools to sharply distinguish cases of ghe-3 from the inherentprefix ghe-2.

However understandable Verwijs & Verdam’s classification may be, itwould be the same as listing the participle of dragen ‘carry’, i.e. gedragen ‘car-ried’ 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.’


(6) a. Ghi laedt de menschen met bordenen dieYou burden the humans with burdens thatsi 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 thateen 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 thelexical/semantic ghe-particle, which usually makes a quite distinct idiosyncraticsemantic contribution, and the syntactic ghe-prefix, which makes hardly any orno semantic contribution at all. Instead of treating ghe-2 en ghe-3 together, asVerwijs & Verdam do, I will make an attempt at unravelling the distinct usesof ghe-, using modern linguistic tools. I will start by showing that the threeparticles 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 inde-pendence of ghe-3 and ghe-1. This verb is special in that it lacks ghe-prefixingin the participle.2 This is usually explained on the basis of the inherent perfec-tivity of the verb comen: no additional perfective ghe- is necessary in the par-ticiple. 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 theinfinitive and preterite, where it is found in only a limited number of specificsyntactic 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 theyoccur in a negative clause, in a clause introduced by the conjunction (al)eer‘before’, or in another context.

(7) before other

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

Gertjan Postma

before otherb. 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 constewten boeme ghecomen en konde +So that he with force nor with trick out the treeghe-come could‘In such a way that he could not leave the treeneither 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-youmocht 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 derheeren handen ghecomen conden +all English, (. . . ) so far as they (. . .) in the lords’shands ghe-come could‘all the English people that would fall intothe 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 genomenso has the king flight taken‘Before they could reach the place, the king had escaped.’


before otherj. Hoe ic daer ghecomen sal, . . . +

How I there ghe-come will, . . .‘In whatever way I will come there, . . .’

These examples make it abundantly clear that a block on participial ghe- inperfective verbs does not imply a block on ghe-3. Some additional regularitiesstand 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 spe-cific paradigmatic form of the verb comen. Secondly, we observe that this useof ghe- predominantly occurs in negative contexts: only 10 of the 27 cases arenot negative. Most grammars and dictionaries of Middle Dutch mention thistendency (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-negativecases 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 classifiedas ‘other’, to which we will return in Section 4.

It is a striking fact that [ghe-3 + infinitive] always occurs with a modalverb (Streitberg 1891:144; Van Swaay 1899:55) in the deontic reading. In mostcases the modal verb is connen ‘can’, but [ghe- + infinitive] also occurs in thecomplement 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 connenmore than I can ghe-mention‘More than I can mention’

b. By desen lande ende by der sonnen sweer ic u,mocht so ghecommen moghenBy 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 thatif it might happen so . . . ’

c. Hoe ic daer ghecomen sal sullenHow I there GHE-come shall‘How I will ever get there.’

d. Dat hijt wale doerste ghedoen dorrenThat he-it well dared GHE-do‘That he did dare to do it.’

Gertjan Postma

e. Dander en wilde niet hem geantwerden willenThe-other wanted not him GHE-answer‘The other one did not want to answer him.’

f. (not attested)3 laten

What is crucial here is that not a single case has been found where ghe3+inf oc-curs in the complement of moeten ‘must’, or in the complement of non-modalECM verbs such as sien ‘see’ en horen ‘hear’.4 This shows that a specific typeof modality is necessary to license the occurrence of ghe-3. Only CAN modalityallows 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 demon-strated best with the help of the verb sijn ‘be’. As shown in (9), the ModernDutch copular verb zijn is one of the few verbs that does not allow for a lexicalprefix at all: the prefixes be-, ver-, ge- or ont- are inadmissible with zijn. Thesame 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-benI 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 exam-ples in (10).

(10) a. Hy en can niet ver gezijn negationHe 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 concessiveHow great, how wise, how holy they ever could GE-be‘However great, however wise, however holy they could be’


e. Dat onder dat volc nu meer heyligher liede sijndant hier voermaels ye ghewaren comparativeThat under that people now more holy men arethan-it before ever GE-were‘That there are more holy men among those people than there wereever before.’

f. 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 mustchoose the eleventh’

Parallel to the result in (8), ghe- is only found in infinitives and preterites, indi-cating that we are dealing with ghe-3.6 Apparently, ghe-3 is possible in contextswhere lexical ghe-2 is blocked. It is striking that negation shows up again asa licensing factor of ghe-3. Moreover, [ghe + infinitive] only occurs in modal-ity contexts, in this case only in the complement of connen ‘can’. Apart fromnegative contexts, ghesijn occurs in the restrictive set of the universal quanti-fier (10b), in rhetorical questions (10c), in concessive contexts (10d), in thescope of the comparative (10e), and in conditional clauses (10f). In the follow-ing 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. Thisghe-3 prefix has specific properties that make it distinct from participial ghe-on the one hand, and from lexical/semantic ghe- on the other hand. I list itsproperties 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 ofa. 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 thescope 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 usedin an affirmative sentence such as (12a), it is perfectly acceptable in a negativeclause such as (12b).

(12) a. *Jan had ook maar één boek gelezenJohn 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 in-tensively. One of the points at issue has been why the set of contexts allowingthe presence of items like ook maar is wider than merely the predicative scopeof ordinary negation. The restrictive set of the universal quantifier, conditionalsentences, comparatives, rhetorical questions, and concessive clauses all pro-vide a good context for the use of ook maar. A number of instances are givenin (13).

(13) a. Niemand had ook maar iets gedaan. NEGnobody 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 boughtvoelden zich bedrogen. restrict. set ∀felt cheated‘All tourists who had bought anything at all felt cheated.’

c. Geen toerist die ook maar iets begrepen hadno tourist who at all anything understood hadvond de gids redelijk. restr. set ¬found the guide reasonable‘No tourist who had understood anything at all considered the guideto be reasonable.’

d. Hij ging vaker naar de bioscoophe went more often to the cinemadan ook maar iemand vermoed had. comparative


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, . . . concessivemight 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, . . . conditionalif 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 com-mon. Seuren (1975) assumes that they contain a covert negative operator.Ladusaw (1980) proposes that they share the logical property of allowingfor downward entailments. An entailment F(X)→ F(Y) is called ‘downward’if X ⊂ Y. In (14a) for instance, the restrictive set ‘left-handed Frisian’ isproperly included in ‘Frisian’. Similarly, the nuclear scope of ‘sing loudly’is contained in that of ‘sing’. Hence, the inferences in (14a) and (14b) areboth downward.

(14) a. Every Frisian sings → Every left-handed Frisian singsb. Every Frisian sings *→ Every Frisian sings loudly.

A downward inference is licit in (14a) but not in (14b): the quantifier everyis evidently monotone-decreasing in its restrictive set, but not in its nuclearscope. 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 Dutchgeen ‘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 singsb. No Frisian sings→No Frisian sings loudly

Correspondingly, the Dutch negative quantifier geen and English no allow forthe presence of negative polarity items like Dutch ook maar or English ever, inthe 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 aboutFrance.

b. No Frisian will ever sing a song.

Hoeksema (1983) shows that comparatives create monotone-decreasing con-texts, and hence allow for negative polarity items. Sanchez, van der Wouden& Zwarts (1993) argue that contexts with voor/eer ‘before’ are also monotone-decreasing. Van der Wouden (1994) also argues for subjunctive contexts be-ing monotone-decreasing. Although our understanding of negative polarity isstill fragmentary, it has become clear that negative polarity is a linguistic phe-nomenon showing up in a wide range of languages, with a by-and-large similarpattern, i.e. the same type of contexts seem to license negative polarity itemsacross 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 withalmost the same meaning, but without polarity sensitivity: English some andany both indicate existential quantification (∃). The essential difference is thatany 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 ofcontexts 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 polarityitem: the function of ghe-3 is to turn a verb into negative polarity verb. Wecapture this in (19).


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

Modern Dutch no longer has this or any other productive strategy for formingnegative polarity verbs.8 However, Dutch has verbs and verbal expressions thatare 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, how-ever: they are modal in nature. This shows up quite clearly in those expressionsthat convey modality explicitly, such as kunnen uitstaan ‘can stand’ in (20b) andkunnen schelen ‘care’ in (20c). The modality, however, can also be implicit, asin donderen ‘matter’ in (20e). That there is an underlying modality in donderen‘matter’ and talen ‘want/ask for’ is shown conclusively by the impossibility ofembedding these verbs under kunnen ‘can’.9

(21) a. *Niemand kon daar naar talennobody could there for want

b. *Dat kan niet donderenthat could not thunder

Apparently, negative polarity in the verbal domain goes hand in hand withmodality. Whatever the reason for this (see Section 6 and 7.5.1), the fact thatthe Middle Dutch form [ghe-3 +infinitive] only occurs in modal contexts sup-ports 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 ‘syn-tactic ghe-’ without any scruple. The reason for calling ghe-3 syntactic is that itis completely productive, and is only found in specific syntactic environmentslike negative polarity and modality. Moreover, it is virtually impossible to at-tach any precise semantic content to ghe-3. Ghe-3 clearly encodes somethingsyntactic. It would be desirable, however, to be able to point to a real syntacticprocess in which ghe-3 is involved. Consider the fact that ghe- can move fromthe embedded infinitive to the dominating modal verb. Some instances of thisare given in (22).

(22) a. Dat negeen broeder negeen knecht, die den huse omme sout of inkaritaten dient ummer ghedorre slaan.that no brother no servant who works in the house for money or incaritas serves ever GHE-dares hit‘That no brother should ever dare to hit a servant who works in thehouse 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 nochachter no besidenso-that you those people let not can beside you walk before norbehind 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’.


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-machsien 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’. Thisseems to be the correct intuition. There is some evidence for this intuition: theoccurrence of ghe- with a present tense modal, as in (22b) and in (22g) im-plies that the modal itself cannot be the underlying source of ghe-, as we haveseen that ghe- only occurs with infinitives and preterites. In a generative frame-work, this would mean that Stoett’s ‘anticipation’ can be analysed as involvingsyntactic movement of ghe- from the main verb to the modal verb, as shownschematically 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 inItalian 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. Notethat movement of ghe- seems to be obligatory in those cases in which theembedded 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 ob-tained from locality effects. The licensing of the ghe-prefix is blocked across anopaque (tense) domain: no barrier may be present between ghe- and its licens-ing factor. This can be inferred from the absence of ghe- in gerunds, i.e. theform of the verb that occurs after the infinitival particle te, ‘to’ as in Englishto go. In other words, there are no examples like (25a, b), not even in a modalreading of the construction. I have represented this formally in (25c).

(25) a. Doe was daer niemen sonder die das, hine hadde te claghene/*ghe-claghene over ReynaerdeThen was there nobody but the badger, he had to complain aboutReinaerd

Gertjan Postma

‘Then there was nobody but the badger who had no complaints aboutReinaerd’

b. dat hij jeghen tvoorbod van den stede meer souts zood dan hij ver-mochte te ziedene/*gheziedene naar dordinanchethat 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 thanhe was allowed to produce according to the rules.’

c. *ne [ te [ ghe-Vgerund . . . ]

Locality effects also show up in constructions with multiple embeddings with-out an infinitival particle. In these cases, ghe- appears on the infinitive that ishierarchically closest to the modal verb.11 We can represent this abstractly as in(27a, b).

(26) a. Sine mochten . . . den muer niet ghedoen vallenThey 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 Vb. *ne doen ghe-V

If we accept this argument ex silentio, we conclude that once again syntacticlocality appears to be necessary for the licensing of ghe-3, on peril of ungram-maticality. Yet, we do encounter constructions of the type in (28), where thereis no local relation between the negation and ghe but where a local relationdoes exist between modality and ghe:

(28) Ic waende, men niet en soude hebben ghe-i+connen ti+vinden manI 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. Thenegative lexeme only needs to c-command the polarity item. This is entirelyanalogous 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.’


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, forthe licensing of the polarity item uitstaan, the negation may reside in a higherclause than the polarity item, and that the relation between modal and polarityitem is bound to strict locality. We conclude that ghe- behaves in all respectsas a negative polarity item. We formulate the locality requirement on ghe-3 asin (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 itemsand the modality requirement

Negative polarity items can be subdivided into two major classes. The first typeincludes those expressions that can occur in any monotone-decreasing envi-ronment. Examples are Dutch hoeven ‘need’, kunnen schelen, ‘care’ en kunnenuitstaan ‘can stand’. Monotone-decreasing contexts fulfill two formal require-ments, i.e. L1 en L2 (for details, see Zwarts 1986:357).12 The second type ofNPI imposes an additional requirement on its environment in that it can onlyoccur in a specific type of monotone-decreasing contexts, namely those that areanti-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 re-quirements on their environment. We should note, however, that in additionto their monotone decreasing characterisation, they impose a requirement ofmodality. The other type of additional requirement is anti-additivity. We nowobtain the scheme in (32).

(32) NPIs can occur in contexts that are monotone decreasing {L1,L2} + addi-tional requirement

additional requirement in the scope of examples of NPI

a. modality nauwelijks ‘hardly’ hoeven ‘need’, kunnenniet elke ‘not every’ uitstaan ‘can stand’,slechts/niet dan ‘only’ kunnen schelen ‘matter’, . . .niemand ‘nobody’niet ‘not’. . .

b. anti-additivity *nauwelijks ‘hardly’ ook maar ‘at all’, enig [e:nıx]*niet elke ‘not every’ ‘any’, ooit ‘ever’, . . .*slechts/niet dan ‘only’niemand ‘nobody’niet ‘not’

I leave for further research the question as to whether these two additionalrequirements reduced to a single formal condition.

In the case of verbal NPIs, the additional requirement can be satisfied bymodality alone. In the case of non-verbal NPIs, modality licensing is excludedand anti-additivity is required.

Let us now address the question to which type ghe-3 belongs: is it anti-additive or not? If ghe-3 is a prefix that turns a verb into an NPI, we expect thatit belongs to the first type since other verbal NPIs of Modern Dutch belong tothis class as well. If we find a context that has ghe-3 in the scope of the Mid-dle Dutch equivalent of nauwelijks ‘hardly’ or niet elke ‘not every’, we couldconclude that ghe- belongs to the monotone-decreasing subtype that does notimpose a further requirement on its environment. Unfortunately, I have notfound a single construction of the type: niet elc . . . ghe-V. However, this seems


to follow from the total absence of niet elc in the Middle Dutch corpus.14 We aretherefore left with contexts with nauwelijks ‘hardly’ and slechts ‘only’. In (33) Igive 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 ghecomenfrom 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’ andniet dan ‘only’, proving that ghe-3 indeed belongs to the NPI-type that requiresmonotone-decreasing contexts where the additional requirement is satisfiedby modality. We are therefore justified in concluding that Middle Dutch ghe+Vbehaves like Modern Dutch negative polarity verbal expressions.

. Summary

In the previous sections, we have seen that one of the uses of the Middle Dutchverbal prefix ghe- is that of a negative polarity item. In this use, it behaves likeModern Dutch NPI verbal expressions. We have found two additional restric-tions: 1. if ghe- is prefixed to infinitives, it must be embedded under a modalityof the CAN type (cf. Section 2); 2. if it is prefixed to finite verbs, the verb maynot be a present tense (cf. Section 2). In Section 7.5.3, we will identify a thirdrestriction: the verbal form with ghe- never undergoes V-second. These highlyspecific restrictions seem to be of a syntactic nature. In order to account forthese restrictions, we need to accommodate the behaviour of ghe- in a syntacticformat. As yet, the only syntactic condition on NPIs reported in the literatureis that they should be c-commanded by a licenser. Fortunately, for our analysisof ghe- we can make use of a syntactic condition that has been formulated fornegative items, especially in the Romance languages, where negative items canpile 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 makean 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 negativespread. N-words in Middle Dutch behave as negative polarity items: when theyare in the scope of an n-item they have an indefinite reading, not a negativereading. The negative clitic seems to function as a scope marker, as is illustratedin (34).

(34) a. Personne n’a vu riennobody not has seen nothing‘Nobody has seen anything.’

b. Niemen en hevet ghesien dinc negheennobody not has seen thing none‘Nobody has seen anything.’

When n-items occupy a scope position, they clearly have negative force, suchas personne and niemen in (34a, b). Apart from these n-items, Middle Dutchhas true NPIs, such as een quade bone ‘a bad bean’, which must always be in thescope of an n-item.

(35) ic en weet daer af niet/*ø ene quade boneI 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 headof a negative chain. In this respect it is similar to the expression a red cent inEnglish. 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 theFrench 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, butoccasionally they carry negative force. The sentential negator niet can be absentalthough the NPI itself never occupies a higher position than the verbal ne-clitic (We may assume that they obtain their operator force in a lower NegPposition). A typical example is een twint ‘nothing at all’, an expression withoutany real meaning, which can best be compared to English squat or fuck all orother taboo items (Postma 2000b). Een twint can be used without niet if it hasa preverbal position (see Postma 2002 for an inventory of the Middle Dutchitems that behave like een twint).15


(36) a. Gine sout macht een twint hebben te anegroetene hare preverbalYou- 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 postverbalc. . . . , sine connen minnen niet een twint

. . . , they could love not a bit‘they could not make love a bit’

Items like een twint can thus take over the function of the negator niet, thenormal carrier of the negation in Middle Dutch. We can now see that ghe-behaves like een twint: there are occurrences where ghe+V can be used in anegative sense without the strong negator niet being present, as in (37).

(37) Mine bliscap es so groot, dat ict gesegghen en machMy 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-itemsneed to meet the Neg-criterion (Van Kemenade 2000). In what follows I discussthe 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 onlylogically, but also syntactically: interrogative and negative constituents maymove to an operator position; both may cause inversion; both may licensenegative polarity items, etc. This has led syntacticians to treat these opera-tors and/or quantifiers as syntactically equal. The proposal by Haegeman &Zanuttini (1991), to the effect that, parallel to the WH-criterion there is aNEG-criterion, has been very influential. According to the WH-criterion (Rizzi1991), a WH-feature must be involved in a specifier-head agreement rela-tion. A WH-word such as where must move to a sentence-peripheral posi-tion and induces movement of the finite verb to the second position (V2).In terms of X-bar theory (Jackendoff 1977; Chomsky 1982), movement ofwhere is XP movement to SpecCP whereas movement of the verb constituteshead-movement to C. According to Rizzi (1991), the verb has an abstractWH-feature in such cases.

(38) a. [Where[+WH]

has[+WH]

[he t gone ]]

Gertjan Postma

b. [Neither[+ NEG]

has [he t gone]][+NEG]

c. CP

SPEC C’

C0 IP

has][+WH

Where[ ]+WH

...

d. CP

SPEC C’

C0 IP

has][+NEG

Neither[ ]+NEG

...

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 neg-agreement takes place.

. The Neg-criterion in the IP domain

In lower domains of the sentence, for instance in the IP domain, neg-agreementphenomena can be observed as well. In (40a, b), a Middle Dutch and a Frenchsentence, negation is not only realised in the subject but also on the verb as aclitic (en/ne).

(40) a. niemen en ginc te dier stedenobody neg went to that place‘Nobody went that way’.


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

The clitic is frequently taken to be a kind of a verbal inflection. Two reasonsare provided in the literature for this. According to Weerman (1988) and vanGestel et al. (1992), the demise of the clitic negation goes hand in hand withthe gradual decline of the verbal inflection with regard to person and numberfeatures (‘deflection’). These authors argue that dropping the negative clitic isan instance of deflection. Haegeman (1995:146–149) reports that in MiddleDutch the absence of person and number features, as in infinitival subclauses,prevents negative inflection by means of the negative clitic en. Since in the IPdomain Neg-agreement is not realised abstractly as verb movement, but overtlyas inflection, there needs to be agreement with respect to number/person spec-ifier features and the head of the projection that has the [+] feature. Ac-cording to Pollock (1989), the IP domain is split into AgrP, NegP and TP, buta 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 IPcan be used as domains for negative agreement: nothing in the Neg-criterionforces 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 projectionjust above the predicative domain, i.e. just above VP in verbal clauses.16 She de-rives this from the distribution of not/naught in early Middle English. Basically,however, her claim applies to all languages. The reinforcing negator would thenbe realised in the lower negation projection. Taboo items like (didley) squat andfuck-all can only be in (an underlying) object position (Paul Postal and JohnRobert Ross, pers. comm; McCloskey 1993; Postma 2000b). They can occasion-ally take over the function of the negator not (Postma 2002): they have negativepotentiality.

(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 po-sition lower in the tree, say specFP, where they enter into NEG-agreement witha head F. Let us label this projection NegP2 (presumably it can be identified asAgrOP). 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 thatthis 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. Inobservational terms, we then see a Neg-V order, i.e. the negator surfaces infront of the verb, provided V2 or other verb movements do not blur this pic-ture. In this way, van der Wurff (1999) and Beukema & Van der Wurff (this vol-ume) explain the preverbal position of negative objects in late Middle English.The made-up sentences in (42) illustrate the difference between non-negativeand negative objects:

(42) a. John will visit Peterb. John will nobody visit

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

(43) NegP

Neg’

Neg

visit

VP

Spec

nobody

...

It will be clear that this theory makes predictions with respect to the locationand behaviour of elements playing the role of the reinforcing negator in MiddleDutch, such as een twint. As we have seen, this NPI is realised preverbally (seeSection 7.1), i.e. in front of the non-finite verb, and may take over the roleof the reinforcing negator niet. It may therefore be necessary to allow for notonly reinforcing elements to move to specNegP, but also (some) NPIs. If thesedo, they acquire the force of a negative item. Laka (1994) also suggests thatin the Romance languages there is no difference between negative elementsand 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 likea negative operator. While Laka’s claim merely involves movement of maximalprojections and their interpretation, the NEG-criterion imposes the restrictionthat there must be a correlation between XP-movement and head-movement.Because of the word order phenomena observed in Middle Dutch and MiddleEnglish, and because of theoretical considerations of symmetry, we need toapply the Neg-criterion to Van Kemenade’s NegP2 as well.


. The Neg-criterion and verbal NPIs

Let us now consider what happens if a verbal NPI takes over the role of thereinforcing negator, as ghe+V in (37) shows. We now arrive at the followingsituation:

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 restrictionThe first possibility of complying with the Neg-criterion is that the verbal pro-jection itself consists of two lexical heads, the lower of which carries the NPIproperty. These are cases with an auxiliary, as in (44). In (44a) the VP in whichthe NPI ghecomen ‘ghe-come’ is included moves to SpecNegP and the auxiliarycan moves to the head of NegP2. In (44b), the finite verb can moves on to theV2 position.

(44) a. dat daer niemen ghecomen en canthat there nobody ghe+come neg can‘that nobody can come there’

b. niemen en can daer ghecomennobody neg can there ghe+come

In order to be able to move to the head of NegP2, the auxiliary must at somestage be below NegP2. Modals appear to take two types of scope with respect tonegation: high scope and low scope. This can be seen in (45), in which Dutchmoeten and English must are typical high-scope modals. The combinationsmoet . . . niet and must not can be paraphrased as ‘having the obligation tonot X’.17

(45) a. Jan moet dat niet doenb. 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 doenb. 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 doenb. 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 com-bined with the NPI ghe-particle. This is indeed the case, as we have seen inSection 2. What is more, we now predict that a Middle Dutch counterpart ofEnglish need, Dutch hoeven might also be combinable with ghe-. Unfortunately,Middle Dutch does not have a separate lexeme for low-scope MUST. The sen-tence in (48), though, might serve as an indication that the correct generali-sation is not to be formulated in terms of a restriction to CAN modality, butrather in terms of a restriction to low-scope modality.

(48) Noyt ne moeste van hem varen ridder no seriant negheen aen anderenhere te soekene leenNever neg must- by him go knight nor sergeant no-one to other lordto seek fief‘As for him, never did knight or sergeant need to go and ask for a fief ofanother lord’.

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

(49) a. DefinitionA wide/narrow scope modal is a modal that has wide/narrow scopewith respect to negation.

b. GeneralisationThe NPI ghe- requires licensing by a narrow-scope modal.

For verbal NPIs I derived this effect from the Neg-criterion as it applied toNegP2, 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 ofNPIs and modality. Not only the Middle Dutch ghe- + infinitive is sensitive toscope effects, negative polarity verbs like English need and Dutch hoeven realisea low-scope modality as well. There are many verbal expressions in Dutch withexplicit or implicit CAN-modality (kunnen uitstaan, kunnen velen ‘can stand’,kunnen schelen ‘care’, kunnen maken, etc.), but not a single NPI with MUST-modality. 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 DutchNPI-verb talen naar, in the context niet talen naar (this expression does not


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 asWANT > NOT > HAVE). Whereas talen naar is an NPI, the high-scope modallusten is polarity-insensitive. Such scope effects in verbal expressions in modernDutch and English might leave us with the suspicion that our generalisation istoo weak. Can we not make verbal NPIs such as need and hoeven sensitive tothe Neg-Criterion and derive their low-scope nature from this?

Despite the initial attractiveness of such a move, I do not think that thisis correct. To begin with, it is theoretically undesirable. The Neg-criterion isapplicable to expressions that have negative potential, in the sense that theycan carry negation. Dutch hoeven and English need have a different status. Sec-ondly, such a generalisation would not be maintainable in view of the freez-ing effects of ghe-preterites discussed in Section 7.5.3. As we will see there, theNeg-criterion applied to verbal NPIs causes freezing effects with respect to V-second. No such freezing effects of verbal NPIs can be found with hoeven ortalen naar. Thirdly, it should be noted that the low-scope modal requirementis absolute in Middle Dutch ghe-infinitives but there are systematic exceptionsin the case of the modern NPI verbal expressions. It is not immediately clearwhether there is any modality in verbal NPIs such as donderen in dat dondertniet ‘that does not matter’. These verbs seem to be stative in nature rather thanmodal. The scope condition in these cases is more like a conditional statement:if modal in nature, modern Dutch NPI verbal expressions realise low-scopemodality. 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 ifwe extend our modality considerations to all NPIs, whether verbal or not. Thispoint can best be illustrated in a table, in which NPIs are classified with respectto 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 ofghe-V to (kunnen) uitstaan and explain it by the same mechanism, i.e. move-ment to NegP2. If we do so, negative potentiality is dissociated from movementto SpecNegP. However, we should then do the same in the nominal domain andextend 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 nominalNPIs also have a modality requirement, although of a slightly different type.This can be seen from the paradigm in (50), which shows the behaviour ofnegative polarity taboo items such as Dutch ene bal or English shit, fuck-all (cf.Postma 2000b).

(50) a. Niemand begrijpt er ene bal vannobody 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 begrijpennobody 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 canbe seen from (50a). Significantly, this NPI noun phrase is blocked by MUSTmodality (50b), although it does tolerate CAN modality (50c). This behaviourseems to be systematic. If we now return to the verbal NPI expressions, we cansubsume these restrictions under the general incompatibility of negative polar-ity and high-scope modality. In the verbal domain, this shows up in a relativelystrong tendency for low-scope modality. The restriction on NPI complementa-tion can be captured in the following two generalizations.

1. NPI verbal expressions with negative potentiality require low-scope rootmodality

2. NPI complements (verbal as well as nominal) exclude MUST modality.

Further investigation of the nature of the general incompatibility of high-scope 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 Dutchghe + infinitive is by no means a fundamental semantic property; rather, itseems to be an accidental syntactic consequence of the Neg-criterion which,felicitously but accidentally, coincides with the general block on high-scopemodality in NPIs.

.. The preterite restrictionLet us now turn to the preterite restriction of ghe-prefixation. The nature ofghe- as an NPI requires it to move to the specifier of NegP. A corollary of theNEG-criterion is that the verbal head to which ghe- is prefixed must move to


the head of NegP. This puts a double burden on ghe+V. It follows that simplextenses cannot undergo ghe-prefixation unless they are complex at some level. Afirst 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 ghewarenall people who there ever ghe+were‘all people who were ever there’.

b. *alle menschen die daer oyt ghesijnall people who there ever ghe+are

It is therefore natural to assume that the present tense is a simplex tense but thepreterite is a complex tense, which may have incorporated a kind of aspectualprojection. This would explain why in finite tenses, the NPI ghe- only attachesto preterites. The NPI ghe-V must not only undergo XP movement to Spec-NegP, but also head movement to Neg, because of the NEG-criterion. Onlycomplex tenses can play this double role. The complex nature of the preterite isat present no more than a stipulation, but it is a reasonable one, with attractiveconsequences. These will be discussed in the next section.

.. Freezing effectsIf we adopt the line of reasoning sketched above, we could say that syntacticallythe 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, thefirst of which is that head-to-head incorporation takes place. The complex isonce 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 bythe ghe-preterite. The lexical part, ghe+V, which carries the NPI-feature, mustmove to SpecNegP, while the V-head marked with PAST undergoes head move-ment to Neg, as required by the NEG-criterion. The structure is illustratedin (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 alexical preterite. It cannot move to a higher functional projection as it is a spell-out of a hybrid syntactic entity: the ghe+V as the remaining part of VP and theabstract PAST tense as the adjacent head. We therefore predict that the ghe-preterite 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- canbe isolated in its purest form in the verb sijn ‘to be’. In (53), I have listed alloccurrences 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 remainher 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 wasor will be.’

e. Overmits des vaders cracht, die al hevet in sijnre macht, wat ye ghewasoff sel gheschien;because the father- force that all has in his power what ever GHE-was or will happen‘Because the father’s, who has everything in his power that ever wasor 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’


g. . . . thooste bot (. . . ) dats minne in caritas, die ewich blijft ende yeghewas. . . the-highest commandment (. . . ) that-is love in charity that foreverremains and ever GHE-was‘the highest commandment that is love in charity which remains for-ever and ever was’

h. . . . dat dat volc van ertricke also goed es gemeynlike ia dleeke volc alstye ghewas. . . that the people of the earth as good is common or-rather lay peo-ple 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 . . . , vanenen troncke . . . , die edelste die ye ghewas.We are all equally of king’s descent come . . . from one stem the no-blest that ever GHE-was‘We are all equally of royal descent, from one stem, the noblest thatever was.’

k. . . . dat onder dat volc nu meer heyligher liede sijn . . . , dant hier voer-maels ye ghewaren.. . . that under that people now more holier men are (. . . ) than-therehere before ever GHE-were‘that there are now more holier men among that people than thereever were before.’

l. Haer onbesceidenheit is soe groet datse (. . . ) meer versmaden wordentwaren, dan si te voren ye ghewarenTheir immodesty is so great that-they . . . more despised become werethan they before ever GHE-were‘Their (i.e. the people’s) immodesty is so great that they have beendespised 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 V2-position. 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 dootNobody 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 (seeAppendix for more details on freezing and V2). The explanation that was givenabove crucially depends on the nature of ghe- as an NPI which can take overthe function of the reinforcing negator niet ‘not’.23

. Conclusions

In one of its uses, the Middle Dutch prefix ghe- is sensitive to the negativepolarity of the context. This use as an NPI is most clearly attested in a verbthat does not allow other uses of ghe: the copular verb sijn ‘to be’. This NPI usecan also be seen in other verbs, although in a less pure form. Negative polarityghe- has all the properties of verbal negative polarity items in Modern Dutch,such as kunnen uitstaan ‘can stand’, most clearly shown in the infinitival verbalform, which requires low-scope root modality (can and need). The low-scopemodality restriction, the preterite restriction and the non-V2 restriction werederived by applying the Neg-criterion to those NPIs that can take over the roleof 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 (main clause with V2)John re-up-brought his daughter‘John re-educated his daughter’


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 instranding/pied-piping properties of her- and the following particle (here: op-).Koopman’s data are not exhaustive, however: similar effects can be foundin double-particle constructions in which both particles have identical pied-piping properties separately, like for example her- and be-.

(ii) a. *Gisteren herbegonnen ze de procedureyesterday 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), thereis no freezing effect.

(iii) Gisteren herverdeelden ze de buityesterday re--divided they the loot‘They redivided the loot yesterday’.

No ablaut effect is present in her+strong-particle constructions in V2 construc-tions: they have a degraded grammaticality already. We thus have the patternsin (iv):

(iv) *her + heavy Prt +V*her+light Prt + strong V

iff in C (Koopman’s effect)iff in C

Similar freezing effects are found in compound verbs consisting of a stressedadverbial or noun as the first element, and a strong verb as the second element:

(v) a. *Jan HARDliep twee uurJan 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 uurJan 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 thatthere is a block on double stress if the verbal complex moves to C, but not ifit 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 procedureÝesterday they restarted the procedure.’

In some cases, the switch to a non-ablauting paradigm can make a complexverb well-behaved.

(vii) a. . . . omdat Jan toen STOF-ZOOGbecause Jan then dust-sucked‘because Jan was hoovering then.’

b. *Jan STOF-ZOOG toenc. Jan STOFzuigde toen

and hence, by analogy:

d. omdat Jan toen stofzuigde

Preferably, Koopman’s freezing effects should get an account of the kind dis-cussed in this paper. Since Ablaut expresses tense-oppositions, we suggest thatgrammatical stress ([+s]), licensing Ablaut, is available in TP. The particle her-does not have inherent stress but acquires it from T by the two mechanismsavailable: 1. head adjunction to T (after which the complex can move furtherto C) or 2. under adjacency to T in SpecTP at spell-out. In both cases, stressconditions are met in the phonology-syntax interface. Adjacency, in fact, dou-bles the stress assignment: [TP X [T Y]], where X and Y are both adjacent toT. In the presence of her+stressed particle/ablaut, only the second strategy isavailable, which results in the freezing configuration.

This account predicts that triple-stressed verbal complexes should exhibitdegraded grammaticality, even when they stay in the freezing position. Thisprediction is borne out by (viii):


(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 Ablautingverb, 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 andcomments.

. 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 non-occurrence of ghe- in the present tense is not an artefact of this corpus: statistical investi-gations of the corpus Gysseling, which includes charters and administrative texts, show thesame 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 underconnen or sullen as licensing modals, as in (i)–(ii).

(i) Dat men die stede ghesien en can aldaar hi esThat 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 wilthat 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 andwith 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, seePostma (1993).

. As Verwijs & Verdam note, the verb ghesijn ‘only occurs in the infinitival and the preteriteforms’. The same cannot be said for the other ghe-verbs since lexical ghe- en syntactic ghe-are subsumed under the same lexical item. In the case of sijn, the form ghesijn can only beinterpreted 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 verbaldomain (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 belgeme-- 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 itsnegative 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 ProParticipio) effect, where an underlying participle (i.e. a form containing the equivalent ofMiddle Dutch ghe-2) shows up as an infinitive:

(i) omdat ik hem dat heb *gezien/zien doenbecause 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 for-mal equations. The four formal properties of contexts defining a classification of quantifiersare listed in (i).

(i) a. f(X)∪f(Y)⊆ f(X∩Y) L1b. f(X∪Y)⊆ f(X)∩f(Y) L2c. f(X)∩f(Y)⊆ f(X∪Y) L3d. f(X∩Y)⊆ f(X)∪f(Y) L4

{L1 and L2} can be considered the core properties of negative polarity.


. Van der Wouden (1994:37) adds L4 as a property relevant to the typology of NPIs, andthus distinguishes two additional types of NPIs: those that impose {L1,L2,L4} on their con-texts, and those that impose {L1,L2,L3,L4} on their contexts. The set of NPIs of the first typeis completely or virtually empty (Van der Wouden 1996:59ff), while the set of NPIs of thesecond type is strongly collocational (e.g. niet pluis ‘fishy’, niet mals ‘trenchant’). It cannot beruled 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 Ger-manic and Romance languages. This might be related to the obligatory negative doubling insubject position in French en Middle Dutch (see (34)). The French negative clitic ne and theMiddle Dutch negative clitic en must be licensed under c-command by a negative operatorat 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 likeYiddish and Middle German in that it allows for both OV and VO surface orders. Here Iassume an underlying VO order, as argued in Zwart (1986).

. Further investigation will be necessary to show whether this lower NegP is the syntacticrealisation 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 scalarcomplementation. If this is the case, the striking interaction of (low scope) modality andnegative polarity (as discussed in this paper) would be identical to the interaction of (lowscope) modality and scalarity studied by Barbiers.

. Both in the subject-directed reading and the non-subject-directed reading, the scoperelations 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 thatthe modal is part of the verbal expression. But this turns out to be unattractive, for threereasons. First, it does not provide a deep explanation for the modality of verbal expressionat all. Second, it is uitstaan rather than kunnen uitstaan that is listed as an NPI, as can beseen from the paradigm in (i).

(i) a. Niemand kan hem uitstaannobody can him out-stand‘Nobody can stand him’

b. Hij is niet om uit te staanHe is not out to stand‘He is unbearable’

Gertjan Postma

c. Hij is onuitstaanbaarhe is un-out-standable‘He is unbearable’

The absence of lexical modality in (i b, c) shows that uitstaan is the NPI rather than kunnenuitstaan, which does not require kunnen lexically, but selects modality on a more abstractlevel. Thirdly, in the case of Middle Dutch ghe-, which behaves in all respects as a NPI verbalexpression, 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 in-terpretations are strongly favoured when there is a perfect tense in the embedded clause,as in (i).

(i) Niemand moet er ene bal van begrepen hebbennobody 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-movementallows for scopal reconstruction (Fox 1999; Bobaljik & Wurmbrand 1999).

(ii) a. niemand moet [ t er ene bal van begrepen hebben] SSb. – 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 thestructural 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 lopenall people who there ever walk

(ii) Alle mensen die er ooit zullen lopenall people who there ever will walk

(iii) Alle mensen die er ooit liepenall people who there ever walked

A parallel account can be given if we assume that Modern Dutch ooit ‘ever’ has absorbed ghe-at the time that the NPI ghe- disappeared in the late 15th century. While absorbing the verbalghe- particle, the newly created NPI ooit ‘ever’ came to mark the VP as expressing negativepolarity. As a result, the VP must move to SpecNegP, and the NEG-criterion requires thatthere be a head to move to Neg. (The phonological shape of ooit is noteworthy: all Dutchwords 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 besimplex: no new noun can exist such as *mijn plooit “my PLOOIT”. This may be a furtherargument for the composite nature of ooit.)

. I omitted one example with a strong Middle High-German colouring, although it fullycomplies with the generalisation.


. The Middle Dutch verb roeken ‘care’ is an NPI which can take over the role of reinforcingnegator just like ghe-V. Nevertheless, it can undergo V2, as shown in (i)

(i) Mi en roeket (niet) ware si gaenMe 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 NPImoving to SpecNegP. No double role is played by the verb and, hence, no freezing effectsshow 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 ofLeiden.

Beukema, F. & W. van der Wurff (this volume). Modals, Objects and Negation in LateMiddle-English.

Bobaljik, J. & S. Wurmbrand (1999). Modals, raising, and A-reconstruction. Paper presentedat 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: Gram-

maticale 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: Syntaxand Semantics (pp. 139–156). Amsterdam: Benjamins.

Jackendoff, R. (1977). X-bar Syntax: A Study of Phrase Structure. Cambridge, Mass: MITPress.

Kayne, R. (1999). Overt vs. covert movement. Syntax, 1, 128–191.Kemenade, A. van (2000). Jespersen’s cycle revisited: Formal properties of grammati-

calization. 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. Pots-

dam (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 andPolarity Items (pp. 285–322). Amsterdam: Benjamins.

Postma, G. (2000b). Anthropological taboo, grammatical mismatches, and the dynamics oflanguage. Paper read at Leiden University, HIL seminar, 5 April 2000.

Postma, G. (2002). De enkelvoudige clitische negatie in het Middelnederlands en de Jesper-sen-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: Mar-

tinus 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 Gro-

ningen.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 Milojevic Sheppard and Marija GoldenUniversity of Ljubljana

This paper provides a syntactic account of Slovene imperative sentences. Twotypologically relevant aspects of imperatives are considered: (i) negation;(ii) pronominal clitic placement. We attempt to account for the Slovene datain terms of strength of the imperative feature located in Co. We show that inthis approach the unacceptability of sentence-initial Clitic– Imperative Verborder is unexpected since Slovene clitics can be either enclitic or proclitic.

. Introduction

In this paper we investigate Slovene imperative sentences with verbs inflectedfor imperative mood. Our principal aim is to present the relevant Slovene dataagainst the background of recent generative approaches to cross-linguistic vari-ation in imperative systems, and to call attention to two theoretically inter-esting properties of Slovene: (i) while clitics are otherwise allowed to open asentence, in imperatives this possibility vanishes – the obligatory order is Verb-Clitic; (ii) contrary to standard assumptions that the imperative is strictly amain-clause item, imperatives in Slovene can be embedded.

For languages whose mood systems provide their verbs with an imperativeparadigm, Rivero and Terzi (1995) propose a typology with respect to whetherthe imperative verb exhibits a syntax distinct from the indicative. They discussin detail two such distinguishing properties: (i) sentential negation, and (ii)clitic placement. Thus, Modern Greek, Spanish and Italian belong to Class Ilanguages, where the syntax of imperative verbs differs from the syntax of in-dicative verbs. With respect to the first property, sentential negation, these lan-guages allow only affirmative imperatives, as in (1a); prohibitions cannot beexpressed by negating the imperative verb, as shown in (1b). With respect tothe second property, clitic placement, Class I languages allow the clitic to pre-cede the indicative verb as in (2a), but they do not allow the clitic to precede the


imperative verb, as shown in (2b). In the French example (2b), the pronomi-nal clitic le must follow the verb, when it is imperative, although it precedes itwhen 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. Loit

leiste/*Leísteloread-.2

‘You read it.’ (Spanish; Rivero & Terzi 1995: (4d))b. Faites

Do..2

le!/*Leit

faites!

‘Do it.’ (French)

On the other hand, in Class II languages, among them Bulgarian and Serbo-Croatian (SC), the syntax of imperative verbs does not differ from the syntax ofindicative verbs. Imperative verbs can be negated to express prohibition, as in(3b); clitics are placed according to the second-position constraint, irrespectiveof verbal morphology, as shown in (4).

(3) a. Ne

citate.read..2

‘You (.) are not reading.’ (SC; Rivero & Terzi 1995: (11b))b. Ne

citajte!read..2

‘Do not read.’ (SC; Rivero & Terzi 1995: (11a))

(4) a. CitateRead..2

je.it

‘You (.) are reading it.’ (SC; Rivero & Terzi 1995: (12b))b. Citajte

read..2

je!it

‘You () read it!’ (SC; Rivero & Terzi 1995: (12a))

Our account of Slovene imperatives begins with a brief overview of the relevantparts of Rivero and Terzi’s (1995) syntactic analysis and Han’s (1998) semanticproposal for the (un)availability of negative imperatives (Section 2). In the dis-cussion that follows we shall argue that both approaches fail to account for thefull range of Slovene data. Section 3 outlines the basic descriptive properties


of Slovene sentential negation and imperatives and examines the syntax of theSlovene imperative verb with respect to pronominal clitic placement. We showthat root imperatives are different from indicatives in that at the beginning ofthe sentence, clitics cannot precede the imperative verb. In Section 4 we focuson negation and clitic placement. With respect to these two properties, Slovenelooks like a mixed language. It belongs together with Serbo-Croatian and Bul-garian (Class II) in having negative imperatives and Wackernagel clitics, yet itappears to have one property of Class I languages, namely, clitic placement inimperatives is not identical to the clitic placement in indicatives. In view of thefact that Slovene clitics are prosodically either enclitic or proclitic, the crucialquestion is what makes the order sentence-initial Clitic-Verb unacceptable inroot 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 neg-ative imperatives and the unique position of imperative verbs with respect topronominal clitics to the location of the imperative operator in syntax.

In Class I, a strong V-feature in Co encodes the logical mood (illocutionaryforce) of imperatives, and the verb, inflected for the imperative morphologymust raise overtly to Co, obeying Greed (Chomsky 1994:14). The absence ofnegative imperatives is attributed by Rivero and Terzi to the Head MovementConstraint. The Neg head acts as a relativized minimality barrier, past whichVo 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 inIo. They reserve Co as a last resort landing site of the verb (both indicative andimperative) to provide the PF-licensing of second-position (2P) clitics. If nophrase fronts, and the clitic is stranded sentence-initially, then all verbs move toCo. 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. How-ever, in Class I languages, the imperative Vo is in Co as a result of feature-checking (Greed); in Class II, the imperative verb ends in Co as a last resortto provide the first-position (1P) constituent for the Wackernagel pronom-inal clitics.


(5) a. Class I: [CP

[imperative]Co [FP Clo [NegP Nego [IP V] ] ]]

b. Class II: [CP

[CP [SpecCP

[Co Vi]X”

[WP Clo

[C Ø]][IP ti] ] ]][WP Clo [NegP

orNeg [IP V] ] ]]

. A semantic account: Han (1998)

Han (1998) derives the unavailability of negative imperatives in Class I lan-guages from their ill-formed semantic representation, with respect to two sen-tential operators, the negative operator and the imperative operator. Accord-ing to her analysis negative imperatives are unavailable in languages where thenegative operator ends up having syntactic scope over the imperative operator.In such structures, the directive force contributed by the imperative operatorwould be negated, which would not be the intended meaning. Negating theimperative does not remove the directive illocutionary force.

(6) CP

C’

Co IP

I Co

Imp-Op

Neg I

In the LF structure suggested in (6), negation asymmetrically scopes over theimperative operator.1 This structure occurs when the negation marker cliticisesonto the imperative verb, and both raise overtly to Co for feature checking.In languages without negative imperatives, the preverbal negation marker andthe verb form a unit in syntax. If negation is present, it moves with the verbas a unit, due to its clitic-like nature. If the Neg-V complex moves to Co forimperative mood feature checking, the imperative in Co ends up within thescope of Neg.


. Negation, imperatives and clitics in Slovene

. Sentential negation

Slovene expresses sentential negation by a negation particle (ne) in preverbalposition. In discourse-neutral sentences, the negation particle appears early inthe clause and is adjacent to the tensed verb. In the embedded clause (7a),the negation particle occurs after the complementiser, the clitics, and the sub-ject DP.

(7) a. dathat

muhim..

gahim/it..

JanezJanez

šeyet

nenot

zaupa.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 verbeven by clitic-like adverbials (še). The negation marker is not part of the cliticcluster; in (7a), it is separated from the pronominal clitic cluster by the subjectand an adverbial.

In the spirit of Pollock (1989) and much subsequent work, we propose totreat 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 inindicative interrogative sentences such as (8), in which the subject DP precedesthe negation and the tensed verb.

(8) [C Aliwhether

gait..

[AgrSP [Spec AgrSP Špela] [NegP

Špelane [TP

not[VP bere . . . ?

read‘Doesn’t Špela read it?’

The negated tensed verb functions as a 1P constituent with respect to cliticplacement, 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 withit. (9) is thus evidence that the complex verbal head [Neg+V ne dam] counts as asingle syntactic constituent with respect to the 2P constraint on Slovene cliticplacement.

(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.’


The placement of the negation particle in negative verbal questions is shownin (10). In a discourse-neutral verbal question like (10b), the auxiliary clitic sicomes first in the clitic cluster (occupying the position of Aux1 in clitic cluster(10a)). In a discourse-neutral negative verbal question (10c), the auxiliary cliticsi 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. Aliwhether

siare.

muhim..

gait..

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 tensedverbal form. The adjacency requirement between the negation marker and theverb is shared by indicative (11) and imperative verb alike (12).

(11) a. Nenot

opravicujeexcuse..3

se

mime..

vec.more

‘He doesn’t apologise to me any more.’b. *Ne se mi vec opravicuje.

(12) a. Nenot

opravicujexcuse..2

se

mime..

vec!more

‘Do not apologise to me any more!’b. *Ne se mi vec opravicuj!

. Imperatives

In Slovene, the imperative is one of the three morphosyntactic mood categoriesexpressed in the inflectional paradigm of the verb. The paradigm in (13) showsthat the imperative forms of the Slovene verb are not morphologically identicalto its indicative forms. Furthermore, there are selectional restrictions on theimperative verb such that it can have only 2nd person singular and 1st and 2ndperson plural and dual subjects.


(13) Indicative and Imperative Forms

2-Sg 1-Pl 2-Pl 1-Du 2-Du

present indicative delaš delamo delate delava delataimperative delaj delajmo delajte delajva delajta

‘you.work’ ‘let.us.work’ ‘you.work’ ‘let.us.two. ‘you.two.work’ work’

In (14) we show that in Slovene imperatives can be embedded. They are not re-stricted to appearing only under reporting verbs, as in (14a), but occur in suchunambiguous structures of embedding as argument clauses (14b), restrictiverelative clauses (14c) and adnominal complement clauses (14d).

(14) a. Rekelsaid

je,is

dathat

delajwork..2

bolje.better

‘He said that you must work better.’b. Peter

Petervztraja,insists

dathat

pridicome..2

jutri.tomorrow

‘Peter insists that you must come tomorrow.’c. To

thisjeis

film,film

kiwhich

siyourself..

gait..

oglejsee..2

cimprej.as.soon‘This is a film which you must see as soon as you can.’

d. Zakajwhy

teyou..

mojmy

nasvet,advice

dathat

bodibe..2

pameten,sensible

takoso

jezi?angers

‘Why does my advice that you must be sensible make you so angry?’

The imperative feature differs from the V-features of other morphologicalmoods in that it signals logical mood. It is the formal ingredient that con-tributes to the determination of illocutionary force, i.e. an illocutionary (logi-cal mood) operator (Rivero and Terzi 1995). However, contrary to Rivero andTerzi’s claim that imperative verbs can only be used in imperative sentences, astheir morphology carries intrinsic logical mood, Slovene imperatives are notrestricted to sentences with imperative illocutionary force, as shown in (15).

(15) a. Potemthen

pa

še

komuwhom

zaupaj!trust.


‘And then trust anybody.’(Intended meaning: ‘It is impossible to trust anybody.’)

b. Recitell.

bedaku,fool.

dathat

jeis

pameten,clever

pa

tiyou.

bowill

verjel.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 necessarilycorrelate with imperative illocutionary force. We identify the root Co as thenatural syntactic location in the clause for the logical mood.

. Clitic placement

As argued in Golden and Milojevic Sheppard (2000), while both Serbo-Croatian and Slovene are 2P clitic languages, only Serbo-Croatian clitics needto be lexically specified as prosodically inherently enclitic (Boškovic 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 cliticsis to be syntactically identified as right adjunction to Co, and concluded thatwhereas the relevant domain for Serbo-Croatian 2P clitics is the intonationphrase, for Slovene clitics, it is CP.

(16) Muhim..

gait..

bošyou.will.

pokazal?show.

‘Will you show it to him?’

(16) is a verbal question beginning with a pronominal clitic cluster. The factthat the clitic cluster in (16) is sentence-initial cannot be attributed to prag-matically controlled omission of the overt interrogative complementiser (ali‘whether’: Ali mu ga boš pokazal?) since Slovene clitics are frequently sentence-initial in declaratives, as well. Thus (17a) can be a response to a question suchas Kje je pismo? ‘Where is the letter?’, and (17b) is a threat.

(17) a. Semam..

gait..

žealready

oddal.mailed

‘I have already mailed it.’b. Ti

you..

bomI.will.

žealready

pokazal!show.

‘I will teach you a lesson!’


Slovene clitic pronouns can be prosodically either enclitic or proclitic, yet wedo not find them in initial position in root imperatives. The imperative verbcannot follow a sentence-initial pronominal clitic (18a, b) whereas the indica-tive verb can ((16), (17b)). Note that in embedded clauses, there is no dif-ference in clitic placement, irrespective of whether the verb is indicative orimperative (18c).

(18) a. *Gahim..

poslušaj,you.listen.

ceif

hoceš!you.want.

b. Poslušaj ga, ce hoceš.‘Listen to him, if you want.’

c. Vztrajalinsisted.

je,is...3

dathat

gahim..

poklici/call..2

poklicecall..3

sam.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 (prag-matically controlled) ellipsis of 1P material: the complementiser ali in (16) andthe subject NP in (17). However, a pragmatic approach cannot account for theunacceptability of (18a). While it is true that Slovene has no lexical comple-mentiser in root imperative clauses, the subject of root imperatives is, as in thecase of non-imperatives, normally unexpressed, so the question remains unan-swered: why can the imperative response in (19b) not have the ordering of theindicative sentence (19a)?

(19) Kje so moja ocala?‘Where are my glasses?’

a. Jihthem..

že

išcem.I.look.for

‘I am already looking for them.’b. *Jih

them..

išci!look.for..2

‘Look for them!’

. Analysis

Intuitively, Slovene looks like a candidate for a mixed language with respect tothe properties investigated in Rivero and Terzi’s (1995) typology. The fact thatit has negative imperatives puts Slovene in Class II languages. Slovene is also


a Wackernagel language, yet it sides with Class I languages in that clitic place-ment in imperatives is not identical to the clitic placement in indicatives. Cru-cially, clitics can appear initially in indicative root sentences but they cannot beinitial in root imperatives.

. Alternative 1

Let us first try to account for the Slovene data in terms of Rivero and Terzi’sanalysis of imperatives in Class II languages. Their approach turns out to beproblematic for several reasons.

i. In Rivero and Terzi’s analysis of Class II imperatives, Vo-to-Co raising is con-sidered to be a last-resort, phonologically-triggered operation. Furthermore,they crucially assume the mutually exclusive application of Vo-to-Co raisingand Topicalisation. In Slovene, however, the rule that raises Vo to Co cannot bea last-resort movement, since the raised verb can co-occur with a topicalisedphrase, as (20a, b) shows.

(20) a. Vrnireturn..2

muhim..

knjigobook.

še

danes.today

‘Return the book to him today!’b. Knjigo

book.

vrnireturn..2

muhim..

še

danes.today

[Spec CP [Knjigo] [Co vrni [C mu [AgrSP . . .

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 isobligatory only with imperative verbs, as shown in (18a, b).

iii. Vo-to-Co is conceived of as a last-resort movement with a phonological trig-ger. Vo moves to Co in order to provide phonological support for clitics. As hasalready been noted, Slovene clitics may, in contrast to SC clitics, be prosodicallyeither enclitic or proclitic. As a last-resort operation, Vo-to-Co is obligatory; ifit 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 nec-essary; unnecessary applications are ruled out by the principle of economy ofderivation. If Vo moves to Co, with SpecCP being filled, the resulting sentenceis predicted to be ungrammatical. In Slovene, this prediction is not borne out,witness the grammaticality of (20b).


v. Rivero and Terzi propose that in Class I languages, Co is the locus of imper-ative force, whereas in Class II languages, Co is empty, reserved for last-resortV-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 locusof the question-force operator, also in Class I languages. In the simplest theory,one would expect that operators signalling the illocutionary force of a sentenceshould have a single location in the CP clausal structure, with root Co beingthe obvious candidate.

. Alternative 2

Proceeding from the assumption that root Co is the locus of the imperativemood 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 languagesin 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 thatshe discusses, Bulgarian and Serbo-Croatian, do have negative imperatives, inspite of the fact that the negation marker affixes onto the imperative verb. Hansuggests that in these two languages the imperative verb remains in Io in syntaxand then moves alone to Co at LF. Since the negation thus stays low in theCP structure, it does not scope over the imperative operator at LF, renderingnegative imperatives possible.

To account for the obligatory Imperative Verb-Clitic order in Bulgarianand Serbo-Croatian, Han draws upon Embick and Izvorski’s (1997) analysisof Verb-Clitic orders in Slavic. They propose that when the clitic has beenstranded sentence-initially, Morphological Merger (Marantz 1989) applies onthe branch to PF as a last-resort post-syntactic mechanism to invert the cliticwith an adjacent element, thus resolving the clitic’s leftward phonologicaldependency. Following Embick and Izvorski, Han argues that in Bulgarianand Serbo-Croatian ((21a) and (22a) respectively) the clitic has affixed ontothe imperative verb in Io at a post-syntactic level. The presence of the nega-tion marker in (21b) makes Morphological Merger unnecessary. The obliga-tory encliticisation of the clitic in negative imperatives in Serbo-Croatian (22b)is to be attributed, according to Han, to an independent constraint of thelanguage.


(21) a. Cetiread.2.

ja!it

(Bulgarian; Han 1998: (22))

‘Read it!’b. Ne

jait

ceti!read.2.

‘Don’t read it!’

(22) a. Citajread.2.

je!it

(SC; Han 1998: (23))

‘Read it!’b. Ne

citajread.2.

je!it

‘Don’t read it!’

In Slovene, the distribution of imperative verbs with respect to clitic placementand negation is identical to that in Serbo-Croatian (cf. (23) and (24)).

(23) a. Beriread.2.

jo! (Slovene)it

‘Read it!’b. *Jo beri!c. Ne

beriread.2.

je!it

‘Don’t read it!’d. *Ne jo beri!

(24) a. Citajread.2.

je! (SC)it

‘Read it!’b. *Je citaj!c. Ne

citajread.2.

je!it

‘Don’t read it!’d. *Ne je citaj!

Adopting Han’s idea of LF imperative verb movement to Co and applyingthe standard feature-checking theory (Chomsky 1995), let us now consider apossible derivation of Slovene (negative) imperatives.

Given the clause structure CP > AgrSP > Neg > TP, (negative) imperativeswould be derived as follows: the imperative verb, having raised overtly out ofVP to T (for T-feature checking), attaches (in negative sentences) to Neg, thetwo forming a single complex head, which subsequently raises to AgrSP (forphi-feature checking). All the movements are overt since all the features of


functional heads within AgrsP are assumed to be strong. The weak imperativefeature 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 move-ment involves formal features only, the V-feature moves to Co alone, leavingthe negation marker behind, hence Neg does not scope over the imperativeoperator in Co. Negative imperatives are thus not ruled out.

The presence of a weak imperative feature in Co predicts that imperativeverbs distribute like indicative verbs with respect to negation and clitic place-ment. This prediction is borne out in the case of negation (cf. 3.1) as well asclitic placement when the clitic is not sentence-intial (cf. 3.3). However, withsentence-inital clitics, the Clitic-Verb order in imperatives is unacceptable, butis restored to full acceptability when an item is inserted in 1P (cf. (23b), re-peated below as (25a), and (25b)). Since Slovene clitics can be prosodicallyeither enclitic or proclitic, this cannot be explained by appealing to a phono-logical constraint, i.e. the leftward phonological dependency of the clitic, to beresolved by a post-syntactic mechanism of Morphological Merger, as proposedby Han (1998) for Bulgarian and Serbo-Croatian.

(25) a. *Joit

beri!read.2.

‘Read it!’b. Sam

yourselfjoit

beri!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 sentence-initial Clitic-Imperative Verb order – under control.

. Conclusion

In this paper, we have examined the syntax of Slovene imperatives. We havefocused on negative imperatives and clitic placement in imperatives, negationand clitic placement being the two aspects of imperatives in which typolog-ical differences in the syntax of imperatives have been observed. An analy-sis of Slovene imperatives in terms of standard feature-checking theory hasbeen shown to be able to account for the relevant data concerning negativeimperatives as well as clitic placement when clitics do not occur in sentence-


intial position. However, the question of what makes the sentence-inital Clitic-Imperative Verb order unacceptable remains unsolved and merits further re-search.

Notes

* We would like to thank Frits Beukema and two anonymous reviewers for valuable com-ments and suggestions.

. In (6), under Kayne’s (1994) definition of c-command, Neg c-commands Co but Co doesnot 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 AgrsPmight be sought in the rich morphology of the Slovene verb. However, the same motivationfor the strength of the imperative feature in Co is absent. The feature involved is a logicalmood feature, different from ‘ordinary’ morphological features that are related, for exam-ple, to Tense and Agr. It would seem that it is more naturally expected to be strong in lan-guages 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 imperativemorphology are not restricted to sentences with imperative logical mood (cf. (15)).

References

Boškovic, Ž. (2000). Second position cliticisation: Syntax and/or phonology? In F. Beukemaand 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 FormalApproaches to Slavic Linguistics: The Cornell Meeting 1995 (pp. 210–239). Ann Arbor:Michigan Slavic Publications.

Golden, M. and M. Milojevic Sheppard (2000). Slovene pronominal clitics. In F. Beukemaand 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 im-peratives. In S. Blake, E.-S. Kim and K. Shahin (Eds.), West Coast Conference on FormalLinguistics 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.), Al-

ternative Conceptions of Phrase Structure (pp. 99–116). Chicago: Chicago Press.


Pollock, J.-Y. (1989). Verb movement, Universal Grammar, and the structure of IP. LinguisticInquiry, 20, 365–424.

Rivero, M.L. and A. Terzi (1995). Imperatives, V-movement and logical mood. Journal ofLinguistics, 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.


Olga Mišeska TomicUniversity of Novi Sad/University of Leiden

The paper focuses on the syntactic behaviour of the Macedonian lexical andauxiliary modals and the interaction of negation with mood and modality.While lexical modals take as their complements Mood Phrases headed by thesubjunctive marker da, auxiliary modals, having become tenseless and havinglost their t’-features, project Modality Phrases subcategorising for the TensePof the verb which in previous stages of the language had appeared in thesubjunctive clause in their complement.

The Mood node projected by the da-complementiser is also the nodewhere imperative mood is checked. On the basis of the syntactic andphonological behaviour of pronominal clitics, the paper argues that theimperative verb raises overtly to this node. In negative imperative clauseswith pronominal clitics, there is a dialectal distinction in the positioning ofthe pronominal clitics relative to the verb. This distinction follows from adifference 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 singlephonological word with the pronominal clitics and the verb. InNorth-Western Macedonian, on the other hand, Neg is strong and occupiesthe specifier position of NegP; at PF the clitics encliticise to it.

. Introduction

Recent generative work has dealt with an increasing number of languages. Witheach new language analysed, new insights concerning the structures and pro-cesses that are part of Universal Grammar emerge. This paper provides an ac-count of the relationship of modals to mood and modality phrases, emergingfrom the syntactic behaviour of Macedonian lexical and auxiliary modals. Itwill be argued that Mood is the node where imperative verbs check their fea-tures overtly. The relationship of mood and modality to negation will also beanalysed, with special emphasis on negative imperatives.

Olga Mišeska Tomic

In Section 2 the syntactic behaviour of the Macedonian lexical and aux-iliary modals will be discussed; evidence will be given for the existence ofModality and Mood Phrases. In Section 3 it will be argued that, in Macedo-nian positive imperative clauses, the verb raises to MoodP. In Section 4 thespecific phonological behaviour of clitics in the environment of the raised im-perative verb will be described. In Section 5 the relationship of mood andmodality to negation will be dealt with: it will be shown that the differences be-tween the behaviour of pronominal clitics in Standard Macedonian and North-Western Macedonian negative imperative clauses follow from differences in thestrength of the negation operator, rather than from the strength of the modalityoperator. 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 aux-iliaries – кÜe ‘will/shall’ and bi ‘would’. The lexical modals are tensed verbswhich take as complements tensed phrases introduced by the subjunctivemarker (.) da. Saka always inflects for person and number, its per-son/number markers being independent of the person/number markers of theclause in its complement:

(1) Sakamwant.1

da.

dojdam/(dojdeš). (Mac)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 complementclauses):

(2) a. Možatcan.3

da.

dojdat. (Mac)come.3.

‘They can come.’b. Može

may.

da.

dojdete. (Mac)come.2.

‘You may come.’c. Moravme

must.1.

da.

dojdeme. (Mac)come.1.

‘We had to come.’


d. Morašemust..

da.

goit.

storimedo.1.

toa. (Mac)that

‘It had to be done.’

Treba and ima are regularly impersonal:

(3) a. Treba(*m/*š)should (1/2)

da.

odam/odiš. (Mac)go.1/2.

‘I/you should go.’b. Ima(*š/*at)

have (2/3)da.

dojdeš/dojdat. (Mac)come.2/3.

‘You/they have to come.’

The modal auxiliaries кÜe and bi are non-inflecting clausal clitics with a fixed po-sition in the Macedonian clausal clitic cluster; they are always accommodatedto the right of the negation operator and to the left of the be-auxiliary andpronominal clitics.2 While bi occurs in clauses in which V is instantiated by anl-participle and cooccurs with an auxiliary clitic, кÜe occurs in clauses in whichV is instantiated by any of the items that can be instantiated in V. Examples ofthe use of bi and кÜe are given in (4):

(4) a. Nenot

biwould..

sumam..

muhim..

goit..

dalgiven.l-..

proektot. (Mac)project-the

‘I would be unwilling to give him the project.’b. Ne

notкÜewill..

muhim..

eis.

izpraznetemptied....

stanot. (Mac)apartment-the‘His apartment will not be vacated.’

As argued in Tomic (1996), the clausal clitics are heads of functional projec-tions. 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, andhas as its complement a Tense/AgrSP,3 projected by the first person be-auxiliaryclitic sum. The relevant structure of (4a) is given in (5):

Olga Mišeska Tomic

(5)

The relevant structure of (4b), where the head of the VP is instantiated by apassive participle, and the modal clitic is preceded by the negation operatorand followed by the pronominal clitic mu and the third person be-auxiliaryclitic e, is represented in (6):4

(6) NegP

Neg’

Neg ModP

ne Mod

e

Tense/AgrSP

Tense/AgrS AgrIOP

AgrIO

mu

AuxP

Aux

e

VP

V XP

izpraznet


The subcategorisation properties of lexical modals differ from those of auxil-iary (clitic) modals: rather than Tense/AgrSPs, the lexical modals take as com-plements tensed subjunctive phrases introduced by subjunctive markers. Thesecomplements, however, differ from ‘canonical’ CPs.

In her analyses of the Balkan clause structure, Rivero (1994) places theBalkan subjunctive markers in the same position as modal auxiliaries. How-ever, in Aromanian – a typically Balkan Romance language, spoken in Mace-donia, Albania and Greece – subjunctive complements occur in complementpositions of auxiliary modals, and so they do in dialectal Macedonian, thoughin the latter case the auxiliary modal is a non-inflecting clitic:

(7) a. Vawill.3..

s-yin.-come.1.

s-ti.-you..

vedsee.1.

mãne. (Arom)tomorrow.

‘I will come to see you tomorrow.’b. КÜ e

will..

da.

dojdecame.3.

nekoj. (Dialectalsomebody

Mac)

‘Somebody seems to have come.’5

The subcategorisation properties of the Balkan subjunctive markers, and thoseof the Macedonian subjunctive markers in particular, differ from the subcat-egorisation properties of that-complementisers: while that-complementiserssubcategorise for IPs with subjects in specifier positions, the subjunctive mark-ers never occur to the immediate left of subjects, though subjects of subordi-nate clauses can occur to the immediate left of the subjunctive markers. This isillustrated in (8):

(8) a. Recesaid.3

(ti)you

dekathat.

(ti)you

siare.2..

muhim..

gohim..

dal. (Mac)given..

‘(S)he said that you (yourself) have given it to him.’b. Saka

wants(ti)you

da.

*tiyou

muhim..

gohim..

dadeš. (Mac)give.2.

‘(S)he wants you (yourself) to give it to him.’

The non-occurrence of subjects between auxiliaries and subjunctive comple-ments in Romanian led Dobrovie-Sorin (1994:15–47) to analyse Romanian

Olga Mišeska Tomic

IPs as V-initial and Romanian structures with Aux followed by a subjunctivemarker as CP/IPs. Thus, for (9a) she proposes the structure (9b):

(9) a. Arehas

sa.

vinacome.3.

Ion. (Rom)John

‘John has to come.’b. CP/IP

Auxi CP/IP

Ci IP

IPCli

V-Infli VP

NP V�

tv NP

The verb of the higher clause – the tensed auxiliary – is here adjoined to theCP/IP,6 while Infl is projected by the tensed verb of the lower clause. The mainargument for the analysis is the fact that in interrogative clauses the auxiliarycannot raise by itself, which Dobrovie-Sorin illustrates by the ungrammatical-ity of sentences such as (10):

(10) *Arehas

IonJohn

sa.

cînteplay...

laon

pian. (Rom)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’ occursto the left of the interrogative clitic li, which, in all the Slavic languages thathave it, is typically generated in C:

(11) Sakawants

li.

(toj)he

tiyou

da.

muhim..

gohim..

dadeš? (Mac)give.2.

‘Does he want you (yourself) to give it to him?’

Moreover, in Macedonian (and in the Balkan languages in general), structureswith post-verbal subjects are marked.


In view of the above facts, the Macedonian subjunctive markers are bestanalysed as heads of Mood Phrases to the left of TenseP. The structure of (12a)would be (12b):

(12) a. Sakawants

tiyou

da.

dojdeš. (Mac)come.2.

‘(S)he wants you (yourself) to come.’b.

CP

C’Spec

C AgrS/TenseP

Spec AgrS/Tense’

AgrS/Tense vP

v’Spec

v VP

V’Spec

pro V

saka

MoodP

Mood’

Mood

da

TenseP

Tense’

Tense vP

v’

v VP

V’Spec

ti V

dojdeš

Spec

Olga Mišeska Tomic

In sentences in which the subjunctive structure is preceded by a modal clitic,MoodP would appear in the complement position of a modality phrase andthe upper IP would be absent. The structure of (7b), for convenience repeatedas (13a), is given in (13b):

(13) a. КÜ ewill..

da.

dojdecame.3.

nekoj. (Dialectalsomebody..

Mac)

‘Somebody seems to have come.’b. ModP

Mod’Spec

Mod

e

MoodP

Mood’

Mood

da

TenseP

Tense’

Tense vP

v’

v VP

Spec V’

nekoj V

dojde

At an earlier stage, the Balkan Slavic modal clitics were tensed modal auxil-iaries, subcategorizing for subjunctive complements introduced by the sub-junctive marker da. A vestige of this subcategorisation has remained in theBulgarian modal šte ‘will’, which can have past tense morphology and, whenit does, inflects for person and number and takes da-complements:7

(14) a. Štjaxwould.1

da.

dojda. (Bulg)come.1.

‘I would have come.’


b. Štešewould.3

da.

dojde. (Bulg)come.3.

‘(S)he would have come.’

Having become tenseless and having lost their φ-features, the Macedonian aux-iliary modals turned into modal clitics, clinging to the verb which had previ-ously appeared in their da-complement. The latter verb remained as the onlyverb in the sentence and the da-marker disappeared. Accordingly, the loss ofthe subjunctive complements came as a consequence of the loss of tense andφ-features.

. Imperatives

The Mood node projected by the da-marker is also the node where impera-tive mood is checked. As illustrated in (15), the position of pronominal cliticsin imperative clauses differs from their position in clauses with modal clitics:in clauses with modal clitics the pronominal clitics are wedged between theseclitics and the verb, whereas in imperative clauses (in which no modal cliticsoccur), they appear to the right of the verb:

(15) a. КÜ ewill..

muhim..

goit..

dadešgive.2.

podarokot. (Mac)present-the‘You will give him the present.’

b. Dajgive.2.

muhim..

goit..

podarokot! (Mac)present-the

‘Give him the present!’

At first glance, one might assume that the Macedonian imperative verb be-haves like predicate nouns in that language, which, unlike the tensed verbs andl-participles, do not form local domains with the clitics and, when in clause-initial position, host the clitics to their right. Nevertheless, the imperative verbalways appears to the left of the pronominal clitics, whereas the predicate nom-inal does so only when there is no other constituent in this position. The po-sition of the imperative verb relative to the pronominal clitics is invariant,whether the verb occurs in clause-initial position, as in (15b) or is precededby one or more other constituents as in (16a) and (16b), respectively:

Olga Mišeska Tomic

(16) a. Tiyou

dajgive.2.

muhim..

goit..

podarokot! (Mac)present-the

‘You give him the present!’b. Ti

youutretomorrow

akoif

sakašwant.2

dajgive.2.

muhim.

goit..

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 sce-nario: The imperative verb is a [+V,+N] head and as such it does not form ex-tended local domain with the clitics to its left and is free to long-head move toMood, attracted by its strong features. Subsequently, in PF, the moved imper-ative verb and the pronominal clitics, which in syntax sit in their base-derivedpositions, form a single phonological word with the antepenultimate stresspattern characteristic for the language.

. Specific phonological behaviour

The strength of Mood has a distinct phonological reflex. Macedonian clausalclitics have been treated as prototypical examples of proclitic verbal clitics (cf.Spencer 1991). Tomic (1999, 2000) and Franks (2000), however, observe thatMacedonian clausal clitics are inherently neutral with respect to directional-ity of cliticisation. As argued in Tomic (1997, 2000, 2001), the directionalityof cliticisation in Macedonian depends on the values for the verbal features,[±V], and the nominal features, [±N], of the head of the clause: in clauseswith [+V,–N] heads, they are proclitic, in clauses with [–V,+N] heads, they areenclitic, while in clauses with [+V,+N] heads we find dual behaviour.8 Nev-ertheless, in clauses in which V is instantiated by imperative verbs, there isneither procliticisation nor encliticisation. In such clauses, the verb and the cl-itics form a single phonological word with the antepenultimate stress patterncharacteristic for the language:9

(17) a. DAJ –give.2.

mu –him..

go! (Mac)it..

‘Give it to him!’b. ZeMI –

take.2.

mu –him..

go! (Mac)it..

‘Take it from him!’


c. PokažiTE –show.2.

mu –him..

go! (Mac)it..

‘Show it to him!’d. Na –

‘here you are’.

MI –me..

ti –you.2..

go! (Mac)it..

‘Take it (my dear little one)!’

In (17a) the stress falls on the first and only syllable of the verb; in (17b) itfalls on the second syllable of the two-syllabic verb; while in (17c) it falls onthe last syllable of the four-syllabic verb. In each case, the stress falls on theantepenultimate syllable of the verb + clitics complex. The same stress-strategyis applied in (17d), where a monosyllabic verb is followed by three pronominalclitics and the stress falls on the clitic to the immediate left of the verb, whichconstitutes the antepenultimate syllable of the verb + clitics complex.

As argued in Tomic (2001), the clitics in the scope of an operator forma single phonological word with their host. The distinct phonological be-haviour of the clitics in the environment of the imperative verb follows fromthe fact that the imperative verb has moved to an operator in whose scope theclitics are.

. The relation between mood and negation

In Macedonian, the phonological behavior of the verb and the clausal clitics inthe scope of the negation and wh- operators is the same as that of the verb andthe clausal clitics in the scope of the imperative mood operator. As illustratedin (18) and (19), in clauses with [+V,–N] heads, the operators and the verb, orthe operators, the verb and the pronominal clitics wedged between them, formsingle antepenultimately stressed phonological words:

(18) a. ŠTO –what

sakaš?want.2

‘What do you want?’b. Što –

whatMU –him..

dade? (Mac)gave.2/3.

‘What did you/(s)he give to him?’

(19) a. NE –not

baramseek.1

NIšto. (Mac)nothing

‘I am not asking for anything.’

Olga Mišeska Tomic

b. Ne –not

mu –him..

GO –it..

davam. (Mac)give.1

‘I am not giving it to him.’

In (18a) and (19a), where the operators immediately precede the disyllabicverbs, 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 projectedby a modal (non-inflecting) clitic, we have analogous phonological behavior.In clauses with [+V,–N] heads both the negation operator and the modal cliticform a single antepenultimately stressed word with the pronominal clitics andthe verb:10

(20) a. Ne –not

КÜ E –will..

dojdam. (Mac)come.1.

‘I won’t come.’b. Ne –

notкÜe –will..

mu –him..

GO –it..

dadam. (Mac)give.1.

‘I won’t give it to him.’

In both (20a) and (20b) the stress falls on the syllable to the immediate left ofthe 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 ante-penultimately stressed single phonological word with the verb and the cliticsto its right. In this case, however, the verb occurs to the immediate right ofthe negation operator, while the clitics, if any, follow the verb. Examine thefollowing examples:

(21) a. NE –not

davajgive.2.

NIšto! (Mac)nothing

‘Don’t give (away) anything!’b. Ne –

notDAvaj –give.2.

muhim..

NIšto! (Mac)nothing

‘Don’t give him anything!’c. Ne –

notdaVAJ –give.2.

mu –him..

go! (Mac)it..

‘Don’t give it to him!’

In (21a) the disyllabic verb is not followed by any clitics and the stress falls onthe negation operator, which actually constitutes the antepenultimate syllable


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 stressfalls on the first syllable of the verb, which occupies the antepenultimate sylla-ble of the phonological word. In (21c), where the disyllabic verb is followed bytwo pronominal clitics, the stress falls on the second syllable of the verb, whichalso represents the antepenultimate syllable of the phonological word.

This pattern does not obtain in all the Macedonian dialects. Thus, inNorth-Western Macedonian the pronominal clitics occur between the nega-tion operator and the verb; they actually encliticise to the negation operator,which is stressed independently of the head of the clause. The North-WesternMacedonian counterparts of (21b) and (21c) are given in (22a) and (22b),respectively:

(22) a. NE ←not

muhim..

DAvajgive.2.

NIšto! (NWMac)nothing

‘Don’t give him anything!’b. NE ←

notmu ←him..

goit..

DAvaj! (NWMac)give.2.

‘Don’t give it to him!’

Franks (1998:55) sees two ‘realistic’ options for the cliticisation strategy inMacedonian imperative clauses: ‘Either (i) there is some kind of strong imper-ative 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 nonfiniteverb forms should move further up the tree than finite ones, Franks settles foroption (ii), assuming that ‘the functional impoverishment of nonfinite formscould 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 actionthemselves’ (p. 57). He also argues that the negation marker ne must in someway be attracting the verb and the relevant feature could be either strong orweak. In the former case the imperative verb raises to ne, picking up the cliticsas it goes; in the latter, the imperative verb does not raise and the clitics loweronto it instead.

Franks’ analysis is in many respects unsatisfactory. First of all, the dis-tinction finite: nonfinite verb forms has no significant import in Macedonian:with respect to cliticisation the l-participles behave like tensed verbs; the pastand passive participles have dual behavior – like tensed verbs or like predicatenouns; and imperative verbs in positive clauses occur to the left of the pronom-inal clitics, while in negative clauses we have different behaviours in differentdialects. Franks does not explain why the clitics encliticise rather than procliti-

Olga Mišeska Tomic

cise in either Standard or North-Western Macedonian, and, why in the lattercase, illustrated in (22), they encliticise to ne, leaving the verb, which ne shouldhave attracted, by itself. Neither does he explain why clitics in Standard Mace-donian negative imperative clauses, exemplified in (21), do not procliticise tothe verb (if they cannot encliticise to a weak ne) and have to lower. But aboveall, 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 Macedonianand North-Western Macedonian negated imperative constructions does followfrom a difference in the strength of the negation operator. As argued in Tomic(2001), in Standard Macedonian, Neg is weak and occupies the head positionof NegP. At PF, the negation operator forms a single phonological word withthe pronominal clitics and the verb, which, due to the strength of Mood, ispronounced to the left of the clitics. In North-Western Macedonian, on theother hand, Neg is strong and occupies the specifier position of NegP.11 Thepronominal 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 thesubjunctive marker da, auxiliary modals project a Modality Phrase and subcat-egorise for TenseP. At an earlier stage in the history of the language, all theBalkan Slavic modals were tensed and subcategorised for subjunctive com-plements introduced by subjunctive markers. Having become tenseless andhaving lost their φ-features, the Macedonian auxiliary modals stopped takingsubjunctive complements.

The Mood node projected by the subjunctive marker da is the node whereimperative mood is overtly checked; material evidence for this comes from thedistinct syntactic and phonological behaviour of the clitics in the environmentof the imperative verb. In negative imperative clauses with pronominal clitics,there is a dialectal distinction in the positioning of the pronominal clitics rel-ative to the verb. This distinction follows from a difference in the strength ofthe negation operator. In Standard Macedonian, Neg is weak and occupies thehead position of NegP; at PF, it forms a single phonological word with thepronominal clitics and the verb. In North-Western Macedonian, on the otherhand, Neg is strong and occupies the specifier position of NegP; at PF the cliticsencliticise to it.


Notes

. Macedonian has no infinitives. The third person singular present tense forms are used ascitation forms.

. The Macedonian clausal clitic cluster has six slots:

(i) 1

2

31/2. ⁄.

4.

5.

63. ⁄.

The third and the fifth slot are never filled in the same clause. The clitics in the other slotscan occur simultaneously and combine with either the clitics in slot 3, or with those in slot5. The order is always fixed.

. A joint Tense and Subject Agreement Phrase is argued for by the portmanteau morphwhich in all Balkan Slavic languages represents tense, person and number.

. Whereas the first and second person auxiliary clitics (singular and plural) are analysed asheads of Tense/AgrSP, the third person auxiliary clitics (also singular and plural) are analysedas heads of AuxP. Crucial evidence for the distinct analysis of the third person auxiliary cliticsis the fact that, in the clitic cluster, they occur to the right of the pronominal clitics, whilethe first and second person clitics occur to the left of the pronominal clitics. Moreover, thirdperson auxiliary clitics occur in clauses in which V is instantiated by passive participles, suchas (4b), but not in clauses in which V is instantiated by l-participles, such as (i).

(i) Nenot

biwould..

muhim..

goit..

*eis.

dalgiven.l-..

proektot. (Mac)project-the‘(S)he wouldn’t give him the project.’

. Macedonian sentences in which a subjunctive expression occurs in the complement po-sition of кÜe do not have future reference.

. Dobrovie-Sorin (1994:17) argues that the adjunction analysis of the auxiliary is possibleif we assume the folowing conventions:

i. Functional Coindexation: Coindex adjacent functional X0 categories.ii. 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, conse-quently, cannot take a da-complement:

(i) Štewill

(*da).

dojda. (Bulg)come.1.

‘I will come.’

(ii) Štewill

(*da).

dojde.come.3.

‘(S)he will come.’

Olga Mišeska Tomic

. 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 be-havior of clitics in clauses headed by l-participles, nouns and past participles is illustrated in(i), (ii) and (iii), respectively:

(i) Ste →are.2.

mu →him..

go →him..

SKInaletorn.l-.

PALtoto. (Mac)coat-the

‘You have, reportedly, torn his coat.’

(ii) BRAкÜa ←brothers

ste ←are.2.

mu. (Mac)him..

‘You are his brothers.’

(iii) Mu →him..

e →is.

SKInatotorn....


‘His coat is torn.’

(iii′) SKInato ←torn....

mu ←him..

eis.


‘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 oper-ators and the pronominal and/or auxiliary clitics. This distinction follows from the natureof the head of the clause: as argued in Tomic (2001), the head of the clause forms or doesnot 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 distinctfrom the cliticisation strategy in clauses without operators. In the former case the opera-tor(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 pronominalclitics procliticise to the head of the clause or encliticise to an element to their left, the stresspattern of the host remaining as it had been originally.

. The structural differences between Standard Macedonian and North-West Macedoniannegated imperative structures can be observed by comparing (i) and (ii):

(i) [NegP [ Neg’ [Neg ne] Tense/AgrSP]] (StMac)

(ii) [NegP[Spec ne] [NegP’ [Neg ] Tense/AgrSP]] (NWMac)

I am grateful to Lisa Cheng for pointing out that my analysis of the structural differencebetween Standard Macedonian and North-Western Macedonian negated clauses is reminis-cent of Zanuttini’s (1991, 1997) analysis of the differences between a variety of Italian di-alects, thus turning my attention to the possibility of analyzing the strong NWMac negationoperator as an XP in the Spec of NegP.


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 Morpho-

syntax Workshop. Bloomington: Indiana University, June 1998. (downloadable athttp://www.indiana.edu/∼slavconf/linguistics/index.html)

Franks, S. (2000). Clitics at the interface. In F. Beukema and M. den Dikken (Eds.), CliticPhenomena in European Languages (pp. 1–46). Amsterdam: Benjamins.

Rivero, M.-L. (1991). Long head movement and negation: Serbo-Croatian vs. Slovak andCzech. 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.Tomic, O.M. (1996). The Balkan Slavic clausal clitics. Natural Language and Linguistic

Theory, 14, 811–872.Tomic, O.M. (1997). Non-first as a Default Clitic Position. Journal of Slavic Linguistics, 5(2),

1–23.Tomic, O.M. (1999). On Clitichood. In I. Kenesei (Ed.), Crossing Boundaries (pp. 9–32).

Amsterdam: Benjamins.Tomic, O.M. (2000). On clitic sites. In F. Beukema and M. den Dikken (Eds.), Clitic

Phenomena in European Languages (pp. 293–317). Amsterdam: Benjamins.Tomic, 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

Aacquisition 10, 95–98, 105, 106, 113,

114, 126, 158, 186–190, 193–195,199, 200

adjectives 21, 188, 276adjunct 5, 6, 15, 57, 99, 134, 138,

145, 148, 153, 157, 160adjunction 238, 252, 275adjuncts 5, 57, 138, 145, 148, 153,

157adverb 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, 72agreement 35, 225–227, 229, 275AgrOP 78, 79, 91, 227, 228AgrSP 78, 79, 83, 84, 91, 92, 249,

254, 256–258, 263, 265, 275, 276Aktionsart 21–23, 25–27, 35, 40, 41,

43–47alethic 30, 38, 42, 46, 135, 154anti-additive 221, 222Aromanian 265aspect 21–23, 25–27, 33, 36, 45, 47,

60, 105, 109, 110, 115aspect phrase 44aspectual class 61aspectuals 155autism 11, 187, 195–199autistic 195–200Aux 8, 81, 140, 141, 144, 158, 159,

245, 250, 261, 263, 265, 266, 268,275

auxiliary modal 8, 261, 262, 265,269, 274

BBalkan 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

CCatalan 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 218cliticisation 270, 273, 274, 276complement 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

DDanish 36, 37, 79, 160definiteness 63deontic 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, 160dispositional 62, 63, 66, 67disyllabic 272, 273double-modal construction 86downward entailment 215dubitative 165, 166, 171, 176, 177,

182Dutch 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, 67dynamic 21, 33, 34, 61, 87, 95, 115,

116, 141, 186, 189

EEcho 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

Ffinite 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

GGerman 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

Hhead-movement 134, 225, 228head-to-head 142, 233hedging on truth 171historical 20–22, 76, 110hypothetical 86, 128, 166, 173, 175,

176, 180, 182hypothetical modality 165, 171–174,

180

IIcelandic 7, 8, 16, 79, 82, 87, 88, 154,

160identity function 143imperative 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 29indicative 165–173, 175, 177, 178,

180, 181, 183, 245–247, 249–251,253, 254, 257

individual-level 12, 14, 55, 61

infinitival morphology 11, 107, 117,120, 121, 127

input-based hypothesis 104intransitive 3, 4, 31, 34, 35, 41irony 157Italian 136, 151, 152, 159, 200, 219,

245, 276

JJamaican creole 145

LLanguage of Thought 155learnability 136, 145left periphery 83Lele 14, 166, 167, 169, 170, 176, 177,

183lexical modal 262, 265, 266, 274lexicalist 142LF 2, 9, 67, 133–135, 137, 140–144,

146–148, 150, 153, 156, 158, 241,242, 248, 255–257

LF-movement 148licensing 43, 106, 213, 219–222, 230,

238, 239, 247local domain 269, 270, 276locality 219–221

MMacedonian 5, 6, 261–263, 265–267,

269–271, 273–276marginal modal 84, 85Merge 109, 134, 143, 147, 148, 156Middle 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 83Minimalist 31, 76, 78, 79, 133, 134modal 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, 154modal reference effect 103, 107, 111,

113, 114, 120, 125modal 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, 272modality phrase 261, 263, 268, 274monadic 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, 274MoodP 262, 268morphological marking hypothesis

104morphophonological 158morphophonological 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

Nnecessity 1, 8, 9, 105, 115, 116,

138–141, 146, 151, 155, 157, 158,187, 188, 192, 194, 205, 273

necessity modals 139, 157neg-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 complements12

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

Oobligation 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

QQR 83, 99quirky case 7

Rraising 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, 155remnant VP-preposing 92, 97, 98Romanian 265, 266root 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, 254root infinitive 11, 103, 104, 106, 117root interpretation 1–3, 6–9, 12–14,

51, 52, 58–61, 63–65, 67, 70, 186,194, 195, 242

Sscope 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, 154Scots 144, 145, 152, 153, 159Scots English 144select 3, 67, 113, 134, 139, 142, 144,

148, 160, 242selection 3, 21, 38, 41, 140, 141, 144,

153semantic role 36–39, 41sequence of tenses 157Slavic 47, 255, 266Slovene 14, 245–247, 249–258

Subject index

Southern States English 145Specifier 78, 83, 84, 93, 96, 156, 157,

225, 227, 228, 232, 265, 274speech act 29, 42Spell-out 79, 83, 233, 234, 238Split Sign 134splitting 150stage-level 55, 59stress 159, 238, 241, 270–273, 276structural position 7–9, 36, 48subcategorisation 265, 268subject-orientation 52, 58, 63, 66subject-oriented 52, 58, 62, 66, 67,

90subjunctive 5, 165, 166, 177–179,

181, 207, 216, 246, 262, 265–269,274, 275

sympathy 52, 62systems interaction 166

Ttaboo 224, 227, 232tag 25, 137–139, 146, 147, 149, 157temporal adverb 137, 143, 157tense 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 158tensed verb 249, 262, 266, 269, 273,

276thematic 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

UUG 96, 134, 142, 145, 151

VV-to-I 96, 97, 99, 106V2 8, 77, 78, 86, 92, 225, 228, 229,

234, 236, 237, 241, 243verbal NPI 222, 229–232VO 14, 76, 78, 79, 82, 87, 88, 94, 99,

241

WWackernagel 247, 248, 254West Frisian 7, 15, 21, 40wide scope 35, 87–91, 93, 146, 149word order 14, 72, 76–79, 82, 166,

228

YYiddish 7, 15, 21, 40, 241

Name index

AAbraham, W. 3, 13, 19, 21, 24, 26,

27, 31–33, 200Akmajian, A. 128Anderson, L. 128Anderson, S. R. 158Astington, J. 189, 192Atance, C. 188Atkinson, M. 97Avrutin, S. 110

BBailey, B. L. 145, 158, 159Barbiers, S. vii, 1, 6–8, 12, 13, 21, 25,

30, 32, 51, 61, 65, 67, 69–72, 112,128

Baron-Cohen, S. 195, 198Battistella, E. L. 145Beal, J. C. 138, 156Behrens, H. 103, 104, 113, 123Belletti, A. 83Bellugi, U. 97Bennis, H. 6Beukema, F. vii, 14, 75Blom, E. 11, 103, 104, 107, 113, 116,

129Bloom, L. 111Bobaljik, J. 7Boser, K. 11, 103, 106Boškovic, Ž. 252Brennan, V. 6, 10, 15, 57, 58, 70, 156Brinton, L. 46Brody, M. 156Brown, K. 141, 144–146, 152, 153,

156, 158, 159Bryant, P. 190

Burzio, L. 3, 6Butterworth, G. 189Bybee, J. 128Byrnes, J. 188

CCarruthers, P. 189Cassidy, K. 200Charman, T. 198Choi, S. 193Chomsky, N. 13, 65, 76–78, 133,

247, 256Chung, S. 109, 128Cinque, G. 8, 9, 15, 83, 133, 136,

151–153, 156, 159Clark, H. 128Coates, J. 20, 141, 157, 158, 195Collins, C. 65Conradie, J. 31Cormack, A. vii, 9, 10, 14, 133, 134,

136, 142–144, 151, 156–160, 200

DDavidge, J. 192De Roeck, A. 196Denison, D. 76, 77, 86Diewald, G. 13Dobrovie-Sorin, C. 265, 266, 275Duff, M. 188Durbin, J. 24, 27, 30, 31

EEisenmajer, J. 198Embick, D. 255Ernst, T. 9, 143Espinal, M. T. 159

Name index

Estes, D. 190Extepare, R. 128

FFeldman, F. 6, 57, 58Ferdinand, A. 128Fischer, A. 32Flavell, J. 191Fleischman, S. 128Fodor, J. 155Foster, T. 95, 96Fox, D. 82, 87, 100Frajzyngier, Z. vii, 14, 165, 166,

170–172, 183Franks, S. 270, 273, 274Frith, U. 195, 198Frye, D. 189Furrow, D. 192

GGarrigues-Cresswell, M. 175, 183Gazdar, G. 146, 158Gee, J. 51, 53, 187, 214Geerts, G. 51, 53Gelman, S. 189Gelderen, E. van 46Gerhardt, J. 111, 187Gillette, J. 195Gillis, S. 107, 113Ginneken, J. van 103Giorgi, A. 107, 110Givón, T. 128Gleitman, H. 201Gleitman, L. 195, 200Golden, M. 14, 245, 252Gonsalves, J. 115Gopnik, A. 190, 191, 193, 200Gordon, P. 190Grimm, H. 194Guo, J. 189

HHaan, F. de 117, 128, 158, 159

Haegeman, L. 104, 106, 137, 150,159

Hale, K. 13, 65Han, C.-H. 246, 248, 255–257Happé, F. 198, 200Harris, P. 156, 189Heim, I. 66Hirschfeld, L. 189Hirst, W. 188Ho, S. 188Hoeksema, J. 70Hoekstra, T. 3, 11, 57, 103, 104, 108,

114, 127–129Hofmann, T. 2, 52, 56Hogrefe, J. 191Holmberg, A. 78, 99Hornstein, N. 83, 99Hurlburt, R. 198Hyams, N. 11, 103, 104, 108, 114,

127–129, 200

IIngram, D. 103, 106–108, 116Izvorski, R. 255

JJackendoff, R. 99, 143, 159Jacobs, J. 159Jespersen, O. 110Jordens, P. 117, 121

KKayne, R. 77, 78, 87, 88, 90–92, 94,

150, 258Keyser, S. J. 13, 65Klima, E. 97, 137, 150, 158, 214Klinge, A. 112Klooster, W. 6Koopman, H. 236–239Koopman, W. 76, 78Krämer, I. 104Kratzer, A. 66, 152, 186Krifka, M. 159Kursawe, C. 106

Name index

LLadusaw, W. 215Lahey, M. 111Larson, R. 13, 65Lasnik, H. 97, 247Lass, R. 96, 103–105, 127, 128Lasser, I. 103–105, 127, 128Lechner, W. 159Lederer, A. 195Leekam, S. 198Leslie, A. 189, 190, 195, 198, 200Li, P. 194Lightfoot, D. 76, 99, 144, 157Lust, B. 103Lyons, J. 1, 70, 128

MMacWhinney, B. 114Marantz, A. 255May, R. 87McDowell, J. 9, 10McGinnis, M. J. 6Menaugh, M. 146Millar, M. 145Miller, J. 100, 145Moerenhout, M. 82Moore, C. 189, 192Mycielski, J. 183

NNagle, S. J. 156Niemeier-Wind, K. 194Noveck, I. 188Nuyts, J. 196–198

OO’Neill, D. 187, 191, 193Olson, D. 189, 192Overdiep, G. S. 211

PPalmer, 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. 129Perkins, M. 188Perlmutter, D. 2, 52, 56Perner, J. 191, 198Pianesi, F. 107, 110Picallo, C. 8, 134, 144, 156, 157Pintzuk, S. 99Platzack, C. 16, 99Poeppel, D. 106Pollock, J.-Y. 144, 249Postma, G. 9, 14, 205Pratt, C. 190Premack, D. 189Prior, M. 198Pullum, G. K. 142Pure, K. 192

QQuirk, R. 141

RRadford, A. 10Reis, M. 19, 24, 32Riemsdijk, H. van 86, 99Rivero, M. L. 245–247, 251,

253–255, 257, 258, 265, 270Rizzi, L. 83Roberts, I. 8, 77, 78, 94, 96, 99, 144Roeper, T. 129Ross, J. R. 2, 52, 56, 129Roth, D. 195, 198Roussou, A. 8, 154Rowlett, P. 159Rullman, H. 159

SSantelmann, L. 103, 106Savasir, I. 187Sera, M. 188Seuren, P. 215Shatz, M. 187, 190, 192–194, 200

Name index

Shepherd, S. 187Sheppard, M. M. 14, 245, 252Silber, S. 192Slaughter, V. 191Smith, N. vii, 9, 10, 14, 133, 134,

136, 142–144, 151, 157–160, 200Smith, P. 189Smoczynska, M. 189Sodian, B. 191Spencer, A. 270Sperber, D. 147Sprouse, R. 24, 27, 30, 31Stephany, U. 98, 115, 187–189Stoett, F. A. 211Stowell, T. 5Streitberg, W. 211Swaay, H. A. J. van 211

TTackeff, J. 111Tager-Flusberg, H. 199, 200Terzi, A. 245–247, 251, 253–255,

257, 258Thaiss, L. 195, 198Thompson, W. 103, 106–108, 116Thráinsson, H. 6–8, 15, 154, 160Tieken-Boon van Ostade, I. 97Timberlake, A. 109, 128Toman, J. 252Tomic, O. M. 5, 261, 263, 270, 271,

274, 276Traugott, E. 86Tsimpli, I.-M. 158

UUd Deen, K. 107, 114, 129

VVerdam, J. 208, 209, 212Verwijs, E. 208, 209, 212Vikner, S. 6–8, 15, 99, 154, 160

WWarner, A. 6, 85, 99Weibegué, C. 183Weil, J. 29, 188Wellman, H. 189, 190, 192Wells, G. 187, 188, 194Weverink, M. 104Wexler, K. 106Whiten, A. 189Whitman, J. 103Wijnen, F. 103, 104, 107, 113, 114,

117, 127, 129Wilcox, S. 187, 193, 194Williams, E. 146Wilson, D. 147, 200Wimmer, H. 191Woodruff, D. 189Woolley, J. 190Wurff, W. van der vii, 14, 75–79, 82,

95, 96, 99Wurmbrand, S. 7, 242Wyngaerd, G. vanden 51, 53

YYaniv, I. 190

ZZanuttini, R. 159, 276Zifonun, G. 13Zwicky, A. M. 142

In the series LINGUISTIK AKTUELL/LINGUISTICS TODAY (LA) the following titleshave 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 dreiBeiträge von Helene Malige-Klappenbach. 1980.

2. EHLICH, Konrad & Jochen REHBEIN: Augenkommunikation. Methodenreflexion undBeispielanalyse. 1982.

3. ABRAHAM, Werner (ed.): On the Formal Syntax of the Westgermania. Papers fromthe 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 Barri-ers. 1990.

6. BHATT, Christa, Elisabeth LÖBEL and Claudia SCHMIDT (eds): Syntactic PhraseStructure 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 Swed-

ish. 1998.20. LAENZLINGER, Christopher: Comparative Studies in Word Order Variation. Ad-

verbs, 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 Move-

ment 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. R°UZICKA, Rudolf: Control in Grammar and Pragmatics. A cross-linguistic study.1999.

28. HERMANS, Ben and Marc van OOSTENDORP (eds.): The Derivational Residue inPhonological Optimality Theory. 1999.

29. MIYAMOTO, Tadao: The Light Verb Construction in Japanese. The role of the verbalnoun. 1999.

30. BEUKEMA, Frits and Marcel den DIKKEN (eds.): Clitic Phenomena in EuropeanLanguages. 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, Morphol-

ogy 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 Ger-

manic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax(Groningen, May 26-27, 2000)(Workshop). n.y.p.

barbiers,et.al.2002.modality and its interaction with the verbal system

Documents