Perspectives on Arabic linguistics VII: papers from the Seventh Annual Symposium on Arabic Linguistics

PERSPECTIVES ON ARABIC LINGUISTICS VII

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F. KONRAD KOERNER (University of Ottawa) Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Henning Andersen (Los Angeles); Raimo Anttila (Los Angeles) Thomas V. Gamkrelidze (Tbilisi); John E. Joseph (Hong Kong) Hans-Heinrich Lieb (Berlin); Ernst Pulgram (Ann Arbor, Mich.) E. Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Tokyo)

Volume 130

Mushira Eid (ed.) Perspectives on Arabic Linguistics VII

PERSPECTIVES ON ARABIC LINGUISTICS VII PAPERS FROM THE SEVENTH ANNUAL SYMPOSIUM ON ARABIC LINGUISTICS

Edited by

MUSHIRA EID The University of Utah, Salt Lake City

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

The paper used in this publication meets the minimum requirements American National Standard for Information Sciences — Permanence Paper for Printed Library Materials, ANSI Z39.48-1984.

Mushira Eid (ed.) Perspectives on Arabic Linguistics VII Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 130) ISBN 90 272 3633-X (Eur.) / 1-55619-584-2 (US) (alk. paper) © Copyright 1995 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 75577 • 1070 AN Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

Table of Contents

Foreword Introduction L

vii 1

AGREEMENT IN ARABIC

Arabic and Parametric VSO Agreement Naomi Bolotin

7

Internal and External Agreement in Quantified Construct States Mark S. LeTourneau

29

Parasitic Gaps in Arabic Wafaa Batran Wahba

59

Negation and Modality in Early Child Arabic Ibrahim Mohamed & Jamal Ouhalla

69

H. PERSPECTIVES FROM EXPERIMENT-BASED STUDIES

Morphological Structure and Lexical Processing: Evidence from Arabic Sabah Safi-Stagni

93

Experimental Investigations of Arabic Syllable Structure Bruce L. Derwing, Dilworth B. Parkinson & Richard A. Beinert

107

The Timing Structure of CVVC Syllables Ellen Broselow, Marie Huffman, Su-I Chen & Ruohmei Hsieh

119

III. DISCOURSE AND PHONOLOGICAL PERSPECTIVES

Topic Continuity in Arabic Narrative Discourse Ahmed Fakhri

141

On Vowel Shortening in Palestinian Arabic Munther A. Younes

157

Morpheme Edges and Arabic Infixation Michael L. McOmber

173

Index of Subjects

191

FOREWORD On March 5-6, 1993, the Seventh Annual Symposium on Arabic Linguistics was held at the University of Texas at Austin. The symposium was sponsored by the Arabic Linguistic Society, The University of Texas College of Liberal Arts, its Middle East Center, and its Departments of Linguistics and Oriental and African Languages and Literatures, and by the University of Utah's College of Humanities and the Department of Languages and Literature. A total of eighteen papers were presented at the symposium; of these, eight are published in this volume. Two other papers (Safi & Broselow et. al) were presented at the Eighth Annual Symposium but are included here for thematic reasons. The papers presented at the symposium were selected on the basis of an anonymous review of abstracts submitted to the Program Committee. The papers included in the volume were further reviewed by the editor before their final acceptance for publication. The transcription of all Arabic materials in the body of the papers follows the International Phonetic Alphabet or standard equivalents. The Arabic emphatics, however, are represented by a dot underneath the symbol, and long vowels as sequences of two vowels. The transliteration of Arabic names and titles follows accepted formats, with some simplification in the use of diacritics. We have used ' and ' for hamza (glottal stop) and 'ayn (pharyngeal approximant), respectively. The preparation and printing of the final manuscript was done using facilities available at the Middle East Center of the University of Utah, Salt Lake City. I would like to thank Joseph Aoun and Keith Walters for reading and commenting on some of the papers in this volume. I am also indebted to Tessa Hauglid, who has served as assistant editor as well as copy editor for this volume.

(

INTRODUCTION MUSHIRA EID University of Utah

For the first time in this series we have a volume that includes a section devoted to experiment-based studies of Arabic. Section Two consists of three such studies, reporting on results from different types of experiments designed to test the status of certain aspects of Arabic morphology (Safi-Stagni) and phonology (Derwing et al., Broselow et al.), and their implications for Arabic and other languages as well. Sabah Safi-Stagni's experiment is designed to provide evidence regarding the nature of morphological representation, i.e., how morphological information is stored in the (mental) lexicon, and the status of inflectional and derivational mophemes in Arabic. She focuses on the lexical processing of Arabic during a reading task performed by an aphasic patient suffering from both a reading disorder (Alexia) and a writing disorder (Agraphia). On the basis of the types of errors produced by this patient, she argues that lexical items are represented in their morphemic constituents. But, as she points out, her results are not as conclusive with respect to the second question she raises in her paper, namely, the status of inflectional vs. derivational morphemes. The experiments conducted by Bruce Derwing, Dilworth Parkinson, & Richard Beinert are designed to assess the status of various phonological units, e.g., segment, syllable, and tiers, crosslinguistically. Their paper reports on three experiments they carried out in Cairo, using Arabic disyllabic words to determine, among other things, the status of intervocalic consonants, both single and double, within syllable structure. The results point to an interesting difference in the contribution made by consonants vs. vowels: consonants appear

2

MUSHIRA EID

to count more more than vowels, and the C-Tier appears to make independent contribution whereas the V-Tier does not. The paper by Ellen Broselow, Marie Huffman, Su-I Chen, & Ruohmei Hsieh reports on results of an experiment designed to test the timing properties of 'heavy' and 'super heavy' syllables (CVV, CVC, CVVC) as recorded in the speech of four speakers from Cairo, Alexandria, Beirut, and Damascus. They find that long vowels tend to be shorter in closed than in open syllables, and coda consonants tend to be shorter after long vowels than after short vowels. This, they argue, would be expected if long vowels and coda consonants share the same mora. Their results, then, confirm the analysis of Arabic word-internal CVVC syllables as being bimoraic, with the final consonant sharing the second mora of the second vowel. In addition, this volume has a section devoted to agreement in Arabic. The papers in Section One deal with various aspects of agreement and other functional categories in Arabic. Naomi Bolotin's and Mark LeTourneau's papers both deal with agreement asymmetries of sorts. Bolotin addresses the well-known problem of subject verb agreement in SVO and VSO word orders, where the verb fully agrees with its subject in the former order but only partially (person and gender) in the latter. She relates this asymmetry in Arabic agreement to other agreement asymmetries across languages and proposes an explanation in terms of a parametric model of feature variation that correlates richness of morphology with overt movement of constituents. LeTourneau focuses on asymmetries in agreement between the verb and a quantified construct state (or quantifier-headed idafa), cases where the verb may agree with the quantifier or with its complement. He proposes to explain the asymmetry and the various agreement properties of this construction (construction specific idiosyncrasies, variable number agreement, and the dissociation of number agreement and case assignment) on the basis of NP-internal and NP-external agreement and the interaction of the three modules of GB: X-bar theory, government theory, and case assignment. Wafaa Batran Wahba examines the licensing of parasitic gaps in Arabic. She argues that three potential licensers may authorize the existence of parasitic gaps in a number of Arabic dialects: wh-trace, resumptive pronoun, and wh-in-situ. The licensing of a parasitic gap in Arabic, she claims, is a property of the location occupied by the trace,

INTRODUCTION

3

rather than the wh-trace itself. Both wh-resumptive pronouns and whin-situ's behave as variables at surface structure and logical form, respectively, and they may both license parasitic gaps in Arabic. She concludes that licensing of parasitic gaps by wh-resumptive pronouns entails that logical form is directly involved in accounting for the behavior of parasitic gaps. Her arguments are based on data from Standard and Jeddah Arabic. Ibrahim Mohamed & Jamal Ouhalla's paper is a welcome contribution to an area not sufficiently researched in Arabic, namely, child language acquisition. The paper is an investigation of the status of negation as a functional category in early child language (Palestinian). It introduces to the literature a set of new data which the authors use to then attempt to choose between two hypotheses regarding the status of functional categories in child language: the Continuity vs. Maturation Hypotheses. The two hypotheses differ in their account of differences between child and adult language with respect to functional categories. Is the difference to be attributed to difference in the feature composition of functional categories? Or is it attributed to a reduction in the actual sentence structure in early child language, a structure that lacks functional categories. Their findings appear to be consistent with either hypothesis. The analysis they provide, however, makes their results more in line with the Maturation Hypothesis since they account for the unique features of child language in terms of the absence of certain functional categories, e.g. AgrS, and the presence of others, e.g. Neg. The papers in Section Three deal with various aspects of Arabic structure, specifially, discourse, phonology, and morphology. They differ from the papers in Section Two in that they are not experimentoriented and from the papers in Section One in terms of their topics . Using Givón's (1983) approach to topic continuity, Ahmed Fakhri analyzes a story, al-Hajalah, by Al-Kabisi and compares the results with those obtained from an analysis of two expository texts. In addition to describing the nature of discourse topics in Arabic narrative, he contends that the discourse notion of topic continuity can be used to help motivate certain morphosyntactic properties of Arabic, including the neutralization of agreement, the presence or absence of waaw alhaal, and the usage of inna. Munther Younes examines vowel shortening in three varieties of Arabic. He argues that closed syllable

4

MUSHIRA EID

shortening is a phonological process that operates within the syllable to block the creation of impermissible syllable types. Open syllable shortening, on the other hand, operates within sequences of two syllables to maximize occurrences of the iambic foot. The last paper in this section, by Michael McOmber, takes issue with nonlinear morphological analysis. McOmber argues that nonlinear tiers and template driven systems are not necessary in a theory which includes 'morpheme edges'. He shows how infixation, which is usually problematic for linear models, would be handled in a theory with morpheme edges and illustrates the efficiency of this approach by providing derivations for Standard Arabic verb morphology. Some of the papers in this volume, then, represent research in the traditional areas of phonology, morphology, and syntax while some others represent areas that are relatively new as far as current Arabic linguistic research is concerned.

I

AGREEMENT IN ARABIC

ARABIC AND PARAMETRIC VSO AGREEMENT*

NAOMI BOLOTIN Harvard

University

0. Introduction Standard Arabic exhibits an inflectional pattern in which the verb agrees with the subject fully for SVO order, but only partially for VSO order (in person and gender, but not number). While the gender agreement for VSO order appears to be unique to Arabic, the existence of an agreement asymmetry between preverbal and postverbal word orders occurs crosslinguistically. In Section 1, data is presented to show that agreement alternations occur in a variety of languages and constructions. In Section 2, by way of historical background, a number of previous solutions for the Arabic agreement facts are described. In Section 3, a new solution is presented which can account for both the Arabic agreement facts, as well as the agreement facts from other languages, by explaining them in terms of a parametric model of feature variation that correlates richness of morphology with overt movement of constituents. 1. Agreement Asymmetries 1.1 Verbal agreement 1.1.1 Arabic Standard Arabic exhibits an agreement asymmetry in verbal sentences with full NPs. With SVO order, there is person, number,

*This paper builds on work in Bolotin (1992). I wish to thank Erich Groat and participants at the Symposium for helpful discussion.

8

NAOMI BOLOTIN

and gender agreement; with VSO order, there is only person and gender agreement. (1) a.

b.

(2) a.

b.

al-cawlaad-u naamuu the-boys-Nom slept-3mp "The boys slept." al-banaat-u nimna / the-girls-Nom slept-3fp / "The girls slept." naama / *naamuu slept-3ms / slept-3mp "The boys slept." naamat / *nimna slept-3fs / slept-3fp "The girls slept."

/ /

*naama slept-3ms *naamat slept-3fs

al-cawlaad-u. the-boys-Nom al-banaat-u. the-girls-Nom

1.1.2 Other languages There are also other VSO languages with an agreement asymmetry. In negative clauses in the Celtic language Breton, there is number agreement with SVO order, but not with VSO order.1 (3) a.

b.

ar vugale ne lennont /*lenn ket levrioù. the children particle read-3pl/ read-3sg not books "The children do not read books." ne lenn / *lennont ket ar vugale levrioù. particle read-3sg / read 3pl not the children books "The children do not read books." (Stump 1984:293)

A similar agreement asymmetry is also found in a number of SVO languages. For instance, several dialects of Italian have full agreement with SVO order, but only person agreement when the verb precedes the subject. Shown below are Fiorentino and Trentino, respectively.2 1Affirmative clauses in Breton exhibit no agreement with either order. Although Irish and Welsh also have no agreement with VSO order, no asymmetry occurs since SVO order is not permitted. Note that there are also VSO languages with full agreement, such as Berber. This will be discussed further in Section 3.2.2. (i) sqad-n timgharin tabrat. sent-3p women letter "The women sent the letter." (Ouhalla 1991:124) 2 The Rhaeto-Romance languages are another group of SVO languages that exhibit asymmetrical agreement behavior (Haiman & Benincà 1992:203-26).

ARABIC AND PARAMETRIC VSO AGREEMENT

(4) a.

b.

(5) a.

b.

Gli è venuto delle ragazze. is come some girls "Some girls have come." *Le son venute delle ragazze they are come some girls "Some girls have come." E' vegnú qualche putela. is come some girls "Some girls have come." *L'è vegnuda qualche putela. they are come some girls "Some girls have come."

9

(Brandi & Cordin 1987:121-22)

And in English, when the subject occurs postverbally, particularly when the verb form is contracted, there are speakers who use a singular verb form, as in (6c). (6) a. Three candidates are/*is running for office. b. There are/*is three candidates running for office. c. There're/(?)there's three candidates running for office.

There are also SOV languages with agreement asymmetries. In many dialects of Dutch and in the second person singular forms of Standard Dutch, shown below, when the subject precedes the verb, the verb exhibits agreement; when the subject follows the verb, the bare stem is used. (7) a.

b.

c.

d.

.. dat jij naar huis gaat /*ga that you to house go (2s) / go (inf) "that you are going home" Jij gaat /*ga naar huis. You go (2s) / go (inf) to house "You are going home." Vandaag ga /*gaat jij naar huis. today go (inf) / go (2s) you to house "Today you are going home." Wanneer ga /*gaat jij naar huis. When go (inf) / go (2s) you to house "When are you going to the house?" (Zwart 1993:309)

10

NAOMI BOLOTIN

1.2 Adjectival agreement Asymmetrical agreement patterns are not restricted to verbal elements alone. Fassi Fehri (1989) notes that predicate adjectives that follow the subject in Arabic must agree with them in number and gender, while those that precede the subject need only agree in gender.3 (8) a.

b.

c

antunna nabiilaat-un. you-fem.pl. generous.fem.pl.-nom. "You are generous." c ?a nabiilat-un /nabiilaat-un antunna? Q generous-fem.sing.-nom. / generous-fem.pl.-nom. you.fem.pl. "Are you generous?" (Fassi Fehri 1989:99)

1.3 Generalizations From this data, we can make three generalizations about the agreement asymmetries: 1) they occur crosslinguistically in languages unrelated to one another and with different word orders, 2) they occur crossconstructionally, and 3) the asymmetry is unidirectional (i.e., the impoverished agreement order is always the order with the verb or adjective preceding the other constituent).4 These facts suggest that this phenomenon is subject to some kind of structural constraint. And, since not all languages exhibit agreement asymmetries, this phenomenon must be one possible parametric option that is available to languages. In Section 3, a solution which captures these generalizations will be presented which correlates richness of agreement morphology with overt verb movement to particular projections.

3 Note that the attribute-head relationship does not appear to be subject to the same configurational constraint that the specifier-head relationship does: impoverished agreement can occur either with prenominal adjectives, as in Armenian, or with postnominal adjectives, as in German. 4 One crucial distinction between the Arabic agreement asymmetry and the agreement asymmetries found in the other languages is that for Arabic the impoverished agreement is the normal situation since VSO is the unmarked order, whereas for the other languages the VS order with impoverished agreement is the marked option. Note, however, that this is a separate issue from that of determining what licenses these agreement configurations.

ARABIC AND PARAMETRIC VSO AGREEMENT

11

2. Previous Solutions This section includes several previous syntactic solutions that have been proposed to account for the Arabic agreement phenomenon, grouped by the general types of mechanisms that they employ, by way of historical background.5 2.1 Incorporation Since VSO is the normal order in Standard Arabic, one common analysis of the agreement that occurs for SVO order is to assume it a pronoun serving as a placeholder for a topicalized subject that has subsequently been incorporated onto the verb. 6 This is the view adopted by the early Arab grammarians, who considered VSO structures to be verbal sentences with a verb-subject pattern and SVO structures to be nominal sentences with a topic-comment pattern (Levin 1985). The comment clause, in turn, consisted of a verb followed by a pronoun that was either implicit (as in the case of the masculine singular) or explicit (as with the feminine singular). Fassi Fehri (1989), working within a Goverment-Binding model (and Fassi Fehri (1988), within a Lexical-Functional framework), also accounts for the Arabic agreement as an incorporation phenomenon. He proposes a rule that case must be assigned consistently to the right in Arabic, hence requiring the verb to raise to I in both SVO and VSO structures, where it acquires inflectional features.

5

Although there may be pragmatic factors that affect a speaker's choice of agreement patterns, these are independent of the syntactic mechanisms which license such configurations. See Belnap (1991) for a sociolinguistic study of agreement variation in Cairene Arabic. 6 An incorporation analysis has also been proposed for Breton (Anderson 1982).

12

NAOMI BOLOTIN

DP (full NP)

SVO agreement and the agreement found in null subject sentences occurs when the pronominal subject in [Spec, VP] incorporates onto the verb. Since [Spec, VP] is the position of the thematic subject, Fassi Fehri assumes that any DP occurring in [Spec, IP] must be a topicalized element which is assigned nominative case by default. As a result, he maintains that all SVO sentences in Arabic have a focus interpretation to them.7 Demirdache (1991), too, adopts an incorporation analysis. She assumes that nominative subjects are topics base-generated in [Spec, TP] which are bound at LF to a resumptive pronoun originating in [Spec, VP] that subsequently moves to [Spec, AGRP] to receive case.8

7

Although Fassi Fehri (1989) does not address the gender agreement that occurs with VSO order, Fassi Fehri (1988) assumes that it comes from an affix that is specified only for gender which attaches to the verb whenever it occurs with a non-pronominal subject. 8 Her reason for assuming that [Spec, TP] is a topic position is because only definite NPs can occur preverbally.

ARABIC AND PARAMETRIC VSO AGREEMENT

13

T

T

AGRP Spec

AGR'

To account for the agreement facts, she assumes that there are two different agreement affixes in Arabic, an affix for number and an affix for person and gender.9 The number affix attaches to the subject noun phrase; when there is no lexical subject, as in the case of topicalized and null subject structures, this bound morpheme incorporates onto the verb.10 The person and gender affix, in contrast, is located in AGR, and the verb adjoins to it when it raises to this position.11 2.2 Different verb movements for each word order Another approach taken to explain the Arabic agreement facts is to posit different movements for the verb for each word order. Mohammad (1990) assumes that the verb raises from V to I for the 9

Benmamoun (1992) also proposes a solution in which there are separate agreement affixes, in his case one for number and one for person; both are specified for gender. 10 Thus, she proposes that what were formerly nominal plural markers (0 for singular, -aa for dual, -uu for plural) have been reanalyzed as verbal agreement. 11Note that while diachronically it may be that SVO order in Standard Arabic evolved by a topicalization/incorporation strategy, as proposed by these three analyses, it does not necessarily follow that synchronically all Arabic subjects are also topics. It is possible that at a certain stage in the language, what were formerly topics became reanalyzed as subjects. Note, too, that subjects may serve as focus elements (i.e., in structures involving movement) as well as topics; since the subject appears in the nominative in both constructions and the resumptive signalling topicalization may be deleted, the two may appear identical structurally (Cantarino 1974:455).

14

NAOMI BOLOTIN

VSO order, but lowers from I to V for the SVO order.12 He proposes that all Arabic sentences have a thematic subject in [Spec, VP] and an optional null expletive in [Spec, IP]; the verb agrees with the closest subject that c-commands it. For VSO sentences, V raises to I and thus agrees in features with the null expletive pro, which he assumes is third masculine singular. Since I is not c-commanded by the thematic subject that occurs in [Spec, VP], it cannot agree with it. For SVO sentences, I lowers to V, making the lexical subject in [Spec, VP] the closest c-commanding antecedent and hence resulting in full agreement. Mohammad assumes that the gender agreement is due to a non-syntactic rule.13

VP Spec V' thematic subject V

NP

Abd El-Moneim (1989) also explains the agreement facts by proposing different verb movements for each order. She assumes that VSO order is derived by having V raise and adjoin to I, and then having this structure raise to C so as not to violate an Empty C Filter she formulates for Arabic. 14 Agreement results when I copies the features of the subject, which she proposes is possible only in non12 In more recent frameworks (Chomsky 1991, 1992; Chomsky & Lasnik 1993) this option has been abandoned, and only raising is allowed. The reason for having a uniform direction of movement is that the structure of sentences with the same meaning will then be the same at the level of interpretation (LF) in all languages, a desired result. 13 He bases this on the fact that in conjoined NPs a linear rule operates that has the verb agree with the first NP in the sequence. 14 Movement of V to a presentential landing site also been proposed for Welsh (Sproat 1985; Sadler 1988).

ARABIC AND PARAMETRIC VSO AGREEMENT

15

adjoined structures. Since V has adjoined to I for VSO order, no agreement occurs.

SVO structures, which she assumes are IPs, involve no syntactic movement, but at PF, I and V are rebracketed. 15 Since I is not adjoined to V, it may copy the agreement features of the subject, and hence full verbal agreement results. The gender agreement that occurs with VSO order is assumed to be a lexical property that is basegenerated on the verb. 2.3 A parametric solution A third approach taken to account for Arabic agreement is to explain it within the context of a parameter controlling some other factor. 16 Ouhalla (1988, 1991, 1994) explains the agreement facts within a model in which the ordering of functional categories such as TNS and AGR is assumed to vary across languages.17 Following the 15

As Abd el-Moneim herself notes, she is treating the same process differently for each word order: "... I will handle INFL +VERB merging as a PF process in certain cases and as a syntactic process in other cases" [26]. 16 Although Koopman & Sportiche (1991) do not directly address the agreement facts in Arabic, they propose a parametric solution to account for the SVO/VSO word order alternation. They assume that there are two possible manners in which case may be assigned, in a specifier-head configuration or under structural government. A language like Arabic allows both, and hence movement of the subject is optional. Huybreght (1992) also proposes a parametric solution in which differences between Arabic and Celtic have to do with the status of operator bound variables. 17 Since Ouhalla (1988) developed the idea of having separate tense and agreement projections independently of Pollock (1989) and Chomsky (1991), he uses TNS rather than T to represent the tense projection.

16

NAOMI BOLOTIN

Mirror Principle of Baker (1988), which says that the ordering of syntactic categories should mirror the order of attachment of morphological affixes, he posits that TNS occurs higher than AGR in Arabic, but that the reverse is the case for a language like Italian.18 For both VSO and SVO word order in Arabic, the verb moves to AGR to acquire agreement, which consists of a default set of features in the case of VSO order (third person masculine singular), and a full set of agreement features for SVO order. The V+AGR complex then moves to TNS in order to receive tense features. For VSO order, the full NP occurs in the thematic [Spec, VP] position. He assumes that this position is assigned default nominative case, and thus there is no motivation for the NP to move.19

Spec (full NPi)

TNS' TNS

AGRP Spec

AGR' AGR

VP Spec PF°i

V'

J

NP

Like Demirdache (1991), he assumes SVO order consists of a preverbal topic base-generated in [Spec, TNSP] which is coindexed with a 18

Tense is assumed to occur higher than agreement in Arabic because in negated sentences tense is realized on the negative particle and agreement on the verb. (i) t-tullaab-u lam / Ian ya-ðab-uu the-students-Nom NEG.Past / NEG.Fut 3MImp-go-Agr (Benmamoun 1991:18-19) 19 At S-structure, a non-thematic null expletive pro must be inserted in [Spec, AGRP] in order to satisfy the Principle of Full Interpretation, which requires that a VP be licensed by a subject; at LF, the NP raises and replaces the expletive in order to form a semantic relation with the VP.

ARABIC AND PARAMETRIC VSO AGREEMENT

17

resumptive pronoun that originates as the thematic subject in [Spec, VP] and then raises to [Spec, AGRP] in order to be licensed. The topic agrees with AGR by virtue of being coindexed with this pronoun and is assigned default nominative case in [Spec, TP]. Null subject sentences are identical in structure to SVO sentences, but without an NP occupying the [Spec, TNSP] position, and hence also exhibit full agreement. For VSO languages which have full agreement, such as Berber, Ouhalla assumes that they do not have a default mechanism for case assignment as Arabic does, and hence the subject must raise to [Spec, AGRP] to receive case. Since it is then in a specifier-head relation with AGR, the two agree in features.20 Ouhalla assumes that the gender agreement that occurs with VSO order is a property of the default AGR, rather than a case of partial agreement of the verb with it subject. The default setting has the option of choosing either masculine or feminine gender. 2.4 Summary While all of the solutions summarized above account for the Arabic data, with the exception of Ouhalla (1988, 1991, 1994), none accounts for other languages or constructions in which an agreement asymmetry occurs. An incorporation analysis such as that assumed for Arabic, as Demirdache (1991:20) notes, will not work for SVO languages because this would require the subject to lower, which is precluded by current analyses. And, an analysis which posits different types of verb movements for each order will most likely require adopting separate derivations for SVO, VSO, and SOV languages. One goal of Universal Grammar is to posit solutions that can apply crosslinguistically. With that in mind, a parametric solution that can provide a unified account for all the agreement facts is presented in the following section.

20

Ouhalla assumes that the Celtic languages are structurally AGR-initial instead of TNS-initial because of their ordering of inflectional morphology. To explain their lack of agreement, he posits that they have a language-specific rule like that of Arabic which assigns default nominative case to [Spec, VP], thereby preventing the subject from raising to [Spec, AGRP].

18

NAOMI BOLOTIN

3. A Parametric Agreement System of Inflectional Features This section describes the Minimalist Program of Chomsky (1992), and then shows how crosslinguistic agreement alternations can be explained within such a model. Section 3.1 presents the necessary background about the framework. Section 3.2 shows how this model can be used to account for the Arabic agreement facts. In Section 3.3, the model is extended to adjectival agreement, and in Section 3.4, to other languages with agreement asymmetries. 3.1 The minimalist program The Minimalist Program (Chomsky 1992) assumes that language consists of a lexical component and a computational component. The lexicon specifies the idiosyncratic properties of words, including their morphology; these items then enter the syntax fully inflected. The purpose of the syntactic or computational component is to ensure that the derivation is legitimate by checking that various conditions are met, among them the presence of appropriate morphological features. In this model, lexical items may be drawn freely from the lexicon throughout the formation of a sentence, and thus a level of Deep Structure is not longer necessary; there are just two syntactic levels: Spell-Out and LF. Following Chomsky (1991), sentences are assumed to have the following structure. There are two agreement projections, one for the subject and one for the object, and a tense projection. Since morphological processes are now relegated to the lexicon, the ordering of the functional categories for the purpose of checking features in the syntax may be assumed to be the same crosslinguistically.

ARABIC AND PARAMETRIC VSO AGREEMENT

19

There are four inflectional parameters, one for each set of features: nominal features of AGR, verbal features of AGR, nominal features of T, and verbal features of T. Features may either be strong or weak; strong features are visible but not interpretable and thus need to be checked before Spell-Out; weak features are not visible and thus need not be checked until LF. LF movements are preferable by Chomsky's Procrastinate principle, which advocates delaying operations as long as possible. Every NP and V must have its features checked with the appropriate functional head; if they match, the features on that head disappear (since their sole role was to mediate checking), the derivation converges and the structure is legitimate. Verbs need to check both tense and agreement, which they do by adjoining to the relevant head, T or AGR. Noun phrases need to check case and agreement by moving to the appropriate specifier position; since nominative case is assigned by T and accusative by V, case is checked in the configuration T+AGRs or V+AGRo, respectively (i.e., T or V must first raise to the relevant AGR projection).21 21 In this framework, all feature checking, including checking of object agreement, is done in a spec-head or head-head configuration and hence the notion of government no longer plays a role. Since the ø features for both AGR projections are assumed to

20

NAOMI BOLOTIN

Although the idea of having strong and weak inflectional features that need checking was originally developed to explain distinctions between English and French, since these parameters are assumed to be part of Universal Grammar, the theory predicts that they should apply to all languages. As the remainder of this section shows, this is in fact the case. 3.2 Parametric VSO agreement There are two types of Arabic agreement variation that occur, an SVO/VSO alternation within the language itself, and a rich VSO/poor VSO agreement alternation between Arabic and a language like Berber. The parametric variations for each alternation will be illustrated, and then generalizations will be drawn from these. 3.2.1 The Arabic SVO/VSO alternation The SVO/VSO alternation that occurs in Arabic can be explained by assuming that there is one set of inflectional parameters for each word order. 22 The tense settings remain the same for both word orders, while the agreement settings change. (15)

VSO order

SVO order

V features of T V features of Agr

strong weak

strong strong

N features of T N features of Agr

weak weak

weak strong

For VSO order, the verbal features for tense are strong and thus the verb must raise to T before Spell-Out to have its tense features checked off. Since the verbal features on AGR are weak, the verb does not raise to this position until LF by the Procrastinate principle.

be the same, it is the presence of the adjoined head (T or V) that ensures that the correct case is assigned. 22 Chomsky (1992:44) himself suggests such a possibility.

ARABIC AND PARAMETRIC VSO AGREEMENT

21

The subject likewise remains in situ at Spell-Out but raises to [Spec, AGRP] at LF to check its features.23

The gender agreement that occurs for VSO order can be accounted for easily in this model: it is part of the AGR features that must occur for this word order, an idiosyncratic property which is specified in the lexicon.24 Since words enter the syntax fully inflected, this feature must be present on the verb in order for it to match the features in AGR, and hence for the derivation to converge. SVO order occurs when the verbal and nominal features on Agr are strong, requiring the verb to move to AGR25 and the subject to move to [Spec, AGRP] before Spell-Out.26 23

0bject agreement and the AGRo projection are ignored throughout since they do not bear directly on the data at hand. For discussion of this projection in Arabic, see Ouhalla (1991b). 24 The reason why gender, rather than number, happens to be the extra feature that is specified seems to be because gender is a more intrinisic property than number: a noun's gender stays constant, whereas its number usually depends on the context. The fact that Arabic also has feminine singular agreement with inanimate plurals may explain historically why gender agreement has been maintained across the verbal agreement system of the language. 25 Note that since the verb must raise and adjoin to T in moving to AGR, so as not to violate minimality (Rizzi 1990), this means that it does not matter whether the setting for the verbal feature of T is strong or weak. Thus, rather than assuming the setting to be the same for both word orders, another option is that the value of this parameter is left unspecified in instances such as this. 26 As formulated in Chomsky (1992), the subject moves directly to [Spec, AGRP] to check both its agreement and case features since T has adjoined to AGR, and thus no [Spec, TP] actually exists. This means that the value of the parameter governing the

22

NAOMI BOLOTIN

(17)

AGRP Spec

AGR' AGR

TP

V

NP

The two derivations for Arabic will thus have the following structures, where the traces of moved elements are indicated by the relevant subscript.27 (18) a. SVO [AGRP NP [AGR'V+T+AGR [TP tT [VP tNP [ V tv]]]]] b. VSO [TP V+T [yp NP [ y tV ]]]

3.2.2 Rich VSO vs. poor VSO agreement The agreement alternation that occurs between poor VSO agreement languages like Arabic and rich VSO agreement languages like Berber can be explained by assuming the following parameter values. All settings remain the same, except for the verbal features of AGR.

nominal features for tense is irrelevant. Note, however, that Bobaljik & Carnie (1992) suggest that [Spec, TP] may in fact be available for the subject to check its case features there. One reason why assuming the existence of a [Spec, TP] position is desirable for Arabic is that this provides a structural way to distinguish focus from topic, the former being located in [Spec, TP] and the latter in some higher projection ([Spec, AGRsP] if all Arabic subjects are in fact topics, [Spec, CP] otherwise). 27 The internal structure of the verb complex is [[V+T]+AGR], assuming left adjunction. Note that when the verb raises to AGR for SVO order, it is actually the complex V+T that raises, rather than just the verb. Since T is the head of this complex, presumably the verb must be able to percolate its features up to T so that they are visible for checking.

ARABIC AND PARAMETRIC VSO AGREEMENT poor VSO (Arabic)

rich VSO (Berber)

V features of T V features of AGR

strong weak

strong strong

N features of T N features of AGR

weak weak

weak weak

(19)

23

For Arabic, the verb moves to T before Spell-Out. All other movements take place at LF: the verb to AGR, and the subject to the specifier position of AGR. The situation is the same for Berber, except that the verb must raise to AGR in the overt syntax. (20) a. poor VSO (Arabic)

b. rich VSO (Berber)

The two VSO derivations will have the following structures. (21) a. Arabic VSO [TP V+T [Vp NP [V tv ]]] b. Berber VSO [AGRP V+T+AGR [TP tT [VP NP [ y t v ]]]]

3.2.3 Generalizations By comparing the settings for all three language types, it can be seen that parametric VSO agreement stems from the setting of the parameter for the verbal features of AGR: if they are strong, there is overt movement of the verb to AGR and hence rich verbal agreement; if they are weak, there is no overt movement, and thus verbal

24

NAOMI BOLOTIN

agreement is poor.28 Word order, variation, in contrast, is determined by the strength of the nominal features of AGR. If these are strong, an SV order occurs; if they are weak, a VS order occurs. poor VSO (Arabic)

rich VSO (Berber)

rich SVO (Arabic)

V features of T Vfeatures of Agr

strong weak

strong strong

strong strong

N features of T N features of Agr

weak weak

weak weak

weak strong

(22)

Summarizing, we get the following generalization regarding agreement for these three language types: (23)

strong V features on AGR weak V features on AGR

Arabic SVO; Berber VSO Arabic VSO

And for word order: (24)

strong N features on AGR weak N features on AGR

Arabic SVO Arabic VSO; Berber VSO

Thus, feature strength necessitates overt movement, which in turn determines word order and agreement properties. 3.3 Adjectival agreement The parametric model of feature variation should also be applicable to other structures that have such inflectional features. 28

The fact that in languages with asymmetries SV order is the one with rich agreement and VS the one with poor agreement also follows from the theory. Strong nominal features on AGR entail not only overt subject movement, but also overt object movement. To explain how the object can cross over the [Spec, VP] subject position as it moves to [Spec, AGRoP] without violating minimality, Chomsky assumes that the verb moves to AGRo first. In doing so, it extends its domain, thereby rendering all potential landing sites in that domain equidistant in terms of movement. In order for such verb movement to occur, the verbal features for AGR must be strong in addition to the nominal features for AGR. This explains why the SV order will always have rich agreement; for the VS order, in contrast, there is no such demand and hence the verbal features for AGR may either be strong or weak.

ARABIC AND PARAMETRIC VSO AGREEMENT

25

Chomsky (1992:12) proposes that there is an adjective agreement phrase ( A G R A P ) located above the adjective phrase (comparable to the AGRoP phrase located above the verb phrase).29 Using the same parameter settings that apply to verbal agreement, Arabic asymmetries involving adjectival agreement can also be accounted for. For subject adjective order, since both the verbal/adjectival features of AGR and the nominal features of AGR are strong, the adjective will raise to A G R A and the subject to [Spec, A G R A P ] . Once again, since the head (in this case, an adjective) is in AGR, rich agreement occurs. No such movement will occur for adjective subject order. (25)

AGRP Spec

AGR" AGRA

Spec NP

AP

A' I

A 3.4 Agreement asymmetries in other languages The agreement asymmetries in the other languages shown in Section 1 can also be characterized by assuming that, as in Arabic there are two different parameter settings, one for each order, which hold consistently across the language. All will share in common the fact that the verbal features on AGR will be strong for the SV order and weak for the VS order. In the case of languages such as Breton, where the asymmetry does not hold consistently across the language and thus cannot be captured by just one parametric alternation, the verb will still be assumed to be in AGR when rich agreement occurs, although perhaps something else might be responsible for triggering movement in such an instance.30 29

This is turn is dominated by the AGRsP and TP verbal projections. Stump (1984) suggests that the subject in Breton is outside the sentence for negative SVO clauses, but inside the sentence for affirmative SVO clauses. In terms of the model presented here, presumably the verb would be in AGR for the former and in some lower position for the latter. 30

26

NAOMI BOLOTIN

4. Conclusion The agreement asymmetry that occurs in VSO languages like Arabic and Welsh, SVO languages like Italian, and SOV languages like Dutch, can now be accounted for by a single common parameter setting: strong vs. weak verbal features on AGR. (26)

weak verbal features on AGR strong verbal features on AGR

poor agreement rich agreement

Strength of nominal features of AGR can likewise be used to account for word order variations. A nice consequence of such a system is that crosslinguistic commonalities across different types of languages can now be accounted for with one solution. As linguistic theory continues to develop, and as new data becomes available, this parametric model will inevitably undergo modification. The hope, however, is that the idea of treating Arabic agreement as part of a larger crosslinguistic phenomenon will remain.

REFERENCES Abd El-Moneim, Aliaa Ahmed. 1989. The Role of INFL. Ph.D. dissertation, University of Connecticut, Storrs. Anderson, Stephen. 1982. "Where's Morphology?" Linguistic Inquiry 13.571-612. Baker, Mark. 1988. Incorporation. Chicago: The University of Chicago Press. Belnap, R. Kirk. 1991. Grammatical Agreement Variation in Cairene Arabic. Ph.D. dissertation, University of Pennsylvania, Pittsburgh. Benmamoun, Elabbas. 1991. ''Negation and Verb Movement". Proceedings of NELS 21 ed. by Tim Sherer, 17-31. . 1992. "Structural Conditions on Agreement". Proceedings of NELS 22, 1732. Bobaljik, Jonathan & Andrew Carnie. 1992. "A Minimalist Approach to Some Problems of Irish Word Order." Paper presented at the Xllth Harvard Celtic Colloquium. To appear in Celtic and Beyond ed. by Robert Borsley & Ian Roberts. Cambridge: Cambridge University Press. Bolotin, Naomi. 1992. "Arabic Agreement and Functional Projections". Harvard Working Papers in Linguistics I ed. by Susumu Kuno & Höskuldur Thráinsson, 33-49. Brandi, Luciana & Patrizia Cordin. 1989. "Two Italian Dialects and the Null Subject Parameter". The Null Subject Parameter ed. by Osvaldo Jaeggli & Ken Safir, 111-42. Dordrecht: Kluwer.

ARABIC AND PARAMETRIC VSO AGREEMENT

27

Cantarino, Vicente. 1975. Syntax of Modern Arabic Prose, Volume III. Bloomington: Indiana University Press. Chomsky, Noam. 1991. "Some Notes on Economy of Derivation and Representation". Principles and Parameters in Comparative Grammar ed. by Robert Freidin, 417-54. Cambridge, Mass.: MIT Press. . 1992. "A Minimalist Approach to Linguistic Theory". MIT Occasional Papers in Linguistics, 1. Reprinted in The View from Building 20: Essays in honor of Sylvain Bromberger ed. by Kenneth Hale & Samuel Jay Keyser, 1-52. 1993. Cambridge, Mass.: MIT Press. Chomsky, Noam & Howard Lasnik. 1993. "The Theory of Principles and Parameters". In Syntax: An international handbook of contemporary research ed. by Joachim Jacobs et al., 506-569. Berlin: Walter de Gruyter. Demirdache, Hamida. 1991. "Nominative Subjects in Arabic". Ms., MIT. Fassi Fehri, Abdelkader. 1988 "Agreement in Arabic, Binding, and Coherence". Agreement in Natural Language ed. by Michael Barlow & Charles Ferguson, 107-58. Stanford: Center for the Study of Language and Information. . 1991. "Generalized IP Structure, Case, and VS Word Order". MIT Working Papers in Linguistics 10 ed. by Itziar Laka & Anoop Mahajan, 75-111. Haiman, John & Paola Benincà. 1992. The Rhaeto-Romance Languages. London and New York: Routledge. Huybregts, Riny. 1991. "Allosteric Agreement in VSO Languages". Linguistics in the Netherlands ed. by Frank Drijkoningen & Ans van Kemenade, 81-90. Amsterdam & Philadelphia: John Benjamins. Koopman, Hilda & Dominique Sportiche. 1991. "The Position of Subjects". Lingua 85.211-258. Levin, Arye. 1985. "The Distinction between Nominal and Verbal Sentences According to the Arab Grammarians". Zeitschrift für arabische Linguistik, Journal of Arabic Linguistics 15.118-127. Mohammad, M. A. 1990. "Subject-Verb Agreement in Arabic: Towards a solution". Perspectives on Arabic Linguistics I ed. by Mushira Eid, 95-125. (= Current Issues in Linguistic Theory, 63.) Amsterdam & Philadelphia: John Benjamins. Ouhalla, Jamal. 1988. The Syntax of Head Movement: A case study in Berber. Ph.D. dissertation, University of Essex. . 1991. Functional Categories and Parametric Variation. Oxford: Oxford University Press. . 1994. "Verb Movement and Word Order in Arabic". In Verb Movement ed. by David Lightfoot & Norbert Hornstein, 41-72. Cambridge: Cambridge University Press. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar, and the Structure of IP". Linguistic Inquiry 20.365-424. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Sadler, Louise. 1988. Welsh Syntax: A government binding approach. London: Croom Helm. Sproat, Richard. 1985. "Welsh and VSO Order". Linguistic Inquiry 16.173-216. Stump, Gregory. 1984. "Agreement vs. Incorporation in Breton". Natural Language and Linguistic Theory 2.289-348. Zwart, C. Jan-Wouter. 1993. "Verb Movement and Complementizer Agreement". Papers on Case and Agreement I ed. by Jonathan Bobaljik & Colin Phillips, 297-340. (= MIT Working Papers in Linguistics, 18.)

INTERNAL AND EXTERNAL AGREEMENT IN QUANTIFIED CONSTRUCT STATES* MARK S. LETOURNEAU Weber State University

1. Introduction to the Problem The construct state in Modern Standard Arabic (hereafter MSA) is a familiar and relatively well-studied construction. A typical example, from al-Aboudi (1987:15), is given in (1): (1) qutila [NP zapiim-u pal-mupaaradat-i] was killed-3ms leader-NOM the-opposition-GEN "The leader of the opposition was killed."

The NP in (1) illustrates the salient properties of the construction, summarized as (2): (2) a. It consists of two members, X and Y, the first of which is the head. b. X may take neither the definite determiner nor nunation, and Y may take either. c. X may be assigned nominative, accusative, or genitive case, depending on the grammatical function of the NP, while Y may be assigned only genitive case (al-Aboudi 1987:15-18).

The first term of the construct state may be a quantifier, as in (3): (3) ra?ay-tu [kull-a ?al-rijaal-i] saw-lsg all-ACC the-men-GEN "I saw all the men." *I would like to thank Professors Joseph Aoun and Mushira Eid for their comments on earlier drafts of this paper.

30

MARK S. LETOURNEAU

I will henceforth refer to a case like (3) as a quantified construct state (QCS). QCSs in MSA exhibit a wider range of agreement patterns than do either ordinary CSs or NPs in general. A subject NP in MSA agrees with the verb only in gender when the verb precedes the subject and in both gender and number when it follows, as in (4) (from Mohammad 1990a:95): (4) a. jaa?-a ?al-?awlaad-u came-3mpl the-boys-NOM b. al-?awlaad-u jaa?-uu the-boys-NOM came-3mpl c. *jaa?-uu ?al-?awlaad-u came-3mpl the-boys-NOM

By contrast, a verb may agree in number and gender with either the quantifier (invariantly masculine singular) or with its complement, yielding the four patterns in (5) (where Q and N abbreviate the first and second terms of the QCS), respectively instantiated in (6a-d), with the agreeing terms in italics:1 (5) a. b. c. d.

Verb preceding QCS and agreeing with Q Verb preceding QCS and agreeing with N Verb following QCS and agreeing with Q Verb following QCS and agreeing with N

(6) a. yahduθu bapd-u ?alqalaaqil-i occur-3ms/IMPF some-NOM the-disturbances-f-GEN "Some disturbances occur". b. kaan-at kull-u muhaadaθ-at-i-n tabtad-ii?-u was-3fsg all-NOM conversation-f-GEN-INDEF begin-3fsg/IMPF "Every conversation used to begin. ..". c. kull-u-haa yuhaddiθu pan-hu all-NOM-her speak-3msg/IMPF about-him "Everything speaks of him."

1(6a) is from Parkinson (1975:60); (6b-d) are from Cantarino (1975:154-55).

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

31

d. kull-u-haa xadap-at li-1-hukm-i ?alislaam-iyy-i all-NOM-her accept-3fsg to-the-rule-GEN the-Islamic-GEN "All of them accepted the Muslim rule".

Furthermore, certain grammatical features on the second term in the QCS license only one agreement option. Thus, if the second term to kull is either an indefinite feminine singular or a definite plural, the verb must agree with the second term (logical agreement, in traditional terms); if bapd has a pronominal suffix and the verb follows, agreement with the quantifier (grammatical agreement) is mandatory (Parkinson 1975:66). The examples (l)-(6) raise several questions. First, how can one account for the idiosyncracies of the QCS listed in (2)? These, like all construction-specific properties, should follow from general principles. Second, why are there more possible agreement patterns in a QCS than in an ordinary CS or NP, and why is there a further narrowing of agreement options for certain types of QCS? Third, in cases in which the verb agrees with the 'logical' subject, we observe a curious dissociation of number agreement from morphological casemarking. Normally, a verb agreeing with the subject agrees with an NP bearing nominative case; in (6b), however, number agreement is with an NP bearing genitive case. Why should this be so? I propose that these three properties—construction-specific idiosyncracies, variable number agreement, and dissociation of number agreement and case assignment—follow from the interaction of three modules of government-binding (GB) theory: X-bar theory, government theory, and Case theory. The discussion divides into two agreement domains: NP'-internal and NP-external agreement (Lehman 1988:58; Barlow 1992:137, citing Moravcsik 1978:340). Section 2 analyzes internal agreement within the QCS between the first and second terms following the Determiner Phrase analysis of NPs. Section 3 analyzes external agreement between the subject and the verb (mediated by functional heads introduced there). The questions posed above are answered in Section 4, and Section 5 summarizes the argumentation of the preceding sections.

32

MARK S. LETOURNEAU

2. Internal Agreement in Quantified Construct States 2.1 Quantifiers as X0 Lexical Heads The traditional parse of quantifiers like bapd "some", and kull "all", as nouns suggests a phrase structure like (7) for QCSs: (7) [NP [N' [N kullu [NP almudanisiina]]

Let us assume, as a first approximation, that kull is the head of the NP for purposes of agreement; we return in 2.2 to the question of whether it is the head for (semantic) selection as well (Fehri 1988:135; Chomsky 1986a:86-92). Taking the quantifier as the head of the phrase implies, in X-bar theoretic terms, that it is the 0-level lexical category from which the NP is projected. The general X-bar schemata are of the form (8), from Chomsky (1986b:4), "where X* stands for zero or more occurrences of some maximal projection and X = X0": (8) a. X' = XX"* b. X" =X"X'*

The schema (8a) indicates that a 0-level category X may combine with one or more maximal projections as its complement to form the intermediate category X'; (8b) indicates that one or more X's may in turn combine with a maximal projection as its (their) specifier to form the maximal projection of X, X". The head of a phrase is the X0 which the phrase immediately dominates; note that the head is obligatory, while specifiers and complements are optional. The optionality of specifiers and complements, at least in the unmarked case, predicts that if the quantifier is the head of its NP, then (8) should license the following sequences of functional categories for QCSs: (9) a. b. c. d.

Spec-Head Comp (SHC) Spec-Head (SH) Head-Comp (HC) Head (H)

If only heads are obligatory, then one would expect, other things being equal, that all four patterns should be attested: the head should

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

33

be able to cooccur with both a specifier and a complement, as in (9a); with either a specifier or a complement alone, as in (9b), (9c), respectively; or by itself, and on a nonelliptical reading, as in (9d). Of the four sequences, (9b-d) are possible, (9a) is not. In this quantifers closely parallel lexical Ns heading ordinary CSs, as also noted in Benmamoun (1993); I return to differences in 2.2. The respective functional sequences, exemplified by bayt "house", and kull for comparison, are given in (10) and (11): (10)

(11)

a. SHC: *al-bayt-u alkunuuz-i the-house-NOM the-treasure-pl-GEN "the house of treasures" b. SH: al-bayt-u the-house-NOM "the house" c. HC: bayt-u al-mudiir-i house-NOM the-director-m-GEN "the director's house" d. H: ra?ay-naa bayt-a-n saw- lpl house-ACC-INDEF "We saw a house". a. SHC: *jaa?-a al-kull-u taalib-i-n cameras the-all-NOM student-GEN-INDEF "Every student came". b. SH: ?akal-at al-kull-a ate-3fsg the-all ACC "She ate the whole thing". c. HC: ?akal-at kull-u altufaah-i ate-3fsg all-NOM the apple GEN "She ate the whole apple". d. H: palaa kull-i-n upon all-GEN-INDEF "in spite of it all" (Cantarino 1975:125)

Putting aside the ungrammaticality of the (a) strings, which are instances of (2b) to be discussed in 2.2, the distributional similarity between quantifiers and lexical nouns establishes that the former, like the latter, are X0 heads of noun phrases, a parallelism which is to be expected if one adopts the traditional parse. There remains, however, a seemingly important difference: lexical nouns are a morphologically open class, while quantifiers are

34

MARK S. LETOURNEAU

closed-class items (Ritter 1991:58). In this respect, quantifiers resemble other closed-class categories such as complementizers and inflection, which head functional rather than lexical projections. Let us now consider whether the similarities between quantifiers and nouns, on the one hand, and between quantifiers and determiners, on the other, can be explained by treating quantifiers as heads of functional projections. 2.2 Quantifiers as X0 Functional Heads Recent research in X-bar theory (e.g., Stowell 1989) has put forth the hypothesis that noun phrases should be analyzed as Determiner Phrases (DPs), the maximal projection of a functional head D and complement NP, on analogy with the reanalysis of clauses as projections of Inflection (yielding Inflectional Phrase, or IP) and sentences as projections of Complementizer (Complementizer Phrase, or CP), respectively (Chomsky 1986a:3). This hypothesis has recently been proposed for Arabic (Benmamoun 1993; Kaplan 1993:202). On this analysis, (7) would have the phrase structure (12): (12)

[DP [SPEC] [D' kullu [NP al-mudarrisiiyna]]]2

Ritter (1991) extends the DP analysis to propose that in Modern Hebrew (MH) construct states (analogous to those in MSA) D0 does not select NP directly as its complement; rather, it selects a second functional head, Number Phrase (NumP). Ritter's analysis of Modern Hebrew construct states offers a way of accounting for the ungrammaticality of the (a) strings in (10) and (11). These data illustrate that, like MH, Arabic does not permit the first term of a construct to take the definite determiner. This fact can be explained as follows. Suppose (i) that construct state DPs are headed by an abstract genitive case-assigner Dgen and (ii) that the definite article al {ha in MH) is a DET which does not assign genitive Case. This latter assumption is natural in view of the fact that the Case of the first term of a QCS is assigned by an external governor 2

Kaplan (1993:204) proposes that in CSs SPEC of DP is generated to the right of D' to accommodate demonstratives which follow a head. I take up a modification of this proposal in Section 3.1.

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

35

independently of its definiteness, which is transparent for government. From (i) and (ii) it follows that al and Dgen are in complementary distribution, since they are both determiners with polar Case-marking properties. Since, however, D g e n is a phonetically null head, it must be licensed by an overt category in order to assign genitive Case to the following NP. To permit this, Ritter assumes that the first term of the construct is in fact right peripheral at D-structure and moved to D0 by head-to-head movement (1991:39-40). The D-structure and Sstructure of (10a) are given as (13a) and (13b), respectively: (13) a. [DP [SPEC] [D' [D e] [NP [SPEC alkunuuz] [N' [N bayt]]]]] b. [DP [SPEC] [D' [D bayti] [NP [SPEC alkunuuz] [N' [N ti]]]]]

To extend Ritter's analysis to explain the inadmissability of strings like (11a), assume that the D-structure of the sentence is (14a), with the quantifier heading NumP, and its S-structure (14b) after head-tohead movement in (14')'. (14) a. [DP [SPEC [D' [D gen ] [NumP [SPEC taalib] [Num' [Num alkull]]]]] b. [DP [SPEC PT [Dgen alkulli] [NumP [SPEC taalib] [Num' [Num ti]]]]]

(140

At S-structure, D will dominate both al and kull, which it cannot do by virtue of the complementary distribution of Dgen (lexicalized as kull) and al. Hence, the derivation and the resulting structure are illicit. We might extend this analysis by proposing that Dgen is in

36

MARK S. LETOURNEAU

complementary distribution with definiteness morphology in general, so that the first term can take neither the definite determiner nor nunation, as (2b) requires. Exploiting the distributional parallels between nouns and quantifiers by classing them, along with the definite determiner, as tokens of D gives a unified explanation of the ungrammaticality of (10a), (11a). Such an explanation is unavailable on the lexical head analysis of 2.1, because on that analysis determiners do not belong to the same category as quantifiers and nouns, so that they cannot occur in complementary distribution with each another as members of a single category DET. The functional head analysis of quantifiers is for this reason preferable to the lexical head analysis. Moreover, it accounts plausibly for the hybrid character of quantifiers as being morphologically functional and distributionally lexical X0 categories. I shall therefore adopt the DP analysis in what follows. Before leaving this topic, we need to revise the statement of the distributional parallels between quantifiers and nouns. (11d), in which kull occurs by itself, is, one might argue, slender evidence that the quantifier may occur without a specifier, forming a minimal pair with (11b). The example palaa kullin translates like an idiom, and if so is opaque to ordinary distributional (or other) conditions. For instance, kull cannot occur as the direct object of a verb, unlike its lexical counterpart tufaah: (15) a. ?akal-at tufaah-a-n ate-3fsg apple-ACC-INDEF "She ate an apple". b. *?akal-at kull-a-n ate-3fsg all-ACC-INDEF "She ate all".

In normally requiring a specifier, as in (11b), or a complement, as in (11c), kull is analogous to a masdar such as tahtiim "destruction", which may only occur with al or in a CS: (16) a. b.

*tahtiimun altahtiimu

AGREEMENT IN QUANTIFIED CONSTRUCT STATES c.

37

tahtiim-u almadiinat-i destruction-NOM the-city-GEN "the city's destruction"

Presumably, verbal nouns and quantifiers form a natural class of DETs in CSs in that both select the second term as a complement.3 In this they differ from other CSs in which no selectional relation holds, as in kitaabu almudiiri "the director's book". In these cases, one might regard the second term as a (post)modifier or perhaps a kind of adjunct, at any rate as a nonselected complement. If this line of reasoning is correct, it would yield an explanation for the inadmissibility of kull in an S-structure like (15b): lacking its complement, the quantifier is syntactically ill-formed, either because it violates a kind of subcategorization restriction or because it is unable to discharge its Case, and probably semantically ill-formed as well in that it lacks altogether an argument to bind.

3Verbal nouns and quantifiers differ, however, in that while the former may 6-mark their complements, the latter cannot. From this distinction follows a difference in their Case-marking properties. Chomsky (1986a: 193) distinguishes inherent Case, including oblique and genitive Cases, which is assigned at D-structure, from the structural Cases nominative and objective, assigned at S-structure. However, Chomsky asserts that "inherent Case is assigned by a to NP if and only if a 0-marks NP, while structural Case is assigned independently of 0-marking". There being no reason to suppose that quantifiers assign G-roles to their NPs, it is necessary to revise Chomsky's taxonomy, thus: genitive Case as assigned by quantifiers must be structural, while genitive Case as assigned by a verbal noun must be inherent. This implies that verbal nouns must be generated in D0 at D-structure to 0-mark their complements, whereas Num0s heads move to D0 to Case-mark theirs. Al-Aboudi (1987:59-61) notes that the Arab grammarians allowed that the terms of a CS may be separated by a parenthetical element, as in (i) (his example (51); I have modified the glosses slightly to agree with those in the rest of this paper): (i)

haaSa [saahib-u [ya-?ab-a pisaam-i] zayd-i-n] this friend-NOM O-father-ACC Isam-GEN Zaid-GEN-INDEF "O father of Isam, this is one of Zaid's friends".

While the resulting sentence (along with the phenomenon itself, Al-Aboudi suspects) is slightly marginal, it can be used to diagnose the function of the second term of a CS. The distinction made in the text between second terms as complements and adjuncts predicts that parenthetical insertion between a quantifier and its complement should be degraded in comparison to insertion between a lexical noun and a nonselected second term. Joseph Aoun (p.c.) informs me that he finds a distinction in the predicted direction in his own Lebanese dialect.

38

MARK S. LETOURNEAU

2.3 Problems with Alternative Analyses A second reason for adopting the DP analysis for QCSs, in addition to its accounting for the distributional parallels in (10) and (11), is that the alternative analyses of their internal structure disallow the second term from receiving genitive Case. Suppose that we take kull not as a head but as the specifier of the term to which it assigns genitive case, following Al-Aboudi (1987:142). Since government of an XP entails government of both its specifier and its head (Chomsky 1986b:ll), this is a tenable move; the specifier could receive nominative Case from I under goverment. There are two ways in which this analysis might be executed: base generation or head-to-head movement. On the former analysis, kull would be generated in the DP SPEC of NT by (8), with the D-structure (17): (17)

[NP [DP [XP] [D' [D kull]]] [N' [al] [N mudarris-pl]]]

In (17), D heads DP, the specifier of the N head almudarris; recall that by (8) the specifier of a phrase must itself be a maximal projection. As before, D must govern the head N in order to assign it Case. However, D does not govern N because it does not even mcommand it, m-command being a necessary condition for government. The definition of m-command—that is, c-command by a maximal projection—is given in (18), from Chomsky (1986b:8): (18) A m-commands B if A does not dominate B and every maximal projection which dominates A dominates B.

In (17), A = D0 and B = N0. The maximal projection DP dominates D = kull but not N; therefore, the former does not m-command the latter and so cannot govern or Case-mark it from this position. Hence, the base generation analysis is untenable. Consider now the movement analysis: (7) would have the Dstructure in (19a), the specifier again being a maximal projection by (8), and the S-structure (19b): (19) a. [DP [XP [X' [e]]] [D' [D kull] [NP almudarris-pi]]] b. [DP [XP [X' [kulli]]] [D' [D ti] [NP almudarris-pl]]]

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

39

(19b) is ruled out on two grounds. Movement to specifier position is available only to maximal projections, under X-bar-theoretic assumptions (Chomsky 1986b:4); since (19b) instantiates X0 movement, the derivation is illicit. The movement (19) must instead satisfy the Head Movement Constraint (HMC) of Travis (1984) (from Baker 1988:53): (20)

AnX0 may only move to a Y0 which properly governs it.

Movement of kull to X 0 in (19b) also violates the HMC: the X 0 to which kull moves in (19b) does not properly govern D0 because, as noted above, it fails by (18) to even m-command it. Hence, the movement in (19b) is illicit. On neither the base generation nor the movement analysis can kull be the specifier of a quantified NP. This being the only alternative to categorizing kull as a head, I conclude that quantifiers in general are the heads of their phrases, which are thus DPs rather than NPs. 2.4 Quantifiers as NumP Heads In 2.2,1 mentioned briefly Ritter's analysis of quantifiers as heads of NumP. Let us now explore this idea in more detail. Ritter's argument for a second functional head Num0 is based on another MH construction, the free genitive, in which the first term bears the definite determiner and the second receives genitive Case from the case marker shel 'of, as in (21): (21) ha-axila ha-menumeset shel Dan et ha-uga the-eating the-polite of Dan ACC the-cake "Dan's polite eating of the cake"

Since the D0 position in (21) is filled by ha, it is not possible to move axila to that position; neither is it necessary, since Dan receives genitive Case from shel and so does not need to receive Case from a lexical category filling the Dgen position. However, since the N0 axila is generated within the complement NP, it must move via head-tohead movement to its S-structure position. Since the head D° is filled, it must move to the head of another phrase, namely, NumP. The derivation is given in (22):

40

MARK S. LETOURNEAU

(22) a. [DP [D ha] [NumP [Num e] [NP [AP ha-menumeset] [NP [DP shel Dan] [N' [N axila] [DP etha-uga]]]]]] b. [DP [D ha] [NumP [Num axilai] [NP [AP ha-menumeset] [NP [DP shel Dan] [N' [N t{] [DP et ha-uga]]]]]]

Though there is no exact analog to the free genitive in MSA, the phrase taabiS. li "belonging to", exemplified in (23) (from McCarus & Ayoub 1962:133), is a close parallel to shel: (23) maphad-u al-huquuq-i al-taabip-u institute-NOM the rights-GEN the belonging-NOM li-l-jaamip-at-i al-parabiy-at-i to-the-university-f-GEN the-Arab-f-GEN "the Law Institute of the Arab University"

Just as Dan in (22b) receives genitive Case from shel rather than from ha-axila, so in (23) aljaamipa(t) receives its genitive Case (and case) from li rather than from alhuquuqi altaabipu, which, being definite, cannot be a Case assigner due to the complementarity between D g e n and al discussed in 2.2. Assuming that the D-structure of (23) parallels (22a), we may tentatively conclude that NumP is a functional category in MSA as well as Modern Hebrew. Assuming that NumP exists in MSA, it is reasonable to follow Ritter (1991:55) in supposing that quantifiers such as kull and bapd are tokens of Num 0 which take no NP complement.4 These quantifiers differ from numbers proper in that they lack a specific cardinality, even when occurring with countable nouns. Compare the differences in meaning and grammaticality among the examples in (24): (24) a. bapd-u al-zayt-i some-NOM the-oil-GEN "some of the oil" b. bapd-u al-mudarris-iina some-NOM the teacher-mpl/GEN "some of the teachers" 4

This supposition does not contradict the suggestion above (Section 2.2) that kull selects a complement. The S-structure complement to a quantifier is its specifier at Dstructure, and it is at D-structure that Num0 takes no complement.

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

41

c. *θalaaθ-at-u al-zayt-i three-f-NOM the-oil-GEN "three of the oil" d. θalaaθ-at-u al-mudarris-iina three-f-NOM the-teacher-mpl/GEN "three of the teachers"

Although both (24a) and (24b) are partitive, only the latter can be equated with a specific cardinality, as (24 c, d) show. It seems, then, that quantifiers like bapd and kull are not inherently specified for a number feature, which they may instead inherit from the noun with which they are in construction. (We will see below, in connection with (29), that another option must also be available.) Two mechanisms license number inheritance: SPEC-head agreement and head-to-head movement. Assume the D-structure of (24b) to be (25): (25)

[DP D' [D e] [NumP [DP almudarris-pl] [Num' [Num bapd]]]]]

Under SPEC-head agreement, the specifier of NumP, almudarris-pl, is coindexed with the head Num 0 bapd so as to share its number feature with the head; bapd then moves into the null head D0 of DP, generating the S-structure (26): (26)

[DP D' [D bapdi] [NumP [DP almudarris PL] [Num' [Num ti]]]]

From the D0 position, bapd assigns genitive Case to almudarris-pl, permitting the latter to pass the Case Filter (the condition that all lexical NPs receive abstract Case) and to be realized morphologically as almudarrisiina. Assuming this analysis, we now return to the agreement data of (6), repeated here: (6) a. yahduθu bapd-u al-qalaaqil-i occur-3msg/IMPF some-NOM the-disturbances-GEN "Some disturbances occur". b. kaan-at kull-u muhaadad-at-i-n tabtad-ii?-u was-3fsg all-NOM conversation-f-GEN-INDEF begin-3fsg/IMPF "Every conversation used to begin. ..".

42

MARK S. LETOURNEAU c. kull-u-h&a. yuhaddiθu San-hu all-NOM-her speak-3msg/IMPF about-him "Everything speaks of him." d. kull-u-haa xadap-at li-1-hukm-i al-islaam-iyy-i all-NOM-her accept-3fsg to-the-rule-GEN the-Islamic-GEN "All of them accepted the Muslim rule".

Of the four examples, only (6a) exhibits deflected number agreement. This is problematic in that if the quantifier head in D° bears the feature [plural] by SPEC-head agreement, it should trigger plural number agreement on the verb, irrespective of whether the verb precedes or follows the noun. We might account for this by assuming the analysis of Mohammad (1990a), independently motivated for variable agreement in nonconstruct states, to the effect that the verb in verb-initial sentences agrees not with the following subject but with a preceding null pleonastic subject with the features third person masculine singular. On Mohammad's analysis, (6a) would have the abbreviated D-structure (27): (27) pro yahduθu bapdu alqalaaqili

More challenging is the matter of deflected gender agreement. In VS-order sentences, the verb agrees with the subject only in gender, or, if the subject is a compound, with the conjunct with which it is string adjacent (Mohammad 1990a:95-97). But string adjacency is irrelevant to gender agreement in QCSs; the verb may agree in gender with either the first or second term. Given this fact, we might suppose that SPEC-head agreement licenses not only transmission of the number feature of the SPEC of NumP but also its gender feature, as in MH (Ritter 1991:56). However, the agreement patterns in MSA are more complicated than in MH. In MH, the complement to a quantifier invariably transmits its gender to the quantifer head, to trigger both gender and number agreement on the predicate (here, an active participle): (28) a.

kol ha-yeled-im ohav-im/*ohav-ot glida all the-boy-mpl love-mpl ice cream "All the boys love ice cream".

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

43

b. kol ha-yeled-ot ohav-ot/*ohav-im glida all the-girl-fpl love-fpl icecream "All the girls love ice cream".

In MSA, by contrast, gender agreement varies between head and complement. Thus, in (6c) gender agreeement is with the quantifier, while in (6d) it is with the complement. Moreover, gender agreement varies independently of number agreement. In (6b), the feminine singular complement muhaadaθatin triggers feminine agreement on the verb kaanat, while in (6a) the nonhuman plural qalaaqil (which by itself would trigger feminine singular agreement) does not. If gender and number agreement operated in concert, as they do in the MH data, one would predict that the verbs in both (6a) and (6b) would be feminine, but they are not. A further complication is that whereas kol may be unmarked for gender, since gender agreement varies only with its complement, kull evidently may be marked as masculine, since there are cases like (6a) and (6c) in which the verb misagrees with the complement and so must agree in gender with the quantifier. While the hypothesis that quantifiers are unmarked for gender cannot be sustained in full generality, a modified form of it appears tenable. The fact that "number and gender agreement surface independently in MSA suggests the following suppositions: (i) that SPEC-head agreement is optional for both gender and number agreement and, as a corollary, (ii) that it applies for each type of agreement independently of the other. The cases in (6) can then be distinguished as follows: (29) a. b. c. d.

5

(6a): (6b): (6c): (6d):

SPEC-head agreement applies for neither number nor gender SPEC-head agreement applies for both number and gender SPEC-head agreement applies for number but not gender SPEC-head agreement applies for gender but not number 5

(6d) is an unusual and arguably unrepresentative datum. Ordinarily, the feminine suffix -haa bears a plural sense when its antecedent is a nonhuman plural, whereas in (6d) it evidently bears a human plural antecedent. A clearer example would be a variety of the standard case of deflected agreement in which a broken plural of a nonhuman feminine noun takes a feminine singular predicate. Belnap & Shabaneh (1992:250) cite an example of this kind from Taha Hussein:

44

MARK S. LETOURNEAU

The SPEC-head agreement analysis entails that, for the relevant examples in (29), internal agreement must hold at D-structure, since it is only at this level that the second term of a QCS is a specifier.6 In (6a), bapd is masculine singular (as agreement with the verb shows), while qalaaqil is feminine plural (singular qalqala(t)); hence, the specifier transmits neither its gender nor its number feature to the head. In (6b), the verb agrees with muhaadaθatin in both gender and number, indicating that it has transmitted both features to the head. In (6c), -haa is feminine singular but transmits only its number feature to hull, with which, being masculine singular, the verb agrees, as in (6a). Note that -haa does not denote plural inanimates here; if it did, it would trigger feminine singular agreement on the verb, as in (6d). That it does not implies that the head and specifier do not agree in gender. In (6d), -haa is singular in form but plural in reference and, exceptionally, denotes human referents; hence, it transmits only its gender feature to the head and, ultimately, to the verb, which receives its singular number from the head in this case. (29a) requires further comment. To say that SPEC-head agreement internal to DP is optional cannot imply that the quantifier, which normally takes its number feature from its complement, receives no number; rather, it entails that when DP-internal SPEChead agreement does not apply, the head must receive its number independently of the complement. This idea might be executed in three ways: (i) adopting Mohammad's null pleonastic analysis just in case internal SPEC-head agreement does not apply; (ii) making (i)

wa innamaa hiya jumal mutaqattip-a and however she sentence-pl choppy-fsg "However, they were short, choppy sentences".

qisaar short/pl

jumal, the broken plural of the feminine jumla(t), agrees in gender but not number with mutaqattipa. 6Should evidence be forthcoming that agreement must be between the head and its (S-structure) complement rather than its (D-structure) specifier, one might pursue a minimalist analysis (Chomsky 1993) of generalizing the relation of object agreement (Chomsky 1991) to characterize DP-internal agreement as that between a head and its complement rather than between the head and its specifier. This would permit the Dstructure specifier to be a complement at that level as well, rendering head-to-head movement from Num0 to D0—and, perhaps, both D- and S-structure as well— otiose. I will put the matter aside here.

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

45

singular number (and masculine gender) a default specification on the quantifier when it does not inherit number from its complement; (iii) making singular number a default specification on the verb, taking the subject as its agreement target (Barlow 1992:276). Appeal to Mohammad's analysis weakens the unity of the paradigm in (29) and creates difficulties for agreement with, and case-marking of, postverbal subjects on the assumptions about clause structure made in Section 3 which recommend it only as a last resort. Barlow's proposal to make singular number a default in VS sentences is redundant with respect to (29b), since the verb would be singular on two counts: a default specification and (external) SPEC-head agreement. I therefore adopt option (ii), assigning singular number and masculine gender to the quantifier just in case it does not inherit these from its complement—that is, in case (29a). Adopting (ii) renders appeal to a null pleonastic subject otiose for QCSs, avoiding the ad hoc extensions would be required to apply it to these constructions; it also explains the observed variation between MH and MSA and within MSA, in that the mandatory agreement patterns noted by Parkinson follow from (29). The cases of mandatory logical agreement with a plural definite or a feminine indefinite complement to hull would fall under (29b), where SPEChead agreement applies for both features; and the case of mandatory grammatical agreement with bapd when it has a pronoun suffix and a verb following would fall under (29a), where SPEC-head agreement does not apply and singular agreement occurs as the default. Cases of agreement variation can thus be handled by one or another of the options in (29). Independent evidence for this analysis comes from two sources, the first of which is the idiosyncratic behavior of bidp(at) "some". Ordinarily, this quantifier exhibits in MSA (though not in Classical Arabic) the kind of anti-agreement with respect to gender which is standard for numerals: the gender of the numeral and its plural complement are opposite (Wehr 1974:62). However, with broken plurals of feminine nonhuman nouns, bidp may agree in gender and number with its complement, as in these examples (from Cantarino 1975:156):

46

MARK S. LETOURNEAU (30)

a. bidp-at-u quruuš-i-n some-f-NOM piasters/f-GEN-INDEF "some piasters" b. bidp-at-u ayyaam-i-n some f-NOM days/f-GEN-INDEF "some days"

The fact that bidp exhibits both agreement and anti-agreement with respect to the gender of its complement suggests that the quantifier is unmarked for gender, just as hull and bapd are unmarked for number, receiving its gender feature in (30) via SPEC-head agreement by (29d). The second sort of evidence pertains to the interpretive possibilities of QCSs. Consider the following paradigm: (31) a. kull-u mudarris-i-n all-NOM teacher-GEN-INDEF "each teacher" b. kull-u al-mudarris-ii-na all-NOM the-teacher-mpl/GEN "all the teachers/all teachers" c. kull-u al-zayt-i all-NOM oil-GENDEF "all the oil" d. *kull-u mudarris-iina all-NOM teacher-mpl/GEN "each teachers"

The interpretations for the permitted QCSs (31a-c) are as follows: (31a) has a distributive reading, (31b) a collective or generic reading, (31c) a qualitative totality reading (Cantarino 1975:124-26). (31d) receives no interpretation. In each of the licit strings, kull agrees with its complement in number; thus, in (31a) and (31c) both quantifier and complement are singular (in construction with a count and a mass noun, respectively), while in (31b) both are plural. The head inherits its number from its complement via SPEC-head agreement in each case, presumably by (29c). (31d) is then ill-formed because the head and complement misagree in number. This would only be possible if the former were assigned its number independently of the latter under the default option (29a); but that option applies for definite plurals

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

47

like (6a), whereas (3ld) is an indefinite plural. Thus, none of (29a-d) may apply to (31d), which consequently receives no interpretation. 3. External Agreement between DP and Inflection So far, we have been concerned with SPEC-head agreement internal to DP. We now turn to external SPEC-head agreement between the DP SPEC of IP and the head I of IP, including nominative Case-marking of the subject DP. To explain the dissociation of number agreement and case-marking noted in Section 1, and to provide the theoretical apparatus needed to specify the mechanisms of subject-verb agreement in 3.2, it will be necessary to discuss Case assignment first. 3.1 Nominative Case Assignment by Inflection0 Under the Split Inflection Hypothesis (Pollock 1992), Inflection divides into Tense and Agreement, which head discrete functional projections, TP and AgrP, respectively. In Pollock's original proposal (1989), TP dominated AgrP; in Rizzi (1990:34), AgrP dominates TP. Inasmuch as it is agreement features rather than tense which are relevant to Case-marking and number agreement, this seems to be the correct move. Chomsky (1991:434-36) proposes to reconcile these views by distinguishing two AGR projections: AGR-S (= I0) for agreement between subject and verb (nominative Case-marking) and AGR-0 for agreement between object and verb (accusative Casemarking). The head AGR-S takes as its complement Finite Phrase (FP), where F = [+ finite]. Under these assumptions, (6c) will have the D-structure (32):

48

MARK S. LETOURNEAU

kulí-haa SING PRES yuhaddiθu panhu

Under standard assumptions (e.g., Safir 1987:79), nominative Case is assigned by AGR-S to the SPEC of IP (the DP subject) under government, specifically, head government. Plunkett (1993:234), citing Koopman & Sportiche (1988), which proposes that Casemarking may occur alternatively under SPEC-head agreement, tentatively concludes that in MSA nominative Case is assigned under government, the view adopted here; however, I will also argue that SPEC-head agreement and government are not complementary options but must apply together, in Arabic at least, so as to void barrierhood of the subject DP and license Case-marking. I adopt the definition of head government of Rizzi (1990:24-25), given in (33): (33) X head-governs Y iff X jointly meets these conditions: a. configurationality conditions (i) X is a head (X0). (ii) X m-commands Y. b. substantive condition: X = {[± V ±N], Agreement], T[ense]} c. locality conditions (i) No barrier [to government] intervenes. (ii) Relativized Minimality is respected.

Relativized Minimality is defined (Rizzi 1990:7) as in (34): (34) X alpha-governs Y only if there is no Z such that a. Z is a typical potential alpha-governor for Y b. Z c-commands Y and does not c-command X c. alpha-government = head government or antecedent government

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

49

Taking X = AGR-S and Y = DP, (32) satisfies the conditions on head government in (33): AGR-S is a head which m-commands DP (though it does not c-command it) (33a); it is a member of set the heads which are noninert for government (33b); and there is no intervening X0 head (Z in (34)) to act as a barrier to head government of DP by AGR S, so that Relativized Minimality is satisfied (33c). The condition (33ci) requires closer attention. Rizzi asserts that "XPs which are not directly selected by [+ V] elements are inviolable barriers for government" (Rizzi 1990:6). One such XP is the subject of a clause. While (34) implies that there is no minimality barrier in (32), satisfying (33cii), a maximal projection may also be a barrier if it is not a 0-governed complement to a verb, in which case it is said to be a blocking category (BC) with respect to government; this is the sense of "barrier" in (33ci). The relevant definitions of 9-government, L-marking, and BC are from Chomsky (1986b: 14-15): (35) A θ-governs B iff A is a zero-level category that 0-marks B, and A, B are sisters. (36) A L-marks B iff A is a lexical category that 6-governs B. (37) B is a BC for C iff B is not L-marked and B dominates C.

The restriction of L-marking to complements jointly entailed by (36), (37) implies that DP in (32) is a barrier as per (33ci). Taking A = AGR-S and B = DP, the former does not G-govern the latter by (35), since DP is not a sister to AGR-S. Since DP is not 0-governed by AGR-S, it is also not L-marked by it by (36), and since it is not Lmarked, DP is a BC by (37), hence a barrier to government and Case assignment, with respect to D 0 . The solution to this difficulty is to void the barrierhood of subjects via external SPEC-head agreement. Chomsky (1986b:24) motivates the relation of SPEC-head agreement as "a form of 'feature sharing' similar to θ-government—in fact, sharing of the features person, number, gender, Case, etc. (the '(j) features' of Chomsky 1981 and other work)." By virtue of its analogy with 0-government and Lmarking, SPEC-head agreement voids barrierhood of a subject, though only with respect to the assignment of grammatical features and not, say, with respect to extraction out of a subject island, which

50

MARK S. LETOURNEAU

is still barred by the failure of L-marking. SPEC-head agreement satisfies (33ci), rendering DP permeable for head government of its D0 head hull to receive nominative Case (and morphological case). This proposal seems more natural than an alternative suggested by the literature. One might adopt an extended definition of sisterhood in which A and B are sisters if they are dominated by the same maximal projections (Chomsky 1986b:13-14); from this it follows that AGR-S and DP will be sisters in (32). Assuming further that AGR-S is a 0marking category for its complement FP (Chomsky 1986b:20) and may even L-mark it (Lasnik & Saito 1992:106), we may take DP to be L-marked by AGR-S, voiding barrierhood and permitting nominative Case to be assigned to the head of DP. This solution is problematic in that it reduces 0-role assignment—and, a fortiori, L-marking—to, in effect, a variety of syntactic selection, vitiating the intuitive semantic import of both concepts. In any event, some qualification of Rizzi's theorem is surely required to accommodate the case-marking facts of Arabic. The inviolable barrierhood of subjects with respect to L­ marking also calls into question Koopman & Sportiche's Casemarking typology: SPEC-head agreement appears to be not an invariable alternative to head governemnt but may be a condition for it, in the case of nominative Case-assignment. The preceding discussion suggests a refinement in the X-bartheoretic notion 'head of a phrase'. If almudarris-pl in (19b) were a head, it would be governed by AGR-S and receive nominative Case; since it does not, the quantifier must be the head. But since we are assuming that DPs are multiply headed, we can no longer unambiguously identify the head of a phrase as the unique daughter of a maximal projection that is not itself a maximal projection (Di Sciullo & Williams 1987:23). Therefore, it may be necessary to restrict government by AGR-S to the "maximal" head D0 of the SPEC of IP, depending on whether NumP, and its NP specifier, if present, is a barrier or not. If NumP were a barrier, then it would be ineligible for nominative Case and the restriction would be otiose. However, since the DP SPEC of NumP receives Case from D, NumP is transparent to government and so must be excluded from government by AGR-S. The obvious move would be to appeal to Relativized Minimality to block head government of NumP by D 0 , the head intervening between AGR-S and NumP; however, D is not a member

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

51

of the set of governors in (33b). At issue is whether functional heads other than AGR and T are inert for government. Rizzi (1990:51-52) supposes that, for example, C0 is normally not a head governor; however, as is well known, the complementizer Pinna Case-marks, hence (head) governs, the lexical or pronominal subject in its clause.7 More to the purpose, case-marking by a quantifier of its complement is prima facie evidence that the former governs the latter. The D0 head thus distinguished by Relativized Minimality can be defined as in (38), with the theorem (39) following from (33) and (34), if the inventory of head governors is expanded to include D and C, and perhaps from X-bar theory as well. Note that (38) is properly neutral as to the nature of the head governor, which may be lexical (e.g., V) or functional (I or C). (38) X0 is the maximal head of XP iff there is no YP such that XP = YP, XP dominates YP, and YP dominates x 0 . (39) Only maximal heads (and specifiers) are accessible to (external) government.

So far, we have accounted for Case-marking in SV sentences. We turn now to VS structures, as in (40), a diagram of (6a): (40)

SING

PRES

yuhaddiG kull

?al

qalaaqil

Some recent work on verb-initial sentences in Arabic (Benmamoun 1993, n.l., citing Mohammad 1990b; Plunkett 1993) assumes the 7

An account of this curious fact, based on Chomsky's version of Minimality ("Rigid Minimality" in Rizzi 1990), can be found in LeTourneau (1993).

52

MARK S. LETOURNEAU

Lexical Clause Hypothesis (LCH), to the effect that subjects originate internal to some lexical projection, normally as the specifier of VP, and move to SPEC of IP to receive Case (Speas 1990:17). The LCH appears incompatible with the assumption adopted here that VP is the complement of AGR-O. In a VS sentence, a VP-internal subject DP would remain in situ while V raises to I; in this local domain it would receive accusative rather than nominative Case from AGR-0 (Chomsky 1991:437). Therefore, the question becomes how the subject in (40) is generated postverbally. A movement analysis is untenable: movement of the subject DP to the SPEC of IP position, generated to the right of V, would entail A-movement within a single clause from one SPEC of IP to another, which appears impossible within any restrictive version of X-bar theory. Such A-movement would also be from one 9-position to another, in violation of the 9Criterion, in addition, I suspect, to crossing branches, the path of the movement being ill-defined. Thus, it seems simplest to suppose that the order of categories in (8) may be parametrized to permit a specifier to precede or follow its head (extending a proposal for DPs made by Kaplan 1993:204; cf. note 2). On this assumption, the command relations between DP and I are undisturbed, permitting Case-marking under government as in SV structures. In summary, the S-structures (32) and (40) satisfy the configurational, substantive, and locality conditions (33) and (34) on head government of the DP SPEC of IP by AGR-S, the head of IP, provided that condition (33ci) is construed to allow SPEC-head agreement to void barrierhood of the subject with respect to Lmarking. Unique assignment of nominative Case to the head D0 of the subject follows from Relativized Minimality and its corollary (39), and entails the definition (38). 3.2 Number Agreement between Subject and Verb We turn next to number agreement. I assume that SPEC-head agreement between DP and AGR-S is a symmetric relation, in the sense that it permits the transmission of (|) features from DP to AGR-S and conversely. Number agreement, though symmetrically bidirectional, like Case-marking, is the mirror image of the latter: from left to right in VS sentences like (6a, b) and from right to left in SV sentences like (6c, d). This directionality thesis follows the

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

53

traditional understanding of the subject as the controller and the verb as the target (in the terminology of Barlow & Ferguson 1988) of this agreement relation. I take it as the null hypothesis that external number agreement, as a feature-sharing relationship like internal agreement within DP, is a reflex of SPEC-head agreement. However, the relevant head is not I0 but now V 0 , since the matching of features between subject and predicate surfaces morphologically on the verb. This relationship between subject and verb is crucially mediated by AGR-S, standing in an agreement relationship with DP under SPEC-head agreement. Making these assumptions, let us now consider the mechanisms of subject-verb agreement in (40). The number feature of the subject is determined by internal SPEC-head agreement, if it applies, and headto-head movement. The number feature of the head then percolates up to its maximal projection DP. By virtue of SPEC-head agreement between DP and AGR-S, the number feature of the former is transmitted to the latter. The number feature must now be transmitted to V 0 . In (40) AGR-S m-commands (and c-commands) FP and its daughters. Assume that head-to-head agreement applies between AGRS (I 0 ) and V0 to license (perhaps force) V Movement (Chomsky 1986b:75-77). By the HMC (20), the V0 yuhaddiθu raises successive cyclically, first to the head AGR-O, from which, being intransitive, it receives no features, then to F 0 , and finally to the head AGR-S of I', amalgamating with the latter two heads to receive its tense and number features (PAST and PLURAL, respectively, the latter from DP). 8 Note that in each case the trace of V0 will be properly governed, as the HMC requires, by virtue of the fact that it will be head governed as per (33), meeting not only the substantive condition on governors (Agr = AGR-S and Tense = F, respectively) but also the additional condition, which distinguishes proper head government from simple head government, that X0 head govern its governee within the same X' projection (Rizzi 1990:31): i.e., within Agr', F',

8 This head-to-head movement analysis incidentally offers a partial solution to an outstanding problem: the need for indirect 0-marking of subjects by VP. By moving to AGR-S, V m-commands the subject DP and so is in position to 6-mark the subject under (head) government, in accord with the normal pattern, if we construe the 0Criterion as a well-formedness condition at LF, as is standardly assumed.

54

MARK S. LETOURNEAU

and F, respectively, for each cycle of V Movement. Similar remarks apply to (32), the mirror image of (40). 4. Implications of the Analysis Let us now return to the questions posed in Section 1 to see how far the foregoing analysis answers them. Consider first the idiosyncratic properties of the QCS construction listed in (2). Regarding (2a), the status of Q as head of a QCS follows from axioms of X-bar theory as realized in the DP analysis proposed here. Regading (2b), the head of a QCS may not take either the definite determiner or nunation on account of the complementary distribution between them and Dgen. Regarding (2c), the possiblity of all three Case options being realized on the head, but only genitive Case on the complement, follows (i) from the structural parallels betweeen quantifiers and nouns (including verbal nouns) and (ii) from government theory, and, in particular, from Relativized Minimality, under which D0 is a barrier to head government by AGR-S of any categories dominated by D 0 . Moreover, if NumP is the complement of D0 at S-structure, as in (22), and if D is the lexicalization of an abstract genitive Case-marker, as its complementary distribution with Pal suggests, then genitive Case-marking of the SPEC of NumP follows without stipulation. The SPEC-head agreement analysis of DP-internal agreement in number, gender, and definiteness accounts for DP-external agreement as well, given the auxiliary assumption that internal SPEC-head agreement is optional. This assumption is required to explain the wider range of agreement patterns in MSA as compared to Modern Hebrew, on the one hand, and, within MSA, between quantified and nonquantified construct states on the other: plainly, nonquantified CSs exhibit fewer agreement patterns because SPEC-head agreement is unavailable in these cases. The highly specific agreement patterns noted by Parkinson seem amenable to differentiation and grouping in terms of the operation of SPEC-head agreement, at least in principle, though further investigation is certainly needed. This conclusion also seems warranted in that it explains both the variation internal to QCSs headed by bidpa in (30) and the ungrammaticality of indefinite plural complements like (31d).

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

55

Ritter's head-to-head movement analysis explains the dissociation of internal Case-marking and external number agreement in QCSs. If D0 inherits the number of its D-structure specifier via SPEC-head agreement, but not its Case, which cannot be assigned until after headto-head movement has taken place, then the dissociation of case and number noted above is explained. That is, Num 0 , the X0 category which receives nominative Case, does not head the YP which controls number agreement, since the latter, alias DP, is SPEC NumP at Dstructure and becomes the complement of D at S-structure. As a result, the head movement analysis permits us to regard number agreement between subject and verb as a head-to-head relation, mediated by AGR-S. It is unnecessary to suppose that V0 agrees directly with the (singular or plural) complement of D 0 ; we can sustain the null hypothesis that agreement is a relation between heads under the assumption that number features are transmitted to a quantifier head via SPEC-head agreement and head-to-head movement. 5. Conclusion I have argued that quantified construct states are structurally parallel to their nonquantified counterparts in that both are heads of phrases and assign genitive Case to the second term. To capture this parallelism, I took the first term of each type of CS to be the head of a functional rather than a lexical projection. I took this projection to be Determiner Phrase, on the basis of the absence of definiteness morphology on the first term; the second term of the QCS I argued to be the specifier of a second functional projection, Number Phrase. The discussion considered both internal agreement between members of the QCS and external agreement between the subject DP and Inflection (ultimately, the verb). Internal agreement between the head Num and its specifier (ultimately, between the head D and the specifier of its complement NumP) is licensed by head-to-head movement and SPEC-head agreement; head-to-head movement satisfies the Case Filter by ensuring that the lexical complement to the functional head receives (structural) genitive Case. Sharing of features of gender and definiteness is also made possible via internal SPEChead agreement. External agreement, including Case-marking, is with AGR-S, the higher of two agreement projections under V, which

56

MARK S. LETOURNEAU

assigns nominative Case to the head of a subject DP, whether initial or noninitial, under head government as licensed by SPEC-head agreement to void barrierhood of the subject, a non-L-marked XP. SPEC-head agreement also licenses external number/gender agreement: the number or gender feature transmitted to the DP from its complement is then transmitted to AGR-S, ultimately to V0 under head-to-head movement as demanded by head-to-head agreement between I0 (AGR-S) and V 0 . Like Case-marking, gender/number agreement is symmetrical, with the same processes applying in both verb-initial and DP-initial clauses.

REFERENCES Al-Aboudi, Fahmi Basil Shukri. 1987. The Syntax of 'Idaafah', Annexation in Arabic. Indiana University Ph.D. dissertation, Bloomington. Baker, Mark. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Barlow, Michael. 1992. A Situated Theory of Agreement. Outstanding Dissertations in Linguistics. New York: Garland. Language: & Charles Ferguson, eds. 1988. Agreement in Natural Approaches, theories, description. Stanford: Center for the Study of Language and Information. . 1988. "Introduction". Barlow & Ferguson 1988, 1-23. Belnap, Kirk & Osama Shabaneh. 1992. "Variable Agreement and Non-Human Plurals in Classical and Modern Standard Arabic". Perspectives on Arabic Linguistics TV ed. by Ellen Broselow, Mushira Eid, & John McCarthy, 245-62. (= Current Issues inLingustic Theory, 85.) Philadelphia: John Benjamins. Benmamoun, Elabbas. 1993. "Null Pronominals in the Context of NP and QP". Proceedings of WCCFL 11:32-43. Cantarino, Vicente. 1975. Syntax of Modern Arabic Prose, vol. II. Bloomington: Indiana University Press. Chomsky, Noam. 1986a. Knowledge of Language: Its nature, origin, and use. New York: Praeger. . 1986b. Barriers. Cambridge: MIT Press. . 1991. "Some Notes on Economy of Derivation and Representation". Principles and Parameters in Comparative Grammar ed. by Robert Freiden, 41754. ( = Current Studies in Linguistics, 20). Cambridge: MIT Press. . 1993. "A Minimalist Program for Linguistic Theory". The View from Building 20: Essays in linguistics in honor of Sylvain Bromberger ed. by Kenneth Hale & Samuel J. Keyser, 1-54. (= Current Studies in Linguistics, 24.) Cambridge: MIT Press. DiSciullo, Anna Maria & Edwin Williams. 1987. On the Definition of Word. (= Linguistic Inquiry Monographs, 14.) Cambridge: MIT Press.

AGREEMENT IN QUANTIFIED CONSTRUCT STATES

57

Eid, Mushira & Clive Holes, eds. 1993. Perspectives on Arabic Linguistics V. (= Current Issues in Linguistic Theory, 101.) Philadelphia: John Benjamins. Fassi Fehri, Abdelkader. 1988. "Agreement in Arabic, Binding, and Coherence". Barlow & Ferguson 1988, 107-58. Kaplan, Tamar. 1993. "Arabic Genitives: A problem for the binding theory". Eid & Holes 1993, 195-208. Koopman, Hilda & Dominique Sportiche. 1988. "Subjects". Ms., University of California, Los Angeles. Lasnik, Howard & Mamoru Saito. 1992. Move Alpha: Conditions on its output and application. (= Current Studies in Linguistics, 22.) Cambridge: MIT Press, Lehmann, Christina. 1988. "On the Function of Agreement". Barlow & Ferguson 1988, 55-66. LeTourneau, Mark S. 1993. "Case-Marking and Binding of Subject Clitics in Arabic Complement Clauses". Eid & Holes 1993, 261-90. McCarus, Ernest N. & Adil I. Yacoub. 1962. Contemporary Arabic Readers, I: Newspaper Arabic. Ann Arbor: University of Michigan Press. Mohammad, Mohammad A. 1990a. "The Problem of Subject-Verb Agreement in Arabic: Towards a solution". Perspectives on Arabic Linguistics I ed. by Mushira Eid, 95-125. (= Current Issues in Linguistic Theory, 63.) New York: John Benjamins. . 1990b. The Sentential Structure of Arabic. University of Southern California Ph.D. dissertation, Los Angeles. Moravcsik, Edith. 1978. "Agreement". Universals of Human Language, vol. 4: Syntax ed. by Joseph Greenberg, 331-74. Stanford: Stanford University Press. Parkinson, Dilworth. 1975. "The Agreement of bapd 'some' and kull 'all' in Modern Literary Arabic". Al-cArabiyya 8:52-68. Plunkett, Bernadette. 1993. "The Position of Subjects in Modern Standard Arabic". Eid & Holes 1993, 231-60. Pollock, Jean Yves. 1989. "Verb Movement, Universal Grammar, and the structure of IP". Linguistic Inquiry 20:365-424. . 1992. "Review of Generalized Verb Movement: Aspects of verb syntax". Language 68:836-40. Ritter, Elizabeth. 1991. "Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew". Perspectives on Phrase Structure: Heads and licensing ed. by Susan J. Rothstein, 37-62. (= Syntax and Semantics, 25.) New York: Academic Press. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge: MIT Press. Safir, Kenneth J. 1987. "What Explains the Definiteness Effect?" The Representation of (In)definiteness ed. by Eric J. Reuland & Alice G. B. ter Meulen, 71-97. (= Current Studies in Linguistics, 15.) Cambridge: MIT Press. Speas, Margaret. 1990. Phrase Structure in Natural Language. (= Studies in Natural Language and Linguistic Theory, 21.) Dordrecht: Kluwer. Stowell, Tim. 1989. "Subjects, Specifiers, and X-Bar Theory". Alternative Conceptions of Phrase Structure ed. by Mark Baltin & Anthony Kroch, 232-62. Chicago: University of Chicago Press. Travis, Lisa. 1984. Parameters and Effects on Word Order Variation. Massachusetts Institute of Technology Ph.D. dissertation, Cambridge. Wehr, Hans. 1974. A Dictionary of Modern Written Arabic ed. by A. Milton Cowan. Beirut: Librairie du Liban.

PARASITIC GAPS IN ARABIC WAFAA BATRAN WAHBA Mansoura University, Egypt

0. Introduction The data investigated in this paper are drawn from two dialects of Arabic: Standard and Jeddah Arabic. Both dialects utilize 'the gap strategy' and 'the resumptive pronoun strategy' in forming whquestions. Jeddah Arabic adopts 'the wh-in-situ strategy' as an extra device to form genuine wh-questions. All three forms of questions will be shown to satisfy the conditions that trigger parasitic gap licensing. The arguments presented in this paper are organized as follows. Section 1 shows that a parasitic gap may be licensed by an A'bound resumptive pronoun in wh-questions in both Standard and Jeddah Arabic. Section 2 examines the fact that a wh-in-situ, an unmoved wh-phrase at SS, may authorize the existence of a parasitic gap in Jeddah Arabic. This implies that the LF level is directly involved in seeking an explanation for such a behavior, since only at that level does a wh-in-situ undergo LF movement to obtain its scope as a wh-operator. Thus, the basic assumption that the S-structure whtrace is the sole licenser of parasitic gaps has to be slightly modified to make a direct reference to the linguistic level of LF where a wh-trace which results from the abstract rule of LF movement may license a parasitic gap. This, in turn, captures the generalization that the Sstructure restrictions on the distribution of parasitic gaps hold of the LF level as well in the grammar of Arabic. A parasitic gap (Chomsky 1982:40-44) is a case in which there is a wh-phrase coindexed with two empty categories at once. The first empty category is a wh-trace that results from movement and as such observes Subjacency. The second empty category, which is the parasitic gap, originates inside an island, gets licensed by the wh-trace,

60

WAFAA BATRAN WAHBA

and behaves as a variable in being c-commanded by the A'-wh-phrase. The following paradigm from English delineates the general parasitic gap diagnostics: (1) a. Which reporti did you file ei after you read pgi?1 b. *The reporti was filed ei after Bill read pgi. (2) a. Whoi did you hire ei because you said pgi would work hard? b. *Whoi did you hire ei because hei said pgi would work hard?

The contrast between (la) and (lb) signifies that a parasitic gap can be licensed by a wh-trace and not an NP-trace. The contrast between (2a) and (2b) confirms the descriptive generalization that a parasitic gap is a variable and as such is not allowed to be A-bound. Following Chomsky, parasitic gaps are assumed to be basegenerated empty pronominals. As such, they reflect the optionality of phonological content for base generated pronominals. That is, if the base-generable pronominals have a phonological content, they may violate all kinds of constraints. These lexically derived pronominals provide indirect confirmation for the existence of the LF level of representation since they do not acquire their index until they get there. If, on the other hand, the base-generable pronominals lack a phonological content, they surface as gaps at SS where they have to abide by some restrictions on their distribution such as the Binding Theory. The core of this paper argues that the distribution of parasitic gaps in Arabic provides two pieces of direct evidence for the existence of the three interdependent levels of linguistic representation: DS, SS, and LF. First, that pg's are licensed by wh-traces as well as resumptive pronouns at SS in Standard Arabic and Jeddah Arabic provides a double argument for the existence of the S-structure as an independent linguistic level of representation. Second, a parasitic gap, which is base-generated at DS, can be licensed by a wh-in-situ at SS, but may not be A'-bound until the abstract rule of wh-movement applies at LF. 1. Resumptive Pronouns as Parasitic Gap Licensers 1Inthese and subsequent examples, e = empty category and pg = parasitic gap.

PARASITIC GAPS IN ARABIC

61

1.1 Resumptive pronouns in wh-questions The data examined in this section argue that all the parasitic gap diagnostics in Standard Arabic are observed by A'-bound whresumptive pronouns, i.e., resumptive pronouns that are coindexed with a wh-phrase via Move Alpha at SS. As a consequence, the descriptive statement that parasitic gaps may only be licensed by a whtrace has to be modified to accommodate this new fact. To form wh-questions, SA employs both the gap strategy and the resumptive pronoun strategy: (3) a.

maðai paptayta ei li-1-walad-i? what gave-you to-the-boy-gen "What did you give to the boy?" b. ma-allaði paptayta-hui 1-il-walad-i? what-that gave-you-him to-the-boy-gen "What did you give to the boy?"

That the wh-trace in (3a) and the wh-resumptive pronoun in (3b) are both derived by Move Alpha can be verified by the following paradigm in which neither may originate inside a syntactic island: (4) a. qaabala omar al-rajul-a allaSi ištara al-bayta al-mujaawir met Omar the-man-acc who bought the-house next door "Omar met the man who bought the house next door." b. *maðai qaabala omar al-rajulaj allaðij ej ištaraa-hui? c. *ma-allaði qaabala omar al-rajulaj allaÖij ej ištaraa-hui? *Whati did Omar meet the man that bought ei?

The object NP inside the embedded relative clause in (4a) is not accessible to Question Formation whether the extraction site is marked by a wh-trace as in (4b) or a resumptive pronoun as in (4c), since this will result in a violation of the Complex NP Constraint. The ungrammaticality of the above sentences confirms the claim made at the outset of this section—that the A'-bound wh-resumptive pronoun observes Subjacency and, as such, cannot be base-generated. To account for its behavior, I will make the assumption that a whresumptive pronoun is originally a wh-trace that is spelled out morphologically at S-structure. To further validate this assumption,

62

WAFAA BATRAN WAHBA

both the wh-trace and the wh-resumptive pronoun may equally license a parasitic gap as elucidated by the following paradigm: (5) a. pomarj faqada al-kitaab-ai qabla arm PROj yaqra?a-hui. Omar lost the-book-acc before that to-read-it "Omar lost the booki before he read iti." b. *pomar faqada al-kitaabai qablaannPRO yaqraPa ei. c. maaðai faqada ei omar qabal ann PRO yaqraPa pgi? d. ma-allaðii faqada-hui omar qabalaannPRO yaqraPa pgi? "Whati did Omar lose ei before he read pgi?"

The empty category in (5b) is ruled out on functional grounds since it is never A'-bound (i.e., a variable) nor A-bound (an NP-trace). It is also not a PRO (since it is governed). Finally, it cannot be a pro because it is not a subject. The parasitic gap in (5d), like the one in (5c), satisfies all the parasitic gap licensing conditions since it is not ccommanded by the licensing wh-resumptive pronoun and it is A'bound by the wh-phrase in the matrix Comp. In sum, a wh-resumptive pronoun syntactically behaves as a whtrace in observing Subjacency and as such may license a parasitic gap. The fact that a resumptive pronoun licenses a parasitic gap in SA constitutes a double argument that the S-structure level is independently needed to account for the behavior of parasitic gaps in the grammar of Standard Arabic. 1.2 Resumptive pronouns in wh-relatives In direct contrast to the behavior of wh-resumptive pronouns with respect to licensing pg's, a relative clause resumptive pronoun freely violates Subjacency, and as such must be assumed to be basegenerable: (6) qara?-tu al-kitaab-ai allaðii qaabalat monaNP [al-kaatib-aj Mona the-writer-ace read-I the-book-acc that met S'tallaSi s[ej ?allafa-hui]]]. who wrote-it. *"I read the book which Mona met the author who wrote."

The resumptive pronoun that occupies the relativized site inside the complex NP island has to be assumed to be a base-generable

PARASITIC GAPS IN ARABIC

63

pronominal due to its free violation of Subjacency. This, in turn, predicts that a relative clause resumptive pronoun cannot license a parasitic gap, whereas a wh-resumptive pronoun can, since the former is generated in the base and the latter is syntactically derived at Sstructure via wh-movement. Such a prediction is borne out as illustrated by the contrasts below: (7) a. *al-mu?allifi allaSii ihtaqar-ta-hui bapda ann qaabalt pgi. the author whom despised-I-him after that met-I "The author whomi I despised after I met pgi." b. mani allaði ihtaqart-a-hui bapda ann qaabal-ta pgi? who that despised-you-him after that met-you "Who did you despise after you met?" (8) a. *al-kitaabi allaöii faqat-ta-hui qabala ann t-aqra?a pgi the-book which lost-you-it before that you-read "The book whichi you lost ei before you read pgi." b. ma allaöii faqad-ta-hui qabla ann t-aqra?a pgi? what lost-you-it before that you-read "What did you lose before you read?"

That the wh-resumptive pronoun may license a parasitic gap (7b, 8b) whereas the relative clause resumptive pronoun cannot (7a, 8a) maintains the crucial property of parasitic gap licensing—namely, that a potential licenser should be a variable which observes Subjacency. This is exactly what we have in (7b) and (8b) where the parasitic gap is legitimately licensed by the wh-resumptive pronoun which is A'bound at SS by observing Subjacency (cf. 4b, 4c). It also explains why the base-generable relative clause resumptive pronoun in (7a) and (8a) is incapable of authorizing a parasitic gap (cf. 6). To summarize this section, a wh-resumptive pronoun may license a parasitic gap, whereas a relative clause resumptive pronoun cannot. This is justifiable since the former is a morphologically spelled-out wh-trace derived at S-structure, whereas the latter is a base-generable pronominal derived at DS. The schemata in (9) below shows that a parasitic gap in Standard Arabic has two potential licensers: a whtrace, and a wh-resumptive pronoun:

(9) wh-phrasei

{

resumptive - pronoun < > wh - trace

pgi

64

WAFAA BATRAN WAHBA

2. A Wh-in-situ as a Parasitic gap Licenser A quite interesting fact about the grammar of Jeddah Arabic is its interchangeable utilization of three wh-question formation strategies which can be represented by the following paradigm: (10) a. miini who b. miini who c. [+wh

gabal-ta ei al-yoom? met-you today illi gabal-t-a-hi il-yoom? that met-you-him today? [gabal-t miin il-yoom]]? met-you whom today "Whom did you meet today?"

The wh-trace in (10a) is coindexed via SS movement with the A'-whphrase. The A'-wh-operator in (10b) is coindexed with a resumptive pronoun that occupies the questioned site via SS movement. Finally, the wh-operator in (10c) occurs in-situ, i.e., occupies its base position at S-structure. In the following paradigm, a parasitic gap is licensed by a wh-trace (11a), a resumptive pronoun (11b), and a wh-in-situ (11c): (11) a. min miini γaarat ei mona [pašaan [pomar yebγa [PRO ye of whom was-jealous Mona because Omar wants to marry tjawwaz pgj]? "Whoj was Mona jealous of ei because Omar wanted to marry pgi." b. miini illi mona γaarat min-hai pašaan pomar yebya who that Mona was-jealous of-her because Omar wants yetjawwaz pgi? to-marry? c. mona γaarat min miini pašaan [pomarj yebya [PROj Mona was-jealous of whom because Omar wants to yetjawwaz pgi?]] marry "Of whomi was Mona jealous ei because Omar wanted to marry pgi?"

The parasitic gap in (11a) is a variable bound by the A'-operator min miin and authorized by the non-c-commanding wh-trace. The parasitic gap licenser is a wh-resumptive pronoun in (11b) and a wh-in-situ in

PARASITIC GAPS IN ARABIC

65

(11c). The wh-operator min miin in (11c) occurs in-situ, i.e., it cannot behave as an operator that might bind a variable since it is not located in Comp. Here, it functions as a mere parasitic gap licenser because it does not c-command the parasitic gap it authorizes. In what follows we discuss the nature of the wh-in-situ as a potential licenser and the implications this may have for the GB framework. There are three main diagnostics that characterize a potential parasitic gap licenser. First, it has to be a wh-trace, i.e., it is derived by Move Alpha at SS. Second, it should not c-command the parasitic gap it licenses. Third, it has to be A'-bound by a wh-operator that does c-command the parasitic gap. Only the second diagnostic is observed by a wh-in-situ as a parasitic gap licenser, whereas the other two (the first and the third) are satisfied at LF. The second diagnostic, which states that a parasitic gap should not be c-commanded by its licenser, i.e., a parasitic gap cannot be Abound, can be illustrated by the following contrasts: (12) a. *pali xabbar miini inn-u biyekrah pgi Ali told whom that-he hates *"Whoi did Ali tell ei that he hated pgi?" b. Sali darab miini pašaan biyekra pgi? Ali hit whom because he-hates "Whoi did Ali hit ei because he hated pgi?"

The verb xabbar "tell" in (12a) subcategorizes for a clausal complement, hence, the parasitic gap inside that clause will be ccommanded by the wh-in-situ, which rules the sentence as ill-formed. The tree in (13) helps to clarify this point:

66

WAFAA BATRAN WAHBA

The verb darab does not subcategorize for a clausal complement. As a result, the parasitic gap in (12b) is not c-commanded by the wh-in-situ in the matrix clause since the parasitic gap is outside the scope of the wh-in-situ as shown by the tree below:

A definition of binding and variable that incorporates the above insight may be formulated as follows: a. Binding: X binds Y if X and Y are coindexed and X c-commands Y. b. Variable: A variable is an empty category whose closest binder is in an A' position (i.e., a variable is locally A'-bound)

Such a definition is connected to a definition of Logical scope—the scope of an operator is the subtree that it c-commands. In light of these definitions, the differences between (12a) and (12b) can be stated as follows: In (12a) the parasitic gap is bound by its licensing wh-insitu, hence it is ruled out as ill-formed. (12b) is grammatical because the parasitic gap is outside the scope of the licensing wh-in-situ.

PARASITIC GAPS IN ARABIC

67

The first diagnostic—that the parasitic gap licenser has to be a trace—can be maintained at LF if we assume with May (1985) and Huang (1982) that an abstract rule of wh-movement applies at that level. The existence of such a rule satisfies the first and third diagnostics mentioned at the beginning of this section. Thus, the LF representations of (12a) and (12b) will be as in (15) and (16), respectively: (15) *miini pali xabbar ei inn-u biyekrah pgi Ali told that-he hates who *"Whoi did Ali ei that he hates pgi?" (16) miini Sali darab ei pašaan biyekrah pgi Who Ali hit because he-hates "Whoi did AU hit ei because he hated pgi?"

The well-formed abstract representation in (16) looks approximately the same as well-formed S-structure questions (cf. 5c). The ill-formed LF representation in (15) looks more or less as an ill-formed Sstructure in Standard Arabic where the parasitic gap is illicitly ccommanded by its licensing trace. In Jeddah Arabic, then, a wh-in-situ functions as a valid parasitic gap licenser whose behavior can be adequately accounted for if we make a direct reference to the LF level. As a justifiable parasitic gap licenser, the wh-in-situ does satisfy all the parasitic gap licensing diagnostics at LF where the parasitic gap is legitimately authorized by an abstract wh-trace and c-commanded by an abstract wh-operator. 3. Conclusion The data examined in this brief survey of the behavior of parasitic gaps in Arabic have significant implications for the GB framework. First, they provide compelling evidence for the existence of the three interdependent levels of linguistic representation: D-structure, the Sstructure, and the LF. First, the parasitic gap is base-generated at DS as an empty category that may optionally have a phonological content. Second, the fact that a parasitic gap may be licensed by a wh-trace or a resumptive pronoun provides a double argument in favor of the existence of the S-structure. Third, the LF level is needed to account for the fact that a parasitic gap licensed by a wh-in-situ at SS may only

68

WAFAA BATRAN WAHBA

be A'-bound at LF since only at that level does the wh-in-situ undergo wh-movement. This strongly supports the hypothesis made in May (1986) and Huang (1982) that there exists an abstract rule of whmovement that emulates that of S-structure wh-movement. Thus, the behavior of parasitic gaps provides insight into the GB theory, especially in the area where the modules interrelate. It is especially revealing, where the movement module is concerned, that the Jeddah Arabic data provide direct evidence that Move Alpha accounts for the behavior of parasitic gap at LF as it does at SS. To maintain the generalization that a parasitic gap may only be licensed by a wh-trace, we have to assume that the linguistic level of LF does exist to capture the requirement that a parasitic gap must be A'-bound, which in the grammar of Arabic is only realized at LF.

REFERENCES Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. . 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. Engdahl, Elizabeth. 1985. "Parasitic Gaps". Linguistics 23.3-44. Huang, James. 1982. "Logical Relations in Chinese and the Theory of Grammar". Ph.D. dissertation, MIT, Cambridge, Mass. Lasnik, Howard & J. Uriagereka. 1988. A Course in GB Syntax. Cambridge, Mass.: MIT Press. May, R. 1985. Logical Form: Its structure and derivation. Cambridge, Mass.: MIT Press. Riemsdijk, H. C. & Edwin Williams. 1986. Introduction to the Theory of Grammar. Cambridge, Mass.: MIT Press. Ross, John R. 1967. "Constraints on Variables in Syntax". Ph.D. dissertation, MIT, Cambridge, Mass.

NEGATION AND MODALITY IN EARLY CHILD ARABIC

IBRAHIM MOHAMED & JAMAL OUHALLA Queen Mary and Westfield College London University

1. Introduction Among the major issues debated in recent literature on language acquisition is one which concerns the status of functional categories in the early stages of acquisition. One view, often referred to as the Continuity Hypothesis, holds that the early stages of development instantiate a full structure of the sentence, identical in its major geometrical properties to the one instantiated by adult languages (e.g., Pierce 1989; Weissenborn 1990). Some of the distinctive properties of child language are attributed to a difference in the feature-makeup of functional categories. Another major view, which came to be known as the Maturation Hypothesis, argues that the early stages of acquisition instantiate a reduced structure of the sentence, more precisely a structure which lacks functional categories altogether (e.g., Radford 1990; Tsimpli 1992). Proponents of both hypotheses have relied on the properties of negative sentences in various languages for evidence. For example, Pierce (1989) argues that the properties of negative sentences in early child French indicate the presence of verb-movement processes, and therefore of functional categories which serve as landing sites for them. On the other hand, Tsimpli (1992) argues that the properties of negative sentences in early Modern Greek and other languages, including early child French, indicate the absence of functional categories. Negation elements are argued to have the status of adjuncts

70

IBRAHIM MOHAMED & JAMAL OUHALLA

of the predicate phrase, an analysis which makes it possible to account for their interaction with modality. In this paper we will examine negative sentences in early child Arabic, in particular the Palestinian dialect spoken in Nablus and surrounding areas. We will argue that the properties of negative sentences in early child Palestinian require an analysis which combines elements from both the Continuity Hypothesis and the Maturation Hypothesis. While there are indications that at least some functional categories are absent at the early stages of development, Neg is not likely to be one of them. The hypothesis we are led to adopt is one which makes a distinction between functional categories which contribute to meaning, e.g., Neg, and those which do not, e.g., AgrS. The former are present, and, moreover, project X-bar structures of their own, whereas the latter are absent. Despite the semantic overtones of the hypothesis, the presence of Neg and its X-bar projections is argued for mainly on syntactic grounds, having to do with the derivation of negative sentences with postverbal negation. Moreover, following conclusions reached in Ouhalla (1992, 1993), we make a distinction between negative sentences which have a neutral reading, and the ones which have a (negative) modality reading, conveyed by the negation elements. In the former, the negation elements project an X-bar structure of their own, whereas in the latter the negation elements occupy a different functional position which we identify as Mod(ality). In other words, in neutral negative sentences the negation elements function as pure sentential negators, whereas in negative sentences with a modality reading they function as markers of negative modality. The discussion of data from early child Palestinian (sections 3 and 4) is preceded by a discussion of the properties of negative sentences and their derivation in the target (adult) language (section 2). The intention is to provide the necessary empirical and theoretical background for a proper appreciation of the properties of negative sentences in the early stages of development. 2. Negation in Palestinian Arabic In most spoken dialects of Arabic, sentential negation is expressed in terms of the particles m- and -sh accompanied by certain phonologically determined vowels. In some contexts the two particles

NEGATION AND MODALITY IN EARLY CHILD ARABIC

71

appear attached to each other in the form of the complex m-š followed by a predicate phrase XP, and in others they appear on opposite sides of a predicative head category, e.g., the verb, forming with it a complex of the form m-V-š (XP). The examples discussed in this paper are all from the Palestinian dialect of Arabic spoken in Nablus and the surrounding areas, although in major respects the facts are not fundamentally different in most other spoken dialects of Arabic. The following examples illustrate the contexts where the pattern m-sXP is found: (1) a. hiyya mis mudarrisa/mabsuuta/raayha/ fil-biit she NEG teacher happy going in the-house "She is not a teacher/happy/going/in the house." b. *hiyya ma mudarrisa-š/ mabsuuta-š/ raayha -š she NEG teacher-NEG happy-NEG going-NEG fi 1-biit-s in the-house-NEG (2) a. (hu) miš rah yštayil lyuum he NEG will (3MS) work today "He will not work today." b. *(hu) ma rah yštayil -š lyuum NEG will (3MS) work -NEG today he c. (hu) ma rahi (-š) yštayil lyuum NEG will -NEG (3MS) work today he

The examples in (la,b) are instances of the so-called nominal sentences, where the head of the predicate is a noun, an adjective, a participle, or a preposition, i.e., a nonverbal category. The examples in (lb) show that the two negation particles cannot attach to a nonverbal head of a predicate phrase in the Nablus dialect, contrary to what is found in some other dialects (e.g., Egyptian Arabic and some varieties of Moroccan Arabic) where this is possible at least with some nonverbal predicative categories (Benmamoun 1992). Examples (2a,b) are verbal sentences which consist of a main verb and a (near) future Modal verb which inflects optionally for subject agreement. Example (2b) shows that in the Nablus dialect the negation particle -5 cannot attach to the main verb even when the future Modal is not inflected for subject agreement, again contrary to what is found in some varieties of Moroccan Arabic where this is possible when the future Modal verb is not inflected for subject agreement (Benmamoun

72

IBRAHIM MOHAMED & JAMAL OUHALLA

1992). Example (2c) illustrates the fact that in this type of sentence the particle -š can attach to the (near) future Modal. In this example, and others below, -š is included between brackets to indicate that its occurrence is optional. This means that the pattern V-m is a variant of the pattern m-V-š. The contexts where the pattern m-V (XP) is found are illustrated in examples (3a-c), (4a-b), and (5a-c): (3) a.

b. c. (4) a.

b. (5) a.

b. c. (6) a.

b.

(hu) ma b- yištiyil (-Š) hassa he NEG PRT-(3MS)work (-NEG) now "He is not working now." *hu miš b-yištiyil hassa he NEG PRT-(3MS) work now (hu) ma rahi (-š) yštayil lyuum he NEG will-NEG (3MS) work today (hu) ma šafi (-Š) Ahmad he NEG see NEG Ahmad "He has not seen Ahmad." *hu miš šaf Ahmad he NEG see Ahmad (hu) ma kaan (-ši) byištayil he NEG be NEG PRT-(3MS) work "He was not working." *hu mis kaan b- yištayil he NEG be PRT-(3MS)work *hu ma kaan b- yištayil -ši he NEG be PRT-(3MS) work-NEG (ma) tudrubi -š paxuuk! NEG (2MS)- hit -NEG your brother "Don't hit your brother." *mis tudrub paxuuk! NEG (2MS)hit your brother

Examples (3 a) and (4a) are simple verbal sentences which include a main verb but not an auxiliary. In this type of sentence the negation elements attach to the verb obligatorily, so that the pattern m-š XP is excluded (see 3b,4b). Example (5a), on the other hand, includes the auxiliary kaan "be", which marks the tense frame of the sentence, in addition to the main verb which is unmarked for tense. In this type of sentence both negation elements attach to the auxiliary verb obligatorily. (5b) shows that the pattern m-s XP, where X is the auxiliary verb, is excluded in

NEGATION AND MODALITY IN EARLY CHILD ARABIC

73

this context, indicating that the auxiliary kaan "be" behaves differently from the future tense Modal rah discussed above. (5c) shows that the particle -š cannot attach to the main verb in this type of sentence. Finally, (6a) is an imperative sentence, where the negation particles attach to the verb, and the pattern m-š XP is excluded (6b). Note that with imperatives it is the preverbal particle m- which tends to be dropped, rather than the postverbal one. Although this distinctive property of imperatives may have some significance, we will assume here that the pattern V-š, like the pattern m-V, is also a variant of the more fundamental pattern m-V-š. Examples (la) and (lb) seem to suggest that the distribution of the negative particles may have to do with their morphological selectional (m-selectional) properties. In the Nablus dialect, they apparently can only attach to a verbal category or to each other. On the other hand, examples (2a-c) and (5a-c), which include a modal/auxiliary in addition to the main verb, indicate that the distribution of the negation particles cannot be exclusively reduced to their idiosyncratic mselectional properties. These examples suggest that distance also plays a role in the distribution of at least -5. Assuming that the underlying position of -š is closely related to that of its correlate m-, the main verb is too far away to serve as a host for -š, presumably because it is separated from the negation elements by another verbal category. In view of these observations, an adequate account of the distribution of the negation particles in the Nablus dialect will have to rely on their peculiar m-selectional properties, as well as on relevant syntactic principles of locality. The former is necessary to explain the fact that the negation particles do not attach to nonverbal predicates even when they are adjacent to them, and the second to explain the fact that -š cannot attach to a verb when the verb is not sufficiently close to it. One such analysis is outlined in the next few paragraphs, and draws on ideas articulated in Benmamoun (1992) in the context of the so-called Split Infl Hypothesis. We will assume that m- is the head of NegP, and -š its specifier. Moreover, -š can cliticize (from SpecNegP) to the head position of NegP or to the head position immediately dominating NegP, but not to a position lower than Neg (i.e., a position which does not c-command SpecNegP). Presumably, the latter restriction follows from some version of the ECP, or a generalized condition on the binding of

74

IBRAHIM MOHAMED & JAMAL OUHALLA

traces (Aoun 1985). Finally, we will assume, for reasons to be spelled out, that NegP is ordered lower than I(nfl)/T(ense), implying that the Nablus dialect has IP-internal negation on a par with Standard Arabic (Benmamoun 1990, 1992). Accordingly, verbal sentences such as (3a), (4a), (5a), and (6a) have the structure and derivation shown in (7) (for arguments that AgrS in Arabic is lower than T, see Benmamoun 1990, 1992): (7)

The verb moves in a stepwise fashion to I/T, and possibly to a higher position (see below). The particle -š then cliticizes (from SpecNegP) onto the verbal complex under I/T (or a higher position), deriving the complex m-V-š, where V is either a main verb, the auxiliary kaan "be", or the future Modal rah. Movement of the main verb to I/T, across the auxiliary, is excluded for reasons having to do with the HMC/ECP (Chomsky 1986). Likewise, movement of the particle -5 down to the main verb is also excluded for reasons having to do with the ECP, and more generally the ban against lowering movements. Note that since it is generally possible for the two negation particles to satisfy their m-selectional properties simply by attaching to each other, the derivation of the complex m-V-š and the exclusion of the illicit pattern *m-š V in verbal sentences cannot be attributed to the need to satisfy the m-selectional properties of the negation particles. It is necessary to assume obligatory movement of the verb to

NEGATION AND MODALITY IN EARLY CHILD ARABIC

75

a position higher than Neg for the desired result to be achieved. This partly accounts for our decision to order Neg lower than I/T. For the purposes of this paper we will adopt the view that movement of the verb to I/T is necessary for the verb to acquire tense properties. The optional movement of the future Modal rah to I/T, indicated by its compatibility with the pattern m-š rah (2a), can be accommodated under this view by assuming that rah is inherently marked for tense, and therefore does not have to move to I/T. When rah fails to move to I/T the negation particle -š cliticizes onto its correlate under Neg, deriving the pattern m-š rah. Nominal sentences of the type illustrated in (la) have the structure and derivation shown in (8), where X is a noun, an adjective, a participle, or a preposition (8)

IP Spec

I' I/T

NegP

Assuming, as we did earlier, that the negation particles in the Nablus dialect m-select verbal categories only, movement of the head of the predicate to Neg (on its way to I/T) is excluded. Instead, the particle -s cliticizes onto m- under Neg, thereby deriving the pattern m-š XP characteristic of negative nominal sentences. Note that it is possible to exclude movement of X in (8) to I/T (via Neg) on the grounds that nominal categories are incompatible with tense. On this view, it is not necessary to appeal to the m-selectional properties of the negation particles, at least as far as the Nablus dialect is concerned. Comparing the derivations outlined in (7) and (8), the crucial generalization seems to be that the pattern m-V-š is derived when the verb moves to a position higher than Neg, whereas the pattern m-š

76

IBRAHIM MOHAMED & JAMAL OUHALLA

(rah) XP is derived when a head category fails to move to a position higher than Neg. This is an important conclusion, which will play a crucial role in the account of negation in early child Palestinian outlined below. To complete the presentation of the paradigm in the target dialect of Nablus, it is important to point out that the pattern m-X-š is also found in apparently nominal sentences which include the existential element/// "there is", the boulemaic modality element bidd+CL(itic) "want", or the possession particle pind +CL(itic) "have/possess": (9) a.

(ma) bidd-ii -š masaaii NEG want -me -NEG money "I don't want (any) money." b. *mis bidd -ii masaari NEG want-me money (10) a. (ma) fii -š šuyl huuna NEG in -NEG work here "There are no jobs here." b. *miš fii šuyl huuna NEG in work here (11) a. ma pind-ii -š masaari NEG have -me -NEG money "I have no money/I don't have any money." b. *miš Sind -ii masaari NEG have-me money

The curious property of these predicates is that they pattern with verbs with respect to their position in relation to the negation particles even though they are apparently nonverbal categories. We saw above that the Nablus dialect generally does not allow nonverbal categories to be flanked by the two negation particles, so that the grammaticality patterns in (9a-b), (10a-b) and (11a-b) are the reverse of what is usually found with nominal sentences. The contrast between these specific predicates and nominal predicates in general is clearly shown by comparing the possession element in (11) to a homophonous element in the language with a partly existential and partly possessive meaning roughly equivalent to that of the French preposition chez "at/in X's place":

NEGATION AND MODALITY IN EARLY CHILD ARABIC

77

(12) a. mis Sind -naa ahmad NEG chez -us Ahmad "Ahmad is not at our place/in our house." b. *ma pind-naa -š ahmad NEG chez -us -NEG Ahmad

As shown by these examples, the element in question behaves more like nominal predicates in general with respect to the negation particles, in contrast to its purely possessive counterpart illustrated in (11a-b) which behaves more like verbal categories with respect to the negation particles. Note that (12b) is grammatical, with a purely possessive, but somewhat incongruous, meaning. The status of the predicates in (9), (10), and (11) is reminiscent of that of the apparently pronominal form which appears in copular sentences in Semitic languages. Like the predicates in (9), (10), and (11), the pronominal form also behaves like a verbal category with respect to the negation particles in the Nablus dialect. In the following examples, the predicate jaay "coming" is a participle: (13) a. (ahmad) ma hu-š jaay "Ahmad is not coming." b. (ahmad) *mis hu jaay Ahmad NEG be coming c. ahmad/hu miš jaay Ahmad/he NEG coming

In (13a-b) the element hu, which is homophonous with the third person masculine singular pronoun illustrated in (13c), behaves like a copula with respect to the negation particles as well as other considerations which we cannot discuss here for lack of space. In view of the parallelism between hu in (13a-b) and the predicates in (9), (10), and (11) it is plausible to conclude here that the latter are also verbal categories. Unlike rah, these elements are not inherently specified for tense, and therefore move obligatorily to I/T in a structure along the lines of (7) above. This scenario will account for their incompatibility with the pattern m-š XP.

78

IBRAHIM MOHAMED & JAMAL OUHALLA

3. The Negation Paradigm in Early Child Palestinian The early stages of the acquisition of Palestinian Arabic are characterized by the appearance of the two major patterns of sentential negation discussed in the previous section. However, the contexts in which the two negation patterns are used do not necessarily correspond to the ones in the target language, so that a given pattern is used ('incorrectly') in a context which requires a different pattern in the target language. We will argue that this apparently random use of the negation patterns reveals significant properties of the structure of the sentence at the early stages of development. We will start by discussing the use of the anaphoric negator la? "no". As shown in the following examples, this element is used both anaphorically, to negate a previous utterance (14a-f) in a given discourse, as well as non-anaphorically, to negative the sentence immediately following it (15a-f) (note that, as a general rule, we provide the full forms of morphemes/words, ignoring the peculiar properties of child pronunciation whenever these do not bear crucially on the issues discussed): (14) a. la? bidd-ii -š NEG want-me -NEG "No, I don't want [it]." b. la? mis habiib-ak ?anas NEG NEG beloved-your ?anas "No, Anas is not your beloved." c. la? fiha-s telfizyuun NEG in -NEG television "No, there is no television." d. la? kola kaman NEG coke again "No, I want more coke." e. la? hayyo fiil NEG this elephant "No, this is an elephant." f. la? kaman minsahwil NEG also slide "No, we also played in the slides." (15) a. la? raah baaba NEG leave Daddy "Daddy has not left."

2.3

2.3

2.3

2.3

1.11

2.3

2.3

NEGATION AND MODALITY IN EARLY CHILD ARABIC b. la? kola hayy NEG coke this "I don't want this coke." c. la? ?aja pind-na NEG come to -us "He has not come to visit us." d. la? dis NEG this "I don't want this." e. la? paarfe NEG knowing 'T don't know." f. la? tayybe NEG nice "It is not nice."

79

2.3

2.3

1.11

1.11

1.10

In (14a-f) la? negates a previous utterance, rather than the sentence following it. This is clearly shown in example (14a-c), where the sentence following la? is independently negated by the negation particle-}šand the complex m-š, respectively. In (15a-f), however, la? negates the sentence following it, and therefore functions as an instance of sentential negation. When la? functions as a sentential negator it can also express boulemaic modality. Thus, (15b,d) have a boulemaic modality reading even though the predicate usually used to express it in the target language (bidd+CL) is absent. Note that (14a) also has a similar negative modality reading, but this time it is due to negation particle -š rather than to the anaphoric negator. The use of negation elements to convey (boulemaic) modality in the early stages of development has been reported in relation to a number of languages (cf. Bloom 1991; Hoekstra & Jordens 1991; Tsimpli 1992, among others). This fact has been argued by Hoekstra & Jordens (1991) and Tsimpli (1992) to indicate a close connection between negation and modality not only at the conceptual level but also at the structural level. We will come back to this point in detail in the next section. Turning now to the purely sentential negators, the following examples illustrate the contexts in which the pattern m-iš XP is used:

IBRAHIM MOHAMED & JAMAL OUHALLA

80 (16) a.

b.

c.

d.

e.

f.

(17) a.

b.

c.

d.

e.

f.

miš paarfe NEG know(participle) "I don't know." miš šaatre NEG clever "I am not clever." mis pilak NEG yours "It is not for you." mis xadra NEG green "It is not green." mis habiib-ak ?ana NEG beloved-your I "I am not your beloved." mis piib NEG shameful "It is not shameful." mis pašuf NEG see "I don't want to see." mis haaki ?ana NEG talking I "I don't want to talk." mis pakalit NEG eat "I haven't eaten." mis rah NEG leave "He hasn't left." miš šalhiin-i NEG undress -me "Don't undress me!" mis sakkr-u NEG shut -it "Don't shut it!'

2.3

2.0

1.11

1.10

1.11

2.0

2.3

2.3

2.3

2.3

2.5 (Abdo & Abdo 1986)

2.7

Examples (16a-f) are nominal sentences, whereas (17a-f) are verbal sentences. While the nominal sentences are 'correctly' negated with the pattern m-š XP, the verbal sentences are not. The pattern m-š XP is used to negate both types of sentence, ignoring the distinction made between them in the target language.

NEGATION AND MODALITY IN EARLY CHILD ARABIC

81

As with the anaphoric negator, the complex m-š is also used to express modality, in addition to negation. Examples (17a, b) have a boulemaic modality reading even though they do not instantiate the predicate bidd+CL "want". Although verbal sentences of the type illustrated in (17a-f) are frequently 'incorrectly' negated with the complex m-s, they are equally frequently 'correctly' negated with the pattern m-V-s: (18) a. ?akalti-š eat -NEG "I haven't eaten." b. ma qarast-u -š NEG pinch -him -NEG "I haven't pinched him." c. ma širbat NEG drink "She has not drunk." d. tudrub-s -NEG hit "Don't hit [him]!" e. ma tukil-s NEG eat -NEG "Don't eat [it]!" f. ma tuxid lktab NEG take the book "Don't take the book."

2.3

2.3

1.11

2.3

2.3

1.11

Examples (18a-c) are indicative sentences, and (18d-f) are imperatives. Both types of sentence can be negated with the pattern mV-š or one of its variants V-s and m-V. As expected, this pattern is also used with sentences which include the boulemaic, existential, or possession predicate, which we have classified as instances of verbal sentences: (19) a. bidd-ii-s samak want-me-NEG fish "I don't want fish." b. ma bidd-ii -š NEG want-me-NEG "I don't want to."

2.3

2.3

82

IBRAHIM MOHAMED & JAMAL OUHALLA c. fi-š tuffaha in-NEG apple "There is no apple (left)." d. ma fi-ha-s NEG in-it-NEG "There isn't any." e. ?inn -aa -š battiix possess-us -NEG melon "We don't have melon." f. ma ?inn-aa-š NEG possess-us -NEG "We don't have any."

2.3

2.3

1.11

1.11

Perhaps significant is the fact that in our corpus these sentences do not appear with the variant m-V of the pattern m-V-š, the most favored variant being V-š. Note, however, that the latter also happens to be the form favored in the target language. 4. The Structure and Derivation of Negative Sentences in Early Child Palestinian Having reviewed the paradigm of negation in early child Palestinian, its most notable property is the (partial) absence of the strict correspondence between the two patterns of negation and specific syntactic contexts which characterizes the target language. While nominal sentences are consistently 'correctly' negated with the pattern m-š XP, verbal sentences are freely negated either with the 'incorrect' pattern m-š XP or the 'correct' pattern m-V-s. The only possible exception to this general picture are verbal sentences which include the boulemaic, existential, or possession elements, which in our corpus do not seem to occur with the pattern m-sXP. In the discussion of the derivation of the two major patterns of negation in the target language, it was concluded that the pattern m-Vš (or any of its two variants) is derived when a verbal head category moves obligatorily to a position higher than Neg, namely I/T. On the other hand, the pattern m-š XP is derived when the head of the predicate fails to move to this position. Viewed in this context, the facts of early child Palestinian suggest the absence of obligatory movement of the verb to a position higher than the position of the negation element. It seems that while this movement is possible, as

NEGATION AND MODALITY IN EARLY CHILD ARABIC

83

shown by the availability of (verbal) sentences with the pattern m-V-š, it is not obligatory. A possible way of explaining this situation is to assume, in the context of the Continuity Hypothesis, that I/T, though present, is underspecified for the tense features, and consequently does not pull the verb to it. Thus, while the verb can formally make use of this position as a landing site, the verb is not required to move to it. As in the target language, movement of the verb to I/T will account for sentences with the pattern m-V-š, and failure of this movement will account for sentences with the pattern m-š XP, subsequent to cliticization of -š onto m-. Alternatively, it could be assumed, this time in the spirit of the Maturation Hypothesis, that I/T is not available at this stage of development. The pattern m-V-š would then be derived by optional movement of the verb to Neg, and the pattern m-š XP would be the result of the verb failing to make use of the option of moving to Neg. It seems to us that either of the two analyses would account for the derivation of negative sentences in early Palestinian Arabic. Although the two hypothesis may be fundamentally different from each other when applied to other properties of child language at this stage (Mohammed, in preparation), they both have the desired effect as far as the derivation of negative sentences is concerned. In the next few paragraphs we will outline an alternative approach to sentence structure at this stage of development, which has the distinctive property of combining elements from both the Continuity Hypothesis and the Maturation Hypotheses. The alternative approach is based on a sub-classification of functional categories, into those which play a (crucial) role in the interpretation (i.e., the representation of the meaning) of the sentence and those which do not. The former are available at the early stage of development, and, moreover, project X-bar structures in the same way they do in the target language. The latter, however, are not available, and therefore do not project in the sentence structure. While the desirability of including functional categories which play a role in interpretation in the sentence structure is self-evident, the idea that they project X-bar structures as in the target language is not necessarily so. It is conceivable that these categories are present in the form of adjuncts of the predicate phrase (Tsimpli 1992). The

84

IBRAHIM MOHAMED & JAMAL OUHALLA

decision to assign these categories X-bar structures is based on purely syntactic grounds having to do with the derivation of the two major patterns of negation. The underlying reasoning is that the pattern mV-š in early child Palestinian is derived in terms of the same mechanism responsible for its derivation in the target language, that is by movement of the verb to an appropriate head position in the sentence structure. Thus, our hypothesis concerning the sentence structure at the early stage is not motivated solely on the basis of considerations having to do with meaning, but also on the basis of purely syntactic grounds. A clear candidate of functional categories which do not play a role in interpretation is AgrS. Although some AgrS morphemes, typically the third person morphemes, appear with most occurrences of the verb at the early stage, there is a general lack of agreement between these morphemes and the subject which arguably indicates the absence of AgrS and its X-bar projections. The following are representative examples: (20) a. tiktib risaale baba (3FS)write letter daddy "Daddy is writing a letter." b. ihkii-lak mama (3MS)talk-to-you mummy "Mummy is talking to you." c. taxxu layla ddiib shoot (3PL) Layla wolf "Layla shot the wolf." d. baba w mama rah daddy and mummy leave(3MS) "Daddy and mummy have left."

2.3

2.1

2.0

1.11

Examples (20a-b) involve an 'error' in gender agreement, and (20c-d) involve an error in number agreement, between the subject and the verb. As Tsimpli (1992) points out on the basis of similar data from early Modern Greek and other languages, these agreement 'errors' are indicative of the absence of AgrS and its X-bar projections insofar as the latter provide the structural domain in which the configurational relationship of Spec-Head agreement holds between AgrS and the subject located in Spec-AgrSP. In other words,

NEGATION AND MODALITY IN EARLY CHILD ARABIC

85

although subject agreement morphemes are present, they display (nonagreement) properties which indicate that they are not structurally in a Spec-Head agreement relation with the subject. Note that once a noun phrase and an AgrS are structurally in a Spec-Head agreement relation, agreement between them in person, number, and gender features is mandatory. The fact that the agreement morphemes do not agree in these features with the subject in early child language indicates that the two categories are not structurally in a Spec-Head agreement relation. This would follow if the subject agreement morphemes are assumed not to project an X-bar structure. The presence of the subject agreement morphemes in early child Greek is attributed by Tsimpli to the fact that in this language the root forms of verbs in general do not constitute well-formed units of morphophonological representation separately from a subject agreement morpheme. In other words, the presence of the subject agreement morphemes is due to morphophonological considerations, rather than to a structurally-based syntactic requirement. In view of the fact that Arabic dialects in general resemble Modern Greek in that the root forms of verbs also do not constitute well-formed units of morphophonological representation separately from subject agreement morphemes, Tsimpli's analysis extends naturally to early child Palestinian. Another functional category which is arguably also absent at the early stage is I/T. This is indicated by the virtual absence of elements associated with this category, namely the (near) future Modal verb rat and the auxiliary verb kaan "be". This view does not necessarily mean that no time distinctions are made at the early stage. There is evidence which suggests that time distinctions at this stage are made in terms of aspectual categories (Tsimpli 1992), which in Arabic dialects are built into the morphology of the verb. For lack of space we cannot discuss the evidence here. The reader is referred to Mohammed (in preparation) for a detailed discussion of the temporal properties of sentences in early child Palestinian. For the moment, note that to the extent that aspectual information can indeed provide a time frame for the interpretation of sentences, the idea that I/T is absent at the early stage is not necessarily in conflict with the hypothesis that categories

86

IBRAHIM MOHAMED & JAMAL OUHALLA

which do not play a crucial role in interpretation are absent from the sentence structure at the early stage of development. Turning now to functional categories which play a role in interpretation, Neg is a clear candidate, for obvious reasons. However, here we will take the view that not all negative sentences include Neg and its X-bar projections. Whether Neg is present or not depends on whether the negation elements have or do not have a modality reading. When the negation elements do not have a modality reading, that is, when they function as pure sentential negators, they project an X-bar structure of their own. However, when they have a modality reading they occupy another functional position which we will call here Mod(ality). This distinction is based on conclusions reached in Ouhalla (1992, 1993) in relation to Standard Arabic which has two negation elements, laa (and its temporal variants lan (future) and lam (past)), and maa which is unmarked for tense. Following a long tradition, Ouhalla argues that while the former functions as a pure sentential negator, the latter functions as a negative modality element, where the notion modality is understood in a broader sense (Searle 1983, Palmer 1990). In structural terms, this means that laa projects an IP-internal X-bar structure of its own and interacts with other functional categories, in particular Tense (Benmamoun 1990, 1992). In contrast, maa occupies a peripheral position usually reserved for certain modality particles. In other words, maa is the morphological realisation of negative modality features. Although the spoken dialects of Arabic generally have done away with laa, and consequently do not make an overt distinction between pure negation elements and negative modality elements, we believe that the particles m-š do have this dual function, at least in Palestinian Arabic. As far as early child Palestinian is concerned, which is the focus of our attention here, the dual function of the negation particles is clear from the data presented and discussed above. Recall that a negative sentence can have a modality reading irrespective of the pattern of negation it instantiates. Presumably, an adequate analysis of the use of negation at this stage should reflect this dual function of the negation elements. Our position here is that this fact is not simply a matter of the possible interpretations which can be assigned to a given uniform structure, but is rooted in a difference in the structural status

NEGATION AND MODALITY IN EARLY CHILD ARABIC

87

of the negation elements. The negation elements have a modality reading when they occupy the Mod position, and do not have a modality reading when they do not occupy the Mod position. Accordingly, negative sentences where the negation elements do not have a modality reading have a structure roughly along the lines shown in (21): (21)

NegP Spec

Neg'

-š Neg

XP

Although movement of the head X of the predicate XP to Neg is generally optional, it is also dependent on whether X is a verbal category. If X is not a verbal category, it cannot move to Neg, for reasons having to do with the m-selectional properties of the negation elements in the Nablus dialect discussed in section 2. Consequently, -š cliticizes onto m- for morphophonological reasons. This scenario accounts for the derivation of neutral negative sentences which instantiate the pattern m-s XP, where X is a nonverbal category. When X is a verbal category, it may or may not move to Neg. If it moves, the pattern m-V-s is derived, subsequent to cliticization of -š onto the complex under Neg. If, on the other hand, X does not move to Neg, the pattern m-š XP is derived, again subsequent to cliticisation of -5. These two scenarios account, respectively, for neutral negative sentences which instantiate the patterns m-V-š XP and m-š XP, where X is a verbal category. Negative sentences where the negation elements have a modality reading have the structure shown in (22):

88

IBRAHIM MOHAMED & JAMAL OUHALLA (22)

ModP Spec

Mod'

The scenarios which can take place on the basis of this structure are the same as with (21), depending on the categorial features of X. Recall that nominal sentences with the pattern m-š XP and verbal sentences with either the pattern m-š XP or the pattern m-V-š (XP) can have a modality reading conveyed by the negation elements. The difference is that while in (21) the negation elements function as sentential negators, in (22) they function as negative modality elements. The position occupied by the negation elements in (21) and (22) can also be occupied by the anaphoric negator la?, given that this element can also function either as a pure sentential negator or as a negative modality element. On the assumption that this element is not a bound morpheme, and therefore cannot attach to any category whether verbal or nominal, the fact that it systematically appears preceding the predicate follows. In other words, movement of a predicative head category to the position occupied by la? is excluded for the same (morphophonological) reason the same process is excluded in nominal sentences negated with the particles m-š. 5. Conclusions In this paper we have examined the properties of negative sentences in early child Palestinian Arabic, and found that their major characteristic is the absence of the strict correspondence between the two major negation patterns and syntactic contexts found in the target language. While nominal sentences are invariably 'correctly' negated with the pattern m-š XP, verbal sentences are freely negated either with the 'incorrect' pattern m-š XP or the 'correct' pattern m-V-š. We have concluded from the properties of negative verbal sentences that their derivation at this stage lacks obligatory verb-movement to a

NEGATION AND MODALITY IN EARLY CHILD ARABIC

89

head category situated higher than Neg, the process responsible for the derivation of the pattern m-V-s in the target language. While the absence of verb-movement to a position higher than Neg in early child Palestinian can in principle be accounted for in Continuity or Maturation terms, we have argued that other considerations seem to point to a position closer to that of the Maturation Hypothesis. On the other hand, the productive nature of the pattern m-V-š with verbal sentences, and its absence with nominal sentences, inevitably point to the presence of a productive process of verb-movement, and therefore to the presence of Neg and its X-bar projections. Consequently, we have been led to adopt the position whereby AgrS and I/T are absent, whereas Neg is present. Furthermore, the fact that negation elements can convey a modality reading has led us to make a structural distinction between negative sentences with a neutral reading and those with a modality reading. The former instantiate Neg and its X-bar projections, whereas in the latter the negation elements occupy a Mod(ality) position. We have indicated that a similar distinction between negative sentences is overtly made in Standard Arabic, as concluded in Ouhalla (1992, 1993) following a long tradition in Arabic linguistics.

REFERENCES Abdo, Dawood & Salwaa Abdo. 1986. Dirasa fi mufradat tiflayn. Kuwait City: Dhaat al-Salaasil Publications. Abu El-Haija, Lutfi. 1981. The Acquisition of the Negation System in Arabic as Spoken in Jordan. Ann Arbor, Mich.: UMI Dissertation Services. Aoun, Joseph. 1985. A Grammar of Anaphora. Cambridge, Mass.: MIT Press. Benmamoun, Elabbas. 1990. "Verb Movement and Negation". Proceedings of NELS 21 ed. by T. Sherer, 17-31. Amherst: University of Massachusetts. . 1992. Functional and Inflectional Morphology: Problems of projection, representation and derivation. Ph.D. dissertation, University of Southern California, Los Angeles. Bloom, Lois. 1991. Language Development: From two to three. Cambridge: Cambridge University Press. Chomsky, Noam. 1986. Barriers. Cambridge, Mass.: MIT Press. Hoekstra, Teun & Peter Jordens. 1991. "From Adjunct to Head". Paper presented at the GLOW Workshop, Leiden.

90

IBRAHIM MOHAMED & JAMAL OUHALLA

Mohammad, Mohammad. 1988. The Sentence Structure of Arabic. Ph.D. dissertation, University of Southern California, Los Angeles. Mohammed, Ibrahim. In preparation. The Grammar of Early Child Arabic. Ph.D. dissertation, Queen Mary and Westfield College, London University, London. Omar, Margaret. 1973. The Acquisition of Egyptian Arabic as a Native Language. The Hague: Mouton. Ouhalla, Jamal. 1992. "Focus in Standard Arabic". Ms., Queen Mary and Westfield College, London University, London. . 1993. "Negation, Focus and Tense: The Arabic maa and laa". To appear in Rivista Linguistica 5.3: Thematic issue on negation ed. by Liliane Haegeman, 275-300. Pierce, Amy. 1989. On the Emergence of Syntax: A crosslinguistic study. Ph.D. dissertation, MIT, Cambridge, Mass. Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax: The nature of early child grammars of English. Oxford: Basil Blackwell. Smadi, Oglah. 1979. The Acquisition of the Jordanian Arabic Interrogation and Negation by a Three-year-old Native Speaker of Jordanian Arabic. Ph.D. dissertation, The University of Texas at Austin. Tsimpli, lanthi-Maria. 1992. Functional Categories and Maturation: The prefunctional stage of language acquisition. Ph.D. dissertation, University College London. Weissenborn, Jürgen. 1990. "Functional Categories and Verb Movement: The acquisition of German syntax reconsidered". Ms., Max-Planck-Institut für Psycholinguistik.

n PERSPECTIVES FROM EXPERIMENT-BASED STUDIES

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING: EVIDENCE FROM ARABIC SABAH SAFI-STAGNI King Abdulaziz University

1. Introduction Two major issues of interest to linguists in morphological theory concern the representation of morphological information, i.e., whether morphological information is represented in the lexicon or is part of the syntactic component of grammar, and the status of inflectional vs. derivational morphology, i.e., whether or not they form distinct autonomous processes. The first of these two issues is less controversial than the second, since there is strong evidence that points towards the lexicon as the seat of most, if not all, morphological information. Lapointe (1980), for example, advocates a strong lexicalist approach in which all morphological information is represented in the lexicon. Lieber (1983) and the lexical phonological literature that followed also assign both derivational and inflectional information to the lexicon. And although there are solid arguments to treat inflectional and derivational material as different in some way in their grammatical status, there is still much debate over an established satisfactory treatment of this distinction. (See Anderson 1988 for a detailed theoretical discussion of both issues). The model of lexical morphology adopted here is the one proposed by Kiparsky (1982) in which morphology is 'level-ordered' and word formation rules apply in a cyclic fashion. Kiparsky (1982) argues for a two level lexical morphology (see Figure 1)..Level I is responsible for irregular morphological forms that trigger phonological modifications in the base, such as keep-kept and divine-

94

SABAH SAFI-STAGNI

divinity. Level II, on the other hand, is responsible for regular morphological forms including affixes that do not trigger any phonological processes that affect the root morpheme, such as walkwalked and sleep-sleeping. Figure 1: Kiparsky's (1982) model of lexical morphology Irregular morphology

Level I

↓ Regular morphology

Level II

Mirroring the interest of formal linguists, the 1980s have witnessed an increased interest in psycholinguistic research regarding the role of morphological structure in lexical processing. (See Butterworth 1983 and Cutler 1983). The question often asked is how can the formal characteristics of linguistic competence account for linguistic performance? In other words, how can the recognized formal structure of lexical items be used to explain lexical processing during comprehension or production? Central to this research is the controversy over the processing of morphologically complex words (defined as words made up of two or more morphemes), and whether these lexical items are represented in the lexicon as wholes or in decomposed form. The two possible treatments offered for the lexical organization and processing of these complex words are the whole word hypothesis and the lexical (de)composition hypothesis. According to the whole word hypothesis, morphologically complex words are represented in the lexicon as wholes, i.e., each word corresponds to one lexical entry. On the other hand, the lexical (de)composition hypothesis states that morphologically complex words are represented in terms of their constituent morphemes, i.e., each morpheme represented separately. In other words, a complex word such as walked has one lexical entry in the lexicon according to the former hypothesis but has a decomposed form with two lexical entries of walk and -ed according to the latter.

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING

95

One experimental paradigm used to investigate lexical processing involves lexical decision tasks. These tasks require subjects to decide whether a visually presented stimulus is a lexical item or not, and their reaction times (RT) are then measured. Within this paradigm, results from whole word priming (Fowler et al. 1985), whole word frequency effects (Taft 1979), stem and surface frequency effects (Andrews 1986; Gordon 1983), repetition (Laudanna & Burani 1985), and prefixed vs. suffixed words (Segui & Zubizarreta 1985) as well as word vs. nonword stimuli (Miceli & Caramazza 1988) have all been used to argue for the crucial role of morphological structure in lexical processing. Another methodological tool often used employs the reading task with normal and language impaired subjects. (For a theoretical discussion on the relevance of data from normal as well as patients with acquired language disorders, see Caramazza 1984). Data from pathology provides an additional source of empirical tests of models of morphological processing. Patterson (1982) and Job & Sartori (1984), for example, use error data from patients with acquired dyslexia to investigate the nature of the language processing impairment that leads to morphological errors in reading aloud single words. Such studies have been used to distinguish between alternative models of lexical representations in the mental lexicon. This research focuses on lexical processing of Arabic, a language rich in morphology, during a reading task. Data collected from an aphasic patient with a case of Alexia (a reading disorder) and Agraphia (a writing disorder) will be used to show the role morphological structure plays in the organization and processing of complex words. With language impaired subjects, the reading difficulty results from damage to reading-specific mechanisms that require access to the lexicon. A possible theoretical framework for the lexicon (Figure 2) identifies two input systems: one visual for visually presented material, for example during a reading task, and the other phonological for material presented using the auditory modality. This model also includes a syntactic-semantic processor which contains information about the meaning and the syntactic characteristics of the word. The two output processing systems include a graphemic as well as a phonologic processor. The former is responsible for the orthographic specifications of words used, for example, during

96

SABAH SAFI-STAGNI

writing, while the latter is responsible for the phonologic specifications used if these words are spoken. Figure 2: Theoretical framework for reading/writing Auditory stimulus

Visual stimulus ↓ Graphemic Input Processor

i Phonological Input Processor Central SyntacticSemantic Processor Phonologic Output Processor

↓ Spoken output

Graphemic Output Processor

↓ Written output

A reading-aloud deficit, therefore, must involve either the visual input, syntactic-semantic, and/or phonologic output processors. (See the bold lined areas in Figure 2). According to the model outlined above, errors such as madrasah "school" for maktabah "library" are considered semantic errors, martabah "job scale" for maktabah are phonologic, kaatib "writer" for maktabah are morphologic, and makiinah "machine" for maktabah are graphemic errors. 2. Case History TA is a 42-year-old Saudi male from Makkah with college level education. He suffered a Cerebral Vascular Accident (CVA) in 1987. Several CT and MRI scans were performed immediately during his hospital stay which showed a large cerebral hemorrhage in the left inferior temporal area. This left him with a right-sided weakness and an expressive aphasia. Since 1987, he has been undergoing intermittent speech therapy which mostly used the written form. His performance on auditory verbal comprehension, visual comprehension, and repetition was within normal ranges. His speech was fluent, characterized by complex sentence structures which were replete with morphological errors (la), some semantic errors (lb) as well as some phonological errors (1c) and some hesitation for word

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING

97

retrieval. (For more details of TA's linguistic performance, see SafiStagni 1991, 1992.) (1)

Target a. yiruuh "he goes" b. musajil "recorder" c. badu "Bedouins"

TA's Productions raah "he went" talafon "telephone" balad "town"

Despite his six years of therapy using the written material, his reading and, even more, his writing abilities were severely impaired and he showed a tremendous amount of frustration during the experiment. 3. The Experiment The single word processing task design was used. The patient was asked to read aloud five groups of single words which were selected to investigate lexical parameters that may affect his performance. All stimuli material were mid-to-high frequency items matched in each category for number of syllables and, almost always, number of letters. The parameters are discussed in the remaining part of this section. Grammatical category Twenty single morphemic bisyllabic words equally distributed among nouns (e.g. safhah "page"), verbs (e.g., lapiba "he played"), adjectives (e.g., pajiib "strange") and function words (e.g., hunaak "there"). Concreteness Bisyllabic monomorphemic pure nouns using ten abstract (e.g., xidmah "service") and ten concrete (e.g., saxrah "rock"). Written vs. unwritten words Bi- and trisyllabic single morpheme words, half of which were common words with written forms (e.g., haqiiqah "truth"), while the other half were common colloquial Hijazi Arabic words that have no written form (e.g., bahlalah "mockery").

98

SABAH SAFI-STAGNI

Length This group consisted of twenty single morpheme nouns equally distributed among monosyllabic (e.g., baab "door"), bisyllabic (e.g., walad "boy"), trisyllabic (e.g., waraqah "paper") and quadrisyllablc forms (e.g., saydaliyyah "pharmacy") and (muqaaranah "comparison"). Simple vs. complex words Simple here is defined as the citation or base verb form "to do", while complex encompasses derived forms such as mafpal and faapil as well as the inflected forms of the verb which include inflections for tense, person, number and gender:fapalti "you (fem.sg.) did", yafpalu and so on. This category made up the largest number of test items. It consisted of ten high frequency triconsonantal verbs treated as the citation or base forms (e.g., daxala "to reach"), (tabipa "to follow"), and "to enter"), {balaya (Sahara "to mention"), and ten of each verb's derived or inflected forms (e.g., daxalna "we entered"), (daxaluu "they entered"), (yadxul "he enters"), (daaxil "in"), (madxuul "entered"), (daxiil "foreigner"), (mudaaxala "entered to"), (madxal "entrance") and so on. All stimuli material were boldly printed on 5 x 6 cards and were randomly presented to TA in isolation on a clear table (see Figure 3). Figure 3: An example of bisyllabic stimuli visually presented

TA's performance was audio- and video-recorded in a sound proof room and his productions were broadly transcribed by two independent transcribers. TA. was given a score of one (+1) if he correctly produced the target item (even after several attempts) and zero (+0) if he produced errors, whether visual, semantic or phonologic.

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING

99

4. The Results and Discussion No significant effects were found for the first three parameters of grammatical category, concreteness and written vs. unwritten forms. With grammatical category, TA. read correctly verbs such as daraba (2a) and lapiba nouns such as saßiah (2b) and kitaab and adjectives such as latiif (2c) and as well as function words such as Paydan (2d) and hatta Target (2) a. daraba b. safhah c. latiif d. Paydan

"he hit" "page" "gentle" "also"

TA's Productions daraba safhah latiif paydan

He also successfully read concrete as well as abstract nouns correctly producing jabal (3a) and qitaar (3b) as well as xabar (4a) (4b). and difaap (3)

(4)

Target a. jabal b. qitaar c. manzil a. xabar b. difaap c. mudiir

"mountain" "train" "house" "report" "defense" "director"

TA's Productions jabal qitaar manzil xabar difaap mudiir

Errors were expected with the third group of words since it involved words that have no legitimate written form, and therefore, have no written representation. Interestingly enough, though, there was also no significant effect due to the Written vs. Unwritten parameter. Although the words with the Unwritten forms were met with questioning and ridicule, he read them with a similar success rate to the regular written forms. For example, he had no difficulty reading (6b) as he forms such as saayip (6a), and majaayah correctly read regular forms such as paalim (5a), and da jaa jah (5b). Target "scientist" (5) a. paalim b. dajaajah "chicken" c. zalzalah "quake"

TA's Productions paalim dajaajah zalzalah

100

SABAH SAFI-STAGNI (6) a. saayip "a bum" b. majaayah "silliness" c. bahlalah "mockery"

saayip majaayah bahlalah

According to the model sketched above, these particular items have no graphemic representation in the lexicon and, therefore, must be accessed through their phonological representation. In other words, there has to be a grapheme-to-phoneme mapping mechanism that converts the orthographic representation to a phonologic representation before accessing the central syntactic-semantic processor. In TA's case, this mechanism seems to be intact. Single morphemic words in Arabic are not, strictly speaking, simple forms but are actually composed of two discontinuous nonconcatenative morphemes: one consisting only of the consonants which carry the primary meaning, while the other consists of vocalic patterns which mark categories such as tense and aspect. (See McCarthy 1981, 1986 for a full discussion.) These results are, therefore, not so insignificant since they can be used to argue that forms are not represented in a decomposed form and imply a whole word representation and processing. TA's reading, however, was slightly affected by the increased length of the stimulus which resulted in phonological errors. He had no difficulty reading monosyllabic items, such as baab (7a) and bint (7b) or bisyllabic words such as tawfiiq (8a) and qaryah (8b). Target (7) a. baab "door" b. bint "girl" c. bayt "home" (8) a. tawfiiq "success" b. qaryah "village" c. markab "ship"

TA's Productions baab bint bayt tawfiiq qaryah markab

He did, however, have some difficulty reading tri- and quadri-syllabic words. For example, (9c) shows the exchange of /h/ and /f/ the two consonants in syllable initial position of the contiguous syllables /hii/ and /fah/ in sahiifah

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING Target a. mahkamah b. maktabah c. sahiifah (10) a. saydaliyyah b. pamaliyyah c. muqaaranah

(9)

"court" "library" "paper" "pharmacy" "operation" "comparison"

101

TA's Productions mahkamah maktabah safiihah sayyaliyyah palamiyyah qa .. na .. nuqaaran

TA produced more phonological type errors on the quadrisyllablc forms. In (10a), for example, he substituted one phoneme for another (/y/ for lál) while in (10b) he exchanged phonemes in contiguous syllables (/m/ in the second syllable with HI in the third syllable). In (10c) he deleted phoneme sequences; but no errors of addition were noted. TA's performance, however, was significantly affected by the last parameter of morphological complexity. His reading attempts of this group of words resulted in the most interesting type of errors. With a 100% accuracy, TA read all simple forms (11) such as, balaya Sahara fataha daxala Target "to reach" (11) a. balaya "to follow" b. tabipa c. öakara "to remind"

TA's Productions balaya tabipa Öakara

With the remaining ninety items representing the morphologically complex forms in this group, his performance dropped to a 31% accuracy, showing a severe impairment in reading complex forms. (Compare examples (12-16) below with the examples above.) This breakdown must occur at a level of processing where complex lexical forms are treated by the processor as different from simple forms. Examination of the patient's errors reveals that they largely affected the inflectional specification of the word—see (12), where he produced correct bases but within category errors on inflections such as tense, number, and gender. For example, while the target in (12a) was (balayna ), TA produced balayti "you (fem.sg.) reached", balayni (it reached me), and balaya (he reached). The same within category substitution is true for (12b) substituting the 2nd masc. pi. target (-turn) with 1st pi. (-na) in his first attempt, 1st sg. (-t) in the second attempt before finally settling for the 3rd masc. sg. or the base

102

SABAH SAFI-STAGNI

form. In (12c) he even attempted to spell out the word but again finally settled for (dakar). He opted more often than not to produce the simple or Base form. Target (12) a. balayna "we reached" b. hadartum "you (masc. pl.) attended" c. ðakartunna "you (fem.pl.) mentioned" d. xara jat

"she went out"

TA's Productions balaγti.. balaγni.. balaya hadarna .. hadart.. hadar ðakart.. ðak.. ðaa . kaa . raa . nu .. ðakarnu .. ðakar xara jaa .. xara jt.. xara jti.. xaraj

The discrepancy in his performance on simple vs. complex forms first points out the significance of the morphological structure during processing, and second indicates (as opposed to what has been suggested above) that words are not processed as wholes. If they were processed as wholes, his performance would be the same on simple as well as on complex forms. TA has no difficulty accessing stems and no difficulty accessing inflectional or derivational morphemes. He does, however, have great difficulty in combining the two types of information. The locus of his deficit seems to be at the level of combining stems with the appropriate inflectional affixes. These errors in (12-16) could be interpreted as evidence that lexical production includes some processing stage at which inflected and derived forms are morphologically decomposed, thus supporting the lexical (de)composition hypothesis. Note that these type errors are not due to the length parameter since his errors are not reduced in syllable length and are often phonologically sound. Also note that the stems of these complex forms, or what is known as the Base forms, remain intact. It is reasonable, therefore, to assume that derivational and inflectional affixes are represented separately from the base morphemes. Further analysis of his total productions shows that the drop in accurate performance is significantly different for derivationally versus inflectionally complex words, with a more than 50% accuracy for the derivational forms but less than 2% for inflectional forms (compare the derived forms in (13) below with the inflected forms in (12) above). Thus, these two types seem to be treated differently by

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING

103

the processor. Derivational information appears to be more tightly attached to the base form than inflectional information. Target (13) a. kaatib "writer" b. mahdar "report" c. maftuuh "open (adj.)"

TA's Productions kaatib mahdar maftuuh

This dissociation has already been reported by Miceli & Caramazza (1988) with the spontaneous speech and single word repetition performance of patient FS, who had a selective impairment in producing inflectional morphology while derivational morphology was left intact. Furthermore, within the derivational group there was also a noted differential performance. Derived forms such as kaatib mahdar and maftuuh were produced without much hesitation or difficulty. However, forms such as mukaatabah (14a), tahaddara (14b), and Pistixraa (14c) were often rejected and the attempts abandoned. Target (14) a. mukaatabah "reciprocal writing" b. tahaddara "he became civilized" c. pistixraaj "he derived"

TA 's Productions kaatab .. mu .. ka.ta.bah .. hadara .. tahdur .. ?a .. sa .. ta .. xaaraajaa ..

The success rate seemed to be a function of frequency of the derivational structure. Although it is premature to speculate on the nature of the representations of derived forms, it is possible to suggest that high frequency and highly productive patterns such as faapil and maflala appear to be processed as wholes while low frequency and less productive patterns such as mufaapalah and pistifîaal are treated as decomposed forms resulting in errors during their production. According to Kiparsky's (1982) model sketched above, high frequency derived forms constitute Level I morphology and would have separate entries in the lexicon. These forms can be accessed without morphological parsing. Inflectional affixes and low frequency derived forms belong to Level II, would have a single entry, and would involve morphological decomposition during access.

104

SABAH SAFI-STAGNI

Further evidence for the (de)composition of morphologically complex lexical items comes from illegal forms produced by TA during reading. While attempting to read jalastum (15), for example, he produced jalashum. Target (15) jalastum "you (masc. pl.) sat"

TA's Production jalasna .. ja . la .. jalas .. jaa . laa .. Jalas Pintu .. jalas .. jalashum

TA tried several times, even providing the lexical form of the pronoun (visual recognition and comprehension of the affixed form appear to be intact). However, he finally produced jalashum combining a stem such as jalas with a possible but illegal inflectional form (-hum) If complex lexical items were represented and processed as wholes, the chance for such an error would not arise. To further probe the status of morphological constituents during production, an additional short experiment was conducted. TA was asked to read an additional list which consisted of Non-Arabic words. The list contained twenty possible but nonexistent forms, some simple and some complex. The complex forms were partly derived and partly inflected using genuine Arabic inflectional and derivational patterns. After much protest, TA again read accurately all the simple forms and a few derived ones predominantly producing the base forms for the complex nonwords. Target (16) a. xulum b. maatix c. malpuus d. lapasna e. sarab f. taγdiidahum

Written

TA's Production xulum maatix lapuus lapas sarab γadiih

5. Conclusion The preliminary results shown here suggest that there is a stage in lexical processing at which morphologically complex lexical items are represented in their morphemic constituents. Such errors as the ones produced by TA during a production task can only arise if component morphemes are represented separately. At this processing stage,

MORPHOLOGICAL STRUCTURE AND LEXICAL PROCESSING

105

inflectional, but not high frequency derivational, affixes are represented separately from their morphological bases. Reading error data from Arabic provide support for the lexical decomposition hypothesis. The status of inflectional vs. derivational morpheme representations in Arabic, however, remains open for further investigation.

REFERENCES Anderson, Stephen R. 1988. "Morphological Theory". Linguistics: The Cambridge Survey I ed. by Frederick Newmeyer, 146-191. Cambridge: Cambridge University Press. Andrews, Sally. 1986. "Morphological Influences on Lexical Access: Lexical or nonlexical effects?" Journal of Memory and Language. 25.726-740. Butterworth, Brian. 1983. "Lexical Representation". Language Production II ed. by Brian Butterworth, 257-294. New York: Academic Press. Caramazza, Alfonso. 1984. "The Logic of Neuropsychological Research and the Problem of Patient Classification in Aphasia". Brain and Language 21.9-20. Cutler, Anne. 1983. "Lexical Complexity and Sentence Processing". The Processes of Language Understanding ed. by Giovanni B. Flores D'Arcais and Robert J. Jarvella, 43-79. London: John Wiley & Sons. Fowler, Carol A., Shirley E. Napps, & Laurie Feldman. 1985. "Relations Among Regular and Irregular Morphologically Related Words in the Lexicon as Revealed by Repetition Priming". Memory and Cognition 13.241- 255. Gordon, Barry. 1983. "Lexical Access and Lexical Decision: Mechanisms of Frequency Sensitivity". Journal of Verbal Learning and Verbal Behavior. 22.146-160. Job, Remo & Giuseppe Sartori. 1984. "Morphological Decomposition: Evidence from crossed phonological dyslexia". The Quarterly Journal of Experimental Psychology 36A.435-458. Kiparsky, Paul. 1982. "Lexical Morphology and Phonology". Linguistics in the Morning Calm. Seoul: Hanshin Publishing Company. Lapointe, Steven G. 1980. A Theory of Grammatical Agreement. Ph.D. dissertation, University of Massachusetts, Amherst. Laudanna, Alessandro & Cristina Burani. 1985. "Address Mechanisms to Decomposed Lexical Entries". Linguistics 23:5.775-792. Lieber, Rochelle. 1983. On the Organization of the Lexicon. MIT Ph.D. dissertation, Cambridge, Mass. McCarthy, John. 1981. "A Prosodic Theory of Non-concatenative Morphology". Linguistic Inquiry 12.373-418. . 1986. "OCP Effects: Gemination and Antigemination". Linguistic Inquiry. 17.207-263. Miceli, Gabriele & Alfonso Caramazza. 1988. "Dissociation of Inflectional and Derivational Morphology". Brain and Language 35:1.24-65.

106

SABAH SAFI-STAGNI

Patterson, Karalyn E. 1982. "Neuropsychological Approaches to the Study of Reading". British Journal of Psychology 72:2.151-174. Safi-Stagni, Sabah. 1991. "Agrammatism in Arabic". Perspectives on Arabic Linguistic III ed. by Bernard Comrie & Mushira Eid, 251-270. (= Current Issues in Linguistic Theory, 80). Amsterdam & Philadelphia: John Benjamins. . 1992. Normal and Pathological Breakdown in Arabic. Ph.D. dissertation, Louisiana State University, Baton Rouge. Segui, Juan & Maria-Luisa Zubizarreta. 1985. "Mental Representation of Morphologically Complex Words and Lexical Access". Linguistics 23:5.759774. Taft, Marcus. 1979. "Recognition of Affixed Words and the Word Frequency Effect". Memory and Cognition 7.263-272.

EXPERIMENTAL INVESTIGATIONS OF ARABIC SYLLABLE STRUCTURE*

BRUCE L. DERWING, DILWORTH B. PARKINSON & RICHARD A. BEINERT University of Alberta, Brigham Young University, University of Alberta

1. Overview In this paper we will report on three experimental studies of Arabic disyllabic words: one to clarify questions about syllable boundaries and two others to assess the status of a variety of proposed phonological units, such as the segment, rime, syllable, and tier. 2. Syllable Boundaries in Arabic 2.1 Background on the pause-break task The first phase of this investigation was inspired by the work of Treiman & Danis (1988), who performed a series of experiments to assess the syllabic status of single intervocalic consonants in English, using both an oral string-inversion task and a written slash-insertion task. We concentrated first on string-inversion, knowing that many of our subjects would likely not be fully literate, but even this task proved to be too difficult for the first group of subjects, literate and semiliterate alike, who were tested in Cairo. When the same thing The authors wish to acknowledge the important assistance provided to them by Terry Nearey and Grace Wiebe (of the University of Alberta) and Sayyid Rawash (of Cairo University). The first author also acknowledges the financial support provided by the Social Sciences and Humanities Research Council of Canada (Grant No. 41091-0249).

108

DERWING, PARKINSON & BEINERT

happened with Derwing's first group of Blackfoot speakers in Alberta, the decision was made that some new task would have to be devised that would be easier for subjects and would still yield the kind of data that we were interested in. What we settled on was the 'pausebreak' task, which was an oral version of Treiman & Danis' slashinsertion technique. Rather than inserting slash marks to indicate syllable divisions in written representations, subjects were asked instead to choose among three alternative locations of pause-breaks in spoken strings. This technique is illustrated in (1) below, using the English word lemon. First, the word was pronounced with normal 'list' intonation, then repeated three times with short pauses inserted in different positions. (1) (a) le...mon (b) lem...on (c) lem...mon

Choice (a) here is le...mon, where the pause is inserted before the medial consonant, such that the /m/ in this case is treated as the onset of the second syllable. By contrast, in choice (b) lem...on, the break occurs after the consonant, which is now treated as the first syllable coda. Finally, choice (c) lem...mon, permits an ambisyllabic response, where the consonant appears in both of these positions (see Derwing 1992 for a fuller description of this task). Before proceeding with this technique with Arabic and the other languages, Derwing first conducted a test using a variety of English words like this one in order to compare the pause-break findings with those achieved earlier by Treiman & Danis (1988), and the results were quite satisactory, in that all of Treiman & Danis' main original findings emerged in this replication. Furthermore, a pilot study with Blackfoot speakers in southern Alberta indicated that this new technique was readily understood and quickly carried out even by speakers with very low educational levels who were completely unfamiliar with any writing system for Blackfoot; for this reason we were encouraged to try it out on Egyptian Arabic, as well.

EXPERIMENTS IN ARABIC SYLLABLE STRUCTURE

109

2.2 The Arabic pause-break study A sample of word-types used in the Arabic version of this test is shown in Table 1 below. Table 1 Arabic Word-Types for Pause-Break Task A . Words with Medial Single C (33 items, with 17 different segment-types) 1. libad 1 2. daras (a) l i . . . bad (a) da . . . ras (b) lib . . . bad (b) dar . . . ras (c) lib . . . ad (c) dar . . . as B . Words with 3. (a) (b) (c)

Medial CC Clusters (25 items) akbar 4. nidris ?a . . . kbar (a) ni . . . dris ?ak . . . bar (b) nid . . . ris ?akb . . . ar (c) nidr . . . is

C . Words with Medial Geminates /Long C's (12 items) 5. zakkar 6. darris (a) za . . . kkar (a) da . . . rris (b) zak . . . kar (b) dar . . . ris (c) zakk . . . ar (c) darr . . . is

Part A illustrates words with single intervocalic consonants. Choice (a) here is the 'onset' response, choice (c) the 'coda' response, and the middle choice, (b), the ambisyllabic response. Part B illustrates words with medial consonant clusters, and in these cases choice (b) involves dividing the syllables between these consonants, while the other two choices keep them together as either a complex onset, as in (a), or as a complex coda, as in (c). Finally, in Part C, are examples of words with medial long or geminate consonants, which are treated in ways paralled to the clusters. Half of the subjects had the order of the (a) and (c) choices reversed on the test, while the (b) choices remained the same throughout. The data presented in Table 2 were obtained from two groups of subjects tested in 1991. 1Libad, "mane of a lion, felt skullcap worn under or without a tarboosh, or a felt 'dervish hat".

DERWING, PARKINSON & BEINERT

110

Table 2 Results of Pause-Break Task Single Cs CC Clusters Geminates

Coda .06 (.09) .04 (.14) .07 (.16)

Onset .78 (.70) .08 (.17) .18 (.40)

Ambisyllabic .17 (.21) .88 (.69) .75 (.43)

The main group of 48 subjects were all students at Cairo University, and they were tested for us by Prof. S. Rawash. The smaller group, whose results are shown in parentheses, were 13 semiliterate subjects tested earlier in the year by Parkinson. As can readily be seen from this table, both the educated and semiliterate subjects converge on the following points: (1) Both show an overwhelming preference to treat single intervocalic consonants as syllable onsets. With one very specific exception in English, in fact, this was the same result Derwing has found in every other language he has tested so far, including Blackfoot, Korean, Swiss German, and, for the most part, English (see Derwing 1992 for this and the other cross-linguistic results to be discussed immediately below). So the obligatory onset principle is, so far at least, ALMOST universal. (2) Both Arabic subject groups also show a very strong preference to divide syllables between the two members of intervocalic CC clusters. This is scarcely surprising for Arabic, which normally allows no complex onsets or codas. Interestingly, however, Derwing has also found this result with all of the other languages he has looked at so far, even some that DO allow syllable initial and final clusters, so this result casts some doubt on the generality of the maximal onset notion. (3) Finally, on the third category of words we found an interesting divergence of response between our two subject groups. As seen in the table, the vast majority of the educated, highly literate subjects preferred the ambisyllabic or split response, thus treating geminate segments in much the same way as consonant clusters. For the less educated, semiliterate subjects, however, opinions were about evenly split between this response and an alternative one where the long consonants were accepted as syllable onsets, thus treating geminates in the same way as single segments. Though only 13 subjects are involved here, the validity of this result is attested to by the fact that Derwing's Blackfoot subjects, who were

EXPERIMENTS IN ARABIC SYLLABLE STRUCTURE

111

all completely unfamiliar with any writing system for their language, treated geminates in almost exactly the same way (see Derwing 1992). This raises interesting questions about the possible role that formal education might play in affecting judgments of this kind. 3. Intrasyllabic Units in Arabic: Evidence from Global Sound Similarity Judgments (SSJs) 3.1 Background on the SSJ task While the study just described was concerned with the question of syllable boundaries, the second one involved the question of the internal structure of syllables, or intrasyllabic units. A variety of experimental tasks have been employed to investigate this question in English, including productive word-blending by Treiman (1986), substitution-by-analogy by Dow (1987), as well substitution pattern identification and the use of global sound similarity judgments (or SSJs) by Derwing & Nearey (1986). We chose to go with SSJs on our first attempt with Arabic for two main reasons, both of which can be illustrated from a recent thesis by Bendrien (1992), carried out under Nearey's direction. First of all, our experience with English has been that this task is not only one that is very easy for subjects to perform, but also one that yields extremely reliable results. Basically, all subjects have to do is to listen to a pair of words or pseudo-words and to rate them for overall similarity in sound on some scale, usually 0-9 or 1-10. For some reason, subjects seem to be very good at performing this task, and different subjects are remarkably consistent in the kind of responses they give. This point is aptly illustrated in Bendrien's study of English CVC word-pairs, which were systematically varied in the following ways: each segment was matched by an identical segment in one or more of the three CVC positions, mismatched by a single or minor feature difference (e.g., /p/ vs. /b/ for either consonant, or /e/ vs. /æ/ for the vowels), or mismatched by a triple or major feature difference (e.g., /p/ vs. /z/ for consonants or /æ/ vs. /ay/ for the vowels). This study confirmed and extended the main prior results that Derwing & Nearey had found, which can be summarized as follows: (1) any error or mismatch lowers similarity judgments (no big surprise here); (2) a major error or multiple-feature mismatch produces a greater effect than a minor error or single-feature

112

DERWING, PARKINSON & BEINERT

mismatch (again, no big surprise, but it is gratifying that the technique is sensitive to such sub-segmental differences); (3) mismatches at the ends of words, and in vowels, produce larger effects than mismatches in the inital C; and, most importantly, (4) small one-feature mismatches in TWO segment positions have a greater effect than a single many-feature difference in one segment position. (Thus /pæp/:/bæb/ is judged less similar than /pæp/:/pæz/.) A linear regression analysis of Bendrien's data showed that a seven-factor model (including both the rime-match and three feature indicators) gives the best predictions overall, at r 2 = 94.4%, but the three-factor segment model also does a very good job on its own (at 85.8%); clearly, the segment contrasts are doing the vast bulk of the work here, while the other variables merely add fine tuning. Interestingly, however, the rime element VC also makes an independent and significant contribution. One important new finding also came out of the Bendrien study: by adding an 'acoustic tag' condition to drown out any transient acoustic effects (e.g., "/pæp/—OK, GO; /pæz/—OK, NEXT"), Bendrien also showed that the increased weight attached to the ends of words could not be dismissed as an artifact of a precategorical acoustic store, suggesting that stable phonological representations are accessed in the SSJ task, not transient phonetic or auditory ones. The use of this technique is thus indicated as an appropriate one for assessing the status of lexical representations in other languages, as well. 3.2 The Arabic SSJ study To turn now to Arabic, the phonological representation of Arabic words and syllables is a matter of some theoretical dispute. Onsetrime, body-coda, and non-hierarchical string structures have all been proposed, as well as an underlying form that separates C- and Velements out onto separate planes or tiers. The purpose of the study to be described was to clarify the status of these various hypothesized units and levels, by determining which, if any, of them participate in the prediction of SSJs in this language. 3.2.1 Procedure A list of 68 Arabic word-pairs was constructed by Beinert (1992), around 20 systematic variations, which normally involved one-, two-,

EXPERIMENTS IN ARABIC SYLLABLE STRUCTURE

113

or three-segment mismatches between the members of each pair, e.g., sakan-makan (one segment difference), rasal-rikal (two segments), and darab-dukub (three segments). About two-thirds of the pairs had the shared canonical form CVCVC, while the remainder all contained at least one word with a medial cluster or geminate. Each pair was scored on a scale from zero (least similar) to nine (most similar). Twelve subjects participated in the experiment—all native speakers of Egyptian, Lebanese, or Palestinian Arabic. The forms presented were from Educated Spoken Arabic, a "levelled" dialect familiar to all subjects. (See Beinert & Derwing 1992 for the detailed instructions and for the full list of Arabic words used.) 3.2.2 Analysis and results The first analysis involved making a priori predictions about SSJs on the basis of a simple match-mismatch scoring scheme for each of the four models under consideration. Thus, as illustrated in Table 3 below, for the non-hierarchical Segment model, similarity predictions were based on shared phoneme counts; on shared onset and rime unit counts for the On+Ri hierarchical model; on shared body and coda counts for the Bo+Co model; and on shared C-tier and V-tier counts for the nonlinear Tier model. Table 3 Illustrative Similarity Predictions for Four Alternative A Priori Models Word-Pair /ka$tab/-/ka$tib/ (Mismatches are underlined) Segment Model: Cl=k:k; Vl=a:a; C2=t:t; V2=a:i; C3=b:b Match=4/5= 80 Onset-Rime Model: Onl=k:k; Ril=a:a; On2=t:t; Ri2=ab:ib. Match=3/4=75 Body-Coda Model: Bol=ka:ka; Col (none); Bo2=ta:ti; Co2=b:b. Match=2/3=.67 Tier Model: C-Tier=ktb:ktb; V-Tier=aa:ai. Match=l/2=.50

114

DERWING, PARKINSON & BEINERT

The overall level of success achieved for each of these models for the full 68 test items was as follows, in terms of the proportion of variance accounted for: Segment (r2=71.0%), On+Ri (60.6%), Bo+Co (46.5%) and Tier (44.2%). Clearly, of the four a priori models considered, the Segment Model yields by far the best overall account of the variance, which is roughly on the same level as that achieved earlier in the English studies reported. Given the inherent arbitrariness of some of the scoring procedures used, however, and also considering the possibility that none of the four 'pure' models tested need necessarily supply a complete answer, a series of linear regression analyses were also performed on these data. The first of these corresponded to a segmental approach and involved inserting the variables C1 (for first consonant, scored as either same [=1] or different [=0] for each pair), V1 (first vowel), C2 (second consonant), etc., first limiting the comparisons to those 47 test items that employed the five-segment CVCVC structures throughout. The result was a very impressive r 2 of 92.0%. The parallel results for On+Ri (88.5%), Bo+Co (71.0%), Tiers (70.5%), and (especially) whole Syllables (44.7%) all fall short of this. Another notable feature of these results, however, is the rather impressive coverage achieved by the Tier model (r 2 = 70.5%), considering the fact that only two variables are involved (C-Tier and V-Tier), as compared to five in the pure Segment model. Interestingly, if C-Tier is added to the variable list along with the five segments, the coverage for the CVCVC subset increased to 93.4%, which was the highest figure achieved by any of the various combinations of variables that we tried in this exploratory study. Adding the V-Tier variable contributed nothing, however. This analysis is presented in detail in Table 4 below, which shows the coefficients for each variable, as well as the t-ratios (all significant at p < .01).

EXPERIMENTS IN ARABIC SYLLABLE STRUCTURE

115

Table 4 Linear Regression Analysis for Mean SSJ Similarity Ratings in Arabic (47 CVCVC stimuli only) Source Regression Residual Variable Constant C1 C2 C3 V1 V2 C-Tier

Sum of Sq 180.778 12.847 Coefficient -0.50699 1.52308 2.03831 1.49338 1.04570 0.650865 1.00028

df 6 40

Mean Sq 30.1 0.32117

S.E. of Coeff 0.3322 0.2453 0.2981 0.2419 0.1964 0.1943 0.3443

F-ratio 93.8

t-ratio -1.53 6.21 6.84 6.17 5.33 3.35 2.91

Assuming that this is about the best model we are going to find for predicting our SSJ results, a few very interesting conclusions follow from this: (1) Consonants count more than vowels, judging by the coefficients—this makes a great deal of sense for a language like Arabic, where only consonants are normally represented in spelling and the vowels taken for granted, as largely predictable from context. (2) Of the three consonants represented in CVCVC structures, the middle one (C2) counts the most; interestingly, it is only in the middle, intervocalic position where contrasts with consonant geminates and clusters are possible. (3) Of the two vowels represented, the first vowel (V1) counts more than the second, though both are significant; taken together with the previous observation related to C2, however, this may be an indication that a rime unit is involved, but we did not have sufficient variation in the shape of Ri in our stimuli to explore this possibility any further. Finally, (4) C-Tier (scored as a match only if all three consonants were the same) continues to make a significant, independent contribution, in addition to the individual contributions of each separate consonant. While this might be interpreted as evidence in support of the Tier model, the lack of relevance of the V-Tier variable detracts from this. Since C1...C2...C3 corresponds to the root in all of these words, we suspect

116

DERWING, PARKINSON & BEINERT

that this is an ancillary semantic or morphological effect. (This could also, of course, be an orthographic effect.) Finally, we have made some progress in extending the preferred models to cover some of the comparisons involving the six-segment CVCCVC items, as well, i.e., words with consonant clusters or geminates in medial position. Under the simplifying assumption that C2 comparisons be scored as mismatches whenever medial single vs. double comparisons were involved, the Segment +C-Tier Model outlined above yields an r 2 of 81.6% for all 68 words. An examination of residuals, however, showed four clear outliers, all involving underprediction of similarity scores, and all four of these turned out to be the cases where the single vs. double comparisons involved single vs. geminate counterparts. By rescoring these four items as matches, rather than mismatches, and by adding the variable 'Gem(inate) Match' to the analysis, the coverage increases to almost 89%, the best we achieved for the full set of items. As in the English case, however, predictions based on simple mismatches among segments account for the bulk of the variance in this study. Finally, in a new study currently in progress, we are now systematically extending the SSJ task to a fully balanced set of sixsegment CVCCVC words, permitting a better assessment of units like the syllable rime (VC) or body (CV). Shown in Table 5 is the linear regression analysis for the 13 subjects we have tested so far.

EXPERIMENTS IN ARABIC SYLLABLE STRUCTURE

117

Table 5 Linear Regression Analysis for Mean SSJ Similarity Ratings in Arabic (36 CVCCVC stimuli only) Source Regression Residual

Sum of Sq 63.4576 18.3060

Variable Constant C1 C2 C3 C4 V1 V2 C-Tier V-Tier

Coefficient -0.01410 1.52739 1.38095 1.34262 1.32047 0.60852 0.63076 0.95548 0.38786

df 8 27

Mean Sq 7.93200 0.67800

S.E. of Coeff 0.6845 0.3588 0.3481 0.3481 0.3379 0.5104 0.4955 0.4686 0.6351

F-ratio 11.7

t-ratio -0.02 4.26 3.97 3.86 3.91 1.19 1.27 2.04 0.61

As can be seen, the results are much the same as those found earlier for the five-segment CVCVC word-pairs: the Segment model gives the best set of predictions overall (r2=77.6%), with consonants again counting more than vowels and with C-Tier again making a significant independent contribution, but not V-Tier (as shown). Nor is there any evidence in these data of a significant contibution from either rime or body units. 4. Summary and Conclusions To summarize, the 'pause-break' technique appears to be a useful procedure for eliciting cross-linguistic information about syllable boundaries. It is simple, efficient, and, where comparisons are available, it seems to yield the same kind of results as more cumbersome production tests. It has also yielded a widespread crosslinguistic pattern in which single intervocalic consonants are preferentially treated as syllable onsets. There is a similar crosslinguistic preference to split intervocalic consonant clusters. In languages which have intervocalic geminates, however, such as Arabic and Blackfoot, while educated subjects treat these as ambisyllabic bisegments, less educated, semiliterate subjects are less certain of this,

118

DERWING, PARKINSON & BEINERT

as about half treat them as single-segment onsets. Finally, experimental evidence is accumulating to support the idea that the segment is the preferred representational unit in many languages, with both Arabic and English included. In line with this, good results have also been achieved in predicting SSJ results in Arabic on the basic of a pure segment model, but other factors (such as morphology) also seem to play some role. More data are required to assess the viability of the segment in certain other languages, where other units so far seem to be indicated (such as the mora for Japanese and the potentially indivisible syllable for Chinese).

REFERENCES Beinert, Richard A. 1992. Phonological Units in Arabic: An experimental study. B.A. Honors Thesis, University of Alberta, Edmonton. Beinert, Richard A. & Bruce L. Derwing. In press. "Segment, Rime, Syllable, Tier or Root? Evidence from Global Sound Similarity Judgments in Arabic". ESCOL '92 ed. by M. Bernstein, 1-10. Ithaca, N.Y.: DMLL Publications. Bendrien, Tracey A. 1992. Sound Similarity Judgments in English CVC's. B.A. Honors Thesis, University of Alberta, Edmonton. Derwing, Bruce L. 1992. "A 'Pause-Break' Task for Eliciting Syllable Boundary Judgments from Literate and Illiterate Speakers: Preliminary results for five diverse languages". Language and Speech 35.219-235. Derwing, Bruce L. & Terrance M. Nearey. 1986. "Experimental Phonology at the University of Alberta". Experimental Phonology ed. by John J. Ohala & Jeri J. Jaeger, 187-209. Orlando, Flor.: Academic Press. Dow, Maureen L. 1987. On the Psychological Reality of Sub-syllabic Units. Ph.D. dissertation, University of Alberta, Edmonton. Treiman, Rebecca. 1986. "The Division Between Onsets and Rimes in English Syllables". Journal of Memory and Language 25.476-491. Treiman, Rebecca & Catalina Danis. 1988. "Syllabification of Intervocalic Consonants". Journal of Memory and Language 27.87-104.

THE TIMING STRUCTURE OF CVVC SYLLABLES* ELLEN BROSELOW, MARIE HUFFMAN, SU-I CHEN, & RUOHMEI HSIEH State University of New York at Stony Brook

1.0 Introduction This paper reports on an investigation of the timing properties of particular syllable types in different dialects of colloquial Arabic. Our data are productions of sentences containing CVV, CVC, and CVVC syllables recorded by four speakers from Cairo, Alexandria, Beirut, and Damascus. We focus particularly on word-internal CVVC syllables, which are permitted in the Lebanese and Syrian dialects, but not in the Egyptian dialects. Different versions of phonological theory make possible several competing analyses of the structure of CVVC syllables; we consider these competing representations and their match with the durational properties of segments within CVVC syllable rhymes. We argue that both the phonological and the durational data support an analysis of word-internal CVVC syllables in Arabic as bimoraic, with the final consonant sharing the second mora of the long vowel. 1.1 Syllable Weight and Syllable Structures The opposition between heavy and light syllables has long been recognized as a crucial factor in numerous domains of the grammar. (See, for example, Hyman 1985, McCarthy & Prince 1986, and Hayes 1989). The opposition is generally encoded in mora count, where a short vowel is associated to a single mora, as in (la), and a long vowel to two moras, as in (2a). In many languages, a consonant in the coda of a syllable is also moraic (as in 2b), rendering CVV and CVC *This work was supported by grant number SBR-9310058 to the first author from the National Science Foundation.

120

BROSELOW, HUFFMAN, CHEN, & HSIEH

syllables equivalent with respect to processes that respect syllable weight: (1) Light syllable:

(2) Heavy syllables:

In a language in which coda consonants are moraic, CVVC syllables should be trimoraic. However, the status of trimoraic syllables is controversial, for several reasons. First, while it is common to find monomoraic (CV) syllables patterning differently from bimoraic (CVV and CVC) syllables, evidence for a three-way distinction in syllable weight is relatively scarce. Furthermore, CVVC syllables are frequently restructured to bimoraic CVC syllables, or are limited to the edges of constituents where their final consonant can be analyzed as extraprosodic. (For arguments that word-final Arabic consonants are extraprosodic, see McCarthy & Prince 1990.) Both phenomena are illustrated by Cairene Arabic, in which CVVC is permitted word-finally but is shortened to CVC word-internally, as in kitaab "book" but kitabna "our book". Facts such as these have led various researchers to posit a universal constraint prohibiting trimoraic syllables. If this constraint is absolute (McCarthy & Prince 1986; Steriade 1991), then all CVVC syllables must actually be bimoraic; possible analyses of non-peripheral CVVC syllables are then as in (3a) or (3b), where the final consonant does not occupy a mora of its own. (3) a.

THE TIMING STRUCTURE OF CVVC SYLLABLES

121

But if the constraint against trimoraic syllables may be violated under certain conditions (i.e., where CVVC syllables provide the best means of satisfying more powerful constraints, as Optimality Theory predicts (Prince & Smolensky 1993; McCarthy & Prince 1993)), then CVVC syllables may in fact be trimoraic, as in (4): (4)

In the absence of a universal, inviolable constraint against trimoraic syllables, phonological theory provides (3a), (3b), and (4) as competing candidates for the representation of CVVC syllables. In the following section we examine phonological arguments for considering phrase-internal CVVC syllables 1 in Arabic to be bimoraic, rather than trimoraic, offering an account of cross-dialectal variation in Arabic syllable structures in terms of Optimality Theory. Then in section 3, we present phonetic evidence supporting the bimoraic analysis of CVVC syllables in Arabic. We argue that the durations of vowels and consonants in CVC and CVVC syllables produced by the Syrian and Lebanese speakers are consistent with the hypothesis that the vowel and the coda consonant in these syllables share a mora. In contrast, duration patterns for the Egyptian dialects are consistent with a shortening analysis, in which the vowel and the coda consonant are associated with separate moras. In section 4, we contrast Arabic durational patterns with those in Hindi, a language with a three-way distinction in syllable weight. We argue that Hindi CVVC syllables, in contrast to their counterparts in Arabic, show timing properties consistent with a trimoraic analysis. 2. The Phonology of CVVC Syllables 2.1 CVVC/CVCC Asymmetry One major argument for assigning a bimoraic structure to CVVC syllables in Arabic involves the asymmetry between these syllables and CVCC syllables, which can be accounted for by assigning different 1 ln this paper we restrict our attention to phrase-internal (actually, word-internal) syllables, assuming that different factors operate in phrase-final position.

122

BROSELOW, HUFFMAN, CHEN, & HSIEH

moraic structures to the two syllable types. As noted by Broselow (1992) and Farwaneh (1992), a survey of Arabic dialects reveals that CVVC syllables are much more likely to surface phrase-internally than are CVCC syllables. CVVC syllables may be restructured by vowel shortening, as in the dialects of Lower Egypt (Broselow 1992), or by epenthesis of a vowel, as in Makkan Arabic (Abu-Mansour 1992). But in a majority of dialects, word-internal CVVC syllables are preserved intact:2 (5) CVVC Syllables a. Shortening Cairene, Alexandrian kitaab+na -> kitabna b. Epenthesis Makkan kitaab+na -> kitaabana c. VVC preserved Sudanese, Iraqi, Syrian, Lebanese, Gulf, Upper Egypt 3 k(i)taab+na -> k(i)taabna

CVCC syllables, on the other hand, are much more likely to be restructured through insertion of a vowel (where A represents the inserted vowel, while V represents a vowel that is present in this position in some dialects and not others; where present, it varies in quality across dialects): (6) CVCC Syllables a. Epenthesis Cairene, Sudanese, Makkan katabt+na -> katabtAna Iraqi, Abu Dhabi kVtabt+na -> kVtabAtna Others: epenthesis or preservation of CVCC, depending on quality of CC

2

The reader is referred to Broselow (1992) and Farwaneh (1992) for references on particular dialects. 3In some dialects, "book" is monosyllabic, as the parenthesization of the i indicates.

THE TIMING STRUCTURE OF CVVC SYLLABLES

123

The following chart, summarizing the surface occurrence of CVVC and CVCC syllables, makes clear that within a given dialect, CVCC syllables are more susceptible to restructuring than are CVVC syllables: (7) Surface Word-internal CVVC, CVCC Cairene Makkan Sudanese Abu Dhabi Levantine other Gulf

CVVC no no yes yes yes yes

CVCC no no no no only some CC clusters only some CC clusters

Farwaneh (1992) argues that in those dialects that permit CVCC syllables, the coda clusters that are permitted are generally those in which the first consonant is significantly more sonorous than the second. Since the situation is fairly complex, we will not for the purposes of this paper attempt to provide an account of the constraints permitting some CVCC syllables to surface in some dialects. Thus far, we have found no dialect that allows CVCC but not CVVC. Broselow (1992) proposes to account for the asymmetry between CVVC and CVCC syllables as an effect of a prohibition on trimoraic syllables. CVVC syllables escape this prohibition because they are actually bimoraic, with the structure in (3a), in which the consonant shares a mora with the preceding vowel. CVCC syllables, on the other hand, are generally trimoraic (with the possible exception of those CVCC syllables permitted in particular dialects). We now turn our attention to how this analysis can be implemented within the framework of Optimality Theory, in which languages may differ in terms of their rankings of different output constraints. 2.2 A constraint-based analysis Within the framework of Optimality Theory, surface configurations are accounted for by assuming that all possible candidates for pronouncing a given input form are evaluated with respect to a set of surface constraints. The form that best satisfies these constraints is the preferred form; where there is no form that satisfies all constraints, the form that satisfies the stronger (that is,

124

BROSELOW, HUFFMAN, CHEN, & HSIEH

higher ranking) constraints wins. We propose to account for the asymmetric distribution of CVVC and CVCC syllables by assuming that all the dialects have two high-ranking constraints: (i) NO TRIMORAIC SYLLABLE, which forbids syllables dominating three moras; and (ii) PARSE SEGMENT, which forbids deletion of a vowel or consonant as a means of enforcing conformity to constraints on prosodic structure. Thus, the only acceptable structure for CVVC and CVCC syllables is a bimoraic one in which all segments are incorporated into prosodic structure and in which all coda consonants are dominated by a mora. We therefore need a set of constraints that will generate the appropriate bimoraic structures for each dialect. The input form k(i)taab+ha will have two moras associated with the vowel of its second syllable, as shown in (8): (8) input:

Assuming that trimoraic syllables and deletion of segments are ruled out, three possible surface analyses of this input form are as in (9): (9) a.

ki-taab-ha

b. ki-tab-ha

c. ki-taa-ba-ha

(9a) corresponds to (3a), given earlier, and is consistent with the pronunciation found in Syrian, Lebanese, Sudanese and the majority of other dialects. (9b) involves a failure to incorporate the second mora of the long vowel into higher prosodie structure; the failure to parse this mora (indicated by angled brackets) will result in pronunciation of the vowel as short, as in Cairene and Alexandrian dialects. In (9c), a vowel has been inserted to provide a separate syllable for the /b/, resulting in the pronunciation found in Makkan Arabic. Before discussing the different rankings of constraints that will produce these three different surface configurations, we must deal with a fourth logical possibility, the structure in (3b), in which the /b/ is assigned to the coda but is not dominated by a mora. The surface form corresponding to this structure would be essentially the

THE TIMING STRUCTURE OF CVVC SYLLABLES

125

same as that produced by the structure in (9a=3a). We have no phonological arguments to choose between (9a=3a) and (3b) at this point, and since the phonetic facts discussed in section 3 appear to favor (9a), we will assume that all the dialects also require that all coda consonants be moraic (via a constraint entitled No Appendix in Sherer (1994)). We are now prepared to account for the dialectal variation in the treatment of word-internal CVVC sequences: Cairene and Alexandrian kitaab+ha -> kitabha, Makkan kitaab+ha -> kitaabaha, and all other dialects k(i)taab+ha -> k(i)taabha. We assume three additional constraints: a constraint against mora sharing No Branching Mora (= NBM); and the familiar constraints Parse Mora, which requires moras to be incorporated into higher prosodic structure, and Fill, which forbids insertion of new material into the string. The three patterns illustrated in (9) will result from different rankings of these constraints, where we adopt the precept of Optimality Theory that the form that satisfies the higher ranking constraint is the winning candidate. Ranking of NBM below Parse Mora and Fill will give the surface form in (9a), with vowel and consonant sharing a mora; the Syrian, Lebanese, and many other dialect forms satisfy the higher ranking constraints while violating the weaker NBM. Ranking of Parse Mora below Fill and NBM gives (9b), the Cairene/Alexandrian form, in which the least offending form leaves the second vowel mora out of the structure, resulting in a short vowel. And ranking of Fill below Parse Mora and NBM will give the Makkan form, (9c), in which a vowel is inserted (violating FILL) but the moraic structure of the long vowel remains intact (satisfying Parse Mora and NBM). To assure that CVCC syllables cannot be assigned a bimoraic structure, we must rule out the assignment of a bimoraic structure to CVCC sequences; these generally undergo vowel epenthesis, in violation of Fill. Possible bimoraic analyses of these syllables are those shown in (10a) and (10b):

(10)

126

BROSELOW, HUFFMAN, CHEN, & HSIEH

(10a) can be ruled out by assuming a highly ranked constraint specifying that the head mora of a syllable—that is, the first mora— may dominate no more than one segment (NO BRANCHING HEADMORA=NBHM). This constraint would permit the mora sharing illustrated in (9a), where the nonhead mora dominates two segments, but not that in (10a). (10b) can be ruled out by a constraint forbidding mora sharing unless the segments sharing the mora maintain a particular sonority profile (with the required sonority distance varying across the dialects).4 Optimality Theory therefore makes possible an account of the distribution of CVVC and CVCC syllables across dialects. This account supposes a bimoraic analysis for CVVC syllables. We will now turn to the phonetics of these syllables. 3. The Phonetics of CVVC Syllables: Duration data In this section we consider the match between the competing representations of Arabic CVVC syllable rhymes and the durations of segments in the syllable rhyme. One would expect the different structures (3a) and (4) to be associated with different timing patterns. Specifically, if the long vowel shares a mora with the following consonant, as in (3a), that vowel should be shorter than a long vowel followed by a heteromoraic consonant, but longer than a short vowel in a CVC syllable (2b). Similarly, a final consonant sharing a mora with a vowel, as in (3a), should be shorter than one with exclusive occupation of a mora, as in (2b).5 In contrast, we would not expect to find such a close relationship between vowel and consonant duration if the vowels and consonants have independent moraic structure, as in 4

This analysis raises an interesting problem: the various constraints against branching moras cannot be combined, since they sometimes must be ranked differently with respect to other constraints. Thus, in dialects like Syrian, Lebanese, and Sudanese, which generally insert a vowel into CVCC sequences but maintain CVVC sequences intact, NBHM and NO CC MORA must be ranked above FILL to allow epenthesis in CVCC sequences, while the general constraint against branching moras (NBM) must be ranked below FILL to prevent epenthesis in CVVC. 5 Note that the prediction of consonant shortening is crucial to the argument. Shortening of a long vowel in a closed syllable would not of itself argue for a difference in moraic association, since Closed Syllable Vowel Shortening is a common phonetic process (Maddieson 1985) that appears to operate independently of mora count and moraic association.

THE TIMING STRUCTURE OF CVVC SYLLABLES

127

(2b) and (4). We will show data consistent with analyzing Syrian and Lebanese word-internal CVVC syllables as having the branching bimoraic structure in (3a): vowel and coda consonant durations are shorter than those in comparable CVV and CVC syllables, respectively, though vowels remain significantly longer than underlying short vowels. In contrast, underlying word-internal CVVC syllables in Cairene and Alexandrian were clearly restructured, resulting in the non-branching bimoraic structure in (3b). 3.1 Methods Speakers of the four Arabic dialects—Alexandrian, Cairene, Syrian, and Lebanese—were recorded reading a list of words and sentences containing examples of CVVC syllables in word internal and word final position, plus CVC, CVV, and CV syllables in the same positions, for comparison. The lists were written in English, to avoid the possible influence of Arabic spelling on the speaker's pronunciation. Before recording, the speaker and experimenters reviewed the list together, to ensure that the speaker understood which Arabic word was intended by each English word on the list. To ensure that the speakers used colloquial Arabic, they were asked to translate into Arabic as they would speak with their friends. Subjects were carefully monitored for use of Modern Standard Arabic forms and were asked to redo any form they produced in Modern Standard Arabic as they would say it in casual conversation with a close friend. An example excerpt from the recording materials is given in Appendix I. Each speaker read through the full list three times, pausing for a brief break after each reading. In preparation for durational analysis, the tape recordings were then digitized on a PC using Kay Elemetrics' Computer Speech Lab (CSL) speech analysis system. Durations were measured from wide-band spectrograms generated with the CSL system. Segmentation of the acoustic signal was made following common practice, in which the boundary between vowels and voiced consonants is taken to be the point at which formant amplitude changes dramatically—decreasing going from V to C and increasing going from C to V. For stop+vowel sequences, a stop burst, if visible, was taken to be the end of the stop. For the syllables containing voiceless aspirated stops (e.g., Ill of kitaabi), the stop burst

128

BROSELOW, HUFFMAN, CHEN, & HSIEH

was considered the end of the consonant, and the aspiration was measured as a separate element. For such cases, the onset of voicing for the vowel was then taken as the beginning of the vowel. In most cases, segmentation according to these principles was straightforward. Defining the boundary between two adjacent consonants (specifically, /bn/ of kitaabna and Pinabna) was sometimes more difficult, however, either because a /b/ release was not clearly visible, or because nasalization for the /n/ seemed to overlap the articulatory closure for the /b/. However, usually either /b/ release or onset of nasal formant structure for /n/ was visible and could be used as the criterion for segmentation. 3.2 Analysis Since only one speaker of each dialect was recorded, duration measures were analyzed separately for each speaker. As illustrated in Appendix I, each of the words analyzed occurred in three different contexts, all of which involve phrase-final position. The segment durations for these three contexts were very similar, with no observable effects of sentence type, so for each speaker the nine duration measures (three repetitions x three sentence contexts) for each segment were averaged. The vowel and consonant averages are presented in the graphs in Figures 1-4. In these and all following graphs, the vertical "I" bars superimposed indicate the range of values one standard deviation above and below the mean duration. Figures 1 and 2 show the results for the shortening dialects (Alexandrian and Cairene). We see that both speakers show the duration patterns predicted by the analysis in (9b). On the shortening analysis, the /a/ in kitaabna should be shorter than the /a/ in kitaabi (the phonologically long vowel) and similar in duration to the /a/ of Pinabna (the phonologically short vowel). For the Cairene speaker (Figure 2), the /a/ in kitaabna is 78.1 msec, about 40 msec shorter than the /a/ in kitaabi, and very close in duration to the /a/ in Pinabna (71.4 msec). For the Alexandrian speaker (Figure 1), the /a/ of kitaabna is very short (55.6 msec) less than half the duration of the long vowel in kitaabi, but also shorter than the phonologically short vowel in Pinabna. The consonant durations also pattern as predicted by the analysis. On the assumption that in both kitaabna and Pinabna the /b/ is associated with one mora, consonant durations for the two words

THE TIMING STRUCTURE OF CVVC SYLLABLES

129

should be similar. For the Cairene speaker, the /b/ of kitaabna is 85.5 msec, versus 85.9 msec for /b/ of Pinabna; and for the Alexandrian speaker, the /b/ of kitaabna is 72.6 msec, compared with 83.7 msec for Pinabna. To summarize so far, the duration patterns for the speakers of Egyptian Arabic are consistent with the shortening analysis proposed in (3b). Though the two speakers show differences in the absolute durations of the segments involved, the patterns of relative segment durations for each speaker behave as predicted. The only surprising result was the fact that, for the Cairene speaker, the /a/ in kitaabna was in fact shorter than the phonologically short vowel in Pinabna. This fact does not contradict the shortening analysis, but it does suggest that additional prosodic and phonetic factors not discussed here are also contributing to patterns in segment durations. Because we have only one speaker of each dialect, we can not say at the moment whether the difference in /b/ durations between our two Egyptian speakers represents a general difference between Cairene and Alexandrian or is due to idiosyncrasies of one or both speakers. Figures 3 and 4 show the results for the CVVC-preserving dialects (Syrian and Lebanese). We see that both speakers show the duration patterns predicted by our analysis. Assuming that the CVVC syllable is bimoraic, without phonological vowel shortening, we proposed that the second mora associated with the syllable would be shared by the /a/ and the /b/, as in (9a). This predicts that the vowel of kitaabna will be shorter than a long vowel in an open syllable and longer than a short vowel in an open syllable. This is exactly what happens. For the Lebanese speaker, the vowel of kitaabna (97.8 msec) is shorter than the vowel of kitaabi (115.4 msec) and longer than the vowel of Pinabna (81.2 msec). For the Syrian speaker, the vowel of kitaabna is 112.2 msec, while the vowel of kitaabi is 123.9 msec, and the vowel of Pinabna is 65 msec. Our analysis also predicts that the coda consonant /b/ of kitaabna will be shorter than a coda consonant that has its own mora (Pinabna). Again, the duration data are consistent with our analysis. For the Lebanese speaker the /b/ of kitaabna is 66.7 msec, while the /b/ of Pinabna is 81.1 msec. For the Syrian speaker, the /b/ of kitaabna is 80.8 msec, while the /b/ of Pinabna is 114.4 msec. Two-tailed t-tests were done to evaluate the significance of the duration comparisons discussed above. All of the differences in vowel

130

BROSELOW, HUFFMAN, CHEN, & HSIEH

duration which we have attributed to differences in moraic association (e.g. /a(a)/ of kitaabna vs. Pinabna in Lebanese and Syrian) were significant, while vowel durations found to be similar, attributed to segments having the same moraic representation, were not found to be significantly different (vowel results significant to at least the .001 level). In addition, the very short /a/ of kitaab+na for the Alexandrian speaker was significantly shorter than the phonologically short vowel of ?inab+na. Though we do not yet have a definite explanation of this additional patterning, the important thing to note at this point is that it is consistent with a shortening analysis. For the consonants, the differences in consonant duration attributed to a difference between single-mora association vs. mora sharing (the coda /b/s in Lebanese and Syrian) were significant, and the durations of consonants associated with single moras (coda /b/s in Cairene and Alexandrian) were not significantly different (consonant results significant to at least the .01 level). While the duration patterns of CVVC syllables in Lebanese and Syrian are consistent with our bimoraic mora-sharing analysis, the possibility remains that these syllables could be trimoraic, and the seemingly shorter durations of the vowel and the coda consonant are simply the natural consequence of accommodating more moras in one syllable. To rule out this possibility, we turn to another language, Hindi, for purposes of comparison. 4. Hindi Hindi is one of several languages that have been reported to show a three-way contrast in syllable weight. Gupta (1987) and Pandey (1989) both argue that stress assignment in Hindi is sensitive to (at least) three degrees of syllable weight: light, heavy, and superheavy. According to Gupta (1987), stress is assigned to the heaviest syllable in a word (when equal, stress the leftmost). (11) Hindi a. ántarim b. áandoolan c. cimpáanjii

"internal" "movement" "chimpanzee"

So in example (11a), stress is assigned to the leftmost heavy syllable /an/, instead of the light syllable /ta/ or the heavy syllable

THE TIMING STRUCTURE OF CVVC SYLLABLES

131

/rim/ to its right. Similarly, in (11b), the superheavy syllable /aan/ receives the stress, not the heavy syllable /doo/. Furthermore, (11c) shows the superheavy CVVC syllable /paan/ stressed, rather than the heavy syllables /cim/ or /jii/. Assuming that syllable weight is represented by mora count in the phonological structure, this threeway contrast in Hindi would require a trimoraic representation for superheavy CVVC syllables, as in (4), where the long vowel is associated with two moras, and the coda consonant with a separate mora. To determine the duration patterns of these trimoraic CVVC syllables in Hindi, we recorded a Hindi speaker reading a list of words containing CVVC syllables in both word-internal and word-final positions, and, for comparison, CV, CVV, and CVC syllables in the same positions. Appendix II gives the Hindi words with their translations. Note that, for reasons mentioned earlier, the actual list used for the recording was written in English. The Hindi speaker read through the full list three times, and the tape recordings were then digitized and analyzed, using the same method described earlier. The results of segment durations for some of the words can be found in Figures 5 and 6. Figure 5 shows the durations of the vowels and following consonants in word-internal position for four different types of syllables: CV, CVV, CVC, and CVVC. The long vowel /aa/ in the CVVC syllable (192.3 msec) is slightly more than twice as long as the short vowel /a/ in the CVC syllable (87.5 msec). The same durational contrast between long vowel /aa/ and short vowel /a/ can be observed in the open syllables (179.4 msec for /aa/ vs. 69.2 msec for /a/). These duration patterns are what would be expected if the long vowels have two moras and the short vowels have only one. They are also consistent with our trimoraic analysis for CVVC syllables in Hindi, where the long vowels are bimoraic. Similarly, the consonant durations are consistent with the analysis of CVVC syllables as trimoraic rather than bimoraic with sharing of a consonant and vowel mora. In Syrian and Lebanese, the consonant following a tautosyllabic long vowel was significantly shorter than a consonant following a tautosyllabic short vowel. In Hindi, coda consonants show no shortening after long vowels; in fact, the HI following the long vowel is actually longer than the /l/ following the short vowel: 122.9 msec

132

BROSELOW, HUFFMAN, CHEN, & HSIEH

for the /l/ in daaldaa vs. 58.5 msec for the DJ in kalcii. The differences between the durations of the /l/s in daaldaa and kalcii may be due in part to segmental differences between the word-medial clusters in the two items. In homorganic sonorant + voiced stop clusters in other languages, it is common for the sonorant to be quite long and the stop to be quite short. Thus, the HI in daaldaa may be especially long because it precedes a /d/. When we compare CVC/CVVC pairs in which the coda sonorant is followed by voiceless segments in both words, we find much closer consonant durations. Figure 6 shows the segment duration contrast for the words parcai and paartii. For vowel duration, just as we expect, the bimoraic /aa/ in /paar/ (186.4 msec) is a little more than twice as long as the monomoraic /a/ in /par/ (82.7 ms). The durations of the coda consonants are also as expected: /r/ in /paar/ (66.6 ms) is similar to, and certainly no shorter than /r/ in /par/ (59.1 ms). Thus the segment durations of both vowels and coda consonants in Hindi CVVC syllables are consistent with the trimoraic representation shown in (4), where the long vowel has two moras, the coda consonant has one, and no moras are shared. These duration patterns of CVVC syllables in Hindi, as shown in Figures 5 and 6, are clearly different from those in Syrian and Lebanese (see Figures 3 and 4). If Hindi CVVC syllables show the duration patterns of true trimoraic syllables, then a different moraic representation is appropriate for CVVC syllables in Lebanese and Syrian. Given the durational patterns in these languages, the bimoraic mora-sharing representation in (9a) is the obvious candidate. 5. Conclusion We have argued that a bimoraic analysis of Arabic CVVC syllables is justified both on phonological and phonetic grounds. A bimoraic analysis provides an account of the asymmetry between CVVC and CVCC in these dialects. And the phonetic facts of vowel and consonant duration are consistent with such an analysis: long vowels tend to be shorter in closed than in open syllables, and coda consonants tend to be shorter after long vowels than after short vowels; this is expected if long vowels and coda consonants share a mora. This analysis was confirmed by comparison with the durational facts of Hindi CVVC syllables. In Hindi, where there are independent phonological reasons for assuming a trimoraic analysis of CVVC

THE TIMING STRUCTURE OF CVVC SYLLABLES

133

syllables, we found no shortening of either long vowels or coda consonants in these syllables. The Arabic and Hindi durational patterns therefore provide a reassuring match between the representations motivated by purely phonological considerations and the phonetic details of the utterance.

134

BROSELOW, HUFFMAN, CHEN, & HSIEH Appendix I EXAMPLE OF MATERIALS USED IN RECORDING ARABIC

Subjects translated English phrases into their native dialect of Arabic, putting the words for book and grape in the position marked by underlining: (a) my our their And then I said "(a) And then I said "my And then I said "our And then I said "their And then I saw (a) And then I saw my And then I saw our And then I saw their

" " " "

Appendix II EXAMPLE OF HINDI WORDLIST Hindi 1. dál 2. dáal 3. kálaa 4. káalaa 5. kálcii 6. dáaldaa 7. adáalat 8. váardaat 9. párcai 10. páartii 11. kámal 12. káamal 13. baláatkaar 14. camátkaar

gloss "group" "pulses" "art" "black" "a ladle" "vegetable shortening" "court (of law)" "event" "shadow" "party" "lotus" "wonders" "rape" "miracle"

THE TIMING STRUCTURE OF CVVC SYLLABLES

Fig. 1

135

Alexandrian

Vowels Consonants

Fig. 2

Cairene

Vowels Consonants

136

BROSELOW, HUFFMAN, CHEN, & HSIEH

Fig. 3

Syrian

Vowels Consonants

k(i)taabi k(i)taabna

Fig. 4

pinabi

pinabna

Lebanese

Vowels S

Consonants

k(i)taabi k(i)taabna

pinabi

pinabna

THE TIMING STRUCTURE OF CVVC SYLLABLES

137

Fig 5. V/C Duration Contrast in Hindi

V duration /l/ duration

kálaa

káalaa

kálcii

dáaldaa

Fig 6. CVC /CVVC Contrast in Hindi

V duration /r/ duration

pár-cai

páar-tii

138

BROSELOW, HUFFMAN, CHEN, & HSIEH

REFERENCES Abu-Mansour, Mahasen. 1992. "Vowel Shortening in Two Arabic Dialects". Broselow, Eid & McCarthy 1992, 47-75. Broselow, Ellen. 1992. "Parametric Variation in Arabic Dialect Phonology". Broselow, Eid & McCarthy 1992, 7-45. Broselow, Ellen, Mushira Eid & John McCarthy, eds. 1992. Perspectives on Arabic Linguistics IV. (= Current Issues in Linguistic Theory, 85.) Amsterdam & Philadelphia: John Benjamins. Farwaneh, Samira. 1992. "Directional Syllabification and Syllable Structure in Arabic Dialects". Paper presented at 66th Annual Meeting of the Linguistic Society of America. Gupta, Abha. 1987. "Hindi Word Stress and the Obligatory-Branching Parameter". Proceedings of the Chicago Linguistic Society 23, 134-148. Chicago: Chicago Linguistic Society. Hayes, Bruce. 1989. "Compensatory Lengthening in Moraic Phonology". Linguistic Inquiry 20.253-306. Hyman, Larry. 1985. A Theory of Phonological Weight.. Dordrecht: Foris. Maddieson, Ian. 1985. "Phonetic Cues to Syllabification". Phonetic Linguistics ed. by Victoria Fromkin, 203-221. New York: Academic Press. McCarthy, John & Alan Prince. 1986. Prosodie Morphology. Ms., University of Massachusetts at Amherst and Brandeis University. . 1990. "Prosodic Morphology and Templatic Morphology". Perspectives on Arabic Linguistics II ed. by Mushira Eid & John McCarthy, 1-54. (= Current Issues in Linguistic Theory, 72.) Amsterdam & Philadelphia: John Benjamins. . 1993. Prosodic Morphology I: Constraint Interaction and Satisfaction. Ms., University of Massachusetts at Amherst and Rutgers University. Pandey, Promod. 1989. "Word Accentuation in Hindi". Lingua 77.37-73. Prince, Alan & Paul Smolensky. 1993. Optimality Theory. Ms., Rutgers University. Sherer, Timothy. 1994. Prosodie Phonotaeties. Ph.D. dissertation, University of Massachusetts, Amherst. Steriade, Donca. 1991. "Moras and Other Slots". Proceedings of the Formal Linguistics Conference of the Midwest 1.254-280.

III

OTHER PERSPECTIVES

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE AHMED FAKHRI West Virginia University

1. Background and Purpose The purpose of this study is to investigate the distribution of topics in Arabic narrative discourse and identify morphosyntactic correlates of topic maintenance and shifts, using Givón's approach of topic continuity in discourse (Givón 1983a). The notion of topic has received a great deal of attention in modern research. In the tradition of Prague School linguistics (Vachek 1966), the sentence is divided into two components, the topic or theme, which is what the sentence is about, and the comment or rheme, which refers to what is said about the theme. The topic is thus viewed as a discrete constituent of the sentence. This topic-comment description of sentences seems to be adequate in the case of topic-prominent languages which have special surface coding devices for topics (Li & Thompson 1976). However, in subject-prominent languages such as English, the notion of topic is problematic since it is not clear what criteria are used to identify this entity in a sentence which includes more than one referent. Given the definition of topic above, in a sentence like (1), (1)

Bill sent a book to Mary.

there exist three potential candidates for topic: Bill, Mary, and book. The question then is whether (1) is about Bill, Mary or the book. Problems such as this one have motivated the shift from the narrow view of topic as an intrasentential entity to the consideration of the discourse-functional dimension of 'topic'. Givón (1983a) proposes the notion of degree of topicality according to which the status of a

142

AHMED FAKHRI

referent as topic is determined by the extent to which that referent is continuous in a stretch of discourse. In the following text (2), for instance, the referent BOOK is the topic since this is a referent that is continuously mentioned in the various sentences. In other words, this passage is more likely to be interpreted as being about a book rather than Bill or Mary. (2) Bill sent a book to Mary. It deals with the Civil War. It contains maps and pictures of historical figures. It was published in the U.S. in 1990 and costs thirty dollars.

Thus the continuity of a referent determines the importance of its contribution to the identification of what a particular text is about, in other words, the topic of discourse. Givón's topic continuity approach to discourse has several advantages. First, the continuity of referents plays an important role in discourse since it provides a major means for achieving textual coherence, which is the essence of connected discourse. A sequence of sentences would constitute a text only if it is coherent. Second, the application of this approach has yielded significant insights into the interface between the discourse properties of texts and their localized morphosyntactic features. For example, topic continuity has been used to explain the behavior of passive and ergative constructions in Chamorro (Coorman 1983), to determine the properties of ergative NPs in Hindi-Urdu (Kachru 1987), and to identify constraints on word order variation in biblical Hebrew, Ute, and Mandarin Chinese (Fox 1983; Givón 1983a; Sun & Givón 1985, respectively). Third, topic continuity seems to have psychological correlates, and thus plays a major role in discourse processing (Fletcher 1984; Kertoy 1991). Continuous topics are presumably easy to process, since they are immediately accessible to the reader/hearer, and consequently require minimal linguistic marking such as agreement and clitics. Discontinuous and less predictable topics, on the other hand, must be brought to the consciousness of the reader/hearer, as it were, by means of heavy encoding devices such as full noun phrases with various types of modification. Finally, from the point of view of methodology, topic continuity is amenable to quantitative analysis,

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

143

which makes it more rigorous than other approaches to discourse analysis. 2. Data and Measurements The data for this study consist of a short story by the Iraqi writer Jihaad Abduljabbar Al-Kabisi entitled al-Hažalah "The Hopscotch." It is about a group of children who were playing hopscotch when another group came along, gradually forced themselves into the game, and eventually took over the hopscotch squares that the first group had traced, and expelled the latter from the game. The story is roughly 1300 words long, excluding the dialogues (see discussion below). In addition to this narrative text, a sample of expository discourse of approximately the same length was used for comparison.1 It consists of two passages which dealt with the advantages of science and technology and the status of women in society. The respective authors of these passages are Abdelaziz Amin and Allai Al-Fassi.2 Two measurements of topic continuity used in Givón (1983a) were applied to the data: referential distance and persistence of referents in discourse. Referential distance measures the gap in number of clauses between the previous mention of a referent and its current occurrence in a clause. The minimal value for referential distance is one clause when the referent in the current clause has also been mentioned in the immediately preceding clause. Since the maximal value can theoretically be infinite, an arbitrary cut-off point of twenty clauses was used (cf. Givón 1983a), under the assumption that referents which have not been mentioned for twenty clauses or more are equally unaccessible to the reader. As has been shown in other studies of topic continuity (Givón 1983a), referential distance determines the linguistic means of encoding referents in discourse. In the case of Arabic, the major means of encoding referents are grammatical agreement and clitics, independent pronouns, and full 1 Expository discourse is a type of discourse which is subject-matter oriented and displays logical rather than chronological linkage (Longacre 1976:200; Hinds 1979:136). 2 The narrative and the expository texts were published in the following collection of readings for secondary students in Morocco: al-mutalapatu wa lnusuusu lissanati rraabipatial pipdaadiyyati. Dar al-našr al maghribiyya, 1990. Al Kabisi's text is on pages 117-123, Al Fassi's on pages 93-96, and Amin's on pages 130-132.

144

AHMED FAKHRI

noun phrases. As suggested in Givón (1983b) such devices can be assumed to form a hierarchy on the following topic continuity scale: Most continuous topic grammatical agreement and clitics independent pronouns full noun phrases Least continuous topic

The assumption, then, is that in Arabic the most continuous referent will be encoded through grammatical agreement and clitics, the least continuous through full noun phrases, with independent pronouns encoding moderate continuity. In this study, the referential distance measurement was applied only to human referents and two major props in the story, the hopscotch squares and the stone that the children toss onto the squares, since these are the most frequent referents in the story. It should also be noted that dialogues in the story were excluded from the analysis because they fall outside the line of the narrative proper. The second measurement, the persistence of referents, assesses the extent to which a particular referent once introduced into the discourse is continuously mentioned thereafter in adjacent clauses without interruption. This measurement thus reflects the importance of the referent in the discourse, and is presumed to enhance the identifiability of discourse topics. Referents which have high persistence contribute more to the determination of what a particular text is about. 3. Results 3.1 Referential distance The results in Table 1 clearly show that grammatical agreement and clitics encode maximally continuous referents with an average value of 1.23 clauses between the current mention of a referent and its previous one.

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

145

Table 1. Correlation between Referential Distance and Topic Encoding Devices Encoding Devices Grammatical Agreement & Clitics Independent Pronouns Full Noun Phrases

Average Referential Distance 1.23 1.12 10.40

The average referential distance for agreement and clitics indicates that there are instances where agreement was used even though the referential distance exceeded one clause. Many of these instances can be accounted for through the absence of potential interference (Givon 1983b). The use of agreement or clitics in this case without any risk of ambiguity is allowed because there is no possibility of interference from other referents in the neighboring discourse. Consider the following example from the story. The double slashes indicate clause boundaries. (3)

xaraža lhažaru pafwan Sani lmustatiili// hiina left the stone accidentally form the rectangle when šarapa pahaduhum billapibi// faqafaza waahidun began one of them with the play jumped one mina lsabiyati// yatanaawaluhu.3 of the children pick up it. "The stone left the rectangle accidentally when one of them started to play, and one the children jumped to pick it up."

The clitic -hu "it" in yatanaawaluhu refers to lhažaru three clauses back. In spite of a referential distance value of 3, there is no ambiguity of reference because the other referents (e.g., the rectangle or one of the children) in this stretch of discourse are semantically implausible as objects of the verb yatanaawalu. The least continuous referents which are interrupted by a significantly larger number of clauses (X =10.40) are reintroduced through heavier linguistic encoding, i.e., as full noun phrases. Independent pronouns do not fit in the scale of encoding continuity since, like grammatical agreement and clitics, they are shown to mark 3Transcription does not show assimilation of the definite article. It is irrelevant to the data presented and will only serve to confuse readers.

146

AHMED FAKHRI

maximally continuous referents (X =1.12). The reason for this is that in Arabic independent pronouns do not serve a purely anaphoric function. They are used either for contrast as in (4) or are required in particular grammatical structures such as circumstantial clauses introduced by waaw lhaal as in (5), regardless of the discourse environment. In both (4) and (5), the independent pronoun under consideration is hum "they." (4)

(5)

laakinnahum surpaana maa yataðakkaruuna Panna but-they soon they realize that lmalpaba ma lpabuhum hum. the field their field they. "But they realized soon that the playing field was THEIRS." tatapaalaa dahakaatuhum wa hum yalpabuun. rise their laughter while they play. "Their laughter became louder as they played."

3.2 Persistence As mentioned above, the persistence measurement assesses the importance of the contribution of a referent to the identification of the discourse topic. For the purpose of this study, I computed the frequency of clause sequences with high persistence referents. A high persistence referent is operationally defined here as any referent which persists in three successive clauses or more without interruption. This measurement was applied to the narrative sample and the expository prose sample. The results are presented in Table 2. Table 2.

Text Type Narrative Expository

The Frequency of Clause Sequences with High Persistence Referents Persistence Values in Number of Clauses 3 Cl. 8 5

4 Cl. 6 3

5 C1.+ 9 4

Total 23 12

These results indicate that the narrative text includes almost twice as many clause sequences with high persistence referents as the expository sample (23 versus 12). Thus, it seems that the continuity of

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

147

referents is a major device for encoding discourse topics in the narrative. In other words, the narrative is about those referents that are continuously mentioned. The expository texts do not rely to the same degree on this device for signalling discourse topics. A follow-up analysis of the data suggests that in the expository texts, discourse topics are signalled through the promotion of subjects to preverbal position, in other words SV word order, frequently accompanied by the particle Pinna. In this analysis, the frequency of SV and VS word orders were computed for both types of texts. Word orders that are dictated by the syntactic properties of the language are excluded from the computation, since they involve no choice, so to speak. For example, the SV word order in some circumstantial clauses introduced by waaw lhaal (6) and the VS word order that is required after the complementizer Pan (7) are excluded. (6)

(7)

γaadara lbayta wa lmataru yasqutu. the rain fall (3MS). left (3MS) the house while "He left the house while the rain was falling." paržuu Pan yaxruža ltaalibu. want (1S) that leave (3MS) the student. "I want the student to leave."

Only instances involving choice, as illustrated in (8) and (9), are included in the computation. Although these sentences are syntactically different, they are semantically equivalent. (8) (9)

tdahwara lwadpu lsiyaasiyu kaθiiran deteriorated the situation political a lot (Pinna) lwadpu(/a) lsiyaasiyu(/a) tadahwara kaØiiran PARTICLE the situation political deteriorated a lot "The political situation has deteriorated a lot."

VS SV

Table 3 gives the frequency of the VS and SV word orders in both text types.

148

AHMED FAKHRI Table 3. The Frequency of VS and SV Orders Text Type Narrative Expository

Word Order VS SV 46 (96%) 2 (4%) 9 (45%) 11(55%)

The results in Table 3 show that in narrative discourse the SV word order is marked and accounts for only 4% of word order types, whereas in expository prose both the VS and SV word orders are used. Although the percentages are not the same, this distribution of SV and VS is to some extent consistent with data presented in Parkinson (1981). In that study, short stories consistently exhibit a predominance of VSO, whereas in some non-narrative texts, the SVO order is used more frequently than the VSO one (Parkinson 1981:3032). Nine of the eleven instances of SV word order in the expository sample are accompanied by the particle Pinna. We will examine the implication of this result for the status of Pinna in the last section of the paper. One of the two tokens of SV word order in the narrative begins a paragraph in which the author suspends the events of the story in order to explain the rules of the hopscotch game. The paragraph includes features of logical type of organization such as conditionals, which characterize expository texts (Longacre 1976). This constitutes further evidence that the SV word order is more likely to be found in expository prose. In brief, then, these findings indicate that there exist in Arabic two devices for encoding discourse topics: (a) (b)

VS word order plus high persistence of referents. SV word order frequently accompanied by Pinna plus moderate persistence of referents,

and that these devices are constrained by discourse type since (a) is typical of narrative discourse whereas (b) is typical of expository texts.

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

149

4. Morphosyntactic Correlates of Topic Continuity In the following discussion, the discourse notion of topic continuity will be shown to motivate morphosyntactic properties of Arabic. As Givón observes, the correlation between discourse function and its grammatical coding is never perfect but the search for aspects of their association is motivated by the "fundamental assumption that there exists a systematic correlation between message and code." (Givón 1983b:15). The morphosyntactic properties of Arabic that are investigated here are: the neutralization of agreement in VS structures, the deletion of waaw Ihaal, a connector which introduces circumstantial clauses, and the particle Pinna. 4.1 The neutralization of agreement Subject-predicate agreement is a complex phenomenon in Arabic and no claim is made that all its details can be accounted for in terms of discourse function. What is suggested here is that the pattern of agreement reduction exhibited in the contrast between the VS and SV word orders (see the discussion of examples 10-12 below) can be viewed as a reflection of the discourse function of agreement. As can be gathered from the previous discussion (section 3.1), the discourse function of agreement is a derivative of topic continuity, which requires that only a subset of semantic features that characterize nouns such as plurality and gender be used to encode continuous referents. The discourse function of this subset, which constitutes agreement, is to single out a particular referent among a potential group previously mentioned (Givón 1976). In Arabic, the grammatical marking of subject-verb agreement is richer in SV structures than in their VS counterparts, as shown in the following examples (10)-(12). 4 Sentences (11) and (12) indicate neutralization with respect to number agreement: the verb is singular regardless of whether the post-verbal subject is plural as in (11) or singular as in (12). 4Excluded from this discussion are sentences in which the NP subject is (or includes) a plural independent pronoun as in: žaa?uu hum wa pawladuhum came (3MP) they and children-their "They came with their children." In this case, the verb is always plural, regardless of the position of the subject NP.

150

AHMED FAKHRI

(10)

(11)

(12)

al-?awlaadu xaražuu the boys (MP) went out (3MP) "The boys went out." xaraža al-?awlaadu went out (3MS) the boys (MP) "The boys went out." xaraža al-waladu went out (3MS) the boy (MS) "The boy went out."

Since the referent al-Pawlaadu "the boys" is overtly encoded in (11), and thus immediately accessible in its entirety, the referent identification function of agreement is obviated, resulting in the reduction of agreement features. The encoding of a referent as an independent constituent is a necessary but not sufficient condition for agreement reduction as shown in (10) where full agreement is also redundant. However, a crucial difference between (10) and (11) is that the use of plural agreement in (10) is consistent with the anaphoric or back reference function of agreement in discourse; marking plural agreement in (11) would have violated this discourse function of agreement. In other words, given the discourse function of agreement described above, the agreement reduction patterns in (13) and (14) are impossible. (13) (14)

*xaražuu went out (3MP) *al-?awlaadu the boys (MP)

al-?awlaadu the boys (MP) xaraža went out (MS)

In (13) the plural marking on the verb is not only functionally redundant, but is also at odds with the anaphoric use of agreement. The point made here is that the discourse function of agreement can account partially for the direction of agreement reduction in Arabic: the agreement patterns in (10) and (11) are preferred because they are more congruent with the discourse properties of agreement than the patterns in (13) and (14) are.5 5Arabic dialects may exhibit subject-verb agreement as in (13). This is the case, for example, in Moroccan Arabic as sentence (a) shows:

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

151

4.2 Topic continuity and waaw lhaal In Arabic, circumstantial clauses are connected to main clauses by the use of waaw lhaal, the form wa in (15). (15)

γaadara lbayta wa lmataru left(3MS) the house while the rain "He left the house while it was raining."

yasqutu. fall (3MS).

However, there are instances where the circumstantial clause is simply appended to the main clause without this connector. The data in (16)(19) show that deletion of waaw lhaal is constrained by topic continuity. (16)

žaa?a samiirun yartapišu came (3MS) Samir shaking (3MS) "Samir came shaking." (17) žaa?a samiirun yadaahu tartapišaani came (3MS) Samir his two hands shaking (3FD) "Samir came, his hands shaking." (18)? žaa?a samiirun qamiisuhu yurafrifu came (3MS) Samir shirt-his waving (3MS) "Samir came, his shirt waving." (19)* žaa?a samiirun pumarun yartapišu came (3MS) Samir Omar shaking (3MS) "Samir came while Omar was shaking."

The grammaticality of these sentences correlates with the continuity of the referent in the main clause.6 In (16), (17), and (18), went out (3 MP) the boys (MP) However, in this case, it should be impossible to have subject-verb agreement reduction when the subject occupies pre-verbal position as in (14). Indeed, in Moroccan Arabic, the SV order also shows subject-verb agreement as in (b): (b) draari xeržu The boys (MP) went out (3 MP) The Moroccan Arabic data simply shows that subject-verb agreement is totally grammaticalized. The contention still holds that no Arabic variety should exhibit full agreement in the VS word order (as in 13) and agreement reduction in the SV order (as in 14). I thank Dr. Eid for raising this issue of comparative data from Arabic dialects. 6Note that the following version of (16) with waaw lhaal in it would include an independent pronoun in the circumstantial clause:

152

AHMED FAKHRI

the referent Samir is mentioned in the main clause as a full NP and in the circumstantial clause through agreement or the clitic -hu "his." On the other hand, in the ungrammatical sentence (19) this referent is not mentioned in the circumstantial clause. The slight difference in grammaticality between (17) and (18) (based on my own judgment) may be due to the fact that yadaahu "his hands" expresses inalienable possession whereas qamiisuhu "his shirt" expresses alienable possession. In short, the continuity of referents serves as a linking device between the main clause and the circumstantial clause appended to it, obviating the need for an overt connecting particle such as waaw lhaal. 4.3 Topic continuity and the particle Pinna As has been shown earlier, discourse topics in expository texts are signalled by the use of SV word order frequently accompanied by the particle Pinna. Traditionally, Pinna has been considered as a sentence modifier whose function is to emphasize the content of the whole sentence it introduces (Moutaouakil 1984:131). However, from the point of view of topic continuity, Pinna exhibits distinct discourse properties. First, it introduces discontinuous referents encoded as full NPs which are marked as definite, a feature typical of topics (Li 1976). In Arabic this can be accomplished through the use of the definite article or idaafa "genitive". All the nine tokens of Pinna mentioned above (see the discussion of the results of Table 3) introduce definite full NPs. Second, its use is often associated with important discourse boundaries, typically the beginning of paragraphs where topic shifts are likely to occur. In fact, five of the nine tokens (a)

žaa?a samiirun wa huwa yartapišu came (3MS) Samir while he shake (3MS) "Samir came shaking." The deletion of waawu lhaal alone would yield the following ungrammatical result: (b) *žaa?a samiirun huwa yartapišu However, the ungrammaticality of (b) is independent of the deletion of waaw lhaal and is due to the more general constraint that independent pronouns in Arabic cannot be used for a purely anaphoric purpose. In other words, (b) is disallowed for the same reason as (c), which does not include a circumstantial clause: (c) *daxala samiirun Gumma huwa žalasa Sala lkursiyyi enter (3MS) Samir then sit (3MS) on the chair "Samir came in and then sat on the chair."

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

153

of Pinna occur in paragraph-initial position. These facts suggest that Pinna serves as a topic marking device in connected discourse. Given the above analysis, it is not difficult to discern the relationship between the function of Pinna as a topic marker and its more grammaticalized use. Complement clauses introduced by Pinna require the promotion of subjects to preverbal position as in (20). (20) a.

b.

xaraža lwaladu went out (3MS) the boy "The boy went out." qaala Pinna lwalada xaraža said(3MS) that the boy went out (3MS) "He said that the boy went out."

Furthermore, in some instances, the particle Pinna may introduce a VS clause, but in this case a non-referential pleonastic pronoun is cliticized to it, as shown in the following examples adapted from Fassi Fehri (1990:17). (21)* Pinna zaaratni ØalaaØu šaapiraatin that visited (3FS)-me three poets (F) "Three female poets visited me." (22) pinnahaa zaaratni ØalaaØu šaapiraatin that-her visited (3FS)-me three poets (F) "three female poets visited me."

The fact that grammaticalized Pinna requires a (pro) nominal form immediately after it can be construed as a reflection of its discourse function as a topic marking device. 5. Conclusion The present study suggests the usefulness of discourse notions such as topic continuity in explaining morphosyntactic properties of Arabic. While the distribution of the various linguistic devices for encoding referents in discourse (i.e., full NPs, independent pronouns, and agreement) was for the most part predictable, the other findings of the study provide important insights into the interface between discourse phenomena and more localized morphosyntactic features. The analysis of topic continuity has enabled us to identify constraints on word order variation in Arabic discourse. It has been

154

AHMED FAKHRI

shown that the VS word order is predominant in narrative discourse, whereas the SV word order is typical of expository discourse. Topic continuity has also been used to provide a unified account for the properties of Pinna both at the discourse and syntactic levels, for the patterns of subject-verb agreement reduction, and for the deletion of waawu lhaal. These findings are encouraging, given the limited size of the data considered. Further research using a larger and more varied sample of texts is needed to corroborate and fine-tune these findings, as well as explore other aspects of the relationship between discourse function and morphosyntactic properties of the Arabic language.

REFERENCES Coorman, Ann. 1983. "Topic Continuity and the Voicing System of an Ergative Language: Chamorro". Givón 1983a. 425-489. Fassi Fehri, Abdelkader. 1990. "On Pleonastics in Arabic". La Linguistique au Maghreb ed. by Jochen Pleines, 15-36. Rabat: Editions Okad. Fletcher, Charles. 1984. "Markedness and Topic Continuity in Discourse Processing". Journal of Verbal Learning and Verbal Behavior 23.487-493. Fox, Andrew. 1983. "Topic Continuity in Biblical Hebrew Narrative". Givón 1983a. 215-254. Givón, Talmy. 1976. "Topic, Pronoun, and Grammatical Agreement". Li 1976. 149-188. , ed. 1983a. Topic Continuity in Discourse: A quantitative cross-language study. Amsterdam: John Benjamins. . 1983b. "Topic Continuity in Discourse: An introduction". Givón 1983a. 141. Hinds, John. 1979. "Organizational Patterns in Discourse". Syntax and Semantics 12: Discourse and syntax ed. by Talmy Givón, 135-158. New York: Academic Press. Kachru, Yamuna. 1987. "Ergativity, Subjecthood and Topicality in Hindi-Urdu". Lingua 17.223-238. Kertoy, Marilyn. 1991. "Listening Comprehension for Sentences: The accessibility of referents for pronouns as a function of age, topic continuity, and pronoun emphasis". Journal of Experimental child Psychology 52.344-353. Li, Charles N., ed. 1976. Subject and Topic. New York: Academic Press. & Sandra A. Thompson. 1976. "Subject and Topic: A new typology of language". Li 1976. 457-490. Longacre, Robert. 1976. An Anatomy of Speech Notions. Lisse: Peter de Ridder Press. Moutaouakil, Ahmed. 1984. "Le Focus en Arabe: Vers une analyse fonctionnelle". Lingua 64:2/3.115-176.

TOPIC CONTINUITY IN ARABIC NARRATIVE DISCOURSE

155

Parkinson, Dilworth B. 1981. "VSO to SVO in Modern Standard Arabic: A study in diglossia syntax". Al-cArabiyya 14.24-37. Sun, Chao-Fen & Talmy Givón. 1985. "On the So-Called SOV Word Order in Mandarin Chinese: A quantified text study and its implications". Language 62.329-351. Vachek, Josef. 1966. The Linguistic School of Prague. Bloomington: Indiana University Press.

ON VOWEL SHORTENING IN PALESTINIAN ARABIC MUNTHER A. YOUNES Cornell University

1.0 Introduction This paper discusses two types of word-internal vowel shortening in Arabic. The two types are known as closed syllable shortening (CSS) and open syllable shortening (OSS). It will be shown that whereas CSS is a straightforward phonological process that operates within a syllable to block the creation of an impermissible syllable type, OSS operates within a two-syllable sequence to maximize occurrences of the iambic foot, a dominant prosodic constituent in Arabic. CSS will be discussed briefly with examples from three Arabic dialects: Modern Standard Arabic (MSA), Cairene Arabic, and Palestinian Arabic, but the focus of the paper will be on OSS in a dialect of Palestinian. 1.1 Vowel shortening in Arabic Consider the following examples from MSA, Cairene, and Palestinian Arabic: (1)

MSA Cairene Palestinian kaan kaan kaan "he was" kunti1 kunti kunti "you (FS) were" pistašaar pistašaar pistašaar "he consulted" Pistašarti Pistašarti Pistašarti "you (FS) consulted"

1The shortened vowel of hollow verbs of Form I (the basic, nonderived form) in the perfect, which is typically aa, is raised to either i or u in certain persons, depending on the individual verb.

158

MUNTHER A. YOUNES

A rule of vowel shortening is clearly at play, as shown in the forms above. This rule is well-documented in Arabic and its operation is straightforward: the long vowel is shortened when a consonantinitial suffix is added to a hollow verb. In some modern Arabic dialects, such as those of Egypt and Syria, it is also operative before the negative suffix š and the preposition l "to, for". This is shown in the following examples from Egyptian: (2)

šaaf gaab ruuh

"he saw" "he brought" "go!"

mašafš gabli ruhlu

"he did not see" "he brought for me" "go to him!"

Two of the better-known detailed studies of CSS in Arabic are Abu-Salim (1982) and Abu-Mansour (1992). Abu-Salim's study focuses on a dialect of Palestinian Arabic, while Abu-Mansour's compares the operation of the rule in Makkan and Egyptian. The two studies reach the same general conclusion, namely, that the rule operates to eliminate CVVCs word-internally.2 Now consider the following set of forms: (3) MSA Cairene Palestinian baapu baapu baapu baapuuha bapuuha bapuuha mabapuhaaš mabapuhaaš/mabaapuhaaš baapatha bapitha baapatha suuratayn surteen suurteen

"they sold" "they sold it (FS) "they did not sell it (FS) "she sold it (FS) "two pictures"

These forms show a different process of vowel shortening. Comparing baapu to bapuuha in Egyptian, for example, shows that a long vowel in an open syllable is shortened, and comparing the two forms to mabapuhaaš shows that two such vowels in two adjacent syllables are shortened. The forms also show that the three Arabic dialects differ in the operation of the rule: there is no shortening in the case of MSA and there are more instances of shortening in Egyptian than in Palestinian, as shown by forms like marametihaaš and marameetihaaš, respectively. A careful examination of the 2

This type of vowel shortening operates in a similar way to the English rule that shortens vowels in such pairs of related forms as: deep/depth, broad!'breadth, etc. (Myers 1987).

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

159

Egyptian forms in (3) and other relevant examples shows that the operation of the rule in that dialect is in fact straightforward: a long vowel is shortened when it loses main stress due to the addition of stress-attracting suffixes. Other examples are shown in (4). (' indicates stress.) (4) beet "house" masy'uul "busy (MS) š'aafit "she saw"

beťeen (or biteen) "two houses" mašyul'iin "busy (P)" šafithum "she saw them"

The operation of the rule in Palestinian is, however, not as straightforward as it is in Egyptian. Consequently, the focus of this paper will be on OSS in a dialect of Palestinian. 1.2 Rural Palestinian Arabic The dialect examined in the paper, which is that of the author, is spoken in the rural areas in the north of the West Bank. It will be referred to here as Rural Palestinian Arabic (RPA). A number of observations regarding its syllable structure and related rules are in order before proceeding with a detailed account of OSS. A. Syllable types: Like the majority of Arabic dialects, RPA permits five syllable types. Three basic types, CV, CVC, and CVV, are not restricted in their distribution, and two secondary types, CVVC and CVCC, are generally restricted to word-final position. Examples of the five types are shown in (5). ($ indicates a syllable boundary.) (5) katab maktab maktuub mafiall

CV$CVC CVC$CVC CVC$CVVC CV$CVCC

"he wrote" "office" "written, letter" "place, store"

CV syllables have traditionally been referred to as light, CVC and CVV as heavy, and CVCC and CVVC as superheavy. It should be noted in this context that every syllable in Arabic obligatorily starts with a consonant, so in a VCCV sequence the syllable boundary lies between the two Cs.

160

MUNTHER A. YOUNES

B. Stress: The stress rules of RPA can be summarized as follows: a. Stress a final superheavy syllable if there is one, (6) jaw'aab "answer

mafi'all "place, store"

b. Otherwise, stress the penultimate syllable if it is heavy or if it is the first syllable in the word. (7) makťabna "our office" k'atab "he wrote"

makťuubak "your letter"

c. Otherwise, stress the antepenultimate syllable. (8) k'atabu "they wrote"

C. Syncope: A short, high vowel is deleted in open unstressed syllables: (9) fihim fihmat

"he understood" +at "she" → "she understood"

D. Epenthesis: A short, high vowel is inserted to break up sequences of three or four consonants utterance-intemally and sequences of two consonants, other than geminates, utterance-finally: (10) dar[i]s "lesson" dar[i]skum "your (MP) lesson" dars [i] lparabi "the Arabic lesson' ' but, (11) darsak

"your (MS) lesson"

2. Discussion 2.1 Open Syllable Shortening The most comprehensive treatment of OSS in Palestinian Arabic is found in Abu-Salim (1982). After refuting previous segmental accounts of stress, syncope, epenthesis, and vowel shortening in Palestinian Arabic, he argues for a metrical analysis of these

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

161

phenomena. According to his analysis, "the Vowel-Shortening rule [OSS] is sensitive to the relative strength of segments in metrical trees, and . . . only vowels dominated by weak nodes [in a given word] will undergo shortening" [122]. This is shown in (12) (Abu-Salim 1982:123). (12)

"happy (MP)"

"they did not see her"

The vowel of the second syllable in mabsuutiin as well as the second and third syllables of mašaafuuhaaš, are shortened because they are dominated by weak nodes. Apparent exceptions like the word suurt'een "two pictures", where the rule does not apply to the long vowel uu, are accounted for by assigning uu a strong node in the stress tree: (13)

Words like ïaalameen "two worlds" and kimbyuutareen "two computers" (Abu-Salim 1982:119), which do not undergo OSS, are treated the same way.

162

MUNTHER A. YOUNES

One class of exceptions to Abu-Salim's metrical analysis is represented by words with four open syllables in which the first one has a long vowel, such as the word haarabato "she fought him" [130]. He accounts for these exceptions by stating that "words consisting of four syllables, the first three of which are light CV, are not permissible in PA [Palestinian Arabic] as well as in other modern Arabic dialects" [131]. Consequently, the vowel in the first syllable of such words remains long. One weakness of Abu-Salim's analysis is in his treatment of forms like suurteen mentioned above. With no explanation given, the tree used to represent the word is drawn in such a way as to make the node dominating the first syllable strong. Another way of drawing the tree, which would be more consistent with the way he draws the other trees, such as those shown above for mabsutiin and mašafuhaaš, is as follows: (14)

According to the labeling of this tree, uu of suurteen would have to undergo OSS. A careful examination of all the forms used in Abu-Salim's discussion of OSS shows that the rule, as formulated, operates consistently in one environment only, specifically, when a long, unstressed vowel occurs immediately to the left of a stressed syllable. The following is a list of all the forms in Abu-Salim's study to which OSS applies: (15) baab'een → bab'een pajiir'een → ajir'een muftaah'een → muftah'een bantaloon'een → bantalon'een pariib'iin→parib'iin

"two doors" "two employees" "two keys" "two pairs of pants" "near (MP)"

ON VOWEL SHORTENING IN PALESTINIAN ARABIC Pariib'aat → parib'aat mabsuutlin → mabsut'iin mabsuuťaat → mabsuťaat makaaťibna → makatibna makaaťibkum → makat'ibkum saah'ibhum → safi'ibhum saah'ibkum → safi'ibkum madaar'ishum → madar'ishum šaaf 'uuha → šaf 'uuha masafuuh'aas → mašafuh'aas

163

"near (FP)" "happy (MP)" "happy (FP)" "our offices" "your offices" "their friend" "your friend" "their schools" "they saw her" "they did not see her"

And the following is a list of the forms in which it does not apply: (16)

suurťeen (/suurateen/) jaarťeen (/jaarateen/) Paalb'een (/Paalibeen/) taabp'een (/taabrpeen/) saatr'iin (/šaatiriin/) saatr'aat (/šaatiraat/) paalam'een moolad'een kumbyuutar'een haar'abato

two pictures "two female neighbors" "two molds" "two stamps" "smart (MP)" "smart (FP)" "two worlds" "two birthdays" "two computers" "she fought him"

The rule could be reformulated to make it operative only when a long, unstressed vowel is followed immediately by a stressed one. But Abu-Salim rules out such a possibility when he states that the metrical approach is superior to previous segmental approaches in that it "is no longer sensitive to the adjacent environment" [122]. The rest of the paper will be devoted primarily to showing that Abu-Salim's data as well as additional data from RPA clearly demonstrate the importance of the adjacent environment in the operation of OSS, that a two-syllable sequence is the domain of the application of the rule, and that it operates to maximize iambic feet in the language. I will also show that a distinction needs to be made between the two types of heavy syllables (CVC and CVV), since one type triggers the rule while the other does not.

164

MUNTHER A. YOUNES

2.2 Open syllable shortening in Rural Palestinian Arabic For the most part, OSS operates similarly in the dialect examined by Abu-Salim and in RPA, so the rule applies to the following forms in RPA (compare with (15) above): (17) baab'een pPajiir'een muftaah'een bantaloon'een Pariib'iin Pariib'aat mabsuuťiin mabsuuťaat

"two doors" "two employees" "two keys" "two pairs of pants" "near (MP)" "near (FP)" "happy (MP)" "happy (FP)"

and it does not apply to any of the forms shown in (16) or any of the following forms shown in (15), repeated in (18) for convenience. (18) makaat'ibna makaaťibkum saah'ibhum saah'ibkum madaar'ishum

"our offices" "your offices" "their friend" "your friend" "their schools"

Consequently, OSS in RPA can be stated as follows: a long vowel in an open syllable is shortened if the vowel in the syllable immediately to its right is long and has primary stress. (19) VV-A7

$CW

Evidence in support of this formulation of the rule can be found in uninflected words, in which a constraint exists against forms with two long vowels in adjacent syllables. A comparison of forms from RPA and their equivalents in MSA (which does not have such a constraint) illustrations this: RPA masammiir jamuus jiraan

MSA masaamiir jaamuus jiiraan

"nails" "water buffalo' "neighbors"

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

165

OSS takes place regardless of whether the following long vowel is in an open or a closed syllable, as shown in (21): (21) sayyaara "car" sayyaraatak "your (MS) cars" sayyaraathum "their (M) cars"

2.3 OSS before the nisba suffix Vowel shortening also takes place before the feminine form of the so-called "relational adjective" or nisba suffix, as the following examples show: šaami šamiyyi nuuri nuriyyi piid pidiyyi paadi padiyyi

"from Damascus (MS)" "from Damascus (FS)" "gypsy (MS)" "gypsy (FS)" "the Eid, feast" "for the Eid" "normal (MS)" "normal (FS)"

The application of OSS in this environment would be inconsistent with the formulation of the rule given above if the syllable following the target vowel is considered to have a CVC structure. However, there is evidence in favor of treating the iyy of the nisba as phonologically equivalent to iiy. For example, the language does not show a phonological contrast between the two, which are phonetically identical (John McCarthy, p.c.). Consequently, vowel shortening before the nisba suffix could be considered part of the general rule of shortening before long vowels, or before V [-Consonantal] sequences. (23) VV→ V/_$C'V [-consonantal]

2.4 OSS before CVCCC's Long vowels in open syllables are also shortened before syllables of the form CVCCC. These 'super' superheavy syllables are produced by attaching the negative suffix š to a verb followed by an object pronoun of the form CVCiCi. There are four object pronouns with a CVCiCi structure in RPA: humm "they (M)", hinn "they (F)", kumm

166

MUNTHER A. YOUNES

"you (MP)", činn "(FP)". The final geminate is degeminated wordfinally.3 (24) ma not

šaaf saw

humm them(M)

mašaafhummiš

š not "He did not see them (F)"

The i preceding negative š is epenthetic, as evidence from other areas of RPA phonology shows. For example, the stress rule, which is known to apply before epenthesis treats combinations like Cï+negative š as CC clusters and assigns primary stress to the vowel immediately to their left: /makaťabš/ "he did not write", which surfaces as [makaťabiš]. (See the stress rules given in section 1.) Examples illustrating the application of OSS before CVCCC syllables4 are given in (25). (25) ramuuhin maramuhinniš šuftuuhum mašuftuhummiš

"they threw them (F)" "they did not throw them (F)" "you (MP) saw them (M)" "you (MP) did not see them (M)"

Hence, the OSS rule can be modified to accommodate forms like mašaafhummiš as follows: (26)

3

VV→V/_$

{C'V[consonantal]} CVCCC

Alternatively, one may argue for a rule that geminates a word-final consonant when the negative suffix is added. This, however, is not supported if other data are considered: šaafak "he saw you (MS)" /mašaajakiš "he did not see you (MS)", but not *mašaafakkiš. Furthermore, a look at the corresponding 'independent' pronouns confirms the de-gemination rule, at least in part. These pronouns are hummi, hinni, Pintu, Pintin. The two which correspond phonologically to the object pronouns show geminates in the same position. 4Itis possible to argue that OSS has been generalized to apply before all object suffixes followed by negative š. Four of the ten object suffixes show a long vowel in this position: huuš "him", haaš "her", m/s "me", naaš "me". Shortening also takes place before another two of the ten suffixes as a result of the CSS rule: mašafuukš→mašafuk(i)š "they did not see you (MS), mašafučš→mašaffuuč(i)š "they did not see you (FS)". It is interesting to note that in some Palestinian dialects other than the one examined here kiš and čiš are pronounced kiišand čiiš, with long vowels.

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

167

2.5 Exceptions to OSS: The case of haamiihalharamiiha The rule of vowel shortening as discussed above is not without exceptions. These exceptions include cases in which a long vowel is not shortened before another long vowel. They are clearly limited both phonologically and morphologically. Consider, for example, the following two pairs of forms: (27) h'aami har'aami

"protector" "thief

haam'iiha haramliha

"its protector" "its thieves"

In the first pair, the long vowel aa is not shortened while in the second it is. Upon further examination, it turns out that this resistance to shortening is found in two main categories of words: the first includes lame verbs in Form III and their active participles, and the second includes the active participles of lame verbs in Form I. Examples are given in (28). (28) paada(a) (Form III verb, root: pdw) paadaani mapaadaniiš mpaadiini (active participle) šaari(i) (active participle, root: šry) šaariiha

"he considered an enemy" "he considered me an enemy" "he did not consider me an enemy" "considering me an enemy" "buying" "buying it (F)"

An explanation for the absence of shortening in the above forms may be sought on functional grounds. If OSS were to apply to these limited morphological categories then they would lose their distinctive identities. It is the long vowel in the first syllable of these forms that gives them these identities. Note also that the vowels which resist OSS are exactly those of the same morphological categories which are considered 'derived' by McCarthy & Prince (1990b:29). He can, therefore, claim that they constitute a marked class of forms to which certain regular phonological processes do not apply. 2.6 Domain of OSS: The iambic foot According to Abu-Salim (1982), OSS is a word-level rule which applies to shorten any unstressed long vowel in an open syllable that is dominated by a weak node in the word metrical tree. But it should be clear that in RPA and even in the dialect examined by Abu-Salim the

168

MUNTHER A. YOUNES

domain of application of the rule is restricted to a sequence of two syllables, as is shown by forms like paalam'een "two worlds" and suur(a)teen "two pictures". The first long vowel aa of the first word and uu of the second are not shortened although they are in open syllables and are in weak positions (i.e., not receiving primary stress). An examination of the verb negation paradigm, which offers the longest concatenations of morphemes and syllables in the language, confirms the conclusion that OSS has a limited domain of two syllables. Consider the following forms: (29) rameena ramenaaha marameenahaaš habbeetu habbetuu mahabbeetuhuuš

"we threw" "we threw it (F)" "we did not throw it (F)" "you (MP) liked" "you (MP) liked it (M)" "you (MP) did not like it (M)"

When three long vowels occur in a sequence as in marameenahaaš and mahabbeetuhuuš, the rule applies only once. If it were a word-level rule, then both ee of the antepenultimate and uu of the penultimate syllables would be shortened. A counterexample is to the two-syllable domain is presented by the negative paradigm of hollow verbs such as the paradigm of the verb šaaf shown in (30). (30) šaafu "they saw" mašafuuš "they did not see" mašafuhaaš/mašaafuhaaš "they did not see her"

As the forms mašafuhaaš/mašaafuhaaš show, there is a pattern of free variation when two potential targets for OSS are found in a sequence. The form with no shortening (mašaafuhaas) follows the pattern shown by the lame (ramay) and doubled (habb) verbs shown in (29), while the other free variant does not. This poses a problem for Abu-Salim's analysis as well as the analysis proposed here. One way to solve the problem is to state that in a syllable sequence of the form ABC in which A and B are potential targets for shortening while B and C are possible triggers, both A and B and B and C are scanned as sequences of two syllables each, with A

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

169

shortening before B and B shortening before C. This solution, however, will lose any appeal once the lame and doubled verbs are taken into consideration. I would like to suggest that RPA might be undergoing a change that is affecting different types of verbs differently. If MSA is assumed to represent the conservative end of the spectrum, where no shortening of vowels in open syllables takes place, and Cairene the innovative end, where all vowels preceding a stressed syllable are shortened, then RPA would occupy a place between the two but would be moving in the direction of Cairene. In such a scenario one could predict that the free variant mašafuhaaš, in which two vowels are shortened, will be the dominant one. Shortening may then be generalized to all vowels in open syllables which precede the stressed syllable of a word in the lame and doubled verb paradigms, as is the case in Cairene. But regardless of what may happen in the future, there is overwhelming evidence indicating that the domain of OSS in RPA at this point is the two-syllable sequence. The rule operates to change a CVV$C'VV(C) or a CVV$C'VCCC sequence to a CV$C'VV(C) or a CV$C'VCCC one, and in the case of Abu-Salim's dialect it operates to change CVV$C'VX to CV$C'VX. In all these cases a heavy-heavy or heavy-superheavy sequence is changed to a light-heavy one. McCarthy & Prince (1990a, b) argue on the basis of a wide range of data from Arabic nominal and verbal patterns, for the central role that the iambic pattern plays in Arabic morphology. They demonstrate its dominance of the Arabic lexicon through a comprehensive analysis of broken plural formation. They also point out (McCarthy & Prince 1990b:44) that the CVVCVVC pattern(e.g.,Jaamuus "water buffalo") is not used in any systematic way by Arabic morphology and that it is lexically marked. OSS can be viewed as a phonological rule which operates to eliminate marked phonological sequences and to generate those of the dominant type, i.e., the iambic foot. 2.7 OSS and syllable types A distinction is traditionally made in Arabic phonology between light (CV) and heavy (CVC, CVV) syllables, or monomoraic and bimoraic syllables (McCarthy & Prince 1990b; Broselow 1992). Such a classification of syllable types in Arabic is necessary for the stress

170

MUNTHER A. YOUNES

rules in which members of each category behave in an identical manner. However, as the above discussion shows, a distinction needs to be made between the two types of heavy syllables in RPA, since CVVs trigger OSS but CVCs do not. In fact, the operation of the rule suggests the following classification of syllable types: (31)

Group 1 Group 2

CV, CVC, CVCC CVV, CVVC, CVCCC

The syllables in Group 2 trigger shortening of a long vowel in a preceding syllable, while those in Group 1 do not. Another issue raised by the operation of the OSS rule described above relates to what Broselow (1990) refers to as the Bimoraicity Constraint, which specifies that "syllables are maximally and optimally bimoraic" [10]. Broselow argues that the different phonological processes of syncope, epenthesis and CCS operate in conformity with this principle. The findings of this paper, however, show that, the Bimoraicity Constraint holds true only in the case of CVC syllables, but not CVV syllables, since the latter are reduced by OSS to monomoraic syllables. This provides further support for placing CVCs and CVVCs in two different groups. 3. Conclusion I have argued that the rule of Open Syllable Shortening is triggered by a following syllable with either of the following structures: CV[-consonantal] or CVCCC. The rule generally applies within a bi-syllabic, or foot, domain, so only a long vowel in the syllable immediately preceding the trigger is shortened. The operation of the OSS rule could be viewed as a means of maximizing the iambic pattern in the language. The exceptional behavior of Form III verbs, their active participles, and the active participles of Form I lame verbs is attributed to the need to preserve certain morphological categories. Finally, the operation of the OSS rule as discussed in the paper challenges the traditional classification of syllable types in Arabic into light (CV), heavy (CVC, CVV), and superheavy (CVCC, CVCCC).

ON VOWEL SHORTENING IN PALESTINIAN ARABIC

171

REFERENCES Abu-Mansour, Mahasen. 1992. "Closed Syllable Shortening and Morphological Levels". Perspectives on Arabic Linguistics TV ed. by Ellen Broselow, Mushira Eid & John McCarthy, 47-75. (=Current Issues in Linguistic Theory, 85.) Amsterdam & Philadelphia: John Benjamins. Abu-Salim, Issam. 1982. A Reanalysis of Some Aspects of Arabic Phonology: A metrical approach. Ph.D. dissertation, University of Illinois, Urbana. Broselow, Ellen. 1992. "Parametric Variation in Arabic Dialect Phonology". Perspectives on Arabic Linguistics IV ed. by Ellen Broselow, Mushira Eid & John McCarthy, 7-45. (=Current Issues in Linguistic Theory, 85.) Amsterdam & Philadelphia: John Benjamins. McCarthy, John & Alan Prince. 1990a. "Foot and Word in Prosodic Morphology: The Arabic broken plural". Natural Language and Linguistic Theory 8:209-283. . 1990b. "Prosodic Morphology and Templatic Morphology". Perspectives on Arabic Linguistics II ed. by Mushira Eid & John McCarthy, 1-54. (=Current Issues in Linguistic Theory, 72.) Amsterdam & Philadelphia: John Benjamins. Myers, Scott. 1987. "Vowel Shortening in English". Natural Language and Linguistic Theory 5:485-518.

MORPHEME EDGES AND ARABIC INFIXATION* MICHÆL L.McOMBER University of Utah

0. Introduction Within the context of theories that refer to edges in syntax and phonology, it is a reasonable question to ask whether there are edges in morphology as well. Semitic languages have a long-standing tradition of unusual discontinuity. We explain here that discontinuity results naturally from infixation, and is an effect rather than a cause of particular morpheme edge sequences. We define morpheme edges in §1, showing how they handle infixation in §2. Infixation is concatenatory by means of these edges. We claim in §3 that the affixations will proceed most efficiently if the lexical root, or citation form for Arabic verbs is CCVC1, noting the lexical savings provided such a shape. It allows a reduction in the diacritic system for the six basic ablaut groups. Sample verb derivations for Arabic are given in §§4-8. These illustrate the generality in §9 that we may affix at all five possible points along CCVC. We conclude that morpheme edges are insightful. A shorter version of this paper was also read at the 23rd North American Conference on Afroasiatic Linguistics, March 1995, Salt Lake City, Utah. I am grateful for reviews from Mushira Eid, Ray Freeze, Carol Georgopoulos, Samira Farwaneh, and Mauricio Mixco, on various earlier versions. The usual disclaimers apply. Funding was provided by the Language Science Institute. 1Beyond the three consonants of the traditional CCC root for Semitic, we claim that there is one more lexical segment, a vowel, in position 3. CvCC nouns are broken CCvC roots, derived by a reclustering of the consonants. Beyond the issue of lexical savings mentioned here, this view is based on the predictability of aspectual ablaut (also noted by others) and the high correlation between root consonant distance and homorganicity. See McOmber (1989a) for further discussion, and note 4 below.

174

MORPHEME EDGES

1. Defining Morpheme Edges Every morpheme has a first and last segment. From the definition of a morpheme as a string of phonological material, a string of length n tells us that the n\h segment is the last. We designate this last segment the right edge, and the first segment the left edge. This property is quite economical: we need no new machinery in the grammar to recognize or refer to these edges, but only a slightly more precise notation regarding the plus used in concatenation. In the American tradition of the 1950s (Gleason 1955:63, Francis 1958:174ff) morphemes were notated in braces: {jump}. For us here, these braces mark edges. The left edge, j , is marked by the left brace, {, and the right edge, p, is marked by a right brace, }. This near trivial move allows us keep concatenation for infixation. 2. Infixation with Edges The famous Tagalog example from Gleason (1955b:32) shows an infixation after the first2 consonant of the root: (1) (urn) 'active' (Ramos 1985:80) {sulat} 'write' {s(um)ulat} 'write active'

The containment of the infix (urn) within the root {sulat} is shown by the nesting of the edges: we meet a second left edge '(' before we meet the matching right edge '} ' which is paired with the first morpheme and which closes that first morpheme. These other boundaries are single occurrences of an edge within the string. It is not the case, then, that all internal boundaries are pluses (right-left edge sequences). It follows that specific single edge references represent discontinuities. 2.1 Infixation causes discontinuity Notice also in (1), s(um)ulat, that the infixation process creates an inevitable discontinuity and that the edges mark the boundaries of the discontinuity. The discontinuity is predictable and thus absent from Jensen (1990:76) tells us that the commonest sites for infixation are after the first and before the last segment (or syllable). This place is just inside the morpheme edge (or higher-level prosodic edge). Like this Tagalog morph, Arabic voice infixes after the first segment.

MICHÆL L. McOMBER

175

the lexicon. Since it is predictable, there is no need to abstract it away on a separate C or V tier, with morpheme status of its own. It is rather the ipso facto result of the infixation process itself. Other forms of the Tagalog verb order a reduplication before the infixation. It is ordinarily a moot point whether we suppose that the reduplication, shown in braces in (2), copies to the right or left, before the infixation, which is shown in parentheses: (2) Left Reduplication: {su}sulat Infixation: {s(um)u}sulat

Right Reduplication: su{su}sulat Infixation: s(um)u{su}lat

If we are looking to minimize discontinuity of a root then the leftward version of the rule shows an intact sulat. However, our trade-off is to have pieces of the reduplicated syllable su separated into individual segments. This is the same discontinuity as shown in the root for the rightward version of the reduplication. So, we can place the discontinuity in either morpheme we choose. One version keeps the root simple but complicates the affixes, while the other equally complicates the continuity of the root in order to leave the affixes whole. To us, the leftward version is more natural, but our point is the same (regardless of which analysis the reader prefers) infixation causes discontinuity. Semitic discontinuity has been given such daunting titles as interdigitation (Schramm 1962:497), intercalation (Hammond 1988:248), and transfix interspersion (Bauer 1988:255) as if it represented an original process in its own right. Rather we show it as a predictable consequence and effect (rather than some cause) of the infixation process itself. We submit that there are no discontinuous roots in the lexicon, (nor therefore also any true circumfixes3), in Arabic or any other language for that matter. We need single-edge boundaries to mark discontinuity. The edges simultaneously track the points of origin and extent of the discontinuities. Circumfixes are two separate morphemes with opaque analysis. The {ge} in German past participles, for example, is an absolutive usage handling cases where no other inseparable prefix is available. Bauer's (1988:239) analysis of a ge- -t circumfixed onto mach to get gemacht is unnecessarily complicated since some sort of {t} is required for other past morphology in any event.

176

MORPHEME EDGES

2.2 Abbreviation conventions With the braces as primitives, we take the outermost edges for granted ad libitum: (3) Outside-edge Convention In a morpheme string, the outside edges may optionally be omitted. For example, {jump} {ing} = jump} {ing.

We also formalize the plus as an abbreviation of a } { sequence with the following notation convention: (4) The Plus Convention In a morpheme string, the right-left edge sequence } { may optionally be notated + for convenience. For example, {jump} {ing} = {jump + ing}

These two conventions together show a principled formalized equivalence of {jump} {ing} and jump + ing as in (5): (5) a. b. c.

{jump} {ing} jump} {ing jump + ing

Given (fully notated) Outside-edge Convention Plus Convention

This equivalence takes us back to where we started with jump + ing but with the crucial difference of having the edges formalized and nestable. It has also been a common practice to use a hyphen instead of a plus: jump-ing. We will use the hyphen and plus interchangeably here as well. A hyphen is a separation symbol in other usages anyway, and has the advantage of (when lecturing at the blackboard, for example) not being confused with the letter t! 2.3 Infixation is concatenatory McCarthy (1981:380) discusses the assimilation rule of the eighth binyan (a specific derivational paradigm) of Arabic verbs, noting that it is morpheme-specific; that is, it fails to apply in phonologically similar environments, such as the inflection -ta "you (MS): The upshot of these facts is that, to apply the assimilation rule correctly, the grammar must be able to identify the t-infix of [Form Eight] exclusively. Under a boundary-based theory, though, there is no way to locate an infix as distinct

MICHÆL L. McOMBER

177

from the unit that contains it. Infixes are not delimited by +-boundary—this is an incoherent and ad hoc suggestion that would lead to such absurdities as a morpheme apparently composed solely of the first root consonant, preceding the infix: + w + t + apad. McCarthy (1981:380)

McCarthy interpreted concatenation as consisting of only a plus. With morpheme edges, however, the plus is a particular sequence of two edges. These edges are the boundaries which delimit infixation. With boundaries available, infixes is just as concatenatory as other affixes. In the Arabic case, the reflexive r-infix interrupts waHad thus: w(t)apad. The glide-to-t assimilation then applies: t(t)apad. This edgesensitive notion of boundary likewise contradicts McCarthy's (1981:381) statement that the "rule could not be expressed in a boundary-based theory." As a result, infixation is still available within concatenation, and the morphology remains simpler, because it requires less innovation to represent discontinuities. The crux of the matter, then, is to recognize the edges within the plus as separate entities. In McOmber (1989a:llff, 1989c:13ff) we discuss some examples of direct infixation. Following the root given above, wp?id, here is a Form Eight which alters the meaning from "promise" to "come to terms": (6)

w?id w(a)?id wa?ad w(t)a?ad t(t)a?ad it(t)a?ad

Root lexis "promise" Voice infix: {a} = active Aspect ablaut: i > a Infix for Form Eight Leftward spreading (Regressive Assimilation) #CC repair (Prothesis)

Flop and erasure are not required. The putative template builds up as the derivation proceeds, because all affixes come with their CV structure intact. The infix (a) is active voice, a generalization missed by McCarthy's allomelodies in both the CV and moraic treatments, and {X} after root C2 for intensive. Such Xs are available at every affix position of ktub to accomplish the various derivations. And a single segment more clearly specifies the binyan. To say that Form Five is

178

MORPHEME EDGES

CvCvCCvC obscures the fact that the only difference beyond the ta prefix is the extra C in the middle of the root. We have shown a preview of how a derivation proceeds using morpheme edges. The analysis is also dependent on the CCVC lexical root shape itself. That shape also promotes savings in the lexicon. 3. The Lexical Root In order to set the stage for a demonstration of morpheme edge derivation in Arabic, we present the starting point, the lexical root or citation form we will build from. It is the traditional CCC root, but further augmented by the imperfective vowel in position three. The resulting CCVC4 shape follows Fronzaroli (1965:246ff) and Levy (1971:103) and is further supported by the Root Consonant Distance Correlation (McOmber 1989b) which shows that a CCXC distribution explains 99% of Greenberg's homorganicity constraints on choices for these Cs, whereas the CCC and CVCVC shapes correlate poorly. In the traditional ablaut analysis there are six prominent categories for the ablaut vocalisms. The verb is cited in the perfect active, third person singular, i.e., kataba. The vowel needed for the imperfect is shown diacritically in parentheses off to the side. Below is a chart of the six types, grouped first by the perfective then imperfect vowel. Here, however, we will not show the inflectional suffix, the final -a of kataba. We also include McCarthy's (1989:8) frequency figures based on the Wehr dictionary, followed by a looser diacritic which hints at membership in the category: (7)

Traditional Ablaut Groups (by frequency order based on McCarthy 1989:8) Group Base Frequency Hints a/u katab(u) 1029 — a/i darab(i) 842 — i/a

šarib (a)

518

—

In working from the surface data back to underlying representations, there is no justification for removing that V. Yet it persists throughout the paradigms in that very position. It is not predictable and thus cannot be written out to a rule. Traditionalists show it as a diacritic in the lexicon: kataba (u) instead of the more economical ktub.

179

MICHÆL L. McOMBER Group

Base

a/a u/u i/i

fapal(a) balud(u) hasib (i)

Frequency 436 191 9

Hints Guttural Stative —

The most efficient way to mark these six groups is to make the larger groups the predictable one, and the smaller groups the exceptions. Beyond that, it is helpful to reduce the diacritics as much as possible in order to simplify lexical entries. By inspection, the katab group should be a predictable one. So we take a > u ablaut as normal or standard. That leaves the darab. group as exceptional since it ablauts a > i instead of a > u. The statives are the only group with u in the perfective, so they can be considered predictable also. The šarib group with its i in the perfective is overwhelmingly larger than hasib with only nine members. We now show the chart in (8) with the top three groups as the default cases for a, u and i. The second group contains the less numerous exceptions: (8) Traditional Ablaut Groups (sorted by predictability) Base

Frequency

Hints

Predictable a i u

Group

katab (u) šarib (a) balud(u)

1029 518 191

— — Stative

Unpredictable a/i a/a iA

darab (i) fapal(a) hasib (i)

842 436 9

— Guttural —

In this system, barely half (56%) of the verbs are predictable (shown with bolded frequency values). To optimize predictability, we choose the largest group of each ablaut as the default. If we include the hint

180

MORPHEME EDGES

for guttural verbs, and say that the presence of a guttural consonant predicts the ala vocalism, then we raise predictability5 to -72%. If instead of basing the lexical citation form on the traditional perfect we allow the imperfect CCVC instead, we have only one diacritic to deal with for our six groups,6 verbs which do not ablaut. Of these two groups, ablauting and non-ablauting, the former is more numerous. Now the group vowel is the imperfect: (9) CCVC Groups (ablauting vs. non-ablauting) Group Base Frequency Predictable (Ablauting) a srab 518 u ktub 1029 i drib 842

Hints — — —

Unpredictable (Non-ablauting vowel is underlined as a diacritic) a fpal 436 Guttural u blud 191 Stative i hsib 9 —

Ironically, the imperfect pattern reveals some interesting statistics about Arabic Verbs. First of all, most verbs (≈75%) do in fact ablaut. Secondly, when seen from the imperfect, all vowel groups are represented with and without the (non-ablaut) diacritic. There are no gaps in the distribution: each vowel can participate in ablauting or not. The hinting 7 increases predictability beyond 90%. Both types of hinting are productive toward predictability as well. 5 There is the further complication that fhim "understand" has a guttural but does not obey the lowering. These would have to be pulled out and marked diacritically. Such marking is an expense to the lexicon. There are additional minor groups where verbs take multiple ablauts (Al-Ahmadi AlHarbi 1992:18ff)—reducible to membership in more than one established category. To date we follow Brame (1970: passim) to resolve hollow and lame phonology. 7 The hints are additional data about the category that act as a diacritic in the lexicon. Here, for example, the fact that the verb root contains a guttural "hints" to us that it will be a member of the a/a ablaut group, and thus contributes to predictability (when taken as the default as we do here). Verbs like ƒhim that were exceptions in the perfective-based model are not a problem here because the base form includes the very vowel that tells us this is not a guttural-lowered verb! But the perfective faham does not tell us that: we would have to diacritically mark it as an exception to show it does not behave like a guttural verb.

MICHÆL L. McOMBER

181

We have shown some lexical advantages of the CCVC root. Now we will show some edge-based derivations which are based on it. This is not difficult to do, since there is nearly always at least a vowel between the second and third consonants of all the derived verbs. 4. Edge-based Derivations: Form I There is a universal tendency for abrupt8 imperatives to have little or no morphology on them. Where there is morphology, there is still another tendency that the morpheme marks is the imperative mood itself, rather than person, number, etc. Likewise in Arabic, the CCVC mentioned in the last section is the imperative as well. Phonotactics require a prothetic vowel depending on environment. That will be discussed in §5 below. First we will discuss deriving the imperfect active from the imperative/base form. The imperfect active derives the root ktub into aktub. We prefix the active voice morpheme (a), and then the verb is ready for personal affixes: (10)

Form I — Imperfect Active ktub Root a-ktub Prefix active voice (a)

The aspect itself, imperfect, is left unexpressed just as in the imperative. We have not changed voice or aspect, only mood, from imperative to indicative. When the passive is formed with (u) in as the voice morpheme, *uktub the imperfect aspect is shown in the stem vowel as a height dissimilation from the voice vowel. Since u is high, the stem vowel becomes low. The stem vowel u of ktub changes to low a. Feature changing rules on an already existing segment do not show additional edge boundaries. In the notation a diacritic shows that the segment has been altered. Here we will capitalize such a segment:

An abrupt imperative is one without courtesy or honorific semantics and is often restricted to the most informal registers. It includes the masculine and singular where number and gender is marked elsewhere in the language. Even languages with consistent second person markers will suddenly show no endings at all for this form.

182

MORPHEME EDGES (11)

Form I — Imperfect Passive ktub Root u-ktub Prefix passive voice (u) u-ktAb Height dissimilation of stem vowel for imperfect aspect

Forming the perfect active from the root is a change of aspect but not voice. First, the stem vowel of the imperfect root is ablauted to the perfect, then the active voice morpheme (a) is infixed: ( 12)

Form I — Perfect Active ktub Root ktAb Stem vowel ablaut for perfect aspect k(a)tAb Active voice infixed

The perfect passive shows the (u) morpheme for passive. The perfect passives for all the forms use (u) for passive and / in the stem vowel for perfect. The stem vowel must be front for either aspect. Beyond that, it is assigned to be the opposite height for the imperfect, and the same height for the perfect. The heights were imperfect-ly level for the imperfect aspect, but are perfect-ly level (same height) for the perfect aspect: (13)

Form I — Perfect Passive ktub Root ktAb Stem vowel ablaut for perfect aspect k(u)tAb Passive voice infixed k(u)tlb Height assimilation for perfect aspect

The participles in the derived forms (i.e., higher than Form I) have a (mu) prefix. The active participle for Form I is just an adjectivalization without this (mu). The adjectivalization consists of an (aa) infix after the first consonant. Notice that this is a long version of the active voice morpheme (a). The passive also shows a lengthened voice morpheme. The stem vowel is adjusted to i by analogy from the imperfect active verb of the same form: ( 14)

Form I — Active Participle ktub Root (ad) infix: lengthened active morpheme k(aa)tub kaatlb Stem vowel to i

MICHÆL L. McOMBER

183

The passive participle has the (mu) but with an odd displacement in mirror image to the active kaatib. In the passive, the (u) morpheme is lengthened (as was the active!) but it is now placed in the stem vowel position (see §9). It is also odd that the vowel in the participial prefix (mu) adjusts away from u. ( 15)

Form I — Passive Participle ktub Root kt(uu)b Stem vowel replaced by lengthened passive morpheme mu-kt(uu)b (mu) prefix mA-kt(uu)b Vowel in prefix dissimilated

5. Form II In the derived forms, that is, forms II and higher, the morphemes for voice are passive (u) and active (a). In these forms, the stem vowel is conformed to be front (regardless of its original lexical quality) and then the height is opposite for imperfect, yet the same in the perfect. We end up with all four possibilities of combining the aspect and voice: (16)

Derived Vowel Melodies: Affix Active Passive Imperfect (a) — i (u) — a Perfect (a) — a (u) — i

(voice) and Stem Vowel (aspect) Mnemonic different heights: imperfect-ly level same heights: perfect-ly level

The characteristic of Form II is just the extra segment slot at position three of the root (hereinafter X3) to which it assimilates: (17)

Form II — Imperfect Active ktub Root kt(X)ub Extra segment infix: X3 k(a)t(X)ub Active voice infix k(a)t(X)Ib Imperfect aspect i in stem vowel position k(a)t(t)Ib X assimilation

The addition of personal prefixes such as y for third person singular, promotes epenthesis, because the initial clusters will be repaired: *ykattib. A vowel is added between the morpheme edges: (y)V(kattib). This vowel is filled with u by analogy with the participles, yukattib, but this copied u does not mean anything. That

184

MORPHEME EDGES

is, since this u is phonological, it does not carry morpheme edges of its own and has no semantic value associated with it, i.e., it does not convey grammatical information. This form uses the (u)—a melody but has shifted it over to the left, where the extra u for the personal prefixes is. Notice that the left edge is where an u had been copied from the participle (mu). In the imperfect passive, that u now counts as the passive u and the following a is copied over to the left as well. (18)

Form II — Imperfect Passive ktub kt(X)ub k(u)t(X)ub k(u)t(X)Ab k(u)t(T)Ab

Root Extra segment infix Passive voice infix Imperfect a in stem vowel X assimilation

We would be done, but we will prefix personal inflections that will generate an epenthetic vowel, such as y for third person singular masculine. Now for the perfects: (19)

Form II — Perfect Active ktub kt(X)ub k(a)t(X)ub k(a)t(X)Ab k(a)t(T)Ab

(20)

Root X3 infix Active voice infix Perfect aspect melody in stem vowel X assimilation

Form II — Perfect Passive ktub Root kt(X)ub X3 infix Passive voice infix k(u)t(X)ub Perfect i melody in stem vowel k(u)t(X)Ib X assimilation k(u)t(T)Ib

6. Form III Form III verbs have an X2 infix. The perfect active of Form III ends up with a long vowel after the first consonant. That is because the X2 infix, which assimilates late in the derivation, picks up the voice morpheme:

MICHÆL L. McOMBER

(21)

Form III — Perfect Active ktub Root X2 infix k(X)tub Active voice infix k(a)(X)tub k(a)(X)tAb Perfect a in stem vowel X assimilation k(a)(A)tAb

(22)

Form III — Perfect Passive ktub Root k(X)tub X2 infix Passive voice infix k(u)(X)tub k(u)(X)tIb Perfect i in stem vowel: same height as voice X assimilation k(u)(U)tIb

(23)

Form III — Imperfect Active ktub Root k(X)tub X2 infix k(a)(X)tub Active voice infix Imperfect i in stem: opposite height of voice k(a)(X)tIb X assimilation k(a)(A)tIb

185

Form III — Imperfect Passive ktub Root X2 infix k(X)tub k(u)(X)tub Passive voice infix k(u)(X)tAb Imperfect a in stem: opposite height of voice X assimilation k(u)(U)tAb

As we saw for the imperfects above, the personal inflections promote an epenthesis and a melody shift: (24b)

(C){kuutab) (C)V(kuutab} (C)u(kuutab} Cukaatab

Inflection #CC repair u-analogy Melody shift

7. Form IV Form IV is the last of the X affix types. Form II had the X3 affix, creating an extra segment in the 3rd position, Form III had the X2 affix creating an infix to the 2nd position. Form IV affixes an extra segment as well, but into position one. Let us call this an X1 affix.

186

MORPHEME EDGES

Within the parameters of regressive assimilation, however, there will not be a source consonant to assimilate or spread from. In such a default case, the glottal stop is used, as elsewhere in Arabic. The rest of the morphology proceeds as usual, and a voice morpheme inserts into position two: (25)

Form IV — Perfect Active ktub Root (X)ktub X1 affix for Form IV (X)(a)ktub Active voice infix (X)(a)ktab Perfect a stem vowel: same height as voice (?)(a)ktab X assimilation (defaulted)

(26)

Form IV — Perfect Passive ktub Root X1 affix for Form IV (X)ktub (X)(u)ktub Passive voice infix Perfect i stem vowel: same height as voice (X)(u)ktIb (?)(u)ktIb X assimilation (defaulted)

(27)

Form IV — Imperfect Active ktub Root (X)ktub X1 affix for Form IV Active voice infix (X)(a)ktub Imperfect i stem vowel: opposite height as voice (X)(a)ktIb X assimilation (defaulted) (?)(a)ktIb u(?)(a)ktIb «-analogy (for personal inflection)

(28)

Form IV — Imperfect Passive ktub Root X1 affix for Form IV (X)ktub Active voice infix (X)(u)ktub (X)(u)ktAb Imperfect a stem vowel: opposite height as voice (?)(u)ktAb X assimilation (defaulted) u(?)(u)ktAb «-analogy (for personal inflection) u(?)(A)ktAb Melody shift

There are two common features to Forms II-IV. All three forms use an X affix, and all three borrow a meaningless u from their respective participles for the #CC repair in the imperfect. The higher forms, V+, have neither of these two features. The mismatch of names

MICHÆL L. McOMBER

187

II and III from the point of view of the X affix location, along with the unity of these forms because of their similar behavior, will eventually prompt us to eventually abandon Form numbers except as a trivial (and uninsightful) reference to traditional grammar. 8. Forms V and VI Forms V and VI show a t- prefix on form X3 and X2, respectively. As usual, this consonant creates a #CC violation with the following root consonant, *tktub. This is repaired at the morpheme edge with a vowel insertion. The quality of the vowel is copied from the right in all cases, tVkattab > takattab and tVkuttib > tukuttib. In the case of the imperfect we have the #CC repair once again for the personal inflections. In the active, the quality is again copied from the right. In the passive, the melody shifts the (u) into this position, dragging the a with it in the following position: y(u)takaatab. A more insightful nomenclature would be t-X3 for Form V and t-X2 for Form VI. Quibling over names is immaterial from the point of view of pure theory, since these are just labels. Yet effective labelling becomes helpful in pedagogy when they are mnemonic and analogous to the items they refer to. More importantly, they help reveal a generality about these X affixes which we now explain. 9. The X-Affix System We are seeing a pattern emerge where a single segment, notated X because it is underspecified, is the target of progressive assimilation (alias leftward spreading and perseverative coarticulation). Since there are cases shown of X1, X2, and X3, it is reasonable to look for a generality across the rest of the root. A lexical root shape CCVC predicts insertable Xs at two more locations, X4 with CCVXC and X5 CCVCX. In fact, these both occur. The X4 form, CCVXC should assimilate its X as a vowel and show as CCVVC. That is the root shape in the plain (Form I) passive participle such as (ma)ktu(u)b "having been written". The X5 form, CCVCX should assimilate the final X as a copy of the third consonant. Forms Nine, Eleven, and Fourteen all show this copy. Though all these do not actually occur with the root ktub they would appear as ktabab, ktaabab and ktanbab respectively. MSA thus

188

MORPHEME EDGES

requires a subsequent epenthesis to break up the bb cluster, although dialects such as Egyptian are more lenient on such matters. Using the CCVC lexical root has uncovered the generality that an X-affix is productive across the entire root, and can insert at any and all of the five possible points, as shown in display (29): (29)

All Five Affix Positions across the CCVC Lexical Root 1

K2T3U4B

5

10. Summary The use of morpheme edges builds upon concatenation and clarifies it as an instance of two adjoining morpheme edges, a right edge followed by a left edge. Simplifying infixation, particularly in a Semitic language and undoubtedly others as well, makes intricate morphology all the more tractable. It further supports the notion that discontinuity is a predictable by-product of such a process and can thus be freed from encumbering the lexicon. Our guiding light throughout has been economy and simplicity. The CCVC shape was used as a base form for sample derivations for the more common verb forms. Along the way, we discovered the generality that an X-infix can occur at any position in the lexical root. Since morphemes have edges whether we refer to them or not, it is a small and hence economical innovation to the theory to include them and allow infixation to remain as a type of concatenation.

MICHÆL L. McOMBER

189

REFERENCES Al-Ahmadi Al-Harbi, Awwad Ahmad. 1992. "The Classical Arabic Vowel Plane: Implications for non-linear phonology". Paper presented at the Sixth Annual Symposium on Arabic Linguistics, Ohio State University, Columbus. Bauer, Laurie. 1988. Introducing Linguistic Morphology. Edinburgh: Edinburgh University Press. Brame, Michael K. 1970. "Arabic Phonology: Implications for phonological theory and historical Semitic". MIT Ph.D. dissertation, Cambridge, Mass. Francis, W. Nelson. 1958. The Structure of American English. New York: The Ronald Press Company. Fronzaroli, Pelio. 1965. "Studi sul lessico comune semitico". Rendiconti: Classe di scienze morale e storiche, Academia dei Lincei, XX:246-269. Bollettino del Comitato per la Preparazione del Edizione Nazionale dei Classici Greci e Latini. Rome: Academia dei Lincei. Gleason, Henry A., Jr. 1955. An Introduction of Descriptive Linguistics. New York: Henry Holt & Company. . 1955b. Workbook in Descriptive Linguistics. New York: Holt, Rinehart & Winston. Hammond, Michael. 1988. "Templatic Transfer in Arabic". NLLT 6:247-270. Jensen, John T. 1990. Morphology. (= Current Issues in Linguistic Theory, 70.) Amsterdam: John Benjamins. Levy, Mary M. 1971. "The Plural of the Noun in Modern Standard Arabic". Ph.D. dissertation, University of Michigan, Ann Arbor. McCarthy, John J. 1981. "A Prosodic Theory of Nonconcatenative Morphology". LI 12:373-418. . 1989. "Guttural Phonology". Ms., University of Massachusetts, Amherst. McOmber, Michæl L. 1989a. "The Aprothetic Imperative as Theoretic Base Form in Modern Standard Arabic". Ms, University of Utah, Salt Lake City. . 1989b. "The Inverse Correlation of the Homorganic Cooccurrence Constraint with Root Consonant Distance in Arabic". Ms., University of Utah, Salt Lake City. Also presented at the Fifth Annual Symposium on Arabic Linguistics, 1991, University of Michigan, Ann Arbor. . 1989c. "Verb Vocalism in Modern Standard Arabic". Ms., University of Utah, Salt Lake City. Ramos, Teresita. 1985. Conversational Tagalog: A functional-situational approach. Honolulu: University of Hawaii Press. Schramm, Gene M. 1962. "An Outline of Classical Arabic Verb Structure". Readings in Arabic Linguistics ed. by Salman H. al-'Ani, 497-516. Bloomington: Indiana Linguistics Club, 1978. [Repr. from Language 38:36075.]

Index of Subjects A ablaut 173, 177-80, 182 agreement 7- 20, 22- 26 adjectival 10, 18, 24 asymmetry 7-8, 17 external 31, 47, 54 gender 8, 14-15, 17,21 internal 31-32, 44, 53-55 poor 20, 22-24, 26 SPEC-head 41-50, 53, 55-56, 84-85 subject-verb agreement 47, 53 rich 20, 22-23, 25 verbal 7, 15,23 ambisyllabic 108-10, 117 Arabic Abu Dhabi 122, 123 Alexandrian 122, 124, 125, 127130 Cairene 107-13, 115-17, 120, 122-125, 127-130 Gulf 122-23 Iraqi 122 Jeddah 59-60, 64, 67 Lebanese 119, 121-22, 124-25, 127, 129, 130-32 Lower Egypt 122 Makkan 122-23, 124-25 Modern Standard 7, 10-18, 20, 22-26, 29-30, 34, 40, 42-43, 45, 48, 54, 59-63, 67, 74, 86, 89, 127 Nablus 70-71, 73-77, 87 Palestinian 70, 71, 76, 78, 82-86, 88, 157-58, 159, 166-67 Sudanese 122-24 Syrian 119, 121-22, 124-25, 127, 129, 130-32 C case 11, 12, 16, 17, 19, 31,34-35, 37-41,45,47

categories empty 59, 62, 67 functional 69-70, 83, 86 child language 69, 83, 85 coda 119-21, 123-25, 127, 129, 130132, 108-10, 112-13 concatenation 174, 177, 188 construct state 29-32, 34, 36-37, 5455 D derivation 173, 176-78, 181, 184, 188 discontinuity 173-77, 188 E edges 173-77, 188 epenthesis 122, 125, 166, 170 errors morphological 95-96 phonological 96, 100 semantic 96 I infixation 173 L languages Blackfoot 108, 110, 117 Breton 8, 25 Chamorro 142 Chinese 117, 142 Dutch 9, 26 English 9, 20, 107-8, 110-11, 114, 116-17 French 69, 76 Modern Greek 69, 84-85 Modern Hebrew 34, 39, 42, 43, 45, 54,142 Hindi-Urdu 142 Japanese 117 Korean 110 Semitic 77

192

INDEX OF SUBJECTS

Swiss German 110 Tagalog 174-75 Ute 142 Welsh 26 G government 31, 38-50, 52, 54 H head functional 31, 34, 51 government 48-50, 52-54, 56 head-to-head movement 35, 39, 41, 53, 55-56 I incorporation 11, 12, 17 inflection 7, 11, 18, 19,20,24 L language disorders 95 lexical (de)composition hypothesis 94, 102 lexical processing 94-95, 104 licensing 59, 62- 63, 66-67 M modality 69-70, 76, 79, 81, 86, 88-89 mora 117, 119-21, 123-26, 128-32 morpheme edges 173, 174, 177, 178, 183, 184, 188 morphological selection 73-75, 87 morphology derivational 93, 102-105 inflectional 93,101-105 lexical 93-94 movement 74-75, 82-84, 87-88 N negation 69-79, 82, 84, 86-89 S subject null 12-13, 17 pronominal 12 thematic 12, 14, 17 O onset 108-10, 112, 113, 117 open syllable shortening 157, 160, 166-70

Optimality Theory 121, 123, 125-26 P pause-break 107-10, 117 parametric 7, 10, 15, 17, 20, 23, 24, 26 parasitic gap 59-63, 65, 67 prosodic structure 124-25 Q quantifier 29-37, 39-46, 50-51, 54 R resumptive pronoun 12, 17, 59, 6165,67 rhyme 107, 112-13, 115-17, 119, 126 S sound similarity judgments 111-13, 115-117 syllable 107-12, 114, 116-17 boundaries 107 duration 119, 121, 126-29, 13132 structures 107, 119, 121 weight 119-21, 130 syncope 170 T tier 107, 112-17 topic continuity 141-44, 148-49, 15153 V verb movement 10, 13, 14, 17, 69, 88 vowel shortening 122, 129, 157 W wh-in-situ 59-60, 64-67 wh-resumptive pronoun 61-63 wh-trace 59-65, 67 whole word hypothesis 94 word order 7, 10, 13, 20-21, 23-24, 26 SOV 9, 17, 26 SV0 7-8, 11-12, 14-17,20-21, 26 VSO 7- 8, 11, 14-17,20,22-23, 26