afroasiatic grammar

559

Upload: raubn111

Post on 22-Nov-2015

112 views

Category:

Documents


3 download

TRANSCRIPT

  • RESEARCH IN AFROASIATIC GRAMMAR II

  • AMSTERDAM STUDIES IN THE THEORY AND

    HISTORY OF LINGUISTIC SCIENCE

    General Editor

    E. F. KONRAD KOERNER(Zentrum fr Allgemeine Sprachwissenschaft, Typologie

    und Universalienforschung, Berlin)

    Series IV CURRENT ISSUES IN LINGUISTIC THEORY

    Advisory Editorial Board

    Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.)Sheila Embleton (Toronto); John E. Joseph (Edinburgh) Manfred Krifka (Berlin); Hans-Heinrich Lieb (Berlin)

    E. Wyn Roberts (Vancouver, B.C.); Hans-Jrgen Sasse (Kln)

    Volume 241

    Jacqueline Lecarme (ed.)

    Research in Afroasiatic Grammar II

    Selected papers from the Fifth Conference on Afroasiatic Languages, Paris, 2000

  • JOHN BENJAMINS PUBLISHING COMPANYAMSTERDAM/PHILADELPHIA

    RESEARCH INAFROASIATIC GRAMMAR II

    SELECTED PAPERS FROM THE FIFTHCONFERENCE ON AFROASIATIC

    LANGUAGES, PARIS, 2000

    Edited by

    JACQUELINE LECARMELaboratoire de Linguistique Formelle,

    CNRS Universit Paris 7

  • The paper used in this publication meets the minimum requirements of AmericanNational Standard for Information Sciences Permanence of Paper for PrintedLibrary Materials, ANSI Z39.48-1984.

    8 TM

    Library of Congress Cataloging-in-Publication Data

    Conference on Afroasiatic Languages (5th: 2000: Paris, France)Research in Afroasiatic Grammar II: Selected papers from the Fifth Conference on Afroasiatic Lan-guages, Paris, 2000 / Edited by Jacqueline Lecarme.

    p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Currentissues in linguistic theory, ISSN 0304-0763 ; v. 241)

    Includes bibliographical references and index.1. Afroasiatic languages--Grammar--Congresses. I. Title: Research in Afroasiatic grammar two. II.

    Title: Research in Afroasiatic grammar 2. III Lecarme, Jacqueline. IV. Title. V. Series.PJ993. C66 2003492'.045--dc21 2003048150

    ISBN 90 272 4753 6 (Eur.) / 1 58811 386 8 (US) (Hb; alk. paper)

    2003 John Benjamins B.V.No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any othermeans, without written permission from the publisher.

    John Benjamins Publishing Co. P.O.Box 36224 1020 ME Amsterdam The NetherlandsJohn Benjamins North America P.O.Box 27519 Philadelphia PA 19118-0519 USA

  • Table of contents

    Acknowledgements vii

    Alternation of state in Berber 1Karim Achab

    Anti-faithfulness: An inherent morphological property 21Outi Bat-El

    The internal structure of the determiner in Beja 35Sabrina Bendjaballah

    Reciprocals as plurals in Arabic 53Elabbas Benmamoun

    Modern Hebrew possessive yeS constructions 63Nora Boneh

    The thematic and syntactic status of Ps: The Dative, Directional,Locative distinction 79

    Irena Botwinik-Rotem

    Emergent vowels in Tigrinya templates 105Eugene Buckley

    Transitivity alternations in the Semitic template system 127Edit Doron

    Verbal plurality, transitivity, and causativity 151Abdelkader Fassi Fehri

    Ex-situ and in-situ focus in Hausa: Syntax, semantics and discourse 187Melanie Green and Philip J. Jaggar

    The metathesis effect in Classical Arabic and the representationof geminates 215

    M. Masten Guerssel

    Omotic: The empty quarter of Afroasiatic Linguistics 241Richard J. Hayward

  • Table of contents

    Demonstratives and reinforcers in Arabic, Romance and Germanic 263Tabea Ihsane

    Tonal alternations in Somali 287David Le Gac

    Verb conjugations and the Strong Pronoun declensionin Standard Arabic 305

    John S. Lumsden and Girma Halefom

    The historical dynamics of the Arabic plural system: Implicationsfor the theory of morphology 339

    Robert R. Ratcliffe

    The syntax of special inflection in Coptic interrogatives 363Chris H. Reintges

    Indexicality, logophoricity, and plural pronouns 409Philippe Schlenker

    Vowel innovation in Arabic: Inductive groundingand pattern symmetry 429

    Kimary Shahin

    Phrasal movement in Hebrew DPs 447Ivy Sichel

    Prosodic Case checking domain: The Case of constructs 481Tal Siloni

    Templatic effects as fixed prosody: The verbal system in Semitic 511Adam Ussishkin

    Index 531

  • Acknowledgements

    The 22 articles in this volume grew from papers presented at the Paris Conferenceon Afroasiatic Languages (CAL 5), which was held at University of Paris 7, France,on June 2830, 2000. It was the fitfh of a series initiated in Paris 1992, continued inSophia Antipolis 1994, Sophia Antipolis 1996, and London 1998.

    I wish to thank the invited speakers, Richard Hayward, Mohand Guerssel andAlec Marantz, the referees who evaluated the abstracts, and Jean Lowenstamm andUr Shlonsky for their help in reviewing the papers in this volume.

    Financial support from Centre National de la Recherche Scientifique (CNRS),University of Paris 7, and Laboratoire de Linguistique Formelle is gratefully ac-knowledged.

    Jacqueline Lecarme

  • Alternation of state in Berber*

    Karim AchabUniversity of Ottawa

    . Introduction

    State alternation refers to the morphological change between two alternative formsof nouns, traditionally called free state and construct or bound state (henceforth FSand CS respectively). These forms are illustrated with the feminine and masculineexamples given in Table (1) below:1

    (1) Masculine Feminine

    Free state a-mghar t-a-mghar-t

    Construct state w-mghar t-mghar-t

    Nouns take the FS form when they occur as (lexical) preverbal subjects, objects ofgenuine prepositions, and object of verbs. These situations are illustrated in (2ac)below, respectively (FS nouns in bold):

    (2) a. a-rgaz y-eccaFS.man 3m.sg.ateThe man has eaten

    b. Y-ewwet s a-zru3m.sg.hit to FS.stoneHe launched (something) towards the stone

    c. Y-engha a-zrem3m.sg.killed FS.snakeHe killed the snake

    Nouns take the CS form when they occur as (lexical) postverbal subjects, ob-jects doubling accusative clitics, and objects of non-genuine prepositions. Thesesituations are illustrated in (2ac) below, respectively.

  • Karim Achab

    (3) a. Y-ecca w-rgaz3m.sg.ate CS.manThe man has eaten

    b. Y-engha-t w-zrem-nni3m.sg.killed-cl CS.snake-thatHe killed the snake

    c. Y-ewwet s w-zru3m.sg.hit with CS.stoneHe hit with a stone

    Note that the element s in (2b) and the one in (3c) are distinct and have differentmeanings. In this study, we will refer to the elements patterning with the latter asnon-genuine prepositions and to those patterning with the former as (true) preposi-tions. This difference will be dealt with in more detail in Section 3. The clitic objectdoubled by a lexical object is specific to a few dialects only, its discussion will berestricted to its occurrence in Kabyle Berber.

    The FS form is characterized with the initial vowel a- in both masculine andfeminine, and the loss of the morpheme w- in the masculine. In contrast, the CSform is characterized with the loss of the initial vowel in both masculine and fem-inine, and the prefixation of the morpheme w- in the masculine. In a recent study(Achab 2001), we showed on the basis of facts well established in the literature, thatthe morpheme w- is a former masculine gender marker, which parallels the femi-nine prefix t- and whose survival in the CS form is specialized in the morphologicalreinforcement of the masculine construct state form. It will become clear later onthat the syntactic difference between CS and FS is indeed restricted to the initialvowel in both masculine and feminine forms.

    From the morphological point of view however, the CS morphology of themasculine form is more complex than its feminine counterpart in that the formerinvolves the loss of the initial vowel and the prefixation of the morpheme w- (or itsvariant y-, see Guerssel 1983; Dell & Jebbour 1991), while the latter involves the lossof the initial vowel only, both as compared to their FS counterparts. The manifes-tation of the morpheme w- in the masculine CS form led almost all Berber scholarswho considered this topic, with the notable exception of Guerssel (1987), to erro-neously consider the CS form as marked and the FS form as unmarked (see for in-stance Chaker 1995:45). Therefore, we agree with Guerssel (op. cit.) in consideringthe reverse situation as true.

    The loss of the initial vowel in CS is partially accounted for in the literature interms of its affixal origin. Indeed, in cases where the initial vowel is maintained inthe CS form, it is not a prefix but part of the word stem (see particularly Guerssel,op. cit., and Dell & Jebbour, op. cit.). The so-called initial vowel will be discussed inSection 2 where we will argue in favour of its head status as a determiner. Accord-

  • Alternation of state in Berber

    ingly, this will infer that the internal structure of FS nouns is composed of a head D,represented by the initial vowel, and a complement NP in conformity with the hy-pothesis developed by Abney (1987), as illustrated in (4) below with the masculineFS noun a-rgaz (see (2a)):

    (4) DP

    D NP

    rgaza-

    Alternatively, we will argue in Sections 3 and 4 that CS nouns are bare NPs that areselected by a non-genuine preposition or an agreement morpheme as illustratedin (3) above. Together, these elements and their CS NPs form full DPs. We willfurther argue that these elements act as head determiners in the same way as theinitial vowel does with respect to FS nouns, as illustrated in (5) below with themasculine CS noun s w-zru (see (3c)):

    (5) DP

    D NP

    w-zrus

    The discussion will have a lot in common with the one developed by Guerssel(1987, 1992) in which these elements as well as the initial vowel are consideredas head case markers.

    At first glance, the agreement morphemes (as in 3ab) and the non-genuineprepositions (as in 3c) do not seem to constitute a natural class and behave differ-ently with respect to adjacency. Guerssel (op. cit.) considered the hypothesis thatthese elements do constitute a natural class by assuming that they are all case mark-ers. However, his analysis did not go as far as to explain why some (case marker)elements appear adjacent to the CS nouns while others are realized on the verb asagreement morphemes. In Section 4, we will argue that this state of affairs resultsfrom the incorporation of the agreement morphemes onto the verb, from a headposition adjacent to its CS complement NP. This proposition will constitute themain innovation of the present study.

    This line of reasoning will account for important diachronic changes in someEastern Berber dialects such as Libyan Berber dialects and Egyptian Berber of Siwa.In these dialects, the FS is generalized to situations corresponding to the ones illus-trated in (5) above as a consequence of the CS loss. As may be expected, this changehad particular consequences on the status of elements considered as determiners.

  • Karim Achab

    In other words, we expect the subject agreement morpheme and the elements weconsider as non-genuine prepositions to no longer act as determiner heads of DPsin these dialects. We will show in Section 5 that this prediction is correct. The sub-ject agreement morpheme has acquired an inflectional status while the elementsconsidered as non-genuine prepositions have evolved to (genuine) prepositions se-lecting a DP instead of an NP. Moreover, we will see that these changes had otherconsequences with respect to other phenomena such as topicalization.

    If we believe that state alternation is associated with definiteness, as many au-thors have been supposing in the literature, we would also expect it to be affectedby CS loss in eastern Berber dialects. So far, confusion has prevailed in the debateas to which of CS or FS is to be considered as marked [+definite]. However, We willargue in Section 6 that definiteness is neither a property of CS nor that of FS nounsas both may be associated with the definite as well as the indefinite reading. To ac-count for this disparity of situations, we will adopt the idea developed by Chierchia(1998) relating definiteness to a [+definite] operator.

    . The initial vowel is a determiner

    In this section, we are going to argue in favour of the head Det status of the initialvowel, which is characteristic of FS nouns as shown in examples (2) above. Theinitial vowel has been the focus of much attention in Berber linguistics since Basset(1932). It has been supposed by different linguists to contain all sorts of syntac-tic and semantic features among which gender marker (far too many to be listed),definiteness (Vycichl 1957; Chaker 1995); nominalizer (El Moujahid 1993); deter-miner (Vycichl 1957; Prasse 1974; Ouhalla 1988); number (Basset 1942, 1945; ElMoujahid 1993); default case marker (Guerssel 1987, 1992) and free state marker(Prasse 1974). Vycichl (1957) argued that it used to form a definite article in asso-ciation with the gender markers t- (for feminine) and w- (for masculine), later onincorporated onto the rest of the noun. Guerssel (1987) treated the initial vowelas a default case marker heading a maximal projection called KaseP, which selectsthe rest of the word as a DP headed by the gender markers t- (feminine) and w-(masculine), although the masculine marker does not show up in FS nouns. Sucha proposal gives rise to order conflict between the initial vowel and the gendermarker t- in the feminine form. In Guerssels account, the initial vowel a-, consid-ered as a Kase head, will precede the feminine gender morpheme t-, and give riseto the non-existent form *a-t-mghart as opposed to the correct form t-a-mghartas given in Table (1) above. To account for this conflict, the author attributed thisproperty to a phonological process similar to the one at the origin of certain for-mations known as portmanteau words such as du and aux in French for instance,

  • Alternation of state in Berber

    which are derived from the combination of a preposition de/ with a determiner, leand les, respectively. Note that supposing the opposite order between the Kase headand the determiner head will only transfer the problem to CS since the elementsconsidered by Guerssel as Kase heads with respect to CS nouns do precede gendermarkers.

    We will get back to the CS part of this question in Section 3 dedicated to CS.For the moment, we will proceed with the FS form and show how the problemraised in Guerssels study finds a solution in the present analysis. In the presentstudy, we assume that FS nouns are DPs headed by the initial vowel a-which isa determiner. This determiner inflects for gender, by attaching to it the femininemarker t- or the masculine w-, yielding ta- or wa-. Such a hypothesis is closely akinto the idea suggested by Vycichl (1957) that the elements wa- and ta- are formermasculine and feminine determiners (definite articles in Vicychls analysis), agglu-tinated to the stem later on. Bearing in mind that the masculine morpheme w-disappeared in the FS form, we get the following structures (6a) and (6b) for theFS masculine noun azru and its feminine2 counterpart tazrut, respectively:

    (6) b.a. DP[+fem.]DP

    D[+fem.]D NPNP

    zrutzru [+fem.]a- D

    t- a-

    In these structures, the determiner a- provides the syntactic category feature [+D]while the NP-object is presented by the stem. In Berber, stems/roots are syntacti-cally unmarked, therefore cannot head a maximal projection of their own. In (6b)both the syntactic category feature [+D] and the gender feature [+feminine] per-colate to the maximal projection DP along the lines suggested by Selkirk (1982:76).

    Note that unlike the structure illustrating the feminine DP in (6b), the struc-ture given in (6a) does not contain a gender feature. In a recent study (Achab 2001)we concluded that what is referred to as the masculine form in Berber is in fact anunmarked form and that the masculine interpretation is obtained by contrast withthe feminine form.

    . Non-genuine prepositions are head determiners

    Lets compare the examples given in (2b) and (3c) reproduced as (7a) and (7b)below respectively:

  • Karim Achab

    (7) a. Yewwet s azru3m.sg.hit to FS.stoneHe launched (something) towards the stone

    b. Yewwet s w-zru3m.sg.hit with CS.stoneHe hit with a stone

    The two sentences form a perfect minimal pair and contrast on the basis of theirmeaning and the state form of their selected objects. The main difference betweenthese two examples is that the element s with the directional meaning in (7a) isfollowed by a FS object, while the one in (7b) which has the instrumental meaningis followed by a CS object. For reasons we have not discussed yet, we restricted thestatus of preposition to the former, while referring to the latter as a non-genuinepreposition. Given that nouns in Berber are DPs, the DP status of the object se-lected by the element s in (7a) infers that the latter is a true preposition. In con-trast, the object in (7b) lacking the initial vowel is not a full DP but a bare NP andas such, it has to be associated with a head D, which is represented by the element s.Note that the gender morpheme w- in (7b) cannot be the head D or else we wouldend up with two determiners in FS feminine nouns that contain both the initialvowel and the feminine gender morpheme t- (see Table 1).

    Consequently, unlike the DP [D a-[NP zru]] in (7a) which is the object of thepreposition s, the DP [D s[NP w-zru]] in (7b) is the object of the verb y-ewwet.This view contrasts with Guerssels (op. cit) who supposed an empty prepositionbetween the verb and the DP (KaseP in Guerssels analysis). The structures of thesentences given in (7a) and (7b) are illustrated in (8a) and (8b) respectively:

    (8) b.a. VP

    V PP

    DP

    NP

    Yewwet P

    s D

    a- zru

    VP

    V DP

    NPYewwet D

    s w-zru

    The syntactic difference between these two types of s elements was also discussedby Guerssel (1987, 1992) and Ouhalla (1988). Guerssel considered the element s in(8a) as a true preposition as opposed to the one in (8b) which he considered as acase marker, heading a maximal projection referred to as KaseP. Ouhallas view isdifferent in that he considers as genuine prepositions what we consider here as non-genuine prepositions while he calls disguised adverbials what we here refer to as

  • Alternation of state in Berber

    genuine prepositions. However, this terminological divergence should not botherus much in that the author got to the conclusion that the elements we consider hereas non-genuine prepositions are nominals, thus marked [+N]. Such a conclusiondoes not contradict the Det status that we have assumed for these elements in thepresent analysis.

    Now that we have identified the position of the determiner D in both FS andCS forms, lets get back to the question relevant to the CS part of the order conflictbetween the determiner and the gender morpheme let in suspense in Section 2in our discussion of Guerssels analysis, and suggest an alternative to Guersselsportmanteau solution. The structure given for the feminine FS DP in (6a) showsthe gender morpheme t- under D. It contrasts with the structure given for CS DPsas illustrated in (8a), which shows the gender morpheme w- not under D but underits complement NP. What has to be explained is then why it is that the gendermorpheme is generated in distinct positions in FS and CS.

    To answer this question, we will first assume that CS nouns are derived througha two-step process. The gender morpheme is first associated with a word stemas illustrated in (9a) below, and the resulting inflected stem is then selected by adeterminer as illustrated in (9b):

    (9) b.a. NP[+fem.]

    NP[+fem.][+fem.] NP

    zrutt-

    DP

    D

    t-zruts

    Note that the CS derivation in (9) parallels the FS derivation as elaborated in (6b)in some aspects, but also contrasts with it in some other aspects. They are anal-ogous in that they both involve two distinct morphological processes, the first ofwhich consists of gender morpheme prefixation and the second of a selection ofthis stem by a head D. They are divergent because the determiner represented bythe element s in (9b) is outside the gender morpheme t-, while the determiner rep-resented by the initial vowel a- in (6b) is inside the morpheme t-. Note that if weline the derivation in (6b) up with the one in (9b), we would obtain the erroneousFS form *a-t-zrut instead of the right form t-a-zrut in (6b). Compare the deriva-tion of the ungrammatical FS form *a-t-zrut with the correct CS form s tzrut asillustrated with the structures (10a) and (10b) below, respectively:

  • Karim Achab

    (10) b.a. DP

    D NP

    zruts t-

    *DP

    D NP

    zrut*a t-

    Instead of keeping the ungrammatical structure in (10a) and postulate a possibleportmanteau-like phonological reordering of the morphemes a-t- into t-a- as sug-gested by Guerssel (1987, 1992) in the way recalled above (see Section 2), we willrather maintain that FS and CS DPs are derived differently, exactly as illustratedin (6b) and (9) respectively. Considered this way, the contrast between FS and CSderivations is then precisely located in the first process illustrated in (9a) above, in-volving the prefixation of the gender morpheme. That is, while the morpheme gen-der is associated with the determiner in the FS derivation, it is associated with thestem in the CS derivation. This conclusion perfectly conforms to the general ideaadopted in this study, supposing that the initial vowel is the element that makesthe difference between FS and CS forms. In other words, when the initial vowela- is present, gender prefixation obtains with the determiner a-. In contrast, whenthe initial vowel a- is not present, gender prefixation obtains with the stem. Whatthis conclusion amounts to is that words in Berber inflect for gender regardlessof state in that the gender morphemes either associate with the determiner a- orwith the word stem, but never with elements belonging to the class of non-genuineprepositions, hence the asymmetry observed between (6b) and (9) above.

    This analysis also accounts for the other asymmetry observed between the FSmasculine form and the CS masculine form. The derivation of the CS masculineform goes through the two processes mentioned above in the same way as illus-trated in (9) for the feminine form. It contrasts with the derivation of the FS mas-culine form that contains only the second process associating the determiner a-with the word stem as depicted in (6a). Contrary to the account provided in mosttraditional analyses, the masculine and the feminine CS forms are not derived fromtheir corresponding FS forms by deleting the initial vowel a- and by prefixing themorpheme w- in the masculine. Instead, both CS forms are directly derived byprefixing the morphemes t- and w- to word stems, which means that CS forms arenever associated with the determiner a- in the course of their derivation.

    In this section we have shown that the class of elements we referred to asnon-genuine prepositions are in fact determiners, which select NPs to form fullDPs. The syntactic status of these determiners was contrasted with that of the ini-tial vowel we had assumed to be a determiner of FS nouns. We concluded thatboth types of determiners have the same function, which is the licensing of theirrespective NPs.

  • Alternation of state in Berber

    The initial vowel and the non-genuine prepositions have in common to beadjacent to their NPs and this property differentiates them from the other class ofelements upon which depend CS nouns and which we referred to as agreementmorphemes as shown in sentences given in (3ab). In the following section, wewill argue that this difference results from the subsequent incorporation of theagreement morphemes onto the verb.

    . Subject and object morphemes are former determiners

    Berber being a pro-drop language, the subject morpheme on the verb suffices tosatisfy the subject requirement (Extended Projection Principle (EPP)). When thelexical subject, co-indexed with the agreement subject morpheme, is realized post-verbally, it has to occur in the CS form as shown in the example (11) below:

    (11) Yecca w-rgaz3m.sg.eat CS.manThe man has eaten

    This sentence may be opposed to the one in (12) below, in which the lexical subjectoccurring in the pre-verbal position has to take the FS form.

    (12) agaz yeccaFS.man 3m.sg.eatThe man has eaten

    On the other hand, in the dialects allowing clitic-doubling objects such as KabyleBerber, the lexical object has to take the CS form when it doubles a clitic as shownin example (13):

    (13) Yecca-t Yidir w-ghrum-nni3m.sg.eat.cl Yidir CS.bread-thatYidir ate the bread

    This example may be opposed to the one in (14) below, in which the object doesnot double a clitic, and to the one in (15) in which the clitic-doubling object isleft-dislocated. In both situations, the object takes the FS form.

    (14) Yecca Yidir a-ghrum3m.sg.eat bread CS.manThe man ate bread

  • Karim Achab

    (15) A-ghrum yecca-t YidirCS.bread 3m.sg.eat.cl YidirYidir ate the bread

    As we can see, unlike what we observed with respect to the class of non-genuineprepositions dealt with in the previous section, adjacency does not hold betweenthe agreement morphemes and the CS nouns when lexical subjects and objects oc-cur in the CS form. For instance, in (11) the verb intervenes between the (subject)morpheme y- and the CS noun w-rgaz while in (13) the subject Yidir intervenesbetween the cltic object t- and the CS object weghrum-nni.

    In the present section we are going to generalize the D hypothesis discussed inthe previous section to the morpheme agreements and provide an account for thefacts just mentioned. We will first discuss the subject morpheme in Section 4.1 andthen the object morpheme in Section 4.2.

    . Subject morpheme

    In Section 3, we assumed that CS nouns are complement NPs selected by a headdeterminer D. We noticed that two conditions had to be met for a complement CSNP to be projected: (i) it has to be selected by a head D and (ii) be adjacent to thehead that selects it. Although the examples given in (11) and (13) show that theagreement morphemes are not adjacent to their CS NPs, we will nevertheless showthat adjacency does hold at some level of derivation.

    Lets start by assuming the following derivation for the example given in (11)above (with irrelevant details omitted) by analogy with the structure given in (8a):

    (16) b.a. VP VP

    V VDP DP

    NP NP-cca y -ccaiD D

    y- tiw-rgaz w-rgaz

    (16a) shows the lexical subject DP as the complement of V prior to the incorpora-tion of the subject morpheme y- onto the verb. At this stage, the subject morphemeis the head D of the maximal projection DP and selects its CS NP w-rgaz to whichit is adjacent. (16b) shows that the morpheme y- has incorporated onto the verb,leaving a trace in its original position. The incorporation satisfies locality conditionand Head Movement Constraint or its equivalent. Note that we have omitted theTNS layer so as not to encumber the structure. The order of the morphemes consti-

  • Alternation of state in Berber

    tuting the inflected verb reflects the mirror principle in that the subject morphememoves first to V, which subsequently moves to TNS and both the morpheme y- andthe verb adjoin to the left of their landing site. Remember that in the present anal-ysis, CS nouns, which are considered here as bare NPs, cannot be licensed withouta Det head. If the morpheme y- is not generated as a determiner under D, the adja-cency condition mentioned above would be violated and there would be no otherway to license the CS noun.

    To show that the agreement morpheme y- appearing on the verb is indeed de-rived by movement, lets consider another verb inflectional form in the languagecalled the participial. The participial form in Berber is invariable with respect toperson and is derived by adding the (participial) suffix -n to the third masculinesingular perfect form of the verb. That is, the (masculine singular) subject mor-pheme y- appears as a prefix and the participial morpheme n- appears as a suffixin the way illustrated in (17):

    (17) win y-ecca-nWho 3m.sg.eat.partThat who has eaten

    The negative counterpart of the example in (17), found cross-dialectally, is givenin (18):

    (18) win ur y-ecca-n (ara)Who neg 3m.sg.eat.part (neg)The one who has not eaten

    However, in a few Berber dialects such as Kabyle, there exists an alternative formto (18) in which the morpheme n appears as a prefix, instead of the morpheme y-,as opposed to its suffix position in (18). This possibility is illustrated below:

    (19) Win ur n-ecci (ara)That who neg part.eat (neg)That who has not eaten

    These examples combined with the derivation we suggested in (16) constitute ev-idence that the subject morpheme y- can be singled out from the verb inflectionand we take this to infer that this morpheme is more likely to be a clitic rather thana fossilized inflectional morpheme. Guerssel (1995) also concluded to the cliticstatus of the subject morpheme on the basis of DP extraction out of clitics. Theposition of the morpheme y- is more obvious in the derivation given in (16) aboveas the analysis developed here necessitates a D element in a position that is adjacentto the CS NP.

  • Karim Achab

    We believe that the movement of the subject morpheme in (16) as well as itsinsertion (or movement) in (17) and (18) has to do with the satisfaction of theExtended Projection Principle (EPP). Viewed this way, the movement of the par-ticipial morpheme n from the lower position in (19) may be motivated by the samereasons. The crucial difference between the 3rd person singular morpheme y- in(11) and its participial variant in (17) and (18) is that its projection and movementare compulsory in the former situation but not in the latter. In (11) its projection iscompulsory as a D category because it has to licence the lexical NP subject, while itsincorporation onto the verb is required by the EPP. This situation contrasts with(17) and (18) in which the subject morpheme y- is motivated by the subject re-quirement only. The alternative example in (19) from Kabyle Berber shows thatthis requirement can be satisfied by the participial morpheme n-, which in (17)and (18) appears as a suffix. This latter possibility dismisses the hypothesis that theoccurrence of the morpheme y- in the subject position as in (16), (17) and (18)has to do with nominative case assignment as supposed by Guerssel (1992, 1995).On the other hand, it provides more support to the idea according to which theEPP can be satisfied by categories other than (nominal) subjects as suggested byHolmberg (2000). The clitic status of the subject morpheme puts it on a par withthe object clitic whose status is more obvious and to which we turn presently.

    . Object morpheme

    In this section, we will argue that the object clitic is also a determiner incorporatedonto the verb from a lower D position in the same way as argued for the subjectmorpheme in the foregoing. Consider the following example from Kabyle Berber,a dialect that allows lexical objects to be doubled with a clitic.

    (20) Yecca-t w-ghrum-nni3m.sg.eat.cl CS.bread-thatHe ate the bread

    On the basis of the same arguments as the ones put forward with regard to thesubject morpheme, we will assume that the object clitic originates in a D positionadjacent to its complement NP as in the following structure:

  • Alternation of state in Berber

    (21) b.a. VP VP

    V VDP DP

    NP NPy-cca y-cca-tD D

    -t tiw-ghrum-nni w-eghrum-nni

    (21a) shows that the object clitic is first base-generated as the head D selecting aCS NP to which it is adjacent. Like in the subject situation, the clitic subsequentlyincorporates onto the verb leaving a trace in its original position as shown in (21b).The reason why the object clitic appears on the right side of the verb in (21b) hasprobably to do with the saturation of the left site by the subject morpheme. Theincorporation of the clitic onto the verb as shown in (21b) makes it clear why theadjacency condition does not hold on the surface between D and the CS noun asshown by the example (13) above.

    The structures we have so far postulated for the agreement morphemes in as-sociation with their CS nouns show that the former are determiners base-generatedin a D position, which selects CS objects as NPs. The same D position we consid-ered to be occupied by elements referred to as non-genuine prepositions, and bythe initial vowel with regard to FS nouns in Sections 3 and 2 respectively.

    So far, we have considered the relation between agreement morphemes andtheir respective lexical CS NP complements with which they are co-indexed. Thisrelation of co-indexation, but not adjacency, also holds between agreement mor-phemes and their corresponding lexical FS nouns. This is an interesting point toconsider in that the subject and object agreement morphemes associated with FSnouns cannot be base-generated in the D position since this position will be occu-pied by the initial vowel which, as argued in Section 2, is also a determiner head.This topic will be dealt with in the following section.

    . FS DPs

    In this section, we will consider two types of different situations where DPs occurin the FS form: topicalized subjects and objects, as well as FS nouns as they occurin Eastern Berber dialects in which the FS form has disappeared. The structure ofthese DPs and their relation with agreement morphemes will be compared to thoseof CS nouns as analyzed in the foregoing. In Section 5.1 we will consider topic DPsas they occur in Kabyle Berber. In Section 5.2 we will consider FS DPs as they occurin the Egyptian Berber dialect of Siwa.

  • Karim Achab

    . FS DPs as topics

    The following example shows the topicalized subject in FS:

    (22) A-rgaz y-eccaFS.man 3sg.ateThe man has eaten

    The topic a-rgaz in (22) is a DP headed by the initial vowel a- and co-indexed withthe subject agreement morpheme y- showing up on the verb. A question that ariseshere is whether the topicalized subject is base-generated in the topic position (to beidentified later) or derived by movement from a VP-internal position. Rememberfrom our earlier discussion that the morpheme y- is a determiner incorporatedonto the verb from a VP- internal D position. This entails that the DP a-rgaz cannotbe base-generated VP-internally since the D position that would host the initialvowel is taken by the morpheme y-. Accordingly, the only remaining possibility forthe DP to surface as a FS noun is to be base-generated in the surface topic position,under spec, TP or TOPP.

    Guerssel (1987), Shlonsky (1987) and Ouhalla (1991) identified the topic posi-tion such as the one in (22) as spec of INFL. However, if we have to restrict the roleof [spec, TNS] to EPP satisfaction, this position will not be projected since this rolewill be taken by the subject agreement morpheme y- subsequent to its incorpora-tion onto the verb. In view of these considerations, we will assume that the subjecttopic in (22) is base-generated in [spec, TOPP] as illustrated in the structure below:

    (23) [TopP A-rgaz [TP yi-eccay [VP ti ty]]]

    However, the argument above does not explain why the topic cannot be base-generated in the CS form as a NP-complement of the morpheme y-, and thenmoves to the preverbal position where it would surface in the CS form. Such anoption would give raise to an ungrammatical sentence as the stared example in(24a) shows with its structure illustrated in (24b) with irrelevant details omitted:

    (24) a. *w-rgaz y-eccaCS.man 3sg.ateThe man has eaten

    b. *[TopP w-rgazy [TP yi-ecca [DP ti ty]]]

    The main difference between the erroneous example in (24) and the grammaticalexamples involving the CS form discussed so far has to do with the order of themorphemes. In the latter examples, the D element always precedes its complementNP whereas in the ungrammatical example in (24) the D element y- is preceded bythe NP w-rgaz. This difference finds explanation in the notion of c-command. Inother words, the D element c-commands its CS NP-complement in all the gram-

  • Alternation of state in Berber

    matical examples but not in the ungrammatical example in (24). As we can see,c-commanding the CS NP-trace does not rescue the structure. Therefore, gener-ating the topic as a (FS) DP with the initial vowel as the D head remains the onlypossibility to avoid violating c-command of the NP by its head. The same analysisequally accounts for object topics.

    . FS DPs in eastern Berber dialects

    Lets now consider other FS situations, provided by Eastern Berber dialects such asthe Egyptian Berber dialect of Siwa. In this dialect, FS is generalized to all nominalsituations, including those corresponding to the Kabyle CS nouns discussed so far,that is NPs associated with non-genuine prepositions and the subject agreementmorpheme.

    In Siwa Berber, elements corresponding to the Kabyle non-genuine preposi-tions discussed above select full FS DPs, which suggests that they evolved fromdeterminer to (genuine) prepositions, a view also shared by Guerssel (1995:131).FS DPs selected by these elements have a structure similar to the one elaborated inSection 2.

    The status of the subject morpheme and its relation with the lexical subjectneeds more explanation. To illustrate this situation, lets consider the examplein (25) from Egyptian Berber of Siwa (reproduced from Laoust 1931:147) andcompare it to its Kabyle counterpart given in (26):

    (25) I-umm-as a-zidi3m.sg. told-dat FS.jackalThe jackal told her/him

    (26) Y-enna-yas u-zirdi3m.sg. told-dat CS.jackalThe jackal told her/him

    The lexical subject a-zidi in (25) is in FS and contrasts with the CS form u-zirdi in(26). The semi-vowel u- is a variant of the morpheme w-. The initial vowel i on theverb I-umm-as in (25) is a notational variant of the morpheme y- in (26) dealt withabove. The element -as in (25), realized as -yas in (26), is a dative clitic occupyingthe same position as the object clitic -t discussed above, which does not affect theissue under discussion. The CS noun in (26) is an NP occupying a complementNP position. The head D with which it forms a full DP is the same as the oneoccupied by the trace in the structure given in (16) subsequent to the morpheme y-incorporation onto the verb. This contrasts with the FS form a-zidi in (25), which isa full DP headed by the initial vowel a-, which in turn means that the morpheme y-cannot be base-generated under the same D. In view of that, the subject morpheme

  • Karim Achab

    in (25) is directly base-generated on the verb, unlike the situation in (26) in whichthe subject morpheme is derived by movement from a lower D position. This leadsus to conclude that the morpheme i- in (25) is inflectional, and contrasts in thisrespect with the clitic status of the morpheme y- in (26).

    The generalization of FS in Eastern Berber dialects presents one more challengeto another idea well established in the linguistic literature, which we have not dealtwith so far. State forms have often been related to definiteness by many authors ina way or another. If such an association were true, we would expect nouns in SiwaBerber to be all definite or all indefinite, or else developed another device to encodedefiniteness. In the following section, we will show that definiteness is neither aproperty of CS nouns nor that of FS nouns but a feature structurally associatedwith nouns regardless of their state form.

    . Definiteness

    The debate on definiteness and state has been controversial and the views suggestedin the literature are contradictory. On the one hand, traditional linguists considerthe FS form as definite. We have recalled that Vycichl (1957) associated definitenesswith the initial vowel, therefore with the FS form, by considering the composed ele-ments ta- and -wa as former feminine and masculine definite articles. This idea hasrecently been explicitly reasserted by Chaker (1995:49). Such a hypothesis wouldmean that the opposition between FS and CS is based on definiteness, which isvery unlikely.

    In contrast, Guerssel (1995) stated that CS subjects always have definite readingas opposed to FS objects that might be definite or indefinite according to the con-text. Nevertheless, the author refuted the direct association between definitenessand the CS form on account of the fact that CS nouns associated with non-genuineprepositions may also have a definite or an indefinite reading according to the con-text. The author interpreted the definite reading of CS subjects and objects as aproperty of clitic doubling in general given that they are associated with agreementmorphemes, which is not the case with CS nouns associated with non-genuineprepositions in the way we stated above. Considered this way, both definitenessand indefiniteness may be associated with either CS or FS forms.

    There is another important fact suggesting that definiteness is not a propertyof nouns per se. Sentences involving subject and object morphemes without theirlexical NPs are always definite as shown in (27ab) below:

    (27) a. Y-ecca1sg.eatHe (def.) has eaten

  • Alternation of state in Berber

    b. Yecca-t3sg.ate-3.sg.clHe ate it (def.)

    Consequently, the answer to this question must reside elsewhere than within eitherof the state forms. A possibility we might consider is the hypothesis developed byChierchia (1998) postulating an autonomous definite operator in sentences con-taining definite DPs. Chierchia argued that languages that do not have the def-inite article have a non-overt operator with the same function. Accordingly, thestructure of definite DPs, FS and CS alike will have the following structure:

    (28) OP

    OP DP

    NPD

    Considered this way, the definite and indefinite readings of both CS and FS nounswill be explained by the presence or absence of the definite operator accordingly.Such a hypothesis provides a coherent account of definiteness in both FS and CSsituations as mentioned earlier. Note also that the structure in (28) is compatiblewith all the structures involving CS and FS forms suggested in this analysis.

    . Conclusion

    In this paper we have dealt with syntactic and morphological facts underlying al-ternation of state in Berber. The morphological data relevant to the oppositionbetween FS and CS was presented in Section 1.

    In Section 2 we argued that FS nouns are DPs whose head D is representedby the prefixed initial vowel. The hypothesis constitutes an alternative to the onesuggested by Guerssel (1987, 1992) viewing this vowel as a default case marker.

    FS DPs were opposed to CS nouns in Sections 3 and 4 where we claimed thatthe latter are bare NPs, which have to be selected by a head Determiner category inorder to form a full DP precisely because they lack the initial vowel. These head de-terminers are of two types, prepositions selecting CS nouns, which we referred toas non-genuine prepositions (Section 3) and agreement morphemes (Section 4).We argued that subject and object agreement morphemes appearing on the verbare derived by incorporation from a lower position we identified as D. This incor-poration account has the advantage of lining up the class of agreement morphemeswith the class of elements referred to as non-genuine prepositions and it explains

  • Karim Achab

    why adjacency holds on the surface between the latter elements and their CS nounsbut not between the former and their CS nouns.

    After dealing with the different situations involving the CS form, we consid-ered some of the instances in which nouns occur in FS (Section 5). In Section 5.1we examined why subject topics require the FS form as opposed to non-topicalizedlexical subjects occuring in CS. We concluded that this requirement is motivatedby the c-command principle. We assumed without arguing that the analysis carriesover to clitic doubling objects as they occur in Kabyle Berber.

    In Section 5.2, situations requiring the CS form in dialects such as Kabyle arecompared with their FS counterpart in the Egyptian Berber of Siwa. FS nouns cor-responding to Kabyle CS complement of non-genuine prepositions were accountedfor by simply assuming that the elements considered as determiners in Kabyle havebecome genuine prepositions in the dialect of Siwa. As far as the subject agreementmorpheme is concerned, we concluded that it is not base-generated as a Det headbut directly realized as a verb inflectional morpheme as supposed to its clitic statusin other dialects with CS.

    We ended this study by examining a controversial issue associated with alterna-tion of state, namely definiteness. We assumed, after Chierchia (1998), the existenceof a definite operator in sentences involving definite FS and CS nouns alike.

    Notes

    * I am very grateful to L. Galand, M. Guerssel, J. Ouhalla, M. Rivero, P. Hirshbller,J. Jensen, J. Lecarme, U. Shlonsky, and M. Allaoua, for their comments and suggestions.

    . Unless specified otherwise, all examples are from Kabyle Berber. For the sake of clarity, wewill separate the initial vowel a-, the feminine gender marker t- and the (former masculinegender) morpheme w- from the rest of the word with a dash.

    . Although the feminine marker is represented by the discontinuous morpheme tt inboth FS and CS forms, we will here be concerned by the prefix part only, the suffix partbeing irrelevant to the discussion.

    References

    Abney, S. (1987). The English Noun Phrase in its Sentential Aspect. Doctoral dissertation.Cambridge, MA: MIT.

    Achab, K. (2001). Changement de fonction morpho-syntaxique, le cas du morpheme w- enberbre. Cahiers Linguistiques dOttawa, 29, 5167.

    Basset, A. (1932). Sur ltat dannexion en berbre. Bulletin de la Socit de Linguistique deParis, XXXIII, 173174.

  • Alternation of state in Berber

    Basset, A. (1942). Sur le pluriel nominal en berbre. Revue Africaine, LXXXVI, 255260.Basset, A. (1945). Sur la voyelle initiale en berbre. Revue Africaine, 8289.Chaker, S. (1995). Linguistique berbre. Paris-Louvain: Editions Peeters.Chierchia, G. (1998). Plurality of mass nouns and the notion of semantic parameter. In

    S. Rothstein (Ed.), Events and Grammar. Dordrecht: Kluwer.Dell, F. & Jebbour, A. (1991). Phonotactique des noms voyelle initiale en berbre. Linguistic

    Analysis, 21, 119147.El Moujahid, E. (1993). Syntaxe du Groupe Nominal en Berbre Tachelhiyt. Thse de

    doctorat dtat, Universit de Rabat.Guerssel, M. (1983). A phonological analysis of the construct state in Berber. Liguistic

    Analysis, 11, 309330.Guerssel, M. (1987). The status of lexical category preposition in Berber: implication for the

    nature of construct state. In M. Guerssel & K. Hale (Eds.), Studies in Berber syntax. MITLexicon Project Working Papers.

    Guerssel, M. (1992). On the case system in Berber. The Canadian Journal of Linguistics, 37,113299.

    Guerssel, M. (1995). Berber clitic doubling and syntactic extraction. Revue Qubecoise deLinguistique, 24, 112133.

    Holmberg, A. (2000). Scandinavian stylistic fronting: How any category can become anexpletive. Linguistic Inquiry, 31 (3), 445483.

    Laoust, E. (1932). Siwa: son parler. Paris: Librairie Ernest Leroux.Ouhalla, J. (1988). Syntax of head movement, a study of Berber. Ph.D. dissertation,

    University College London.Ouhalla, J. (1991). Fuctional Categories and Parametric Variation. London: Routledge.Ouhalla, J. (1993). Subject-extraction, negation, and the anti-agreement effect. Natural

    language and Linguistic Theory, 11, 477518.Prasse, K. (1974). Manuel de Grammaire Touargue (tahaggart), IVV: Nom. Copenhague:

    Akademisk Forlag.Selkirk, L. (1982). The Syntax of words. Cambridge, MA: MIT Press.Shlonsky, U. (1987). Focus construction in Berber. In M. Guerssel & K. Hale (Eds.), Studies

    in Berber syntax. MIT Lexicon Project Working Papers.Vycichl, W. (1957). Larticle dfini du berbre. In Mmorial Andr Basset (pp. 139146).

    Paris: Adrien Maisonneuve.

  • Anti-faithfulnessAn inherent morphological property

    Outi Bat-ElTel-Aviv University

    . Introduction

    This paper provides an analysis of the main aspects of imperative truncationin Colloquial Hebrew (also called Spoken Israeli Hebrew), whereby the imper-ative is derived from the future form by truncating V or CV from the pre-fix. Since truncation is a type of non-affixal morphological process, it providesgood empirical grounds for the argument that morphological processes are trig-gered by morphological constraints rather than by listed affixes. These constraintsare argued to be inherently anti-faithful (competing with phonological faithful-ness constraints), and as such, they reflect the purpose of morphology to createparadigmatic contrast.

    The present section reviews the relevant theoretical aspects of morphologywithin Optimality Theory (Prince & Smolensky 1993 and subsequent studies).Section 1.1 presents the view that morphological processes are triggered by con-straints, and Section 1.2 proposes that morphological constraints are inherentlyanti-faithful. Section 1.3 introduces truncation within this view. The next sectionprovides the analysis of imperative truncation in Colloquial Hebrew. Section 2.1presents the constraint interaction involved in truncation and Section 2.2 discussesthe cases where truncation is blocked. The conclusion in the last section pointsout the advantage of the approach advocated here, for truncation as well as formorphology in general.

    . Morphology as constraints

    Studies in morphological theory have been concerned, since Hockett (1954), withthe status of affixes in the grammar. Are they listed items, which concatenate with

  • Outi Bat-El

    the root/stem (item-and-arrangement) or do they form an integral part of theprocess expressed by the morphological rule (item-and-process)? OptimalityTheory (Prince & Smolensky 1993 and subsequent studies) has remained quitesilent with respect to this debate, appealing to both views at the same time; af-fixes are represented in the input along with the base (thus items), as well as in thealignment constraints which place them in the appropriate position with respectto the base (thus process). For example, in McCarthy and Princes (1993) analysisof Tagalog -um- infixation, the affix -um- appears not only in the alignment con-straint but also in the input. This approach raises a duplication problem where anaffix is specified twice.

    The item vs. process debate is concerned with morphology in general and notjust with affixation. As argued in Anderson (1992) the presence of a process-onlymorphology, like ablaut and truncation, as well as segmentally empty affixes, as inreduplication, poses a serious problem to the view that morphology simply con-catenates morphemes. Anderson argues that morphology involves processes, whichexpress relations between words. The processes are formally represented by WordFormation Rules (Aronoff 1976), which can attach an affix, as well as change thequality of the vowel (ablaut) or the consonant (mutation), delete segmental ma-terial (truncation), and so forth. This view is also manifested within the frame-work of Lexical Phonology (Kiparsky 1982), where affixes are integrated within themorphological rules.

    Within the framework of Optimality Theory processes are the outcome of con-straint interaction. In particular, there must be at least one constraint that triggersthe process and this constraint must outrank the constraint that would otherwiseblock the process. This view of grammar can be applicable to morphology, onlyif morphological processes are triggered by constraints as well. Russell (1995) ar-gues for this approach to morphology within the framework of Optimality Theory(see also Yip 1998; Adam & Bat-El 2000): In the morphemes-as-constraints ap-proach, a morpheme is not itself a representation, but rather a set of requirementsthat a representation must satisfy in order for it to become the winning candidate(Russell 1995:17).

    Were morphological processes exclusively affixation, the debate would havebeen merely theoretical. It is process-only morphology, such as truncation, thatprovides the empirical basis for the process approach to morphology. However,Optimality Theory views all morphological processes as triggered by items, whichare then arranged by constraints. Not only are affixes represented in the input alongwith the stem, but also reduplication and truncation are triggered by the abstractmorphemes red (McCarthy & Prince 1993) and trunc (Benua 1997) respectively.One problem with such an approach (see others in Russell 1995) arises in caseswhere the abstract morpheme does not have a surface realization. In Papago, forexample, truncation does not affect vowel final bases; in order to derive the per-

  • Anti-faithfulness

    fective from the imperfective the final consonant is truncated, but if the perfec-tive ends in a vowel the perfective and imperfective are identical (Zepeda 1983;Anderson 1992). The question in such a case is what is the penalty for not trun-cating, i.e. not satisfying trunc? One may suggest a constraint like Faith Trunc(like Faith Affix), which is violated when truncation is blocked. However, faith-fulness constraints require phonological identity between input and output, butsince trunc lacks phonological material there is nothing to be faithful to.

    In this paper I eliminate the abstract morpheme trunc, arguing that trun-cation is due to a violable constraint. When this constraint is dominated by acompeting constraint, it is violated, and truncation is then blocked.

    . Morphology as anti-faithfulness constraints

    [A]ny linguistic system is in a state of tension arising from the competing forcesof the semantic and the phonetic poles of language (Kisseberth 1976:45). InOptimality-Theoretic terms (Prince & Smolensky 1993; McCarthy & Prince 1993,1995) the semantic pole is faithfulness constraints and the phonetic pole is marked-ness constraints. When a faithfulness constraint outranks a competing markednessconstraint (Faithfulness Markedness) the output is faithful to the input,and when a markedness constraint outranks a competing faithfulness constraint(Markedness Faithfulness), the output is unfaithful to the input. Only inthe latter case do we observe phonological alternation.

    Markedness constraints are not by themselves anti-faithful, as they do not referto the input. It is their conflict with faithfulness constraints which may give riseto an unfaithful output. There are, however, anti-faithfulness constraints in thegrammar, and these are the morphological constraints, which trigger the variousmorphological processes. Both faithfulness and anti-faithfulness constraints serveto provide contrast: faithfulness constraints maintain phonemic contrast and anti-faithfulness constraints create paradigmatic contrast.

    Faithfulness violation has thus two sources: phonology and morphology.Faithfulness violation in phonology is forced by a higher-ranked markedness con-straint. For example, if the input ends in d and the output in t, faithfulness isviolated due to a higher-ranked markedness constraint requiring voiceless ob-struents at the end of the syllable/word (Final Devoicing). In morphology thereare no higher-ranked constraints triggering such a faithfulness violation. Sincethe function of morphology is to create contrast, faithfulness violation is an in-herent property of morphological constraints. That is, morphology is by natureanti-faithful.

    The notion of anti-faithfulness is introduced in Alderete (1999), where it is ar-gued that every faithfulness constraint has an anti-faithful counterpart. Crucially

  • Outi Bat-El

    though, anti-faithfulness is restricted to morphology; that is, only morphologi-cal constraints are anti-faithful. The (anti-faithful) morphological constraints aredrawn from the set of phonological constraints, or as stated in Martin (1988),whatever can be done by phonology, can also be done by morphology.

    The logic behind the inherent anti-faithfulness of morphological constraintslies in the contrastive function of morphology. The purpose of morphology is tocreate a contrast between lexical categories and therefore a morphological con-straint must require the output to be phonologically distinct from (or unfaithfulto) the input. Affixation, the most common way to encode morphological contrast,is also anti-faithful since it adds phonological material to the input and thus vio-lates Dep (every segment in the output has a correspondent in the input). This isprobably why affixes usually consist of a few segments (sometimes only one).

    . Truncation

    Truncation, the morphological process that is the concern of this paper, involvesdeletion of segmental material. Truncation here refers only to non-templatic trun-cation, as templatic truncation is triggered by prosodic constraints on the output(as in hypocoristics such as Robert Rob). Truncation is a type of morphologi-cally conditioned process where segmental material is deleted in a particular lexicalcategory, where the specific lexical category is the only relevant environment.

    Any type of segmental deletion violates the faithfulness constraint MaxSeg(every segment in the input has a correspondent in the output). In phonologicaldeletion there must be some markedness constraint forcing deletion, where theranking is Markedness Faithfulness (Max). For example, when a consonantis deleted from a base such as CCVCVC resulting in CVCVC, the constraint forcingdeletion, say *Complex (complex syllabic positions are prohibited), is cruciallyranked above MaxSeg.

    In non-templatic truncation, which is a morphological deletion, there is nomarkedness constraint which forces deletion. What would then be the triggeringconstraint? I claim that in every morphological process the trigger is inherent in therelevant morphological constraint. That is, the morphological constraint imposingtruncation is the one that requires deletion.

    Following this view, morphological constraints are stated as the negative ex-pression of their phonological counterparts (Alderete 1999). MaxSeg states thatevery segment in the input has a correspondence in the output and thereforeTruncation, which is anti-Max, states the following:1

  • Anti-faithfulness

    (1) TruncationNot every segment in the input has a correspondent in the output.(i.e. there must be at least one segment in the input that does not appearin the output.)

    What is crucial about the statement of Truncation is that it does not limit trun-cation to one segment. It minimizes truncation to one segment since truncationof one segment suffices to satisfy Truncation and at the same time to minimallyviolate MaxSeg (Benua 1997). However, a markedness constraint, ranked aboveMaxSeg, may force truncation of two segments as will be shown in the discus-sion below. Therefore all cases of non-templatic truncation, within and across lan-guages, are due to the force of this universal constraint (where the relevant lexicalcategory is specified for each particular case).

    In the ensuing sections I will present some data from imperative truncationin Colloquial Hebrew, where sometimes one segment is truncated, sometime two,and sometimes none. This apparent inconsistency of the size of the truncated ma-terial is accounted for by the interaction of the violable morphological constraintTruncation with other constraints.

    . Colloquial Hebrew imperative truncation

    The base of Colloquial Hebrew truncated imperatives (TIs) is the second personfuture form (see arguments in Bolozky 1979; Bat-El 2002). A TI is derived fromthe future form by truncating a CV (2a) or a V (2b) from the future prefix tV-(stress is final unless otherwise specified). This is thus a case of output-outputcorrespondence (McCarthy & Prince 1995; Benua 1997).2

    (2) The data

    Masculine FeminineFuture TI Future TI

    a. CV truncationtivrax vrax tivrexi vrexi to run awaytiftax ftax tiftexi ftexi to opentitfor tfor titferi tferi to sawtigzor gzor tigzeri gzeri to cut

    b. V truncationtikanes tkanes tikansi tkansi to entertegalgel tgalgel tegalgeli tgalgeli to rolltenawek tnawek tenawki tnawki to kisstemale tmale temali tmali to fill

  • Outi Bat-El

    * The normative imperative forms (masculine feminine) are as follows: (2a)brax birxi, ptax pitxi, tfor tifri, gzor gizri; (2b) hikanes hikansi, galgel galgeli, nawek nawki, male mali.

    These data, supported by the table below, suggest that imperative truncation inColloquial Hebrew is not a case of templatic truncation whereby the output has tofit some fixed prosodic structure. The prosodic structure of the output is contin-gent upon that of the input and not restricted by independent prosodic constraints.

    (3) Patterns of TIs

    Future form Truncated Truncated Outputimperative material patterns

    telamed tlamed V CCVCVC to teachtiwava twava V CCVCVC to sweartiftax ftax CV CCVC to opentakum kum CV CVC to get up

    Fricative-initial TIs such as ftax and vrax (2a) support the claim that the base of aTI is indeed the corresponding future form. These fricatives are derived by post-vocalic spirantization whose environment appears in the future form due to thevowel-final prefix; in word initial position they correspond to stops (cf. the pastforms patax and barax). That is, there is no other source for the initial fricativesin the TIs but the future form. Also the initial t in the TIs in (2b) cannot be in-dependently derived since there are no verbal prefixes in the language that form acomplex onset with the stem initial consonant. This t must then be drawn from thefuture base.

    In the following section I provide an analysis of the data given above. Themerit of this analysis is that both CV and V truncation are accounted for by thesame set of ranked constraints (a classical rule-based approach would require twoindependent rules, one for V truncation and another for CV truncation). Such aunified analysis, which reflects the intuition that this is indeed a single process,can be obtained only within a constraint-based approach, where the phonologicalconstraints interact with the violable morphological constraint Truncation.

    . The analysis

    The examples (2a) show that when the first syllable of the future base is CVC the leftmost CV is truncated.3 Truncation is imposed by the universal constraint Trunca-tion (1), specified for the relevant categories (here Imperative Truncation).

  • Anti-faithfulness

    (4) Imperative Truncation (ImpTrunc): Not every segment in the TI (out-put) has a correspondent in the Future base (input).

    Since the future base begins with a CVC syllable, truncation of just a V would resultin an impermissible triconsonantal onset. This representation is ruled out by thefollowing markedness constraint:

    (5) *[CCC: A triconsonantal onset is prohibited.

    Truncation of C only would result in an onsetless syllable. The familiar Onsetconstraint must then be in force.

    (6) Onset: A syllable has an onset.

    The three constraints given above outrank the faithfulness constraint MaxSeg,which penalizes a form for every truncated segment and thus minimizes the effectof truncation.

    (7) Max Segment (MaxSeg): Every segment in the input has a correspon-dent in the output.

    The tableau in (8) demonstrates the effect of the above constraints in selectingthe optimal candidate for bases beginning with a CVC syllable. As in Yip (1998),the input is specified for the category required of the output (Imp), which acti-vates the relevant morphological constraint (ImpTrunc). For ease of expositionthe truncated material is enclosed in angle brackets.

    (8) Future tiftax TI ftax to open

    Turning now to V truncation, the examples (2b) show that when the first syllablein the future base is CV only the V is truncated. As shown below, V truncation isderived by the same constraint ranking as CV truncation.

  • Outi Bat-El

    (9) Future tesarev TI tsarev to refuse

    The above analysis suggests the following ranking:

    (10) Onset, *[CCC, ImpTrunc MaxSegIn both (8) and (9) MaxSeg minimizes the number of truncated segments, but in(8) the prohibition against a triconsonantal onset (*[CCC) forces truncation oftwo segments, rather than one.

    . Blocking truncation

    Truncation is not always surface true. In this section I present two cases wheretruncation is blocked and the output is thus identical to the input (i.e. the futureform is used for the imperative mood).4 These cases show that Truncation is aviolable constraint.

    .. Bases with a stem initial sonorantVerbs with a stem initial sonorant do not have a corresponding TI when the fu-ture base begins with a CVC syllable (the second C is a sonorant). As shown in(11) below, while this type of verbs does not have a TI (i.e. an imperative form inColloquial Hebrew) it does have a normative imperative (NI).

    (11) No truncation in stems beginning with a sonorant

    Future TI NItinvax *nvax nevax to barktimca *mca meca to findtilmad *lmad lemad to learntirkod *rkod rekod to dance

    The verbs in (11) are expected to have a TI with an initial sonorant-C cluster (cf.tigmor gmor to finish Future-TI), but this cluster violates the Sonority Sequenc-ing Generalization (SSG). Violation of SSG is usually rescued by epenthesis, as inlavan levanim white sg.-pl. (cf. katan ktanim small sg.-pl.), and as shown in(11) also in the normative imperatives (NIs). However, such a procedure is notavailable in the grammar of imperative truncation due a strict base-TI faithful-

  • Anti-faithfulness

    ness, which compels a uniformed paradigm (see Bat-El, to appear). The avoidanceof epenthesis in the TIs is expressed by the high ranking of the anti-epenthesisconstraint Dep.

    (12) Dep: Every segment in the output has a correspondent in the input.

    Dep, as well as the undominated SSG, are crucially ranked above ImpTrunc, thusallowing the untruncated candidate to win.

    (13) Future tinvax no TI to bark (SSG, Dep ImpTrunc)

    Candidate (d), the one which wins when the stem initial consonant is not a sono-rant (cf. tigmor gmor), is ruled out by SSG, candidate (c) is ruled out by *[CCC,candidate (b) is ruled out by Onset, and candidate (e) is ruled out by Dep. Theremaining candidate (a), the untruncated (faithful) form, is thus selected as theoptimal candidate. The revised constraint ranking is as follows:

    (14) Onset, *[CCC, SSG, Dep ImpTrunc MaxSeg

    .. Verbs in Binyan hifil (B-III)As pointed out in Bolozky (1979), all hifil (B-III) verbs consistently lack a TI.

    (15) No truncation in B-III

    Masculine FeminineFuture TI Future TItazkir *zkir tazkri *zkri to remindtafsik *fsik tafski *fski to stoptakim *kim takmi *kmi to raisetapil *pil tapli *pli to droptorid *rid tordi *rdi to put downtake *ke taki *ki to hittavi *vi tavi *vi to bring

    * B-III verbs do have an imperative form in the normative language: hazker hazkri, hafsek hafski, hakem hakmi, hapel hapli, hake haki, have havi.

  • Outi Bat-El

    Notice that there is nothing wrong with the surface structure of these TIs; *ke lookslike ce go out! (from tece), *kim looks like sim put! (from tasim), and *fsik is notvery different from ftax open! (from tiftax). The question is, therefore, why doB-III verbs not have TIs?

    Bolozky notes that only te and ti prefixes are affected by truncation while taof B-III is not. To this we should add the to prefix appearing in a several B-IIIverbs, which also resist truncation (e.g. torid *rid to put down). It is not clearwhether Bolozky attributes the absence of truncation in B-III verbs to the vowelquality in the first syllable of the base (i.e. only [-back] vowels are truncated) orto the binyan (i.e. all binyanim except B-III allow truncation). It seems that theabsence of a TI cannot be attributed to the quality of the vowel because ta- foundin a few B-I verbs is freely truncated (e.g. takum kum to get up Future-TI). Inaddition, there is no obvious reason why a is preserved of all the vowels, since inother instances in the language it is freely deleted or altered, certainly more than iwhich is the most stable vowel in the morphophonology of the language (cf. a ande deletion in yarad yarda to descend Past ms.-fm. and gidel gidla to raise Pastms.-fm. vs. i retention in horid horda to put down Past ms.-fm.).

    The absence of truncation in B-III verbs must then be attributed not to thebinyan as a lexical category but to some property of the binyan. I argue below thatthe relevant property of B-III is not the quality of the vowel following the prefixconsonant but rather its source: this vowel is a stem vowel. Stem segments are nevertruncated, due to the dominance of the constraint MaxSegS (recall the distinctionbetween stem and base mentioned in fn. 3).

    (16) Max Segment Stem (MaxSegS): Every stem segment in the input has acorrespondent in the output.

    MaxSegS crucially outranks ImpTrunc since the latter is violated in order to satisfythe former.

    It is assumed that the distinction between the stem and the prefix is accessiblein the surface forms. This distinction is easily accessible in regular verbs, where thebase is trisyllabic, consisting of a disyllabic stem plus a monosyllabic prefix (seeBat-El 2002 for the distinction between regular and irregular verbs). However, inB-III verbs the distinction between stem and prefix is not as obvious because thefuture bases of this binyan are disyllabic, including the prefix.

    When faced with a future base like tazkir or torid, speakers easily recognize theinitial t as a prefix consonant. However, they cannot determine the status of thevowel following the t. On the one hand it could be a prefix vowel since the futureprefix is usually CV. On the other hand it could be a stem vowel since stems areusually disyllabic. In order to verify the source of the vowel speakers refer to thepast form. If this vowel does not have a correspondent in the past form it is a prefix

  • Anti-faithfulness

    vowel (e.g. gadal tigdal to grow Past-Future B-I), and if it has a correspondent inthe past form it is a stem vowel (e.g. hizkir tazkir to remind Past-Future B-III).That is, when the base is disyllabic there are two bases in the input, the past andthe future forms (see another case of this sort in Bat-El 2002).

    All B-III past forms begin with hi (a few with ho or he) which is often consid-ered a prefix. I argue that while the h is a prefix the vowel following it is a stemvowel to which the vowel in the first syllable in the future base corresponds.5 Sinceit has a correspondent it must be a stem vowel, and so is the vowel in the past form.MaxSegS blocks truncation of the stem vowel (candidates (c) and (d) in tableau(17) below) and Onset blocks truncation of the initial consonant (candidate (b)).The optimal candidate is thus the untruncated (faithful) candidate (candidate (a)).

    (17) Future tazkir, Past hizkir no TI to remind

    Similar instances of correspondence of one output to two inputs have been referredto in the recent literature as multiple correspondence (Burzio 1998 for Italian)or split base (Steriade 1999 for English and French). The multiple base hypoth-esis is not always in force; it needs to be activated. What activates reference to anadditional base in this case is the unexpected number of syllables in the future base,i.e. two instead of three syllables.

    . Conclusion

    The analysis of imperative truncation in Colloquial Hebrew provides the groundsto support two theoretical claims: (i) morphological processes are due to mor-phological constraints and not to items (affixes) listed along with the stem; (ii)these constraints are inherently antifaithful, serving the purpose of morphology tocreate contrast.

    As noted earlier, most studies in Optimality Theory assume that morphologi-cal processes are items and thus introduce them as affixes in the input, along withthe stem, where alignment constraints are responsible for their position with re-spect to the base. Most problematic, as argued in Anderson (1992), are non-affixalmorphological phenomena, such as truncation. Benua (1997) abstracts away from

  • Outi Bat-El

    the issue of item vs. process (as she notes on p. 98, fn. 51) assuming throughout thediscussion an abstract morpheme trunc rather than a truncation constraint. Shecircumvents the problems that may arise with this approach by not considering themost faithful candidate, the one identical to the input. As shown in (18) below, thisapproach works when the actual output is not the faithful candidate and trunc isindeed realized by truncation.

    (18) Future tiftax TI ftax to open (cf. (8) above)

    However, such approach cannot account for the cases presented in Section 2, wherethe actual candidate is the faithful candidate ( marks the wrong optimal candi-date and

    the actual form).

    (19) Future tazkir, Past hizkir TI *azkir to remind (cf. (17) above)

    Truncation must then be triggered by a constraint rather than by an abstract affix.As a constraint it is violable and its interaction with other constraints may blocktruncation.

    The view of morphology as constraint interaction allows to account for caseswhere two affixes compete for the same position. Such a case can be found in Geor-gian (Anderson 1992 and earlier studies) where the 2nd per. Object prefix g- (as inmo-g-klav-s he will kill you) and the 1st per. Subject prefix v- (as in mo-v-klav Iwill kill him) cannot co-occur. Therefore in mo-g-klav I will kill you, where bothprefixes are required, only one can survive. When each affix is introduced by a vio-lable constraint requiring (a non-gradient) alignment to the right edge of the root,the higher ranked constraint is satisfied at the cost of violating the lowered rankedone. An analysis of such competition is given in Anderson (1995) with respect toclitics. Anderson presents the clitics in terms of ranked violable constraints suchthat in case of competition only one is surface true.

  • Anti-faithfulness

    All morphological processes are then best introduced by violable constraints,which interact with other constraints in the grammar. The morphological con-straints are inherently anti-faithful and thus compete with their faithful counter-parts.

    Notes

    . Similarly, ablaut and mutation are Anti-ident, where ident requires featural identitybetween input and output, morphologically conditioned stress shift (as in trnsfer transfr)is Anti-FaithHead, where FaithHead requires the stressed (head) syllable in the input tobe stressed in the output.

    . Some of the TIs are in free variation with CV-truncated forms (e.g. tkanes kanes); seeBat-El (2002) for detailed discussion.

    . The distinction between base and stem is crucial. Base refers to the future form in-cluding the prefix and stem refers to the future form excluding the prefix (reference to thefeminine suffix is not relevant since truncation affects the left edge of the base).

    . It should be noted that in all cases the future and the imperative are in free variation.That is, it is always possible to use the future form for imperative mode whether or not a TIis available.

    . Also B-II past forms start with a prefix (e.g. nigmar was finished nixnas entered). How-ever, since the future form of B-II is usually trisyllabic (e.g. tigamer, tikanes) reference to thepast base is not activated.

    References

    Adam, Galit & Outi Bat-El (2000). Morphological knowledge without morphologicalstructure: The emergence of inflectional suffixes in the acquisition of Hebrew verbs.A paper presented in a workshop on Phonology and Prosodic Morphology of SemiticLanguages. Tel-Aviv.

    Alderete, John (1999). Morphologically Governed Accent in Optimality Theory. Ph.D.dissertaion, University of Massachusetts, Amherst.

    Anderson, Stephen R. (1992). A-Morphous Morphology. Cambridge: Cambridge UniversityPress.

    Anderson, Stephen R. (1995). Rules and constraints in describing the morphology ofphrases. CLS 31 Clitics Parasession, 1531.

    Aronoff, Mark (1976). Word Formation in Generative Grammar. Cambridge, MA: The MITPress.

    Bat-El, Outi (2002). True truncation in Colloquial Hebrew imperatives. Language, 78, 133.

  • Outi Bat-El

    Bat-El, Outi (to appear). Competing forces in Paradigm Uniformity. In L. J. Downing,T. A. Hall, & R. Raffelsiefen (Eds.), Paradigms in Phonological Theory. Oxford: OxfordUniversity Press.

    Benua, Laura (1997). Transderivational Identity: Phonological Relations Between Words.Ph.D. dissertation, University of Massachusetts, Amherst. ROA-259-0498.

    Bolozky, Shmuel (1979). On the new imperative in Colloquial Hebrew. Hebrew AnnualReview, 3, 1724.

    Burzio, Luigi (1998). Multiple correspondence. Lingua, 104, 79109.Hockett, Charles (1954). Two models of grammatical description. Word, 10, 210231.Kiparsky, Paul (1982). From cyclic to lexical phonology. In van der Hulst & N. Smith (Eds.),

    The Structure of Phonological Representations (Part I) (pp. 131175). Dordrecht: Foris.Kisseberth, Charles (1976). The interaction of phonological rules and the polarity of

    language. In A. Koutsoudas (Ed.), The Application and Ordering of Grammatical Rules(pp. 4154). The Hague: Mouton.

    McCarthy, John & Alan Prince (1993). Generalized Alignment. In G. Booij & J. van Marle(Eds.), Yearbook of Morphology 1993 (pp. 79153). Dordrecht: Kluwer.

    McCarthy, John & Alan Prince (1995). Faithfulness and reduplicative identity. In J.Beckman, W. Dickey, & S. Urbanczyk (Eds.), Papers in Optimality Theory (pp. 249384). Amherst: GLSA.

    Martin, Jack (1988). Subtractive morphology as dissociation. In H. Borer (Ed.), Proceedingsof the 7th Conference on Formal Linguistics (pp. 229240). Stanford: Stanford LinguisticAssociation.

    Prince, Alan & Paul Smolensky (1993). Optimality Theory: Constraint interaction inGenerative Grammar. Ms., Rutgers University and University of Colorado, Boulder.

    Russell, Kevin (1995). Morphemes and candidates in Optimality Theory. ROA-44-0195Steriade, Donca (1999). Lexical conservatism in French adjectival liaison. In B. Bullock,

    M. Authier, & L. Reed (Eds.), Formal Perspectives in Romance Linguistics (pp. 243270).Amsterdam: John Benjamins.

    Yip, Moira (1998). Identity avoidance in phonology and morphology. In S. G. Lapointe, D.K. Brentari, & P. M. Farrell (Eds.), Morphology and Its Relation to Phonology and Syntax(pp. 216246). Stanford: CSLI Publications.

    Zepeda, Ofelia (1983). A Papago Grammar. Tucson: University of Arizona Press.

  • The internal structure of thedeterminer in Beja*

    Sabrina BendjaballahCNRS Lille

    . Introduction

    Current work in Distributed Morphology (Halle & Marantz 1993 and relatedwork) assumes that the terminal nodes of syntactic trees are bundles of exclusivelygrammatical features; phonological expressions are inserted only after syntax by anoperation called Spell-Out. At Spell-Out the feature bundles specified in the termi-nal nodes are matched against vocabulary items, the phonological strings avail-able in the language for the expression of syntactic terminals. Vocabulary itemscompete for insertion, and the one that matches the highest number of grammat-ical features specified in the terminal node is inserted in that node. The matchingprocedure itself bypasses the need for a richly articulated phonological compo-nent in two respects. First, a crucial property of phonological strings, namely thatthey have internal structure, is not exploited. Second, the relation of a particularphonological string to the context in which it is inserted is simply stipulated inthe vocabulary of the language. It is an empirical question if this system missesany generalization. In this paper I argue that it does: the insertion of phonologicalexponents is a more complex procedure than the mere competition between vo-cabulary items assumed in Distributed Morphology. I examine the phonologicalproperties of the determiner in Beja1 and suggest that an adequate morphologicaltheory must take the internal phonological structure of the determiners exponentsinto account.

    The determiner in Beja is a prefix attached to the noun. Its different forms aregiven in table (1.1) and some examples in (1.2).2

  • Sabrina Bendjaballah

    (1) 1. The determiner in Beja

    Singular PluralSubject Non-subject Subject Non-subject

    Masculine* u:- o:- a:- e:-Feminine tu:- to:- ta:- te:-

    * The masculine determiner prefix is preceded by w in the singular and y in theplural when the noun to which it is attached begins with w, y, , h, or a vowel(e.g. u:-tk the man vs. wu:-b the kid). This phenomenon will not bedealt with in the present paper.

    2. a. u:-me:kdet-donkey

    e:acome.3ms.past

    The donkey came.vs. me:k

    donkeye:acome.3ms.past

    A donkey came. (Almkvist 1881:54)b. t:-fna

    det-spearihyttake.3ms.past

    He took the spear. (Reinisch 1893b:122e)c. :-mana

    det-visceratmyaeat.3ms.past

    He ate the viscera. (Reinisch 1893a:24, 9)

    A possible list of the vocabulary items for Beja masculine determiner within Dis-tributed Morphology would be as follows:

    Signal Context of insertion/u:-/ [+subject, plural, feminine]/o:-/ [subject, plural, feminine]/a:-/ [+subject, +plural, feminine]/e:-/ [subject, +plural, feminine]

    Although this listing correctly describes the facts in (1.1), it fails to capture thefact that for each gender category the four forms generated by the system are onlydifferentiated by their vowel quality. In (1.1) several grammatical features are ex-pressed by one and only one segment, a vowel. I would like to propose an analysisthat accounts for this property of the determiner in Beja.

    I will explore the hypothesis that a grammatical feature may be expressed byan object which is not phonetically realized as a separate segment, but which con-stitutes a segment in composition with another phonological object. In this paperI will defend the most restrictive version of this hypothesis for the Beja determiner:

  • The internal structure of the determiner in Beja

    (2) The One-to-One-Primitive Hypothesis:

    a. Grammatical features, i.e., the primitives of grammatical representa-tions, are expressed by the primitives of phonological representations.

    b. There is a correspondence between the type of grammatical featureand the type of phonological primitive that expresses it.

    This article is organized as follows. In Section 2 I lay out the theoretical framework Iadopt and define the phonological and grammatical primitives I assume. In Section3 I identify the phonological primitives that make up the determiner in Beja. InSection 4 I complete the decomposition analysis laid out in Section 3. In Section 5I discuss the information that has to be lexicalized. Section 6 concludes the paper.

    . Ingredients: The phonological and grammatical primitives

    I assume the autosegmental framework, where phonological representations con-sist of a segmental tier and a skeletal tier linked to each other by lines of association.

    Concerning the segmental tier, I adopt the element theory as introduced inKaye, Lowenstamm and Vergnaud (1985, 1990) and assume that segments are notthe ultimate constituents of phonological representations. Rather, segments areanalysable into smaller units called the elements. Within this theory, elements arethe primitives of segmental representations. An element is a fully specified matrixand is phonetically interpretable.

    Of interest to us here is the representation of Beja vowels. The vocalic systemof Beja is given in (3).

    (3) i, i: u, u:e, e: o, o:

    a, a:

    (cf. Almkvist 1881:5; Hudson 1964:18; Reinisch 1893b:7879)

    The elements involved in Beja vowels are I, A and U. The matrices of these elementsare given in (4).

    (4) I A Uround round +roundback +back +back+high high +highlow +low low

    [i] [a] [u]

    The phonological structures of Beja vowels appear in (5).3

  • Sabrina Bendjaballah

    (5) Phonetic interpretation: [i] [a] [u] [e] [o]Phonological structure: I A U A.I A.U

    Concerning the skeletal tier, I adopt Lowenstamms (1996) CV model (cf. alsoGuerssel & Lowenstamm 1990). Within this model, the skeletal level consists of astrict alternation of C- and V-positions. There is only one syllable type: CV. Theprimitive of the skeletal level is, therefore, the CV unit.

    In brief, I assume three types of phonological primitives:

    (6) a. The element.b. The CV unit.c. The association line.

    I take the grammatical features expressed on the determiner in Beja to be as in (7).

    (7) a. Gender feature: Masculine (Masc), Feminine (Fem).b. Number feature: Singular (Sg), Plural (Pl).c. Case feature: Subject (S), Non-subject (nonS).d. Definiteness feature: The determiner is definite (Def).

    Given the hypotheses laid out above, the internal phonological structure and thegrammatical feature matrix of each form of the determiner are as in (8).

    (8) Phonetic Internal structure Gramm. featuresexponent Cons. Voc.

    a. [u:] U b. [o:] A.U c. [a:] A d. [e:] A.I e. [tu:] t U f. [to:] t A.U g. [ta:] t A h. [te:] t A.I

    We now have all the ingredients we need in order to determine the phonologicalidentity of the grammatical features of the determiner in Beja.

  • The internal structure of the determiner in Beja

    . The phonological identity of the grammatical features

    . A form of correspondence between the grammatical features and theirphonological exponents

    The grammatical features of the determiner may be divided into two sets: gender,number and case features on the one hand and the definiteness feature (hence-forth, [+definite]) on the other. While [+definite] is not inherited because the de-terminer, by definition, expresses definiteness, gender, number and case featuresare inherited from another syntactic node. Gender and number features are inher-ited from the noun via agreement. Case is assigned to the DP; D inherits case bypercolation on a par with all the nodes under DP.

    I propose to correlate this difference in status between [+definite] and the in-herited features with a difference in the nature of the phonological exponents thatexpress these features. Specifically, I suggest that [+definite] is expressed by the CVprimitive on the skeletal tier which is uncontroversially the backbone of phonolog-ical representations. As for the inherited features, they are expressed in one of twoways: as elements or as association lines. This hypothesis is formulated in (9).

    (9) Phonological Exponents Hypothesis:

    a. The phonological identity of the feature [+definite] is skeletal in na-ture.

    b. The phonological identity of number, gender, and case features iseither an element or an association line.

    . Analysis

    Let us assume that the phonological identity of the feature [+definite] is the prim-itive of the skeletal level, a CV unit.

    (10) [+definite] = CV

    Let us then examine the Gender category. The gender opposition is a binary oppo-sition between the masculine and the feminine. I assume the feminine to be markedand encode the gender opposition as [+Fem] vs. [Fem]. The phonological ex-ponents of gender features are as shown in (11).

    (11) Gender: [+Fem] = t [Fem] =

    Now, let us turn to the phonological identity of number and case features. Theonly consonant present in the phonological structure of the determiner is t. Thissegment expresses the feminine only; number and case features must then be ex-

  • Sabrina Bendjaballah

    pressed by the vocalic part of the determiner. Consider the four characteristicvowels of the determiner, which I have given in (12).

    (12)

    Singular PluralSubject Non-subject Subject Non-subject

    Phonetic realization: [u:] [o:] [a:] [e:]Phonological structure: U A.U A A.I

    At first sight the distribution of the vowels seems odd. Whereas the vowels of thesingular constitute a natural class in that they are both labial, the vowels of theplural do not. However, a closer look at the phonological structures in (12) revealsa more significant dis