d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

ARABIC MODIFYING ADJECTIVES AND DP STRUCTURES* Abdelkader Fassi Fehri Abstract. Important characteristics of the Arabic adjectival system are investigated, in view of the question of how the DP system is organized. A first series of issues include: (a) how adjective serialization observes quasi-universal hierarchical ordering restrictions (or their mirror image order), (b) how adjectives and other modifiers and determiners alternate in postnominal and prenominal positions, and (c) how distributional classes of adjectives relate to attributive/predicative, or head/modifier uses. A second series concerns (d) inflectional properties of adjectives (including Case, Definiteness, and Number and Gender features), and their Head/Spec dependent status. Such questions are approached through postulating an articulated (fissioned) DP structure. Structural grounds are provided for checking various inflectional features in hierarchically ordered, but autonomous DP domains. AP (or A) movement, as well as N and Possessor raisings, are independently motivated. Cross-linguistic variation follows, depending on whether all, some, or none of these processes are involved, to yield convergent derivations. APs (along with NPs) are treated as DPs, taking into account their inflectional and interpretational behaviours. Definiteness inheritance and Genitive checking are reanalyzed in view of new empirical and theoretical considerations of synthetic possessive structures.

1. Introduction The study of the system of adjectival modification, in addition to its importance in its own right, is essential for clarifying how the internal structure of nominal phrases is articulated, and what appropriate parallelisms should be established between DP/NP and CP/IP. In investigating the internal syntax of modifying adjectival constructions in Arabic, evidence can be provided for the need of independent AP (or A) raising, as well as N and Possessor raisings.1 Order variation across languages is then characterized depending on (a) how much use is made of these movement processes for (re-) ordering constituents, and (b) how high the targeted positions are located in the structure. In line with Chomsky's (1995) Attract movement theory, I argue that it * The content of this work has been presented in various forms at the 11th Symposium on Arabic Linguistics held at Emory University, Atlanta (March 1997), at the GLOW 20 Workshops (IERA, Rabat, April 1997), the MIT Linglunch (September 1997), and the LSM annual meeting (IERA, Rabat, February 1998). I would like to thank the audiences there, and acknowledge helpful comments and remarks by Ken Hale, Noam Chomsky, Richie Kayne, Anders Holmberg, David Pesetsky, Pino Longobardi, Morris Halle, Noriaki Yusa, Hyon Sook Choe, and two Studia Linguistica reviewers. The usual disclaimers apply. 1 The term Arabic is used in the text to designate Standard Arabic, the unique official language through the Arab world (with only minor regional variation). Spoken Arabic dialects exhibit more significant variation. Studia Linguistica 53(2) 1999, pp. 105±154. # The Editorial Board of Studia Linguistica 1999. Published by Blackwell Publishers, 108 Cowley Road, Oxford OX4 1JF, UK, and 350 Main Street, Malden, MA 02148, USA

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

106

Abdelkader Fassi Fehri

is the strength of the various features in D which triggers A or N raising to D, on the one hand, and AP or Possessor raisings to Spec D, on the other.2 A feature-based split (or fission) of D structure aÁ la Bittner & Hale (1996a) and Lamontagne & Travis (1987) is then postulated, to account for various principled distributions.3 Hierarchical prominence and mirror image ordering restrictions are derived from a base structure in which multiple APs originate as multiple Specs.4 The latter are locally moved to higher functional Specs, thus creating only standard minimal links.5 Adjectives (like nouns) are inherently endowed with Case, Def(initeness), and Number (= Nb) and Gender (= Gr) features, and they project as DPs. They then receive modificational or non-modificational interpretations, depending on the structures in which they are found. Contrary to the traditionally spread view that Arabic adjectives are only postnominal, it is shown that the latter can also be prenominal. The analysis of adjectival synthetic genitives (= SGAs) is typically unified with that of nominal synthetic possessives and/or partitives (= SGNs). Furthermore, Genitive checking is argued to take place in the DP domain, whereas Def inheritance is shown to apply in some SGs, but not all of them. Postnominal AP order is derived, through antisymmetrical left movement aÁ la Kayne (1994), which observes the Minimal Link Condition (= MLC).6 The paper is organized as follows. In section 2, I provide a brief description of the essential properties and alternatives involved in the syntax of Arabic adjectives, and argue for the existence of prenominal adjectives. In section 3, I investigate the structure of postnominal adjectives, and the mechanisms involved in deriving their mirror image (= MIO). In section 4, I propose an analysis of synthetic possessives which is based on a fissioned DP architecture.7 In section 5, I show why APs should be analyzed as DPs. Finally, I discuss some consequences of the approach adopted in section 6. The latter concern crosslinguistic movement variation, the article content, Genitive checking, adjectival mixed order distributions, as well as competing proposals for dealing with Def inheritance. 2

Chomsky (p. 297) defines Attract as follows: K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K. 3 In Bittner & Hale (1996a), Case and Def are realized on independent syntactic heads (namely K and D). I assume that both heads are Ds, as will be made clear below. Other work has directly or indirectly assumed a similar split (see e.g. Holmberg 1993 and Giusti 1995). 4 On the Multiple Spec hypothesis, see Chomsky (1995) and references cited there. 5 Cinque (1994a) argues that adjectives are specifiers, rather than heads of F. But as we will see, Arabic exploits both options. 6 Chomsky (p. 311) provides the following definition of MLC: K attracts A only if there is no B, B closer to K than A, such that K attract B. 7 On split or fissioned categories, in syntax, see Chomsky (1995), Fassi Fehri (1996), and McGinnis (1995). For a quite different view of fissioning and movement, see Nash & Rouveret (1997). As for Halle & Marante (1993), they propose a strictly morphological view of fission. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

107

2. Adjective classes and their distributions Although Arabic modifying adjectives occur productively in postnominal positions, and adjective serialization observes a mirror image ordering of that found in A-N languages, Arabic can be argued to be a A-N language.8 This typology is supported by (a) the distributional properties of attributive adjectives (which differ significantly from those of predicative adjectives, even though both types are normally placed postnominally), (b) the prenominal positioning of various determiners (including numerals, quantifiers, and demonstratives), which conform to non-MIO universal restrictions, and (c) the existence of prenonimal adjectives. 2.1 Serialized postnominal ordering Arabic adjectives occur normally in postnominal positions. There, they agree in Def, Case, and Nb/Gr with the head noun they modify. They also observe hierarchical prominence restrictions on serialized ordering documented for other languages. The following examples illustrate some of these properties. (1) l-kitaab-u l-?axdar-u s-sagÇiir-u Ç ÇÇ the-book-nom the-green-nom the-little-nom `The little green book' (2) sÏaay-un siiniiy-un ?axdar-u jayyid-un Ç Ç tea-nom Chinese-nom green-nom excelllent-nom `An excellent green Chinese tea' (3) ?-al¹ ab-u bi-l-kurat-i l-kabiirat-i l-jamiilat-i I-play with-the-ball-gen the-big-gen the-beautiful-gen `I play with the beautiful big ball.' In (1), the adjective carries the definite article (in addition to Case and Nb/Gr features) in agreement with the modified noun, whereas the adjective in (2) is indefinite. The construction (3) shows how the genitive case on the head spreads to all adjectives. Furthermore, the order of adjectives observed is the mirror image of that found with attributive adjectives in English or French, for example: (4) a beautiful big (round) red ball (5) un joli gros ballon (rond) rouge Indeed, a number of authors have claimed that the relative ordering of the different classes of adjectives is by and large the same across languages. 8 I have shown elsewhere that Arabic is an Adv-V language, and that adverbs are specifiers of extended functional projections of verbs (cf. Fassi Fehri 1997a, b, which builds on Cinque's (1994b, 1997) theoretical and empirical motivations). Functional parallelisms are then established between nominal and verbal modifiers.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

108

Abdelkader Fassi Fehri

Consider e.g. the serialized ordering prominence restriction on object denoting nominal adjectives in (6), proposed in Sproat and Shih (1988, 1990), and that on event nominal adjectives in (7), found in Cinque (1994a).9 (6) object denoting: quality 4 size 4 shape 4 color 4 provenance (7) event nominals: speaker oriented 4 subject oriented 4 {manner, thematic} These orderings are observed by attributive adjectives in direct A-N languages like Germanic, or indirect A-N languages like Romance. The mirror image of this order (i.e. N-A) is observed in V final languages like Indonesian and Thai.10 In examples (1)±(3), we have seen that the ordering restriction in (6) is respected with object denoting nominals, although in MIO. The same is true of event nominals with regard to (7): (8) l-hujuum-u l-?amiriikiyy-u l-wahsÏiyy-u l-baliid-u l-muhtamal-u Ç Ç the-attack the-American the-salvage the-stupid the-probable `The probable stupid salvage American attack' (9) l-intiqaad-u l-?amiriikiyy-u sÏ-sÏadiid-u li-l-muqaawamat-i the-criticism the-American the-violent of-the-resistance-gen `The violent American criticism of the resistance' Here too, the order is also a MIO. It contrasts with the order found in Romance, in which postnominal adjectives occur in the nominal (noninverted) order, although N raises also above (some) adjectives, as shown by Cinque (1994a). The following Italian examples illustrate this variation.11 (10) la sola grande invasione italiane dell' Albania `the single big Italian invasion of Albania' (11) la probabile goffa reazione immediata alla tua lettera `the probable clumsy immediate reaction to your letter' Spanish, however, as described by Bosque and Picallo (1996), instantiates a case of a Romance language in which the order of adjectives is inverted, unlike what happened in Italian or French. This is illustrated by (12) for object denoting nominal constructions (= (1b) and (48a) in BP), and (13) for event nominal ones (= their (26a) ): 9 Cf. e.g. Sproat & Shih's (1990) AOR given in (i), based on ideas found in Bloomfield (1933), Whorf (1945), Quirk et al. (1972), and other references cited there: (i) Restrictions on the ordering of multiple adjectival modifiers ± henceforth AOR ± obtain iff the adjectives involved are hierarchical direct modifiers. (= their (8) ). Cf. also Cinque (1994a) for restrictions on event nominal modification. 10 These restrictions are also observed in Celtic, as argued e.g. by Rouveret (1994) for Welsh. 11 The terminology indirect A-N languages applies to languages in which N has raised past A at PF, although A is prenominal at base structure.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

109

(12) a. una comedia musical americana a comedy musical American `An American musical comedy' b. una guerra religiosa fratricida devastadora a war religious fratricidal devastating `A devastating fratricidal religious war' (13) produccioÂn marisquera gallega production shellfish Galician `Galician shellfish production' It is important to note that although these adjectives occur postnominally (whether in MIO or not), they are interpreted as attributive, as has been argued e.g. by Cinque (1994a) and Bosque & Picallo (1996) for Romance. The same is true of Arabic adjectives. When relevant ordering restrictions are observed, their interpretation is attributive. Other orders are possible, but they are associated with a different interpretation. Thus observing strict ordering restrictions can be taken as a diagnostic for attributive reading, whereas other orders are (nominally) associated with predicative reading.12 In the next subsection, I will examine distributional and interpretive evidence supporting this view.13 14

12 The attributive vs. predicative distinction of adjectives is a classical one (see e.g. Bolinger 1967 and Hawkins 1983, among others). Other (quite) equivalent terminologies used in the literature are: direct vs. indirect, restrictive vs. non-restrictive, and intersective vs. non-intersective. Furthermore, order has been reported to vary depending on what Sproat & Shih (1990) call `the discourse relevant class' to which the speaker wishes to refer, hence the contrast between e.g. `small brown dogs' and `brown small dogs'. For relevant discussion, see also Crisma (1993, 1995). 13 Note that other orders (which do not observe MIO) are not possible with attributive interpretation, although they may have predicative or focus readings (together with comma intonation), as illustrated by the following counterparts of (1), (2), and (8): (i) l-kitaab-u s-sagÇiir-u l-?axdar-u ÇÇ Ç the-book the-little the-green `The little book which is green' (ii) sÏaay-un jayyid-un siiniiy-un Ç tea excellent Chinese `An excellent tea which is Chinese' (iii) l-hujuum-u l-wahsÏiyy-u l-?amiriikiyy-u Ç the-attack the-salvage the-American `The salvage attack which is American' See the following subsections for further clarifications. 14 Sproat & Shih (1988) treat Arabic postnominal adjectives as indirect modifiers (the equivalent of predicatives in our terminology), and they claim that these adjectives are not subject to their AOR ordering restrictions. According to their informant, no order of postnominal adjectives is basic. But this description is incorrect. On the other hand, they take the placement of the synthetic possessor (before adjectives) and the occurrence of two separate articles on nouns and adjectives to be indications of indirect modification. But see below for a different interpretation of these facts.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

110

Abdelkader Fassi Fehri

2.2. Further features of attributives Consider first the case of typical attributives, which occur only prenominally in English, such as those in (14): (14) a. the alleged murder b. the former president These adjectives cannot be used predicatively, e.g. as a complement of a copula, hence the ungrammaticality of (15): (15) a. *The murderer is alleged b. *The president is former Their exact Arabic counterparts, which also fail predicativity tests, occur (only) postnominally, hence the following contrasts: (16) a. l-qaatil-u l-maz¹ uum-u the-killer-nom the-alleged-nom `The alleged killer' b. l-mudiir-u s-saabiq-u the-director-nom the-former-nom `The former director' c. l-xamiis-u l-faarit-u Ç the-thursday-nom the-last-nom `Last Thursday' (17) a. *l-qaatil-u maz¹ uum-un the-killer-nom alleged-nom `The killer is alleged.' b. *l-mudiir-u saabiq-u-n the-director-nom former-nom `The director is former.' c. *l-xamiis-u faaritu-n Ç the-thursday-nom last-nom `Thursday is last.' The fact that typical attributives are placed (only) postnominally is a clear indication that postnominal positioning is not a characteristic property of predicative adjectives.15 But although attributive adjectives occur productively in postnominal positions, they are still distinguishable from predicative ones on other grounds. One important distributional distinction between the two classes has to do with their placement with respect to the complement of the head noun they are modifying: attributives must be placed before that com15 Note that even the counterparts of adjective-noun compounds in English exhibit postnominal adjective ordering: (i) l-bayt-u l-?abyad-u Ç the-house-nom the white-nom `The White House'

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

111

plement, whereas predicatives occur after it. Thus the contrasts in (18) and (19) show that attributives must precede complements: (18) a. l-mudiir-u s-saabiq-u li-l-maktab-i the-director-nom the-former-nom of-the-office-gen `The former director of the office' b. *l-mudiir-u li-l-maktab-i s-saabiq-u (19) a. s-sabab-u r-ra?iisiyy-u li-stiqaalati-hi the-reason-nom the-main-nom of-resignation-his `The main reason of his resignation' b. *s-sabab-u li-stiqaalati-hi r-ra?iisiyy-u On the other hand, predicative adjectives must follow the complement. Compare the following pair of constructions: (20) a. muhaarabat-u l-hukuumat-i l-muntadarat-u ÅÇ Ç Ç fighting-nom the-government-gen the-expected-nom li-l-irtsÏaa?-i of-the-corruption `The expecting fighting of the corruption by the government' b. muhaarabat-u l-hukuumat-i li-l-irtisÏaa?-i Ç Ç fighting-nom the-government-gen of-the-corruption l-muntadarat-u ÅÇ the-expected-nom `The fighting of the corruption by the government, which is expected' In (20a), the AP has only an attributive reading, while in (20b) it has only a predicative reading. As the English translation shows, the interpretation of (20b) is close to that of a reduced non-restrictive relative clause. Likewise, in analytic possessive constructions, attributive adjectives must precede both the possessor and the complement, whereas predicative adjectives follow both of them:16 (21) a. l-hujuum-u sÏ-sÏadiid-u l-muhtamal-u li-?amiriikaa Ç the-attack-nom the-violent-nom the-probable-nom of-America ¹ alaa l-muqaawamat-i on the-resistance `The probable violent attack of the resistance by the U.S.' b. l-hujuum-u sÏ-sÏadiid-u li-?amiriikaa ¹ alaa l-muqaawamat-i the-attack-nom the-violent of-America on the-resistance l-muhtamal-u Ç the-probable-nom `The violent attack of the resistance by the US, which his probable' 16 Constructions like (21) are productive in Modern Standard Arabic, although their grammaticality is questionable in Classical Arabic. See Fassi Fehri (1993) for a discussion. See also Sibawayhi (8th cent.) and Wright (1858/1974).

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

112

Abdelkader Fassi Fehri

A further distinction between attributives and predicatives suggests that they branch differently in the base structure. If attributive APs are left branch specifiers, as in Cinque (1994a), and predicate APs are on the right branch of N, then we expect them to exhibit a different behaviour with respect to a well-known restriction on left branch maximal XPs, namely their inability to take complements to the right. This expectation is borne out. Consider the following contrasts: (22) a. s-suhufiyy-u t-tawiil-u l-faransiyy-u ÇÇ Ç ÇÇ the-journalist-nom the-tall-nom the-French-nom l-?asl-i Ç the-original-gen `The tall journalist who is of French origin' b. *s-suhufiyy-u l-faransiyy-u l-?asl-i ÇÇ Ç Ç the-journalist-nom the-French-nom the-original-gen t-tawiil-u Çthe-tall-nom Ç (23) s-suhufiyy-u l-faransiyy-u t-tawiil-u ÇÇ Ç ÇÇ the-journalist-nom the-French-nom the-tall-nom `The tall French journalist' According to the hierarchy in (6), the order in (23) and (22b) is the normal order of attributives. However, the only possible order when the adjective takes a complement is (22a). In that order, the first adjective is interpreted as attributive, and the second one as predicative. The order in (22b) is excluded if we assume that the left branch specifiers cannot branch to the right, and that predicative adjectives are lower in the structure than attributives are.17 Summarizing what has been said so far, there is clear evidence that both attributive and predicative adjectives can be postnominal, although they belong to different distributional and interpretive classes. One particular diagnostic which discriminates the two classes is whether they comply (or not) with strict ordering restrictions. In the next subsection, I will show that Arabic determiners which occur prenominally typically observe these ordering restrictions, without MIO effects. In subsection 2.4, I provide evidence for the existence of prenominal adjectives in Arabic. The properties of the constructions to be analyzed corroborate the view that Arabic is an A-N language, and that nominal determiners have a prenominal origin.

17 The argumentation based on the Left Branch Condition is adapted from Cinque (1994a). See also Emonds (1978, 1985) for relevant properties of the latter condition.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

113

2.3. Prenominal determiners The placement of numerals supports the view that ordering within DP observes strict hierarchical prominence. When numerals occur prenominally, ordinals must precede cardinals.18 (24) a. ?awwal-u xams-i muhaadaraat-in Ç Ç first-nom five-gen lectures-gen `The first five lectures' b. *xams-u ?awwal-i muhaadaraat-in Ç Ç five-nom first-gen lectures-gen `The five first lectures' Furthermore, as (25) and (26) illustrate, when prenominal and postnominal adjectives are combined, the prenominal `space' observes the ordering of the hierarchy, while the postnominal `space' observes its mirror image: (25) ?awwal-u xams-i suhuf-in faransiyyat-in masÏhuurat-in Ç Ç first-nom five-gen newspapers-gen French-gen famous-gen `The first five famous French newspapers' (26) taalit-u hujuum-in ?amiriikiyy-in muhtamal-in ¹ alaa Åthird-nom Å Ç attack-gen American-gen probable-gen on s-suudaan the-Sudan-gen `The third probable American attack of Sudan' Numerals provide only an instance of prenominal determiners and/or modifiers which observe `direct' hierarchical prominence. Quantifiers and demonstratives represent other well-known instances in traditional grammars: (27) kull-u haadaa l-kalaami all-nom this Å the-speech-gen `All this speech' (28) kull-u talaatat-i rijaal-in Å every-nom Åthree-gen men-gen `Every three men' In these constructions, quantifiers and demonstratives occur prenominally, and the normal (partial) prenominal ordering is as stated in (29), when the placement of adjectives is taken into account: (29) Q 4 Dem 4 Ord 4 Card 4 A 4 N 18 Observe that a prenominal adjective enters necessarily into a synthetic genitive relation with the adjective or the noun following it. See subsection 2.4 and section 5 for discussion.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

114

Abdelkader Fassi Fehri

Note, however, that these determiners (and/or modifiers) can also occur postnominally. In that case, they observe MIO. For example, when numerals are placed postnominally, ordinals occur after cardinals: (30) a. l-muhaadaraat-u l-xams-u l-?uulaa Ç Ç the-lectures-nom the-five-nom the-first `The first five lectures' b. ?? l-muhaadaraat-u l-?uulaaa l-xams-u Ç Ç the-lectures-nom the-first the-first-nom `The five first lectures' Moreover, the classes of adjectives listed in (6) and (7) occur before numerals: (31) l-kutub-u l-faransiyyat-u l-xamsat-u l-?uulaa the books-nom the-French-nom the-five-nom the-first `The first five French books' (32) l-hujuum-u l-?amiriikiyy-u l-muhtamal-u Ç the-attack-nom the-American-nom the-probable-nom t-taalit-u Åthe-third-nom Å Å `The third probable American attack' Note that although scrambling of this order is possible, as in (33) and (34), this order involves different interpretations (including predicative and focus ingredients): (33) a. ?? l-kutub-u l-xamsat-u l-?uulaa l-faransiyyat-u the-books-nom the-five-nom the-first the-French-nom `The first five books which are French' b. ?? l-hujuum-u l-muhtamal-u t-taalit-u Åthe-third-nom Å Å Ç the-attack-nom the-probable-nom l-?amiriikiyy-u the-American-nom `The third probable attack, which is American' The question marks do not indicate that the constructions are illformed, but only that they should have a non-attributive (and/or focused) interpretation. Furthermore, the facts observed indicate that the canonical order (basically (29) ) is respected in the prenominal space, whereas the postnominal space conforms to MIO, as stated in (34):19 (34) N 4 A 4 Num 4Card 4 Ord 4 Dem 4 Q

19

See Fassi Fehri (1997b) for details.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

115

2.4. Prenominal adjectives Arabic adjectives occur essentially in postnominal positions. But there are quite productive cases in which adjectives occur prenominally, as in the following examples: (35) ?akal-tu ladiida Å Å ate-I delicious-acc `I ate the delicious (of

t-ta¹ aam-i Çthe-food-gen Ç the) food.'

(36) ?-aqra?-u jadiid-a l-kutub-i I-read new-acc the-books-gen `I read the new (of the) books.' (37) yahdutu haadaa fii muxtalif-i l-mayaadiin-i Ç Å this Å happens in various-gen the-fields-gen `This happens in various fields' (literally: in the various of the fields). In these constructions, the adjective is heading a synthetic genitive (= SG). It receives external structural Case, whereas the nominal complement receives Genitive.20 Furthermore and quite unexpectedly, the constructions behave more like nominal than adjective SGs. First, they occur in DP (not AP) positions, and they are interchangeable with other DPs, headed by nouns. Thus the following N headed constructions are equivalent to their above counterparts: (35) a. ?akal-tu t-ta¹ aam-a l-ladiid-a Å Å Çthe-food-acc Ç ate-I the-delicious-acc `I ate the delicious food.' (36) a. ?-aqra?-u l-kutub-a l-jadiid-at-a I-read the-books-acc the-new-fem-acc `I read the new books.' (37) a. yahdutu haadaa fii l-mayaadiin-i l-muxtalif-at-i Ç Å this Å happens in the-fields-gen the-various-fem-gen `This happens in various fields.' Constructions (35)±(37) can be distinguished from (35a)±(37a) in that the former, but not the latter, have a sort of partitive reading (although the two constructions may be equivalent in use and actual interpretation). Second, the SG phrase is definite, indicating that Def inheritance has taken place there. The definite character of the phrase becomes clear when it is modified by a definite relative, for example, as (38) illustrates:21 20 It is also worth noting that the adjective is in a neutral form, which does not agree in Gr and Nb with the nominal complement, a situation which recalls that of VS order agreement properties. 21 Def inheritance is presumably forced due to the Definiteness Partitive Constraint, recently discussed in e.g. de Hoop (1997).

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

116

Abdelkader Fassi Fehri

(38) a. ladiid-u t-ta¹ aam-i lladii ?akal-tu-hu Å Å Å ate-I-it Çthe-food-gen Ç delicious-nom that `The delicious food that I ate' b. muxtalif-u l-mayaadiin-i llatii yahdutu haadaa fii-haa Ç Å this Å in-them various-nom the-fields-gen that happens `The various fields in which this happens' But adjectival SGs do not have these properties. In addition to the fact that they occur only in AP positions, they do not trigger Def inheritance (see Fassi Fehri 1993). In order to be definite, an adjectival head must carry a definite article, as exemplified in (39): (39) bahat-tu ¹ ani l-jamiil-i l-wajh-i Ç Å Å for the-nice-gen the-face-gen looked-I `I looked for the one with a nice face.' But the adjectival head of the prenominal adjectival SG cannot carry the article, hence the ungrammaticality of (40):22 (40) *bahat-tu ¹ ani l-waafir-i l-ihtiraam-i ÅÅ Ç looked-I for the-plentiful-gen the-respect-gen Intended to mean: `I looked for the plentiful (of the) respect.' A third piece of evidence indicating that the construction is nominal, and not adjectival, is provided by the fact that it is not compatible with the use of adverbs, as illustrated by the following contrast: (41) a. ?-ukinnu la-hu l-ihtiraam-a l-waafir-a jidd-an Ç I-entertain for-him the-respect-acc the-plentiful-acc lot-acc `I have a very plentiful respect for him.' b. *?-ukinnu la-hu waafir-a l-ihtiraam-i jidd-an Ç I-entertain for-him plentiful-acc the-respect-gen lot-acc `I have plenty of respect for him.' In (41a), the adjective is compatible with a degree adverbial, but in (41b), it is not. In fact, the adjective there is heading a DP which receives a partitive interpretation (i.e. `the plentiful of the respect'), and is not different from nouns heading partitives, which exclude adverb modifiers.23 22

(40) is obviously excluded only in the relevant interpretation. The other interpretation (i.e. `I looked for who is with plenty of respect') is possible, but irrelevant for the discussion. 23 See Fassi Fehri (1990, 1997c). A further test is provided by preadjectival negation (like gÇayr), which is possible with postnominal, but not prenominal adjectives: (i) ?akal-tu t-ta¹aam-a gÇayr-a l-ladiid-i ÇÇ Å Å ate-I the-food-acc non-acc the-delicious-gen `I ate the non-delicious food.' (ii) *?akal-tu gÇayr-a ladiid-i t-ta¹aami-i ÇÇ Å Å ate-I non-acc delicious-gen the-food-gen Note also that prenominal adjectives can occur in a series, provided they observe genitival relations, as already observed. This is illustrated in the following examples: # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

117

Further cases of prenominal adjectives are provided by superlatives, which occur only prenominally. Superlatives come in two forms, exemplified in (42): (42) a. jaa?a ?ahsan-u lugÇawiyy-in Ç came best-nom linguist-gen `The best linguist came.' b. jaa?a ?ahsan-u l-lugÇawiyy-ii-na Ç came best-nom the-linguists-gen `The best of the linguists came.' Superlatives can be interpreted as partitive SGs, and they behave exactly like other prenominal adjectives for the relevant properties, including distribution and Def inheritance. Adjective and determiner uses in prenominal positions with the properties discussed support the view that Arabic is underlyingly a A-N language. In the next section, I propose an analysis of the structure of attributive APs, which implements the view that they originate in prenominal positions. Postnominal placement is then derived, through appropriate movement mechanisms. 3. An antisymmetrical structure of adjectival constructions 3.1 Ordering and left branching specifiers I have shown earlier that prenominal and postnominal determiners and APs observe ordering hierarchical restrictions (eventually in MIO), and that `mixed' cases also comply with these ordering restrictions quite systematically. Consider example (25) above, repeated here as (43), and the construction (44): (43) ?awwal-u xams-i suhuf-in faransiyyat-in masÏhuurat-in Ç Ç first-nom five-gen newspapers-gen French-gen famous-gen `The first five famous French newspapers' (44) kull-u talaatat-i kutub-in xadraa?-a masÏhuurat-in Ç every-nom three-gen books-gen green-gen famous-gen `Every three famous green books' For the sake of concreteness, let us assume, following essentially Cinque (1994a, 1996), that nominal modifiers (like verbal ones) are generated in

(iii) jayyid-u qadiim-i l-manaazil-i best-nom old-gen the-houses-gen `The best of the old (of the) houses' (iv) sagÇiir-u kibaar-i l-¹ulamaa?-i Ç little-nom big-gen the-scholars-gen `The little of the big (of the) scholars' # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

118

Abdelkader Fassi Fehri

the Spec of a functional projection F, of which the NP is a complement. Given the orderings observed, two solutions suggest themselves: (a) specifiers are generated both to the left and to the right of their heads: surface ordering restrictions are then directly accounted for; (b) specifiers are generated only to the left of their heads: only surface ordering restrictions on prenominal modification are then directly accounted for, and postnominal modification must be derived through movement operations. The (a) option is less restrictive than (b), and hence less theoretically attractive. If we adhere to Kayne's (1994) antisymmetrical restrictive theory of phrase structure (= AS), then (b) is the only option.24 Furthermore, (a) appears to be hardly maintainable on empirical grounds. If both orders of specifiers are possible (in the same language), then it is not clear how the restrictions in (6) and (7) and their MIO can be adequately and simply formulated, nor is it clear how the correlated properties of these alternating orders can be naturally treated. As it will turn out, the left specifier hypothesis embodied in (b) (= LSH henceforth) provides an appropriate and motivated solution for dealing with MIO and mixed order cases.25 3.2. Deriving MIO and variation Cinque (1996) put forth a proposal (inspired by Kayne's AS), according to which all adjectives and modifiers (whether prenominal or postnominal) are generated as left specifiers of N, in conformity with the ordering hierarchies (stated above).26 The N-A order is then obtained either 24 In his AS, Kayne proposes that the antisymmetry of linear order be matched by an antisymmetry in underlying hierarchical structure, through asymmetric c-command. Given then two nonterminals X and Y and the terminals they dominate x and y, `if X asymmetrically c-commands Y, x precedes y'. Kayne's AS is based on his Linear Correspondence Axiom (= LCA): (i) LCA d (A) is linear ordering of T. where d is the nonterminal-to-terminal dominance relation, A is a set of pairs of nonterminals (in which the first asymmetrically c-commands the second), and T the set of terminals. If asymmetric c-command maps to linear precedence, then specifiers (or adjuncts) which asymmetrically c-command their heads necessarily precede them, and heads which asymmetrically c-command their complements necessarily precede them. This is how a strictly rigid Spec 4 Head 4 Comp order is respected. 25 Other difficulties for the (a) assumption arise from the right/left asymmetry in order variation stated in Greenberg's (1966) Universal 20, according to which the order to the right is (relatively) `free', whereas the one to the left is severely constrained. See footnote 26. 26 Cinque (1996) exploits a number of left/right asymmetries in favour of AS. He observes, for example, that while the order and its mirror image is found to the right of N (in addition to other variation), the order to the left of N is severely constrained. For example, Greenberg's (1966) Universal 20 expresses a right/left asymmetry which is accommodated naturally in the AS system: Greenberg's (1966) Universal 20

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

119

through (a) N or (b) NP movement, targeting higher left located heads or specifiers, respectively. According to Cinque (1996), N raising occurs in prepositional (or A-N) languages (with no MIO effect because adjectives do not move), and NP raising occurs in postpositioning (or N-A) languages (with the effects of pied piping APs, and placing them in MIO). To see how Cinque's approach derives MIO through XP raising (in postpositional languages), consider a construction like (45), and its presumed base structure (46):27 (45) l-hujuum-u l-?amiriikiyy-u l-wahsÏiyy-u l-muhtamal-u Ç Ç the-attack the-American the-salvage the-probable `The probable salvage American attack' (46)

np 3 l-muhtamal .

np 2

l-wahsii .ˇ

np 1

l-?amirikii

NP l-hujuum

In order to derive the right surface order, the NP is first moved to the left of the lowest AP, then the whole AP (containing NP) to the left of the most proximate higher AP, and so on. Adjectival MIO and N placement before adjectives are then accounted for. The movement proceeds as indicated in (46a):

When any or all of the items (demonstrative, numeral, and descriptive adjective) precede the noun, they are always found in that order. If they follow, the order is either the same or its exact opposite. `The left/right asymmetry', says Cinque, `consists in the fact that while to the right of the N both the order Dem(onstrative) Num(eral) A(djective), and its mirror-image, A Num Dem, are possible, to the left of the N only the order Dem Num A is attested.' (Cinque 1996:453) 27 I have used np as a category projection which includes AP as its specifier, instead of using (less neutral) designated aspectuo-temporal functional categories like those used by Cinque (1995, 1997) for adverbs, leaving the matter for future research. In (46a), NP has moved first to the left of np1, thus forming np1'. The latter moves to the left of np2, creating np2', which later moves to the left of np3. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

120

Abdelkader Fassi Fehri

(46a)

np 3, np 2, k

np 3

np1, j NPi

np 2 np1 l-wahsii .ˇ

l-hujuumi l-?amirikii

NPi

l-muhtamal np 2, k . np1, j

ek

ej

ei Note, however, that NP raising (as described) cannot be adequately applied to Arabic. Since Arabic is not a postpositional (N-A) language, as I have argued, we expect, following Cinque's lines, that only N raising should be relevant to it. This is in fact correct, since it is widely acknowledged in the literature that N raising (to D) is necessary for deriving word order in Arabic DPs.28 But postulating the existence of a NP raising process makes the wrong predictions with respect to ordering in complex cases. For example, it predicts that the DP complement in the possessive DPs in (20a) (repeated here as (47) for convenience) should be pied piped, and be placed higher than attributive APs. This prediction is not borne out: (47) muhaarabat-u l-hukuumat-i l-muntadarat-u ÅÇ Ç Ç fighting-nom the-governmeent-gen the-expected-nom li-l-irtisÏaa?-i of-the-corruption `The expected fighting of the corruption by the government' As observed above, (47) is the only possible order with attributive APs. In that order, the complement must be left behind the possessor (which raises), and APs surface higher than the complement. Recall that the order in (20b), repeated here as (47a), can only be associated with predicative interpretation (which presumably arises from right branching): (47) a. muhaarabat-u l-hukuumat-i li-l-irtisÏaa?-i Ç Ç fighting-nom the-government-gen of-the-corruption l-muntadarat-u ÅÇ the-expected-nom `The fighting of the corruption by the government, which is expected' 28

See references in footnote 33 below.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

121

Thus in order to derive the surface order in (47), and maintain LSH, it is necessary to postulate independent processes of N and Possessor raisings. This is in fact the solution that I adopt in the following subsection, where these two separate movements are given independent motivation. If N and Possessor raise independently, then no solution for Arabic adjectival MIO is readily available in Cinque's (1996) X' type variation. An alternative option that suggests itself is to postulate independent AP (or A) raising, a process which is able to reorder the series of adjectives. A first piece of evidence for the existence of such a process is provided in the following subsection, where structures of postnominal adjectives are examined. 3.3. Raising postnominal APs In approach aÁ la Cinque (1996), adjectives do not move, be they in N-type or NP-type raising languages. In the latter case, they are only pied piped inside a NP which contains them, as exhibited in the structure (46a) above. In this subsection, I would like to provide evidence that adjectives move, and that their movement is independent of that of NP or Possessor. MIO is then seen as an effect of AP movement. The latter is motivated by the richness of the inflectional properties of Arabic adjectives, which match the strength of identical features in the DP domain. APs then target DP, to check their agreeing Case, article, and phi-features against those of a higher functional head, which could arguably be (a segment of) D. I will designate this D with a small d, for ease of reference, giving this notation no theoretical content. 3.3.1. Checking Def, Case, and Agr on adjectives. As exemplified thoroughly above, postnominal adjectives carry Def, Case, and Nb/Gr features which are identical to those on the head noun. Suppose that in order to check the values of these features against those of the noun, APs must raise from the position in which they are generated to that of a Spec of a functional projection F. Call the features involved Agr features. If the F projection is identified as (a segment of) D, then D might be said to have strong Agr features.29 A preliminary independent evidence of AP raising is provided by the placement of adverbial modifiers of adjectives, which are located after the AP, as in (48): (48) a. l-xabar-u l-mudaa¹ u mu?axxar-an Å the-news the-broadcast late-acc `the lately broadcast news' 29 The proper identification of the F category is not a trivial matter. I will follow Chomsky's (1995) logic on the matter, keeping in line with minimalist conceptual requirements.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

122

Abdelkader Fassi Fehri b. l-hukm-u ma¹ ruufun musbaqan Ç the-judgement known advanced-acc `The judgement is known in advance.'

If the adverbial in (48) is originally generated as a Spec of the AP (which it modifies), then the pre-adverbial position of the AP there is presumably derived via raising the AP to the left of the complex AP. Further evidence for AP raising is provided by the existence of constructions like (21a) above, repeated here as (49), for convenience: (49) l-hujuum-u sÏ-sÏadiid-u l-muhtamal-u li-?amiriikaa Ç the-attack-nom the-violent-nom the-probable-nom of-America ¹ alaa l-muqaawamat-i on the-resistance `The probable violent attack of the resistance by the U.S.' In this construction, the manner AP is placed higher than the modal AP and the possessor, both of which are found higher than the AP in the base structure. This indicates that the position of the AP here is derived. On the other hand, the contrast in (50) indicates that adjectives (carrying articles and other agreeing features) cannot raise higher than nouns: (50) a. l-bint-u l-jamiilat-u the-girl the-beautiful `The beautiful girl' b. *l-jamiilat-u l-bint-u the-beautiful th-girl It suggests that the AP raises to a functional projection which is found lower than the D (or its segment) in which N is located. Let us designate the projections involved by dp and DP, respectively. If AP raises to Spec dp, then N raises to D, which then yields (50a). Moreover, if the possessor in a SG like (47) is in Spec DP1 (as I will show below), then the head N is in a higher (segment of) D, presumably D2. Assuming that the APs raise to dp, and that Possessor and N raise to DP1 and DP2, respectively, we are now able to address the following questions: (a) how do the many adjectives raise, and (b) how does MIO obtain? 3.3.2. Deriving multiple AP structures. Suppose that APs originate as left specifiers, and that the multiple movement of APs is targeting one and the same cluster of features, so-called Agr features, which are located in dp. The question is then how these APs move, without producing minimality violations on movement, and at the same time be able to surface in the right (MIO) order. If APs (like possessors) are # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

123

generated as left specifiers of NP or np (possibly in a np shell structure), presumably because they form a unique (though complex) thematico-aspectual category with N and its other arguments, then assuming a multiple Spec structure, the highest AP (in the hierarchy) is generated higher in the left branch, although it surfaces to the extreme right of N. The base structure of a construction like (49) is then depicted in (51): (51)

np 3 l-muhtamal np 2 . li-?amirikaa

np1

ˇ ˇ s-sadiid

NP l-hujuum

Given this structure, AP movement can then proceed as follows: the highest AP3 moves first to the Spec dp1 (targeting Agr there), then AP2 moves to the Spec of the newly formed category, targeting dp2, and so on. The movement process appears then to be operating in a nesting manner, from the closest lexically filled Spec to the nearest Spec DP, created by move a. Furthermore, the movement appears to avoid Specs which are already filled by traces, and to `jump over' filled Specs which do not form a chain with Spec dp. This multiple movement process (from top to bottom) makes multiple checking possible, because the AP moved creates a chain link between the trace it leaves in its original position and its landing site. The movement of the lower AP then becomes possible because it operates in a local manner to the Spec of the newly formed category, jumping over only the chain of the Specs which are coindexed. The resulting structure is given in (52).30

30 I have used dp to host AP movement, and DP to host NP (or N movement), only for convenience sake, and with no theoretical intention. As for np, it is intended to be a thematic extension of NP, recalling the VP/vp extension.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

124

Abdelkader Fassi Fehri

(52)

DP dp 2

D

ˇˇ l-hujuumi s-sadiid j

dp1

l-muhtamal k np 3 . ek

np 2

li-?amirikaa ej

np1 ei

Let us call the mode of movement involved in (52) the External Spec Derivation (= ESD). ESD might be compared to a competitive alternative in which the movement proceeds through a (created) Spec which is internal to the AP, which is itself a Spec of np. This Internal Spec Derivation (= ISD) has been adopted by Bosque & Picallo (1996) for Spanish, following suggestions in Kayne (1994). ISD may or may not be adequate, depending on various theoretical considerations, in particular how locality and cyclicity are construed. For example, limiting the domain of feature checking to a strictly local relation between a Spec and its Head (as in ESD) appears to be much simpler and more restrictive than the more complex Spec of Spec option (embodied in ISD). I will then adopt ESD for the sake of simplicity and concreteness, pending further research on the matter.31 Summarizing, I have shown that there is evidence for AP movement in Arabic, and that the latter operates through external specifiers. In section 5, evidence is provided for the existence of A movement in prenominal modifying constructions, hence corroborating the view that adjectives move (be they heads or maximal XPs). Before that, however, I examine in section 4 how the synthetic possessive structure is organized, and how both AP and Possessor movements take place, without inducing minimality violations.

31 The movement operations used in ESD observe the Extension Condition proposed in Chomsky (1995, pp. 190±191 and 327±329), which derives strict cyclicity, whereas ISD does not.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

125

4. Possessive structure and movement processes In this section, I present an analysis of the structure of Arabic synthetic genitive possessives (the so-called construct states in the Afroasiatic literature). The latter is framed in minimalist terms, along Chomsky's (1995) theoretical lines. It relies on two essential assumptions: (a) Possessor movement is motivated by Def checking, and (b) N to D movement takes place for Case reasons.32 Any analysis of SG will have to take into account the following essential (and well-known) features: (a) (b) (c) (d) (e) (f)

the head N is DP initial the head N is marked by the Case assigned to the whole DP the possessor occurs after N, and before modifying adjectives the possessor is marked with (a non-prepositional) Genitive case the head N cannot carry the article the head N is interpreted as definite or indefinite, depending on the Def feature value of the possessor.

These properties are instantiated in (53): (53) a. htaraqa-t daar-u r-rajul-i l-waasi¹ a-t-u Ç burned-fem house-nom the-man-gen the-large-fem-nom `The man's large house burned.' b. htaraqa-t daar-u rajul-in waasi¹ a-t-un Ç burned-fem house-nom man-gen large-fem-nom `A man's large house burned.' In both constructions, the head N daar satisfies the (a), (b), and (e) requirements, and the possessor the (d) requirement. Moreover, the head N is definite in (53a), and indefinite in (53b), since the adjective (which agrees with it) carries the (suffixed) definite article in (53a), but lacks it in (53b). This is presumably a consequence of (f). Previous analyses of nominal SG in Arabic, Hebrew, and Berber have agreed that its derivation involves N to D raising. As for the possessor, it is either originally placed or raised to a position higher than that of APs. Thus a construction like (53a) is derived from the original base structure (54), by raising N to D, and placing the possessor in a Spec which is higher than that in which APs are located. The resulting structure is then (54a).33 32 Most analyses of Arabic (or Semitic) found in the literature are pre-minimalist, to my knowledge, with the exception of Longobardi (1996). 33 For various proposals, see Fassi Fehri (1987, 1993), Ritter (1987, 1991), Ouhalla (1988, 1996), Mohammad (1988), Siloni (1994), Borer (1996), and Longobardi (1994, 1996). I have taken both AP and Possessor to be specifiers of NP/np at the base, to simplify the picture, although this question is a matter of divergence. Moreover, the use of FP is intended to be neutral with respect to the disputed identity of the category which hosts the possessor (AGRP, in e.g. Siloni 1994 and Longobardi 1996 or PossP, in Fassi Fehri 1993). The generation of the possessor inside the NP/np extension parallels that of the generation of the

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

126

Abdelkader Fassi Fehri

(54)

DP D

np 2 np 1

r-rajul-i

l-waasi a

NP daar-u

(54a)

DP D daar-u i

,

FP r-rajul-i k

np 2

ek

np 1 .

l-waasi a j

NP ei

Divergences, however, become clear when authors address various problems, among which are the following: (g) (h) (i) (j) (k) (l) (m) (n)

how is Def inheritance (as described in (f) ) derived? what is the motivation of possessor movement? what is the landing site of the possessor? what is the motivation for N raising? what is the exact landing site of N raising? how is the restriction in (e) accounted for? how is Genitive assigned/checked? is there a link between Genitive assignment/checking and N-to-D raising, and what is its nature?

subject inside VP/vp. The positioning of some `high' adjectives (e.g. modal adjectives) after the possessor in constructions like (65) below might be taken as evidence that the possessor has raised, since it is lower than those adjectives at the base (see structure (66) ). # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

127

4.1. Def inheritance and strength of Def Consider the question raised in (g), in connection with property (f). Suppose that the possessor in (53) has raised from the position in which it was originally generated to a higher Spec of a functional projection, and that the head N has raised higher, as is currently assumed. It is reasonable to think that in SGs with property (f), the possessor is placed in Spec D, because the head D has a (strong) Def feature to check, assuming a feature attraction theory of movement, aÁ la Chomsky (1995). The head N (which moves to D) `inherits' the Def value of the possessor, in a Spechead DP configuration. Consequently, Def inheritance obtains (naturally) in a DP domain. Suppose that the possessor in (53) is located in Spec D1 since it targets Def (a feature of D1), if APs are located in Spec d1 (which contains Agr, as argued above). Then N must land higher than DP1. Standard analyses of SG assume that D is the highest functional projection of DP (to which N is adjoined by virtue of being initial (property (a) ), and they also assume that the possessor surfaces in a Spec AGR, which is lower than D. Consequently, they take Def inheritance to occur in the AGR domain. But this localization of Def inheritance (as `parasitic' on AGR) does not appear to be motivated.34 For example, the possessor and possessee agree only in Def, and do not agree in Case or phi-features, thus suggesting that these features are split (or fissioned), and checked in autonomous domains, a fact which is corroborated by the difference in the content of agreement between adjectives and nouns. If the possessor has moved to Spec D1, targeting a strong Def feature, then N moves presumably to a higher D2, being attracted by a feature which remains to be determined, but which is different from Def. If D1 has a strong Def feature, then it can be checked either via realization as an article, or via attraction of a possessor DP to its Spec.35 Both situations are found. But note that when D1 is realized as an article, the possessor cannot move to its Spec anymore, because that movement is no longer motivated. The complementary distribution between possessors and realized articles in the same DP domain (property (e) ), is then accounted for by the non-availability of a derivational source for a DN Poss structure like that in (55):

34 See Siloni (1994) and Longobardi (1996) for such a proposal. Longobardi assumes that Def checking occurs in D, although Def inheritance takes place in the AGR domain. Borer (1996), on the other hand, assumes that a feature percolation mechanism is needed for Def matching. For criticism of these approaches, see section 6. 35 I am following here ideas by Holmberg & Sandstrom (1996 = HS henceforth), Delsing (1993), and Holmberg (1993). HS assume the following checking principle (their (11b) ): (i) A strong feature of a functional category F is checked off if a phonetically licit category with a matching feature is adjoined to F, or placed in Spec F.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

128

Abdelkader Fassi Fehri

(55) *d-daar-u -r-rajul-i the-house-nom the-man-gen However, in order to rule out (55) completely, as a potential surface output, we must also rule out another derivational source, namely that in which the noun carrying the article is positioned in some higher D, and the possessor is placed in some relatively lower DP projection. The latter option is available for adjectival SG (as we will see in section 5), and it is also available for nominal possessives in languages like Icelandic, for example (see Holmberg & Sandstrom 1996 and Tarald 1990), or Greek (see Alexiadou & Wilder 1977 and Androutsopoulou 1995) ). Yet this option is excluded for Arabic nominal SGs (for reasons given in section 6). 4.2. N to D2 and strong Case If both Def inheritance and Def checking take place in the DP1 domain, then we have to look for the identity of the head targeted by N. Current analyses motivate N to D movement by a Def feature property.36 But if Def specification is a property of D1 (not D2, in which N lands), then that motivation cannot be maintained. An option that suggests itself is that N is moving to the upper D3 for Case reasons. It is reasonable to think that N raises to a K head, where K is a case head category, as in Bittner & Hale (1996a, b), and Lamontagne & Travis (1987). For conceptual reasons, however, I will explore a notational variant of this proposal, in which N raises to a D head, which contains a Case attracting feature. Adopting an idea put forth by Holmberg & Sandstrom (1996), let us assume that D has a Case feature which may vary in strength, and that the strength can be exploited not only in clausal structure, but also in the internal NP structure. Suppose then that Arabic D2 has a strong Case feature, which attracts a Case-bearing N, to check its feature. For Case purposes, N-D2 movement occurs, and the case feature on N is checked against the feature on D2.37 Assuming then a split or fissioned D (with a number of segments), an expanded inflectional structure for a construction (53) is something like (56): 36

The affixal nature of D and its Def content have been advocated to motivate the movement (see Fassi Fehri 1987, Ritter 1987, Mohammad 1988, and Ouhalla 1988, among others). Longobardi (1996) advocates the strength of N features in D. 37 HS contrast basically Scandinavian languages that allow N-Poss constructions (like Icelandic) with those that do not (like Norwegian and other Mainland Scandinavian), and attribute the variation to the strength of Case in D. They also discuss the problem of N-DPoss, and take the variation (attributed to the strength of Def in D) to a minor parameter. In their system, Case and Def conflate under a single segment category D, while I take the two features to be located into separate categories. The strength of the Case feature on D correlates with the morphological `richness' of case on N (dragged by D), and the absence of a free article (for such a correlation, see HS 1996; see also Giusti 1995). # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures (56)

129

DP2 D daar-u i

DP1 r-rajul-i k

dp

l-waasi a j

np 2

ek

np 1 ej

ei

In this configuration, N has raised to D2, and the possessor to Spec D1. The possessor in Spec D1 transmits its Def feature value to D1 (containing the head moving N) via Spec-Head, and that feature in turn is carried along to D2. The head D2 of DP2 is then equipped with the relevant value of the Def feature for appropriate interpretation. The small dp is introduced to host the moved AP, in line with the ESD analysis argued for above. The fissioned D hypothesis receives further support when the structure of SGs is examined more closely. It thus turns out that Def inheritance does not necessarily take place in all SG cases, contrary to the widely spread view embodied in the property (f) above. Such considerations are examined in the next subsection. 4.3. Definiteness in SG and N movement Definiteness appeals to two distinct notions: one has to do with individual or unique reference, and the other with familiarity.38 It can be noticed (after Holmberg 1993) that genitive constructions without adjective modifiers do not induce individual reference, and they can function as predicates. This observation is illustrated by the following Arabic examples: (57) a. haadaa this Å `This is b. haadaa this Å `This is 38

?ax-ii brother-mine my brother.' bayt-u r-rajul-i house-nom the-man-gen the man's house.'

See Heim (1982) and Renaud (1996), among others.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

130

Abdelkader Fassi Fehri

(58) a. haadaa ?ax-ii wa haadaa (?ayd-an) ?ax-ii Ç this Å brother-mine and this Å (also) brother-mine `This is my brother and this is (also) my brother.' b. haadaa bayt-u r-rajul-i wa-haadaa (?ayd-an) Ç this Å house-nom the-man-gen and-thisÅ (also) bayt-u r-rajul-i house-nom the-man-gen `This is the man's house and this is (also) the man's house.' The use of SGs as predicates in (57) and (58) indicates that the nominal projection there is not a saturated DP, assuming that predication requires an open constituent (as in Higginbotham 1985 and Rothstein 1983). On the other hand, the fact that (58) is well-formed indicates that individual reference is not necessarily obtained in genitive constructions. Suppose we interpret the grammaticality of (58) as follows. If the head N there is not definite (because it has no individual reference), then the possessor could not have raised to Spec D1 (if Def inheritance is an automatic consequence of that raising). N, which raises to D2 (for case reasons), does not then inherit Def from any intermediate D (which agrees with the possessor in the Def value). If N is in D in (57)±(58), although it is not definite, and N is also in D in (53), although it is definite, then the two possessors cannot be located in the same Spec position, and the two Ns cannot be located in the same D chains. If N in (53) is in (a high) D2, and the possessor in Spec D1, then the DP projection is a saturated category.39 In (57)±(58), however, the open character of the phrase (and the absence of individual reference) suggests that the chain of Ds in which N is contained has no Def feature value, and that possessor raising has not occurred there. In other words, predicate (or indefinite) nominals have `less' features and trigger less movement than do argumental nominals. This can be executed by taking predicates/indefinites to project no Def-DP (or DP1) for SGs. Predicative adjectives appear to behave exactly like indefinite nominal possessives in (57)±(58). Adjectives move over adverbs (which modify them) and their genitive subject, as illustrated in (59): (59) danan-tu r-rajul-a tawiil-a l-qaamat-i jidd-an ÅÇ thought-I the-man-acc Çtall-acc the-height-gen lot-acc `I thought the man (to be of) a very tall height.' In (59), the adjective (which is indefinite) moves presumably to D, to check its Case and Def features, from an original AP head position.40 The 39 The saturation obtains presumably via possessor raising and Def activation along the lines discussed in section 6. 40 I am assuming that adjectives do not differ from nouns in receiving Case and Def specification, and that they are also DPs. See section 5 below.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

131

entire predicate phrase carries accusative case, but its internal subject receives Genitive. The adjective is not definite, and has inherited no Def from its genitive complement. If it were to function as a modifier of a definite noun, then it must bear a definite article, as (60) illustrates: (60) ra?ay-tu r-rajul-a t-tawiil-a l-qaamat-i Ç saw-I the-man-acc Çthe-tall-acc the-height-gen `I saw the man with a tall height.' Thus in the case of predicative adjectives, no Def inheritance occurs either, presumably because the possessor DP has not raised to Spec D (or more precisely to Spec D1). I will return to the structure of modifying adjectives like those involved in (60) and (53) in section 5. 4.4. Partitives More evidence for the absence of Def inheritance in SG structures comes from partitive SG constructions. The latter are exemplified in (61)±(62): (61) ?ahad-u r-rijaal-i raja¹ a Ç one-nom the-men-gen came.back `One of the men came back.' (62) kull-u r-rijaal-i raja¹ uu all-nom the-men-gen came.back `All (of) the men came back.' These constructions have presumably the same structure as possessive SGs, as proposed by traditional Arabic grammars. However, in addition to differing from (true) possessives in interpretation, partitives differ from the latter in some (apparently) puzzling and interesting properties. First, the head of the partitive SG in the examples mentioned cannot be modified. Only the complement can, as the following contrast indicates: (63) a. *?ahad-u r-rijaal-i t-tawiil-u raja¹ a Ç Ç one-nom the-men-gen Çthe-tall-nom came.back Intended to mean: `one tall of the men came back.' b. ?ahad-u r-rijaal-i t-tiwaali raja¹ a Ç Ç one-nom the-men-gen Çthe-tall.pl.-gen came.back `One of the tall men came back.' This property is strikingly characteristic of partitive SGs, which differ in this respect from possessives (as we have seen above). It suggests that the partitive head and its Spec cannot raise in the structure as high as those of (modified) possessives. If this is true, then there is no reason to think that the head of the partitive SG has ever inherited Def from its `possessor'. This expectation appears to be borne out. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

132

Abdelkader Fassi Fehri

Consider again (61). An analytic partitive equivalent to this construction with an indefinite head is possible, but a definite head equivalent is ungrammatical: (64) a. ?ahad-un mina r-rijaal-i raja¹ a Ç one-nom of the-men-gen came.back `One of the men came back.' b. *l-?ahad-u mina r-rijaal-i raja¹ a Ç the-one-nom of the-men-gen came.back The ungrammaticality of (64b) suggests that no derivational source with a definite head is available for the partitive in this case. Furthermore, the equivalence of (64a) and (61) casts serious doubt on the definite nature of the head in the latter case. It is then reasonable to think that no Def inheritance takes place in partitive SGs.41 To sum up, I have shown that synthetic possessive and partitive structures involve split DP projections, and they differ in interpretation depending on where the head N and the possessor surface. In the next subsection, I address the question of how possessors and APs land in the right positions, without mismatches of movement or violations of minimality requirements. 4.5. Possessor and AP raisings without minimality violations Consider how the possessor moves. Assuming that it is generated higher than some `lower' APs (e.g. manner), and lower than some `higher' APs (e.g. speaker-oriented) in the base structure, although it surfaces higher than all APs, then this ordering can be explained if we assume (as explained above) that Possessor movement is targeting Spec D1 (in which the Def feature is checked). Furthermore, the latter Spec is higher than any Spec which is targeted by any AP, since APs land in Spec d. Typically, even when crossing is involved (e.g. when the possessor is `sandwiched' between lower and higher APs), the outcome of the movement is uniformly the same, and the possessor must be placed higher than all APs. Consider a complex SG involving more than one modification, as in (65), and its plausible base structure (66): 41

There are SG partitives which have analytic counterparts, like the following: (i) tulut-u r-rijaal-i Åthird-nom Å the-men-gen `The third of the men' (ii) t-tulut-u mina r-rijaal-i ÅÅ Å third-nom of the-men-gen `The third of the men' These contrasts, as well as the question why Definiteness is forced in some SG partitives, need further research. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

133

(65) hujuum-u ?amiriikaa sÏ-sÏadiid-u l-muhtamal-u ¹ alaa Ç attack-nom America the-violent-nom the-probable-nom on l-muqaawamat-i the-resistance `The probable violent attack of the resistance by the U.S.A.' (66)

np 3 l-muhtamal .

np 2

?amirikaa

np 1

ˇ ˇ s-sadiid

NP

N hujuum

PP alaa

l-muqaawama

If the possessor moves directly to Spec D1, then the operation appears to be violating Relativized Minimality. One way out is to allow movement through equidistance, and move first the APs (starting from the highest), until the dp cycle is completed. Then the possessor moves in one step to Spec DP1, jumping over only the chains formed between APs and Specs of dp. The resulting structure is as exhibited in (67):

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

134

Abdelkader Fassi Fehri

(67)

DP2 D

DP1

hujuumii ?amirikaak

dp 2

ˇ ˇ s-sadiid j

dp 1

l-muhtamal l .

np 3

el

np 2 ek

np 1 ej

NP N

PP

ei alaa l-muqaawama In this analysis, Possessor and APs raise independently, through Specs which involve different features (Def and Agr, respectively). The movement then no longer violates locality restrictions on movement, once chain formation and feature targeting are interconnected. A potential alternative to the one-step direct movement of Possessor (to Spec D1) is to move it first to the Spec of the lowest AP. The latter which moves further up pied pipes it to the left of the highest AP, and so on. When AP movement is completed, the possessor then `escapes' to Spec D1. This looks like a simpler analysis at first glance, but it is not so. In fact, it faces a serious problem: if the possessor raises through Spec AP, why does not it agree in Case and Def with the head noun, just like other APs do? In order to avoid potential mismatches of agreement, I have relied on the idea that the series of APs form a sort of a single clustered Spec, which can be jumped over by the possessor, without inducing locality violations, and that any AP would be `close' enough to the Spec it is moving to. A further alternative to moving N, AP, and Possessor separately, as I have done, is to first move all the complements to the left of the NP, # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

135

then move the (whole) NP (except the complements) to the left of the lower AP, and proceed with a Cinquean XP type movement to obtain MIO. As far as I can tell, however, this approach does not make the movement depend on the properties of the various NP constituents in languages. As we will see in section 6, crosslinguistic variation suggests that these processes do occur separately, in lines with what happens in Arabic. Note, finally, that object denoting nominals do not differ from eventive ones in that the same processes apply with multiple AP structures, along the lines suggested above. 5. Adjectives as DPs 5.1. The structure of prenominal adjectives As observed earlier in subsection 2.4 above, prenominal adjective constructions behave in essential ways like nominal SGs. Moreover, they are either interpreted like partitives or like normal postnominal adjective constructions. It is then reasonable to think that the adjective and its complement originate as head and Spec of AP, respectively, and they undergo essentially the same operations which apply to the two constituents of a nominal SG in the DP domain. Thus an `adjectival' possessive in (35) above, repeated here as (68), would have a basic structure like (69): (68) ?akal-tu ladiid-a Å Å ate-I delicious-acc `I ate the delicious (of

t-ta¹ aam-i Çthe-food-gen Ç the) food.'

(69)

IP/ap t-ta . . aam

ladiid

The adjective ladiid, being a head, incorporates first into D1, and then Å Å into D2, just like a `normal' N in a nominal SG does. The two-step movement is motivated by the split/fissioned functions of D, Def and Case, respectively. The NP t-ta¹aam moves through Specs, receiving Ç Spec DP , to check Def in D . Def Genitive case, and surfacingÇ in 1 1 inheritance is then expected. The surface structure is then as follows:

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

136

Abdelkader Fassi Fehri

(70)

DP2 D

DP1

ladiid i t-ta . . aamj

D' IP/ap

D1 ei

ej

ei

If this analysis is correct, then adjectives can also inherit Def from D, and get somehow `nominalized' by D. Note that the (Def) D on the adjective can be taken as `referential', inducing typically individuation, just as a nominal D does. No expletive interpretation is workable here. Further clarification on article properties are provided in the next subsection. 5.2. The adjectival article and its origin Arabic appears to differ significantly from languages like English or French in that both the head noun and the modifying adjective carry (definite) articles, as the examples above show. This difference, I will claim, is only superficial, and APs need to be headed by articles, just like NPs are. If adjectives are DPs, then we expect them to behave like nominal DPs in essential aspects. On the other hand, differences between the two constructions have to be addressed. In line with Szabolcsi (1987, 1994), Kayne (1994), and Smith (1964), let us assume that D and C are parallel, and that D (even in ordinary noun phrases) is a clausal determiner. Kayne (1994) has argued (on the basis of a number of distributional contrasts within Germanic and Romance noun phrase structures) that adjectives, relative clauses, and possessives are basically IP/CP complements (of D). His analysis stemmed from the essential observation that the D-N sequence in noun phrases does not form a constituent. Therefore all phrases (including N, and except D) are generated inside IP/CP (which is a complement of D), and then raised (including the `head' N of the relative clause). If the adjective construction is a reduced relative or CP, and the article is a clausal determiner, then it is conceivable that the sources of the two determiners are both clausal, as illustrated in (71a), the base structure of (71):42 (71) r-rajul-u t-tawiil-u Ç the-man-nom Çthe-tall-nom `The tall man' 42

At S-structure, the adjective incorporates into C/D, and the head noun into D.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures (71a)

137

DP CP/DP

D [l-]

C/D

IP r-rajul

tawiil .

In Kayne's clausal system, however, nothing forces this `double DP' structure, and nothing ensures Def agreement between the two DPs.43 But suppose, in line with Kayne's system, that the base structure of attributive adjective modification is essentially (71a), where two D heads and two Spec Ds must be involved (although they may form only segments of the same category, along the lines to be explained below). If articles are morphologically prefixed on nouns as well as on adjectives in the base, then the syntactic distribution of these two categories will depend on the nature of their inflectional and interpretational properties. The needed derived structure is then (72): (72)

DP D

dp

r-rajulj t-tawiil .. i ei

np NP ej

43 Alexiadou & Wilder (1997), who also assume that the origin of Def is a syntactic D, claim that Kayne's treatment of adjectives as reduced relatives make available (at least) two determiner sources: one is the determiner of the external DP (which contains the relative clause), and the other is the determiner of the DP subject of the AP, contained in CP (see their structures (45) and (46) ). But to my knowledge, Kayne's analysis does not allow such interpretation. See e.g. his analysis of `le livre jaune' on p. 101 (with no article on the DP which is internal to CP). See also his structure (57) on p. 97 for `the book sent to me', and his assumption that `book' there receives Case through incorporation with `the' (p. 98). As far as I understand it, Kayne postulates only one D in (headed) relative clauses, which is equated with C, and the nominal head of the relative clause, which originates in the internal IP without determiner, raises to the domain of the external D/C. The only way then to ensure the existence of two Ds (and their agreement) is to assume that in (headed) relative clauses two base generated DPs (one internal to IP, and one external to it) come to interact in various ways at PF.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

138

Abdelkader Fassi Fehri

Let us now assume that the `small clause' in this structure is of an identificational type, rather than a predicational type. In the former case, both the subject and the predicate of the copulative (or nominal) clause are definite (and hence saturated categories), and one saturated category is `identified' with the other, rather than `predicated' of it (predicates being open, as already explained). The two (small clause) nominal sentences are given in (73) and (74), respectively: (73) ?? r-rajul-u (huwa) t-tawiil-u Çthe-tall-nom Ç the-man-nom (he) `The man is the tall.' (74) r-rajul-u tawiil-un the-man-nom Çtall-nom `The man is tall.' Although (73) appears at first glance to be bad as an identificational sentence, it is not so with appropriate intonation break. Insertion of inflectional huwa makes it better. If attributive adjectives are derived from a predicate position, along Kayne's proposal, and if nominal clauses (in which they originate) are either identificational (definite) or predicational (indefinite), then we expect both types to be relevant for building modification structures. If attributive APs are then generated as predicates, possibly in Spec positions, as a sort of `inverse' identificational clauses, in lines with the LSH, then we expect them to surface with articles. Moreover, the agreement in Def between adjectives and nouns can be thought of as resulting from embedding the two categories in a single functional DP domain, in which the same values of features must be matched.44 The identificational hypothesis is further corroborated by the behaviour of demonstratives. Thus like (73) and (74), the construction in (75) is ambiguous between a noun phrase reading and a sentential reading (intonation apart): (75) a. haadaa (huwa) l-walad-u this Å (he) the-child `This (is) the child.' b. haadaa (huwa) l-mas?uul-u this Å (he) the-responsible `This (is) the responsible.' 44 One consequence of this view is that the attributive AP can be found in an `inverse' structural position of that of the predicate, i.e. as a Spec, rather than a head, of the predication, a well-known property of identificational sentences (see Moro 1991 and Milsark 1976, among others). This appears to be in fact the system proposed by Cinque (1996), in which APs are Specs of NPs, a proposal which appears to be in contradiction to Kayne's, although it is not, if APs are interpreted as predicates.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

139

It is unreasonable to think that the N bearing the article in the construction (75a), when interpreted as a noun phrase, behaves like the `nominalized' adjective in (75b). In both sentence and noun phrase cases (i.e. as predicates or attributes), the definite phrases have an identificational (predicative) origin. It is striking that although the interpretation in these cases is always definite, English and French dispense with the definite article, an option which is impossible in Arabic: (76) this (*the) man (77) ce (*le) jardin This optionality amounts to basically having an optionally realized modifying definite article. In Arabic, however, the definite modifying article must be realized.45 5.3. Modifying and non-modifying D We have seen that Arabic articles can have two different uses: as heads of non-modifying saturated categories (whether nominal or adjectival), or heads of modifying non-saturated categories (be they nominal or adjectival). Let us call the first use a referential use, and the second an anaphoric use. If D is anaphoric, then we expect it to be bound by another D, in a minimal domain. This expectation is borne out, and the two Ds on the noun and the adjective are coindexed. Agreement in features are presumably a consequence of coindexation.46 The `scope' ordering of the two Ds (attached to N and A) can be naturally related to the interpretation of the DP. Suppose that in order 45 I assume e.g. that the derived structure of a demonstrative construction like (75a) is the following: (i) DP

D

haadaa i

,

D l-walad j e i

IP ej

Other identificational/predicational source alternations include the following pairs: (ii) all men (iii) all the men (iv) John is the president (v) John is president Note, by way of comparison, that adjectives (normally) precede nouns in Germanic and (some) Romance, and adjectives precede possessors in Celtic. In these languages, however, adjectives do not carry articles. This situation contrasts with that found in languages like e.g. Greek, where the adjective carrying the article must (normally) follow the noun. 46 Scope is presumably conditioned by c-command, which then motivates that ordering. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

140

Abdelkader Fassi Fehri

to be interpreted as an argument, a DP has to be saturated or closed off. A DP is normally saturated if its Spec contains a realized DP possessor which closes off the open position in the DP, or an empty pronominal (licensed by an article) which has the same function (see e.g. Higginbotham 1985, Holmberg 1993, and Campbell 1993, 1996, among others). In the case of modifying adjectives, the open position in AP is first identified with the variable in NP, and cannot be closed off by its D. Otherwise, the result will be non-interpretable. The anaphoric nature of D on adjectives follows. When D is anaphoric, then it has to be c-commanded by a referential or pronominal D, and it is coindexed with it. Likewise, a similar function can be performed by the possessor, which is placed in the domain of the higher D, presumably for scope reasons.47 5.4. Crosslinguistic evidence I have assumed, following Szabolcsi (1987, 1994), Abney (1987), Kayne (1994), and Smith (1964/1969), among others, that the article is essentially a clausal determiner, which can come to be affixed on nouns and adjectives alike (be they predicates or arguments), and that adjective modification (and/or prediction) licenses two determiner sources. This view has been supported by various modification and predication phenomena in Standard Arabic, but similar phenomena are in fact widely spread cross-linguistically. For example, Greek, Rumanian, Scandinavian, and Cushitic express articles on modifying adjectives (see Alexiadou & Wilder 1997, Dobrovie-Sorin 1987, Holmberg & Sandstrom 1996 and Delsing 1993, and Lecarme 1996, respectively). In French, double determiners occur with superlatives, as in (78), or epithets, as in (79): 47 Among the questions which remains to be addressed is how to account for the fixed ordering of relative clauses (whether restrictive, as in (i)±(ii), or non-restrictive, as in (iii) ). The latter are located after all simple adjectives and modifiers: (i) l-kitaab-u s-sagÇiir-u lladii qara?-tu-hu ÇÇ Å the-book-nom the-little-nom that read-I-it `The little book that I read' (ii) l-kutub-u t-talaatat-u llatii ra?ay-tu-haa ÅÅ Å the-books-nom the-three-nom that read-I-them `The three books that I read' (iii) kitaab-u l-¹aqqad-i haadaa lladii qara?-tu-hu Å read-I-it book-nom al-Aqqad-gen this Å that `This book of al-Aqqad that I read' Orders in which the relative clause precedes modifiers are ill-formed: (iv) *l-kitaab-u lladii qara?-tu-hu s-sagÇiir-u ÇÇ Å read-I-it the-book-nom that the-little-nom (v) *l-kutub-u llatii ra?ay-tu-haa t-talaatat-u ÅÅ Å the-books-nom that read-I-them the-three-nom (vi) *kitaab-u l-¹aqqad-i lladii qara?-tu-hu haadaa Å read-I-it book-nom al-Aqqad-gen that this Å See Fassi Fehri (1997a, b) for suggestive solutions.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

141

(78) le repas le plus deÂlicieux the meal the most delicious `the most delicious meal' (79) l'imbeÂcile de Pierre the idiot of Pierre `that idiot Pierre' In Moroccan Arabic, there is also a Def agreement between the head noun and the adjective or the relative marker heading the relative clause, as is illustrated by the following examples: (80) a. r-rajel t-twiil Ç the-man Çthe-tall `The tall man' b. r-rajel lli jaa the-man that came `The man who came' (81) a. rajel twiil man Çtall `A tall man' b. rajel jaa man came `A man who came' Furthermore, possessive structures are also used as epithets (which involve double determiners), hence the ambiguity of a construction like (82): (82) l-kelb d-hmed Ç the-dog of-Hmed a. `Hmed's dog' b. `that jerk Hmed' These alternative uses of possessives (as nominal expressions or as attributes) become less surprising or accidental once adjectives are taken to be DPs.

6. Consequences and further discussions 6.1. Def inheritance and SGN structure Observing that articles occur on Hebrew adjectives and demonstratives, and assuming that such occurrences do not correlate with semantic values, Borer (1988, 1996) claims that Def is a (base-generated) feature on N in Semitic, and that D has no Def specification. Def is inherited by # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

142

Abdelkader Fassi Fehri

D only from a moved N which is specified for Def.48 Borer also assumes that the salient feature of the nominal construct state (= our SGN) is that its N head is generated without Def specification. Since both D (in general) and (the head) N in SGN have no Def specification, syntactic incorporation (or merger) of N-to-N is required, where the second N (heading the complement) has Def specification, and it transmits its feature to the head N via a complex mechanism of percolation. Then the N-N complex substitutes for D, to make it (in)definite.49 In addition to appealing to a very complex route to ensure Def inheritance, Borer's analysis embodies a number of claims that are questionable: (a) the claim that the article is not a functional affix (and/or a D head, or that D is empty), but only a lexical feature of N; (b) the claim that adjectives do not need articles; the claim that the head N of the SGN and that of its complement form a unique head at Sstructure; (c) the obligatory Def inheritance in SGN, since the only way for the head N of SGN to inherit a Def value is via incorporation with the head N of the complement (which is marked for Def). If (a) and (c) were true, then we would expect the (definite) SGN and the N carrying the article to have basically the same distribution. This expectation is not borne out. As a matter of fact, there is evidence that the two phrases exhibit different distributional behaviours. For example, prenominal demonstratives cooccur only with head nouns on which a definite article is realized, but they cannot be constructed with a (definite) SGN. The latter is treated like bare common nouns (with no article), as in (83a), or bare proper names, as in (83b), which cannot cooccur with a prenominal demonstrative: (83) a. *haadaa Å this-nom b. *haadaa Å this-nom c. *haadaa Å this-nom Intended

rajul-un man-nom zayd-un Zayd-nom bayt-u r-rajul-i house-nom the-man-gen to mean: `this house of the man'

Likewise, demonstratives can precede only nouns bearing articles (whether common or proper): 48 See Borer (1996), p. 53. The author resorts to morphological `secondary percolation' and `feature sharing' mechanisms (see p. 56). I have already called into question the morphological nature of Def inheritance. 49 In contrast to Semitic, English (and more generally Germanic) has a D specified for the Def feature, and hence no CSN is possible. For a different view, see Longobardi (1994, 1996).

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

143

(84) a. haadaa r-rajul-u Å this-nom the-man-nom `This man' b. haadaa l-?azraq-u Å this-nom the-Azraq-nom (blue) `This Azraq' The contrast between (83c) and (84a) suggests that two different syntactic classes of DPs can be distinguished, depending on the presence or absence of a realized D (and/or article). This distributional distinction cuts across the proper/common noun distinction.50 A similar behaviour is observed in vocative constructions. In the latter, SGNs, like bare proper names or bare common nouns, can be used, but nouns with definite articles cannot: (85) a. yaa rajul-u Hey man-nom b. yaa ?azraq-u Hey Azraq-nom c. yaa zayd-u Hey Zayd-nom d. yaa bn-a ?ax-ii Hey son-acc brother-mine `Hey son of my brother' (86) a. *yaa Hey b. *yaa Hey

r-rajul-u the-man-nom l-?azraq-u the-Azraq-nom

It is significant that a proper name loses its article when constructed in the vocative, as the contrast between (85b) and (86b) illustrates. These limitations on demonstrative and vocative constructions indicate that the two distributional classes of constituents involved are differentiated on the basis of whether their head occurs with a realized article or not. If the article is treated as a D head (or realizing the Def feature on D), then these distributional differences are expected. Further evidence for taking the article to be an affix (or realizing/ checking a feature on D), rather than a feature on N (with an empty D) is 50 Note that demonstratives can occur with CSNs or bare proper names, provided they are postnominal: (i) bayt-u r-rajul-i haadaa Å house-nom the-man-gen this-nom `This house of the man' (ii) zayd-un haadaa Å Zayd-nom this-nom `This Zayd'

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

144

Abdelkader Fassi Fehri

its incompatibility with the occurrence of a Genitive possessor.51 If the Def feature is treated as a D head, then the complementary distribution between its realization and that of the genitive possessor can be taken as a complementarity between Head and Spec realization of the same DP category, along the lines suggested above. Consider now the claim that the article on the adjective is not an affix (and hence only a manifestation of a sort of morphological concord; as embodied in (b) ). As we have seen above, the distribution of articles and their interpretation are regulated by syntactic considerations, be they on adjectives or on nouns. For example, the article occurs with postnominal adjectives, but is physically absent on prenominal ones. The same behaviour is observed with numerals, which may or may not bear articles, depending on whether they are postnominal or prenominal. Quantifiers also behave along the same lines (in essential respects). Adjectives, numerals, and quantifiers do not differ therefore from common nouns with respect to article distributions. Furthermore, the interpretation of articles on nouns does not differ from that on adjectives, once we take into consideration the syntactically driven distinction between referential and anaphoric uses of articles in both cases. Finally, let us turn to the correlation of N-N incorporation and Def inheritance (as postulated in (c) and (d) above). As shown in Fassi Fehri (1993), the (phonologically strict) adjacency of the head noun and the possessor is not observed in Arabic. Likewise, adjectives intervene between head nouns and possessors in Celtic, as will be explained below, in subsection 6.3. Moreover, if what has been said in subsection 4.3 is correct, then Def inheritance is not forced in all SGN cases. Therefore, these observations provide reasonable grounds for treating the article as realizing/checking the D head, and taking Def inheritance (when it occurs) to be a structural effect of the higher positions in which the head N and the possessor are found. 6.2. Genitive case Let us start with the idea that Genitive is a structural case, which is assigned/checked under a Spec-Head relation by a functional head (as in Fassi Fehri 1993, Siloni 1994, Ouhalla 1996, and Longobardi 1996, among others). If Nom(inative) and Gen(itive) are structurally parallel, then if Nom is checked in Spec TP (in a system such as Chomsky's 1995 in which AGR is not part of the categorial inventory), then Gen is presumably checked in Spec DP (as has been argued for in Fassi Fehri 1993), and not in Spec of AGR (as proposed by Siloni, Ouhalla, and Longobardi 1996, among others). In this view, DP (or one of its 51 As is known, this incompatibility is not true in all languages (see e.g. Rumanian, Somali, Hungarian, Icelandic, Italian, among other languages).

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

145

segments) is parallel to TP, in that the two categories are endowed with a D feature, through which Cases of external arguments (i.e. possessors and subjects) are checked. Note that since parallels between DP and CP are also widely established in the literature, this further strengthens the need for a layerly fissioned DP, which would then be the equivalent counterpart of the TP/CP extension. If Gen is checked in Spec DP, then a number of questions arise: (a) At which level does the checking take place, is it at PF or at LF? (b) Where is the Gen DP located, in view of other (segments of) DPs? Concerning question (a), there are enough reasons to think that Gen is checked by PF. We have seen above that SGs like (20a) differ from analytic genitives like (21a) in that overt Possessor raising has occurred in the latter, but not in the former, and that the placement of APs depends on whether the mentioned process takes place or not. It is then reasonable to think that, at least in these SGs, the possessor raises. But up till now, we have assumed that Possessor raising is triggered by Def, typically in cases where Def inheritance occurs. There are, however, instances of adjectival as well as nominal SGs, in which there is no Def inheritance, yet the possessor could be argued to have moved. Consider the following event nominal SGs: (87) nasÏr-u l-xabar-i ¹ amd-an l-yawm-a xata?-un Ç publication-nom the-news-gen deliberately today mistake-nom `The publication of the news deliberately today is a mistake.' (88) nasÏr-u xabar-in kaadib-in Å publication-nom the-news-gen false-gen stigÇlaali-hi xata?-un Ç exploiting-it mistake-nom `The publication of a false news in order mistake.'

tama¹ an fii Çpurposing in to exploit that is a

(89) nasÏr-u s-sahiifat-i l-xabar-a ¹ amd-an ÇÇ Ç publication-nom the-newspaper-gen the-news-acc deliberately xata?-un Ç mistake-nom `The deliberate publication of the news by the newspaper is a mistake.' In these constructions, the possessor has moved over all adverbs, including subject-oriented adverbs, which are presumably generated higher than the subject/possessor at the base structure. On the other hand, there is no reason to think that the eventive head N has ever inherited any Def feature from the possessor, since it is impossible to modify this head by nominal modifiers such as adjectives or relative clauses: # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

146

Abdelkader Fassi Fehri

(90) *?idaa¹ at-u l-xabar-i l-mutasarri¹ at-u ¹ amd-an Å broadcasting.f.-nom the-news-gen the-hasty.f.-nom deliberately xata?-un Ç mistake-nom Intended to mean: `the hasty broadcasting of the news deliberately is a mistake.' The absence of evidence for Def inheritance, the presence of adverb modification, in addition to case alternations of the understood NP object l-xabar in (87) and (89) suggest that the movement process takes place there for Case reasons, to check Gen, and not to check Def. Assuming then that Gen checking occurs at PF in Spec DP, triggered by a D feature of D (namely Gen), the next question to address is where Gen DP is located, in particular with regard to Def DP (in which Def agreement is checked), or Agr DP (in which APs are located). Our answer must take into account two previous observations: (a) that agreeing adjectives can never precede nouns (i.e. *D-A D-N order is impossible), and (b) that APs can never precede genitive possessors (i.e. *(D-N) A GenPoss order is impossible). Having these two observations in mind, a plausible solution is to assume that Gen is checked at the lowest (segment of) DP projection, where Agr DP for APs is higher. When APs are generated, Possessor has to raise higher (in order to meet restriction (b) ), presumably for referential reasons, which have been already advocated above, to exclude the order in (a). If this is true, then Possessor may or may not undergo double raising (for Gen and/or Def checking). Longobardi (1996) assumes (after Siloni 1994) that the possessor is in AGRGen, and that its case is checked there. Furthermore, N moves to D, to check its [‹ Def] feature, but its movement is also necessary to check Gen (via a sort of head government). As for Def inheritance, it takes place in Spec AGR, but its effect is carried on to N by stipulating that Gen can only be checked if AGRGen, which is moved to D, has inherited a Def feature through Spec Head agreement.52 This `detour' to Def and Case checking in SGs and their linkage do not appear to be motivated. In my system, there is no direct or necessary correlation between agreement in Def and that in Gen, although it is plausible that N to D incorporation is playing a role in licensing the latter, in much the same lines as those 52 To quote: `suppose [that the article feature] . . . in Semitic is strong [. . . then] it has to be checked before Spell Out, triggering overt movement. Suppose also that AgrG may inherit the definiteness value of the genitive argument in its Spec and that the raising of the noun to D takes place in the following way: N first adjoins to AgrG and then the new complex so formed adjoins to (or substitutes for D). Now the condition on the identification of D will be satisfied: the raising of the lexical noun [. . .] will be functionally motivated and licensed by the need to check the [+ article] feature and the presence of a prepositionless Genitive is explained by the fact that this operation can be performed only by dragging to D an AgrG being in a Spec-Head relation with a Genitive argument' (Longobardi 1996, p. 32).

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

147

proposed by Bittner & Hale (1996a, b).53 Thus the dissociation of Gen and Def in adjectival SGs falls out naturally from my system, but not from that of Longobardi. 6.3. A comparison with Welsh I have taken Possessor movement to operate independently from AP movement and N movement, and I have assumed that there is no (whole) NP movement involved, which would move both the possessor and the head N, and would pied pipe the AP, on its way to DP. Two essential reasons are behind this approach. First, objects, complements and relative clauses are left behind N and Possessor, thus suggesting that no (whole) NP movement is involved. Second, there are cases where possessors are left behind, suggesting that N movement is not linked to Possessor movement. Crosslinguistic variation provides motivation in favour of this analysis. Thus English and Arabic stand to the extremes, since English makes use of none of these processes, while Arabic has all of them. In Romance, Spanish uses N movement and A movement, but not Poss movement, but these movements do not appear to reach D (see Bosque & Picallo 1996). In French, only (a short) N movement is used, but no A movement is documented (see Cinque 1994a). Finally, Celtic which is known to be both V-initial and N-initial like Arabic (and more generally Semitic) provides another interesting case for comparison, since N undergoes a long movement to D, but no A movement or Poss movement occurs. As a matter of fact, Welsh, as described by e.g. Rouveret (1994), exhibits strikingly contrastive properties of nominal modifying constructions, when compared to those found in Arabic. First, the serialization of adjectives observes direct ordering prominence, with no MIO, although adjectives are postnominal, as illustrated in (91): (91) y cwpan mawr gwyrdd Sieineaidd the cup big green Chinese `The big green Chinese cup' 53 Bittner & Hale (1996a, b) propose a `Case binding' theory in which Case is assigned by a head which delimits a small clause, and in which Case competition plays an important role. They define Case binding as follows (their (22) ): (i) Case binding Let a be a head that delimits a small clause, and let b be an argument. Then a Case-binds b, and b's head, iff a. a locally c-commands b; b. a governs a Case competitor for b. They also assume the following general conventions for (direct) Case realizations (their (9) ): (ii) If a Case-binds an overt empty-headed KP b, then the empty K of b is realized as a. ERG, if a is I (or D); b. ACC, if a is V (or P) and has an adjoined D. I assume that the incorporated D (in N-AGR-K) qualifies as a (pseudo-argument) Case competitor, and enables K-D to assign Genitive.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

148

Abdelkader Fassi Fehri

The order observed thus suggests that only N to D raising has occurred, and that APs surface in their initial hierarchical order. Second, adjectives are located after the head noun and before the possessor, even in SGs, as (92) indicates: (92) merch bert brenhines ddoeth daughter beautiful queen wise `The beautiful daughter of a wise queen' The order exhibited here suggests that the possessor has not moved from its base position. Third, Welsh adjectives do not have their own (independent) article, as observed in these examples. The order is then D-N-A, a distribution which cannot be found in Arabic, as already explained above. The poorness of inflection of Welsh adjectives then correlates with the absence of A movement. 6.4. Mixed order cases Consider now examples of mixed orders like those in (43) and (44) above, where numerals and quantifiers are prenominal, and attributive adjectives are postnominal. These cases provide instances of `intermediate' derivations, where various nominal constituents (including the head N and the modifying AP) are located (in MIO) to the right of various prenominal constituents (which observe direct hierarchical ordering). They provide further support for the view that N raising, Poss raising, and AP raising are involved in the derivation of Arabic noun phrases.54 As already explained, evidence for Possessor raising is provided by the positioning of the (genitive) possessor, which is found necessarily preceding adjectives. Evidence for A (or AP) raising is provided (among other things) by the fact that adjectives occur postnominally in MIO, although they occur prenominally in the original order. Finally, evidence for N raising is provided by the fact that the possessor is located to its right. In order to maintain the leftness view of postnominal modifiers (or LSH), it is essential that the division between the prenominal space and the postnominal one be strictly hierarchical. Thus if a modifier X is included in the postnominal space (and hence undergoes leftward raising), then any modifier Y which is included in the prenominal space must belong to a higher functional structure (the converse being also true). Thus given a hierarchy like that in (93):55 54 It might be thought that the movement operates in a `pied piping' fashion, according to which the AP is only dragged by the possessor on its way up. But although pied piping can be invoked in the possessor case, it is not generalizable to cases where no possessor occurs, yet the adjective occurs in a mirror image order. 55 I have included D in parentheses to indicate that it does not move, contrary to all other categories.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

149

(93) Q 4 Dem 4 (D) 4 Ord 4 Card 4 A 4 N and assuming that this hierarchy is basic, then orderings which appear to violate this hierarchy are excluded, unless some motivated extra mechanism operates. Given these mechanisms, we predict that a number of orders which have no derivational sources are excluded. Our predictions are borne out, as we will see. A first set of cases concerns prenominal modifiers which do not respect direct hierarchical ordering, e.g. *Num Q (N) in (94a), or *Num Dem (N) in (95a): (94) a. *talaatat-u kull-i kutub-in Å Å three-nom every-gen books-gen b. kull-u talaatat-i kutub-in Åthree-gen Å every-nom books-gen `Every three books' (95) a. *t-talaatat-u haadihi Å Åthe-three-nom Å Å theseÅ b. haadihi t-talaatat-u Åthe-three-nom Å Å theseÅ `These three' A second set concerns postnominal modifiers which do not observe MIO, e.g. the order *N Q A Num in (96a), or *N A Q Num in (96b), when compared with N A Num Q in (96c): (96) a. *l-kutub-u kull-u-haa l-xadra?-u Ç the-books-nom all-nom-them the-green-nom b. *l-kutub-u l-xadra?-u kull-u-haa Ç the-books-nom the-green-nom all-nom-them c. l-kutub-u l-xadra?-u t-talaatat-u Åthe-three-nom Å Å Ç the-books-nom the-green-nom `All the three green books'

t-talaatat-u Åthe-three-nom Å Å t-talaatat-u Åthe-three-nom Å Å kull-u-haa all-nom-them

The same is true of the order *N Dem Num A in (97a), in which the normal hierarchical order is respected, yet the construction is ungrammatical: (97) a. *s-suhuf-u haadihi t-talaat-u l-jadiidat-u ÅÅ Å ÇÇ Ç the-newspapers-nom theseÅ the-three-nom the-new-nom b. s-suhuf-u l-jadiidat-u t-talaat-u haadihi ÅÅ Å ÇÇ Ç the-newspapers-nom the-new-nom the-three-nom theseÅ `These three new newspapers' The construction (97b), with the N-A-Num-Dem order, is the only acceptable postnominal output. A third set of cases concerns mixed orders, which reflect a combination of the two previous sets. Thus Num-N-Dem, A-N-Q, Q-N-A-Num, or # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

150

Abdelkader Fassi Fehri

Dem-N-A-Num in (98)±(101) are not problematic, but Dem N Num A in (102) is:56 (98) t-talaat-u (s-)suhuf-i haadihi Åthe-three-nom Å Å Ç Ç Ç (the-)newspapers-gen theseÅ `These three newspapers' (99) waafir-u s-sabr-i kull-u-hu ÇÇ plentiful-nom the-patience-gen all-nom-it `All the plentiful patience' (100) kull-u l-kutub-i l-xadra?-i t-talaatat-i Åthe-three-nom Å Å Ç all-nom the-books-nom the-green-gen `All the three green books' (101) haadihi s-suhuf-u l-jadiidat-u t-talaat-u ÅÅ Å ÇÇ Ç theseÅ the-newspapers-nom the-new-nom the-three-nom `These three new newspapers' (102) ?? haadihi s-suhuf-u t-talaat-u l-jadiidat-u Åthe-three-nom Å Å ÇÇ Ç theseÅ the-newspapers-nom the new-nom These three new newspapers' In sum, the various ordering restrictions on prenominal and postnominal determiners and modifiers are adequately accounted for in our analysis, assuming LSH and various left branching raising processes, in conformity with Kayne's (1994) AS theory, and multiple checking processes observing minimality restrictions, along the lines of Chomsky (1995). 7. Conclusion In this paper, I have addressed various descriptive and theoretical questions concerning the base structure ingredients of modifying APs, and how these constituents are reordered within larger DP structures. I have argued for the existence of AP (and A) leftward movement, which is motivated by the strength of D features of Arabic APs. The recursive local application of the latter movement has been shown to be behind MIO effects. Furthermore, AP and Possessor movements interact appropriately, without inducing minimality violations. I have reanalyzed the structure of synthetic possessives in the light of the observation that they may or may not involve Def inheritance. The latter has been argued to be syntactic in nature: it is triggered by Possessor raising, in a split/fissioned 56 The judgements with respect to (101) and (102) are very subtle, and some speakers are unable to make a distinction between the two constructions. When confronted with the difference in interpretation, however, speakers tend to correct their judgements in the right direction.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

151

DP architecture. On the other hand, Case, Def, and Agr features play separate roles with respect to Attract, in autonomous DP domains. Possessive and partitive structures have been examined and contrasted, and the relevance of their structure to that of the treatment of prenominal adjective constructions explored. Finally, Gen checking has been dissociated from Def checking in SGs, thus providing additional motivation for a fissioned DP approach. Further research is obviously needed to detail and refine this picture.57 References Abney, S. 1987. The English noun phrase in its sentential aspect. Ph.D. MIT. Alexiadou, A. & Wider, C. 1997. Adjectival modification and multiple determiners. Ms. Berlin: FAS. Androutsopoulou, A. 1995. The licensing of Adjectival Modification. WCCFL 14, 17±31. Astarabaadii, R. 12th century. SÏarh asÏ-SÏaafiya. Beyruth: daar al-kutub lÇ ¹ ilmiyyah, 1975. Bernstein, J. 1993. Topics in the syntax of nominal structure across Romance. Ph.D. New York: CUNY. Bittner, M. & Hale, K. 1996a. The structural determination of Case and Agreement. Linguistic Inquiry 27, 1±68. Bittner, M. & Hale, K. 1996b. Ergativity: toward a theory of a heterogeneous class. Linguistic Inquiry 27, 531±604. Bloomfield, L. 1933. Language. New York: Holt. Bolinger, D. 1967. Adjectives in English: attribution and predication. Lingua 18, 1±34. Borer, H. 1988. Morphological parallelism: a case study. Morphology yearbook, ed. G. Booij & J. van Marle, 45±65. Dordrecht: Foris. Borer, H. 1996. The construct in review. Studies in Afroasiatic grammar, ed. J. Lecarme, J. Lowenstamm, & U. Shlonsky, 30±61. The Hague: Holland Academic Graphics.

57 Indeed, a number of theoretical and technical questions remains to be solved, in particular with regard to motivation of movement. For example, one might wonder what features are interpretable or non-interpretable in Chomsky's sense, which would then trigger or not trigger movement. I was basically assuming that D has N features (namely Def), and D features (Case and Agr), and that the former are interpretable, whereas the latter are not. If Arabic Case and Agr in D are strong, then they motivate internal movement within DP. What about Def? It is widely accepted in the Arabic and Chamito-Semitic literature that N moves to D (Def), at least in definite SGs. On the other hand, I have shown that N moves to D, even with nouns bearing the prefixed article, as evidenced by the obligatory N-A order. This suggests that Def does also trigger movement, and it may or may not be interpretable. In the case of the N-A order, it is possible that the movement is motivated by scope (and secondarily by Def checking of adjectives). In the case of SGs which undergo Def inheritance, it is possible that D is not included in the numeration, and it is only created by Possessor movement. The N then picks up the Def feature through agreement with the possessor, since it is necessary for its interpretation. As for APs, their movement is basically motivated by the same factors. Other questions arise with regard to the issue of what counts as (lexically) inherent (such as plurality), or (functionally) non-inherent category features (such as definiteness, presumably). Furthermore, alternative technicalities are equally conceivable, but I leave these matters for further inquiry.

# The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

152

Abdelkader Fassi Fehri

Bosque, I. & Picallo, C. 1996. Postnominal adjectives in Spanish DPs. Journal of Linguistics 32, 349±385. Campbell, R. 1993. The occupants of Spec DP. GLOW Newsletter 30, 62±63. Campbell, R. 1996. Specificity operators in Spec DP. Studia Linguistica 50, 161± 188. Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass: MIT Press. Cinque, G. 1994a. On the evidence for partial N-movement in the romance DP. Paths towards universal grammar. Studies in honor of Richard S. Kayne, ed. G. Cinque et al., 85±110, Washington: Georgetown U. Press. Cinque, G. 1994b. On the relative order of certain `lower' adverbs in Italian and French. Linguistique compareÂe et langues au Maroc, ed. A. Fassi Fehri, 11±48. Rabat: Publications of the Faculty of Letters. Cinque, G. 1995. Adverbs and the Universal Hierarchy of Functional Projections. GLOW Newsletter 34, 14±15. Cinque, G. 1996. The `antisymmetric' programme: theoretical and typological implications. Journal of Linguistics 32, 447±464. Cinque, G. 1997. Adverbs and the Universal Hierarchy of Functional Projections. Ms. U. of Venice. Crisma, P. 1993. On adjective placement in Romance and Germanic event nominals. Rivista di Grammatica Generativa 18, 61±99. Crisma, P. 1995. On the configurational nature of adjectival modification. Grammatical theory and Romance languages, ed. K. Zagona, 59±71. Amsterdam: John Benjamins. Delsing, O. 1993. The internal structure of noun phrases in the Scandinavian languages. U. of Lund: Department of Scandinavian Languages. Dobrovie-Sorin, C. 1987. A propos de la structure du groupe nominal en roumain. Rivista di Grammatica Generativa 12, 123±152. Emonds, J. 1978. The verbal complex V'-V in French. Linguistic Inquiry 9, 151±175. Emonds, J. 1985. Syntactic categories. Dordrecht: Foris. Fassi Fehri, A. 1978. Relatives restrictives en arabe. Ms. U. de Paris VIIIVincennes. Fassi Fehri, A. 1981. Linquistique arabe: forme et interpreÂtation. Rabat: Publications of the Faculty of Letters. Fassi Fehri, A. 1987. Generalized IP structure, Case, Inflection, and VS word order. Proceedings of the first International conference of the Linguistic Society of Morocco, ed. A. Fassi Fehri et al. Vol. 1, 189±221. Rabat: Oukad Publishers. Fassi Fehri, A. 1990. l-binaa? l-muwaazii. Casablanca: Toubqal Publishers. Fassi Fehri, A. 1993. Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer. Fassi Fehri, A. 1996. Distributing features and affixes in Arabic subject verb agreement paradigms. Linguistic Research 1.2, 1±30. Rabat: IERA. Fassi Fehri, A. 1997a. Layers in the distribution of Arabic modifiers and their licensing. The 11th Symposium on Arabic linguistics. Atlanta: Emory U. Fassi Fehri, A. 1997b. Licensing Arabic adjectives. GLOW 20 Workshops. GLOW Newsletter 38, 67±68. Fassi Fehri, A. 1997c. Arabic adverbs and their Case. Linguistic Research 3, 1±25. Rabat: IERA. Giusti, G. 1995. A unified structure representation of (abstract) Case and article: evidence from Germanic. Studies in comparative Germanic syntax, ed. H. Haider, S. Olsen, & S. Vikner, 77±93. Dordrecht: Kluwer. Greenberg, J. 1966. Some universals of grammar with particular reference to the order of meaningful elements. Universals of language, ed. J. Greenberg, 73±113. Cambridge, Mass: MIT Press. # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

Arabic modifying adjectives and DP structures

153

Halle, M. & Marantz, A. 1993. Distributed morphology and the pieces of Inflection. The view from building 20, ed. K. Hale, K. & J. Keyser, 111±176. Cambridge, Mass: MIT Press. Hawkins, R. 1983. Word order universals. New York: Academic Press. de Hoop, H. 1997. A semantic reanalysis of the partitive constraint. Lingua 103, 151±174. Heim, I. 1982. The semantics of definite and indefinite noun phrases. Ph.D. Amherst: U. of Mass. Higginbotham, J. 1985. On semantics. Linguistic Inquiry 16, 547±593. Holmberg, A. 1993. On the structure of predicate NP. Studia Linguistica 47, 126± 138. Holmberg, A. & Sandstrom, G. 1996. Scandinavian possessive constructions from a northern Swedish viewpoint. Microparametric syntax and dialect variation, ed. J. R. Black & V. Motapanyane, 95±120. Amsterdam: John Benjamins. Ibn Aqiil, B. 14th century. SÏarh alfiyyat Ibn Maalik. Cairo: daar al-fikr, 1979. Kayne, R. The antisymmetry ofÇ syntax. Cambridge, Mass: MIT Press. Lamontagne, G. & Travis, L. 1987. The syntax of adjacency. WCCFL 6, 173± 186. Lecarme, J. 1996. Tense in the nominal system: the Somali DP. Studies in Afroasiatic grammar, ed. J. Lecarme et al., 159±178. The Hague: Holland Academic Graphics. Longobardi, P. 1994. Reference and proper names. Linguistic Inquiry 25, 609± 665. Longobardi, P. 1996. The syntax of N-raising: a minimalist theory. Ms. U. di Venzia. McGinnis, M. 1995. Fission as feature movement. Papers on minimalist syntax. MITWPI, 27, ed. R. Pensalfini & H. Ura, 165±187. Cambridge, Mass: MIT. Milsark, G. 1976. Existential sentences in English. Indiana, Bloomington: IULC. Mohammad, M. 1988. On the parallelism between IP and DP. WCCFL 7, 241± 254. Moro, A. 1991. The raising of predicates. MITWP, 15, 119±181. Cambridge, Mass: MIT. Nash, L. & Rouveret, A. 1997. Proxy categories in Phrase Structure theory. To appear NELS 27. Ouhalla, J. 1988. The syntax of head movement: a study of Berber. Ph.D. U. College London. Ouhalla, J. 1996. The construct state in Berber. Studies in Afroasiatic grammar, ed. J. Lecarme et al., 278±301. The Hague: Holland Academic Graphics. Quirk, R., Greenbaum, S., Leech, G. & Svartvik, J. 1972. A grammar of contemporary English. Harlow UK: Longman. Renaud, F. 1996. The definite article: code and context. Journal of Semantics 13, 139±180. Ritter, E. 1987. NSO noun phrases in modern Hebrew. NELS 17, 521±537. Ritter, E. 1991. Two functional categories in noun phrases: evidence from modern Hebrew. Syntax and semantics, ed. S. Rothstein. Vol. 25, 37±62. New York: Academic Press. Rothstein, S. 1983. The syntactic forms of predication. Ph.D. MIT. Rouveret, A. 1994. Syntaxe du gallois. Paris: Editions CNRS. Sibawayhi, A. 8th century. al-Kitaab, ed. A. Haaruun. Cario: ¹ aalam al-kutub. 1974. Siloni, T. 1994. Noun phrases and nominalizations. Ph.D. U. de GeneÁve. Smith, C. 1964/1969. Determiners and relative clauses in a generative grammar of # The Editorial Board of Studia Linguistica 1999.

d:/3stud/53 -2/fehri.3d ^ 22/6/99 ^ 14:33 ^ sh/mp

154

Abdelkader Fassi Fehri

English. Language 40, 37±52. Also in Modern studies in English, ed. D. A. Reibel & S. Schane, 247±263. Prentice Hall, New Jersey: Englewood Cliffs. Sproat, R. & Shih, C. 1988. Prenominal adjectival ordering in English and Mandarin. NELS 18, 465±489. Sproat, R. & Shih, C. 1990. The cross-linguistic distribution of adjective ordering restrictions. Interdisciplinary approaches to language. Essays in honor of S. Y. Kuroda, ed. C. Georgopoulos & R. Ishihara, 563±593. Dordrecht: Kluwer. Szabolcsi, A. 1987. Functional categories in the noun phrase. Approaches to Hungarian, ed. I. Kenesei. Vol. 2, 167±189. Jate Szeged. Szabolcsi, A. 1994. The noun phrase. The syntactic structure of Hungarian, ed. F. Kiefer & K. Kiss, 179±274. New York: Academic Press. Tarald, K. T. 1990. D-projections and N-projections in Norwegian. Grammar in progress, ed. J. Mascaro & M. Nespor, 419±431. Dordrecht: Foris. Whorf, B. L. 1945. Grammatical categories. Language 21, 1±11. Wright, W. 1858/1974. A grammar of the Arabic language. Cambridge: Cambridge U. Press. Received November 11, 1997 Accepted December 15, 1998

# The Editorial Board of Studia Linguistica 1999.

Abdelkader Fassi Fehri IERA, Mohammed V University BP 6216 Rabat Institutes, Morocco Fax: (212) (7) 772065 Phone: (212) (7) 773012 E-mail: [email protected]