prosodic case checking domain

41
CONSTRUCT STATES AT THE PF INTERFACE * TAL SILONI Tel-Aviv University The paper claims that the Semitic Construct state defines a prosodic domain of Case checking. It has been common lore in generative grammar that the domain of Case checking is a syntactic one. In concert with recent proposals, I argue that Case can also be checked at PF; the domain of Case checking is then defined in prosodic terms. The properties of constructs follow straightforwardly. The treatment extends naturally to nonnominal constructs, which, in turn, provide additional evidence in favor of the prosodic analysis. A morphological parameter derives the difference between languages allowing construct states and those which do not. Finally, contra standard assumption I show that there is no indefiniteness spread in construct states but only definiteness spread. 1. The Phonological Puzzle 1.1 Background Nominal construct states have been the focus of much research on Semitic languages. 1 They are head-initial structures comprising a head noun and a noun phrase. Standard Arabic marks the noun phrase with genitive Case (Hebrew, by and large, does not show morphological Case). Indeed, the scope of relationships between the members of the Semitic construct covers the spectrum of semantic roles typically attributed by a noun to its genitive complement: Agent, Patient, or Possessor. I will often refer to the nonhead member of the construct as the genitive member. Construct states can also be headed by categories other than nouns. In Siloni (1998,2000) light is thrown on the crosscategorial nature of the phenomenon, * For helpful comments and suggestions, I would like to thank Outi Bat-El, Gabi Danon, Marc- Ariel Friedemann, Alex Grosu, Jacqueline Guéron, Julia Horvath, Idan Landau, Chuck Kisseberth, an anonymous reviewer and the audiences at the fifth Conference on Afroasiatic Languages (Paris 7, 2000), The Noun Phrase Workshop (Tilburg, 2000), and the Research Colloquium at Tel-Aviv University (2000). 1 See Benmamoun (1998,2000), Borer (1984, 1996, 1999), Cinque (2000), Danon (1996), Dobrovie-Sorin (2001), Fassi Fehri (1989,1993), Hazout (1991), Kihm (1999), Mohammad (1988, 1998), Ritter (1988, 1991), Sadock (2000), Shlonsky (1990, 2000), Sichel (2000), Siloni (1991, 1997).

Upload: others

Post on 15-Apr-2022

8 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE*

TAL SILONI Tel-Aviv University

The paper claims that the Semitic Construct state defines a prosodic domain of Case checking. It has been common lore in generative grammar that the domain of Case checking is a syntactic one. In concert with recent proposals, I argue that Case can also be checked at PF; the domain of Case checking is then defined in prosodic terms. The properties of constructs follow straightforwardly. The treatment extends naturally to nonnominal constructs, which, in turn, provide additional evidence in favor of the prosodic analysis. A morphological parameter derives the difference between languages allowing construct states and those which do not. Finally, contra standard assumption I show that there is no indefiniteness spread in construct states but only definiteness spread. 1. The Phonological Puzzle 1.1 Background Nominal construct states have been the focus of much research on Semitic languages.1 They are head-initial structures comprising a head noun and a noun phrase. Standard Arabic marks the noun phrase with genitive Case (Hebrew, by and large, does not show morphological Case). Indeed, the scope of relationships between the members of the Semitic construct covers the spectrum of semantic roles typically attributed by a noun to its genitive complement: Agent, Patient, or Possessor. I will often refer to the nonhead member of the construct as the genitive member. Construct states can also be headed by categories other than nouns. In Siloni (1998,2000) light is thrown on the crosscategorial nature of the phenomenon, * For helpful comments and suggestions, I would like to thank Outi Bat-El, Gabi Danon, Marc-Ariel Friedemann, Alex Grosu, Jacqueline Guéron, Julia Horvath, Idan Landau, Chuck Kisseberth, an anonymous reviewer and the audiences at the fifth Conference on Afroasiatic Languages (Paris 7, 2000), The Noun Phrase Workshop (Tilburg, 2000), and the Research Colloquium at Tel-Aviv University (2000). 1 See Benmamoun (1998,2000), Borer (1984, 1996, 1999), Cinque (2000), Danon (1996), Dobrovie-Sorin (2001), Fassi Fehri (1989,1993), Hazout (1991), Kihm (1999), Mohammad (1988, 1998), Ritter (1988, 1991), Sadock (2000), Shlonsky (1990, 2000), Sichel (2000), Siloni (1991, 1997).

Page 2: Prosodic Case Checking Domain

TAL SILONI 2

which, it is argued, is typical of heads that do not bear a tense specification. The analysis proposed here applies straightforwardly to nominal as well as nononminal constructs. For the sake of presentation, the paper first centers on nominal constructs. In section 5 their nonnominal equivalents are briefly discussed, with particular attention to the evidence they provide in favor of the analysis proposed here. A characterizing property of constructs is their phonological cohesion, which has already provoked interest in traditional grammars. Gesenius' (1910) grammar of Biblical Hebrew, for example, describes the construct state as a "phonetic and rhytmical" phenomenon, and the relation between its members as "...a close connection (or interdependence) of the nomen regens and the nomen rectum" (§89). Gray (1934) observes that a noun in the construct loses “its own accent, becoming proclitic” that has “only a secondary accent” (§77). Let me first concentrate on the phonological facet; the other properties of the construct (concisely summarized in (13), section 3) will unfold in the course of the paper, and be accounted for straightforwardly once the phonological puzzle is understood. The head of the construct does not bear main stress. Main stress (is shifted forward and) falls on the nonhead member. The head is a stressless weak form phonologically dependent on the nonnhead member. Indeed, Hebrew grammars label the head supportee and the nonhead supporter. Loss of stress may cause further phonological changes parallel to reduction phenomena applying within words due to stress shifting caused by suffixation. To some extent loss of stress affects the consonant composition of the head but more often its vocalization. For example, the original /t/ of the feminine ending is retained in the construct (nonstressed) form, giving rise to the form yaldat ('girl'), while the nonconstruct form referred to as the free or absolute form is yalda ('girl'). An initial vowel /a/ is reduced to schwa when nonadjacent to stress, as in pakid resulting in pkid ('clerk').2 As mentioned, similar reduction phenomena arise when a (plural or pronominal) suffix is attached to the noun, as suffixes in Hebrew also trigger stress shifting (either one or two syllables forward). Thus, we find pakid ('clerk') versus pkidim ('clerks') or pkidenu ('our clerk').3

2 This reduction tends to disappear in Colloquial Hebrew singulars, but is retained in the plural pkidey (‘clerks’); notice the construct state masculine plural ending versus the free state ending pkidim. 3 Stress shifting in constructs is a productive phenomenon operative even in words of foreign origin. For example, stress is obligatorily shifted in loanwords such as koncertey (‘concerts’) or salatey (‘salads’) (even in Colloquial Hebrew, where plural suffixation on the free form variant

Page 3: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 3

Mohammad (1998), Benmamoun (1998) and Kihm (1999) report that Arabic shows phonological alternations of a similar kind. Thus, in Palestinian Arabic, as in Hebrew, the feminine ending /t/ surfaces only on construct heads (9amme versus 9ammet, 'aunt'), and in Standard Arabic the dual and sound plural endings are deleted exclusively on heads of construct states (kitaabayni versus kitaabay, 'two books' and mu9allimiina versus mu9allimii, 'teachers').4 Under common assumptions regarding prosodic (phonological) domains, the head of the construct does not constitute a prosodic word in itself as it lacks main stress. Rather, it is part of the subsequent word; that is, it forms a prosodic word with the head of the genitive member. 1.2 Where does it happen? At PF we have a prosodic word containing the head of the construct and the head of its genitive member (the supporter). A priori, this can indicate that a word has been formed in the lexicon or in syntax (through syntactic incorporation), or it is a post spell-out phenomenon, a word only in phonological terms. The three options are summarized below: (1) The construct state word is formed a. in the lexicon. b. through a syntactic operation (of incorporation). c. at PF. In the rest of this section I present evidence against the first two options. Alongside construct states, Hebrew also has compound constructs which share the same morpho-phonological form with regular constructs but constitute a

does not require stress shifting). The only exception I can think of is loanwords such as univérsitat (‘university’) or temperatúrat (‘temperature’) which can keep the original stress in colloquial Hebrew. Idan Landau (personal communication) observes that this behavior seems to be typical of feminine loanwords. Thus, hipotézat (‘hypothesis’), simfónyat (‘symphony’), or láfat (‘a type of pita bread’) all retain their original stress. 4 Mentioning the phonological alternations the head of the construct undergoes, Kihm (1999) also cites Arabic examples where a parenthetical phrase intervenes between the head and the genitive member of the construct. Kihm notes that traditional grammarians had a debate regarding the status and grammaticality of such examples, which seem to appear exclusively in Arabic within the Semitic family. In Hebrew, indeed, such an intervention is completely inconceivable. This exceptional and debated occurrence of parentheticals between the head and nonhead member does not block the phonological alternations.

Page 4: Prosodic Case Checking Domain

TAL SILONI 4

fixed group of combinations whose meaning is idiosyncratic. The formation of regular constructs, in contrast, is productive, and they are semantically compositional combinations showing diverse semantic relations and no semantic drift typical of lexical combinations. Comparing nominal constructs and compound constructs, Borer (1989) shows that, in addition to their semantic opacity, compound constructs are also syntactically inaccessible, while constructs are semantically and syntactically transparent. For example, compound constructs disallow coordination or modification of the supporter and pronominal reference to the head in itself (as shown in (2)). Their opacity follows form the fact that they are lexical products. In contrast, construct states allow coordination, modification, and pronominal reference to the head alone (3). They are not semantically neither syntactically opaque, and cannot be the output of a lexical operation. (2) a. beyt mlaxa house work “workshop” b. beyt sefer house book “school”

c. *beyt mlaxa ve-sefer house work and book d. *beyt mlaxa adina house work fine

e. *hu bana beyt mlaxa exad ve-od exad li-mgurim. he built house work one and-another one for-residence (3) a. bdikat mizvadot ve-tikim examination suitcases and-bags b. bdikat mizvadot xašudot examination suitcases suspicious c. carix la-avor bdikat mizvadot axat ve-od axat šel tikey-yad.

necessary to-pass examination suitcases one and-another one of handbags

Page 5: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 5

Discarding the lexical option, Borer (1989) entertains the syntactic alternative (1b). In later work (1996,1999), she develops her proposal, arguing that construct states are the result of syntactic incorporation of the head of the supporter into the head of the construct (an idea first entertained by Shlonsky 1990). In rough lines, incorporation is forced by the insertion of a head noun (the head of the construct) lacking a definiteness specification. The specification can only be provided by the genitive member of the construct through incorporation of its head with the underspecified head of the construct (as is well known and will be extensively discussed in section 4, heads of constructs cannot realize their article). The reader is referred to Borer's work for details, which I do not discuss here, as I believe there are good reasons not to adopt a syntactic account, independently of Borer’s specific implementation. The argument Borer provides in favor of syntactic incorporation relies on the impossibility of the examples in (4): (4) a. * roš šloša anašim nir'a miba'ad la-xalon. head three people was+seen through the-window b. *lo ra'iti tmunat af exad. not saw(I) picture no one The problem with the examples in (4), Borer argues, stems form the syntactic word status of construct states. Her reasoning is as follows. Both examples require covert raising of the genitive member out of the construct to obtain scope over the head of the construct in (4a) or to adjoin negation in (4b). But following the syntactic operation of incorporation that forms constructs, that is, syntactic words, raising is impossible. Consequently, the quantified noun phrase in (4a) does not obtain wide scope, and one head is interpreted as associated with three people, which is anomalous. In (4b) the negative polarity item cannot raise to adjoin negation and the structure is ruled out. The picture, however, turns out to be less neat than the facts in (4) seem to suggest. First, in Modern Hebrew phonologically invariable heads of constructs sometimes result in odd to ungrammatical constructs.5 roš (‘head’) (4a) or mexonit (‘car’) are invariable heads. In (5) they give rise to unacceptable constructs independently of the question of scope. Neither (5a) nor (5b) require a

5 In particular, this is so when the genitive member is indefinite. The nature of this apparently phonological constraint is unclear to me.

Page 6: Prosodic Case Checking Domain

TAL SILONI 6

distributive reading (wide scope of the genitive member) to obtain a felicitous interpretation. (5a) at any rate discusses a single person, and in (5b) a narrow scope reading is not problematic as three people can own one single car: (5) a. * roš iš nir’a ba-xalon. head man was+seen in+the-window b. *mexonit šloša anašim nicxa ba-meroc. car three people won in+the-race More importantly, the availability of wide (non overt) scope changes considerably with the choice of quantifier and with the context. In all the sentences of (6) the quantified noun phrase has no problem taking scope over the head of the construct: (6) a. ricpat kol bayit ba-šxuna asuya mi-šayiš. floor every house in+the-neighborhood is+made of-marble b. ricpat šloša batim ba-šxuna asuya mi-šayiš. floor three houses in+the-neighborhood is+made of-marble c. be-sof kol mofa mexulakim še’elonim la-kahal. in-end every show are+distributed questionnaires to+the-audience d. ešet kol xayal še-yesayem et ha-limudim tuzman la-tekes. wife every soldier that will+complete acc the-studies

will+be+invited to+the ceremony As non overt scope is possible in the above examples, the problem with the constructs in (4) cannot follow from the fact that they are syntactic units that do not allow excorporation. There must be different restrictions or contextual strategies that dictate scopal possibilities, which this paper does not intend to investigate. Rather, it follows that such scopal phenomena cannot be used to advance the syntactic option of incorporation.6 6 The data are extremely intricate. Some speakers rule out the šel ('of') phrase counterpart of (4a), too, which suggests again that the problem at hand does not stem from the need to excorporate (i). In general, judgments vary considerably: while some speakers accept both (ii) and (iii), some rule them out, and yet others accept only the free state version:

Page 7: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 7

As for negative polarity items, they do occur in constructs in a high register, as shown in the examples of (7), which use the more archaic item iš ('anyone'):7 (7) a. divrey iš lo ricu oto. words anyone neg pleased him b. lo ra’iti ve-lu tmunat iš exad (li-rfu’a). neg saw(I) and-even picture man one (to-remedy) “I did not see anyone’s picture” c. lo nadun be-divrey iš. neg discuss(we) in-words anyone “We will not discuss anyone’s words”

Moreover, under an analysis of constructs in terms of syntactic incorporation, it is not expected that the genitive member could be a coordinate structure. But such constructs are possible (8a). They cast doubt on a syntactic analysis because it cannot be argued that the whole coordinate structure (the maximal projection) incorporates with the head of the construct, nor can it be suggested that only one conjunct (more precisely, its head) is extracted, given the Coordinate Structure Constraint. On that basis, I have argued in earlier work against an incorporation analysis of constructs (Siloni 1997), concluding without elaboration that their cohesive nature can therefore only follow from a PF requirement. Benmamoun (1998) notes that in Arabic, too, constructs can involve a coordinate structure (8b). He further observes that a gap analysis of such constructs (8c), which could reconcile syntactic incorporation with coordinate structures (without violating the

(i) roš šel šloša anašim nir'a miba'ad la-xalon. head of three people was+seen through the-window (ii) zanav šel šlosa tavasim nir'a min ha-xalon. tail of three peacocks was+seen from the-window (iii) znav šloša tavasim nir'a min ha-xalon. tail three peacocks was+seen from the-window And finally, surprising as it is, speakers even find (iv) very odd, as the distributive reading is the salient one, attributing more than one head to each of the children: (iv) ??rašey šloša yeladim nir'u miba'ad la-xalon. heads three boys was+seen through the-window 7 Example (7b) is thanks to Idan Landau.

Page 8: Prosodic Case Checking Domain

TAL SILONI 8

ban to extract from one conjunct) clearly misrepresents the interpretation of collective predicates such as 'meeting' in (8b): (8) a. beyt ha-rabi mi-kiryat arba ve-re'ayat-o ha-nixbada house the-rabbi from-Kiryat Arba and-wife-his the honorable b. ÷ižtimaa؟u l-mudiir-i wa l-kaatib-i (Standard Arabic) meeting the-director-gen and the-secretary-gen c. [N DP & GAP DP] Providing more evidence from Moroccan Arabic against the syntactic view, Benmamoun (2000) proposes that the phonological unit status of the construct state is the result of a post-syntactic merger of its members in the morphological component of PF, as defined under the Distributed Morphology approach (Halle and Marantz 1993). But Benmamoun‘s account does not provide any reason as to why constructs should give rise to morphological words. As is clear, I share with Benmamoun the conclusion that constructs do not form syntactic words. However, I develop a prosodic account of constructs that derives their word status from the need to satisfy Case requirements. The particular properties of constructs are derived from their prosodic structure. 2. Prosodic Case Checking Domain Let me state the query we are facing in a more suggestive fashion. Why is it so that in Semitic languages heads undergo phonological changes when they check Case within the construct state? Does it mean that genitive Case checking in Semitic languages is sensitive to phonological or prosodic structure? In the generative literature the domain of Case assignment or checking (the Case domain) has standardly been defined in syntactic terms. Local hierarchical configurations specified by notions such as government or Spec-head relations have usually been invoked to determine the Case domain. Consequently, linear intervention, for example, which seems to obliterate the Case domain in certain contexts but not in others, has been taken care of by an ad hoc condition requiring linear adjacency between the assigner and the assignee. It has become a common practice to view adjacency as an additional requirement sometimes relevant for the definition of the Case domain (Chomsky 1980, Stowell 1981, Rizzi 1991, and in terms of movement Chomsky 1995). However, neither the nature of the requirement nor its selective manifestation within and across linguistic systems was understood.

Page 9: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 9

In recent years, the increasing attention paid to the interfaces provoked new interest in PF perspectives of Case. Neeleman and Reinhart (1997) have proposed that Case domains can be defined either in syntax or by the prosodic structure of the PF component (see also Friedemann 1995 for a PF Case constraint). Linear intervention between the Case checker and checkee is, of course, liable to break up a prosodic (phonological) unit. In this view, then, adjacency is dictated by a Case domain at PF, not in syntax. I assume that Case domains can be defined in phonological (prosodic) terms. Further, I suggest that the construct state is a pivotal instance of a phonologically defined Case domain. As will become clear below, the Case domain of constructs is a very tight domain. This is the reason why its members not only do not tolerate linear interference but actually form a word at PF. The gist of the proposal can be stated as follows: (9) a. Case domains can be defined in phonological terms. b. The construct state defines the phonological domain of

(genitive) Case checking in Semitic languages. Relying on work by Selkirk (1986,1995) and references cited there, I embrace the following assumptions with regard to prosodic structure. The assumptions are briefly explained in what follows: (10) a. The phonological representation includes a prosodic structure

consisting of categories of different prosodic types. b. The mapping between syntactic structure and prosodic structure aligns

prosodic categories with edges of syntactic constituents. c. The constraints governing the interface between syntactic and

prosodic structure make no reference to function words. The phonological representation includes a prosodic structure consisting of prosodic categories of different types (syllable, foot, prosodic word, phonological phrase, intonational phrase, utterance), which are organized according to the prosodic hierarchy (from syllable to utterance). The mapping between syntactic structure and prosodic structure above the foot and below the intonational phrase refers to edges of syntactic constituents. Prosodic words are aligned with syntactic words and phonological phrases with maximal projections. That is, the prosodic domain has as its terminal string the stretch of the syntactic structure that is demarcated by the right or left edge of the selected constituent. The choice of the relevant syntactic edge (right or left)

Page 10: Prosodic Case Checking Domain

TAL SILONI 10

may be parametric. Finally, the constraints governing the interface between syntactic and prosodic structure do not refer to function words. Typically, function words are words that belong to functional categories, and display phonological properties significantly different form those of lexical words. In the unmarked case, they are stressless reduced forms. They do not dominate a stressed syllable (a foot), and hence cannot form their own prosodic word but are part of the prosodic word determined by an adjacent lexical word. Turning back to constructs, their nominal head is a lexical word, which nonetheless shows phonological properties typical of function words: it is a stressless (reduced) form. I propose that lexical words can also be inserted as weak forms into the computational system. In prosodic structure, then, they are stressless and qualify as function words. Heads of constructs, then, constitute function words at prosodic structure.8 Now, in Hebrew (or Arabic), prosodic words are aligned with right edges (right brackets) of lexical words. If prosodic words were mapped on the basis of left edges of lexical words, the genitive member of the construct would open a new prosodic word (aligned with its left edge) which would have to exclude the head of the construct: (11) Align the right edge of a prosodic word (PWd) ]PWd with the right

edge of a lexical word ]LexWd. The construct in (12) has the depicted prosodic structure. The nominal head beyt (‘house’) is inserted as a weak form; it qualifies as a function word at prosodic structure, and is therefore part of the prosodic word which is formed by the lexical word ha-mora (‘the teacher’). This forms the prosodic Case domain of the construct state. Note that it is the head of the genitive member which checks the Case required by its maximal projection. This is reminiscent of the syntactic configuration of noun incorporation where the incorporated head satisfies the Case requirement of its maximal projection (Baker 1988): (12) beyt ha-mora ]PWd ha-pikxit house the-teacher the-smart Under this view, the prosodic guise is not an arbitrary characteristic of constructs nor an additional morphological requirement, but an elementary trait

8 Thanks to Chuck Kisseberth for pointing out to me the relevance of function words for my prosodic account.

Page 11: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 11

necessary for Case checking to take place. To be a head in the construct form (that is, a weak form) means to be a head endowed with a genitive feature. To be a head that has to check genitive Case may simply mean to be a head marked for transitivity, a transitive head. Checking of Case may be carried out either in the syntactic component or at PF in a prosodic configuration. Construct states are instantiations of the latter sort. The choice between the syntactic or prosodic domain may be parametric depending on the morphological inventory.9 More generally, the question arises as to what role prosodic Case checking plays in grammar. In the Minimalist Program (Chomsky 1995), the rationale for structural Case checking in syntax is the need to eliminate uninterpretable features, as their illegibility to the LF interface makes the derivation crash. The idea is to relate the existence of uninterpretable features to movement. Thus, noun phrases have to move to check structural Case or else the derivation will crash at LF. However, there is accumulating evidence that structural Case can be checked in situ without movement (Chomsky 1998 and others). If so, then the link between displacement and uninterpretable features is less obvious, and the system, in fact, inserts uninterpretable features and eliminates them with no effect on the derivation. The question is then again what the role is of structural Case. Is it nonetheless legible at the LF interface? (Recall that Case tends to be morphologically marked in languages with freer word order.) I will not pursue the issue any further here. Let me nonetheless anchor the proposal of this paper slightly better in the global picture of Case. Under the ’checking for elimination’ approach, Case checking after spell-out at prosodic structure would mean either (i) that certain Case features are uinterpretable at the PF interface, or (ii) that certain PF outputs have automatic consequences on LF representations. The first option seems counterintuitive: why would Case features be illegible to PF? The possibility that PF and LF representations are not entirely opaque to each other (option (ii)) is interesting and worth exploring independently of the present discussion.10 Furthermore, if

9 According to Neeleman and Reinhart (1997), the choice between prosodic and syntactic domains is driven by economy. Preference is given to prosodic domains. When Case checking fails at prosodic structure, the derivation will proceed to try in syntax. 10 Research on other topics has implicitly assumed that the PF and LF interfaces are not entirely unrelated. I have in mind work on ellipsis and quantifier scope (Fox 1995), or work on focus (Reinhart 1995). Note that I assume here that the Case checked in the construct is structural (as argued in Siloni 1997 on the basis of the fact that it is available in ECM configurations). If it turned out that the Case feature under discussion is inherent, it will not pose an LF-legibility problem anyway.

Page 12: Prosodic Case Checking Domain

TAL SILONI 12

structural Case features are legible to the LF interface, the question would be again whether there is a certain link between the two interfaces, or the same feature has effect in parallel at PF and LF. This question, in fact, is raised by any stress phenomenon that affects interpretation (such as focus for example). I leave this issue for future research and turn to derive the properties of construct states. 3. The Syntactic Behavior of Constructs 3.1 Deriving the properties of constructs The properties of (Hebrew) nominal construct states have often been described in the literature. They are summarized in (13) and discussed and illustrated in what follows: (13) a. Prosodic word: The construct state forms a prosodic word. The prosodic

word can contain a construct string of two or more members (each construct containing one head and one genitive member).

b. Adjacency: Adjectives modifying the head cannot immediately follow it as they usually do, but ought to follow the supporter.

c. Nested Configuration: If both the head and the nonhead are modified by an adjective, the configuration is nested.

d. Overt transitivity: The nonhead member must be lexically realized. e. The article constraint: The head of the construct cannot bear the

article. f. (In)definiteness agreement or spread: the definiteness value of the

members of the construct must match (revisited in (43)). g. Thematic hierarchy: Among genitive arguments, the genitive member

of a nominal construct is chosen according to the following hierarchy: Theme > Agent > Possessor (other genitive arguments will be šel (‘of’) phrases).

The prosodic word status of the construct state is the result of the prosodic domain it defines for Case checking, as already discussed in detail. The genitive member of a construct can itself be the head of a subsequent construct, as in gag beyt ha-mora (‘roof house the-teacher’). The string constitutes one prosodic word due to the function word status of the first two members, which allows each of them to check genitive Case with the nominal following it (as expected, all but the first member of a construct are overtly marked for genitive in Standard Arabic). Consider next adjacency. Adjectives modifying the head of the construct cannot immediately follow it but ought to follow the genitive member (14),

Page 13: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 13

although usually they do occur immediately after the head they modify. Clearly, an adjective immediately following the head would break off the prosodic string necessary for Case checking, as it bears main stress and defines its own prosodic word that is closed at its right edge, as schematized in (14b). Under a syntactic definition of the Case domain (agreement, government), it is not obvious why such linear intervention should matter. (14) a. yaldey ha-mora ]PWd ha-yafim children the-teacher the-beautiful(Pl) “The teacher’s beautiful children” b. * yaldey ha-yafim]PWd ha-mora children the-beautiful(Pl) the- teacher As expected, if forms ambiguous between a noun and an adjective are inserted, simply the first one is interpreted as the genitive member, because the prosodic word is aligned with its right edge: (15) a. yaldey ha-gibor]PWd ha-ca’ir children the-hero the-young ‘The young hero’s children’ b. yaldey ha-ca’ir]PWd ha-gibor children the-young the-heroic ‘The heroic young man’s children’ In sum, as the construct defines a prosodic Case domain, the occurrence of adjectives immediately following the head is impossible. Still it should be explained how a modifying adjective that is normally adjacent to the noun it modifies can surface in construct states in a different position. There are two possible paths: either the syntactic structure is one and the same and rearrangements are allowed at PF, or the syntactic component enables the two alternative orders. If the grammar can do without considerable PF rearrangements, it is more elegant. The latter option should therefore be pursued first. There are several proposals in the literature, which I will not investigate in detail here. The major family of accounts has word order explained by movement of the genitive DP for Case reasons higher than modifying adjectives (Ritter 1991, Siloni 1991,1997, Sichel 2000). Note that while this is natural under syntactic Case checking, a parallel syntactic movement for checking at PF is

Page 14: Prosodic Case Checking Domain

TAL SILONI 14

more problematic as it requires global computation. But more crucially, to the best of my understanding, none of the accounts can handle the fact that an adjective modifying the head of the construct cannot, in fact, follow the entire genitive member (16a) nor intervene between the head of the genitive DP and its complement (16b). The occurrence of an adjective, thus, is not compatible with an ‘articulated’ genitive member, but only with a minimal one as in (16c), or with an additional construct as in (16d). This is unexpected and cannot directly follow from the constituent structure under any of the existing analyses.11 (16) a. *madaf ha-sfarim al ha-milxama ha-arox shelf the-books on the-war the-long “The long shelf of books on the war”

b. *madaf ha-sfarim ha-arox al ha-milxama shelf the-books the-long on the-war “The long shelf of books on the war”

11 This is true for the family of accounts mentioned above as well as for the account proposed by Shlonsky (2000), who suggests deriving word order by NP-raising within DP. More specifically, according to his proposal, the entire NP is raised to a specifier position preceding an adjective modifying the head noun, as schematized in (i). This, again, erroneously predicts (16a) to be grammatical, as shown in (ii). Moreover, Shlonsky’s account in terms of NP-raising also erroneously predicts that in nonconstructs, too, any NP-internal material, a PP complement for instance, ought to precede an adjective modifying the nominal head (iii-iv): (i) [DP [NP madaf ha-sfarim ]i ha-arox ti ] shelf the-books the-long (ii) *[DP [NP madaf ha-sfarim al ha-milxama ]i ha-arox ti] shelf the-books on the-war the-long (iii) *[DP [NP sefer al ciporim nodedot ]i yafe ti] book on birds of passage beautiful (iv) [DP sefer yafe al ciporim nodedot ] book beautiful on birds of passage “a beautiful book on birds of passage” Note that Borer (1996) mentions an example of the type in (16a) as marginal (not impossible). The speakers I have consulted (including me) rule it out.

Page 15: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 15

c. madaf ha-sfarim ha-arox shelf the-books the-long “The long shelf of books” d. madaf sifrey ha-milxama ha-arox shelf books the-war the-long “The long shelf of war books” It is not surprising that an adjective modifying the head of the construct cannot split the genitive constituent (16b). But what is the difference between (16a) and (16c-d)? In the latter examples the modifying adjective is one PWd boundary adjacent to the noun it modifies, as it follows the construct prosodic word, which includes one construct in (16c) and a string of two constructs in (16d). In (16a), in contrast, more PWd boundaries separate between the adjective and the head noun, as al ha-milxama (‘on the war’) defines at least an additional prosodic word. Could it be that some prosodic locality is required between nouns and their modifying adjectives? Note that a modifying adjective can follow another modifying adjective as in madaf ha-sfarim ha-lavan ha-arox (‘shelf the-books the-white the long’). I will not pursue the issue any further here. (17) summarizes the empirical generalization that guides the distribution of modifying adjectives in Hebrew:12 (17) An adjective modifying N1 has to immediately follow a. the prosodic word of N1, or b. the prosodic word of N1 followed by another modifying adjective. While deriving (17) is not a simple task, the relative ordering of adjectives illustrated in (18) is expected under the constituent structure. An adjective

12 Heavy adjectives can follow an articulated genitive DP (i). They constitute semi-relatives (in the sense of Siloni 1995), as shown by the fact that they must be introduced by ha- independently of the definiteness value of the head they modify (unlike modifying adjectives, which must show definiteness agreement with the noun they modify). Like other (semi-) relatives they can follow a construct with an articulated genitive member as well as PP complements etc’: (i) talmid tixon le-omnuyot ha-ge’e be-hesegav pupil high-school for-arts the-proud in- achievements-his (ii) talmid šel dina ha-ge’e be-hesegav pupil of Dina the-proud in-achievements-his

Page 16: Prosodic Case Checking Domain

TAL SILONI 16

modifying the genitive member is predicted to precede an adjective modifying the head (as stated in (13c)), as the latter occurs outside the genitive member. Note that the construction is heavy to marginal (18a) but incomparably better than the one exhibiting the reverse order (18b): (18) a. ?beyt ha-mora ha-ce’ira ha-yafe house(M) the-teacher(F) the-young(F) the-beautiful(M) b. * beyt ha-mora ha-yafe ha-ce’ira house(M) the-young(F) the-beautiful(M) the-teacher(F) Likewise, the obligatoriness of the genitive member (19), property (13d), is trivial. The nonhead member must be realized to prosodically support the function word and discharge it of genitive Case:13 (19) a. * yaldey children(CONSTRUCT FORM) b. yeladim children(FREE FORM) Before focusing on the article constraint (13e) and the definiteness value of the members of the construct (13f), let me briefly discuss the thematic hierarchy

13 Albeit the phonologically dependent nature of the head of the construct , in Colloquial Hebrew (but not in Standard Hebrew), it can appear in a coordinate structure. This, however, is morpho-phonologically constrained. Only heads (suportees) with the same morphological ending can be coordinated: (i) morey ve-talmidey ha-tixon teachers(M) and-pupils(M) the-high-school (ii) *morot ve-talmidey ha-tixon teachers(F) and-pupils(M) the-high-school When the nonhead member is a coordinate structure as in (8), the question arises as to how the second (or third) conjunct is associated with Case as only the head of the first genitive member forms a prosodic word with the head of the construct. Noting that, Alex Grosu (personal communication) proposes that a solution can be found if we assume à la Williams (1980) that conjuncts are generated ‘on top of each other’ in a third representational dimension and linearized following prosodic Case checking .

Page 17: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 17

(13g). In Siloni (1994,1997) I observed that when the noun realizes more than one genitive argument as in (20a-b) (this can happen only in concrete nominals), the genitive member of the construct must be the Theme (20a) (in its absence - the Agent, and lastly the Possessor). If the Theme is not a genitive argument, the construct can be formed with the Agent whether the head is an event noun (20c) or a concrete noun (20d):14 (20) a. tmunat ha-praxim šel ha-yeled picture the flowers of the boy b. * tmunat ha-yeled šel ha-praxim picture the boy of the flowers c. harisat ha-cava et ha-ir destruction the army acc the-city d. mixtavey ha-xayal le-imo letters the soldier to-mother-his As in Siloni (1997), I assume that multiple genitive constructions are right-branching constructions ordered according to the following hierarchy: Theme (lowest), Agent, Possessor (higest). The most embedded argument is linearly the closest to the nominal head, and hence the one participating in the construct state (20a). Further I assume for simplicity that šel (‘of’) phrases in Hebrew are only licensed in specifier or adjunct positions (but see Cinque 1994a, Siloni 1997 for other possibilities). Given N-raising to D (which is by now a current assumption15), when the Agent is merged in SpecNP, it becomes the argument

14 Borer (1999) and Shlonsky (2000) claim that the order Noun-Subject-Object is limited to event nominals and is contingent upon the verbal projection that, according to them, is syntactically embedded within such nominals. (19d) shows that the order Noun-Subject-Object also occurs in concrete nouns, and therefore cannot be exclusively attributed to the putative verbal projection. Note that when an event nominal realizes the Theme as the genitive member of the construct, the Agent can surface in an al-yedey (‘by’) phrase, but not in a šel (‘of’) phrase, as event nominals license one genitive noun phrase only. 15 It has become standard practice to assume N-raising to D in Semitic languages (as well as in other languages). Recently objection has been raised against this assumption. The objection relies on the claim that N-raising cannot derive the respective unmarked ordering of modifying adjectives in Hebrew, which is the mirror image of the English ordering (nor the respective

Page 18: Prosodic Case Checking Domain

TAL SILONI 18

immediately following the nominal head. Hence, the construct is formed with the Agent. An internal argument, then, cannot avail itself of genitive Case: it is therefore possible only when introduced either by the accusative (object) marker et (20c) or by a preposition (20d). For more discussion see the above references. Maybe the most puzzling of the properties the construct state shows is the article constraint (13e) illustrated in (21). The head of the construct cannot bear the definite article (Hebrew has only a definite article): (21) (*ha-)yaldey ha-šxenim (the-)children the neigbors “The neighbors’ children” Albeit the constraint, the head clearly has a definiteness value. Over the years it has become standard practice to assume that its value is identical to the definiteness value of the genitive member. The phenomenon is labeled (in)definiteness agreement or spread (13f). Two formal tests have been utilized to motivate the claim. First, as modifying adjectives agree in definiteness with the head they modify, the fact that an adjective modifying the head of the construct shows definiteness agreement when the genitive member is definite indicates that the head is definite, too (22a). Second, as the accusative (or direct object) marker et must precede definite objects, its occurrence preceding a head of a construct ordering of cardinal and ordinal numbers). For a detailed discussion, see Shlonsky (2000). Assuming, following Cinque (1994b), that the ordering of adjectives shows that they are not adjuncts but specifiers, Shlonsky argues that if the ordering reflects a universal structural hierarchy, it is impossible to derive the differences between English and Hebrew via N-raising (in Hebrew), as it would result in the reverse order: (i) brown Swiss cow (ii) para švecarit xuma cow Swiss brown (Shlonsky 2000) I believe the evidence is not conclusive. While the above order is the unmarked order, it is hard to say that the reverse order is impossible. It is a marked order with a different flavor, but not unacceptable (to me, (iii) sounds much better than (18a) for example): (iii) ?para xuma švecarit cow brown Swiss I do not see how a universal hierarchy could still allow the reverse order in certain contexts. Given that, the possibility that the ordering follows from semantic or more general cognitive preferences should be examined before N-raising is abandoned.

Page 19: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 19

whose genitive member is definite points to the same conclusion (22b): (22) a. tmunat ha-more *(ha-)yafa picture(F) the-teacher(M) (the-)beautiful(F) “the teacher’s beautiful picture” b. ra’iti *(‘et) tmunat ha-mora. saw(I) acc picture the-teacher In the subsequent section, the article constraint is derived from the prosodic structure of the construct, and (in)definiteness spread is shown to be a more intricate phenomenon which does not always apply. I conclude the section with a short comparative discussion. 3.2 Some comparative notes The particular behavior of the construct was shown to follow from its prosodic structure, which is the result of the function word status of its head at PF. Thus, all other things being equal, if a language allows lexical words (nouns) to have a phonologically weak alternate, it is predicted to show construct states; in other words, it is expected to allow genitive Case checking in the prosodic structure of PF. Semitic languages allow it, but English or French do not as their lexical inventory does not include the required items. If my proposal is on the right track, a lexical (morphological) parameter derives the relevant distinction between the former and the latter languages. An anonymous reviewer proposes that phonological reduction of nouns may be contingent upon noun raising to D. The idea would be that when a noun substitutes for the functional head D, it is in some sense functional and can therefore have a phonologically weak form. According to his proposal, N-raising to D is required for phonological reduction, but not all cases of N-to-D involve phonological reduction.16 If this is correct, the proposal that lexical words can have a weak form and behave on a par with function words at PF is constrained and less arbitrary. Indeed, in previous work I argued that N-raising is inherent to the formation of constructs. The reviewer further observes that additional evidence in favor of the proposal comes from Romance languages. Longobardi (1994) argues that isolated

16 There is abundant evidence for N-to-D raising that does not result in a construct state. For example, there are good reasons to believe that the free form of Hebrew nominals also raise to D (see Siloni 1997).

Page 20: Prosodic Case Checking Domain

TAL SILONI 20

instances of construct states are found in Romance languages; in particular, with the noun ‘home’, which is the most typical raising common noun in Romance. Roughly speaking, common nouns that are highly ranked on a scale of uniqueness of designation can overtly raise to D in Romance, and raising is a prerequisite for the formation of a construct. Longobardi reports that in standard Catalan, the noun casa (‘home’) has a phonologically reduced form ca, which shows the typical properties of a construct form. It is obligatory transitive, must be followed by a nonprepositional genitive (23a-b) cannot realize the article (23c), and must be adjacent to the genitive following it (23d) (data from Longobardi 1994): (23)a. ca’ *( l metge) home the doctor b. * ca’ del metge home of +the doctor c. * la ca’ l metge the home the doctor d. * ca’ nova l metge home new the doctor A note on transitivity is in order here. According to Longobardi, overt transitivity is not obligatory in Romance (albeit the Catalan paradigm). Longobardi mentions instances of casa (‘home’) in Italian lacking an overt possessor, where the understood possessor receives an arbitrary interpretation or refers to a syntactically or contextually salient individual. He analyzes these cases as constructs whose head in not phonologically reduced and whose possessor is an empty category (pro). But overt transitivity is a crucial trait of constructs in Semitic languages. Under the PF analysis, it is obligatory because the defining property of the construct is the phonological reduction of its head, which enables and requires the latter to be part of the prosodic word of its Case checkee, as constructs are prosodic domains of Case checking. I do not think that the possessorless occurrences of casa just mentioned are construct states. As observed by Longobardi himself, raised kinship nouns such as mamma (‘mom’) display the same behavior, although they cannot head a genuine construct state. Moreover, Hebrew kinship nouns also allow an implicit possessor but only in the free (nonconstruct) form. These cases are simply occurrences of nouns that intrinsically require a possessor and allow it to be implicit under certain

Page 21: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 21

circumstances. Other constructions often compared to Semitic construct states are genitive constructions in Irish and Romanian. Irish displays a head-initial construction comprising a nominal head and a nonprepositional genitive (24a). Unlike construct states, the construction does not require adjacency between the head and its genitive: adjectives modifying the head appear between the members of the construction (24b). This means that Irish does not perform genitive Case checking within prosodic words. Parametrically, then, it is predicted that the head of the construction should not undergo phonological reduction. The prediction turns out to be correct (data taken from Duffield 1995):17 (24)a. teach an fhir house the man-gen “The house of the man”

b. guth láidir an tsagairt voice strong the priest-gen “The priest’s powerful voice” Romanian also manifests nonprepositional genitive constructions (Dobrovie-Sorin 1987, Grosu 1988, 2001 among others). Again, the head of the construction is not phonologically reduced, which means that the Case domain is not a prosodic word built of the noun and its genitive argument. Indeed, according to Grosu (2001), Romanian genitive constructions do not require adjacency between the nominal head and its genitive argument. Nonetheless, Grosu argues, the constructions do show adjacency restrictions, which hold between the genitive argument and an enclitic morpheme (homophonous to the definite article), which is the licenser of genitive Case, regardless of the stem to which it attaches (N, A or P). Interestingly, although the construction is very different from the construct state, it requires adjacency, which Grosu analyzes as a licensing mechanism applying at PF. For more discussion see Grosu (2001).

17 Like the head of the construct, the nominal head in Irish cannot realize its article. In the next section, I argue that the article constraint characterizing constructs follows from the prosodic structure of the construction. A ban of the sort, however, is also manifested in Saxon genitive constructions, where clearly it does not follow from the prosodic structure of the construction. That is to say, the ban against the article does not automatically classify a genitive construction as a construct.

Page 22: Prosodic Case Checking Domain

TAL SILONI 22

4. Definiteness 4.1 The article constraint The fact that the head of the construct cannot bear the article is derived below from its status as a function word at prosodic structure. Recall the constraint (10c) adopted from Selkirk (1995) and repeated in (25): (25) The constraints governing the interface between syntactic and prosodic

structure make no reference to function words. (25) suggests that the boundaries of function words are invisible to processes governing the interface between syntactic and prosodic structure. Now, affixes are located in the scope of some element that constitutes their domain. It may be either a morphological constituent or a prosodic one (see Anderson 1992:210). Suppose the article in Hebrew is a prefix whose domain is the prosodic word. At the interface with prosodic structure, then, the head of the construct does not license an article as it does not constitute a prosodic word; nor can it count for prefixation of the article of the genitive member although it is part of its prosodic word, because as a function word its boundaries are invisible at this level. Arguments in favor of this proposal come from two sources: what one can label juxtaposed constructions and nonprefixal determiners. Some evidence in favor of the prosodic account of the article constraint is suggested by juxtaposed constructions, appositions in Glinert’s (1989) terms (26a). Superficially, they are reminiscent of construct states. Closer inspection, however, reveals that their prosodic representation is different: while the head of the construct is a stressless weak form (26b), its counterpart in juxtaposed constructions is the free (phonologically invariable) form, which dominates a foot and forms its own prosodic word. Indeed, juxtaposed constructions can and must realize the article construction-initially, unlike construct states (27): 18

18 Likewise, as expected given their prosodic structure, the members of juxtaposed constructions bear some intervention: (i) xatixa axat/gdola basar piece one/big meat For a recent cross-linguistic study of such constructions also known as pseudo-partitives, see Corver (1998) and references cited there.

Page 23: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 23

(26) a. mana falafel; ke’ara salat xatixa basar; glida vanil porion falafel; bowl salad piece meet ice-cream vanilla b. mnat falafel; ke’arat salat xatixat basar; glidat vanil portion falafel bowl salad piece meet ice cream vanilla (27) a. hine [ha-mana falafel] šel-xa. here the-portion falafel of-you b. * hine [mana ha-falafel] šel-xa. here portion the-falafel of-you c. tavi kcat salat im [ha-xatixa basar]. bring some salad with the-piece meat d. * tavi kcat salat im [xatixa ha-basar]. bring some salad with piece the-meat e. hine [ha-glida vanil] šel-xa. here the-ice cream vanilla of-you

f. * hine [glida ha-vanil] šel-xa. here ice cream the-vanilla of-you More crucial evidence in favor of the prosodic account comes from nonprefixal determiners. If the article constraint is correctly derived from its prefixal nature (which is incompatible with the prosodic function word status of heads of constructs), it is predicted that nonprefixal determiners (if available in the grammar) should be able to directly modify the head of the construct. This prediction is borne out (contra common claims that the head of the construct cannot be directly modified by a determiner, e.g. Borer 1996:31). As shown in (28), prenominal determiners (including quantifiers and cardinals) can directly modify the head of the construct. This is illustrated below with oto (‘that’, ‘the same’), od (‘another’) me’eyn (‘kind of’), af (‘no’) or harbe (‘many’), which are all nonprefixal, unlike the definite article: 19

19 Nouns bearing pronominal suffixes cannot realize the article either (i). At first, this may seem as a counterargument to my proposal, as the nouns do not seem to involve a function word. However, such nouns do not allow any determiners whatsoever, prefixal (i) or nonprefixal (ii). This shows that the ban in this case is of a different (nonprosodic) origin:

Page 24: Prosodic Case Checking Domain

TAL SILONI 24

(28) a. ota tmunat praxim that/same(F) picture(F) flowers b. od tmunat praxim another picture flowers c. me’eyn tmunat praxim sort+of picture flowers d. af tmunat praxim no picture flowers e. harbe/kama tmunot praxim many/a few pictures flowers Moreover, a second glance at (28a) seems to reveal that a (nonprefixal) determiner can endow the head of the construct with a definiteness value different from that of the genitive member, contra the standard assumption that the members of the construct obligatorily share their definiteness value (13f). In the following subsection this is discussed in detail. Prior to that, however, a word on Colloquial Hebrew is in order. In Colloquial Hebrew there is an ongoing process relocating the article before the head of the construct. The process obeys yet unexplored constraints; it seems to be sensitive to the position of the construct in the sentence, to the type of

(i) *(ha-)digl-o (the-)flag-his (ii) *oto digl-o that flag-his The ban cannot be semantic either, as the šel paraphrases of (i-ii) are possible. The ungrammaticality of (i-ii) probably stems from a structural incompatibility between pronominal suffixes and determiners (suppose each of them is an instantiation of D): (iii) ha-degel šel-o the-flag of-his (iv) oto ha-degel šel-o that/same the-flag of-his

Page 25: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 25

noun at hand (event, concrete), to the phonological variability of the head of the construct etc’. Thus, while (29a) is possible in Colloquial Hebrew, (29b) is ruled out: (29)a. kvar avarti [ta-bdikat rofe]. already passed acc+the-examination physician b. * nora paxadnu me-[ha-pricat/hatxalat milxama]. very afraid(we) from the-outbreak/beginning war I believe this change is part of a more general reanalysis of the article from a prefix whose domain is the prosodic word to a freer clitic that does not select a specific host and can precede any NP (or AP) initial material, somewhat similar to the complementizer še- (‘that’), which attaches to any clause initial material. Indeed, alongside the repositioning illustrated in (29a), the article also appears preceding degree phrases (30b) and other functional elements: (30)a. ha-yeladot [ha-yafot yoter] (normative language) the-girls the-beautiful more b ha-yeladot [ha-yoter yafot] (colloquial language) the-girls the-more beautiful 4.2 New light on (in)definiteness spread As mentioned, on the basis of two formal tests (see (22)) it has standardly been assumed that the members of the nominal construct bear an identical definiteness value.20 Further evidence in favor of (in)definiteness spread comes

20 I am aware of two qualifications. Engelhardt (2000) argues that certain nominals (in her terminology activity nominals) which, she suggests, do not project a DP-level can be unspecified with regard to definiteness even if the genitive member of their construct is definite. Danon (2001) argues that the members of the construct do not have to agree in definiteness, on the basis of examples such as (i), where the intuition is that the head can be interpreted as indefinite although the genitive member is definite. Glinert (1989), discussing such examples, notes that this is possible only in constructs denoting membership. However, as will become clear in the course of the section, constructs whose genitive member is definite fail to appear in contexts reserved to indefinites and pass diagnostics of definites:

Page 26: Prosodic Case Checking Domain

TAL SILONI 26

from examination of environments requiring or excluding definite noun phrases. It is known that indefinite (more precisely, determinerless) noun phrases cannot serve as subjects of predicational sentences (31a) nor do indefinites lend themselves to be topicalized elements (31b), unlike definites (31c-d). As expected if spread is operative, constructs whose genitive member is definite can appear as subjects of predicational sentences and as topicalized elements (32a-b), in contrast with the constructs in (33a-b), whose genitive member is indefinite:21 (31) a. * tmuna hi xadaša. picture is new b. *tmuna, ra’iti ota. picture, saw(I) it. c. ha-tmuna hi xadaša. the-picture is new d. ha-tmuna, ra’iti ota. the-picture, saw(I) it. (32) a. tmunat ha-praxim hi xadaša. picture the-flowers is new b. tmunat ha-praxim, ra’iti ota. picture the-flowers, saw(I) it. (33) a. * tmunat praxim hi xadaša. picture flowers is new

(i) boger ha-universita nasa ne’um. graduate the-university held speech a. A graduate of the university held a speech. b. The graduate of the University held a speech. 21 To be more precise, predicational sentences actually reject bare indefinites (unless they denote kinds), and allow indefinite subjects that are introduced by weak determiners. Interestingly, then, indefinite construct states of the type in (33a) behave on a par with bare nominals. Thanks to Gabi Danon for pointing it out to me.

Page 27: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 27

b. *tmunat praxim, ra’iti ota. picture flowers, saw(I) it. Definite noun phrases, in turn, are ruled out in existential contexts introduced by the existential particle yeš (‘there is’) (34). And again the behavior of constructs in this environment is dependent on the definiteness value of the genitive member; only a construct whose genitive member is indefinite can appear in existential contexts (35):22 (34) a. yeš student ba-xeder. there is student in+the-room b. * yeš (et) ha-student ba-xeder. there is (acc) the-student in+the-room (35) a. yeš talmid tixon ba-xeder. there is pupil high-school in+the-room b. * yeš (et) talmid ha-tixon ba-xeder. there is (acc) pupil the-high-school in+the-room But once it becomes clear that nonprefixal determiners can modify the head of the construct directly (28), the question arises as to whether the addition of such a determiner can endow the head with a definiteness value independent of the value of the genitive member, or whether (in)definiteness spread is obligatory. I first discuss constructs whose genitive member is indefinite and whose head is modified by a nonprefixal definite determiner; I will then turn to constructs whose genitive member is definite. The inflected definite determiner oto (let us concentrate on its anaphoric ‘that’ meaning), used in (28a) above, can optionally cooccur with the definite article without any semantic effect (36a), just like the adjectival demonstrative (36b-c). Both can modify the head of the construct (37):23

22 As is obvious, the insertion of the accusative/definite object marker et in (34b) and (35b) does not make the sentences grammatical, but would enable the list reading that definites can receive in existential constructions. 23 Adjectival demonstratives bear the article in agreement with the noun they modify, hence:

Page 28: Prosodic Case Checking Domain

TAL SILONI 28

(36) a. oto (ha-)yeled that (the-)boy “that boy” b. yeled ze boy this “this boy” c. ha-yeled ha-ze the-boy the-this “this boy”

(i) bayit (*ha-)ze house (the-)this (ii) ha-bayit *(ha-)ze the-house (the-)this (iii) tmunat ha-praxim ha-zot picture(F) the-flowers the-this(F) The demonstrative can also appear in a prenominal position, although this use is archaic (iv). A prenominal demonstrative requires that the noun it modifies bear the definite article. Hence, it is not surprising that it cannot modify the head of the construct (v); the same holds in Standard Arabic (see Mohammad (1998)). In Hebrew, in fact, a noun modified by a prenominal demonstrative cannot be modified by any other element (vi) (the sequences below give rise to grammatical (verbless) sentences, which is irrelevant here): (iv) ze *(ha-)bayit this the-house (v) *ze beyt ha-mora this house the-teacher (vi) ze ha-bayit (*ha-xadaš) (*šel ha-mora) this the house (the-new) (of the-teacher) Additionally, note that the determiner oto is incompatible with a possessor reading of the genitive member independently of its definiteness value. I will not discuss this incompatibility here: (vii) * ota tmunat (ha-)muzeon that picture (the-)museum

Page 29: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 29

(37) a. ota tmunat praxim that(F) picture(F) flowers “that picture of flowers” b. tmunat praxim zot picture(F) flowers this(F) “this picture of flowers” In both examples of (37) the head of the construct is modified by a definite determiner (demonstrative) while its genitive member is indefinite. Now, albeit the indefinite genitive member, these constructs behave on a par with definite noun phrases: they can appear as subjects of predicational sentences (38a-b) and as topicalized elements (38c-d): (38) a. ota tmunat praxim še-hizkarta hi xadaša. that picture flowers that-mentioned(you) is new b. tmunat praxim zot hi xadaša. picture flowers this is new c. ota tmunat praxim še-hizkarta, ra’iti ota. that picture flowers that-mentioned(you), saw(I) it. d. tmunat praxim zot, ra’iti ota. picture flowers this, saw(I) it. Likewise, as expected, they are excluded from existential contexts introduced by yeš (39) just like other definites (if še-hizkarta ‘that-you mentioned’ is dropped in (39b), oto is interpreted as referring to the kind (‘the same kind of picture’), not to a specific (definite) token in the world and the sentence is grammatical): (39) a. * yeš tmunat praxim zot be-xeder. there is picture flowers this in+the-room b. * yeš ota tmunat praxim še-hizkarta ba-xeder. there is that picture flowers that-mentioned(you) in+the-room In brief, a definite determiner (demonstrative) can render the head of the construct definite independently of the definite value of the genitive member.

Page 30: Prosodic Case Checking Domain

TAL SILONI 30

Still, unlike the definite article, the determiner oto and the demonstrative ze never trigger definiteness agreement on modifying adjectives as illustrated in (40a-b). And while oto can cooccur with et (the accusative, direct object marker preceding definite nouns), ze disallows it (40c-d). (40) a. ota tmunat praxim (*ha-)nifla’a that flowers (the-)wonderful b. tmunat praxim (*ha-)nifla’a zot picture flowers (the-)wonderful this c. ra’iti (et) ota tmunat praxim. saw(I) (acc) that picture flowers d. ra’iti (*et) tmunat praxim zot. saw(I) (acc) picture flowers this Thus, although these constructs can appear in syntactic environments reserved to definites (38), adjectival agreement in definiteness and et licensing do not show the behavior triggered by the presence of the definite article (compare (40) to (22)). This suggests a distinction between the definite article ha- and other definite determiners (demonstratives). Along lines proposed by Danon (2001) on different grounds, I suggest that a definite noun phrase can be formally marked with a definiteness feature [+def]. The definite article is the formal marking, but nouns can also be definite without that formal marking, via another definite determiner or demonstrative, for example. This is somewhat reminiscent of gender marking: a noun can refer to a feminine entity and be formally marked as feminine (yaldot ‘girls’ bearing the feminine plural suffix -ot), but a noun can also refer to a feminine entity and lack a feminine formal marking (našim ‘women’, which actually bears a masculine suffix). Definiteness agreement on modifying adjectives is sensitive to the formal feature: it takes place when the modified noun is formally marked [+def] or heads a construct whose genitive member bears [+def]. The marker et can, in addition, be licensed by the definite determiner oto, but not by the adjectival demonstrative.24

24 To be more complete, adjectival agreement in definiteness also takes place when the modifying noun is a proper name or bears a pronominal suffix:

Page 31: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 31

As far as (in)definiteness spread is concerned, then, a clear conclusion is that an indefinite genitive member does not trigger indefiniteness spread. This is congruent with Dobrovie-Sorin’s (2001) claim that there is no indefiniteness spread in Semitic constructs (I will not discuss her arguments here). Moreover, it is important to note that adjectives without the definite article should not be considered as adjectives showing indefiniteness agreement, as they can also modify definite nouns as in (40a-b). Danon (1996,2001) argues in favor of an asymmetric approach to definiteness. According to Danon, syntactically (in Hebrew) there is no indefiniteness feature [-def] but only a definiteness feature [+def]. Semantically, definiteness is defined as a semantic operator, while indefiniteness is a term grouping together the interpretations of noun phrases obtained without application of the semantic operator. If there is no [-def] feature or value, lack of indefiniteness spread or indefiniteness adjectival agreement comes as no surprise. 25 Let us now turn to constructs whose genitive member is marked [+def]. Not only does their head trigger definiteness agreement on modifying adjectives and the occurrence of et (22), but in addition it cannot be directly modified by weak determiners, which are incompatible with definite nouns (41a-b).26 Moreover, as mentioned above, such a construct cannot appear in existential contexts introduced by the existential particle yeš (35b) on a par with definite nouns (34b).

(i) dina ha-yafa Dina the-beautiful (ii) beyt-o ha-xadaš house-his the-new The behavior of et is more cumbersome: it occurs with proper names and nouns bearing a pronominal suffix, but also with wh-elements. A more complete account of these data is beyond the scope of the paper. 25 In this context, it would be interesting to examine the behavior of Standard Arabic, which is generally argued to have an indefinite marker –n. 26 Gabi Danon (p.c.) mentions (i) as an exception. However, the expression xayal cahal behaves on a par with indefinites in other contexts as well. Danon also reports that speakers seem to accept weak determiners with constructs denoting membership (ii), which I do not accept. If this is correct, it is an interesting exception, which will not explore here. (i) af xayal cahal no soldier IDF (ii) (*)harbe bogrey ha-xug many graduates the-department

Page 32: Prosodic Case Checking Domain

TAL SILONI 32

This suggests that definiteness spread is obligatory when the genitive member is marked [+def]: (41) a. *af [tmunat ha-praxim] no picture the-flowers b. * harbe/kama [tmunot ha-praxim] many/a few pictures the-flowers And in fact the same holds when the genitive member is rendered definite by other means. The head cannot be modified by a weak determiner (42a-b), nor can the construct appear in existential contexts (42c-d): (42) a. *af [tmunat otam yeladim] no picture these children b. * af [tmunat yeladim ele] no picture children these c. yeš tmunat (*otam) yeladim ba-xeder. there is picture (those) children in+the-room d. yeš tmunat yeladim (*ele) ba-xeder. there is picture children (these) in+the-room This means that definiteness spread is not limited to the formal feature, but applies whenever the genitive member is definite. (43) summarizes the behavior of constructs with regard to (in)definiteness: (43) (In)definiteness in Constructs: a. There is no indefiniteness spread. b. A definite genitive member requires definiteness spread. c. When the genitive member is not definite, either the head is indefinite by

default or it is endowed with a definiteness value by its own (nonprefixal) determiner/demonstrative.

While there is no indefiniteness spread, definiteness spread is obligatory when the genitive member is definite. Sharing a definiteness value is therefore not a defining property of constructs but rather imposed by the presence of a definite

Page 33: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 33

genitive member. If the scope of the Hebrew definite article is indeed the prosodic word as discussed in section 4.1, (43b) could be explained as follows: both members of the construct are in the scope of the definite article and interpreted as definite because they are part of the same prosodic word. But this does not seem to be an adequate explanation as definiteness spread holds also when the genitive member is made definite by a nonprefixal determiner or demonstrative (42). The question therefore arises as to whether their scope, too, is the prosodic word, whence we get definiteness spread. Or, alternatively, if definiteness spread is a semantic phenomenon, we expect it to be sensitive to the semantic value of the nominal with regard to definiteness, rather than to whether it is marked with respect to the formal (syntactic) feature [+def]. A novel proposal along these lines has recently been put forward by Dobrovie-Sotrin (2001), who derives definiteness spread from the semantic compositionality of the construction. I will not develop this point any further here. Note nonetheless that while under previous analyses of constructs (see note 1 for references) the spread follows from the inability of the head of the construct to have its own definiteness value and is derived from the structural configuration, here the spread is imposed by the presence of a definite genitive member. Recall that Borer (1996,1999) argues that what enables and requires the formation of a construct is the insertion of a nominal head inherently unspecified for a definiteness value, which must therefore be inherited from another DP (the genitive member) through the construct state. As shown above, this proposal could not be on the right track. The members of the construct simply do not always bear an identical definiteness value, and the head of the construct can be endowed with its own definiteness value by its own determiner. In addition, if there is no [-def] feature in Hebrew, as discussed earlier, indefinite nouns are not marked indefinite but rather receive any nondefinite interpretation. Hence, lack of a definiteness specification is not an illegitimate situation, which ought to be mended. In Siloni (1994,1997), I have suggested that heads with no tense specifications can bear a genitive feature and thus head a construct state. Under this approach it is predicted that construct states occur across syntactic categories. Indeed, constructs can be headed not only by nouns but also by gerunds, adjectives and prepositions. The following section briefly discusses nonnominal constructs dwelling in particular on the evidence provided by adjectival and gerundive constructs in favor of the prosodic analysis of constructs. 5. A Cross-categorial Phenomenon Construct states are not limited to nominal heads. A variety of lexical heads can occur in the construct: gerunds (Siloni 1994,1997), quantifiers (Danon 1996),

Page 34: Prosodic Case Checking Domain

TAL SILONI 34

adjectives (Hazout to appear, Siloni 1998, 2000), beynoni forms (Hebrew participles) and most probably even prepositions (Siloni 1998,2000). On a par with nominal heads, these heads occur in a genitival relation with a nominal constituent following them (in parallel constructions in Standard Arabic genitive Case is morphologically marked). (44)a. bi-[r’ot dan] et ha-yeled… (gerund) in-seeing Dan (acc) the-boy… “When Dan saw the boy…” b. [šlošet ha-yeladim] (quantifier) three the-boys “the three boys” c. yalda [yefat mar'e] (adjective) girl(FM SG) beautiful(FM SG) look(MS SG) 'a beautiful looking girl' d. yeladim [nos'ey matanot] (beynoni) children(MS PL) carry(MS PL) gifts(FM PL) 'children carrying gifts' e. [al ha-šulxan] (preposition) on the-table. The constructions show characteristics typical of construct states, which I shortly overview here. They form prosodic words: main stress falls on the genitive member. Quantifiers, Adjectives and beynoni forms show the same phonological alternations nominal heads of construct undergo.27 The members of the constructions have to be adjacent (but see below), and the genitive member is obligatory. Quantifiers heading constructs cannot bear the article, just like nominal heads of constructs. Likewise, modifying constructs headed by an adjective or a beynoni form, which show definiteness agreement with the noun they modify, cannot bear the article, which has to surface on their supporter. All this is natural under the prosodic approach to constructs, as discussed above. For more on definiteness in these constructs, see Siloni (1998,2000). In addition,

27 Gerunds and prepositions are invariable. Note that certain prepositions may bear contrastive stress, which may cast doubts on their status as constructs.

Page 35: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 35

there are constraints on the type of supporter allowed in these various constructs, which are discussed in the above references, and which I will not review here as they are not relevant for our purposes. Adjectival and gerundive constructs provide additional evidence in favor of the prosodic analysis. Or looking at it the other way round, under the prosodic analysis, two apparently unrelated puzzles that these constructs have raised receive a simple and unified account. As noted in Siloni (1998), only simplex adjectives, that is, adjectives that are not formed by an adjectival formative (suffix), can occur in adjectival constructs. For example, 'acbani ('nervous') involves an adjectival suffix and cannot head a construct. (45) presents a minimal pair: the simplex form ge’e (‘proud’) allows a construct, but not the suffixed form ga’avtan (‘proud’). What could be the reason for this difference? (45) a. ge’e levav proud heart

b. *ga’avt-an levav proud-suffix heart The second puzzle is raised by gerundive constructs.28 The subject of the gerund can be either a lexical noun phrase or a pronominal suffix: (46) a. bi-[r’ot dan] et yald-o, alac lib-o. in-seeing Dan acc child-his, exulted heart-his “When Dan saw his child, his hurt exulted” b. bi-[r’ot-o] et yald-o, alac lib-o. in-seeing-his acc child-his, exulted heart-his “When he saw his child, his hurt exulted” Surprisingly, the object cannot be realized as a pronominal suffix, although nonfinite forms are the most friendly verbal environment for pronominal suffixes in Modern Hebrew:

28 Gerundive constructs occur in written language and formal registers. Nonetheless the relevant data are crystal clear for all speakers.

Page 36: Prosodic Case Checking Domain

TAL SILONI 36

(47) a. * bi-[r’ot-o] dan, alac lib-o. in-seeing-himsuffix Dan, exulted heart-his “When Dan saw him…” b. hu raca [li-r’ot-o] he wanted to-see-himsuffix I believe the impossibility of (47a) and (45b) straightforwardly follows from the prosodic structure of the construct. Suffixes in Hebrew attract main stress. The occurrence of an object pronominal suffix (48a) or an adjectival formative (48b), therefore, breaks off the prosodic string of the construct, the prosodic word which is necessary for the supporter to check Case (dan in (48a), levav in (48b)): (48)a. * bi-r’ot-o]PWd dan, alac lib-o. in-seeing-himsuffix Dan, exulted heart-his “When Dan saw him…” b. * ga’avt-an]PWd levav proud-suffix heart Now, Hebrew also has weak object pronouns. If the prosodic account is correct, they ought to be able to intervene between the gerund and its lexical subject as they are stressless elements, which therefore should not affect the prosodic structure. This prediction is born out. (48a) and (49) constitute a minimal pair. While (48a) is impossible as the pronominal suffix attracts main stress and splits the prosodic word, the weak pronoun is stressless and does not interrupt the prosodic sequence (49). There is no other obvious difference between (48a) and (49) except stress assignment. Under a syntactic analysis it would be very mysterious why a suffix has a blocking effect but not a weak pronoun (if at all, the opposite would be more expected): (49) bi-r’ot oto dan] PWd alac lib-o. in-seeing himweak pron Dan, exulted heart-his “When Dan saw him…” Starting with Borer (1984), it has often been argued that Hebrew noun phrases exhibit clitic doubling configurations (50a). The obligatory presence of the dummy preposition (marker) šel (‘of’) (50b) was explained by the need to insert a Case marker for the doubled noun phrase to receive Case because the

Page 37: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 37

Case of the noun was argued to be absorbed by the pronominal clitic (this characteristic of clitic doubling configurations was labeled Kayne’s generalization): (50) a. digl-oi šel dani flag-his of Dan

b. *digl-oi dani flag-his Dan Under the prosodic account, the impossibility of (50b), (47a) and (45b) follows form one single reason. The presence of a suffix interrupts the prosodic sequence necessary for Case checking to take place as schematized in (48a-b) and (51a). Hence, a dummy preposition must be present (51b) to take care of the Case requirement of the possessor noun phrase: (51) a. * [beyt-o] PWd dan house-his Dan b. [beyt-o] PWd šel dan house-his of Dan From the obligatory presence of šel, then, one cannot draw the conclusion that the pronominal suffix requires and checks Case with N and is consequently a pronominal clitic and not an agreement marker (as I have argued in Siloni (1997)). The Case problem in (51a) is independent of the status of the suffix as either an agreement marker or a pronominal clitic (see Engelhardt (1998) and Shlonsky (1997) for the former claim). 6. Conclusions The main claims advanced in the paper are summarized below: • Case can be checked at PF in a prosodic domain. • The choice between syntactic or prosodic Case checking is parametric and

depends on the morphological inventory of the language. • Construct states define the prosodic domain in which genitive Case is

checked in Semitic languages. The domain is the prosodic word. It is formed due to the function word status of the head of the construct in prosodic structure.

Page 38: Prosodic Case Checking Domain

TAL SILONI 38

• Only stressless elements can intervene between the head of the construct and its genitive member as they do not interrupt the prosodic string necessary for Case checking. Thus, an adjective modifying the head of the construct cannot immediately follow it, nor can a pronominal suffix or an adjectival formative be attached to the head of the construct as they attract main stress.

• The definite article cannot be realized on the head of the construct because the latter constitutes a function word at prosodic structure, and function words are invisible to morpho-syntactic rules such as prefixation that operate at the interface between syntax and prosodic structure. Indeed, nonprefixal determiners can modify the head of the construct.

• The members of the construct do not always share the same definiteness value. Definiteness spread is imposed by a definite genitive member. There is no indefiniteness spread in constructs.

References Anderson, S. 1992. A- Morphus Morphology, Cambridge: Cambridge University

Press. Benmamoun, E. 1998. “Apparent Agreement Mismatches: A Question of Spell-

Out”. paper presented at the conference on the Syntax of Semitic Languages, USC.

Benmamoun, E. 2000. “Agreement Asymmetries and the PF Interface”. in J. Lecarme, J. Loewenstam & U. Shlonsky, eds. Studies in Afroasiatic Grammar 2, Amsterdam: John Benjamins.

Baker, M. 1988. Incorporation. A Theory of Grammatical Function Changing. Chicago: The University of Chicago Press.

Borer, H. 1984. Parametric Syntax. Dordrecht: Foris. Borer, H. 1989. "On the Morphological Parallelism between Compounds and

Constructs". in G. Booij et al, eds. Morphology Yearbook 1. Dordrecht: Foris.

Borer, H. 1996. “The Construct in Review”. in J. Lecarme, J. Loewenstam & U. Shlonsky eds. Studies in Afroasiatic Grammar, Holland Academic Graphics. 30-61.

Borer, H. 1999. “Reconstruction the Construct”. in K. Johnson & I. Roberts, eds. Dordrech: Kluwer Academic Publishers.

Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass: MIT Press. Chomsky, N. 1998. Minimalist Inquiries: The Framework. MIT Occasional

Papers in Linguistics 15.

Page 39: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE 39

Cinque, G. 1994a. “Posizione del soggetto nel DP italiano”. in L. Renzi, ed. Omaggio a Florica Dimitrescu e Alexandru Niculescu, Padova: Unipress.

Cinque, G. 1994b. “On the Evidence for partial N-Movement in the Romance DP”. In G. Cinque, J. Koster, J. Pollock, & R. Zanuttini, eds. Paths Towards Universal Grammar: Essays in Honor of Richard S. Kayne. 85-110. Georgetown University Press.

Cinque. G. 2000. “On Greenberg’s Universal 20 and the Semitic DP”. handout of a paper presented at the Antisymmetry Conference, Cortona.

Corver, N. 1998. “Predicate Movement in Pseudopartitive Constructions”, in A. Alexiadou and Ch. Wilder Studies in the Determiner Phrase.

Danon, G. 1996. “The Syntax of Determiners in Hebrew”. M.A. Thesis. Tel-Aviv University.

Danon, G. 2001. “Syntactic Definiteness in the Grammar of Modern Hebrew”. Linguistics 39.6.

Dobrovie Sorin, C. 1987. “A propos de la structure du groupe nominal en roumain”. Rivista di grammatica generativa, 12, 23-152.

Dobrovie Sorin, C. 2001. “From DPs to NPs: A Bare Phrase Structure Account of Genitives”. to appear in the Proceedings of the Antwerps Conference “From DP to NPs”.

Duffield, N. 1995. Particles and Projections in Irish Syntax. Dordrecht: Kluwer Academic Publishers.

Engelhardt, M. 1998. “The Syntax of Nominalized Properties”. PhD Dissertation. The Hebrew University of Jerusalem.

Engelhardt, M. 2000. “The Projection of Argument-taking Nominals”. Natural Language and Linguistic Theory.

Fassi Fehri, A. 1989. “Generalized IP Structure, Case and VS Word Order”. in I. Laka & A. Mahajan, eds. MIT Working Papers in Linguistics, 10, 75-113.

Fassi Fehri, A. 1993. Issues in the Structure of Arabic Clauses and Words. Dordrecht: Kluwer Academic Publishers.

Fassi Fehri, A. 1999. “Arabic modifying adjectives and DP structures”. Studia Linguistica, 53.105-154.

Fox D. 1995. “Economy and Scope”. Natural Language Semantics, 3.283-341. Friedemann, M.A. 1995. “Sujets syntaxiques: positions, inversions et pro”.

Doctoral dissertation, Université de Genève. Gesenius 1910. Gesensius' Hebrew Grammar. Edited by E. Kautsch, revised by

A.E. Cowley, Oxford: Clarendon Press. Glinert, L. 1989. The Grammar of Modern Hebrew. Cambridge: Cambridge

University Press. Gray , L. H. 1934. Introduction to Semitic comparative Linguistics. New York:

Page 40: Prosodic Case Checking Domain

TAL SILONI 40

Columbia University Press. Grosu, A. 1988. “On the Distribution of Genitive Phrases in Romanian”.

Lingustics 26, 931-949. Grosu, A. 2001. “The Distribution of Genitive Phrases in Romanian”. ms. Tel-

Aviv University. Halle, M. & A. Marantz 1993. “Distributed Morphology and the Pieces of

Inflection”. in K. Hale & S.J. Keyser, eds. The View from Building 20. Cambridge, Mass: MIT Press.11-176.

Hazout, I. 1991. “Action Nominalizations and the Lexicalist Hypothesis”. Doctoral dissertation, University of Massachusetts, Amherst.

Hazout, I. to appear. “Adjectival genitive constructions in Modern Hebrew: A case study in coanalysis”. The Linguistic Review.

Kihm A. 1999. “Towards a Predicational Theory of Semitic Construct State Nominals”. ms. CNRS, Paris.

Longobardi, G. 1994. “A Case of Construct State in Romance”. in R. Ajello & S. Sani, eds. Scritti linguistici e filologici in onore di Tristano Bolelli. Pisa: Pacini.

Marantz, A. 1988. “Clitics, Morphological Merger, and the Mapping to Phonological Structure”. in M. Hammond and M. Noonan, eds. Theoretical Morphology, San Diego: Academic Press, 253-270.

Mohammad, M. 1988. “On the Parallelism between IP and DP”. in H. Borer, ed. Proceedings of WCCFL 7, 241-254.

Mohammad, M. 1998. “Checking and Licensing inside DP in Palestinian Arabic”. University of Florida.

Neeleman, E. and T. Reinhart 1996. “Scrambling and the PF Interface”. in W. Gueder and M. Butt, eds. Projecting from the Lexicon, CSLI, Stanford.

Reinhart, T. 1995. “Interface strategies” OTS Working Papers. Utrecht University.

Ritter, E. 1988. “A Head-Movement Approach to Construct-State NPs”. Linguistics 26.6. 909-929.

Ritter, E. 1991. “Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew”. Syntax and Semantics 25. 37-62.

Rizzi, L. 1991. “Residual Verb Second and the Wh-Criterion”. Technical Reports on Formal and Computational Linguistics 2, Université de Genève.

Sadock, G. 2000. “Decomposing the Construct”. ms. University of Chicago. Selkirk, E.O. 1986. “On Derived Domains in Sentence Phonology”. Phonology

Yearbook 3. 371-405. Selkirk, E.O. 1995. “The Prosodic Structure of Function Words”. In J. Beckman,

L. W. Dickey, and S. Urbanczyk eds. Papers in Optimality Theory.

Page 41: Prosodic Case Checking Domain

CONSTRUCT STATES AT THE PF INTERFACE

41

Shlonsky. U. 1990. “Construct State Nominals and Universal Grammar”. ms. University of Haifa.

Shlonsky. U. 1997. Clause Structure and Word Order in Hebrew and Arabic, Oxford: Oxford University Press.

Shlonsky. U. 2000. “The Form of Noun Phrases”. ms. University of Geneva. Sichel. I. 2000. “Remnant and Phrasal Movement in Hebrew Adjectives and

Possessives”. ms. CUNY Graduate Center Siloni, T. 1991. “Noun Raising and the Structure of Noun Phrases”. in J. Bobaljik

& T. Bures eds. MIT Working Papers in Linguistics 14, 255-270. Siloni, T. 1994. “Noun Phrases and Nominalizations”. Doctoral dissertation,

Université de Genève. Siloni, T. 1995. “On Participial Relatives and Complementizer D0: a Case Study

in Hebrew and French”. Natural Language and Linguistic Theory 13.3. 445-487.

Siloni, T. 1997. Noun Phrases and Nominalizations: The syntax of DPs. Dordrecht: Kluwer Academic Publishers.

Siloni, T. 1998. “Adjectival complexes & inalienable constructions”. to appear in J. Ouhalla & U. Shlonsky eds. Themes and issues in the syntax of Arabic and Hebrew. Dordrecht: Kluwer Academic Publishers.

Siloni, T. 2000. “Nonnominal Constructs”. in J. Lecarme, J. Loewenstam & U. Shlonsky, eds. Studies in Afroasiatic Grammar 2. Amsterdam: John Benjamins.

Stowell, T. 1981. Origins of Phrase Structure, Doctoral dissertation, MIT. Williams, E. 1980. “Across-the-board rule application” Linguistic Inquiry 9 31-

43