¹ This research is supported by giants from SSHRCC to the author and to UQAM.

² In fact, the set of nominal features also includes case. Lamontagne and Travis (this volume) argue that noun phrases are KPs, projections of the category K(ase). They note that KP is the nominal counterpart of CP. Pursuing this analogy, the functional categories investigated here are analogous to TP and AGRP (Pollock 1989).

³ The analyses of Hebrew genitives (2.1) and feature mapping (2.2) are from Ritter (1991a).

⁴ Both Irish and Standard Arabic have VSO order in clauses and construct state noun phrases. Guilfoyle (1988) has independently developed a unified analysis of Irish clauses and noun phrases, based on movement of the lexical head (V, N) to the functional head (Infl, D). However, her analysis does not distinguish NumP or its clausal counterpart, AGRP. Fassi Fehri (1989) develops a similar proposal for Standard Arabic.

⁵ In fact, this is the conclusion regarding process nominals with only one DP case-marked by šel. Multiple šel phrases are available in non-argument-taking noun phrases. Cf. Borer (1984), Hazout (1990), Shlonsky (1988) for further discussion of šel. Applying the analysis of Lamontagne and Travis (this volume) to genitive Case assignment in Modern Hebrew, šel may be most perspicuously analysed as genitive K (ase), i.e., the head of the genitive noun phrase. The fact that no overt Kase appears in the CS construction is consistent with the claim of Lamontagne and Travis that the head of KP may be empty iff it is properly (lexically) governed. It appears that this alternative strategy is available inside the NP projection in the FG construction, enabling the subject to remain in situ.

⁶ D_gen is not inherently specified for definiteness. This is evident from the fact that a construct state DP is definite just in case its subject is definite. In Ritter (1991a) I suggest that these noun phrases acquire a definiteness specification from their subjects by the mechanism of Spec-head agreement.

⁷ Jacques Lamarche (personal communication) suggests that certain differences between English and French can be accounted for if grammatical number is lexical in English, but syntactic in French. Thus, he proposes the same analysis for number in French as I have for Hebrew; his analysis of number in English is parallel to my analysis of gender in Hebrew.

⁸ The singular definite determiner has several phonologically conditioned allomorphs: nan, lan, an, a and la.

⁹ These authors consider la a determiner and yo a plural marker which is attached to the same terminal node as la, but they analyze noun phrases as NPs. The discussion in this section incorporates their insights into these particles, but departs from them in proposing that the particles are functional heads dominating the NP, rather than specifiers of N.

¹⁰ In her analysis of verb movement in head final Germanic languages such as Standard Dutch, West Flemish and Z7#x00FC;ritütsch, Browning (1990) assumes that head movement involves adjunction and that this adjunction always locates the moved element to the right side of the target of adjunction. In footnote (2) she proposes the following constraint to account for the non-string vacuous nature of head movement:

The landing site and the launching site for adjunction are on opposite sides of the adjoined to category.

Margaret Speas (personal communication) observes that the same generalization applies to head movement in Navajo, but not Japanese or Basque. One property that distinguishes Navajo, Germanic and Haitian, on the one hand, from Japanese and Basque, on the other, is the fact that the former have some syntactic categories which are head initial and others which are head final, whereas the latter are consistently head initial/final. Since the constraint in the above rule provides a means to determine the headedness of various categories, it may play a crucial role in the acquisition of mixed languages because the headedness of each category must be learned separately. In languages that are uniformly head initial or final it is only necessary to learn that some category is head initial or final. Thus, the constraint in the above rule may constitute a diagnostic for distinguishing these two types of languages.

¹¹ This suggestion is due to Mark Baker (personal communication).

¹² Given this complex structure for pronouns, one might expect third person pronouns of the form la yo, (or yo a) and la li or (li a). Recall that analogous forms consisting of the definite article and third person pronouns do exist in Hebrew, but that they are interpreted as demonstratives rather than personal pronouns (i.e., ha-hu ‘the-he = that’, ha-hi ‘the-she = that’ ha-hem ‘the-they = those’). Lefebvre (this volume) suggests that the Haitian determiner ¡a also has deictic force. Thus, I suggest that the reason that third person personal pronouns do not contain a determiner is that they would then be demonstrative pronouns, like their Hebrew counterparts.

¹³ In fact, Lefebvre analyses la as a deictis element. Since deictics are a subclass of definite elements, I minimally assume that la is specified as [+definite]. It may be further specified as [+deictic] along the lines proposed by Lefebvre. However, nothing in the current analysis hinges on this issue.

¹⁴ Note that I assume that the possessor is base-generated in [Spec, NumP]. This reflects the generally accepted view that possessors are not arguments of the noun they possess. For the purposes of this analysis, possessors could alternatively be generated in [Spec, NP]. It will only be necessary that a possessor which is realized in [Spec, DP] be coindexed with an empty category in [Spec, NumP].

¹⁵ See Cowper (this volume) for an account of lexical insertion in English sentences based on the same reasoning.

¹⁶ I thank Elizabeth Cowper for providing me with the Hungarian data and pointing out its significance.

¹⁷ Szabolcsi’s analysis assumes a single functional category INFL which contains both tense and agreement and assigns nominative case to the subject. Consequently, she assumes that the nominal counterpart of INFL also has two distinct components, possessed, the nominal analogue of tense, and subject agreement. In line with more recent developments, we assume that tense/possession and agreement constitute distinct elements. Chomsky (1989) argues that subject agreement (S-Agr) selects T(ense)P as its complement. Extending this analysis to Hungarian noun phrases, the nominal counterpart of TP should also constitute a distinct head. In other words, there may be yet another functional projection in Hungarian noun phrases, whose head is specified for [±possessed] and whose function it is to assign a possessor interpretation to genitive noun phrases in construction with a non-theta marking noun. I leave this question for future research.