<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1011">
  <Title>Towards a linguistically motivated computational grammar for Hebrew</Title>
  <Section position="4" start_page="82" end_page="82" type="metho">
    <SectionTitle>
2 The framework
</SectionTitle>
    <Paragraph position="0"> HPSG is formulated as a set of constraints on typed (1) feature structures (TFSs) that are used to model linguistic information in all levels: from the lexicon, through grammatical principles, to complete (2) analyses. HPSG &amp;quot;rules&amp;quot; are organized as principles that set constraints on the properties of well-formed phrases, along with ID schemata that license certain phrase structures. The schemata are independent of the categories of the involved phrases; they state general conditions for the construction of larger phrases out of smaller ones, according to the function of the sub-phrases (e.g., subject-head, (3) head-complement, specifier-head etc.) ID schemata only license certain phrase combinations. They do not specify all the constraints imposed on the involved sub-phrases, as these are articulated by the principles.</Paragraph>
    <Paragraph position="1"> Like other current linguistic theories, HPSG is highly lexical: most of the information is encoded in highly articulated lexical entries associated with (4) words. The constraints on the grammar are usually few and very general. An elaborate set of lexicat rules relates lexical entries, either to account for morphology or to introduce changes in the TFSs associated with the basic entries.</Paragraph>
  </Section>
  <Section position="5" start_page="82" end_page="86" type="metho">
    <SectionTitle>
3 The structure of noun phrases
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="82" end_page="82" type="sub_section">
      <SectionTitle>
3.1 The data
</SectionTitle>
      <Paragraph position="0"> Hebrew has one definite article, ha-, which attaches to words (nouns, adjectives, numerals and demonstratives, bencefotl_h nominals), not phrases. Many elements in the noun phrase are marked for, and must agree on, definiteness (1). MH provides two major ways of forming genitive relations: free genitives (FG), in which the genitive phrase is introduced by the preposition Sell 'of' (2); and constructs (CS), in which the head noun is morphologically marked (and is said to be in the construct state, cs) and the genitive phrase must immediately follow it, preceding any other modifiers (3). In FG the definiteness of the possessor is independent of that of the head, allowing for four different combinations of definiteness (both the head and the possessor can each be either definite or indefinite) (2); in CS, the definiteness of the phrase is inherited from the possessor, allowing only two combinations: either both are definite, or both are not (3). The definite article never combines with cs-nouns. A poorly studied yet closely related phenomenon is cs-adjectives, which exhibit the same definiteness behavior (4).</Paragraph>
      <Paragraph position="1"> ha- sepr ha- gadol ha- ze/$1iSi the book the big the this/third 'this big book / the third big book'  (ha-) sparim Sell mSorer (the) books of poet '(the) books of a poet' (ha-) sparim Sell ha- mSorer (the) books of the poet '(the) books of the poet' siprei mSorer xdaSim books-cs poet new 'new books of a poet' siprei ha- mSorer ha- xdaSim books-cs the poet the new 'the new books of the poet' yruqqat (ha-) &amp;einaym green-cs (the) eyes 'a/(thC/) green eyed&amp;quot;</Paragraph>
    </Section>
    <Section position="2" start_page="82" end_page="83" type="sub_section">
      <SectionTitle>
3.2 Are noun phrases NPs or DPs?
</SectionTitle>
      <Paragraph position="0"> Following Abney (1987), analyses carried out in Chomskian frameworks view noun phrases as DPs, headed by the functional category D. The DP hypothesis (DPH) has been applied to a variety of languages and is incorporated into most existing accounts for Modem Hebrew. Originally motivated  by the English '-ing' gerunds, that possess simultaneously properties of both sentences and noun phrases, the importance of the DPH is that it assigns parallel structures to clauses and noun phrases; in particular, both are headed by functional categories. In HPSG, however, functional categories are discouraged: English noun phrases are viewed as NPs, headed by the noun, and determiners - as subcategorized specifiers of nouns (Pollard and Sag, 1994, section 9.4). HPSG analyses for other languages, notably German, consider article-noun combinations to be DPs (Netter, 1994). Preferring either of the two analyses, in the context of HPSG, boils down to deciding whether it is the determiner or the noun that heads a nominal phrase. Applying the criteria of (Zwicky, 1985) we show that in Hebrew it is the noun that heads the noun phrases. Netter (1994) lists several considerations in favor of each of the alternatives. In German, all the morphosyntactic features that must be transferred to the maximal projection of a nominal phrase (for agreement or government purposes) are manifested equally well both on the article and on the noun. Determinerless noun phrases require, in German, disjunctive subcategorization frames for nouns under an NP analyses, and empty categories in a DP analysis. Finally, it is the declension phenomenon that causes Netter 0994) to favor a DP analysis. When applied to MH, these considerations yield a different result: information that is relevant for agr~ment, such as number and gender, is expressed on the noun only; determinerless phrases are always grammatical; and there are no declensions.</Paragraph>
      <Paragraph position="1"> Nevertheless, most existing analyses of MH noun phrases apply the DPH, with the definite article as the D head (Ritter, 1988; Ritter, 1991; Siloni, 1991; Siloni, 1994). For lack of space we cannot survey the motivation for such analyses here; the argumentation relies on derived (deverbal) nouns, especially in CS noun phrases, including the following observations: the inability of cs-nouns to be rendered definite directly (i.e., the fact that ha- never attaches to them); the impossibility of direct modification of such nouns (i.e., the fact the any adjectives must follow the genitive complement in CS); and the inheritance of definiteness from the complement in CS. These, along with theory-internal considerations, yield an analysis by which noun phrases are DPs, headed by the functional, possibly phonologically null, category D, and necessitating a compulsory movement of the head noun. FG noun phrases require yet another functional (and empty) category.</Paragraph>
      <Paragraph position="2"> We show in (Wintrier, 1998) that there is no theory-independent reason to apply the DPH to Hebrew; on the contrary, such accounts miss generalizations and yield wrong predictions. We show below that an NP analysis is not only possible but also plausible, accounting for a body of data, traditionally believed to require functional categories and compulsory head raising in noun phrases.</Paragraph>
      <Paragraph position="3"> Many of the limitations of the analyses mentioned above are listed by Borer (1994), suggesting that definiteness is a feature of nouns, base generated on the N stem. An affixal view of the MH definite article is established in (Wintrier, 1997b), and is the starting point for the analysis we propose here.</Paragraph>
      <Paragraph position="4"> We first account for the fact that cs-nouns must have an immediate complement. We then explain why the article does not combine with cs-nominals. We justify a treatment of possessives as complements, and finally present an analysis for both FG and CS noun phrases as NPs.</Paragraph>
    </Section>
    <Section position="3" start_page="83" end_page="84" type="sub_section">
      <SectionTitle>
3.3 Prosodic dependency
</SectionTitle>
      <Paragraph position="0"> Most subcategorized complements are optional in Hebrew: objectless VPs are grammatical in many contexts, as are subjectless clauses. But compulsory, immediate complementation is not unique to cs-nouns only; it is required in cs-adjectives and cardinals, as well as in prepositions and some quantitiers. In spite of the differences among these elemerits, there are some striking similarities: they can never occur without a complement, which cannot be extracted, or 'moved', but which can be replaced by a pronominal pronoun, which is always realized as a clitic (Borer, 1984, chapter 2). The data are summarized in (5).</Paragraph>
      <Paragraph position="1"> (5) siprei ha- m$or.rim / sipreihem books-cs the poets / books-l-3rd-pl-m 'the poets' books I their books' $1o$t ha- m$or_rim / $1oStam three-cs the poets / three+3rd-pl-m 'the three poets / the three of them' Sell ha- m$or_rim / $ellahem of the poets / of/3rd-pl-m 'of the poets / of them' 'et ha- mSor_rim / 'otam ACC the poets / ACC/3rd-pl-m 'the poets (ACC) / them (ACC)'  &amp;al yad ha- m$or.rim / &amp;al yadam near the poets / near+3rd-pbm 'near the poets / near them' koll ha- mSor_rim / kullam all the poets / all+3rd-pbm 'all the poets / all of them' The need for an immediate complement is a result of these elements being prosodically weak. We do not suggest a theory of prosody in I-IPSG; rather, taking advantage of the observation that the discussed constituents correlate well with phrases in MH, we account for them in the following way: we add a DEpendency feature to the lexical entries of words. The value of this feature can either be an empty list, or a list of one element, in which case the element must be reentrant with some element in some valence list of the word (in other words, DEP points to some element on the ARG_S value of the word). As the only relations between prosodically dependent words and their obligatory complements, in Hebrew, are those of head-complement or specifierhead, the obligatory complement is bound to be a member of the ARG_S of those words. In addition, we introduce the prosodic dependency principle, by which words that are specified as prosodically dependent must first combine with the obligatory complement they depend on; only then can the obtained phrases combine with other modifiers: In a headed phrase, in which one of the  daughters is a word, either the DEP of this daughter is empty, or it is reentrant with (the SYNSEM value of) some other daughter. null 3.4 The morphological nature of definiteness Why doesn't the definite article combine with csnouns? Not only nouns have construct states: adjectives (4) and numerals do, too, and ha- does not  combine with the other cs-nominals either. The rules that govern the combination of ha- with nominals are simple, when the article is viewed as an affix (W'mtner, 1997b): (i) ha- attaches to words, not to phrases; (ii) it attaches only to nominals, and to all kinds of nominals; (iii) it only combines with indefinite words. An additional (boolean) feature, DEFiniteness, is required for encoding the value of definiteness in nominals. As definiteness agreement in Hebrew is not a semantic process, we add this feature to the CATegory of nominals (rather than to their CONTent). Since definiteness is a feature of phrases, inherited from the lexical head, DEF is a head feature, appropriate for all nominals. Viewing definiteness as a lexical process, we introduce the Definite Lexical Rule (DLR, 6). It operates on all nominal words whose DEFiniteness feature is '--'. In all categories its effect on the phonology is determined by the same phonological rules, abstracted over by the function definite. The DLR changes the value of the path SYNSEMILOC\[CATIHEADIDEF from '-' to '+'. Adjuncts specify the heads they select as the value of the MOD feature in their lexical entries. Like any other nominal, they have a DEFiniteness feature, whose value is shared with the value of the path MODILOCICAT\]HEADIDEF. When the DLR operates on adjuncts, it results in a specification of a '+' value for both paths. Thus it is guaranteed that definite adjectives, for example, are not only specified as definite but also select definite heads. As for cs-nominals, these are not indefinite; we show below that they are unspecified for definiteness, and hence the DLR cannot apply to them.</Paragraph>
      <Paragraph position="3"> : cat \[nominal\] L s~nsem : cat head :</Paragraph>
    </Section>
    <Section position="4" start_page="84" end_page="85" type="sub_section">
      <SectionTitle>
3.5 Possessives as complements
</SectionTitle>
      <Paragraph position="0"> In standard I-IPSG (Pollard and Sag, 1994, section 9.4.5) possessives are specifiers: they combine with an N's to form complete NPs through the specifier-head schema, and they express the expectation of an N' as the value of the SPECified feature in their HEADS, just like other determiners do. As Pollard and Sag (1994, p. 375) note, this analysis is valid for German and English, but other languages might require different accounts. We advocate a position by which possessives of all kinds are complements in MH. First, possessives differ from other determiners in their distribution. While most determiners precede the noun, possessives follow it (7). Second, possessives can regularly co-occur with other determiners (8). Thus, if determiners occupy the specifier position in NPs, possessives cannot fill the same function. Third, MH exhibits also eases of  clitic doubled constructions (Borer, 1984), where a genitive pronoun cliticizes onto the head noun and must agree with a doubled possessive on number, gender and person. Agreement is usually associated with complements (including subjects) and not with specifiers.</Paragraph>
      <Paragraph position="1"> (7) koll sepr every book 'every book' koll / $1o$t ha- sparim all / three the books 'all books / the three books' ha- sparim $selli / Sell dan the books my / of Dan 'my/Dan's book' (8) koll sepr $selli / Sell dan every book my / of Dan 'each of my/Dan's books' koll ha- sparim $selli / Sell dan all the books my / of Dan 'all my/Dan's books' $1o$t ha- sparim $selli / Sell dan three the books my / of Dan 'my/Dan's three books' Other arguments for viewing possessives as complements, in two languages that show many similarities to Hebrew, namely Welsh and Arabic, are given in (Borsley, 1995). We therefore view possessors as (most oblique) complements of nouns. When the noun has additional arguments, they are listed in its valence feature preceding the possessor. Thus, in the lexical entry of sept ('book'), the value of the COMPlement list has two members, an agent and an optional I possessor. When two possessives are present, the structure depicted in (9) is obtained.</Paragraph>
    </Section>
    <Section position="5" start_page="85" end_page="86" type="sub_section">
      <SectionTitle>
3.6 The structure of CS
</SectionTitle>
      <Paragraph position="0"> As cs-nominals are words, their lexical entries express an expectation for an immediate complement; that is, an indication (the SYNSEM value) of the compulsory complement of cs-nominals is present in the lexical entry of the nominal. It is thus possible to share, in the lexicon, the values of the definiteness 1Recall that most subeategorized elements are optional in Hebrew.</Paragraph>
      <Paragraph position="1"> feature in both the nominal and its complement.</Paragraph>
      <Paragraph position="2"> This results in only two possibilities of definiteness combinations for constructs, as opposed to the four possible combinations of free genitives. The construct form is generated from the absolute form by means of a morphological process, modelled by a lexical rule (10). Apart from modifying the phonology 2 of the nominal, this process has a double effect. First, the rule picks a genitive complement from the COMP list, replaces it by a nominative noun phrase and unifies the values of the DEr: feature of the nominal and the complement it depends on. In addition, the rule sets the value of 'DEP' to this complement, to indicate that cs-nominals are prosodically dependent. When the nominal is combined with its complement, the resulting phrase inherits the definiteness from the latter. Notice that the results of this process, i.e., the lexical entries of csnouns, are not specified as 'DEF --' (in fact, they are not specified for definiteness at all), and hence the DLR cannot apply to them. The fact that cs-nominals cannot be rendered definite directly is naturally obtained.</Paragraph>
      <Paragraph position="3"> Noun-noun constructs are thus constructed by the head-complement schema. An independent csnoun, with no immediate complement, cannot be promoted to the status of a phrase, as the dependency principle prohibits its combination with other phrases until its DEP requirements axe discharged.</Paragraph>
      <Paragraph position="4"> Since the DEF value of the construct head and its complement are shared, and since DEF is a head feature, it is also shared by the mother; thus, the DEF feature of the phrase is inherited from the complement, as required. This process is depicted in (11); notice in particular how the definiteness of the phrase is inherited from the complement using a reentrancy in the head.</Paragraph>
      <Paragraph position="5"> The similar properties noun-noun and adjective-noun constructs suggest that they are actually only two instances of one process: any analysis that would suggest two different mechanisms to account for both phenomena is bound to be redundant. We simply extend the analysis of noun-noun constructs to cs-adjectives: such adjectives are lexically specified to subcategorize for nouns. They cannot occur independently, with no immediate complement, and hence are marked as dependent; the phrase is constructed through the head-complement schema (12).</Paragraph>
      <Paragraph position="6"> We thus obtain a uniform, principled account for the</Paragraph>
      <Paragraph position="8"/>
    </Section>
  </Section>
class="xml-element"></Paper>