<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1012">
  <Title>Generating Determiners and Quantifiers in Hebrew</Title>
  <Section position="5" start_page="92" end_page="93" type="metho">
    <SectionTitle>
3 Definiteness
</SectionTitle>
    <Paragraph position="0"> Definiteness in NP has two aspects: it corresponds to a complex semantic property of the thing referred to by the phrase and it is syntactically marked in different ways. We will not discuss how a noun phrase is defined as semantically definite (a development on this issue is provided in \[1\] Chap.5), and we assume that the decision that a referent is definite is taken by the content determination module and that the semantic definite feature is given in the input to the generation grammar. We keep the feature definite yes for this purpose.</Paragraph>
    <Paragraph position="1"> In a Hebrew definite NP, the definite marker (ha-) is agglutinated to the head and to most modifiers (adjectives and quantifiers). Not all subconstituents of the NP are marked, however: in a noun-compound construction the head loses the marking and the compounded noun keeps it. When demonstratives are used, a marked and an unmarked form can be used. In addition, pronouns and proper nouns are intrinsically definite and do not receive the ha- marking.</Paragraph>
    <Paragraph position="2"> To explain the distribution of the ha- marker, we have added the feature mark-definite which determines whether an article (definite or indefinite) is to be used for each subconstituent of the NP.</Paragraph>
    <Paragraph position="3"> By default, the mark-definite of a subconstituent is equal to the definite of its mother constituent. However, certain constructions and certain lexical items block this propagation and can force a mark-definite no (for example, proper nouns have a lexical feature of mark-definite no and the noun-compounding construction enforces a mark-definite no on the head).</Paragraph>
    <Paragraph position="4"> The following example illustrates how the feature mark-definite contrasts among the two distinct phrases:  In this example, nawar/boy is not marked with ha- because it enters in a compound construction with cfar. Similarly Toy/good is not marked because it is an adjective compound of Icy/heart.  While there is agreement as to the behavior of the marking of definiteness, the marking of indefinite nouns is quite controversial. Some researchers claim that definite is not a binary but that an additional unmarked value is necessary (cf \[2\] for a review). We mainly adopt Ornan's view \[10\]. In this view, an indefinite NP is marked with a null article by default, but the indefinite can also be marked. In this case, we use the feature (mark-definite indef). This feature explains the contrast between: eyze delet some door mark-definite indef delet door mark-definite no delet aHat door one selective yes We treat delet aHat as marked functionally as a selective phrase, following Halliday's feature as used in SURGE. In our analysis, selective yes implies mark-definite indef. Of the two marked indefinite articles, eyze is non selective while eHad is.</Paragraph>
    <Paragraph position="5"> The following example, taken from an example by Winograd, illustrates the behavior of the selective yes feature: Examining the cabinet, we noticed that a door was marred The Hebrew for it will be: ? cSe-badaknu et ha-aron, ra'ynu S-delet Svura while-checked-us OM the-closet, saw-us that-door broken cSebadaknu et ha-aron, ra'ynu S-delet aHat Svura while-checked-us OM the-closet, saw-us that-door one broken In this context, the existence of a set of 2 or 4 doors is known pragmatically (from world knowledge about closets). The function of the indefinite article is then to select one door out of this known set. In that case, the unmarked indefinite is quite inappropriate, because it would refer to a door that cannot be related to the closet.</Paragraph>
    <Paragraph position="6"> In summary, the complex marking of definiteness in Hebrew has led us to extend the single semantic feature definite yes/no used in SURGE, with the additional feature mark-definite yes/no/indef.</Paragraph>
  </Section>
  <Section position="6" start_page="93" end_page="94" type="metho">
    <SectionTitle>
4 Determining the Order
</SectionTitle>
    <Paragraph position="0"> We have mostly investigated the order of determiners and quantifiers in Hebrew that appear before the head of the NP. We have found that semantics plays a main role in determining the relative order of determiners. This follows Glinert (\[7\] Chap.8 and 9), who distinguishes among the different functions fulfilled by determiners, and predicts the ordering accordingly, as discussed in Sect.2.</Paragraph>
    <Paragraph position="1"> The primary functional distinction is between quantifiers - that express amount or portion and determiners - that express identity. Accordingly, the word col/all for example, has two senses: one as a quantifier (with the all or collective meaning) and one as a determiner (with the every or distributive meaning).</Paragraph>
    <Paragraph position="2"> Quantifiers are classified along two main dimensions: exact/inexact (e.g., three vs. many) and portion/amount. The portion/amount system has the most effect on the structure of the NP, as it can be realized either lexically by certain quantifiers, or by using an explicit partitive syntactic structure X m-Y/X of Y.</Paragraph>
    <Paragraph position="3"> We define amount quantlfiers (cardinals, quantifiers like harbeh/many, yoter/more) as quantifiers that express quantity without explicit relation to the &amp;quot;total&amp;quot; cardinality of the reference set 2 ~Although amount quantifiers are always semantically related to some reference set, this relation is presented  In contrast, partitive quantifiers express a quantity that gets its meaning only in relation to the cardinality of a reference set. For example, roy/most refers to at least half of the reference set. Similarly, col/all refers to all of the reference set, and is also partitive.</Paragraph>
    <Paragraph position="4"> The distinction among partitive quantifiers, amount quantifiers and determiners is important as it predicts the order of the words in the NP. The standard order is: \[partitive deZerminer amount head\] We also note that only partitives can enter into recursive structures.</Paragraph>
    <Paragraph position="5"> Accordingly, our input specification language enforces the constraint that only a single amount and a single identification feature can be present simultaneously. The realization grammar also uses the knowledge of which word realizes which function to determine the ordering. This is illustrated in the following example, where the standard order is obtained for the combination partitive amount head: cat np total + ref-set definite yes cardinal \[ value 10 \] col waseret ha-mafginym all ten the-demonstrators All the ten demonstrators Note that whenever a partitive quantifier is desired, the input specification must include a ref-set construct. This enforces the constraint that partitives yield recursive constructs, as shown in the following example: cat np wasarah me-col ha-mafginym cardinal value 10 \] ten of-all the-demonstrators total -t- 1 \[lex &amp;quot;rnafgyn&amp;quot; \] Ten of all the demonstrators ref-set ref-set definite yes J The decision to build an explicitly partitive construction X m- Y/X of Y is left to the realization grammar and can be quite complex. If a quantifier can be found that has the lexical property of being marked as partitive, then a non-explicitly partitive construction can be used, as in the first example above all the ten demonstrators. Note that in this case, in contrast to English, in Hebrew. the explicitly partitive construction is not possible * col me-waseret ha-mafginym. In contrast. if a feature modifying the ref-set is to be realized by a non-lexically partitive quantifier, then an explicitly partitive construction must be used.</Paragraph>
    <Paragraph position="6"> When the &amp;quot;portion&amp;quot; part is modified with adjectives, then an explicitly partitive construction must be used: ha-roy ha-gadol mi-beyn ha-yeladym the-most the-big of-from the-children The vast majority of the children In that case, the portion functions as the logical head of the NP with the realization of the reference set as a PP - which affects agreement with the verb.</Paragraph>
    <Paragraph position="7"> In summary, we propose a refined classification of the determiners/quantifiers in Hebrew into amount, partitive and determiner. This functional classification determines the relative order of the determiners within the NP and also determines which recursive structures can be constructed. Finally, it is important to take into account the lexical properties of specific determiners which can require or forbid the use of explicitly partitive constructions. Our input specification language encompasses all of the features found in SURGE but organizes them into a new framework which enforces these constraints (with the explicit specification of the ref-set in the input). implicitly for amounts: 3 children refers to 3 children out of some reference set, but the reference set is deliberately left implicit. 95</Paragraph>
  </Section>
class="xml-element"></Paper>