<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0110">
  <Title>Similarities and Differences among Semantic Behaviors of Japanese Adnominal Constituents</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Pustejovsky (Pustejovsky, 1995) proposed the theory of a generative lexicon as a framework by which meanings of words are expressed in one unified representation. This kind ofgenerativity would be very useful for NLP, especially if it is applicable to the complex semantic structures represented by various modification relations.</Paragraph>
    <Paragraph position="1"> In our previous research on adjectives (Isahara and Kanzaki, 1999) we used Pustejovsky's theory to classify adjectives in Japanese. In this paper we take the first steps in a similar classification of the Japanese &amp;quot;noun + NO&amp;quot; construction. Bouillon (Bouillon, 1996) applied this theory to the adnominal constituent of mental states.</Paragraph>
    <Paragraph position="2"> Saint-Dizier (Saint-Dizier, 1998) discussed adjectives in French.</Paragraph>
    <Paragraph position="3"> Isahara and Kanzaki (Isahara and Kanzaki, 1999) treated a much wider range of phenomena of adnominal constituents. They classified the semantic roles of adnominal constituents in .Japanese. where many parts of speech act as adnominal constituents, and discussed a formal treatment of their semantic roles. In their research, adnominal constituents, mainly adjectives which function as adverbials, are discussed. The present paper describes the similarities and differences among adnominal constituents, i.e. adjectives and &amp;quot;noun + NO t (in English &amp;quot;of + noun&amp;quot;)&amp;quot; structures which have a broad range of semantic functions. This paper proposes an objective method for classifying these structures using a large amount of linguistic data. The feasibility of this was verified with a self-organizing semantic map based on a neural network model.</Paragraph>
    <Paragraph position="4"> In section 2, we explain the semantic functions performed by &amp;quot;noun + NO.&amp;quot; In section 3, we discuss how we can semi-automatically obtain and classify examples of adjectives and &amp;quot;noun + NO&amp;quot; structures which have similar semantic functions. In section 4, we introduce a self-organizing semantic map to verify the result of this classification. In section 5, we discuss similarities and differences between adjectives and &amp;quot;noun + NO&amp;quot; structures.</Paragraph>
  </Section>
class="xml-element"></Paper>