<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2154">
  <Title>Translating a Unification Grammar with Disjunctions into Logical Constraints</Title>
  <Section position="2" start_page="0" end_page="934" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The objective of our research is to build a natural language understanding system that is based on unification. The reason we have chosen a unification-based approach is that it enables us to describe grammar declaratively, making the development and amendment of grammar easy.</Paragraph>
    <Paragraph position="1"> Analysis systems that are based on unification grammars can be classified into two groups from the viewpoint of the ways feature structures are represented: (a) those using labeled, directed graphs (Shieber, 1984) and (b) those using first-order terms (Pereira and Warren, 1980; Matsumoto et al., 1983; Tokunaga et al., 1991).</Paragraph>
    <Paragraph position="2"> In addition to internal representation, grammar formalisms can be classified into two groups, (i) those that describe feature structures with path equations and lists of pairs of labels and values (Mukai and Yasukawa, 1985; Ai't-Kaci, 1986; Tsuda, 1994), and (ii) those that describe feature structures with first-order terms (Pereira and Warren, 1980; Matsumoto et al., 1983; Tokunaga et * Presently with Japan Advanced Institute of Science and Technology. al., 1991). Since formalisms (i) are used in the family of the PATR parsing systems (Shieber, 1984), hereafter they will be called PATR-Iike formalisms.</Paragraph>
    <Paragraph position="3"> Most of the previous systems are either ones that generate representation (a) from formalisms (i) or ones that generate representation (b) from formalisms (ii).</Paragraph>
    <Paragraph position="4"> However, representation (b) is superior, and formalism (i) is far better. Representation (b) is superior for the following two reasons. First, unification of terms is more efficient of that of graphs because the data structure of terms is simpler (Sch6ter, 1993). l Second, it is easy to represent and process named disjunctions (DSrre and Eisele, 1990) in the term-based representation. Named disjunctions are effective when two or more disjunctive feature values depend on each other. The treatment of named disjunctions in graph unification requires a complex process, while it is simple in our logical-constraint-based representations. Formalism (i) is better because term-based formalism is problematic in that readers need to memorize the correspondence between arguments and features and it is not easy to add new features or delete features (Gazdar and Mellish, 1989).</Paragraph>
    <Paragraph position="5"> Therefore, it is effective to translate formalism (i) into representation (b). Previous translation methods 2 (Covington, 1989; Hirsh, 1988; SchSter, 1993; Erbach, 1995) are problematic in that they cannot deal with disjunctive feature descriptions, which reduce redundancies in grammar. Moreover, incorporating disjunctive information into internal representation makes parsing more efficient (Kasper, 1987; Eisele and DSrre, 1988; Maxwell and Kaplan, 1991; Hasida, 1986).</Paragraph>
    <Paragraph position="6"> This paper presents a method for translating grammar formalism with disjunctive information based on path equations and lists of pairs of labels and values into term-I Since unspecified features are represented by variables in term unification, when most of the features are unspecified, it is inefficient to represent feature structures by terms. In current linguistic theories such as HPSG (Pollard and Sag, 1994), however, thanks to the type specifications, the number of features that a feature structure can have is reduced, so it does not cause as much trouble.</Paragraph>
    <Paragraph position="7">  based representations, without expanding disjunctions.</Paragraph>
    <Paragraph position="8"> The formalism used here is feature-based formalism with disjunctively defined macros (FF-DDM), an extension of the PATR-Iike formalisms that incorporates a description of disjunctive information. The representation used here is logical-constraint-based grammar representation (LCGR), in which disjunctive feature structures are represented by Horn clauses.</Paragraph>
  </Section>
class="xml-element"></Paper>