<?xml version="1.0" standalone="yes"?>
<Paper uid="J99-2004">
  <Title>Supertagging: An Approach to Almost Parsing</Title>
  <Section position="2" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> In this paper, we present a robust parsing approach called supertagging that integrates the flexibility of linguistically motivated lexical descriptions with the robustness of statistical techniques. The idea underlying the approach is that the computation of linguistic structure can be localized if lexical items are associated with rich descriptions (supertags) that impose complex constraints in a local context. This makes the number of different descriptions for each lexical item much larger than when the descriptions are less complex, thus increasing the local ambiguity for a parser. However, this local ambiguity can be resolved by using statistical distributions of supertag co-occurrences collected from a corpus of parses. Supertag disambiguation results in a representation that is effectively a parse (an almost parse).</Paragraph>
    <Paragraph position="1"> In the linguistic context, there can be many ways of increasing the complexity of descriptions of lexical items. The idea is to associate lexical items with descriptions that allow for all and only those elements on which the lexical item imposes constraints to be within the same description. Further, it is necessary to associate each lexical item with as many descriptions as the number of different syntactic contexts in which the * 180 Park Avenue, Florharn Park, NJ 07932. E-mail: srini@research.att.com t Department of Computer and Information Sciences and Institute for Research in Cognitive Science, University of Pennsylvania, Philadelphia, PA 19104. E-maih joshi@linc.cis.upenn.edu (~) 1999 Association for Computational Linguistics Computational Linguistics Volume 25, Number 2 lexical item can appear. This, of course, increases the local ambiguity for the parser.</Paragraph>
    <Paragraph position="2"> The parser has to decide which complex description out of the set of descriptions associated with each lexical item is to be used for a given reading of a sentence, even before combining the descriptions together. The obvious solution is to put the burden of this job entirely on the parser. The parser will eventually disambiguate all the descriptions and pick one per lexical item, for a given reading of the sentence. However, there is an alternate method of parsing that reduces the amount of disambiguation done by the parser. The idea is to locally check the constraints that are associated with the descriptions of lexical items to filter out incompatible descriptions. 1 During this disambiguation, the system can also exploit statistical information that can be associated with the descriptions based on their distribution in a corpus of parses.</Paragraph>
    <Paragraph position="3"> We first employed these ideas in the context of Lexicalized Tree Adjoining grammars (LTAG) in Joshi and Srinivas (1994). Although presented with respect to LTAG, these techniques are applicable to other lexicalized grammars as well. In this paper, we present vastly improved supertag disambiguation results--from previously published 68% accuracy to 92% accuracy using a larger training corpus and better smoothing techniques. The layout of the paper is as follows: In Section 2, we present an overview of the robust parsing approaches. A brief introduction to Lexicalized Tree Adjoining grammars is presented in Section 3. Section 4 illustrates the goal of supertag disambiguation through an example. Various methods and their performance results for supertag disambiguation are discussed in detail in Section 5 and Section 6. In Section 7, we discuss the efficiency gained in performing supertag disambiguation before parsing. A robust and lightweight dependency analyzer that uses the supertag output is briefly presented in Section 8. In Section 9, we will discuss the applicability of supertag disambiguation to other lexicalized grammars.</Paragraph>
  </Section>
class="xml-element"></Paper>