<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-1058">
  <Title>COORDINATION IN RECONNAISSANCE-ATTACK PARSING*</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
0. Introduclion
</SectionTitle>
    <Paragraph position="0"> Coordinate structures are difficult to parse in part because of the problem ol determining, in a given case, what kinds of constituents are being coordinated. The examples in (1) will illustrate: null  (1) a. John hits Fred and the other guys.</Paragraph>
    <Paragraph position="1"> b. John hits Fred and the other guys attack him. c. When John hits Fred and the other guys attack  him.</Paragraph>
    <Paragraph position="2"> Many variations on this theme are possible, to the point where serious doubts are raised regarding the efficacy in this domain of convention;tl parsers of either the top-down or bottom-up variety. In such parsers, it is necessary either to invoke backtracking to undo the effects of incorrect hypotheses or to store large numbers of alternatives until local indetermlnacies are resolved. In this paper, we will suggest an alternative approach based on the 'Recotmaissance-Attack' model described in Kac et al. 1986 (and more fully :in Rindflesch forthcoming), designed to skirt many of the problems associated with more traditional designs.</Paragraph>
    <Paragraph position="3"> *The work presented here was supported under Control Data Corporation Grant #86M102 to the University of Minnesota (Jeanette Gundel, Larry Hutchinson and Michael Kac, Principal Investigators). Special thanks are due to Nancy Hedberg and Karl Swingle for their xssistance on the project, and to Walling Cyre, technical liaison with CDC. The authors are listed in alphabetical older.</Paragraph>
    <Paragraph position="4"> Our proposal is theoretical in two senses. On the one hand, it does not present a detailed picture of an actual parsing algorithm, being intended rather to show that a significant body of linguistic data supports the contention that rapid, early resolution of local structural indeterminacies of the kind exemplified in (1) is feasible in the vast majority of cases. On the other hand, it is also based on a significant idealization, namely that each word belongs to only one syntactic category. Our intent is, in part, to show the applicability to a difficult parsing problem of a technique which can be found in other AI domains (Kowalski ! 979) but which seems to have been little exploited in work on natural language processing 1.</Paragraph>
  </Section>
class="xml-element"></Paper>