<?xml version="1.0" standalone="yes"?>
<Paper uid="P87-1009">
  <Title>Phrasal Analysis of Long Noun Sequences</Title>
  <Section position="5" start_page="62" end_page="62" type="relat">
    <SectionTitle>
5. Related Work
</SectionTitle>
    <Paragraph position="0"> The problem of noun sequences of the kind common in technical fields like digital system specification has received only limited treatment in the literature. Winograd (Winograd, 1972) presents a more general discussion of Noun Groups, but the type of utterances his system expects does not include extended sequences of nouns as are common in our domain. Winograd therefore does not address the specific ambiguity problems raised here.</Paragraph>
    <Paragraph position="1"> Gershman's Noun Group Parser (NGP) (Gershman, 1979) dealt, among other things, with multiple noun sequences. While our algorithm is consistent with his, our approach differs from NGP in major respects. NGP contains what amount to several different programs for various types of noun groups, while we treat the information needed to analyze these structures as data.</Paragraph>
    <Paragraph position="2"> PHRAN embodies a general approach to language analysis that does not require components specialized to different types of utterances. A clear separation of processing strategies from knowledge about the language has numerous advantages that have been listed elsewhere (Arens, 1986). In addition, our treatment of noun groups as a whole is integrated into PHRAN and not a separate module, as NGP is.</Paragraph>
    <Paragraph position="3"> In evaluating the two systems, however, one must keep in mind that the choice of domain greatly influences the areas of emphasis and interest in language analysis. NGP is capable of handling several forms of noun groups that we have not attempted to deal with.</Paragraph>
    <Paragraph position="4"> Marcus (1979) describes a parsing algorithm* for long noun sequences of the type discussed in this paper. It is interesting to note that the limited lookahead added to the original PHRAN for the purpose of noun sequence recognition is consistent with Marcus' three-place constituent buffer. The major difference between Marcus' algorithm and ours is that the former requires a semantic component that can judge the relative &amp;quot;goodness&amp;quot; of two possible noun-noun modifier pairs. For * Discovered by Finin (Ig80) to be erroneous in some ca.ses.</Paragraph>
    <Paragraph position="5"> example, given the expression transfer block Mzc, this component would be responsible for determining whether block size is semantically superior to transfer block.</Paragraph>
    <Paragraph position="6"> Such a powerful component is not necessary for achieving our present objective - recognizing the presence and boundaries of a noun sequence.</Paragraph>
    <Paragraph position="7"> Our heuristic does not require it.</Paragraph>
    <Paragraph position="8"> A complementary but largely orthogonal effort is the complete semantic interpretation of long noun sequences. There have been several attempts to deal with the problem of producing a meaning representation for a given string of nouns. See (Finin, 19~0) and (Reimold, 1976) for extensive work in this area, and also (Brachman, 1978) and (Borgida, 1975). Such work by and large assumes that the noun sequence has already been recognized as such. I.e., it requires the existence of a component much like the one described in this paper from which to receive a noun sequence for processing. null</Paragraph>
  </Section>
class="xml-element"></Paper>