<?xml version="1.0" standalone="yes"?>
<Paper uid="H92-1029">
  <Title>An Analogical Parser for Restricted Domains</Title>
  <Section position="7" start_page="153" end_page="153" type="concl">
    <SectionTitle>
6. FURTHER WORK
</SectionTitle>
    <Paragraph position="0"> While current results for this parsing model look promising, there are several directions of further exploration.</Paragraph>
    <Paragraph position="1"> Integration in Speech Recognition. There are two obvious ways of incorporating this parser into the speech recognition task. First, it can be used to select among a set of candidate sentences proposed by a recognizer.</Paragraph>
    <Paragraph position="2"> The second, more interesting, approach is to embed the parser in the recognition process. Given the parser's localization of information and its deterministic beginningto-end processing, it can naturally be used to find a locally (where the domain of locality is adjacent trees) optimal path through an (appropriately sparse) lattice.</Paragraph>
    <Paragraph position="3"> Development of Further Processing. This parser rests on the assumption, shared in a variety of recent work from quite different perspectives \[1, 3, 4\], that a level of underspeeified syntactic description is efficiently obtainable and is useful. The current work supports a particular view of what partial syntactic descriptions are obtainable. It remains to show that the further processing components can be constructed to make these pieces useful.</Paragraph>
    <Paragraph position="4"> Implementation Details. A number of decisions in the implementation of the current parser are arbitrary, and further development demands exploring the optimal design. For example, we need to explore what the similarity function should look like, and what function should be used for comparing potential attachments.</Paragraph>
  </Section>
class="xml-element"></Paper>