<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1052">
  <Title>Charting the Depths of Robust Speech Parsing</Title>
  <Section position="8" start_page="411" end_page="411" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> The approach to the robust analysis of spoken language input, that we have described above, exhibits three crucial properties.</Paragraph>
    <Paragraph position="1"> 1. The restrictive parser is given the maximum opportunity of finding a correct analysis for a grammatical sequence of word hypotheses, where this exists.</Paragraph>
    <Paragraph position="2"> 2. The robustness component assembles partial analyses as a fallback, if no grammatical sequence can be found.</Paragraph>
    <Paragraph position="3"> 3. Almost arbitrary time constraints can be supported. Though, obviously, more processing time would usually improve the results. null The latter property depends directly on the chart-like data structures used at each level of processing. Whether it be the input WHG, VHG for robust processing or, most significantly, the parser's chart; each is formally a directed acyclic graph and each permits a selection of the best intermediate result at, virtually, any stage in processing, for a given evaluation function.</Paragraph>
    <Paragraph position="4"> The relatively efficient processing of WHG input achieved by parsing and robustness components working in parallel depends quite heavily on the successive processing of ranked WHG paths, effectively as alternative input strings.</Paragraph>
  </Section>
class="xml-element"></Paper>