<?xml version="1.0" standalone="yes"?>
<Paper uid="W93-0216">
  <Title>Empirical Evidence for Intention-based Discourse Segmentation</Title>
  <Section position="6" start_page="62" end_page="62" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> Our study establishes that a naive notion of speaker intention serves as a reliM)le criterion for identifying discourse segments. Qualitative analysis of annotations of speaker intention supports the conclusion that where subjects agree on segment boundaries, they also agree on the segment's intention. Using the empirically validated seglnents, we can begin to ask specific questions about the relation between segments and their abstract representation in an evolving discourse model. We note that for spontaneous oral narrative, discourse pops 1nay not lie explicitly signaled by cue words, and that structural and semantic relations among distinct segnlents may instead require inference, ht \[Passonneau and Litman, 1993\], we directly address how explicit devices such as pauses, cue words and referential noun phrases correlate with seglnental structure in order to posit constraints between surface structure choices and intentional and seglnental structure.</Paragraph>
  </Section>
class="xml-element"></Paper>