<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-1123">
  <Title>Linear Segmentation and Segment Significance</Title>
  <Section position="4" start_page="203" end_page="203" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we have shown how multi-paragraph text segmentation can model discourse structure by addressing the dual problems of computing topical text segments and subsequently assessing their significance. We have demonstrated a new algorithm that performs linear topical segmentation in an efficient manner that is based on linguistic principles. We achieve a 10% increase in accuracy and recall levels over prior work (Hearst 1994, 1997). Our evaluation corpus exhibited a weak level of agreement among judges, which we believe correlates with the low level of performance of automatic segmentation programs as compared to earlier published works (Hearst 1997).</Paragraph>
    <Paragraph position="1"> Additionally, we describe an original method to evaluate a segment's significance: a two part metric that combines a measure of a segment's generality based on statistical approaches, and a classification of a segment's function based on empirical observations. An evaluation of this metric established its utility as a means of extracting key sentences for summarization.</Paragraph>
  </Section>
class="xml-element"></Paper>