<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1021">
  <Title>Improving Pronoun Resolution Using Statistics-Based Semantic Compatibility Information</Title>
  <Section position="7" start_page="171" end_page="171" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> Our research focussed on improving pronoun resolution using the statistics-based semantic compatibility information. We explored two issues that affect the utility of the semantic information: statistics source and learning framework. Specifically, we proposed to utilize the web and the twin-candidate model, in addition to the common combination of the corpus and single-candidate model, to compute and apply the semantic information.</Paragraph>
    <Paragraph position="1"> Our experiments systematically evaluated different combinations of statistics sources and learning models. The results on the newswire domain showed that the web-based semantic compatibility could be the most effectively incorporated in the twin-candidate model for the neutral pronoun resolution. While the utility is not obvious for personal pronoun resolution, we can still see the improvement on the overall performance. We believe that the semantic information under such a configuration would be even more effective on technical domains where neutral pronouns take the majority in the pronominal anaphors. Our future work would have a deep exploration on such domains.</Paragraph>
  </Section>
class="xml-element"></Paper>