<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0207">
  <Title>Corpus-Based Anaphora Resolution Towards Antecedent Preference</Title>
  <Section position="6" start_page="51" end_page="51" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we proposed a corpus-based anaphora resolution method combining an automatic learning algorithm for coreferential relationships with statistical preference selection in the discourse context. We proved the applicability of our approach to pronoun resolution achieving a resolution accuracy of 86.0% (precision) and 75.9% (recall) for Japanese pronouns despite the limitation of sparse data. Improvements in these results can be expected by increasing the training data as well as utilizing more sophisticated linguistic knowledge (structural analysis of utterances, etc.) and discourse information (extra-sentential knowledge, etc.) which should lead to a rise of the decision tree filter performance.</Paragraph>
    <Paragraph position="1"> Preliminary experiments with nominal reference and ellipsis resolution showed promising results, too.</Paragraph>
    <Paragraph position="2"> We plan to incorporate this approach in multi-lingual machine translation which enables us to handle a variety of referential relations in order to improve the translation quality.</Paragraph>
  </Section>
class="xml-element"></Paper>