<?xml version="1.0" standalone="yes"?>
<Paper uid="P05-1031">
  <Title>Towards Finding and Fixing Fragments: Using ML to Identify Non-Sentential Utterances and their Antecedents in Multi-Party Dialogue</Title>
  <Section position="10" start_page="252" end_page="253" type="concl">
    <SectionTitle>
6 Conclusions and Further Work
</SectionTitle>
    <Paragraph position="0"> We have presented a machine learning approach to the task of identifying fragments and their antecedents in multi-party dialogue. This represents a well-defined subtask of computing discourse structure, which to our knowledge has not been studied so far. We have shown that the task of identifying the antecedent of a given fragment is learnable, using features that provide information about the structure of the discourse between antecedent and fragment, and about semantic closeness.</Paragraph>
    <Paragraph position="1"> The other tasks, identifying fragments and the combined tasks, however, did not perform as well, mainly because of a high rate of confusions between general non-sentential utterances and frag- null ments (in our sense). In future work, we will try a modified approach, where the detection of fragments is integrated with a classification of utterances as backchannels, fragments, or full sentences, and where the antecedent task only ranks pairs, leaving open the possibility of excluding a supposed fragment by using contextual information. Lastly, we are planning to integrate our classifier into a processing pipeline after the pronoun resolution step, to see whether this would improve both our performance and the quality of automatic meeting summarisations.9 null</Paragraph>
  </Section>
class="xml-element"></Paper>