<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1035">
  <Title>Classifying Ellipsis in Dialogue: A Machine Learning Approach</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion and Further Work
</SectionTitle>
    <Paragraph position="0"> In this paper we have presented a machine learning approach to bare sluice classi cation in dialogue using corpus-based empirical data.</Paragraph>
    <Paragraph position="1"> From these data, we have extracted a set of heuristic principles for sluice disambiguation and formulated such principles as probability weighted Horn clauses. We have then used the predicates of these clauses as features to annotate an input dataset, and ran two different machine learning algorithms: SLIPPER, a rule-based learning algorithm, and TiMBL, a memory-based learning system. SLIPPER has the advantage of generating transparent rules that closely resemble our Horn clause constraints. Both algorithms, however, perform well, yielding to similar success rates of approximately 90%. This shows that the features we used to formulate our heuristic principles were well motivated, except perhaps for the feature frag, which does not seem to have a signi cant predictive power. The two algorithms we used seem to be well suited to the task of sluice classi cation in dialogue on the basis of these features.</Paragraph>
    <Paragraph position="2"> In the future we will attempt to construct an automatic procedure for annotating a dialogue corpus with the features presented here, to which both machine learning algorithms apply. null</Paragraph>
  </Section>
class="xml-element"></Paper>