<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-0317">
  <Title>Cue Phrase Selection in Instruction Dialogue Using Machine Learning</Title>
  <Section position="6" start_page="104" end_page="104" type="concl">
    <SectionTitle>
5 Conclusion and Further work
</SectionTitle>
    <Paragraph position="0"> This paper reported the results of using a machine learning algorithm for identifying learning features and obtaining decision trees for selecting cue phrases. It also reported the result of a quantitative evaluation of the decision trees learned. Learning features concerning three factors, discourse structure, task structure, and dialogue context, were examined. By carrying out many experiments in which the combinations of learning features were varied, we found the most simple and effective learning feature set. The accuracy of the best model that uses 6 learning features is about 70%. The error rate is reduced about 25% from the baseline. These results support the claims of previous studies that discourse structure influence cue selection. In addition, it is revealed that task structure and dialogue context are also indispensable factors.</Paragraph>
    <Paragraph position="1"> We focus on predicting the cue phrases that occur at the beginning of discourse segments for signaiing inter-segment &amp;quot;sequence&amp;quot; relation. Elhadad and McKeown (1990), on the other hand, has presented a model for distinguishing connectives, which link two propositions, using some pragmatic constraints. In (Moser and Moore, 1995a; Moser and Moore, 1995b), the relationship between placement and selection of cue phrases was investigated using the core:contributor relations among units within a segment (Moser and Moore, 1995a). Although we discussed only the &amp;quot;sequence&amp;quot; relation between the  segments, the methods presented here will be useful in extending our model so as to select other kinds of cue phrases.</Paragraph>
  </Section>
class="xml-element"></Paper>