<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2318">
  <Title>Prosodic Cues to Discourse Segment Boundaries in Human-Computer Dialogue</Title>
  <Section position="9" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> Based on analysis of more than 450 discourse segment boundary pairs, we found significant increases in maximum pitch, average pitch, and average intensity for segment-initial utterances, with a significant decrease in minimum pitch for segment-final utterances. Consistent with prior work on human monologue, new discourse segments in human-computer dialogue are signaled by increases in pitch, contrastive use of pitch range, and loudness, cues which could serve to attract the attention of the other conversational participants.</Paragraph>
    <Paragraph position="1"> In future work, we plan to apply these features to automatic extraction of discourse boundaries and global discourse structure. These features could also be used in conjunction with phonetic recognition results to enhance confidence scoring for utterances that would cause a topic shift. In systems such as SpeechActs where topic shift often signals an application change, a somewhat time-consuming activity as a new recognizer is swapped in and new data loaded, it is desirable to have additional implicit confirmation that such an action has in fact been requested. Finally we hope to explore cues to more fine-grained hierarchical discourse structure to distinguish full topic shifts from initiation or completion of subdialogues.</Paragraph>
  </Section>
class="xml-element"></Paper>