<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-1005">
  <Title>Identifying Prosodic Indicators of Dialogue Structure: Some Methodological and Theoretical Considerations</Title>
  <Section position="4" start_page="42" end_page="43" type="ackno">
    <SectionTitle>
4. Discussion
</SectionTitle>
    <Paragraph position="0"> The results show that for three of the parameters investigated (break index, timing before and after the target contribution), there was little or no variation according to the type of CGU that the contribution initiated (simple or complex). This finding was consistent with our hypothesis that if there were any measurable differences between these CGU types, they would be found in the first response contn'bution of the CGU, and not in its initiation contn'bution. Also consistent with our hypothesis was the result that all four prosodic and .timing  parameters displayed differences between simple and complex CGUs in the first response contribution. Complex CGUs displayed more overlapped first responses, fi~wer incomplete intonation phrases (131 of less; than 4) and a higher proportion of low falling boundary tones than the response contributions of simple CGUs. Explanations for some of these results were provided in the previous section, but further analysis is required before we can really gain an accurate picture of the formal profile of complex CGUs (cf. simple CGUs). In particular, we require an analysis of the different types of complex CGUs to determine whether they display formal regularities. It 1nay be that only one or two types of complex CGUs are contributing to the patterns observed here. Such an analysis is planned in future work. We also plan to conduct a similar study that takes into account the dialogue act type of initiation and first responses to determine more extensively the full range of factors which influence intonation patterns and the timing of contributions in dialogue, iv Our results do confirm the necessity of developing strategies for the analysis of prosody in dialogue that take into account the effects of sequence. Not only did initiation contributions overall behave differently to response contributions with respect to prosody and timing (a result which we did not address in detail here), but the responses themselves could also be differentiated based on whether they 'finished' a CGU, or whether they were tbllowed by other contributions before grounding was achieved.</Paragraph>
    <Paragraph position="1"> Results such as these suggest that with respect to the formal identification (cPS pragmatic identification) of levels of dialogue structure higher than the dialogue act, structures defined in terms of sequential position, as such as adjacency pairs, might be more compatible with patterns we have observed in our data. The low level of intercoder reliability for CGUs, as reported in Core et al (1999) also suggests the lack of readily identifiable tbrrnal properties. Even taking 'intonation and timing' into account, as we have done here, does not clearly point to a formal profile of CGUs independent of sequences of dialogue acts.</Paragraph>
    <Paragraph position="2"> Empirical investigation\ of the role of prosody in multi-speaker discourse is still in its infancy, and we are still developing tools of analysis that free us from looking only at those aspects of dialogue that best mirror the structure of monologic discourse. We have found that our method of both prosodic and discourse segment annotation in ESPS/Waves+ has provided us with the power to develop this investigation further. In future work we will be able to easily incorporate additional variables, such as dialogue acts, complex CGU type, and a range of other prosodic phenomena into our analysis, while keeping track of the sequences in which these joint actions occur.</Paragraph>
  </Section>
class="xml-element"></Paper>