<?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="3" start_page="40" end_page="42" type="intro">
    <SectionTitle>
3. Results
</SectionTitle>
    <Paragraph position="0"> Chi square tests on the combined totals of all CGUs showed highly significant differences (p&lt;0.001) between initiations and responses for all four prosodic and timing parameters. This result is consistent with those reported in Stirling et al (2000b), where it was shown that initiation type dialogue acts could be prosodically differentiated from response type dialogue acts.</Paragraph>
    <Paragraph position="1"> Here we were more interested in the question of whether the complexity of the CGU was also reflected in initiation and response contributions.</Paragraph>
    <Paragraph position="2"> The results reported in the following subsections therefore refer to the percentage number of each prosodic and timing parameter as a proportion of the CGU type (percentages are in boldface) - the total number of simple CGUs or the total number of complex CGUs - for initiations and responses independently.</Paragraph>
    <Section position="1" start_page="40" end_page="41" type="sub_section">
      <SectionTitle>
3.1 Correspondences with Break Indices
</SectionTitle>
      <Paragraph position="0"> As expected, there was no significant difference between simple and complex CGUs with respect to their initiation contribution break indices.</Paragraph>
      <Paragraph position="1"> Examination of the first res~nse of these units did show a significant difference between simple  and complex CGUs with respect to break indices. In particular, the first responses for simple CGUs had a high proportion of BI3 (corresponding with an intermediate phrase boundary) compared with complex CGUs and a relatively low proportion of BI4 (corresponding with a full intonation phrase). That is, the first response contribution of a complex CGU was more likely to end with a full inttonafional phrase boundary (BI4) than the first response contribution of a simple CGU.</Paragraph>
    </Section>
    <Section position="2" start_page="41" end_page="42" type="sub_section">
      <SectionTitle>
3.2 Correspondences with Boundary Tones
</SectionTitle>
      <Paragraph position="0"> In contrast with the results for BI reported above, there were significant differences found between simple and complex CGUs for boundary tones in both initiation and response contributions. With respect to initiating contributions, a higher proportion of low falling (L-L%) boundary tones (42.1% vs. 29.5%) and a lower proportion of low rising (L-H%) boundary tones (18.5% vs. 29.9%) were found in the complex CGUs, compared with simple CGUs. Proportions of high rising (H-H%) boundary tones were not significantly different and there were proportionally few instances of other types of contours.</Paragraph>
      <Paragraph position="1"> Like initiation conlributions, the results for response contributions also show a higher proportion of low falling tones in complex CGUs than in simple CGUs (30.3% cf. 17.5%). The proportions of both high rising and low rising tones appears stable across the types of CGUs however.</Paragraph>
      <Paragraph position="2"> The numbers of response contn'butions which had other types of boundary tones (including no boundary tone) were much higher than those found for initiating units. This was expected, since responses to initiation units were typically acknowledgments of grounding (like 'okay' or 'yeh'), which were then followed by other talk by the same speaker continuing an intonational phrase. The higher proportion of such 'nonfinal' contours (H-L%, L-, H- and none) for simple CGUs than complex CGUs reflects a set of instances in which the respondent answers a yes-no question, and then makes a further response in the same intonational phrase that is new information which was itself grounded (and was therefore coded as a new CGU).</Paragraph>
      <Paragraph position="3"> These non-final contours typically also corresponded with break indices of less than 4, accounting for the relatively high proportion of BI3s in responses in complex CGUs (noted in the previous section).</Paragraph>
      <Paragraph position="4">  Like the results for break indices, the results for the timing of units with respect to immediately prior talk showed no significant differences between simple and complex CGUs for initiation contributions, but did show significant differences with respect to first responses. In particular, there was a relatively high proportion of responses in complex CGUs whose onset occurred while the other speaker was still talking, resulting in an overlap (18.5% vs. 7.5%). When the first response in a CGU is at least partially overlapping with the initiation contribution, it creates an environment in which both participants must work harder (ie. make more contributions) in order for the acknowledgment of common groundedriess to be clear. At least some of these overlaps resulted in the same speaker repeating an acknowledgment of grounding with no overlap, as in example (3) above. However, they also resulted in complex CGUs of the other kinds.</Paragraph>
      <Paragraph position="5"> 3.3.2. Correspondences with timing relative to following stretch of talk  The results for the timing of the unit following the target contribution also showed no significant difference for initiations between simple and complex CGUs, while the responses did show a difference of distribution. However, the asymmetry in overlapping found in the preceding set of the results did not occur when one considered the contribution foUowing the first response. In this case, the category which showed the greatest degree of difference between simple and complex CGUs was the category of 'latch' (22.5% vs. 12%), where the second speaker timed his/her next turn to follow from the target contribution without any pause or overlap. Responses in complex CGUs showed a higher proportion of latching with the following contribution. This high proportion of latching is perhaps an indication of a smooth transition from first response to timber grounding contributions within the CGU.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>