<?xml version="1.0" standalone="yes"?>
<Paper uid="H93-1063">
  <Title>SESSION 11: PROSODY</Title>
  <Section position="5" start_page="315" end_page="315" type="concl">
    <SectionTitle>
3. IMPORTANT THEMES
</SectionTitle>
    <Paragraph position="0"> Several important and common themes, indicative of recent research trends, cut across subsets of these papers.</Paragraph>
    <Paragraph position="1"> First, it is significant that both synthesis and understanding applications of prosody are represented in this session, and useful since the developments in one field can benefit the other. Second, we see corpus-based analysis and automatic training methods being introduced into many aspects of prosody modeling. Third, Silverman's results argue the case for developing models in constrained domains, but this approach is also supported by the development of automatic training methods and probably used to advantage in the papers focussed on the ATIS domain. Fourth, all of the papers use an intermediate prosodic representation at some level, which raises the issue of representation as an important research question in its own right. Perhaps the most important contribution of this session is the collection of experimental results demonstrating the benefits of prosody in actual synthesis and understanding applications, providing concrete and not just anecdotal evidence that prosody is a useful component of a spoken language system. Since these themes represent relatively new directions in computational modeling of prosody, the applications and modeling possibilities are only beginning to open up and we can expect many more gains in the future.</Paragraph>
  </Section>
class="xml-element"></Paper>