<?xml version="1.0" standalone="yes"?>
<Paper uid="P89-1016">
  <Title>THE EFFECTS OF INTERACTION ON SPOKEN DISCOURSE</Title>
  <Section position="7" start_page="132" end_page="132" type="concl">
    <SectionTitle>
CONCLUSIONS
</SectionTitle>
    <Paragraph position="0"> In summary, the present results imply that near-term spoken language systems that are unable to provide meaningful and timely confirmations may not be able to curtail speakers' elaborations effectively, or the related discourse convolutions typical of noninteractive speech. Current dialogue and text processing systems are not prepared to handle this type of elaborative discourse. Clearly, new heuristics will need to be developed to accomodate speakers who try more than once to achieve their communicative goals, in the process using multiple utterances and varied speech acts. Under these circumstances, models of noninteractive speech may provide a more appropriate basis for designing near-term spoken language systems than either keyboard models or models of fully interactive dialogue.</Paragraph>
    <Paragraph position="1"> To model discourse accurately for interactive SLSs, further research will be needed to establish the generality of these noninteractive speech phenomena across different tasks and applications, and to determine whether speakers can be trained to alter these patterns. In addition, research also will be needed on the extent to which human-computer task-oriented speech differs from that between humans. At present, there is no well developed discourse theory of human-machine communication, and the few studies comparing human-machine with human-human communication have focused on the keyboard modality, with the exception of Hauptmann &amp; Rudnicky (1988). These studies also have relied exclusively on the Wizard of Oz paradigm, although this technique entails unavoidable feed-back delays due to the inherent deception, and it was never intended to simulate the interactional coverage of any particular system. Further work ideally would examine human-computer speech patterns as prototypes of interactive SLSs become available.</Paragraph>
    <Paragraph position="2"> In short, our present research findings imply that designers of future spoken language systems should be vigilant to the possibility that their selected application may elicit noninteractive speech phenomena, and that these patterns may have adverse consequences for the technology proposed. By anticipating or at least recognizing when they occur, designers will be better prepared to develop speech systems based on accurate discourse models, as well as ones that are viable ergonomically.</Paragraph>
  </Section>
class="xml-element"></Paper>