<?xml version="1.0" standalone="yes"?>
<Paper uid="E06-1009">
  <Title>Information Presentation in Spoken Dialogue Systems</Title>
  <Section position="8" start_page="71" end_page="71" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper, we have shown that information presentation in SDS can be improved by an approach that combines a user model with structuring of options through clustering of attributes and successive refinement. In particular, when presented with dialogues generated by a system that combines user modelling with successive refinement (UMSR) and one that uses refinement without reference to a user model (SR), participants reported that the combined system provided them with a better overview of the available options and that they felt more certain to have been presented with all relevant options. Although the presentation of complex tradeoffs usually requires relatively long system turns, participants were still able to cope with the amount of information presented. For some dialogues, subjects even felt they could access relevant options more quickly despite longer system turn length.</Paragraph>
    <Paragraph position="1">  Infuturework,wewouldliketoextendtheclustering algorithm to not use a fixed number of target clusters but to depend on the number of natural clusters the data falls into. We would also like to extend it to be more sensitive to the user model when forming clusters (e.g., to be more sensitive at lower price levels for a user for whom price is very important than for a user who does not care about price).</Paragraph>
    <Paragraph position="2"> The explicit presentation of tradeoffs made by the UMSR system in many cases leads to dialogue turns that are more complex than typical dialogue turns in the SR system. Even though participants did not report that our system was harder to understand, it would be interesting to investigate how well users can understand and remember information from the system when part of their concentrationisabsorbedbyanothertask, forexamplewhen using the system while driving a car.</Paragraph>
  </Section>
class="xml-element"></Paper>