<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-1033">
  <Title>Flexible Guidance Generation using User Model in Spoken Dialogue Systems</Title>
  <Section position="6" start_page="1" end_page="1" type="evalu">
    <SectionTitle>
4.2 Results
</SectionTitle>
    <Paragraph position="0"> All of the subjects successfully completed the given task, although they had been allowed to give up if the system did not work well. Namely, the task success rate is 100%.</Paragraph>
    <Paragraph position="1"> Average dialogue duration and the number of turns in respective cases are shown in Table 3.</Paragraph>
    <Paragraph position="2"> Though the users had not experienced the system at all, they got accustomed to the system very rapidly. Therefore, as shown in Table 3, both the duration and the number of turns were decreased obviously in the latter half of the experiment in either group. However, in the initial half of the experiment, the group 1 completed with significantly shorter dialogue than group 2. This means that the incorporation of the user models is effective for novice users. Table 4 shows a ratio of utterances for which the skill level was identified as high. The ratio is calculated by dividing the number of utterances that were judged as high skill level by the number of all utterances in the eight sessions. The ratio is much larger for group 1 who initially used the system with user  group 1 with UM 0.72 (with UM ! w/o UM) w/o UM 0.70 group 2 w/o UM 0.41 (w/o UM ! with UM) with UM 0.63  was judged as high models. This fact means that novice users got accustomed to the system more rapidly with the user models, because they were instructed on the usage by cooperative responses generated when the skill level is low. The results demonstrate that cooperative responses generated according to the proposed user models can serve as good guidance for novice users.</Paragraph>
    <Paragraph position="3"> In the latter half of the experiment, the dialogue duration and the number of turns were almost same between the two groups. This result shows that the proposed models prevent the dialogue from becoming redundant for skilled users, although generating cooperative responses for all users made the dialogue verbose in general. It suggests that the proposed user models appropriately control the generation of cooperative responses by detecting characters of individual users.</Paragraph>
  </Section>
class="xml-element"></Paper>