<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1302">
  <Title>Hiroshi Tsujino +</Title>
  <Section position="8" start_page="14" end_page="15" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We constructed a multi-domain spoken dialogue system using an extensible framework. Domain selection in conventional studies is based on either the domain based on the speech recognition  +: These include 17 errors because of random selection when there were several domains having the same highest scores. results or the previous domain. However, we noticed that these conventional frameworks cannot cope with situations where neither of these domains is correct. Detection of such situations can prevent dialogues from staying in the incorrect domain, which allows our domain selection method to be robust against speech recognition errors. Furthermore, our domain selection method is also extensible. Our method does not select the domains directly, but, by categorizing them into three classes, it can cope with an increase or decrease in the number of domains. Based on the results of an experimental evaluation using 10 subjects, our method was able to reduce domain selection errors by 18.3% compared to a baseline method. This means our system is robust against speech recognition errors.</Paragraph>
    <Paragraph position="1"> There are still some issues that could make our system more robust, and this is included in future work. For example, in this study, we adopted a grammar-based speech recognizer to construct each domain expert easily. However, other speech recognition methods could be used, such as a statistical language model. As well, multiple speech recognizers employing different domain-dependent grammars could be run in parallel. Thus, we need to investigate how to integrate these approaches into our framework, without destroying the extensibility.</Paragraph>
  </Section>
class="xml-element"></Paper>