<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0702">
  <Title>Topic Detection Based on Dialogue History</Title>
  <Section position="9" start_page="22" end_page="22" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> In this paper, we proposed a topic detection method using a dialogue history to select a scene for the automatic interpretation system. We investigated its limitation in dialogue utterances and provided solutions by clustering training data and utilizing dialogue history. Our method showed topic detection accuracy of at least 50% for both typical and real situation dialogues in 13 topic combinations. For typical dialogues, we found that the best results were obtained when one sentence is used for one cluster, and for real situation dialogues, we found slightly better results were obtained when clustering was introduced. Therefore, it can be argued that the topic detection accuracy is improved for both typical and real situation sentences if an appropriate size cluster is introduced.</Paragraph>
    <Paragraph position="1"> We plan to use our topic detection technique for specifying a scene condition of parallel text based translation in our automatic interpretation system. Detecting topics also helps improve accuracy of the automatic interpretation system by disambiguating polysemy. Topic detection can enhance speech recognition accuracy by selecting the correct word dictionary and resources, which are organized according to the topic.</Paragraph>
    <Paragraph position="2"> Our method is also applicable in determining time series behavior such as topic transition.</Paragraph>
    <Paragraph position="3"> Our future studies will focus on linking the dialogue history and clustering more closely to improve the topic detection accuracy.</Paragraph>
  </Section>
class="xml-element"></Paper>