<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1107">
  <Title>Example-based Speech Intention Understanding and Its Application to In-Car Spoken Dialogue System</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Outline of Example-based
Approach
</SectionTitle>
    <Paragraph position="0"> Intentions of a speaker would appear in the various types of phenomenon relevant to utterances, such as phonemes, morphemes, keywords, sentential structures, and contexts. An example-based approach is expected to be effective for developing the system which can respond to the human's complicated and diverse speeches. A dialogue corpus, in which a tag showing an utterance intention is given to each sentence, is used for our approach. In the below, the outline of our method is explained by using an inference example.</Paragraph>
    <Paragraph position="1"> Figure 1 shows the flow of our intention inference processing for an input utterance &amp;quot;Chikaku-ni chushajo-wa aru-ka-na ? (Is there a parking lot nearby?)&amp;quot;. First, morphological analysis and dependency analysis to the utterance are carried out.</Paragraph>
    <Paragraph position="2"> Then, the degree of similarity of each input utterance with sentences in the corpus can be calculated by using the degree of correspondence since the information on both morphology and dependency are given to all sentences in the corpus in advance. In order to raise the accuracy of the intention inference, moreover, the context information is taken into consideration. That is, according to the occurrence probability of a sequence of intentions learned from a dialogue corpus with the intention tags, the degree of similarity with each utterance is weighted based on the intentions of the last utterances. Consequently, if the utterance whose degree of similarity with the input utterance is the maximum is &amp;quot;sono chikaku-ni chushajo arimasu-ka? (Is there a parking lot near there?)&amp;quot;, the intention of the input utterance is regarded as &amp;quot;parking lot question&amp;quot;.</Paragraph>
  </Section>
class="xml-element"></Paper>