<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-2027">
  <Title>Dialog Navigator : A Spoken Dialog Q-A System based on Large Text Knowledge Base</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> When we use personal computers, we often encounter troubles. We usually consult large manuals, experts, or call centers to solve such troubles. However, these solutions have problems: it is difficult for beginners to retrieve a proper item in large manuals; experts are not always near us; and call centers are not always available. Furthermore, operation cost of call centers is a big problem for enterprises. Therefore, we proposed a spoken dialog Q-A system which substitute for call centers, based on only large text knowledge base.</Paragraph>
    <Paragraph position="1"> If we consult a call center, an operator will help us through a dialog. The substitutable system also needs to make a dialog. First, asking backs for fixing speech recognition errors are needed. Note that too many asking backs make the dialog inefficient. Secondly, asking backs for clarifying users' problems are also needed, because they often do not know their own problems so clearly.</Paragraph>
    <Paragraph position="2"> To realize such asking backs, we developed a system as shown in Figure 1. The features of our system are as follows: AF Precise text retrieval.</Paragraph>
    <Paragraph position="3"> The system precisely retrieves texts from large  text knowledge base provided by Microsoft Corporation (Table 1), using question types, products, synonymous expressions, and syntactic information. Dialog cards which can cope with very vague questions are also retrieved.</Paragraph>
    <Paragraph position="4"> AF Dialog for fixing speech recognition errors.</Paragraph>
    <Paragraph position="5"> When accepting speech input, recognition errors are inevitable. However, it is not obvious which portions of the utterance the system should confirm by asking back to the user.</Paragraph>
    <Paragraph position="6"> A great number of spoken dialog systems for particular task domains, such as (Levin et al., 2000), solved this problem by defining slots, but it is not applicable to large text knowledge base. Therefore, we introduce two measures of confidence in recognition and significance for retrieval to make dialogs for fixing speech recognition errors.</Paragraph>
    <Paragraph position="7"> AF Dialog for clarifying vague questions.</Paragraph>
    <Paragraph position="8"> When a user asks a vague question such as &amp;quot;An error has occurred&amp;quot;, the system navigates him/her to the desired answer, asking him/her back using both dialog cards and extraction of  summaries that makes differences between retrieved texts more clear.</Paragraph>
    <Paragraph position="9"> Our system makes asking backs by showing them on a display, and users respond them by selecting the displayed buttons by mouses.</Paragraph>
    <Paragraph position="10"> Initially, we developed the system as a keyboard based Q-A system, and started its service in April 2002 at the web site of Microsoft Corporation. The extension for speech input was done based on the one-year operation. Our system uses Julius (Lee et al., 2001) as a Japanese speech recognizer, and it uses language model acquired from the text knowledge base of Microsoft Corporation.</Paragraph>
    <Paragraph position="11"> In this paper, we describe the above three features in Section 2, 3, and 4. After that, we show experimental evaluation, and then conclude this paper.</Paragraph>
  </Section>
class="xml-element"></Paper>