<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1152">
  <Title>Ecient Dialogue Strategy to Find Users' Intended Items from Information Query Results</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Dialogue Strategy in General
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Information Query Task
Interactioninaninformationquerytaskcanbe
</SectionTitle>
      <Paragraph position="0"> regarded as a process seeking a common part between the user's request and system knowledge. In order to help users to nd their intended items from the system knowledge, the system has to carry out not only interpreting what users say but also showing the relevant portionofthesystemknowledgetothem.</Paragraph>
      <Paragraph position="1"> We assume that users freely set and retract querykeys based ontheirpreference for informationquerysystems. Ifmanycandidatesstill remainevenafterspecifyingallpossiblehis/her preferencetothesystem, usersmayhave di-culty in narrowing down further the query result. Thus,thesystemshouldgenerateecient guiding questions to help users nd their intendeditems. null  Inthissection,wepresumethesystemknowledgeasapairofanitemandasetofkeywords null (Figure 1). We dene keywords as a set of words representing contents of the items, and their categories such as place, food and so on aregiven. Thisissimilartoindexingwordsin aconventionalinformationretrievaltask. Note thatitisnotneededthatthesystemknowledge isstructuredlikeanRDB.</Paragraph>
      <Paragraph position="2"> Keywords are extracted from a user's utterance,andarematchedwiththesystemknowl- null edge. Here, we adopt the following matching</Paragraph>
      <Paragraph position="4"> a condence measure of speech recognition for</Paragraph>
      <Paragraph position="6"> ,respectively.</Paragraph>
      <Paragraph position="7"> Then, we deneamount of informationthat is obtained when the system generates yes/no questionandtheuseranswersit. Here, C isa current query condition, A is a condition that isaddedbythesystem'squestion,andcount(x) is the numberof items that satisfy the condition x. The conditionconsists of the conjunctionofthekeywordstheuserspecied. Suppose each item occurs by equal likelihood, the fol- null We weight on each item j with the likelihood</Paragraph>
      <Paragraph position="9"> The amount of information that is obtained when the user's answer is \yes&amp;quot; is represented asfollows.</Paragraph>
      <Paragraph position="11"> The following equation gives H(A), the expected value of amount of information that is obtained by generating a question about condition A and getting user's answer (\yes&amp;quot; or \no&amp;quot;).</Paragraph>
      <Paragraph position="13"> By calculating H(A) for all conditions A that can be added to the current query condition, thesystemgeneratesthequestionthathasthe maximumvalueof H(A). Thequestionisgenerated using the category information of each keyword.</Paragraph>
      <Paragraph position="14"> Becausetheobtainedcondition A isselected by a viewpointof narrowingdown the current set of items eciently, the selected condition maybeunimportantfortheuser. Insuchacase, it isnot cooperative to force the user an armativeornegativereply. Oursystemdoesnot forcethereluctantdecisionbyallowingtheuser to say \It doesnot matter anyhow.&amp;quot;. Instead, thesystempresentsthesecondbestproposal.</Paragraph>
      <Paragraph position="15">  Weexplainthemethodwiththefollowingexample in our restaurant query system in the  ciently. If the user thinks that the condition \noodles with boiled-pork-ribs&amp;quot; is not important and tells the system so (for example \Eitherwilldo.&amp;quot;),thesystemcanshowthesecond null best proposal, \Would you like one located in Kabukichoarea?&amp;quot;. Thus, the queryresultcan  of task knowledge is available. The task here is to nd the appropriate item in the manual  ofelectricapplianceswithaspokendialogueinterface. Suchaninterfacewillbeusefulasthe recent appliances become complex with many features and so are their manuals. In the appliances such as VTR (Video Tape Recorder) andFAX machines, thereisnota largescreen todisplaythelistofmatchedcandidatestobe selected by the user. Therefore, we address a spokendialoguestrategytodeterminethemost appropriateonefromthelistofcandidates.</Paragraph>
      <Paragraph position="16">  Analternativesystemdesignistheuseofdirectory search, as adopted in voice portal systems, where the documents are hierarchically  structuredandthesystempromptsuserstoselect one of the menu from the top to the leaf. Themethodisrigidandnotuser-friendlysince users often have trouble in selection and want to specify by their own expression. The proposedsystemallowsuserstomakequeriesspon- null  taneouslyandmakesuseofthedirectorystructureinthefollow-updialoguetodeterminethe null mostappropriateone.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.1 System Overview
</SectionTitle>
      <Paragraph position="0"/>
      <Paragraph position="2"> The inverse document frequency (idf) is weighted with a condence measure CM</Paragraph>
      <Paragraph position="4"/>
    </Section>
  </Section>
  <Section position="4" start_page="0" end_page="3" type="metho">
    <SectionTitle>
3. Generatingdialoguetodeterminethemost
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
appropriateonefromthelistofcandidates
</SectionTitle>
      <Paragraph position="0"> As a result of the matching, many candidatesareusuallyfound. Theymayinclude irrelevant ones because of speech recognitionerrors. Butitisnotpracticaltoread  outalloftheminorderwithaTTS(textto-speech) system. Therefore, dialogue is invoked to narrow down to the intended one. Thisdialogueisrestrictedtosysteminitiated \yes/no&amp;quot; questions in order to play record search setting</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.2 Dialogue Strategy using Structure
of Manual
Ifoneofthecandidatesismoreplausiblethan
</SectionTitle>
      <Paragraph position="0"> under a threshold (). This means the numberofcandidatesthatcanbepresented tousers. Here,weset=3.</Paragraph>
      <Paragraph position="1"> Theseareexperimentallycomparedinthenext subsection.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="3" type="sub_section">
      <SectionTitle>
3.3 Experimental Evaluation
3.3.1 Task and System Implementation
</SectionTitle>
      <Paragraph position="0"> The proposed system is implemented for the query task on a VTR manual that consists of  ios(querysentencesarenotgiven),andseveral spontaneous querieswithoutany scenarios. In total, we had 195 query utterances, of which  ),respectively.</Paragraph>
      <Paragraph position="1"> We have not yet identied the reason why performance by the apparently most accurate function h  is not good. We conjuncture that the dierence of the cost functions does not matter so much in this framework as long as theyarereasonable.</Paragraph>
    </Section>
    <Section position="4" start_page="3" end_page="3" type="sub_section">
      <SectionTitle>
3.3.3 Evaluation with Speech Input
</SectionTitle>
      <Paragraph position="0"> Next, we made experiments using the spoken queriesandthespeechrecognitionsystem. The distributionof recognized keywords and corre- null proposedcostfunctions 2.9 2.9 3.2 dencemeasuresisbetter, thusthecondence measure works well. Summaryof the result is  matching accuracy is lowered with the speech  input,theimprovementbytheproposedstrategyislargerandthenumberofdialogueturns null isclosetothetext-inputcase. Theresultconrms that the proposed framework is eective inspeechinterface.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>