<?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="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In the past years, a great number of spoken dialogue systems have been developed. Their   modelspeechunderstandingprocessasconvertingrecognitionresultsintosemanticrepresenta- null  tionsequivalenttodatabasequery(SQL)commands,anddialogueprocessasdisambiguating null theirunxedslots. Usually,thesemanticslots aredenedaprioriandmanually. Theapproach  isworkableonlywhendatastructureoftheapplication is well-organized typically as a relationaldatabase(RDB). null Dierent and more exible approach is needed for spoken dialogue interfaces to accessinformationdescribedinlessrigidformat, null in particular normal text database. For the purpose,informationretrieval(IR)techniqueis usefultondalistofmatchingdocumentsfrom the input query. Typically, keywords are extractedfromthequeryandstatisticalmatching null is performed. Call routing task (Chu-Carroll and Carpenter, 1998) can be regarded as the specialcase.</Paragraph>
    <Paragraph position="1"> In IR systems, many candidates are usually obtainedasa queryresult, thusthere isa signicant problem of how to nd the user's intendeditemamongthem. Especially,eitheron the telephone or electrical appliances, there is notalargescreendisplayingthecandidates,and all the query results cannot bepresented to a user. Soitisdesirableforthesystemtonarrow downthequeryresultsinteractively. Moreover, interactivequeryismorefriendlytonoviceusers ratherthanrequiringthemtoinputadetailed queryfromthebeginning.</Paragraph>
    <Paragraph position="2"> In this paper, we address a dialogue strategy to nd the user's intended item from the retrieved result, which is initiated by a spontaneous query utterance. In section 2, we describeamethodtogenerateaguidingquestion null thatnarrowsdownthequeryresultseciently, using an example of a restaurant query task.</Paragraph>
    <Paragraph position="3"> The question is selected based on an informationtheoreticcriterion. Insection3,wepresent adialoguemanagementmethodforaquerytask ontheappliancemanualwherestructuredtask knowledge is available. We propose a conrmationstrategybymakinguseofatreestruc- null  tureofthemanual,anddenethreecostfunctionsforselectingquestionnodes. Themethod isevaluatedbythenumberofaveragedialogue turns.</Paragraph>
    <Paragraph position="4"> Although there are previous studies on optimizing dialogue strategies (Niimi and Kobayashi, 1996;; Levin et al., 1997;; Litman et al., 2000), most of them assume the tasks of llingsemantic slots that are denitelyand manually dened, and few focus on follow-up dialogueofinformationretrieval. Forexample, (Denecke,1997)proposedamethodtogenerate  guidingquestionsbymakinguseofatreestructure constructed by unifying retrieved items based on semantic slots. In this paper, we do notassumeanystructureofsemanticslots. Instead,wemakeuseofdistributionofdocument null statisticsorastructureoftaskknowledge. We also investigate cost functionsfor optimaldialogue control by taking into account of speech recognitionerrors.</Paragraph>
  </Section>
class="xml-element"></Paper>