<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2317">
  <Title>Towards automatic addressee identification in multi-party dialogues</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Communication, between humans or between humans and conversational computer agents, involves addressing. Addressing has received attention in the tradition of conversation analysis (Clark and Carlson, 1992; Clark and Schaefer, 1992), but not that much in the community of computational dialogue systems. One exception is (Traum, 2003). An explanation for this lack of attention may be that most research in computational dialogue systems concerns systems that were designed for interaction between one human user and one conversational agent.</Paragraph>
    <Paragraph position="1"> In dialogues in which only two participants take part addressing goes without saying. Addressing becomes a real issue in multi-party conversations and that is the subject of this paper.</Paragraph>
    <Paragraph position="2"> There are a number of application areas that could benefit from studying addressing behavior in human human interactions. It can provide valuable data for learning more about human interaction and the way humans interact with intelligent environments. The result can be used by those who develop communicative agents in interactive intelligent environments, meeting managers and presentation assistants. If we could induce from recorded meetings the &amp;quot;who said what, when and to whom&amp;quot; we can use this information for making summarizations of meetings, and for real-time tracking.</Paragraph>
    <Paragraph position="3"> Research on small group discussions (Carletta et al., 2002) has shown that there is a noticeable difference in the interaction patterns between large and small groups (up to seven participants). A small group discussion looks like two-way conversations but conversations occur between all pairs of members and every member can initiate conversation. A large group discussion is more like a series of conversations between a group leader and various individuals with the rest participants present but silent. We will focus our research on small group discussions in meetings.</Paragraph>
    <Paragraph position="4"> In this paper we propose research that aims at the automatic determination of the addressee of a speaker in small meetings. Analysis of the mechanisms that people use in identifying their addressees leads to a model of a conversation that describes the features that play a role in these mechanisms. These features can be of several types: verbal, non-verbal, and features of the situation.</Paragraph>
    <Paragraph position="5"> Our research is partly based on analysis of the IDIAP multi-modal meeting data corpus made available through the Media File Server 1.</Paragraph>
  </Section>
class="xml-element"></Paper>