<?xml version="1.0" standalone="yes"?>
<Paper uid="E06-1024">
  <Title>Keeping the initiative: an empirically-motivated approach to predicting user-initiated dialogue contributions in HCI</Title>
  <Section position="3" start_page="0" end_page="185" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> It is increasingly recognised that human-computer dialogue situations can benefit considerably from mixed-initiativeinteraction(Allen, 1999). Interaction where there is, or appears to be, little restriction on just when and how the user may make a dialogue contribution increases the perceived naturalness of an interaction, itself a valuable goal, and also opens up the application of human-computer interaction (HCI) to tasks where both system and user are contributing more equally to the task being addressed.</Paragraph>
    <Paragraph position="1"> Problematic with the acceptance of mixed-initiative dialogue, however, is the radically increased interpretation load placed on the dialogue system. This flexibility impacts negatively on performance at all levels of system design, from speech recognition to intention interpretation. In particular, clarification questions initiated by the user are difficult to process because they may appear off-topic and can occur at any point. But preventing users from posing such questions leads to stilted interaction and a reduced sense of control over how things are proceeding.</Paragraph>
    <Paragraph position="2"> In this paper we pursue a partial solution to the problem of user-initiated contributions that takes its lead from detailed empirical studies of how such situations are handled in human-human interaction. Most proposed computational treatments of this situation up until now rely on formalised notions of relevance: a system attempts to interpret a user contribution by relating it to shared goals of the system and user. When a connection can be found, then even an apparently off-topic clarificationcanbeaccomodated. Inourapproach,  weshowhowthesearchspaceforrelevantconnections can be constrained considerably by incorporating the generic conversation analytic principle of recipient design (Sacks et al., 1974, p727). This treats user utterances as explicit instructions for how they are to be incorporated into the unfolding discourse--an approach that can itself be accomodated within much current discourse semantic work whereby potential discourse interpretation is facilitated by drawing tighter structural and semantic constraints from each discourse contribution (Webber et al., 1999; Asher and Lascarides, 2003). We extend this here to include constraints and conditions for the use of clarification subdialogues. null Our approach is empirically driven throughout. In Section 2, we establish to what extent the principles of recipient design uncovered for natural human interaction can be adopted for the still artificial situation of human-computer interaction. Although it is commonly assumed that results concerning human-human interaction can be applied to human-computer interaction (Horvitz, 1999), there are also revealing differences (Amalberti et al., 1993). We report on a targetted comparison of adopted dialogic strategies in natural human interaction (termed below HHC: human-human communication) and human-computer interaction (HCI). The study shows significant and reliable differences in how dialogue is being managed. In Section 3, we interpret these results with respect to their implications for recipient design.</Paragraph>
    <Paragraph position="3"> The results demonstrate not only that recipient design is relevant for HCI, but also that it leads to specific and predictable kinds of clarification dialogues being taken up by users confronted with an artificialdialoguesystem. Finally, inSection4, we discuss the implications of the results for dialogic  system design in general and briefly indicate how the required mechanisms are being incorporated in our own dialogue system.</Paragraph>
  </Section>
class="xml-element"></Paper>