Avoiding and Resolving Initiative Conflicts in Dialogue

24189

Authors

nan

Abstract

In this paper, we report on an empirical study on initiative conflicts in human-human conver- sation. We examined these conflicts in two corpora of task-oriented dialogues. The re- sults show that conversants try to avoid initia- tive conflicts, but when these conflicts occur, they are efficiently resolved by linguistic de- vices, such as volume. 1

Keywords

nan

Full Content

Proceedings of NAACL HLT 2007, pages 17–24, Rochester, NY, April 2007. c⃝2007 Association for Computational Linguistics Avoiding and Resolving Initiative Conflicts in Dialogue∗ Fan Yang and Peter A. Heeman Center for Spoken Language Understanding OGI School of Science & Engineering Oregon Health & Science University {fly,heeman}@cslu.ogi.edu Abstract In this paper, we report on an empirical study on initiative conflicts in human-human conver- sation. We examined these conflicts in two corpora of task-oriented dialogues. The re- sults show that conversants try to avoid initia- tive conflicts, but when these conflicts occur, they are efficiently resolved by linguistic de- vices, such as volume. 1 Introduction Current computer dialogue systems tend to be system- initiative. Although there are some mixed-initiative sys- tems that allow the user to make a request or state a goal, such systems are limited in how they follow natural ini- tiative behavior. An example is where the system always releases the turn whenever the user barges in. However, in a complex domain where the computer system and hu- man user are collaborating on a task, the computer sys- tem might need to interrupt the human user, or might even need to fight with the human user over the turn. Thus the next generation of computer dialogue systems need a better model of initiative (Horvitz, 1999). In what situations can the system try to take initiative from the user? What devices can the system use to fight for ini- tiative? We propose examining human-human conversa- tion to answer these questions. Once we understand the conventions people adopt in negotiating initiative, we can implement them in a computer dialogue system to create natural interactivity. In this research work, we examined two corpora of human-human conversation: the Trains corpus (Heeman and Allen, 1995) and the MTD corpus (Heeman et al., 2005). The research purpose is to understand conver- sants’ behavior with initiative conflicts, which we define a situation where both conversants try to direct the con- versation at the same time, but one of them fails. We ∗This work was funded by the National Science Foundation under IIS-0326496. found that (1) conversants try to avoid initiative con- flicts; and (2) initiative conflicts, when they occur, are efficiently resolved by linguistic devices, such as volume. In Section 2, we review related research work on mod- eling initiative and turn-taking. Dialogue initiative and turn-taking are two intertwined research topics. When conversants fight to show initiative, they are also fighting for the turn to speak. In Section 3, we describe the two corpora and their annotations. In Section 4, we define initiative conflict and give an example. In Section 5, we present the evidence that conversants try to avoid initia- tive conflicts. In Section 6, we present evidence that ini- tiative conflicts are efficiently resolved by linguistic de- vices. We discuss our findings in Section 7 and future work in Section 8. 2 Related Research 2.1 Initiative Models Researchers have been investigating how people man- age dialogue initiative in their conversation. Whittaker and Stenton (1988) …

Content truncated to 500 words

Labeled Text

In this paper, we report on an empirical study on initiative conflicts in human-human conversation. We examined these conflicts in two corpora of task-oriented dialogues. The results show that conversants try to avoid initiative conflicts, but when these conflicts occur, they are efficiently resolved by linguistic devices, such as volume. 1

Publication Details

Paper ID:
24189
Year:
2007
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan