Clarification in task-oriented dialogues

In this project, we investigate how people react to impossible, ambiguous, and implausible instructions in a task-oriented dialogue setting. In order to examine this issue, we will conduct an experiment (on English) where participants receive various instructions to place furniture onto a given ground plan. This will be done prior to the meeting. The aim of the experiment is to elicit clarification questions from participants in four different situational contexts:

• instruction to place a non-existing piece of furniture on the ground plan (impossible reference condition)
• instruction to place an ambiguously described piece of furniture on the plan (ambiguous reference condition, e.g. "the sofa" when there is a green and a blue sofa available)
• instruction to place a piece of furniture on an already occupied spot (impossible placement condition)
• instruction to place a piece of furniture into a room where it would not be expected (implausible placement condition, e.g. "the bed into the kitchen")

During the meeting, we will first read some background literature on clarification and generation in dialogue. We will then analyze the experimental data and try to establish correlations between situational contexts (experimental conditions) and clarification questions' forms (syntax, intonation) and meanings.

Depending on participants and their background/interests, we can also focus more on the actual generation of clarification questions for spoken dialogue systems.

Preliminary references:

Matthew Purver, Jonathan Ginzburg and Patrick Healey. On the Means for Clarification in Dialogue. To appear in Current and New Directions in Discourse and Dialogue, Kluwer Academic Publishers, 2002. (available at http://www.dcs.kcl.ac.uk/pg/purver/publications.html)

Staffan Larsson (2003): Generating Feedback and Sequencing Moves in a Dialogue System. In Freedman and Callaway (eds.): Natural Language Generation in Spoken and Written Dialogue - Papers from the 2003 AAAI Symposium. AAAI Press, Menlo Park, California. (available at http://www.ling.gu.se/~sl/papers.html)