DETECTING, REPAIRING, AND PREVENTING HUMAN--MACHINE MISCOMMUNICATION

AAAI '96 Workshop
Portland, OR
August 1996

Any system that communicates must be able to cope with the possibility of miscommunication---including misunderstanding, non-understanding, and misinterpretation:

• In misunderstanding, one participant obtains an interpretation that she believes is complete and correct, but which is, however, not the one that the other speaker intended her to obtain.
• In non-understanding, a participant either fails to obtain any interpretation at all, or obtains more than one interpretation, with no way to choose among them.
• In misinterpretation, the most likely interpretation of a participant's utterance suggests that their beliefs about the world are unexpectedly out of alignment with other's.

All three forms of miscommunication can eventually lead to repair in a dialogue; however, misinterpretations and non-understandings are typically recognized immediately, whereas a participant is not aware, at least initially, when a misunderstanding occurs. Additionally, misinterpretation can be a source of misunderstanding.

Successful communication requires that participants share considerable knowledge. For example, they must share some knowledge about the state of their interaction and about the physical and social situation in which they are communicating. Knowledge of their interaction includes the current topic under discussion (often a shared task), the focus of attention, and the relevance of each utterance to the previous interaction. In practice, no two participants start with an identical understanding of their task or of the situation---nor can they take the time to identify and resolve discrepancies beforehand. As a result, participants must be prepared to handle miscommunication during dialogue.

Research related to achieving robust interaction is an important subarea in Artificial Intelligence (AI). Early work concerned the correction of spelling or grammatical errors in a user's utterance so that the system could more easily match them against a fixed linguistic model; work has also been done in the area of speech recognition, attempting to find the best fit of a sound signal to legal sequences of linguistic objects. Other systems have attempted to detect misconceptions in the user's model of the domain of discourse. All of these approaches have assumed that the system's model is always correct. More recently, researchers have been looking at detecting and correcting errors in the system's model of an interaction. This work includes research on speech repairs, miscommunication, misunderstanding, non-understanding, and related work in planning, such as plan misrecognition and plan repair.

The focus of this workshop is to bring together researchers interested in developing theoretical models of robust interaction or in designing robust systems. Topics of interest include, but are not limited to, the following:

• Theories that delineate what knowledge must be represented, how it will be obtained and updated, and how responsibility for achieving robustness might be distributed among the interactants.

• Strategies for identifying potential causes of breakdowns, such as ambiguities, misconceptions, and plan misrecognition, in order to avert miscommunication.

• Strategies for identifying symptoms of actual breakdowns, such as deviations from expected behavior, unresolvable ambiguities, and speech errors.

• Techniques for correcting errors in interpretation that have been used in other areas of AI, such as plan recognition and computer vision, and in related areas, such as human-computer interaction and multimedia.

• Approaches to minimizing and correcting miscommunication in tutoring systems and education.

• Empirical data regarding the occurrence of miscommunication and approaches to robust communication that derive from empirical methods.

• Research in knowledge representation that would be useful in detecting, repairing, and preventing miscommunication.

We solicit papers that explore these issues, and papers that discuss implementations of solutions to the problems of detecting, repairing, and preventing human--machine miscommunication. Papers submitted to the workshop should address these topics explicitly.

COMMITTEE:

Susan McRoy, chair
(University of Wisconsin--Milwaukee)

Graeme Hirst
(University of Toronto)

Ronnie Smith
(East Carolina University)

Brad Goodman
(Mitre Corporation)

David Traum
(TECFA, Universite de Geneve)

Kathleen McCoy
(University of Delaware)

Susan Haller
(University of Wisconsin--Parkside)

SCHEDULE:

Submission deadline: March 18, 1996
Author notification: April 15, 1996
Camera-ready copy due: May 13, 1996
Conference dates: August 4--8, 1996

SUBMISSIONS:

Submit an extended abstract. Abstracts should not exceed 10 pages, exclusive of references, in 12 point, double-spaced text, with one-inch margins.

We strongly encourage electronic submissions, either plain text or postscript. Emailed submissions should be emailed to mcroy@cs.uwm.edu with a subject heading "ATTN: AAAI MNM"

In the event that electronic submission is not possible, send 6 copies to:

Susan McRoy
ATTN: AAAI MNM Workshop
Computer Science, University of Wisconsin--Milwaukee
3200 North Cramer Street, EMS Room 503
Milwaukee, WI 53211