COST278 and ISCA Tutorial and Research Workshop (ITRW) on Robustness Issues in Conversational Interaction

University of East Anglia, Norwich, UK
August 30-31, 2004

Flexible Knowledge Representation for Robust Dialogue Management

Melita Hajdinjak, France Mihelic

Laboratory of Artificial Perception, Systems and Cybernetics, Faculty of Electrical Engineering, University of Ljubljana, Slovenia

We give the findings and the results of applying the PARADISE framework to the data from Wizard-of-Oz experiments of a developing, bilingual, spoken natural-language dialogue system for weather-information retrieval. An important conclusion is that directing the user to select relevant, available data makes a significant contribution to user satisfaction when dealing with a sparse and dynamical information source that has a timedependent data structure. Therefore, a flexible knowledge representation that will handle this kind of data is needed. We propose a knowledge representation using two relations between temporarily available pieces of information, i.e., the partial order relation set inclusion and the symmetric and reflexive neighbourhood relation, which enable the data model to relate as many data objects as needed and thus give the dialogue manager a greater robustness in considering and offering relevant data.

Full Paper

Bibliographic reference.  Hajdinjak, Melita / Mihelic, France (2004): "Flexible knowledge representation for robust dialogue management", In Robust2004, paper 23.