Handling out of domain topics by a conversational character

  • Authors:
  • Manish Mehta;Andrea Corradini

  • Affiliations:
  • Georgia Institute of Technology, Atlanta, GA;University of Southern Denmark, Kolding, Denmark

  • Venue:
  • Proceedings of the 3rd international conference on Digital Interactive Media in Entertainment and Arts
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

Unlike traditional task-oriented dialog systems, one of the goals of conversational applications is to keep the user engaged and interested for as long as possible. This paper describes a technique that we implemented for dealing with the problem of out-of-domain topics and successfully integrated in a fully functional conversational system. In our approach, we exploit a simple and freely available coarsegrained ontology, namely Google's directory' structure, in order to automatically categorize unknown words and combines it with the existing ontological properties and dialog acts to create an automated semantic representation. We also resort to available web-based question and answering (QA) systems to generate answers to user's questions that are outside the domains covered by the application. Further, we make a first attempt at categorizing the retrieved information to generate appropriate non-verbal behaviors synced up with spoken utterances. The evaluation of the complete system shows promising results for the future and suggests that we are on the right track in terms of our approach for a better handling of out-of-domain topics.