In human-human conversation, people use linguistic expressions that are flexibly tailored to context as a matter of course, and they expect their conversational partners to do likewise. Towards our goal of helping computers achieve natural linguistic interactions with human users, we have developed an application-independent dialogue system integrating the COLLAGEN collaboration manager with the SPUD natural language generation algorithm. In this paper, we show how this integration supports investigation into strategies for distractor set based generation in dialogue. We discuss, as a case study, our experience in using the integrated system to generate contextually appropriate noun phrases in an email and scheduling application. While we have achieved natural, context-sensitive output in certain cases, our work has also revealed several issues facing systems wishing to exploit distractor sets to fine-tune their linguistic output. We discuss our approach to handling these issues and suggest areas where further research is required.
Published Date: May 2004
Registration: ISBN 978-1-57735-201-3
Copyright: Published by The AAAI Press, Menlo Park, California.