|
Conversational interfaces allow human users to interact with computer systems using spoken language in order to retrieve information and perform problem-solving tasks. A crucial part of developing such a system is devising a conversational strategy. This strategy is effectively a mapping between anticipated user inputs and appropriate system outputs. In this paper we present some preliminary experiments that show how XCS can generate optimal conversational strategies. We also explore the effect of the magnitude and structure of reward functions on the strategies learned by XCS.
|
