Paper: Evolving Optimal Inspectable Strategies For Spoken Dialogue Systems

ACL ID N06-2044
Title Evolving Optimal Inspectable Strategies For Spoken Dialogue Systems
Venue Human Language Technologies
Session Short Paper
Year 2006
Authors

We report on a novel approach to gener- ating strategies for spoken dialogue sys- tems. We present a series of experiments that illustrate how an evolutionary rein- forcement learning algorithm can produce strategies that are both optimal and easily inspectable by human developers. Our ex- perimental strategies achieve a mean per- formance of 98.9% with respect to a pre- defined evaluation metric. Our approach also produces a dramatic reduction in strategy size when compared with conven- tional reinforcement learning techniques (87% in one experiment). We conclude that this algorithm can be used to evolve optimal inspectable dialogue strategies.