Total: 1
In this paper we describe a statistical shared plan-based approach to dialog modeling and dialog management. We apply this approach to a corpus of human-human spoken dialogs. We compare the performance of models trained on transcribed and automatically recognized speech, and present ideas for further improving the models.