Investigating a mixed-initiative dialogue management strategy

This paper discusses how to design mixed-initiative spoken dialogues with only a partial recognition of the user utterances (recognition of concepts or phrase spotting). The objective is to investigate the potential of such a technique and in particular to develop a corresponding dialogue model. The work has been carried out within the ESPRIT OVID project, which addressed a voice controlled home banking task. A number of experiments have been carried out within the project and the paper discusses the results of these. A mixed-initiative dialogue management model has been developed and implemented, and the experiments have shown that users to a very high degree are able to grab the initiative at natural points in the dialogue.