DIA-MOLE: An Unsupervised Learning Approach to Adaptive Dialogue Models for Spoken Dialogue Systems

摘要

The DIAlogue MOdel Learning Environment supports an engineering-orientedapproach towards dialogue modelling for a spoken-language interface. Majorsteps towards dialogue models is to know about the basic units that are used toconstruct a dialogue model and possible sequences. In difference to many otherapproaches a set of dialogue acts is not predefined by any theory or manuallyduring the engineering process, but is learned from data that are available inan avised spoken dialogue system. The architecture is outlined and the approachis applied to the domain of appointment scheduling. Even though based on a wordcorrectness of about 70% predictability of dialogue acts in DIA-MOLE turns outto be comparable to human-assigned dialogue acts.