Recent Trends in Spoken Language Dialogue Systems

Summary form only given. We present an overview of our current research activities in the area of advanced spoken language dialogue systems (SLDSs). One of the key considerations is the use of appropriate system architectures, since the increasingly resource-consuming technology requires a careful investigation on where to perform the data processing. Another issue is that of appropriate decoding strategies. Current approaches to automatic speech recognition use statistical modelling techniques based on the automatic analyses of large corpora of speech data. Similarly, these techniques can be used at the natural language understanding level to classify the utterance, i.e., to convert the speech recogniser output into a semantic meaning representation. We also argue that SLDSs need to become more intelligent and empathic, i.e., they should be more collaborative to find solutions fitting both, the user requirements and application constraints. The support of multi-user communication seems to be of equal importance. In the case where several users communicate with each other the interface should be empathic and act as an autonomous dialogue partner and intervene proactively, i.e., track the dialogue and react as soon as it is required by the conversational situation. Finally, emotion recognition and adaptive dialogue management help to enhance the user-friendliness of SLDSs.