Towards Learning Transferable Conversational Skills using Multi-dimensional Dialogue Modelling

Statistical approaches to dialogue management have brought improvements in robustness and scalability of spoken dialogue systems, but still rely heavily on in-domain data, thus limiting their crossdomain scalability. In this paper, we present a new multi-dimensional, statistical dialogue management framework, in which transferable conversational skills can be learnt by separating out domainindependent dimensions of communication. Our preliminary experiments demonstrate the effectiveness of such transfer.