Partitioning POMDPs for multiple input types, and their application to dialogue managers

Abstract

This research looks at the POMDP models used for dialogue management within Spoken Dialogue Systems (SDS). In particular, it examines the difficulty of handling multimodal inputs. It proposes a generalisation of the POMDP model in order to tackle this. This model is shown, via a Car Advisory system, to improve tractability for multimodal inputs.