Generation of collaborative spoken dialogue contributions in dynamic task environments

Abstract

We explain generation methods in a dialogue system al-lowing humans and complex devices or applications to collaborate in real-time dialogue about ongoing activ-ities in dynamic environments. The generation com-ponent must be able to handle contexts where there are multiple topics being co-ordinated by the conver-sation and where world and system-states can vary independently. We describe turn-management, truth-checking, “relevance”-checking, and the incremental message selection, aggregation, and generation meth-ods employed in this context. We demonstrate that these techniques are viable in a demonstration dia-logue system for multi-modal conversations with semi-autonomous mobile robots.