Pragmatics Research and Non-task Dialog Technology

Abstract

Interest is growing in dialog systems which engage users in conversations that are not entirely focussed on the immediate performance of clearly defined practical tasks. Interest is also growing in data-driven methods for dialog system design, with increasing focus on sequence-to-sequence deep learning models, inspired by success in the machine translation sphere. In this position paper, we discuss research on casual conversation or social talk and current methods in design of systems, highlighting areas which need reconciliation.