Using Generalised Dialogue Models to Constrain Information State Based Dialogue Systems

Abstract

While Information State (IS) based techniques show promise in the construction of flexible, knowledgebased dialogue systems, the many declarative rules that are used to encode Dialogue Theories often lead to opaque systems that are difficult to test and potentially unintuitive to users. In this paper, we advocate the application of explicitly defined Generic Dialogue Models (GDMs), encoded as recursive transition networks (RTNs), to the structuring of information state-based dialogue managers. To this end, we review the state of GDM approaches, comparing and contrasting them against the Dialogue Theories which are typically implemented using information state approaches. Furthermore, to support our approach, we present an extension of the ALPHA (A Language for Programming Hybrid Agents) language, which has been enhanced to support information state and GDM concepts directly.