CLARIE: Handling Clarification Requests in a Dialogue System

This paper sets out a approach to clarification requests (CRs) general enough to cover all the major forms found in corpus data and specific enough to analyse the questions they ask about individual words and phrases. Its main features are a view of utterances as contextual abstracts with a radically abstracted semantic representation, and a view of CRs as standard utterances asking standard questions, but showing a particular kind of contextual dependence. It shows how it can be implemented computationally within a prototype text-based dialogue system, CLARIE, allowing it not only to generate CRs to clarify unknown reference and learn new words, but also to interpret and respond to user CRs, with both capabilities integrated within the standard dialogue processes and governed by empirical evidence.