The Machinery of Natural Conversation and the Design of Conversational Machines

• In Event: 298 - Section on Ethnomethodology and Conversation Analysis Paper Session and Business Meeting. New Directions in EM/CA Research

Abstract

A new wave of conversation-based, virtual agents has emerged in recent years among major computer and Internet companies, as well as startups. With persistent Internet connections and statistical algorithms, virtual agents are much smarter today than they were 20 years ago. Unlike graphical user interfaces, with their windows and icons, conversational interfaces rely heavily on natural language, either typed or spoken. Building such interfaces requires not only an understanding of the systematics of language, but also of the systematics of conversation. Rather than starting from scratch, UI designers can draw on the literature of Conversation Analysis for the latter. For over 50 years, conversation analysts have been documenting in detail how naturally occurring forms of talk and action are organized and specifying these organizations into formal, qualitative models, such as 'expandable adjacency pairs' and '4-position repair.' In this paper, we describe our efforts, within IBM Research and the IBM Watson Group, to begin to integrate findings from CA into the design of Watson-powered conversational applications.

Author

• Robert J. Moore, IBM Research - Almaden