Introduction

Abstract

Natural language dialogue is a continuous, unified phenomenon, not simply a series of unrelated sentences and phrases. Unfortunately, most computational systems for interpreting natural language treat individual utterances as isolated events. A comprehensive computational system for participating in natural language dialogue should be designed so that it can apply a well-integrated set of operations in a uniform fashion, with a continuous context that easily reaches across individual utterances. This work describes such a system, Psli3 (pronounced “sly 3”), and the analysis behind it, concentrating primarily on the computational architecture required for uniform processing of contextual phenomena such as elliptical utterances.