Anaphora for Limited Domain Systems

Abstract

This paper presents a simple mechanism for the resolution of anaphora in limited domain natural language systems. For such domains. this mechdnism provides functionality equivalent to th e natural communication mechanism of anaphora as used and understood by people. but without the deep inferencing or cognitive modelling required for full simulation of human performance. The mechanism covers simple pronoun anaphora, and set selection anaphora (e.g. "last one". "one before". " others"). It was developed to provide the most efficient and effective communication between system and user . even if this meant divergrng significantly from human performance when this performance was impractical to reproduce . In cases of radical divergence. we were careful to make the behaviour of the mechanism very simple and easy to predict. In this way, the user can rely either on his experience of human performance or his knowledge of the artificial, but simple. substitute to predict the behaviour of the system jn response to his inputs. An algorithmic description of an implemented version of the mechanism is presented. A similar approach to other aspects of man·machine interfaces is recommended as a promising way to address the problem of habitability that still plagues all natural language computer interfaces.