Making sense of corrupt input: contextual interpretation in LOQUI

LOQUI is the name of a natural language interface to databases developed during the course of the ESPRIT project LOKI. LOQUI is written in BIM prolog. The purpose of the paper is to illustrate LOQUI´s exploitation of context in its treatment of corrupt input. This is an important issue in LOQUI, since the system has been designed to have a rich dialogue capability that will allow conversations to proceed with the fluidity that is characteristic of natural human conversation, thereby allowing both system and user the possibility of exploiting contextual information in order to minimise constraints on expressiveness. This is used not only to overcome problems of corrupt input, of course, but also to allow the system to be more than just a question-and-answer system. The paper considers how LOQUI handles reference resolution, spelling correction, semantic deviance, and presupposition failure