Response Assertiveness in Human-Computer Dialogues

This paper describes an attempt to determine expectations in human-computer dialogue through an experiment in which human subjects predict the responses of another human or a computer to natural language input. The experiment consists of a questionnaire based on discourse patterns that exhibit slight differences in their syntactic structures, and which, as a consequence, represent distinct levels of linguistic meaning. The specific speech acts defined by these levels: directives, assertions, requests, etc., are used as a basis of categorization for a series of multiple choice questions and answers which are designed to measure a subject's relative predispositions toward human and computer speakers. Results of the test are compared to intuitive expectations and then analysed in terms of potential application to a natural language processing system.