Pragmatics in human-computer conversations

Abstract

This paper provides a pragmatic analysis of some human-computer conversations carried out during the past six years within the context of the Loebner Prize Contest, an annual competition in which computers participate in Turing Tests. The Turing Test posits that to be granted intelligence, a computer should imitate human conversational behavior so well as to be indistinguishable from a real human being. We carried out an empirical study exploring the relationship between computers' violations of Grice's cooperative principle and conversational maxims, and their success in imitating human language use. Based on conversation analysis and a large survey, we found that different maxims have different effects when violated, but more often than not, when computers violate the maxims, they reveal their identity. The results indicate that Grice's cooperative principle is at work during conversations with computers. On the other hand, studying human-computer communication may require some modifications of existing frameworks in pragmatics because of certain characteristics of these conversational environments. Pragmatics constitutes a serious challenge to computational linguistics. While existing programs have other significant shortcomings, it may be that the biggest hurdle in developing computer programs which can successfully carry out conversations will be modeling the ability to 'cooperate'. © 2002 Elsevier Science B.V. All rights reserved.