Intentions are an important concept in Artificial Intelligence and Cognitive Science. We present a formal theory of intentions...

Indirect speech acts, like the request Do you know the time?, have both a literal meaning, here “I ask you whether you know the time, ” and an indirect meaning “I request you to tell me the time. ” In this paper I outline a model of how listeners understand such speech acts and plan responses to them. The main proposals are these. The literal meaning of indirect speech acts can be intended to be taken seriously (along with the indirect meaning) or merely pro forma. In the first case listeners are expected to respond to both meanings, as in Yes, I do-it’s six, but in the second case only to the indirect meaning, as in It’s six. There are at least six sources of information listeners use in judging whether the literal meaning was intended seriously or pro forma, as well as whether there was intended to be any indirect meaning. These proposals were supported in five experiments in which ordinary requests for information were made by telephone of 950 local merchants. Most sentences can be used to convey meanings indirectly. IS Julia at home? can be used in its literal sense to ask a question, a direct speech act. On the telephone it can also be used as a request to call Julia to the

The article surveys a little of the history of the technology, sets out the main current theoretical approaches in brief, and discusses the on-going opposition between theoretical and empirical approaches. It illustrates the situation with some discussion of CONVERSE, a system that won the Loebner prize in 1997 and which displays features of both approaches. 1 Introduction Practical and theoretical investigations into the nature and structure of human dialogue have been a topic of research in artificial intelligence and the more human areas of linguistics for decades: there has been much interesting work but no definitive or uncontroversial findings. The best performance overall in HMC (Human-machine conversation) has almost certainly been Colby's PARRY program [1] since its release on the (then ARPA) net around 1973. It was robust, never broke down, always had something to say and, because it was intended to model paranoid behaviour, its zanier misunderstandings could always be ...

Linguistic Resources for Text Planning. In Proceedings of the Fifth International Natural Language Generation Workshop, Dawson, Pa., June 3 6, 1990, pp62 68.

this dissertation research. Her current inter- ests are robust models of plan recognition and the application of user models to the generation of helpful responses

Preparation of this paper was supported by the under contract N00039-79-C-0118 with the Naval expressed are those of the author.