The organisation of purposeful dialogues

The first part of this paper describes a computer program which models a conversation between two robots. The robots co-operate to achieve a simple practical goal in a world of a few objects, and their co-operation is genuinely facilitated by a conversation in which every remark is ultimately related to the goal they are trying to achieve. They conduct several types of conversational exchange (e.g. agreeing plans, obtaining information), each exchange being controlled by a 'conversationalprocedure', which is a list of instructions to be performed by two agents, a 'first speaker' and a 'second speaker'. These procedures are called by the procedures which control the underlying planning. The second par t of the paper discusses some deficiencies of the program. In particular, it is argued that the robots lack an explicit representation of how utterances achieve their purposes, and with the aid of some recent philosophy of language a possible representation is suggested. Part I. A Computer Program Which Conducts Simple Conversations