Introduction

This chapter describes the desired functionality and the resulting design of the planner. The chapter begins by setting out requirements for the planner, which are chosen such that the motivations set out in chapter 1 are satisfied. It will be argued that a planning system based on the evaluation of Bayesian game trees satisfies the motivation of producing a planner that produces plans that are more efficient than those planners that have come before. Then some assumptions about the planning model, about the rationality of both the user and the planning agent, and about the cooperative setting of the dialogue are made. The design of the planner is then given. The structure of the planner's state is defined, the algorithm that computes its strategy is described, as are the algorithms that revise the agent's state as the dialogue progresses. A simple example will be used to illustrate these algorithms. The reader might be interested in looking at the Prolog source code for the planner, which is available on the world wide web at http://planeffdia.sourceforge.net/.