Multilogues

A multilogue is a discourse with contributions from more than two agents, in contrast to a dialogue, in which there are always two agents. Multilogues are not considered in this thesis, nor is the planner capable of planning them, but a sketch of multilogue planning is given here. To plan a multilogue, the game tree is much the same as that used in the two-party case, with each agent adding a level using its plan rules. Consider a three-party multilogue. Assuming that the agents take orderly turns, agent 1 would plan level 1, agent 2 level 2, agent 3 level 3, agent 1 level 4 and so on. Evaluation of the game tree is somewhat different however. Since there are three agents, each agent maintains a model of not one other, but two other agents. This means that the nested belief model is a binary tree rather than a list. To compute the best play for the game tree in a two party dialogue, the evaluating agent would alternate between calling minimax at levels 1 and 2 as a path is followed in the game tree. In the three-party case, the agent alternates between its base beliefs, then the first of the nested models, then the second of the nested models, and then back to its base beliefs. In multilogues, turn taking may not pass in a circle, since speakers can address particular hearers. For the example of a question addressed to one hearer, and for the purpose of dry-land belief revision, the hearer who has been addressed should come to believe that the speaker wants its answer, whereas the hearers who have not been addressed should come to believe that the speaker wants the addressed hearer's answer, rather than theirs. This would not prevent an unaddressed hearer, who believes that the addressed hearer has the wrong answer, from attempting to take the floor. An example in a two-party case of agents holding the floor by considering the utility value of their own and the expected contribution of the other agent in a dialogue will be given later, in chapter 5.