Rationality, Cooperation and Conversational Implicature

beliefs explaining the role the first set of beliefs play in the speaker's plan. Pollack's idea of ascribing a set of additional beliefs based on the intentions contained in the recognised plan is similar to our idea of ascribing additional conversational goals to explain why a plan is apparently inefficient. However, her motivation is to show how mistaken beliefs in dialogue can be recognised and dealt with and is, therefore more concerned with the validity of the agent's beliefs rather than the overall rationality of their plans. In addition, she is not concerned with how conversational implicature can be understood. Green and Carberry [Green and Carberry, 1992] provide an alternative analysis of indirect answers to yes-no questions. They represent turns in the modelled dialogue as plan operators encoding rhetorical structure theory relations. Such relations feature a nucleus which is the central communicative goal plus satellites which feature additional pieces of information which can be communicated. Green and Carberry argue that if a nucleus is missing but a satellite is given then the nucleus can be inferred as a conversational implicature. Green and Carberry's work provides a practical method of planning conversational implicature. However, we believe our work has the advantage that we derive implicature not from the higher level of rhetorical structures but a lower level of analysis based on beliefs, intentions and goals. This provides our approach with a more general method which can be applied to other classes of conversational implicature. In the sections above, we have shown a method of deriving conversational implicature based on the apparent irra-tionality of a speaker's plan behind his or her utterance. In our approach, conversational implicatures are not represented by the addition of semantic meaning but by the addition of conversational goals and intentions. Implicatures are recognised by the inefficiency of the speaker's dialogue plan. Such an inefficiency suggests that the utterance has a greater relevance to the speaker than just the initially recognised goal of the speaker. This greater relevance can be understood by additional goal ascription to the speaker's plan. It is an open question whether our intuitions fully capture plan rationality or cover the full set of conversational implicatures captured by Grice's Principle. We hope that further work involving corpus analysis with provide further insight. Given the System's assumption of rationality on the part of the Expert, there must be some ulterior reason why the Expert choose the …