Belief Modelling for Discourse Plans

The process of communication between agents is originated by the need of exchanging concepts. In natural language dialogues, the speaker tries to express his/her internal concepts into words and phrases; the hearer’s task consists of understanding the message, i.e. translating back from natural language into his/her internal representation (‘language of thought’) [1]. This description underlines an important fact: when generating natural language expressions, an agent supposes that the addressee is able to understand correctly the utterances. In other words, the speaker makes hypothesis about the hearer beliefs. In fact, (s)he is assuming that the hearer believes that the terms used have a specific ‘meaning’ associated, and, most importantly, (s)he assumes to know which are the specific meanings held in the hearer’s mind. This suggests that the process of natural language generation requires an agent to hypothesize a ‘model’ of, at least, the hearer beliefs and knowledge.