Recent progress of spoken dialogue systems now allow the extension of their abilities, and a richer interaction will need the use of language generation techniques in order to output appropriated messages. Additionally, prosody can indicate speech act, modality, relative salience of concepts, or play discourse functions. Thus, the same words can be pronounced differently according to context, and system's intentions. We introduce a framework linking different communicative goals to different prosodic patterns. We present a two-level of commands for French intonation : the first level (phonological), allowing a fine control of low-level prosodic properties of utterances, the second level (conceptual) made of dialogue act, information structure and enunciation schemes. We then expose our solution to the problem of mapping those two levels introducing the phonostrategies. A phonostrategy is a plan consisting of rules which map the discourse function of prosody and the prosodic commands, like macro-commands that are applied to sequences of symbols representing the concepts of the utterance.