This paper describes a proposal to improve the communication process and the engineering features in VoiceXML dialogues. This proposal is focused on the efficient and reusable representation of the knowledge involved in communication. In order to achieve a good dialogue design we propose the explicit representation of the conceptual information the application needs form the user in an ontology. A syntactic-semantic taxonomy is provided to facilitate the generation of the grammars and the system's messages. Using this separate, reusable knowledge organization, VoiceXML dialogues for different applications and in different languages can be generated semi-automatically.