Human-machine dialogue aims at providing natural dialogue in natural language, i.e., allowing the user to speak using his own language (natural language), following a structure of moves and exchanges that is similar to human dialogue structure (natural dialogue). Related research works feed on linguistics, which analyses language markers, and pragmatics, which analyses language use in context. Two important sectors of linguistic pragmatics focus on referring phenomena, for instance referring to the objects that are accessible in the dialogue context, and speech acts, or dialogue acts, i.e., communicative actions carried out by utterances. We present our modelling and formalizing works dealing with these two aspects and with their application to dialogue systems, particularly when a visual scene is implied or when co-verbal gestures are processed together with speech (multimodal dialogue). Human-machine dialogue also aims at facilitating the design and development of effective systems, with methodologies and means like software architectures. We present our theoretical and practical works in this way, illustrated by our participation to several European projects. We then propose some future works that focus on the integration of linguistic and pragmatic phenomena like salience and ambiguity to human-machine dialogue.
