The design of multimodal dialogue systems requires a particular attention on the way of managing dynamic heterogeneous information. We present a theoretical model of a multimodal dialogue history, that includes a global history and local histories linked to the various modalities. We describe the nature of these components and the relations they entertain. To save the information and to exploit and confront them during the interpretation of an utterance from the user, we need an unified representation format. We developed the MMIL (MultiModal Interface Language) model that we present here with two main aspects: the representation of a simple utterance and the representation of a dialogue structure. Then, we draw some conclusions concerning the exploitation of this framework in the OZONE system, with an interest on the management of attentional scores inside the dialogue history.