This paper argues that the integration of multimodal resources (i.e. talk, gesture, gaze, facial expressions, bodily postures) within grammatical description would change both our conception of what language is and our way of modelizing it. Starting with a discussion about the very notion of resource, the paper shows, on the basis of the study of a collection of cases, that the choice and use of a particular syntactic form is carefully adjusted to the ecology of the ongoing action. The paper focuses on two formats available for the introduction of new referent; it shows that the progressivity of the turn is strongly related to the monitoring by the speaker of the bodily posture of the co-participants-. By showing the systematicity of these emerging multimodal Gestalts, the paper argues that a linguistic description should integrate these complex multimodal dynamics within the very definition and modelization of referential processes.