During spontaneous conversation, interlocutors have three possible actions: speak, be silent or produce feedback. In order to better understand the mechanisms that render spontaneous interactions successful, this PhD research focuses on conversational feedback. It is the reactions/responses produced by an interlocutor in a listening position. Feedback is a phenomenon of deep importance for the quality of the interaction. It allows interlocutors to share relevant information about understanding, establishment/upgrading of the common ground, engagement and shared representations. The objective of the PhD is to propose a multimodal model of conversational feedback. The methodological approach is interdisciplinary, combining a corpus analysis, based on machine learning enhanced by a linguistic interpretation. The resulting model will be evaluated through its integration in an Embodied Conversational Agent (ECA) with perspective studies.