This paper presents an experimental design and setup that explores the interaction between two children and their tutor during a question–answer session of a reading comprehension task. The multimodal aspects of the interactions are analysed in terms of preferred signals and strategies that speakers employ to carry out successful multi-party conversations. This analysis will form the basis for the development of behavioral models accounting for the specific context. We envisage the integration of such models into intelligent, context-aware systems, i.e. an embodied dialogue system that has the role of a tutor and is able to carry out a discussion in a multiparty setting by exploring the multimodal signals of the children. This system will have the ability to discuss a text and address questions to the children, encouraging collaboration and equal participation in the discussion and assessing the answers that the children give. The paper focuses on the design of the appropriate setup, the data collection and the analysis of the multimodal signals that are important for the realization of such a system.