Résumé :
When interacting with students, teachers usually combine several communication modalities (speech, hand gestures, gaze, posture, facial expression, graphics on a blackboard, slides…) and have to adapt their communication to the lecture settings (computer knowledge of the students, duration of the lecture…). Although educational resources and intelligent tutoring systems are developing, they are seldom being used as language resources per se, nor based on real-world pedagogical recording. Even in the field of pedagogical agents where a graphical persona is used as a complementary means of communication, the multimodal and adaptative behaviour of the graphical agent is often based on general knowledge about communication studies rather than on the annotation of pedagogical behaviour observed in video corpora. In this paper, we describe how an educational video corpus is being collected and how it is planned to be used for improving the existing on-line tutorial with multimodal and adaptative hypermedia features.