This research work introduce virtual embodied tutors in Virtual Environments for learning devoted to learning of procedures for industrial systems. We present a communicative behavior which, integrated in pedagogical scenario, permits on the one hand to realize the pedagogical communicative actions at a semantic level (e.g., the tutor explains the goal of an action) and on the other hand to realize such actions through human-like communicative channels (i.e., the virtual tutor's voice, facial expressions and gestures). The communicative behavior relies on a taxonomy of questions in order to interpret the learner's communicative actions and to generate the tutor's own questions.