This study proposes a corpus linguistic method for multimodal text analysis -in this case of video streams with synchronized sounds-, that begins with the analytic unit of social situation, to go further with that of activity type, task or episode, and the participants' behavior.