Since the last decade, online synchronous audio-graphic learning environments, which enable participants to communicate and collaborate in order to achieve a common task, have become increasingly popular, particularly in the language learning domain. These online learning environments generate complex interactions between participants. Their analysis renews research on traces and their processing. The work presented in this paper concerns the methodology used to transcribe and analyse synchronous multimodal interactions coming from a particular online training. We describe the protocol to collect data, the original coding scheme to transcribe the video data into a set of participant's actions, part of them being speech acts which occurred into the different modalities of the system (audio, textchat, text editing, and websites), the organization, and the representation of such data. Two examples of analysis, at a macro scale and at a micro scale, give a glimpse of our results, from which the benefits and limits of such approaches and tools are debated.