We focus on an XML-based video annotation system for language learning using a videoconference system. The environment supports the native speaker as teacher and non-native learner. We extend the videoconference system so the system can record the learning scene as a video file. The learning scene recorded in the videoconference system can be used for instruction. For instance, the teacher can explain by inserting feedback as annotation in the scene. If a teacher edits the video for explanation, it is very useful as a teaching material. However, an ordinal video editing system creates quite a new video file so it cannot change the contents for playback. In addition, the learner can hardly grasp the teacher's task. Therefore, we propose a video correction and VCML (Video-based Correction Markup Language) based on XML to represent corrected information. We call the prototype Viclle (Video-based Communicative Language Learning System). With the system, both teacher and learner can communicate via e-mail without sending video files. In addition, video correction resembles retrieving information from a database with keywords. Therefore, Viclle is regarded as an error database.