In this paper, we present our work on context aware semantic adaptation of multimedia structured documents. We propose semantic annotation of multimedia scenes, expressing semantic information on each media object, as well as on the dependencies between all the media objects of the scene. We use these annotations in order to perform a semantic adaptation on multimedia presentations. We use our proposed description tools under the framework of MPEG-21, and we show that, in order to preserve the consistency and meaningfulness of the adapted multimedia scene, the adaptation process needs the semantic information of the presentation.