We use rhetorical annotations to specify a generation process that can assemble meaningful video sequences with a communicative goal and an argumentative progression. Our annotation schema encodes the verbal information contained in the audio channel, identifying the claims the interviewees make and the argumentation structures they use to make those claims. Based on this schema, we construct a semantic graph which is traversed by rhetoric-based strategies selecting video segments. The selected video segments are edited to form a meaningful video sequence.