Continuous Annotations for Dialogue Status and Their Change Points

This paper presents an attempt to continuously annotate the emotion and status of multimodal corpora for understanding psychotherapeutic interviews. The collected continuous annotations are then used as the signal data to find change points in the dialogues. Our target dialogues are carried between clients with some psychological problems and their therapists. We measured two values, namely the degree of the dialogue progress and the degree of clients being listened to. The first value reflects the goal-oriented nature of the target dialogues. The second value corresponds to the idea of active listening that is considered as an important aspect in psychotherapy. We have modified an existing continuous emotion annotation toolkit that has been created for tracking generic emotion of dialogues. By applying a change point detection algorithm on the obtained annotations, we evaluated the validity and utility of the collected annotation based on our method.