The cartoon task - Exploring auditory-visual prosody in dialogs

This paper introduces and analyses a collaborative task for eliciting auditory-visual dialogs based on the viewing of two versions of the same cartoon film. The original film was edited and cut in such a way that the story must be reconstructed by joining information from two incomplete versions which however share between them all the scenes in a consecutive fashion. Our intention is to elicit a relatively balanced dialog between the two participants throughout the conversation as they are piecing together the story from the beginning to the end. The current paper describes the production of the auditory-visual corpus using audio, video and motion capturing of 22 pairs of Australian English speaking participants, and presents first results regarding turndistribution and raw prosodic features. Our analysis shows that the task is indeed relatively balanced between talkers though this does not apply equally to all pairs. Analysis of raw prosodic features does not suggest convergence throughout the conversation, but replicates, for instance earlier findings of similarity between partners as compared to others.