Fig. 10. For each corpus sentence, the correlation is shown between the average evaluation of the corresponding eliciting video and the average evaluation of the sentence pronounced by the speakers after watching the video. High values correspond to agreement in the evaluations.