Evaluating a dynamic timewarping based scoring algorithm for facial expressions in asl animations

Abstract

Advancing the automatic synthesis of linguistically accurate and natural-looking American Sign Language (ASL) animations from an easy-To-update script would increase information accessibility for many people who are deaf by facilitating more ASL content to websites and media. We are investigating the production of ASL grammatical facial expressions and head movements coordinated with the manual signs that are crucial for the interpretation of signed sentences. It would be useful for researchers to have an automatic scoring algorithm that could be used to rate the similarity of two animation sequences of ASL facial movements (or an animation sequence and a motioncapture recording of a human signer). We present a novel, sign-language specific similarity scoring algorithm, based on Dynamic Time Warping (DTW), for facial expression performances and the results of a user-study in which the predictions of this algorithm were compared to the judgments of ASL signers. We found that our algorithm had significant correlations with participants comprehension scores for the animations and the degree to which they reported noticing specific facial expressions.