Comparison of Finite-Repertoire and Data-Driven Facial Expressions for Sign Language Avatars

To support our research on ASL animation synthesis, we have adopted and enhanced a new virtual human animation platform that provides us with greater fine-grained control of facial movements than our previous platform. To determine whether this new platform is sufficiently expressive to generate understandable ASL animations, we analyzed responses collected from deaf participants who evaluated four types of animations: generated by our old or new animation platform, and with or without facial expressions performed by the character. For animations without facial expressions, our old and new platforms had equivalent comprehension scores; for those with facial expressions, our new platform had higher scores. In addition, this paper demonstrates a methodology by which sign language animation researchers can document transitions in their animation platforms or avatar appearance. Performing such an evaluation enables future readers to compare published results over time, both before and after such a transition in animation technology.