Effect of Displaying Human Videos During an Evaluation Study of American Sign Language Animation

Abstract

Many researchers internationally are studying how to synthesize computer animations of sign language; such animations have accessibility benefits for people who are deaf and have lower literacy in written languages. The field has not yet formed a consensus as to how to best conduct evaluations of the quality of sign language animations, and this article explores an important methodological issue for researchers conducting experimental studies with participants who are deaf. Traditionally, when evaluating an animation, some lower and upper baselines are shown for comparison during the study. For the upper baseline, some researchers use carefully produced animations, and others use videos of human signers. Specifically, this article investigates, in studies where signers view animations of sign language and are asked subjective and comprehension questions, whether participants differ in their subjective and comprehension responses when actual videos of human signers are shown during the study. Through three sets of experiments, we characterize how the Likert-scale subjective judgments of participants about sign language animations are negatively affected when they are also shown videos of human signers for comparison -- especially when displayed side-by-side. We also identify a small positive effect on the comprehension of sign language animations when studies also contain videos of human signers. Our results enable direct comparison of previously published evaluations of sign language animations that used different types of upper baselines -- video or animation. Our results also provide methodological guidance for researchers who are designing evaluation studies of sign language animation or designing experimental stimuli or questions for participants who are deaf.