Development and Evaluation of System for Automatically Generating Sign-Language CG Animation Using Meteorological Information

Abstract

People who are born with hearing difficulties often use sign language as their mother tongue. The vocabulary and grammar of sign language differ from those of aural languages, so it is important to convey information in sign language. To expand sign-language services, we developed a system that accurately generates Japanese-Sign-Language computer-graphics (CG) animation from weather data. The system reads weather-forecast data coded in XML format distributed by the Japan Meteorological Agency and automatically generates CG animation clips to present them in sign language. We conducted two experiments to evaluate the system’s performance in conveying weather information in sign-language CG animation to deaf participants, one was a comprehension evaluation to answer multiple-choice questions on the content and the other was a subjective evaluation on how easy the sign language was to understand and how natural it was on a 5-point scale (1: not understandable and unnatural and 5: understandable and natural). The overall percentage of correct answers was 96.5%. In the subjective evaluation, the average understandability was 4.43, and average naturalness was 4.13, suggesting that the participants highly appreciated the quality of sign-language CG animation. We also published a website in 2017 on which anyone can evaluate such animation regarding the latest weather forecast.