ASL Sea Battle: Gamifying Sign Language Data Collection

  • ,
  • Naomi Caselli ,
  • John W. Gallagher ,
  • Miriam Goldberg ,
  • Courtney Oka ,
  • Bill Thies

CHI 2021 |

Organized by ACM

The development of accurate machine learning models for sign languages like American Sign Language (ASL) has the potential to break down communication barriers for deaf signers. However, to date, no such models have been robust enough for real-world use. The primary barrier to enabling real-world applications is the lack of appropriate training data. Existing training sets suffer from several shortcomings: small size, limited signer diversity, lack of real-world settings, and missing or inaccurate labels. In this work, we present ASL Sea Battle, a sign language game designed to collect datasets that overcome these barriers, while also providing fun and education to users. We conduct a user study to explore the data quality that the game collects, and the user experience of playing the game. Our results suggest that ASL Sea Battle can reliably collect and label real-world sign language videos, and provides fun and education at the expense of data throughput.