YouTube-ASL: A Large-Scale, Open-Domain American Sign Language-English Parallel Corpus

06/27/2023

∙

by   David Uthus, et al.

∙

0

∙

share

Machine learning for sign languages is bottlenecked by data. In this paper, we present YouTube-ASL, a large-scale, open-domain corpus of American Sign Language (ASL) videos and accompanying English captions drawn from YouTube. With  1000 hours of videos and >2500 unique signers, YouTube-ASL is  3x as large and has  10x as many unique signers as the largest prior ASL dataset. We train baseline models for ASL to English translation on YouTube-ASL and evaluate them on How2Sign, where we achieve a new finetuned state of the art of 12.39 BLEU and, for the first time, report zero-shot results.