Research map / Large language models / Language, safety and interpretability
YouTube-ASL: A Large-Scale, Open-Domain American Sign Language-English Parallel Corpus
David Uthus, Garrett Tanzer, Manfred Georg
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add YouTube-ASL training then finetune on How2Sign (vs H2S only) (ablation) | pretrained T5.1.1-Base · How2Sign | BLEU: 1.22 -> 12.39 | Improved |
| finetune on How2Sign after YT-ASL instead of mixing datasets (ablation) | pretrained T5.1.1-Base · How2Sign | BLEU: 11.89 -> 12.39 | Improved |
| finetune on How2Sign after YT-ASL pretraining (vs zero-shot YT-ASL only) (ablation) | pretrained T5.1.1-Base · How2Sign | BLEU: 3.95 -> 12.39 | Improved |
| initialize from pretrained T5 instead of training from scratch (ablation) | YT-ASL -> H2S · How2Sign | BLEU: 6.26 -> 12.39 | Improved |
| initialize from pretrained T5 instead of scratch (ablation) | YT-ASL zero-shot · How2Sign | BLEU: 1.41 -> 3.95 | Improved |
| initialize from pretrained T5 instead of scratch (ablation) | H2S only · How2Sign | BLEU: 0.86 -> 1.22 | Improved |
| YT-ASL pretrained T5 pipeline vs prior state of the art | ASL-to-English translation · How2Sign | BLEU: 8.03 -> 12.39 | Improved |
| train on YouTube-ASL only (zero-shot) vs How2Sign only, no pretraining (ablation) | T5 from scratch · How2Sign | BLEU: 0.86 -> 1.41 | Improved |
Closest papers by meaning
- YouTube-SL-25: A Large-Scale, Open-Domain Multilingual Sign Language Parallel Corpus (ICLR 2025 · similarity 0.95)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.90)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.88)
- LLMs are Good Sign Language Translators (CVPR 2024 · similarity 0.87)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.87)
- Lost in Translation, Found in Context: Sign Language Translation with Contextual Cues (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
