Research map / Large language models / Language, safety and interpretability
BackTranslation2.0 - A Linguistically Motivated Metric to Assess Sign Language Production
Oliver Cory, Maksym Ivashechkin, Oline Ranum, Jian He Low, Edward Fish, Anton Pelykh, Karahan Sahin, Ozge Mercanoglu Sincan, Richard Bowden
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 Phase 2 language model reasoning (vs deterministic fallback) (ablation) | BT2 · Synthetics (understandability) | Spearman rho: 0.60 -> 0.09 | Improved |
| add LLM (vs word heuristic) (ablation) | BT2 · Synthetics (understandability) | Spearman rho: .60 -> .37 | Improved |
| add spot-gloss tool (ablation) | BT2 · Synthetics (understandability) | Spearman rho: .60 -> .37 | Improved |
| add manual features tool (ablation) | BT2 · Synthetics (understandability) | Spearman rho: .60 -> .37 | Improved |
| add non-manual tool (ablation) | BT2 · Synthetics (understandability) | Spearman rho: .60 -> .54 | Improved |
| add topographic directional tool (ablation) | BT2 · Synthetics (understandability) | Spearman rho: .60 -> .43 | Improved |
| add non-manual tool (ablation) | BT2 · Human SRT (understandability) | Spearman rho: -.10 -> -.30 | Hurt |
| add any component (ablation) | BT2 · Known Corruptions (understandability) | Spearman rho: 0.50 -> 1.00 | Improved |
| use full aggregation instead of Grammar sub-score alone | BT2 · Known Corruptions (understandability) | Spearman rho: -1.00 -> 1.00 | Improved |
| use BT2 instead of SiBLEU | BT2 · Human SRT (understandability) | Spearman rho: 0.90 -> -0.30 | Hurt |
| use BT2 Phonology sub-score on real proficiency variation | BT2 · Human SRT (understandability) | Spearman rho: -0.30 -> 0.90 | Improved |
| use BT2 Fidelity sub-score | BT2 · Synthetics (quality) | Spearman rho: 0.66 -> 0.94 | Improved |
Closest papers by meaning
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.89)
- Lost in Translation, Found in Context: Sign Language Translation with Contextual Cues (CVPR 2025 · similarity 0.89)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.87)
- KD-MSLRT: Lightweight Sign Language Recognition Model Based on Mediapipe and 3D to 1D Knowledge Distillation (AAAI 2025 · similarity 0.87)
- LLMs are Good Sign Language Translators (CVPR 2024 · similarity 0.86)
- YouTube-SL-25: A Large-Scale, Open-Domain Multilingual Sign Language Parallel Corpus (ICLR 2025 · similarity 0.86)
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.
