Research map / Large language models / Language, safety and interpretability
SignBind-LLM: Multi-Stage Modality Fusion for Sign Language Translation
Marshall Thomas, Edward Fish, 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 lipreading stream (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 14.6 -> 23.1 | Improved |
| add fingerspelling expert (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 19.9 -> 23.1 | Improved |
| add fusion module (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 9.6 -> 23.1 | Improved |
| add LLM decoder (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 8.8 -> 23.1 | Improved |
| add dynamic router (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 15.3 -> 23.1 | Improved |
| gated fusion instead of concat + MLP (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 12.4 -> 22.3 | Improved |
| add cross-attention on top of gated fusion (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 22.3 -> 23.1 | Improved |
| learned temporal alignment instead of fixed offset (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 18.2 -> 23.1 | Improved |
| shift lips -5 frames relative to hands (fixed) (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 18.2 -> 19.1 | Improved |
| shift lips +10 frames (fixed) (ablation) | SignBind-LLM · How2Sign test | BLEU-4: 18.2 -> 8.8 | Hurt |
| scale decoder from Flan-T5-Base to XL | SignBind-LLM · How2Sign test | BLEU-4: 16.5 -> 23.1 | Improved |
| use multi-stream architecture at matched 250M decoder vs SSVP-SLT | SignBind-LLM base · How2Sign test | BLEU-4: 15.5 -> 16.5 | Improved |
Closest papers by meaning
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.92)
- Sign2GPT: Leveraging Large Language Models for Gloss-Free Sign Language Translation (ICLR 2024 · similarity 0.91)
- Improving Gloss-free Sign Language Translation by Reducing Representation Density (NeurIPS 2024 · similarity 0.91)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.90)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.90)
- LLMs are Good Sign Language Translators (CVPR 2024 · similarity 0.90)
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.
