Research map / Large language models / Language, safety and interpretability
SIGNER: Temporally Grounded Sign Language Generation via Time-Resolved Conditioning
Taeryung Lee, Hyeongjin Nam, Gyeongsik Moon, Kyoung Mu Lee
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 |
|---|---|---|---|
| replace cross-attention with AdaLN + 1D conv (LTF) for condition fusion (ablation) | SIGNER · CSL-Daily | WER: 84.90 -> 55.05 | Improved |
| replace cross-attention with LTF (ablation) | SIGNER · CSL-Daily | BLEU-4: 5.81 -> 15.60 | Improved |
| replace fully connected layer with temporal 1D convolution in LTF (ablation) | SIGNER · CSL-Daily | WER: 72.52 -> 55.05 | Improved |
| replace fully connected layer with temporal 1D convolution in LTF (ablation) | SIGNER · CSL-Daily | BLEU-4: 7.85 -> 15.60 | Improved |
| add inter-part attention (B->H, H->B, B->F) (ablation) | SIGNER · CSL-Daily | WER: 61.31 -> 55.05 | Improved |
| add inter-part attention (all flows) (ablation) | SIGNER · CSL-Daily | BLEU-4: 13.27 -> 15.60 | Improved |
| add only H->B attention (ablation) | SIGNER · CSL-Daily | BLEU-4: 13.27 -> 14.52 | Improved |
| add body-to-face attention on top of B->H (ablation) | SIGNER · CSL-Daily | WER: 57.01 -> 55.31 | Improved |
| add H->B and B->F without B->H (ablation) | SIGNER · CSL-Daily | BLEU-4: 13.27 -> 13.41 | Hurt |
| use fine-tuned mT5 text-to-gloss instead of off-the-shelf gloss model (ablation) | SIGNER · CSL-Daily | WER: 55.05 vs 58.70 | No clear effect |
| perturb estimated gloss duration by +4 (ablation) | SIGNER · CSL-Daily | WER: 55.05 -> 56.60 | No clear effect |
| perturb estimated gloss duration randomly in U(-4, 4) (ablation) | SIGNER · CSL-Daily | WER: 55.05 -> 60.74 | Hurt |
Closest papers by meaning
- Bridging Sign and Spoken Languages: Pseudo Gloss Generation for Sign Language Translation (NeurIPS 2025 · similarity 0.90)
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.88)
- Improving Gloss-free Sign Language Translation by Reducing Representation Density (NeurIPS 2024 · similarity 0.87)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.87)
- Conditional Variational Autoencoder for Sign Language Translation with Cross-Modal Alignment (AAAI 2024 · similarity 0.87)
- Advanced Sign Language Video Generation with Compressed and Quantized Multi-Condition Tokenization (NeurIPS 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.
