Research map / Large language models / Language, safety and interpretability
Addressing Resource Scarcity across Sign Languages with Multilingual Pretraining and Unified-Vocabulary Datasets
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| pretrain on diverse multi-language SignCorpus instead of ISL data only (UFT vs ISL) (ablation) | SL-GCN finetuned individually · MultiSign-ISLR average | accuracy: 62.5 -> 66.4 | Helped |
| UFT vs ISL on low-accuracy datasets (ablation) | SL-GCN · ASLLVD | accuracy: 27.7 -> 32.5 | Helped |
| UFT vs ISL on MSASL (ablation) | SL-GCN · MSASL (1000) | accuracy: 18.2 -> 34.0 | Helped |
| UFT vs ISL on LSA64 (ablation) | SL-GCN · LSA64 | accuracy: 96.3 -> 97.5 | Helped |
| UFT vs ISL on BosphorusSign22k (ablation) | SL-GCN · BosphorusSign22k | accuracy: 88.3 -> 87.2 | Hurt |
| joint multilingual finetuning with aligned vocabulary (MFT) instead of individual finetuning (UFT) (ablation) | SL-GCN pretrained on SignCorpus · MultiSign-ISLR average | accuracy: 66.4 -> 77.6 | Helped |
| MFT instead of UFT on AUTSL (ablation) | SL-GCN · AUTSL | accuracy: 92.1 -> 91.4 | Hurt |
| MFT instead of UFT on LSA64 (ablation) | SL-GCN · LSA64 | accuracy: 97.5 -> 95.9 | Hurt |
| MFT instead of UFT on ASLLVD (ablation) | SL-GCN · ASLLVD | accuracy: 32.5 -> 50.1 | Helped |
| MFT instead of UFT on PHEONIX-W-S03 (ablation) | SL-GCN · PHEONIX-W-S03 | accuracy: 48.7 -> 48.5 | No clear effect |
| multilingual pretraining plus joint finetuning (MFT) vs ISL baseline | SL-GCN · MultiSign-ISLR | accuracy: 13.4% improvement | Helped |
| MFT vs prior SOTA | SL-GCN · MultiSign-ISLR average (without Pheonix) | accuracy: 71.1 -> 77.6 | Helped |
Closest papers by meaning
- Scaling Sign Language Translation (NeurIPS 2024 · similarity 0.91)
- Lost in Translation, Found in Embeddings: Sign Language Translation and Alignment (NeurIPS 2026 · similarity 0.91)
- YouTube-SL-25: A Large-Scale, Open-Domain Multilingual Sign Language Parallel Corpus (ICLR 2025 · similarity 0.91)
- Uni-Sign: Toward Unified Sign Language Understanding at Scale (ICLR 2025 · similarity 0.90)
- Sign2GPT: Leveraging Large Language Models for Gloss-Free Sign Language Translation (ICLR 2024 · similarity 0.90)
- Lost in Translation, Found in Context: Sign Language Translation with Contextual Cues (CVPR 2025 · 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.