Sign Language Translation on PHOENIX-2014T (test)
29.91BLEU-4SignX
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SignXCategory=Sign2Text2025.04 | 29.91 | 55.15 | 55.08 | 42.92 | 35.28 | — | — | — | — | — | — | — | |
| CorrNet+Category=Sign2Text2025.04 | 29.42 | 53.76 | 55.32 | 42.74 | 34.86 | — | — | — | — | — | — | — | |
| CV-SLTSign Representation=Gloss-Based2025.07 | 29.27 | — | 54.88 | 42.68 | 34.79 | — | — | — | — | — | 54.33 | — | |
| MSKA-SLTInput Modality=Keypoint2024.05 | 29.03 | 53.54 | 54.79 | 42.42 | 34.49 | — | — | — | — | — | — | — | |
| MSKA-SLTCategory=Gloss-Based, Extra Technique=3D, Pretrained; LM; Embedding, Params (M)=4042024.11 | 29.03 | 54.04 | — | — | — | — | — | — | — | — | — | — | |
| TwoStream-SLTLLM=true2023.08 | 28.95 | 53.48 | 54.9 | 42.43 | 34.46 | — | — | — | — | — | — | — | |
| TS-SLTSetting=Gloss-based2024.04 | 28.95 | 53.48 | 54.9 | 42.43 | 34.46 | — | — | — | — | — | — | — | |
| TwoStream-SLTInput Modality=RGB2024.05 | 28.95 | 53.48 | 54.9 | 42.43 | 34.46 | — | — | — | — | — | — | — | |
| TwoStream-SLTCategory=Sign2Text2025.04 | 28.95 | 53.48 | 54.9 | 42.43 | 34.46 | — | — | — | — | — | — | — | |
| TS-SLTSign Representation=Gloss-Based2025.07 | 28.95 | — | 54.9 | 42.43 | 34.46 | — | — | — | — | — | 53.48 | — | |
| SLTUNETSetting=Gloss-based2024.04 | 28.47 | 52.11 | 52.92 | 41.76 | 33.99 | — | — | — | — | — | — | — | |
| SLTUNETCategory=Sign2Text2025.04 | 28.47 | 52.11 | 52.92 | 41.76 | 33.99 | — | — | — | — | — | — | — | |
| SLTUNETSign Representation=Gloss-Based2025.07 | 28.47 | — | 52.92 | 41.76 | 33.99 | — | — | — | — | — | 52.11 | — | |
| TwoStream-SLTInput Modality=Keypoint2024.05 | 28.42 | 53.19 | 54.22 | 41.72 | 33.82 | — | — | — | — | — | — | — | |
| Ourspipeline_strategy=Sign2Text (end-to-end)2022.03 | 28.39 | 52.65 | 53.97 | 41.75 | 33.84 | — | — | — | — | — | — | — | |
| MMTLBLLM=true2023.08 | 28.39 | 52.65 | 53.97 | 41.75 | 33.84 | — | — | — | — | — | — | — | |
| MMTLBSetting=Gloss-based2024.04 | 28.39 | 52.65 | 53.97 | 41.75 | 33.84 | — | — | — | — | — | — | — | |
| MMTLBInput Modality=RGB2024.05 | 28.39 | 52.65 | 53.97 | 41.75 | 33.84 | — | — | — | — | — | — | — | |
| VL-MapperCategory=Gloss-Based, Extra Technique=Extra 3 Datasets; Pretrained, Params (M)=6102024.11 | 28.39 | 53.1 | — | — | — | — | — | — | — | — | — | — | |
| MMTLBCategory=Sign2Text2025.04 | 28.39 | 52.65 | 53.97 | 41.75 | 33.84 | — | — | — | — | — | — | — | |
| MMTLBSign Representation=Gloss-Based2025.07 | 28.39 | — | 53.97 | 41.75 | 33.84 | — | — | — | — | — | 52.65 | — | |
| IP-SLTLLM=false2023.08 | 27.97 | 53.72 | 54.25 | 41.51 | 33.45 | — | — | — | — | — | — | — | |
| IP-SLTInput Modality=RGB2024.05 | 27.97 | 53.72 | 54.25 | 41.51 | 33.45 | — | — | — | — | — | — | — | |
| SignBERT+Input Modality=RGB, Refined Pose (R)=true2023.05 | 25.7 | 50.63 | 52.01 | 39.19 | 31.06 | — | — | — | — | — | — | — | |
| SignBERT+(+ R)Input Modality=RGB2024.05 | 25.7 | 50.63 | 52.01 | 39.19 | 31.06 | — | — | — | — | — | — | — | |
| Stochastic Transformer Networks with Linear Competing UnitsL=8, Bit-width=32 bit, Ensemble-Decoding=true2021.09 | 25.59 | — | — | — | — | — | — | — | — | — | — | — | |
| ConSLTTranslation Pipeline Setting=Cascading: Sign2Gloss2Text2022.04 | 25.48 | 47.65 | — | — | — | — | — | — | — | — | — | — | |
| ConSLTInput Modality=RGB2024.05 | 25.48 | 47.65 | 51.57 | 38.81 | 30.91 | — | — | — | — | — | — | — | |
| S2G2T-STMC ensembletype=sign-to-gloss-to-text, ensemble=true2021.09 | 25.4 | — | — | — | — | — | — | — | — | — | — | — | |
| STMC-TCategory=Gloss-Based, Extra Technique=Pretrained; Transfer Learning + Ensembling, 8 Models, Params (M)=192*2024.11 | 25.4 | 48.24 | — | — | — | — | — | — | — | — | — | — | |
| XmDAInput Modality=RGB2024.05 | 25.36 | 49.87 | — | — | — | — | — | — | — | — | — | — | |
| Stochastic Transformer Networks with Linear Competing UnitsL=8, Bit-width=Reduced, Ensemble-Decoding=true2021.09 | 25.33 | — | — | — | — | — | — | — | — | — | — | — | |
| STN-LCUCategory=Gloss-Based, Extra Technique=Pretrained Embedding + Ensembling, 8 Models, Params (M)=2802024.11 | 25.33 | — | — | — | — | — | — | — | — | — | — | — | |
| Stochastic Transformer Networks with Linear Competing UnitsL=4, Bit-width=32 bit, Ensemble-Decoding=true2021.09 | 24.84 | — | — | — | — | — | — | — | — | — | — | — | |
| Ourspipeline_strategy=Sign2Gloss2Text (two-stage)2022.03 | 24.6 | 49.59 | 49.94 | 37.28 | 29.67 | — | — | — | — | — | — | — | |
| G2Ttype=gloss-to-text2021.09 | 24.54 | — | — | — | — | — | — | — | — | — | — | — | |
| Stochastic Transformer Networks with Linear Competing UnitsL=4, Bit-width=Reduced, Ensemble-Decoding=true2021.09 | 24.52 | — | — | — | — | — | — | — | — | — | — | — | |
| BN-TIN-Transf. + SignBTParadigm=S2T, SignBT=true2021.05 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| BN-TIN-Transf + BT*pipeline_strategy=Sign2Text (end-to-end), semi-supervised=true2022.03 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| SignBTInput Modality=RGB2023.05 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| BN-TIN-Transf.+SignBTSetting=Gloss-based2024.04 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| SignBTInput Modality=RGB2024.05 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| BN-TIN-Tran.+SignBTCategory=Gloss-Based, Extra Technique=Dataset; Pretrain; DataAug2024.11 | 24.32 | 49.54 | — | — | — | — | — | — | — | — | — | — | |
| SignBTCategory=Sign2Text2025.04 | 24.32 | 49.54 | 50.8 | 37.75 | 29.72 | — | — | — | — | — | — | — | |
| SignBTSign Representation=Gloss-Based2025.07 | 24.32 | — | 50.8 | 37.75 | 29.72 | — | — | — | — | — | 49.54 | — | |
| LLM-Guided Video StitchingSetting=Augmentation (Synthetic), Epochs=2002026.06 | 24.3 | — | 50.04 | 37.3 | 29.41 | — | — | — | — | — | 46.09 | — | |
| SAGESign Representation=Gloss-Free2025.07 | 24.1 | — | 49.69 | 37.01 | 29.21 | — | — | — | — | — | 48.86 | — | |
| PETLLM=false2023.08 | 24.02 | 49.97 | 49.54 | 37.19 | 29.3 | — | — | — | — | — | — | — | |
| PETSetting=Gloss-based2024.04 | 24.02 | 49.97 | 50.29 | 37.19 | 29.3 | — | — | — | — | — | — | — | |
| S2G2T-STMCtype=sign-to-gloss-to-text, ensemble=false2021.09 | 24 | — | — | — | — | — | — | — | — | — | — | — | |
| STMC-Transfpipeline_strategy=Sign2Gloss2Text (two-stage)2022.03 | 24 | 46.77 | 48.73 | 36.53 | 29.03 | — | — | — | — | — | — | — | |
| STMC-TransformerTranslation Pipeline Setting=Cascading: Sign2Gloss2Text2022.04 | 24 | 46.77 | — | — | — | — | — | — | — | — | — | — | |
| STMC-TCategory=Gloss-Based, Extra Technique=Pretrained; Transfer Learning, Params (M)=65*2024.11 | 24 | 46.77 | — | — | — | — | — | — | — | — | — | — | |
| STMC-TParadigm=S2T, multi-cue setting=true2021.05 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STMC-Tpipeline_strategy=Sign2Text (end-to-end)2022.03 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STMCInput Modality=RGB2023.05 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STMCLLM=false2023.08 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STN-SLTSetting=Gloss-based2024.04 | 23.65 | — | 48.61 | 35.97 | 28.37 | — | — | — | — | — | — | — | |
| STMC-TSetting=Gloss-based2024.04 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STMCInput Modality=RGB2024.05 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| STN-LCUCategory=Gloss-Based, Extra Technique=Pretrained Embedding, Params (M)=352024.11 | 23.65 | — | — | — | — | — | — | — | — | — | — | — | |
| STMC-TCategory=Sign2Text2025.04 | 23.65 | 46.65 | 46.98 | 36.09 | 28.7 | — | — | — | — | — | — | — | |
| BN-TIN-Transf.2 + BTParadigm=S2G2T, back-translation=true2021.05 | 23.51 | 49.35 | 48.55 | 36.13 | 28.47 | — | — | — | — | — | — | — | |
| BN-TIN-Transf + BT*pipeline_strategy=Sign2Gloss2Text (two-stage), semi-supervised=true2022.03 | 23.51 | 49.35 | 48.55 | 36.13 | 28.47 | — | — | — | — | — | — | — | |
| Joint-SLT / Self-Mutual KD + SignCLFeature Type=Gloss-based, SignCL integration=Both2024.05 | 23.46 | — | — | — | — | — | — | 62.18 | — | 24.58 | — | — | |
| Signformer - FullCategory=Ours (Gloss-Free), Extra Technique=Scratch, Params (M)=3.882024.11 | 23.43 | 48.53 | — | — | — | — | — | — | — | — | — | — | |
| SignLLMSetting=Gloss-free2024.04 | 23.4 | 44.49 | 45.21 | 34.78 | 28.05 | — | — | — | — | — | — | — | |
| SignLLMCategory=Gloss-Free, Extra Technique=LM + Pretraining; LLM, Params (M)=70122024.11 | 23.4 | 46.88 | — | — | — | — | — | — | — | — | — | — | |
| SignLLMSign Representation=Gloss-Free2025.07 | 23.4 | — | 45.21 | 34.78 | 28.05 | — | — | — | — | — | 44.49 | — | |
| Joint-SLT / Self-Mutual KD + SignCLFeature Type=Gloss-based, SignCL integration=Feature Extraction2024.05 | 23.23 | — | — | — | — | — | — | 62.18 | — | 24.76 | — | — | |
| FLa-LLMSign Representation=Gloss-Free2025.07 | 23.09 | — | 46.29 | 35.33 | 28.03 | — | — | — | — | — | 45.27 | — | |
| Joint-SLT / Self-Mutual KD + SignCLFeature Type=Gloss-based, SignCL integration=Finetuning2024.05 | 22.92 | — | — | — | — | — | — | 66.23 | — | 25.12 | — | — | |
| Joint-SLT / Self-Mutual KDFeature Type=Gloss-based, SignCL integration=None2024.05 | 22.79 | — | — | — | — | — | — | 66.23 | — | 25.38 | — | — | |
| GFSLT-VLP + SignCLSignCL integration stage=Two State2024.05 | 22.74 | — | 49.76 | 36.85 | 29.97 | — | — | 62.32 | 49.04 | — | — | — | |
| GFSLT-VLP + SignCLSignCL integration stage=Pretraining2024.05 | 22.69 | — | 49.99 | 36.73 | 29.76 | — | — | 62.68 | 49.25 | — | — | — | |
| BN-TIN-Transf.2Paradigm=S2G2T, baseline=true2021.05 | 22.54 | 47.98 | 47.74 | 35.27 | 27.59 | — | — | — | — | — | — | — | |
| BN-TIN-Transfpipeline_strategy=Sign2Gloss2Text (two-stage)2022.03 | 22.54 | 47.98 | 47.74 | 35.27 | 27.59 | — | — | — | — | — | — | — | |
| Sign2GPT(w/PGP)2024.05 | 22.52 | — | 49.54 | 35.96 | 28.83 | — | — | — | 48.9 | — | — | — | |
| Sign2GPTCategory=Gloss-Free, Extra Technique=Pseudo Gloss Pretraining; LLM, Params (M)=17722024.11 | 22.52 | 48.9 | — | — | — | — | — | — | — | — | — | — | |
| Sign2GPTSign Representation=Gloss-Free2025.07 | 22.52 | — | 49.54 | 35.96 | 28.83 | — | — | — | — | — | 48.9 | — | |
| SL-Transf.Paradigm=S2G2T2021.05 | 22.45 | — | 48.47 | 35.35 | 27.57 | — | — | — | — | — | — | — | |
| SL-Transfpipeline_strategy=Sign2Gloss2Text (two-stage)2022.03 | 22.45 | — | 48.47 | 35.35 | 27.57 | — | — | — | — | — | — | — | |
| SL-TransformerTranslation Pipeline Setting=Cascading: Sign2Gloss2Text2022.04 | 22.45 | — | — | — | — | — | — | — | — | — | — | — | |
| GFSLT-VLP + SignCLSignCL integration stage=Finetuning2024.05 | 22.16 | — | 48.56 | 35.04 | 27.73 | — | — | 62.73 | 48.17 | — | — | — | |
| S2(G+T)type=sign-to-gloss-and-text2021.09 | 21.8 | — | — | — | — | — | — | — | — | — | — | — | |
| SignBTModality=Feature stitching, Sign source=Same corpus, Augmentation Level=Feature-level (V2T)2026.06 | 21.68 | — | — | — | — | — | — | — | — | — | — | 2.64 | |
| BN-TIN-Transf.Paradigm=S2T, baseline=true2021.05 | 21.68 | 46.98 | 47.57 | 34.64 | 26.78 | — | — | — | — | — | — | — | |
| BN-TIN-TransInput Modality=RGB2023.05 | 21.68 | 46.98 | 47.57 | 34.64 | 26.78 | — | — | — | — | — | — | — | |
| BN-TIN-Transf.LLM=false2023.08 | 21.68 | 46.98 | 47.57 | 34.64 | 26.78 | — | — | — | — | — | — | — | |
| BN-TIN-TransInput Modality=RGB2024.05 | 21.68 | 46.98 | 47.57 | 34.64 | 26.78 | — | — | — | — | — | — | — | |
| BN-TIN-Tran.+SignBTCategory=Gloss-Based, Extra Technique=Scratch2024.11 | 21.68 | 46.98 | — | — | — | — | — | — | — | — | — | — | |
| ConSLTTranslation Pipeline Setting=End-to-end: Sign2Text2022.04 | 21.59 | 47.69 | — | — | — | — | — | — | — | — | — | — | |
| ConSLTSetting=Gloss-based2024.04 | 21.59 | 47.69 | — | — | — | — | — | — | — | — | — | — | |
| Signformer - F w/ CoPECategory=Ours (Gloss-Free), Extra Technique=Scratch, Params (M)=6.442024.11 | 21.57 | 47.41 | — | — | — | — | — | — | — | — | — | — | |
| Signformer - M w/ CoPECategory=Ours (Gloss-Free), Extra Technique=Scratch, Params (M)=2.702024.11 | 21.56 | 47.98 | — | — | — | — | — | — | — | — | — | — | |
| GFSLT-VLPSetting=Gloss-free2024.04 | 21.44 | 42.49 | 43.71 | 33.18 | 26.11 | — | — | — | — | — | — | — | |
| GFSLT-VLPNote=Reported in original paper [52]2024.05 | 21.44 | — | 43.71 | 33.18 | 26.11 | — | — | — | 42.49 | — | — | — | |
| GFSLT-VIPCategory=Gloss-Free, Extra Technique=Pretrained; LM; Data Aug; Knowledge + Embedding; Transfer Learning, Params (M)=2172024.11 | 21.44 | 43.72 | — | — | — | — | — | — | — | — | — | — | |
| GFSLT-VLPSign Representation=Gloss-Free2025.07 | 21.44 | — | 43.71 | 33.18 | 26.11 | — | — | — | — | — | 42.49 | — | |
| LLM-Guided Video StitchingModality=RGB video stitching, Sign source=Same corpus, Augmentation Level=RGB video stitching, this work (V2T)2026.06 | 21.38 | — | — | — | — | — | — | — | — | — | — | 2.92 |