Semantic Role Labeling on CoNLL 2005 (Brown)
92.1F1 ScoreDEEPSTRUCT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DEEPSTRUCTmode=w/ finetune2022.05 | 92.1 | — | |
| DEEPSTRUCTtraining protocol=w/ finetune2022.05 | 92.1 | — | |
| DEEPSTRUCTmode=multi-task2022.05 | 92 | — | |
| DEEPSTRUCTtraining protocol=multi-task2022.05 | 92 | — | |
| TANLTraining Setting=multi-dataset2021.01 | 84.3 | — | |
| TANLTraining Setting=multi-task2021.01 | 84.1 | — | |
| TANL2022.05 | 84.1 | — | |
| TANLtraining protocol=multitask2022.05 | 84.1 | — | |
| BERT SRLTraining Setting=standard2021.01 | 82 | — | |
| TANLTraining Setting=standard2021.01 | 82 | — | |
| Shi and Lin, 20192022.05 | 82 | — | |
| BERT SRL2022.05 | 82 | — | |
| TANL2022.05 | 82 | — | |
| He+ 18Model Type=Single2018.10 | 80.4 | — | |
| ELMO-SPANModel Type=Ensemble2018.10 | 79.6 | — | |
| ELMO-SPANModel Type=Single2018.10 | 78.7 | — | |
| Dep and SpanTraining Setting=standard2021.01 | 76.4 | — | |
| Dep and Span2022.05 | 76.4 | — | |
| Tan+ 18Model Type=Ensemble2018.10 | 74.8 | — | |
| Tan+ 18Model Type=Single2018.10 | 74.1 | — | |
| He+ 17Model Type=Ensemble2018.10 | 73.6 | — | |
| Strubell+ 18Model Type=Single2018.10 | 72.6 | — | |
| FitzGerald+ 15Model Type=Ensemble2018.10 | 72.2 | — | |
| He+ 17Model Type=Single2018.10 | 72.1 | — | |
| Täckström+ 15Model Type=Single2018.10 | 71.3 | — | |
| FitzGerald+ 15Model Type=Single2018.10 | 71.2 | — | |
| Zhou+ 15Model Type=Single2018.10 | 69.4 | — | |
| Toutanova+ 08Model Type=Ensemble2018.10 | 68.8 | — | |
| Toutanova+ 08Model Type=Single2018.10 | 67.8 | — | |
| Punyakanok+ 08Model Type=Single2018.10 | 67.8 | — | |
| Punyakanok+ 08Model Type=Ensemble2018.10 | 67.8 | — | |
| Full Fine-tuningBase Model=RoBERTa-large2023.03 | — | 85.6 | |
| P-tuningBase Model=RoBERTa-large2023.03 | — | 70.7 | |
| P-tuning V2Base Model=RoBERTa-large2023.03 | — | 84.3 | |
| P-tuning V2 (reproduced)Base Model=RoBERTa-large, reproduced=true2023.03 | — | 83.9 | |
| Zero-init.Base Model=RoBERTa-large2023.03 | — | 84.7 |