Span SRL results without pre-identified predicates on the CoNLL-2005 and CoNLL-2012 data sets. The “PLM” column indicates whether and which pre-trained language model is used, the “SYN” column indicates whether syntax information is employed, and “+E” in the “PLM” column shows that the model leverages the ELMo for pre-trained language model features. [Ens.] is used to specify the ensemble system and [Joint] means joint learning with other tasks.
| System . | PLM . | SYN . | CoNLL05 WSJ . | CoNLL05 Brown . | CoNLL12 . | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| P . | R . | F1 . | P . | R . | F1 . | P . | R . | F1 . | |||
| (He et al. 2017) | N | 80.2 | 82.3 | 81.2 | 67.6 | 69.6 | 68.5 | 78.6 | 75.1 | 76.8 | |
| [Ens.] (He et al. 2017) | N | 82.0 | 83.4 | 82.7 | 69.7 | 70.5 | 70.1 | 80.2 | 76.6 | 78.4 | |
| (He et al. 2018a) | +E | N | 84.8 | 87.2 | 86.0 | 73.9 | 78.4 | 76.1 | 81.9 | 84.0 | 82.9 |
| (Strubell et al. 2018) | Y | 84.0 | 83.2 | 83.6 | 73.3 | 70.6 | 71.9 | 81.9 | 79.6 | 80.7 | |
| +E | Y | 86.7 | 86.4 | 86.6 | 79.0 | 77.2 | 78.1 | 84.0 | 82.3 | 83.1 | |
| [Joint] (Zhou, Li, and Zhao 2020) | N | 83.7 | 85.5 | 84.6 | 72.0 | 73.1 | 72.6 | − | − | − | |
| +E | N | 85.3 | 87.7 | 86.5 | 76.1 | 78.3 | 77.2 | − | − | − | |
| Sequence-based | +E | N | 84.4 | 83.6 | 84.0 | 76.5 | 73.9 | 75.2 | 81.7 | 82.9 | 82.3 |
| +GCN Syntax Encoder | +E | Y | 85.5 | 84.3 | 84.9 | 78.8 | 73.4 | 76.0 | 83.1 | 82.5 | 82.8 |
| +SA-LSTM Syntax Encoder | +E | Y | 85.0 | 84.2 | 84.6 | 74.9 | 76.7 | 75.8 | 83.1 | 81.9 | 82.5 |
| +Tree-LSTM Syntax Encoder | +E | Y | 84.7 | 84.1 | 84.4 | 76.2 | 75.2 | 75.7 | 82.7 | 81.9 | 82.3 |
| Tree-based | +E | N | 85.4 | 83.6 | 84.5 | 76.1 | 75.1 | 75.6 | 83.3 | 81.9 | 82.6 |
| +GCN Syntax Encoder | +E | Y | 84.5 | 85.9 | 85.2 | 76.7 | 75.9 | 76.3 | 82.9 | 83.3 | 83.1 |
| +SA-LSTM Syntax Encoder | +E | Y | 85.0 | 85.0 | 85.0 | 77.2 | 75.0 | 76.1 | 83.5 | 82.7 | 83.1 |
| +Tree-LSTM Syntax Encoder | +E | Y | 85.7 | 84.1 | 84.9 | 75.5 | 75.9 | 75.7 | 83.0 | 82.4 | 82.7 |
| Graph-based (2019a) | +E | N | 85.2 | 87.5 | 86.3 | 74.7 | 78.1 | 76.4 | 84.9 | 81.4 | 83.1 |
| +Constituent Soft Pruning | +E | Y | 87.1 | 85.7 | 86.4 | 77.0 | 76.2 | 76.6 | 83.4 | 83.2 | 83.3 |
| +GCN Syntax Encoder | +E | Y | 86.9 | 86.5 | 86.7 | 77.5 | 76.3 | 76.9 | 84.4 | 83.0 | 83.7 |
| +SA-LSTM Syntax Encoder | +E | Y | 87.3 | 85.7 | 86.5 | 76.0 | 77.2 | 76.6 | 83.8 | 83.2 | 83.5 |
| +Tree-LSTM Syntax Encoder | +E | Y | 85.8 | 86.6 | 86.2 | 76.9 | 76.1 | 76.5 | 83.6 | 82.8 | 83.2 |
| System . | PLM . | SYN . | CoNLL05 WSJ . | CoNLL05 Brown . | CoNLL12 . | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| P . | R . | F1 . | P . | R . | F1 . | P . | R . | F1 . | |||
| (He et al. 2017) | N | 80.2 | 82.3 | 81.2 | 67.6 | 69.6 | 68.5 | 78.6 | 75.1 | 76.8 | |
| [Ens.] (He et al. 2017) | N | 82.0 | 83.4 | 82.7 | 69.7 | 70.5 | 70.1 | 80.2 | 76.6 | 78.4 | |
| (He et al. 2018a) | +E | N | 84.8 | 87.2 | 86.0 | 73.9 | 78.4 | 76.1 | 81.9 | 84.0 | 82.9 |
| (Strubell et al. 2018) | Y | 84.0 | 83.2 | 83.6 | 73.3 | 70.6 | 71.9 | 81.9 | 79.6 | 80.7 | |
| +E | Y | 86.7 | 86.4 | 86.6 | 79.0 | 77.2 | 78.1 | 84.0 | 82.3 | 83.1 | |
| [Joint] (Zhou, Li, and Zhao 2020) | N | 83.7 | 85.5 | 84.6 | 72.0 | 73.1 | 72.6 | − | − | − | |
| +E | N | 85.3 | 87.7 | 86.5 | 76.1 | 78.3 | 77.2 | − | − | − | |
| Sequence-based | +E | N | 84.4 | 83.6 | 84.0 | 76.5 | 73.9 | 75.2 | 81.7 | 82.9 | 82.3 |
| +GCN Syntax Encoder | +E | Y | 85.5 | 84.3 | 84.9 | 78.8 | 73.4 | 76.0 | 83.1 | 82.5 | 82.8 |
| +SA-LSTM Syntax Encoder | +E | Y | 85.0 | 84.2 | 84.6 | 74.9 | 76.7 | 75.8 | 83.1 | 81.9 | 82.5 |
| +Tree-LSTM Syntax Encoder | +E | Y | 84.7 | 84.1 | 84.4 | 76.2 | 75.2 | 75.7 | 82.7 | 81.9 | 82.3 |
| Tree-based | +E | N | 85.4 | 83.6 | 84.5 | 76.1 | 75.1 | 75.6 | 83.3 | 81.9 | 82.6 |
| +GCN Syntax Encoder | +E | Y | 84.5 | 85.9 | 85.2 | 76.7 | 75.9 | 76.3 | 82.9 | 83.3 | 83.1 |
| +SA-LSTM Syntax Encoder | +E | Y | 85.0 | 85.0 | 85.0 | 77.2 | 75.0 | 76.1 | 83.5 | 82.7 | 83.1 |
| +Tree-LSTM Syntax Encoder | +E | Y | 85.7 | 84.1 | 84.9 | 75.5 | 75.9 | 75.7 | 83.0 | 82.4 | 82.7 |
| Graph-based (2019a) | +E | N | 85.2 | 87.5 | 86.3 | 74.7 | 78.1 | 76.4 | 84.9 | 81.4 | 83.1 |
| +Constituent Soft Pruning | +E | Y | 87.1 | 85.7 | 86.4 | 77.0 | 76.2 | 76.6 | 83.4 | 83.2 | 83.3 |
| +GCN Syntax Encoder | +E | Y | 86.9 | 86.5 | 86.7 | 77.5 | 76.3 | 76.9 | 84.4 | 83.0 | 83.7 |
| +SA-LSTM Syntax Encoder | +E | Y | 87.3 | 85.7 | 86.5 | 76.0 | 77.2 | 76.6 | 83.8 | 83.2 | 83.5 |
| +Tree-LSTM Syntax Encoder | +E | Y | 85.8 | 86.6 | 86.2 | 76.9 | 76.1 | 76.5 | 83.6 | 82.8 | 83.2 |