Dependency Parsing on WSJ (test)
95.74UASDozat and Manning
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Dozat and Manningyear=20172016.11 | 95.74 | 94.08 | |
| JMTalltraining=joint training on all tasks2016.11 | 94.67 | 92.9 | |
| Our GlobalNormalization=Global, Beam size=322016.03 | 94.61 | 92.79 | |
| Andor et al.year=20162016.11 | 94.61 | 92.79 | |
| Our PerceptronParser Category=Tri-training, Beam=82015.06 | 94.26 | 92.41 | |
| Alberti et al. (2015)2016.03 | 94.23 | 92.36 | |
| Alberti et al.year=20152016.11 | 94.23 | 92.36 | |
| Zhang et al.year=20172016.11 | 94.1 | 91.9 | |
| Our PerceptronParser Category=Transition-based, Beam=82015.06 | 93.99 | 92.05 | |
| Weiss et al. (2015)2016.03 | 93.99 | 92.05 | |
| Weiss et al.year=20152016.11 | 93.99 | 92.05 | |
| Our LocalNormalization=Local, Beam size=322016.03 | 93.59 | 91.7 | |
| Our GreedyParser Category=Tri-training, Beam=12015.06 | 93.46 | 91.49 | |
| Singletraining=single-task training2016.11 | 93.35 | 91.42 | |
| Bohnet and Kuhn (2012)Parser Category=Transition-based, Beam=402015.06 | 93.27 | 91.19 | |
| Zhang and McDonald (2014)Parser Category=Graph-based, Beam=n/a2015.06 | 93.22 | 91.02 | |
| Zhang and McDonald (2014)2016.03 | 93.22 | 91.02 | |
| S-LSTM (Dyer et al., 2015)Parser Category=Transition-based, Beam=12015.06 | 93.2 | 90.9 | |
| Our GreedyParser Category=Transition-based, Beam=12015.06 | 93.19 | 91.18 | |
| Dyer et al.year=20152016.11 | 93.1 | 90.9 | |
| Zhang and Nivre (2011)Parser Category=Transition-based, Beam=32, re-implementation=true2015.06 | 93 | 90.95 | |
| Our LocalNormalization=Local, Beam size=12016.03 | 92.95 | 91.02 | |
| Zhang and Nivre (2011)Parser Category=Tri-training, Beam=32, re-implementation=true2015.06 | 92.92 | 90.88 | |
| Martins et al. (2013)Parser Category=Graph-based, Beam=n/a2015.06 | 92.89 | 90.55 | |
| Martins et al. (2013)2016.03 | 92.89 | 90.55 | |
| Bohnet (2010)Parser Category=Graph-based, Beam=n/a2015.06 | 92.88 | 90.71 | |
| Bohnetyear=20102016.11 | 92.88 | 90.71 | |
| Chen and Manning (2014)Parser Category=Transition-based, Beam=12015.06 | 91.8 | 89.6 | |
| Proposed Selected Ensemble w/ society entropy (including Sib&L-NDMV)Type=Selected ensemble, Selection Metric=Society entropy (ours), Model Set=Initial + Sib&L-NDMV2024.12 | 68.8 | — | |
| Proposed Weighted Ensemble (including Sib&L-NDMV)Type=Ensemble, Aggregation Strategy=Weighted, Model Set=Initial + Sib&L-NDMV2024.12 | 68.4 | — | |
| Proposed Selected Ensemble by validation (including Sib&L-NDMV)Type=Selected ensemble, Selection Strategy=Ensemble validation, Model Set=Initial + Sib&L-NDMV2024.12 | 68.4 | — | |
| Sib&L-NDMVType=Additional individual2024.12 | 67.9 | — | |
| Proposed Unweighted Ensemble (including Sib&L-NDMV)Type=Ensemble, Aggregation Strategy=Unweighted, Model Set=Initial + Sib&L-NDMV2024.12 | 67.9 | — | |
| Selected ensemble (Ensemble validation)Type=Selected ensemble, Selection Strategy=Ensemble validation (Caruana et al. 2004), Aggregation Strategy=Proposed2024.12 | 67.8 | — | |
| Joint training: sibling-NDMV + L-NDMVinitialization=Naseem et al. (2010)2020.10 | 67.5 | — | |
| Selected ensemble w/ society entropyType=Selected ensemble, Selection Metric=Society entropy (ours), Aggregation Strategy=Proposed2024.12 | 67.3 | — | |
| Our weighted aggregationType=Ensemble, Aggregation Strategy=Weighted MBR2024.12 | 66.6 | — | |
| Lexicalized sibling-NDMVinitialization=Naseem et al. (2010)2020.10 | 66.4 | — | |
| MaxEncadditional_training_data=true2020.10 | 65.8 | — | |
| Selected ensemble w/ Kuncheva's diversityType=Selected ensemble, Selection Metric=Kuncheva's diversity, Aggregation Strategy=Proposed2024.12 | 65.8 | — | |
| Our unweighted aggregationType=Ensemble, Aggregation Strategy=Unweighted MBR2024.12 | 65.7 | — | |
| sibling-NDMVinitialization=Naseem et al. (2010)2020.10 | 64.5 | — | |
| CSadditional_training_data=true2020.10 | 64.4 | — | |
| Joint training: grand-NDMV + L-NDMVinitialization=Naseem et al. (2010)2020.10 | 64.3 | — | |
| Sib-NDMVType=Ensemble individual2024.12 | 64.3 | — | |
| Selected ensemble w/o diversityType=Selected ensemble, Selection Metric=None, Aggregation Strategy=Proposed2024.12 | 63.8 | — | |
| L-NDMVadditional_training_data=true, initialization=Naseem et al. (2010)2020.10 | 63.2 | — | |
| L-NDMVType=Ensemble individual2024.12 | 62.4 | — | |
| Deterministic variant D-NDMVinitialization=Naseem et al. (2010)2020.10 | 61.4 | — | |
| Variational variant D-NDMVinitialization=Naseem et al. (2010)2020.10 | 60.4 | — | |
| L-NDMVinitialization=Naseem et al. (2010)2020.10 | 59.5 | — | |
| CIM aggregationType=Ensemble, Aggregation Strategy=CIM2024.12 | 58.8 | — | |
| Neural DMV2020.10 | 57.6 | — | |
| grand-NDMVinitialization=Naseem et al. (2010)2020.10 | 57.3 | — | |
| UR-A E-DMV2020.10 | 57 | — | |
| CRFAE2020.10 | 55.7 | — | |
| PR-S2020.10 | 53.3 | — | |
| TSG-DMV2020.10 | 53.1 | — | |
| CRFAEType=Ensemble individual2024.12 | 53 | — | |
| Lexicalized grand-NDMVinitialization=Naseem et al. (2010)2020.10 | 52.6 | — | |
| NE-DMVType=Ensemble individual2024.12 | 51 | — | |
| Convex-MST2020.10 | 48.6 | — | |
| NDMVType=Ensemble individual2024.12 | 48.1 | — | |
| Shared LN2020.10 | 41.4 | — | |
| LN2020.10 | 40.5 | — | |
| DMV2020.10 | 39.4 | — | |
| NVTP2020.10 | 37.8 | — |