Dependency Parsing on English PTB Stanford Dependencies (test)
96.04UASLeft-to-Right Dependency Parser (This work)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Left-to-Right Dependency Parser (This work)paradigm=transition-based, implementation_framework=current, model_type=single-model, repetitions=52019.03 | 96.04 | 94.43 | |
| STACKPTRModel Type=Transition-based, Variant=Full2018.05 | 95.87 | 94.19 | |
| Ma et al.paradigm=transition-based, implementation_framework=same as this paper, model_type=single-model2019.03 | 95.87 | 94.19 | |
| STACKPTRModel Type=Transition-based, Variant=+sib2018.05 | 95.85 | 94.18 | |
| BiAFModel Type=Graph-based, Variant=re-impl2018.05 | 95.84 | 94.21 | |
| Ma et al.paradigm=graph-based, implementation_framework=same as this paper, model_type=single-model2019.03 | 95.84 | 94.21 | |
| STACKPTRModel Type=Transition-based, Variant=+gpar2018.05 | 95.78 | 94.12 | |
| STACKPTRModel Type=Transition-based, Variant=Org2018.05 | 95.77 | 94.12 | |
| BiAFModel Type=Graph-based, Variant=Original2018.05 | 95.74 | 94.08 | |
| Dozat and Manningparadigm=graph-based, model_type=single-model2019.03 | 95.74 | 94.08 | |
| Ma and HovyModel Type=Graph-based2018.05 | 94.88 | 92.98 | |
| Ma and Hovyparadigm=graph-based, model_type=single-model2019.03 | 94.88 | 92.96 | |
| Andor et al.Model Type=Transition-based2018.05 | 94.61 | 92.79 | |
| Andor et al. (2016)Parsing Approach=Transition (beam), Pretrained word embeddings=true2016.09 | 94.61 | 92.79 | |
| Andor et al.paradigm=transition-based, model_type=single-model2019.03 | 94.61 | 92.79 | |
| Ensemble MST ParserParsing Approach=Transition (greedy), Pretrained word embeddings=true, Ensemble Size=202016.09 | 94.51 | 92.7 | |
| Fernández-G and Gómez-Rparadigm=transition-based, model_type=single-model2019.03 | 94.5 | 92.4 | |
| Qi and Manningparadigm=transition-based, model_type=single-model2019.03 | 94.3 | 92.2 | |
| Zhang et al.paradigm=graph-based, model_type=single-model2019.03 | 94.3 | 91.95 | |
| Weiss et al.Model Type=Transition-based2018.05 | 94.26 | 92.41 | |
| Kuncoro et al.Model Type=Graph-based2018.05 | 94.26 | 92.06 | |
| Neural FOG, CD (distilled)Parsing Approach=Graph (1st order), Pretrained word embeddings=true, Distilled=true2016.09 | 94.26 | 92.06 | |
| Kuncoro et al.paradigm=graph-based, model_type=single-model2019.03 | 94.26 | 92.06 | |
| Alberti et al.paradigm=transition-based, model_type=single-model2019.03 | 94.23 | 92.36 | |
| Cheng et al.Model Type=Graph-based2018.05 | 94.1 | 91.49 | |
| Cheng et al.paradigm=graph-based, model_type=single-model2019.03 | 94.1 | 91.49 | |
| Wang and ChangModel Type=Graph-based2018.05 | 94.08 | 91.82 | |
| Wang and Chang (2016)Parsing Approach=Graph (1st order), Pretrained word embeddings=true2016.09 | 94.08 | 91.82 | |
| Wang and Changparadigm=graph-based, model_type=single-model2019.03 | 94.08 | 91.82 | |
| Weiss15method=transition (beam), representation=large feature set (dense), pretrained_embeddings=true2016.03 | 94 | 92 | |
| Weiss et al. (2015)Parsing Approach=Transition (beam), Pretrained word embeddings=true2016.09 | 94 | 92 | |
| Weiss et al.paradigm=transition-based, model_type=single-model2019.03 | 93.99 | 92.05 | |
| BiLSTM Parsermethod=transition (greedy, dyn-oracle), representation=11 BiLSTM vectors, pretrained_embeddings=true2016.03 | 93.9 | 91.9 | |
| Kiperwasser and GoldbergModel Type=Transition-based2018.05 | 93.9 | 91.9 | |
| Kiperwasser and Goldberg (2016)Parsing Approach=Transition (greedy), Pretrained word embeddings=true2016.09 | 93.9 | 91.9 | |
| Kiperwasser and Goldbergparadigm=transition-based, model_type=single-model2019.03 | 93.9 | 91.9 | |
| LeZuidema14method=reranking /blend, representation=inside-outside recursive net, pretrained_embeddings=true2016.03 | 93.8 | 91.5 | |
| Le and Zuidema (2014)Parsing Approach=Reranking/blend, Pretrained word embeddings=true2016.09 | 93.8 | 91.5 | |
| Neural FOG, CHParsing Approach=Graph (1st order), Pretrained word embeddings=true2016.09 | 93.76 | 91.6 | |
| BiLSTM Parsermethod=transition (greedy, dyn-oracle), representation=4 BiLSTM vectors, pretrained_embeddings=true2016.03 | 93.6 | 91.5 | |
| Ballesteros16method=transition (greedy, dyn-oracle), representation=Stack-LSTM + composition, pretrained_embeddings=true2016.03 | 93.6 | 91.4 | |
| Ballesteros et al.Model Type=Transition-based, Year=20162018.05 | 93.56 | 91.42 | |
| Ballesteros et al. (2016)Parsing Approach=Transition (greedy), Pretrained word embeddings=true2016.09 | 93.56 | 91.42 | |
| Ballesteros et al.paradigm=transition-based, model_type=single-model2019.03 | 93.56 | 91.42 | |
| Bohnet and NivreModel Type=Transition-based2018.05 | 93.33 | 91.22 | |
| BiLSTM Parsermethod=transition (greedy, dyn-oracle), representation=11 BiLSTM vectors, pretrained_embeddings=false2016.03 | 93.2 | 91.2 | |
| Weiss15method=transition (greedy), representation=large feature set (dense), pretrained_embeddings=true2016.03 | 93.2 | 91.2 | |
| AndorKte=12016.06 | 93.17 | 91.18 | |
| BiLSTM Parsermethod=graph, 1st order, representation=2 BiLSTM vectors, pretrained_embeddings=false2016.03 | 93.1 | 91 | |
| BiLSTM Parsermethod=transition (greedy, dyn-oracle), representation=4 BiLSTM vectors, pretrained_embeddings=false2016.03 | 93.1 | 91 | |
| Martins13 (TurboParser)method=graph, 3rd order+, representation=large feature set (sparse), pretrained_embeddings=false2016.03 | 93.1 | — | |
| Dyer15method=transition (greedy), representation=Stack-LSTM + composition, pretrained_embeddings=true2016.03 | 93.1 | 90.9 | |
| Dyer et al.Model Type=Transition-based2018.05 | 93.1 | 90.9 | |
| Kiperwasser and GoldbergModel Type=Graph-based2018.05 | 93.1 | 91 | |
| Dyer et al. (2015)Parsing Approach=Transition (greedy), Pretrained word embeddings=true2016.09 | 93.1 | 90.9 | |
| Martins et al. (2013)Parsing Approach=Graph (3rd order), Pretrained word embeddings=false2016.09 | 93.1 | — | |
| Kiperwasser and Goldberg (2016)Parsing Approach=Graph (1st order), Pretrained word embeddings=true2016.09 | 93.1 | 91 | |
| Dyer et al.paradigm=transition-based, model_type=single-model2019.03 | 93.1 | 90.9 | |
| Kiperwasser and Goldbergparadigm=graph-based, model_type=single-model2019.03 | 93.1 | 91 | |
| BiLSTM Parsermethod=graph, 1st order, representation=2 BiLSTM vectors, pretrained_embeddings=true2016.03 | 93 | 90.9 | |
| Ballesteros16method=transition (greedy, dyn-oracle), representation=Stack-LSTM + composition, pretrained_embeddings=false2016.03 | 92.7 | 90.6 | |
| Dyer15method=transition (greedy), representation=Stack-LSTM + composition, pretrained_embeddings=false2016.03 | 92.4 | 90 | |
| Chen and ManningModel Type=Transition-based2018.05 | 91.8 | 89.6 | |
| Chen and Manning (2014)Parsing Approach=Transition (greedy), Pretrained word embeddings=true2016.09 | 91.8 | 89.6 | |
| Chen and Manningparadigm=transition-based, model_type=single-model2019.03 | 91.8 | 89.6 | |
| Ballesteros et al.Model Type=Transition-based2018.05 | 91.63 | 89.44 | |
| Ballesteros et al. (2015)Parsing Approach=Transition (greedy), Pretrained word embeddings=false2016.09 | 91.63 | 89.44 | |
| ConBSOKte=10, Ktr=62016.06 | 91.57 | 87.26 | |
| ConBSOKte=5, Ktr=62016.06 | 91.25 | 86.92 | |
| BSOKte=10, Ktr=62016.06 | 91.17 | 87.41 | |
| BSOKte=5, Ktr=62016.06 | 91 | 87.18 | |
| seq2seqKte=10, Ktr=62016.06 | 88.66 | 84.33 | |
| seq2seqKte=5, Ktr=62016.06 | 88.53 | 84.16 | |
| seq2seqKte=1, Ktr=62016.06 | 87.33 | 82.26 | |
| BSOKte=1, Ktr=62016.06 | 86.91 | 82.11 | |
| ConBSOKte=1, Ktr=62016.06 | 85.11 | 79.32 |