Part-of-Speech Tagging on Penn TreeBank (test)
97.96AccuracyMeta-BiLSTM
Evaluation Results
| Method | Links | |
|---|---|---|
| Meta-BiLSTMmodel type=sentence-based character model2018.05 | 97.96 | |
| C2W(tanh)+word (sskip)hand-crafted features=no, additional data=yes2015.08 | 97.78 | |
| Ling et al. (2015)2016.03 | 97.78 | |
| Ling et al. (2015)2017.03 | 97.78 | |
| Ling et al. (2015)2016.11 | 97.78 | |
| Choi2018.05 | 97.64 | |
| C2W+featureshand-crafted features=yes, additional data=no2015.08 | 97.57 | |
| DMNEnsemble=top 4 models2015.06 | 97.56 | |
| Kumar et al. (2016)2016.11 | 97.56 | |
| Huang et al.2018.05 | 97.55 | |
| Bi-directional LSTM-CNNs-CRFneural network based=true2016.03 | 97.55 | |
| Huang et al. (2015)2017.03 | 97.55 | |
| Ma & Hovy (2016)2017.03 | 97.55 | |
| Transfer Learning ModelTransfer setting=w/o transfer2017.03 | 97.55 | |
| Transfer Learning ModelTransfer setting=w/ transfer2017.03 | 97.55 | |
| JMTallTasks=Joint all tasks2016.11 | 97.55 | |
| Ma and Hovy (2016)2016.11 | 97.55 | |
| C2W+word (sskip)hand-crafted features=no, additional data=yes2015.08 | 97.54 | |
| Søgaard2018.05 | 97.5 | |
| SCNN2015.06 | 97.5 | |
| SCCNhand-crafted features=yes, additional data=yes2015.08 | 97.5 | |
| Søgaard (2011)2016.03 | 97.5 | |
| Søgaard (2011)2016.11 | 97.5 | |
| Our Local (B=8)Normalization Strategy=Local, Beam Size (B)=82016.03 | 97.45 | |
| Andor et al.2018.05 | 97.44 | |
| Spoustova et al.2015.06 | 97.44 | |
| Morčehand-crafted features=yes, additional data=yes2015.08 | 97.44 | |
| Our Local (B=1)Normalization Strategy=Local, Beam Size (B)=12016.03 | 97.44 | |
| Our Global (B=8)Normalization Strategy=Global, Beam Size (B)=82016.03 | 97.44 | |
| word (sskip)hand-crafted features=no, additional data=yes2015.08 | 97.42 | |
| Dozat et al.2018.05 | 97.41 | |
| Suzuki et al.2015.06 | 97.4 | |
| C2Wrepresentation=compositional character-to-word2015.08 | 97.36 | |
| C2Whand-crafted features=no, additional data=no2015.08 | 97.36 | |
| structReghand-crafted features=yes, additional data=no2015.08 | 97.36 | |
| Sun (2014)2016.03 | 97.36 | |
| word+featureshand-crafted features=yes, additional data=no2015.08 | 97.34 | |
| Shen et al. (2007)2016.03 | 97.33 | |
| Stanfordfeatures=default set of features2015.08 | 97.32 | |
| Stanford 2.0hand-crafted features=yes, additional data=no2015.08 | 97.32 | |
| CNNhand-crafted features=no, additional data=yes2015.08 | 97.32 | |
| Santos and Zadrozny (2014)neural network based=true2016.03 | 97.32 | |
| Collobert et al. (2011)neural network based=true2016.03 | 97.29 | |
| Collobert et al. (2011)2017.03 | 97.29 | |
| Collobert et al. (2011)2016.11 | 97.29 | |
| Manning (2011)2016.03 | 97.28 | |
| Tsuruoka et al. (2011)2016.11 | 97.28 | |
| Sogaard2015.06 | 97.27 | |
| Toutanova et al. (2003)2016.03 | 97.27 | |
| Toutanova et al. (2003)2016.11 | 97.27 | |
| Linear CRF2016.03 | 97.17 | |
| Giménez and Màrquez (2004)2016.03 | 97.16 | |
| SVMTool2015.06 | 97.15 | |
| Wordrepresentation=word lookup tables2015.08 | 96.97 | |
| wordhand-crafted features=no, additional data=no2015.08 | 96.7 | |
| BI-LSTM-CRF (Senna)extra data=Yes2015.08 | 0.9755 | |
| Semi-supervised condensed nearest neighborextra data=Yes2015.08 | 0.975 | |
| BI-LSTM-CRFextra data=No2015.08 | 0.9743 | |
| CRFs with structure regularizationextra data=No2015.08 | 0.9736 | |
| Bidirectional perceptron learningextra data=No2015.08 | 0.9733 | |
| Conv network tagger (senna)extra data=Yes2015.08 | 0.9729 | |
| Maximum entropy cyclic dependency networkextra data=No2015.08 | 0.9724 | |
| SVM based taggerextra data=No2015.08 | 0.9716 | |
| Conv network taggerextra data=No2015.08 | 0.9637 |