Multilingual Constituency Parsing on SPMRL 2013 2014 (test)
87.51French ScoreNFC
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| NFC2021.09 | 87.51 | — | 91.73 | 90.43 | — | 94.95 | 89.07 | 96.33 | — | 91.67 | 89 | 94.34 | |
| Multilingual Self-Attentive Parsermodel_configuration=one model per language2018.12 | 87.42 | 87.97 | 91.63 | 90.2 | 92.99 | 94.9 | 88.8 | 96.36 | 88.86 | 91.01 | — | — | |
| Kitaev and Klein (2018)2021.09 | 87.42 | — | 91.63 | 90.2 | — | 94.9 | 88.8 | 96.36 | — | 91.55 | 88.81 | 94.3 | |
| Kitaev and Klein (2018)Reproduced baseline=true2021.09 | 87.38 | — | 91.66 | 90.25 | — | 94.56 | 88.91 | 96.14 | — | 91.48 | 88.85 | 94.12 | |
| Multilingual Self-Attentive Parsermodel_configuration=joint multilingual model2018.12 | 87.35 | 87.44 | 90.7 | 88.4 | 92.95 | 94.6 | 88.96 | 96.26 | 89.94 | 90.73 | — | — | |
| Nguyen et al. (2020)2021.09 | 86.69 | — | 92.02 | 90.28 | — | 94.24 | 88.71 | 96.14 | — | 91.34 | 88.56 | 94.13 | |
| Self-Attentive EncoderSelection criteria=Best performance on dev set per language2018.05 | 84.06 | 85.61 | 89.71 | 87.69 | 90.35 | 92.69 | 86.59 | 93.69 | 84.45 | 88.32 | — | — | |
| Kitaev and Kleinuses word embeddings=false2018.12 | 84.06 | 85.61 | 89.71 | 87.69 | 90.35 | 92.69 | 86.59 | 93.69 | 84.45 | 88.32 | — | — | |
| Cross and Huang2018.05 | 83.31 | — | — | — | — | — | — | — | — | — | — | — | |
| Björkelund et al.Ensemble=true, Lexical representation=Character LSTM2018.05 | 82.53 | 81.32 | 88.24 | 81.66 | 89.8 | 91.72 | 83.81 | 90.5 | 85.5 | 86.12 | — | — | |
| Björkelund et al.architecture=character LSTM2018.12 | 82.53 | 81.32 | 88.24 | 81.66 | 89.8 | 91.72 | 83.81 | 90.5 | 85.5 | 86.12 | — | — | |
| Coavoux and CrabbéPart-of-speech tags=predicted2018.05 | 82.49 | 82.92 | 88.81 | 85.34 | 89.87 | 92.34 | 86.04 | 93.64 | 84 | 87.27 | — | — | |
| Coavoux and Crabbéinput=predicted part-of-speech tags2018.12 | 82.49 | 82.92 | 88.81 | 85.34 | 89.87 | 92.34 | 86.04 | 93.64 | 84 | 87.27 | — | — |