Skip to Main Content
Table 3: 
Comparison of our models to previous state-of-the-art models on English (PTB) and Chinese (CTB5.1) Penn Treebanks, and German CoNLL 2009 shared task treebank. “T” and “G” specify “Transition-based” and “Graph-based” models. Bold scores are not significantly different from the best score in that column (with α = 0.01).
ModelEnglish (PTB)Chinese (CTB)German (CoNLL)
TypeUASLASUASLASUASLAS
Chen and Manning (2014) 91.8 89.6 83.9 82.4 − − 
Dyer et al. (2015) 93.1 90.9 87.2 85.7 − − 
Ballesteros et al. (2016) 93.56 91.42 87.65 86.21 88.83 86.10 
Cross and Huang (2016) 93.42 91.36 86.35 85.71 − − 
Weiss et al. (2015) 94.26 92.41 − − − − 
Andor et al. (2016) 94.61 92.79 − − 90.91 89.15 
Mohammadshahi and Henderson (2020) 96.11 94.33 − − − − 
Ma et al. (2018) 95.87 94.19 90.59 89.29 93.65 92.11 
Fernández-González and Gómez-Rodríguez (2019) 96.04 94.43 − − − − 
 
Kiperwasser and Goldberg (2016) 93.1 91.0 86.6 85.1 − − 
Wang and Chang (2016) 94.08 91.82 87.55 86.23 − − 
Cheng et al. (2016) 94.10 91.49 88.1 85.7 − − 
Kuncoro et al. (2016) 94.26 92.06 88.87 87.30 91.60 89.24 
Ma and Hovy (2017) 94.88 92.98 89.05 87.74 92.58 90.54 
Ji et al. (2019) 95.97 94.31 − − − − 
 
Li et al. (2020)+ELMo 96.37 94.57 90.51 89.45 − − 
Li et al. (2020)+BERT 96.44 94.63 90.89 89.73 − − 
 
Biaffine (Dozat and Manning, 2016) 95.74 94.08 89.30 88.23 93.46 91.44 
Biaffine+RNGTr 96.44 94.71 91.85 90.12 94.68 93.30 
 
SynTr 96.60 94.94 92.42 90.67 95.11 93.98 
SynTr+RNGTr 96.66 95.01 92.98 91.18 95.28 94.02 
ModelEnglish (PTB)Chinese (CTB)German (CoNLL)
TypeUASLASUASLASUASLAS
Chen and Manning (2014) 91.8 89.6 83.9 82.4 − − 
Dyer et al. (2015) 93.1 90.9 87.2 85.7 − − 
Ballesteros et al. (2016) 93.56 91.42 87.65 86.21 88.83 86.10 
Cross and Huang (2016) 93.42 91.36 86.35 85.71 − − 
Weiss et al. (2015) 94.26 92.41 − − − − 
Andor et al. (2016) 94.61 92.79 − − 90.91 89.15 
Mohammadshahi and Henderson (2020) 96.11 94.33 − − − − 
Ma et al. (2018) 95.87 94.19 90.59 89.29 93.65 92.11 
Fernández-González and Gómez-Rodríguez (2019) 96.04 94.43 − − − − 
 
Kiperwasser and Goldberg (2016) 93.1 91.0 86.6 85.1 − − 
Wang and Chang (2016) 94.08 91.82 87.55 86.23 − − 
Cheng et al. (2016) 94.10 91.49 88.1 85.7 − − 
Kuncoro et al. (2016) 94.26 92.06 88.87 87.30 91.60 89.24 
Ma and Hovy (2017) 94.88 92.98 89.05 87.74 92.58 90.54 
Ji et al. (2019) 95.97 94.31 − − − − 
 
Li et al. (2020)+ELMo 96.37 94.57 90.51 89.45 − − 
Li et al. (2020)+BERT 96.44 94.63 90.89 89.73 − − 
 
Biaffine (Dozat and Manning, 2016) 95.74 94.08 89.30 88.23 93.46 91.44 
Biaffine+RNGTr 96.44 94.71 91.85 90.12 94.68 93.30 
 
SynTr 96.60 94.94 92.42 90.67 95.11 93.98 
SynTr+RNGTr 96.66 95.01 92.98 91.18 95.28 94.02 
Close Modal

or Create an Account

Close Modal
Close Modal