Skip to Main Content
Table 1: 
Unlabeled constituency scores in the corrected RST-PARSEVAL (Morey et al., 2018) against non-binarized trees. UP and UR represent Unlabeled Precision and Unlabeled Recall, respectively. For reference, we also show the traditional RST-PARSEVAL Micro F1 scores in parentheses. Asterisk indicates that we have borrowed the score from Morey et al. (2018).
Method UP UR Micro F1 
Unsupervised 
RB 7.5 7.7 7.6 (54.6) 
〈RBs,RBd〉 47.9 49.7 48.8 (74.8) 
〈RBs,RBp, RBd〉 57.9 60.2 59.0 (79.9) 
 
LB 7.5 7.7 7.6 (54.6) 
〈LBs, LBd〉 41.7 43.3 42.5 (71.7) 
〈LBs, LBp, LBd〉 50.5 52.5 51.5 (76.2) 
 
BU 19.2 19.9 19.5 (60.5) 
〈BUs, BUd〉 47.9 49.8 48.8 (74.9) 
〈BUs, BUp, BUd〉 54.5 56.6 55.5 (78.1) 
 
RBs*,RBp,RBd⋯ (a) 64.5 67.0 65.7 (83.2) 
RBs*,RBp,LBd⋯ (b) 65.6 68.1 66.8 (83.7) 
 
Kobayashi et al. (2019) − − − (80.8) 
 
Ours, initialized by (a) 66.2 68.8 67.5 (84.0) 
Ours, initialized by (b) 66.8 69.4 68.0 (84.3
Ours (b) + Aug. 67.3 69.9 68.6 (84.6
 
Supervised 
Ours, supervised 68.3 70.9 69.6 (85.1) 
Feng and Hirst (2014)* − − − (84.4) 
Joty et al. (2015)* − − − (82.5) 
 
Human − − − (88.7) 
Method UP UR Micro F1 
Unsupervised 
RB 7.5 7.7 7.6 (54.6) 
〈RBs,RBd〉 47.9 49.7 48.8 (74.8) 
〈RBs,RBp, RBd〉 57.9 60.2 59.0 (79.9) 
 
LB 7.5 7.7 7.6 (54.6) 
〈LBs, LBd〉 41.7 43.3 42.5 (71.7) 
〈LBs, LBp, LBd〉 50.5 52.5 51.5 (76.2) 
 
BU 19.2 19.9 19.5 (60.5) 
〈BUs, BUd〉 47.9 49.8 48.8 (74.9) 
〈BUs, BUp, BUd〉 54.5 56.6 55.5 (78.1) 
 
RBs*,RBp,RBd⋯ (a) 64.5 67.0 65.7 (83.2) 
RBs*,RBp,LBd⋯ (b) 65.6 68.1 66.8 (83.7) 
 
Kobayashi et al. (2019) − − − (80.8) 
 
Ours, initialized by (a) 66.2 68.8 67.5 (84.0) 
Ours, initialized by (b) 66.8 69.4 68.0 (84.3
Ours (b) + Aug. 67.3 69.9 68.6 (84.6
 
Supervised 
Ours, supervised 68.3 70.9 69.6 (85.1) 
Feng and Hirst (2014)* − − − (84.4) 
Joty et al. (2015)* − − − (82.5) 
 
Human − − − (88.7) 
Close Modal

or Create an Account

Close Modal
Close Modal