Skip to Main Content
Table 1: 

Soft-match span F1 on the test set, calculated using the precision and recall of predicted tokens vs. gold truth tokens; allows for partial matches. Word-level baselines are compared against our best performing BERT-based span alignment model.

MethodPRF1
DiscAlign 34.11 39.69 36.69 
FastAlign 78.64 72.13 75.25 
 
Df+Mx+Cue+SBCE 96.75 88.24 92.30 
MethodPRF1
DiscAlign 34.11 39.69 36.69 
FastAlign 78.64 72.13 75.25 
 
Df+Mx+Cue+SBCE 96.75 88.24 92.30 
Close Modal

or Create an Account

Close Modal
Close Modal