Skip to Main Content
Table 4: 
Direct optimization for consistency. We compare training (§5.1) and inference (§5.2) approaches. Bold font indicates the best score.
ModelWERBLEU↓Lex↑Sur
Tri 22.2 19.9 9.72 11.54 
training 24.0 19.9 9.84 12.09 
inference 22.6 19.5 8.79 13.17 
 
DirMu 23.6 18.4 13.89 7.07 
training 23.2 18.9 13.94 7.94 
inference 24.0 18.7 12.63 9.29 
ModelWERBLEU↓Lex↑Sur
Tri 22.2 19.9 9.72 11.54 
training 24.0 19.9 9.84 12.09 
inference 22.6 19.5 8.79 13.17 
 
DirMu 23.6 18.4 13.89 7.07 
training 23.2 18.9 13.94 7.94 
inference 24.0 18.7 12.63 9.29 
Close Modal

or Create an Account

Close Modal
Close Modal