Skip to Main Content
Table 8 

%BLEU on tune and test sets for UR→EN translation, using our unsupervised Urdu parser to incorporate source syntactic features. The two QPD rows are statistically indistinguishable on both test sets. Both are significantly better than all Moses results, but Hiero is significantly better than all others.

Urdu→English
model
notes
tune
test 1
test 2
test avg.
Moses Rampion, S = 200 24.6 24.6 24.5 24.5 
Rampion, S = 500 24.7 24.8 24.9 24.8 
SSVM reranking 24.9 24.4 24.7 24.6 
 
QPD TgtTree 25.8 25.4 25.5 25.4 
TgtTree + (unsup.) TreeToTree 25.8 25.4 25.6 25.5 
 
Hiero Rampion 25.7 26.4 26.7 26.6 
Urdu→English
model
notes
tune
test 1
test 2
test avg.
Moses Rampion, S = 200 24.6 24.6 24.5 24.5 
Rampion, S = 500 24.7 24.8 24.9 24.8 
SSVM reranking 24.9 24.4 24.7 24.6 
 
QPD TgtTree 25.8 25.4 25.5 25.4 
TgtTree + (unsup.) TreeToTree 25.8 25.4 25.6 25.5 
 
Hiero Rampion 25.7 26.4 26.7 26.6 
Close Modal

or Create an Account

Close Modal
Close Modal