machine translation 0.0021606159999999998
english translation 0.002009393
translation tasks 0.001896645
translation probabili 0.0018754520000000001
training bleu 0.001775551
translation 0.00163846
sentence bleu 0.0015622890000000001
such feature 0.0015540039999999999
model loss 0.001504704
feature vector 0.001503572
ing algorithm 0.001430458
dense features 0.001423567
basic features 0.001404322
large feature 0.001380681
small feature 0.001373879
group features 0.001355728
feature size 0.001348295
feature vectors 0.001348102
bleu score 0.0013251880000000001
provenance features 0.001319563
relaxed algorithm 0.00131534
batch training 0.001312083
invoke algorithm 0.001291574
model parameters 0.001252099
maximum mert 0.001236996
test set 0.0012277
bleu scores 0.001224571
feature settings 0.0012210749999999999
same loss 0.001213369
level bleu 0.001207093
feature templates 0.001206725
language model 0.001200716
model score 0.001185394
bleu difference 0.001180025
bleu calcula 0.001166072
other hypotheses 0.001151804
single sentence 0.001145145
bleu differentials 0.001144529
source sentence 0.0010957200000000001
features 0.0010826
sentence pairs 0.001073593
other forms 0.00107327
other cor 0.001058591
algorithm 0.00105445
standard mira 0.001053679
sentence hypothesis 0.001049504
next sentence 0.001046561
batch tuning 0.001043267
other optimiza 0.001033928
parameter optimization 0.00102899
model losses 0.001026163
tuning methods 0.001017912
online learning 0.001008316
hinge loss 0.001005584
dev set 0.001005229
target side 9.85156E-4
loss definitions 9.80355E-4
first case 9.75301E-4
fbis data 9.7461E-4
tion system 9.670319999999999E-4
mira method 9.64303E-4
beam search 9.639170000000001E-4
mira approach 9.61319E-4
source side 9.58236E-4
testing set 9.57679E-4
sentence level 9.52478E-4
input sentence 9.51209E-4
feature 9.47808E-4
reference translations 9.467829999999999E-4
only test 9.21935E-4
hypothesis space 9.217159999999999E-4
source corpus 9.20071E-4
new mira 9.08855E-4
mira batch 9.0817E-4
batch mira 9.0817E-4
last sentence 9.07787E-4
simple search 9.065410000000001E-4
test sets 9.03917E-4
tuning procedure 9.0211E-4
polated sentence 8.905440000000001E-4
baseline system 8.887299999999999E-4
english side 8.872859999999999E-4
sparse test 8.86097E-4
corpus hypothesis 8.73855E-4
tion objective 8.71946E-4
parallel sentences 8.695560000000001E-4
training 8.67099E-4
tuning strategy 8.58737E-4
statistical machine 8.58677E-4
parameter updates 8.577890000000001E-4
following experiments 8.43657E-4
large number 8.43656E-4
process sentences 8.429850000000001E-4
standard miras 8.40368E-4
tuning pattern 8.371230000000001E-4
future work 8.27178E-4
few parameter 8.23895E-4
baseline mira 8.21064E-4
maximum number 8.15863E-4
machine transla 8.15629E-4
