machine translation 0.003330334
translation forests 0.002980666
translation sys 0.002978721
translation pipeline 0.0029635950000000003
translation 0.00263421
consensus decoding 0.002407182
model score 0.002392566
model weights 0.002114756
decoding techniques 0.002035325
viterbi decoding 0.002030867
decoding procedures 0.001968168
rate training 0.0018415200000000001
set sentence 0.001734445
bleu score 0.001725076
objective function 0.00168303
model 0.00167877
decoding 0.00163645
lation models 0.001574692
minimum error 0.001549256
bleu tuning 0.0015292130000000002
reference translations 0.001507483
bleu scores 0.001475845
test set 0.001424095
training 0.00140748
set performance 0.001406764
error rate 0.0013576719999999999
max bleu 0.001353257
suming bleu 0.001340667
entiable function 0.0013311640000000001
mert con 0.0013127219999999999
imum error 0.001309134
new dynamic 0.001291078
coupling mert 0.001288139
tuning set 0.001266976
parameter tuning 0.0012436600000000002
efficient consensus 0.0012259390000000001
features 0.00120695
posterior probability 0.001202907
tuning objective 0.001200793
evaluation measure 0.001189869
models 0.00117336
recent work 0.00115714
ment consensus 0.001144421
same quantities 0.0011317200000000001
fast consensus 0.001116256
downstream consensus 0.00110384
machine trans 0.001100415
previous work 0.001047612
alternative objective 0.001035845
novel objective 0.001030155
simple gradient 0.0010202879999999998
continuous objective 0.001013672
viterbi derivation 0.0010066839999999999
gradient ascent 0.001006452
minimum bayes 0.001004109
function 0.00100017
standard viterbi 9.97735E-4
sentence 9.85402E-4
standard max 9.452950000000001E-4
error 9.23632E-4
translations 9.19721E-4
lation decoders 9.05255E-4
methods leverage 8.969080000000001E-4
ascent procedure 8.61854E-4
recent con 8.55726E-4
similar sta 8.5315E-4
critical quantities 8.48859E-4
second benefit 8.38994E-4
efficient computation 8.0531E-4
empirical benefit 8.02E-4
dynamic program 8.01399E-4
single top 7.97206E-4
evaluation 7.91318E-4
complementary change 7.7987E-4
novel dynamic 7.76585E-4
consensus 7.70732E-4
predictive distribution 7.68537E-4
packed forests 7.6402E-4
principle advan 7.6016E-4
leverage compact 7.59589E-4
compact repre 7.591589999999999E-4
bility benefits 7.591119999999999E-4
expected counts 7.57862E-4
functional form 7.47068E-4
probability 7.37505E-4
viterbi predictions 7.32325E-4
parameter 7.25727E-4
score 7.13796E-4
machine 6.96124E-4
objective 6.8286E-4
performance 6.57721E-4
minimum 6.25624E-4
derivation 6.12267E-4
reference 5.87762E-4
gradient 5.60069E-4
approach 5.42361E-4
methods 5.30004E-4
tuning 5.17933E-4
decoders 5.03923E-4
approximations 4.9707E-4
