machine translation 0.002452931
training data 0.002319847
training set 0.002312616
feature vector 0.002219314
english translation 0.002217762
mert bleu 0.0021596899999999997
training error 0.002146585
feature weights 0.002083835
training score 0.002062695
feature vectors 0.0018727470000000001
training sample 0.001804599
translation 0.001794
rate training 0.00178963
theoretical training 0.001782503
line search 0.0016904659999999998
training bounds 0.001676258
bleu score 0.0016641450000000001
optimization error 0.001607647
mert ranker 0.001604307
hypothesis features 0.001596713
standard mert 0.0015672149999999998
test set 0.001553713
multiple mert 0.0015404949999999998
feature 0.00153075
linear model 0.0014943259999999998
mert procedure 0.0014574479999999999
mert boost 0.001437264
mert rankers 0.0014294339999999998
training 0.00140567
few features 0.001397779
similar algorithm 0.001391661
bleu statistics 0.001385354
train bleu 0.001378159
general algorithm 0.001359877
general function 0.001354101
bleu scores 0.001333525
above algorithm 0.001322254
development set 0.001312006
small set 0.001305432
actual bleu 0.001292989
ing algorithm 0.001291743
best bleu 0.001284226
train set 0.001277985
bleu statistic 0.001277725
selected bleu 0.001277725
optimization problem 0.001277719
vector output 0.0012568969999999999
output vector 0.0012568969999999999
reranking model 0.00122541
efficient optimization 0.001203985
low optimization 0.001198966
dev set 0.0011981749999999999
optimization properties 0.001182513
optimization procedure 0.00117161
optimization problems 0.001164322
error rate 0.001124875
such complex 0.001075161
features 0.00106953
minimum error 0.001053014
sample weight 0.001031917
ranking output 0.001025817
generalization error 0.001013508
large gap 0.001011481
zero vector 0.001011441
oracle score 0.001008023
machine trans 0.0010054039999999999
natural language 9.98184E-4
effective approach 9.93083E-4
language processing 9.84004E-4
score gap 9.813859999999999E-4
various methods 9.756490000000001E-4
other ways 9.70952E-4
other measures 9.67867E-4
large body 9.60483E-4
large smorgasbord 9.60483E-4
algorithm 9.55678E-4
model 9.5497E-4
sample weights 9.52014E-4
function 9.49902E-4
search space 9.41327E-4
foreign sentence 9.293610000000001E-4
cle score 9.28912E-4
next iteration 9.12295E-4
weak ranker 8.994890000000001E-4
weighted sample 8.90388E-4
maximum value 8.87518E-4
preliminary results 8.870079999999999E-4
future work 8.83663E-4
oracle hypothesis 8.78181E-4
optimization 8.66732E-4
search step 8.657179999999999E-4
this performance 8.63724E-4
performance gap 8.63545E-4
inferior results 8.62009E-4
local optimum 8.56495E-4
weighted combination 8.53154E-4
ture weights 8.467150000000001E-4
dim output 8.424369999999999E-4
much work 8.39993E-4
overall ranker 8.27238E-4
