ing features 0.002670956
simple features 0.002560858
pairwise features 0.002545659
complex features 0.0025109
base features 0.0024687479999999998
word error 0.002457852
motivated features 0.00242456
recognition word 0.002403352
average word 0.002288354
word length 0.002265014
word pairs 0.0022240899999999997
features 0.00216651
training data 0.00215377
feature value 0.002131464
feature selection 0.002081177
feature vector 0.00207068
individual words 0.002039748
language model 0.002034984
simple feature 0.002013598
single feature 0.001966713
actual words 0.001953822
dence feature 0.001936081
corresponding feature 0.00191887
feature families 0.001918025
content words 0.001891873
coherence feature 0.001890928
tent words 0.001885679
consensus words 0.001885679
feature family 0.001876271
feature fam 0.001874106
linear model 0.001727578
ing training 0.001704076
ranking model 0.001700548
probabilistic model 0.001641408
markov model 0.0016376960000000001
words 0.00163036
network model 0.001625715
feature 0.00161925
reranking results 0.001607803
novel model 0.0016045500000000002
training size 0.00160034
data set 0.0015767210000000001
reranking models 0.001552521
training pairs 0.00155183
training list 0.001550232
training instances 0.001507738
stochastic training 0.001488059
training strategies 0.001467223
baseline system 0.001461123
reranking methods 0.001427592
linguistic data 0.001423922
translation system 0.001414471
reranking performance 0.001402685
pairwise reranking 0.0013583450000000002
model 0.00134661
reranking approach 0.0013364190000000001
baseline hypothesis 0.0013354970000000002
list reranking 0.001329798
discriminative reranking 0.0013193100000000002
test set 0.001306684
data sets 0.001290525
other information 0.0012663969999999998
language models 0.0012616989999999998
anfal data 0.001249778
ensemble reranking 0.0012457030000000001
ious reranking 0.001233923
ocr system 0.0012146589999999999
handwritten data 0.0012138890000000001
baseline wer 0.001213148
data consortium 0.001210008
typewritten data 0.001208522
training 0.00119963
semantic information 0.0011961329999999998
different sizes 0.001157742
different combinations 0.001154043
byblos system 0.001139761
overall results 0.001120666
additional information 0.001120323
hypothesis length 0.0011064360000000001
such information 0.0011001169999999998
learning method 0.0011000089999999999
system benefits 0.001094944
ocr hypothesis 0.001089033
wer scores 0.0010818400000000001
other methods 0.001071685
language modeling 0.001068229
hypothesis pairs 0.001065512
error rate 0.001037001
natural language 0.001035081
standard baseline 0.0010247260000000001
hypothesis lists 0.0010111270000000001
baseline ocr 9.97906E-4
language processing 9.91961E-4
hypothesis confi 9.86809E-4
sri language 9.86322E-4
reranking 9.79196E-4
baseline ranking 9.76123E-4
information con 9.74643E-4
each hypothesis 9.732180000000001E-4
machine translation 9.71012E-4
