#9985We use a maximum likelihood criterion to train a log-linear block bigram model which uses real-valued features (e.g. alanguage model score) as well as binary features based on the block identities themselves, e.g. block bigram features.
measure(ment),3-3-P05-1069,ak
phrase re-ordering
</term>
. We use a
<term>
maximum likelihood criterion
</term>
to train a
<term>
log-linear block
#9968We use amaximum likelihood criterion to train a log-linear block bigram model which uses real-valued features (e.g. a language model score) as well as binary features based on the block identities themselves, e.g. block bigram features.
model,1-2-P05-1069,ak
machine translation ( SMT )
</term>
. The
<term>
model
</term>
predicts
<term>
blocks with orientation
#9954Themodel predicts blocks with orientation to handle local phrase re-ordering.
model,12-1-P05-1069,ak
novel
<term>
training method
</term>
for a
<term>
localized phrase-based prediction model
</term>
for
<term>
statistical machine translation
#9941In this paper, we present a novel training method for alocalized phrase-based prediction model for statistical machine translation (SMT).
model,9-3-P05-1069,ak
likelihood criterion
</term>
to train a
<term>
log-linear block bigram model
</term>
which uses
<term>
real-valued features
#9974We use a maximum likelihood criterion to train alog-linear block bigram model which uses real-valued features (e.g. a language model score) as well as binary features based on the block identities themselves, e.g. block bigram features.
other,10-5-P05-1069,ak
obtains a 18.6 % improvement over the
<term>
baseline
</term>
on a standard
<term>
Arabic-English
#10026The best system obtains a 18.6% improvement over thebaseline on a standard Arabic-English translation task.
other,14-5-P05-1069,ak
<term>
baseline
</term>
on a standard
<term>
Arabic-English translation task
</term>
. Despite much recent progress on
#10030The best system obtains a 18.6% improvement over the baseline on a standardArabic-English translation task.
other,15-3-P05-1069,ak
block bigram model
</term>
which uses
<term>
real-valued features
</term>
( e.g. a
<term>
language model score
#9980We use a maximum likelihood criterion to train a log-linear block bigram model which usesreal-valued features (e.g. a language model score) as well as binary features based on the block identities themselves, e.g. block bigram features.
other,27-3-P05-1069,ak
language model score
</term>
) as well as
<term>
binary features
</term>
based on the
<term>
block identities
#9992We use a maximum likelihood criterion to train a log-linear block bigram model which uses real-valued features (e.g. a language model score) as well asbinary features based on the block identities themselves, e.g. block bigram features.
other,3-2-P05-1069,ak
</term>
. The
<term>
model
</term>
predicts
<term>
blocks with orientation
</term>
to handle
<term>
local phrase re-ordering
#9956The model predictsblocks with orientation to handle local phrase re-ordering.
other,32-3-P05-1069,ak
binary features
</term>
based on the
<term>
block identities
</term>
themselves , e.g.
<term>
block bigram
#9997We use a maximum likelihood criterion to train a log-linear block bigram model which uses real-valued features (e.g. a language model score) as well as binary features based on theblock identities themselves, e.g. block bigram features.
other,37-3-P05-1069,ak
identities
</term>
themselves , e.g.
<term>
block bigram features
</term>
. Our
<term>
training algorithm
</term>
#10002We use a maximum likelihood criterion to train a log-linear block bigram model which uses real-valued features (e.g. a language model score) as well as binary features based on the block identities themselves, e.g.block bigram features.
other,8-4-P05-1069,ak
</term>
can easily handle millions of
<term>
features
</term>
. The best system obtains a 18.6
#10014Our training algorithm can easily handle millions offeatures.
tech,1-4-P05-1069,ak
<term>
block bigram features
</term>
. Our
<term>
training algorithm
</term>
can easily handle millions of
<term>
#10007Ourtraining algorithm can easily handle millions of features.
tech,17-1-P05-1069,ak
phrase-based prediction model
</term>
for
<term>
statistical machine translation ( SMT )
</term>
. The
<term>
model
</term>
predicts
<term>
#9946In this paper, we present a novel training method for a localized phrase-based prediction model forstatistical machine translation ( SMT ).
tech,8-1-P05-1069,ak
In this paper , we present a novel
<term>
training method
</term>
for a
<term>
localized phrase-based
#9937In this paper, we present a noveltraining method for a localized phrase-based prediction model for statistical machine translation (SMT).
tech,8-2-P05-1069,ak
blocks with orientation
</term>
to handle
<term>
local phrase re-ordering
</term>
. We use a
<term>
maximum likelihood
#9961The model predicts blocks with orientation to handlelocal phrase re-ordering.