P10-2056 |
binary ) feature functions . During
|
ME training
|
, the optimal weights Ai corresponding
|
W06-2601 |
leaving-one-out method into the standard
|
ME training
|
algorithm . Experimental results
|
P10-2002 |
challenge might exist when running the
|
ME training
|
toolkit over a big size of training
|
W14-1710 |
maximum-likelihood parameter estimate in the
|
ME training
|
. Therefore a feature selection
|
P10-2002 |
3.2 Context-based features for
|
ME training
|
ME approach has the merit of
|
J13-2001 |
for log-linear combination or
|
ME training
|
without derivation . In contrast
|
P01-1027 |
the 348 models obtained with the
|
ME training
|
. For an hypothesis sentenceand
|
W03-1020 |
parameters in the conditional
|
ME training
|
. Specifically , we use array
|
W04-3248 |
穆斯塔菲兹拉赫曼 " correctly . Since in
|
ME training
|
we use iterative bootstrapping
|
P10-2002 |
Petrov and Klein , 2007 ) . The
|
ME training
|
toolkit , developed by ( Zhang
|
P06-2028 |
of words may help , and use the
|
ME training
|
process to weed out the irrelevant
|
P02-1021 |
well as the cutoff used during
|
ME training
|
. It will also be necessary to
|
N06-1026 |
To express tree structure for
|
ME training
|
, we extract path information
|
N04-1034 |
each English sentence we keep as
|
ME training
|
instances its Arabic equivalent
|
W13-3610 |
Training data refinement . •
|
ME training
|
. The test step includes three
|
N04-1034 |
Depending on the domain of the
|
ME training
|
corpus and the size of the filter
|
S01-1032 |
The set of features defined for
|
ME training
|
is described below and it is
|
W06-2601 |
log-probability distributions .
|
ME training
|
with the so-obtained real-valued
|