J13-4009 we used the German part of the SMT training corpus . Table 19 shows the translation
D14-1174 using the standard phrase-based SMT training method . 3.1 Phrase Translation
P04-3007 way of reducing data sparsity in SMT training . Finally , evaluation of induced
D14-1014 submodular functions useful for SMT training data subset selection . By staying
D10-1061 this list to augment the existing SMT training corpus . Assuming the pool contains
J13-4009 control for the effect of reusing SMT training data . For the machine learning
J12-4004 translation quality , we design SMT training corpora to be oblivious to the
D11-1034 translation quality , we design SMT training corpora to be oblivious to the
D11-1034 completely disjoint from the LM and SMT training sets and comprises only original
P07-1059 parallel training data into an SMT training pipeline . This training procedure
J13-4009 is feasible to use translated SMT training data for the sequence labeler
D10-1104 their original counterparts for SMT training . The SMT approach on the artificial
D14-1173 caculated from the source side of SMT training corpus . The character-level
P09-1090 source language data prior to the SMT training and decoding cycles . NieBen
D10-1061 and word alignment in subsequent SMT training , these sentences provide maximum
J13-4009 we used the target side of the SMT training data . In these experiments we
N13-1036 trained the Arabic side of our SMT training data . The use of the latter
P09-1089 not from the same domain as the SMT training corpus , it is likely that paraphrases
D14-1173 performance of unsupervised WA in the SMT training procedure was measured through
P04-3006 is most suitable for filtering SMT training data : Two texts are " comparable
hide detail