N13-1002 trained on a development set using MER training to maximize the BLEU score of
N07-1063 calculate P ( D ) are trained using MER training ( Och , 2003 ) on development
W05-0836 also report MAP scores using the MER training described above to determine
D09-1008 However , due to the limitation of MER training , only part of the feature space
W05-0836 each decision rule . The MAP - MER training was performed to evaluate the
P09-1103 tree sequence pairs . For the MER training , we modify Koehn 's version
P07-1059 Och , 2003 ) . Training data for MER training were taken from multiple manual
P08-1115 the interpolation weights using MER training ( Och , 2003 ) . Evaluation was
P09-1020 - n word alignments . For the MER training ( Och , 2003 ) , Koehn 's MER
D10-1043 m-to-n word alignments . For the MER training ( Och , 2003 ) , Koehn 's MER
N07-1063 either approach is effective for MER training . 6.2 Results Figure 3 and Figure
W05-0836 highest Score , just as in the MER training process . The exact method of
P08-1064 m-to-n word alignments . For the MER training ( Och , 2003 ) , we modified
D09-1073 Kenser and Ney , 1995 ) . For the MER training ( Och , 2003 ) , we modify Koehn
D08-1060 weights optimized in standard MER training . The combinatorial effects of
N07-1063 which have been trained using MER training . The parameters used for these
N07-1063 varied pruning parameters run MER training and still generate parameters
W05-0836 algorithm for MAP . Note that the MER training approach can not be performed
D11-1044 understand what our model learns during MER training , we computed the feature vectors
P12-2062 straightforward . At each iteration of MER training , we run the parser and decoder
hide detail