P10-1112 , packed graph encoding to the treebank training trees . We collapse the duplicate
P02-1018 patterns extracted from the Penn Treebank training corpus . The algorithm also records
J10-2004 estimate the split PCFG on the Penn Treebank training trees . After that , 50 % of
P06-2016 observation data from the Lancaster Treebank training set . The training set and testing
D11-1012 the rest of the available Penn Treebank training data for each task . For the
P01-1053 the training procedure . In a treebank training step we observe for each rule
P01-1053 pronunciation dictionary CELEX . In a treebank training step we obtain a probabilistic
E09-1033 train BitPar on the remaining Penn treebank training sentences . The average Fi parsing
P01-1053 follows . Section 2 refers to treebank training . In section 3 we introduce the
P06-1023 parser using the standard PENN treebank training and test data . The labeled bracketing
P01-1053 for the experiments is based on treebank training as well as bracketed corpora
P01-1053 , we can apply the formula of treebank training given by ( Char - ( 1.1 ) 0.1774
P02-1018 theory , but it does require a treebank training corpus from which the algorithm
H94-1051 P ( r ) . reD ( t ) Using the Treebank training corpus , P ( ~ a ) is estimated
D09-1087 the resulting parses with the treebank training data to retrain the parser .
J98-4004 representational scheme used to construct the treebank training data ? Mark Johnson PCFG Models
D09-1087 both languages in Figure 2 with treebank training , it is clear that they perform
D09-1087 60 % , 80 % , and 100 % of the treebank training data to evaluate the effect of
D15-1159 we train our model only on the treebank training set and do not use tri-training
D09-1087 a fraction of the WSJ or CTB6 treebank training data is sufficient to train a
hide detail