Fig. 5From: Context-dependent factored language modelsOracle error rate on the BNSI evaluation set with bigram models and different vocabulary sizesBack to article page