Fig. 6From: Context-dependent factored language modelsOracle error rate on the BNSI evaluation set with trigram models and different vocabulary sizesBack to article page