Utterance classification with discriminative language modeling

This paper investigates discriminative language modeling in a scenario with two kinds of observed errors: errors in ASR transcription and errors in utterance classification. We train joint language and class models either independently or simultaneously, under various parameter update conditions. On a large vocabulary customer service call-classification application, we show that simultaneous optimization of class, n-gram, and class/n-gram feature weights results in a significant WER reduction over a model using just n-gram features, while additionally significantly outperforming a deployed baseline in classification error rate. A range of parameter estimation approaches, based on either the perceptron algorithm or conditional log-linear models, for various feature sets are presented and evaluated. The resulting models are encoded as weighted finite-state automata, and are used by intersecting the model with word lattices.