Discriminative training in natural language call routing

In this paper, we show how discriminative training can be used to improve classifiers used in natural language processing, using as an example the task of natural language call routing. In natural language call routing, callers are routed to desired departments based on natural spoken responses to an open-ended “How may I direct your call?” prompt. With vector-based natural language call routing, callers can be transferred using a routing matrix that is trained based on statistics of occurrence of words and word sequences in a training corpus after morphological and stop-word filtering. New user requests are represented as feature vectors and are routed based on the cosine similarity score with the model destination vectors encoded in the routing matrix. The present paper proposes the use of discriminative training on the routing matrix to improve routing accuracy and robustness. By retraining the routing matrix, a relative error rate reduction of 13-19% was achieved. Increased robustness was demonstrated in that with 10% rejection, there was a relative error rate reduction of 40%. The proposed formulation is equally applicable to algorithms addressing a broad range of speech understanding, information retrieval, and topic identification problems.