Improved lsi-based natural language call routing using speech recognition confidence scores

In most natural language call routing applications, the sole purpose of any automatic speech recognizer (ASR) is to transcribe a user´s spoken request into text, so that the user can reach their desired destination based upon the analysis of the transcribed text. Given the level of uncertainty in correctly recognizing words with an ASR, calls can be incorrectly transcribed, raising the possibility that a caller is routed to the wrong destination. To reduce the potential for errors in classification, we propose a technique for incorporating confidence scores reported by an ASR to reweigh query vectors in a latent semantic indexing (LSI) classifier. Our results show that this technique is capable of reducing the number of misrouted calls by a significant amount