Active learning for spoken language understanding

Abstract

Disclosed is a system and method of training a spoken language understanding module. Such a module may be utilized in a spoken dialog system. The method of training a spoken language understanding module comprises training acoustic and language models using a small set of transcribed data S, recognizing utterances in a set Sthat are candidates for transcription using the acoustic and language models, computing confidence scores of the utterances, selecting k utterances that have the smallest confidence scores from Sand transcribing them into a new set S, redefining Sas the union of Sand S, redefining Sas Sminus S, and returning to the step of training acoustic and language models if word accuracy has not converged.