Tim Peek, Sudeep Gandhe, David Maxwell Chickering: “Rapidly Deploying Grammar-Based Speech Applications with Active Learning and Back-off Grammars”

June 20, 2008 | Columbus, OH

Speaker: Tim Peek, Sudeep Gandhe, David Maxwell Chickering
Host: 9th SIGdial Workshop on Discourse and Dialogue

Grammar-based approaches to spoken lan-guage understanding are utilized to a great ex-tent in industry, particularly when developers are confronted with data sparsity. In order to ensure wide grammar coverage, developers typically modify their grammars in an itera-tive process of deploying the application, col-lecting and transcribing user utterances, and adjusting the grammar. In this paper, we ex-plore enhancing this iterative process by leve-raging active learning with back-off grammars. Because the back-off grammars expand coverage of user utterances, develop-ers have a safety net for deploying applica-tions earlier. Furthermore, the statistics related to the back-off can be used for active learning, thus reducing the effort and cost of data tran-scription. In experiments conducted on a commercially deployed application, the ap-proach achieved levels of semantic accuracy comparable to transcribing all failed utterances with 87% less transcriptions.