Similarity Based Language Model Construction for Voice Activated Open-Domain Question Answering
István Varga, Kiyonori Ohtake, Kentaro Torisawa, Stijn De Saeger, Teruhisa Misu, Shigeki Matsuda, Jun’ichi Kazama · 2011
This paper describes a novel method of constructing a language model for speech recognition of inputs with a particular style, using a large-scale Web archive. Our target is an open domain voice-activated QA system and our speech recognition module must recognize relatively short, domain independent questions. The central issue is how to prepare a large scale training corpus with low cost, and we tackled this problem by combining an existing domain adaptation method and distributional word similarity. From 500 seed sentences and 600 million Web pages we constructed a language model covering 413,000 words. We achieved an average improvement of 3.25 points in word error rate over a baseline model constructed from randomly sampled Web sentences. 1