Enhancing Speech Based Information Service with Natural Language Processing

Speech based information service systems are much friendlier and more accessible than keyboard type-in based one. But a key problem for speech recognizers to be used in realistic applications is its low precision. Existing efforts are made to improve its performance through using statistical language knowledge in the phase of transformation from syllables to words in a speech recognizer. This paper proposes a new way of error detection and correction using Comprehensive Information based natural language processing. A module of text post-processing is added after a Speech Recognizer. In this module, syntactic, semantic and pragmatic informationare analyzed to check the result sentence text of the Speech Recognizer. If there are some errors, the module also tries to correct them. The original test is done in the application of City Guide, which is a demo system for the National 863 project of "2008 Beijing Olympics oriented Multilingual Intelligent Information Service System". Results have shown that for Chinese language, the precision could be improved by 26%, and for English language, the improvement is about 31.9%.