Using high-level linguistic knowledge for Chinese speech recognition

A linguistic processor using syntactic, semantic and vocabulary knowledge as constraints to improve the performance of a Chinese speech recognition system is described. The processor can accept not only sentences but also phrases and words as speech input. Some characteristics of Chinese are taken into account and a knowledge representation framework based on a case frame is developed. Affirmative, negative, interrogative and elliptical sentences, etc., can be represented easily in this framework. The parsing algorithm, having no direct relation to tasks, depends only on the knowledge-representation form. Consequently, it is convenient to change the task-domain by the aid of knowledge acquisition tools. The processor can also distinguish Chinese homonymic characters