Chinese Intelligent Input Method

摘要

This paper proposes a new Chinese intelligent input method based on Markov language model. The kernel engine is composed of three stages: word lattice formation, linguistic decoding, and cache self-learning. Different candidate space search algorithms adopted in linguistic decoding are discussed. Samples containing millions of characters are used to test the kernel engine, and 95.50% (with tone) and 88.61% (without tone) syllable-text conversion accuracy of open test is achieved. The syllable-to-text conversion engine addressed in this article has been applied in several commercial products including Microsoft Intelligent Input Method Editor (IME) -MS PY upon Windows and BOPOMOFO upon Macintosh. This technique can also be applied in speech recognition and Chinese character recognition system.