Toward a unified approach to statistical language modeling for Chinese

This paper presents a unified approach to Chinese statistical language modeling (SLM). Applying SLM techniques like trigram language models to Chinese is challenging because (1) there is no standard definition of words in Chinese, (2) word boundaries are not marked by spaces, and (3) there is a dearth of training data. Our unified approach automatically and consistently gathers a highquality training data set from the web, creates a high-quality lexicon, segments the training data using this lexicon, and compresses the language model, all using the maximum likelihood principle, which is consistent with the trigram model training. We show that each of the methods leads to improvements over standard SLM, and that the combined method yields the best pinyin conversion result reported.

• Type: Journal
• Year: 2002
• Where: TALIP
• Authors: Jianfeng Gao, Joshua Goodman, Mingjing Li, Kai-Fu Lee
• Keywords: Language Model, TALIP 2002, Training Data, Unified Approach