Although a lot of progress has been made recently in word segmentation and POS tagging for Chinese, the output of current state-of-the-art systems is too inaccurate to allow for syntactic analysis based on it. We present an experiment in improving the output of an off-the-shelf module that performs segmentation and tagging, the tokenizer-tagger from Beijing University (PKU). Our approach is based on transformation-based learning (TBL). Unlike in other TBL-based approaches to the problem, however, both obligatory and optional transformation rules are learned, so that the final system can output multiple segmentation and POS tagging analyses for a given input. By allowing for a small amount of ambiguity in the output of the tokenizer-tagger, we achieve a very considerable improvement in accuracy. Compared to the PKU tokenizertagger, we improve segmentation F-score from 94.18% to 96.74%, tagged word F-score from 84.63% to 92.44%, segmented sentence accuracy from 47.15% to 65.06% and tagged sentence accuracy from 14.07% to 31.47%.
Forst, M.; Fang, J. TBL-improved non-deterministic segmentation and POS tagging for a Chinese parser. 12th Conference of the European Chapter of the Association for Computational Linguistics. 2009 March 30 - April 3; Athens, Greece.