Some improvements on maximum entropy based Chinese POS tagging

This paper explores issues related to part-of-speech tagging for Chinese language using maximum entropy technique, in which we first introduced our feature selection strategy based on incremental experiments and error-driven analysis. Then making use of the knowledge from a syntactic dictionary, we created pseudo-events for external lexicon and restricted tags of words to a specific subset, which shrinked the search space greatly. Experiments on the simplified Chinese corpus of China Peking University show that significant improvements are obtained by our approach.