Enhancing a Korean Part-of-Speech Tagger Based on a Maximum Entropy Model

The most core task for Korean text processing is to recognize the lexical morphemes in sentences and to determine their part of speeches. This task is called the part of speech tagging. We present several effective schemes to enhance Korean part-of-speech tagging systems that are based on a Maximum Entropy model. We employ two levels of tags, the inner and outer tags. A probability of a morpheme sequence is computed to augment the probability of the Maximum Entropy model. Special feature functions are employed to exploit co-occurrence of multiple lexical items, which seems effective for lexical ambiguity resolution. Experimental results demonstrate usefulness of these schemes.