A Korean Part-of-Speech Tagging System for Minimizing Human Intervention and Maintaining Tagging Consistency

摘要

A large POS(part-of-speech) tagged corpus plays an important role in many areas of natural language processing. Therefore, a POS tagged corpus should be made with high accuracy and consistency. In order to construct a tagged corpus with high accuracy, human intervention should be required. However, a large amount of cost is needed and it is not easy to maintain consistency between the tagging results when human intervention occurs. In this paper, we propose an efficient system which can tag a corpus by using lexical rules and a stochastic POS tagger. Lexical rules are acquired manually to minimize human efforts and improve tagging accuracy of automatic tagger. And also, by using lexical rules along with a stochastic tagger, we can maintain the consistency between the tagging results. Experimental results show that we can reduce human intervention, and the accuracy of an automatic tagger can be improved continuously as the more lexical rules are acquired.