This thesis presents a north korean part-of-speech tagger using the Word-Formation Transition Network and the Hidden Markov Model(HMM).The Word-Formation Transition Network is designed in north korean regulation. In a morphological analysis step, it p...
This thesis presents a north korean part-of-speech tagger using the Word-Formation Transition Network and the Hidden Markov Model(HMM).The Word-Formation Transition Network is designed in north korean regulation. In a morphological analysis step, it performs morphological analysis using the Word-Formation Transition Network and a north korean dictionary and north korean morphological allomorph and a north korean irregular verb, etc. As a result of morphological analysis, a sentence is converted to a morpheme sequence. In a part-of-speech tagging step, given morpheme sequence, it performs part-of-speech tagging based on a statistical method, that is, Hidden Markov Model. A context probability and a lexical probability of Hidden Markov Model is trained as supervised learning. A context probability uses bigram. The Size of a training corpus is about sixty thousand words that have been morphologically analyzed. This training corpus is a part of the corpus that have been constructed by Special Corpus Construction Division of The 21st Century Sejong Project. In this thesis, the number of tag-set for a part-of-speech tagging is 29.The Accuracy is 95.60% on a experimental corpus fitting to North Korean Regulation In 1988, 93.34% on a experimental corpus fitting to North Korean Regulation In 2003.