HMM 품사 태깅에서 신뢰성 척도와 규칙을 이용한 오류 수정

Abstract

Tagging system is significantly effected by corpus in statistical part-of-speech tagging. Since tagged corpus is made by many people together and their linguistic knowledge is different from each other, it is hard to guarantee coherency. Therefore, tagged corpus always has some errors. In this paper, we propose a measure to detect conditions to make error, and find potential tagging error by using the measure, and correct the potential tagging error by predefined rule. Experimental results showed that the precision of tagging system using our method is improved more than the precision of pure HMM tagging system.