An Integrated Error Detection System For POS-tags In Korean
Duhyeon Jin · Journal of Physics Conference Series · 2019
A POS-tagged corpus is valuable data for NLP tasks. However, a large amount of POS-tagged corpus contains errors and errors debase the quality of corpus. To deal with this problem, we adopt Loftsson(2009)'s approach for POS-tagging error detection with the consideration of linguistic features of Korean. We introduce an integration of two methods to detect errors. One is variation n-gram method and another is multiple taggers' disagreement method. Then, we reconsider 'Sejong tagset' that is used in Sejong corpus which is a representative Korean corpus. By measuring error detection accuracy in POS-tagging result per each tagset candidates with our system, we find that combination of two methods can enhance the accuracy of error detection.