오류 분석을 통한 한글 맞춤법 검사기의 정확성 검토와 제언

초록

This study selected and tested a total of 90 items consisting of 30 items for each category, including spelling, vocabulary, and sentence dimensions, to review the accuracy of 4 types of Korean orthography checkers. In addition, the accuracy of the Korean orthography checkers was examined by analyzing the errors that appeared in the test results. Out of the 30 spelling items, the checker Ⓐ corrected 25 items, the checker Ⓑ corrected 28 items, the checker Ⓒ corrected 23 items, and the checker Ⓓ corrected 21 items. Among the 30 vocabulary items, Ⓐ corrected 16 items, Ⓑ corrected 19 items, Ⓒ corrected 17 items, and Ⓓ corrected 15 items. In the sentence category, Ⓐ corrected 4 items, Ⓑ corrected 5 items, Ⓒ corrected 3 items, and Ⓓ corrected 4 items out of the 30 items. Based on the results, it could be confirmed that the correction rate was highest in the spelling category, but they were less accurate in the vocabulary category; they were significantly less accurate in the sentence category. In other words, the current system shows high performance in correcting simple notation errors, but it was confirmed that there is a problem in that the other errors cannot be corrected in many items because they fundamentally utilize rule-based correction and statistical correction. In order to solve this problem, it is necessary to apply machine learning techniques such as deep learning, which have been used recently in natural language processing fields such as automatic translation and chat in the field of artificial intelligence.

키워드