Young Sin Lee, Young Ja Park, Man Suk Song

The KIPS Transactions:PartB , Vol. 8, No. 1, pp. 98-104, Feb. 2001
10.3745/KIPSTB.2001.8.1.98,   PDF Download:

Abstract

In this paper, we propose an algorithm for Korean spelling correction using the eojeol generation dictionary. Since the eojeol generation dictionary is searched by edit distance based on the syllabic properties of target languages, it generates candidate eojeols independently of languages and error patterns. In addition, it provides morphological analyses of generated eojeols, and all of candidates can be effectively ranked on the basis of linguistic information without re-morphological analyses. The proposed correction algorithm consists of two steps First, it computes all possible corrected stems from incorrect eojeols. Second, we use the eojeol generation dictionary and the stems to generate candidates for original eojeols. We combined the model with POS tagging and co-occurrence information to rank the recovered eojeols. In order to evaluate the performance of the automatic spelling correction, we have conducted an experiment on 3,000 eojeols and accuracy is around 93% at the word level.

Statistics
Cite this article