Detection and Correction of Accidental Semantic Errors

One of the most important characteristics of accidental errors to result from simple mistyping is that there is serious discrepancy between erroneous words and their surrounding context. To detect and correct these errors, contextual information such as n-gram or co-occurrence frequency is needed. However, despite these and other advances, detection and correction of accidental errors in agglutinative languages such as Korean is crawling yet due to relatively freely movable component. This paper presents another method to detect and correct accidental errors using abstract dependency structure to remove functional dependency relations. To capture contextual information of word from abstract dependency structure, our method uses the co-occurrence frequency for words in immediate dependency relations between content words.