A hybrid approach for correcting grammatical errors

This paper presents a hybrid approach for correcting grammatical errors in the sentences uttered by Korean learners of English. The error correction system plays an important role in GenieTutor, which is a dialogue-based English learning system designed to teach English to Korean students. During the talk with GenieTutor, grammatical error feedback and better expressions are offered to learners. We surveyed the grammatical mistakes that occurred in the English sentences uttered by Korean learners. These errors involve preposition errors, verb form errors, agreement errors, noun countability errors and determiner errors. The hybrid error correction system consists of 5 components: an error memory based correction system, a machine learning based correction system, an n-gram based correction system, an edit distance based correction system and a selector. The correction performance of each component is different depending on error types. To evaluate the hybrid system, we used a test set comprising of 858 sentences extracted from utterances by Korean learners. The test set includes not only ungrammatical sentences, but also correct sentences. We conducted various experiments and examined the effect of the hybrid approach on grammatical error correction. The experiments show promising results for correcting grammatical errors.