Over the last few years, there is an increasing recognition of the flaws of the existing automated grammar correction tools used in language classes. Such results as poor accuracy in identifying errors and providing helpful feedback are normal consequences of the failure of the current methods to understand the context and work with the complicated form of the Korean language. To surmount these issues, we suggest a framework of Bidirectional Long Short-Term Memory (BiLSTM) that takes into account dependencies among the context in both forward and backward directions to be able to recognize grammatrical structures and errors more accurately. This framework enables the recognition of grammar rules and errors better than the earlier models of grammar because it examines the context preceding and following a word, as opposed to the previous models, which only examine a single direction. Besides identifying errors, this innovative approach offers valuable feedback and easy to understand explanations to enhance the performance of learning. Evaluations on a Korean learner corpus demonstrate that the BiLSTM model significantly outperforms traditional baselines. The proposed method gradually improves accuracy by 89%, precision by 88%, recall by 85%, F1 score by 86.5%, error type coverage by 86%, a feedback relevance score of 4.4/5" response time by 170 ms. This framework's integration with AI tutoring systems and CAT tools for language acquisition resulted in an impressive 32% improvement in student performance.
Suping Zhan (Fri,) studied this question.