DEEP LEARNING FOR GRAMMER AND SYNTAX CORRECTION IN ENGLISH LANGUAGE TEACHING

Abstract

The improvement of communication through better grammar and syntax is highly important in both education and workplace settings. This research introduces an innovative deep learning model for correcting grammar, which also has a user-friendly and interactive web-based dashboard as part of its user interface (UI). This dashboard (built with current web platforms) allows users to submit text or audio for correcting, view the corrected output, and track performance by the number of Correct Responses (Precision), Incorrect Response of Correct Cause (Recall), Correct versus Incorrect Responses (F1-Score) and Percentage of Accurate Responses (BLEU Score) measured almost instantly while performing their task using the dashboard. The proposed deep-learning model was evaluated using both custom data and publicly available data from previously established benchmark datasets including JFLEG and CoNLL-2014. Fine-tuning the model with custom data that reflects a wide range of real-life cases provided the highest F1-score of 91.56% and BLEU-score of 85.00% compared to leading models such as T5, GPT-3 and GECToR. For benchmark datasets, the model's F1-scores on JFLEG and CoNLL-2014 were 89.05% and 90.00%, respectively, clearly demonstrating that the proposed deep learning model provides superior handling of various types of grammatical errors (ex: subject-verb agreement, verb forms & punctuation) using similar examples from different grammatical categories. Integration of the explainability features in conjunction with the dashboard promotes confidence and transparency for the end users. This study provides an effective solution that has an easy-to-use interface for providing high-quality grammar corrections in a variety of application settings and thus enhances the accuracy and intelligibility of written communication.