A lightweight transformer based system for real time grammatical error correction on mobile devices

For better precision and efficiency in correcting grammatical errors on mobile platforms, this research proposes a new Transformer-based grammatical error correction model using efficient techniques for token-level detection and correction. Following this new and efficient framework for grammatical correction, this research proposes a new mobile-friendly grammar correction system. Furthermore, this research proposes using a hybrid deployment strategy using edge and cloud computing for better efficiency and reduced computational costs. Regarding the performance, this research achieves a Precision of 90.8%, Recall of 82.4%, and F0.5 score of 88.9%, which are significantly better than other baseline options like using BERT, T5, GPT, and rule-based models. In addition, regarding efficiency, this research achieves a reduction in latency to 35 ms and keeps the model size small at just 95 MB, making it highly efficient for mobile platforms. Furthermore, user satisfaction tests show high satisfaction rates for correction quality, response time, and overall satisfaction. The results demonstrate that the proposed grammar error correction system offers high accuracy in correcting grammar errors and real-time responsiveness, thus offering a promising solution for mobile language learning and smart writing support. The system offers a framework for deploying efficient natural language processing systems in real-world applications.