Leveraging Reinforcement Learning in Intelligent Tutoring Systems for Real-Time Feedback in English Grammar Practice

English grammar learning can be challenging due to a lack of real-time, personalized feedback, which is often missing in traditional tutoring methods. This paper proposes a Reinforcement Learning (RL)-based Intelligent Tutoring System (ITS) using a Deep Q-Network (DQN) to provide real-time, adaptive feedback to learners during grammar exercises. The system represents the interaction between the learner and the tutor as a Markov Decision Process (MDP), where each learner's grammatical response is treated as a state, and the system's feedback choices are actions aimed at optimizing learning outcomes. The DQN continuously improves feedback strategies by learning from past interactions and updating its predictions to provide more accurate and personalized feedback. Through a comparative analysis with a rule-based tutoring system (RBT), the results show significant improvements in feedback timeliness, relevance, and learner engagement. Personalized feedback based on learner proficiency resulted in increased engagement and overall satisfaction, highlighting the benefits of integrating RL in tutoring systems. The DQN-based ITS shows a significantly shorter average response time of 2.1 seconds compared to 3.5 seconds for the RBT system. The reduced response time indicates that the DQN-based system is more efficient at processing learner inputs and providing real-time feedback. This study demonstrates that the DQN-based ITS offers a more efficient and adaptive learning experience, outperforming traditional methods in terms of accuracy, relevance, and personalization of feedback.