Answers are grounded in this paper and cite their sources.
Accurate assessment of English learning outcomes increasingly requires objective, scalable, and multimodal approaches. Traditional evaluation methods relying on manual scoring or simplistic automated models fail to capture temporal learning patterns and nuanced speech and semantic characteristics. Research aims to develop an intelligent system for the quantitative evaluation of English proficiency while providing adaptive feedback to support individualized learning progression. A novel framework, termed the Narwhal-Tuned Temporal Deep Q-Network (NTDQN), integrates sequential modeling with reinforcement learning. The English Learning Outcome Dataset captures learner English proficiency across profile information, textual responses, speech features, and temporal logs. Data preprocessing involves handling missing values, tokenization, and lemmatization to ensure that inputs are normalized. Semantic information is extracted using GloVe embeddings, while prosodic features represent the dynamics of speech. The model incorporates along short-term memory (LSTM)-based temporal module for modeling learner behavior sequences, a Deep Q-Network (DQN) for selecting optimal feedback actions, and a nature-inspired Narwhal optimization algorithm for hyperparameter tuning. The system is implemented in Python using modern deep learning and reinforcement learning frameworks. The proposed NTDQN achieves an accuracy of 98.1%, precision of 97.2%, recall of 95.2%, and F1-score of 94.6% in proficiency classification. For outcome prediction, it attains a root mean square error (RMSE) of 1.08 and a mean absolute error (MAE) of 0.91, indicating high predictive reliability. The proposed approach offers a scalable and robust multimodal solution for quantitative English assessment, enabling real-time evaluation and adaptive learning support while significantly outperforming conventional models. Multimodal deep reinforcement learning enables accurate quantitative English proficiency evaluation. LSTM–DQN with narwhal optimization provides adaptive and personalized learning feedback. Proposed system achieves high classification accuracy and low prediction error.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: