Towards Cognitive-Friendly L2 Learning: A Transformer-GAT Fusion Model with Neural Forgetting Simulation

Recent advancements in deep learning have significantly enhanced natural language processing (NLP), with transformer-based architectures achieving state-of-the-art performance in various linguistic tasks. However, these models often struggle with long-term memory retention, particularly in second language (L2) acquisition, where forgetting is a major challenge. Traditional computer-assisted language learning (CALL) systems often fail to address these issues effectively, resulting in suboptimal long-term learning outcomes. This paper introduces a novel Transformer-GAT Fusion Model with Neural Forgetting Simulation (NFS) to enhance cognitive-friendly L2 learning. The proposed model integrates the Transformer’s self-attention mechanism for sequential language processing with Graph Attention Networks (GAT) to capture semantic relationships within a structured knowledge graph. Additionally, NFS is designed to simulate human forgetting patterns based on Ebbinghaus’ forgetting curve, dynamically adjusting learning schedules to optimize retention. The model is evaluated on multiple L2 learning tasks, including vocabulary retention, syntactic structure acquisition, and listening comprehension, using benchmark datasets such as Common European Framework of Reference for Languages (CEFR)-aligned English learning corpora and HSK Chinese proficiency exams. Experimental results demonstrate that the Transformer-GAT fusion model, combined with NFS, significantly outperforms baseline methods in long-term retention and adaptive learning effectiveness. This research highlights the potential of integrating deep learning with cognitive-inspired mechanisms for L2 acquisition, paving the way for more adaptive and personalized language learning systems.