International Journal of Computational Linguistics Research Shi Chen 17 1 2026 https://doi.org/10.6025/ijclr/2026/17/1/35-53 https://www.dline.info/jcl/fulltext/v17n1/jclv17n1_3.pdf This study explores the integration of cognitive science and artificial intelligence to enhance English writing instruction and assessment. It frames writing as a complex, recursive cognitive process involving planning, generation, and revision mirroring functional mechanisms in modern AI systems, such as large language models (LLMs). The paper proposes a conceptual framework that aligns human cognitive models with AI architectures through components such as goal representation, memory (long-term and working), attentionbased context handling, and feedback driven revision. This alignment is formalized mathematically using state transition systems and probabilistic generation models. An intelligent writing system was implemented in a 12-week online doctoral course (N=22) to support idea generation and coherence development. Text coherence was measured via average degree centrality in graph based representations. Experimental results showed that the proposed feature selection method significantly outperformed LSTM across training set sizes (85-99% vs. 40-90% accuracy), particularly with limited data. Adversarial training further improved robustness. Frequent subgraph analysis enabled effective discrimination between coherent and incoherent texts, with low performing subgraphs filtered to preserve reliability. The study concludes that hybrid AI-cognitive models enhance writing quality, engagement, and efficiency while underscoring the need for ethical, explainable, and human centred AI design in educational contexts.