<![CDATA[This article develops a human-mediated artificial intelligence (AI) feedback architecture for Arabic language learning in Islamic higher education. The study addresses two persistent problems in current AI-assisted language education: tool-centred adoption that lacks an explicit learning mechanism and conceptual ambiguity surrounding the term “deep learning.” The paper reports the conceptual and prototype-development phase of Design and Development Research (DDR); ADDIE is used only as a procedural logic for analysis, design, and development. A design-oriented synthesis of constructive alignment, formative feedback, feedback literacy, self-regulated learning, recent systematic reviews of generative AI in language education, Arabic-specific AI studies, and responsible-AI guidance was translated through a problem mechanism component traceability process. The resulting model comprises seven recursive stages: skill-specific diagnostic profiling, meaningful alignment, differentiated task pathways, AI-supported deliberate practice, human-verified feedback and revision, reflection and self-regulation, and competency assessment followed by re-looping. The architecture explicitly separates pedagogical decision rights from AI affordances, uses transparent rubric-based pathway rules, and reserves high-stakes linguistic judgments for lecturers. Five design propositions specify how diagnosis, aligned practice, feedback uptake, lecturer mediation, and reassessment are expected to interact. The contribution is therefore a reproducible instructional architecture and a staged empirical-validation agenda rather than evidence of causal effectiveness.]]>
Copyrights © 2026
