Governed Syntax-Informed AI Adaptation for Sustainable Academic Reading: Fidelity, Comprehension, and Near-Transfer

AI-mediated text adaptation may improve access to sustainability knowledge when linguistic support preserves disciplinary meaning and remains subject to human review. This study examined the Syntactic Accessibility for Sustainable Digital Texts framework in a within-participant mixed-methods experiment with 120 Arabic-speaking university EFL learners and six source passages. Participants read original, generically simplified, and syntax-informed AI-adapted texts; both AI conditions passed a common fidelity gate. Crossed comprehension models were supplemented by seven-day assessments, passage ratings, stimulated recall, and workflow summaries. Adjusted comprehension probabilities were 0.61, 0.69, and 0.76, respectively; near-transfer means were 6.15, 6.84, and 7.55 on a 0–12 scale. Initial syntax-informed outputs retained 62/64 technical terms and 58/61 epistemic qualifiers, compared with 54/64 and 50/61 under generic simplification, and required less downstream validation. The contrast concerns complete human-reviewed workflows, including lexical support and unequal expert repair, rather than an isolated syntactic mechanism. The findings support a corpus-bounded account of syntactic accessibility and short-term learning; feature associations and learner explanations remain exploratory. Longer-term independence, model portability, institutional transformation, and environmental benefit require further evidence.