Hybrid AI-Augmented Assessment in University English L2 Classrooms: A Tripartite Framework for Teacher Judgement, AI Evidence, and Learner Agency

Abstract

Generative artificial intelligence (GAI) has entered English L2 classrooms faster than assessment frameworks have adapted, leaving teachers to make consequential evaluative decisions without theoretical guidance. Existing accounts document what automated evaluation does well; few specify how its use can remain compatible with the validity-argument tradition on which defensible assessment rests. This conceptual paper addresses that gap. Synthesising recent empirical evidence on AI-mediated writing assessment, the paper proposes the Hybrid AI-Augmented Assessment (HAIAA) model, which distributes evaluative work across three interdependent components. Automated Diagnostic Evaluation confines AI to rapid formative diagnosis of surface features; Human Interpretive Assessment reserves judgements of argument, rhetoric, and cultural appropriateness for instructors; Learner Reflective Verification makes learners' documented engagement with both feedback streams an assessable warrant of authorship. Operating as a recursive cycle rather than a sequence, the components protect the inference from score to construct while restoring the learner to the centre of the feedback process. Implications are drawn for classroom practitioners and curriculum designers, and the conditions under which the model would require empirical validation are specified.