Innovation in Language Learning

Artificial intelligence-based corrective feedback tools are increasingly embedded in second and foreign language instruction, offering learners instant responses to linguistic errors. The pedagogical value of corrective feedback, however, is widely held to depend on its calibration to learner readiness, a premise rooted in Vygotskian notions of the zone of proximal development and Krashen's input hypothesis. Whether AI-based feedback tools honour this premise, or instead apply a uniform response regardless of proficiency, remains largely unexamined in a systematic way. This paper reports a scoping review, following the PRISMA-ScR protocol, of studies published between 2019 and 2026 that investigate AI-generated corrective feedback in second or foreign language learning contexts. A structured search of Scopus, Web of Science, ERIC, and ProQuest identified studies screened against predefined inclusion and exclusion criteria. Included studies were coded for target proficiency level, feedback type following Lyster and Ranta's taxonomy, and degree of specificity. Preliminary synthesis suggests that most AI feedback tools do not differentiate specificity by proficiency level, with empirical attention concentrated on intermediate learners and beginner and advanced populations comparatively underexplored. The review maps these patterns against theoretical expectations of developmentally appropriate scaffolding, revealing a clear misalignment between design practice and pedagogical theory. The paper concludes with design implications for developers of adaptive feedback systems and a research agenda for testing proficiency-calibrated feedback algorithms in authentic classrooms.

Keywords: artificial intelligence; corrective feedback; proficiency level; scoping review; adaptive learning; second language acquisition