Artificial Intelligence in Second Language Learning: A Critical Systematized Reassessment of Empirical Research, 2023–2026

Abstract

Language practice and feedback can be increased with the help of artificial intelligence, but enhanced assisted performance does not necessarily prove second language acquisition. The current syntheses focus less on what each result that has been mentioned can be in support of. The review investigated the functions and contexts of various artificial intelligences that a re-evaluated corpus represents, the findings provided by the reports on assisted performance, independent performance and retention, the relation between affective outcomes and measured performance and the design and reporting features that limit interpretation. On 10 September 2026, a frame of 40 DOI-linked published records, published since 2023, was reassessed by targeted publisher and repository retrieval. Outcomes were coded as assisted, independent, retained, or transferred, and reports were analyzed at the domain level. PRISMA 2020 and PRISMA-S were used to inform reporting. Twenty-two reports were evaluated, 20 were retained, two were discarded, and 18 could not be located. It was predominantly writing (nine reports) and speaking or pronunciation (eight). Artificial intelligence was not necessarily compared more favourably with active human feedback; multiple interventions were co-modulated with other support, and three reports incorporated a late assessment. In one assessment-condition study, anxiety levels were lower but had no higher scores. It was not a reproducible database search; only part of the retrieval was realized, and all the stages were completed by one reviewer. The material in the re-evaluated corpus affirms the organized, instructor-assisted applications of synthetic intelligence to training and assessment, although not recommendations of lasting independent scholarly learning. Matched comparators and delayed, no-assistance assessment should be the priority of primary research.