Work

From Cefr Descriptors to Ai-generated Feedback: Reliability, Validity, and Research Priorities for Efl Writing Assessment in Uzbekistan

Year: 2026
Type: article
Abstract: Generative artificial intelligence (GenAI) is increasingly used to support English as a Foreign Language (EFL) writing through automated comments, language correction, and rubric-based scoring. Yet the usefulness of these applications depends on whether their outputs are consistent, aligned with intended assessment criteria, and pedagogically appropriate. This critical narrative review examines selected research on large language models in writing assessment and considers implications for CEFR-aligned assessment in Uzbek higher education. The discussion is organized around three issues: rating reliability, validity of score interpretations, and the quality and use of AI-generated feedback. Existing evidence indicates that large language models can produce consistent rubric-based ratings under particular conditions, but consistency alone does not establish that a score validly represents writing proficiency or a CEFR level. Feedback may support revision, particularly when it addresses language form and textual organization, but learners and teachers still need to evaluate the relevance and accuracy of suggestions. The review identifies a context-specific research priority: comparing AI-generated judgments with trained human ratings using transparent CEFR-informed criteria and investigating how learners use the resulting feedback. Because this article is a critical narrative review rather than a systematic review or an original empirical study, its conclusions are provisional and should be followed by a documented literature search and local validation study.
Language: English
Open Access status: hybrid
OpenAlex ID: W7225905108
FWCI:
Cites: 3
Cited by:
Related to: 10