The emergence of Generative Artificial Intelligence (GAI)—particularly Large Language Models (LLMs) such as ChatGPT—is transforming the educational landscape, especially in the field of foreign language instruction.
This article explores the potential of these technologies to automate the assessment of writing proficiency in Spanish as a Foreign Language (SFL), a task that is especially time-consuming at the beginning of university-level courses for Erasmus students.
The study is based on three experiments conducted using the Spanish Learner Corpus compiled by the Instituto Cervantes. The first experiment applied a zero-shot learning approach by prompting the model with level descriptors from the Instituto Cervantes’s Curriculum Plan. In the second and third experiments, the model was adjusted through fine-tuning using 90% and 80% of the corpus, respectively, with the remaining data reserved for testing and validation.
The results indicate that the fine-tuned models significantly outperform the zero-shot configuration in identifying the correct proficiency levels of learner texts.
These findings demonstrate that LLMs can be effectively employed to streamline the initial placement process in SFL courses, thus reducing the workload of instructors and improving efficiency.
The study concludes that GAI can serve as a valuable complementary tool in multilingual and multicultural educational settings, provided its use is guided by sound pedagogical principles.
---
How to Cite: Cantero Romero, M.-V., Martín-Valdivia, M.-T., Ortiz-Colón, A. M., & Jiménez-Zafra, S. M. (2026). Measuring writing skills in Spanish as a foreign language with generative artificial intelligence. RIED-Revista Iberoamericana de Educación a Distancia, 29(1), 353–379. https://doi.org/10.5944/ried.45486
