Evaluación de la competencia léxica de los LLM: Un estudio comparativo con el corpus CAES basado en el MCER

Resumen

Large language models (LLM) have come to play a significant role in the field of education, particularly in the teaching of Spanish as a foreign language, where they are used for content generation, automatic correction and the development of conversational systems. However, it is necessary to assess the suitability of their lexical output for the proficiency levels defined by the Common European Framework of Reference. This study analyses the lexical complexity of six LLM by comparing them with the CAES corpus. To this end, texts were generated based on tasks equivalent to those performed by students, without specifying the level of difficulty. Furthermore, a lexicon annotated by levels (A1–C1) was constructed from the reference corpus. The results indicate that, although LLM reproduce trends similar to those of learners, they demonstrate a broader use of specialised vocabulary at advanced levels.