Performance of Generative AI Models in Three Teaching Roles: Task-Based Analysis with Prepositional Regime Complements in ELE

Abstract

Abstract: This study analyzes the performance of six generative artificial intelligence (AI) models in teaching tasks related to prepositional regime complements (CRP) in the context of Spanish as a foreign language (ELE). The focus is on Chinese university students. Based on a didactic corpus of 28 items, an experimental evaluation was designed involving three task types—item resolution, metalinguistic explanation, and oral activity design—each simulating a different instructional role. The same items were submitted to all models, varying only the instructions, in order to assess their ability to adapt to distinct educational demands. The results show that all models completed the tasks as instructed, though with notable differences. Overall accuracy did not reach 100%, and performance varied according to task type: errors were more frequent in open-ended production than in closed-choice items. Some models also showed limitations in interpreting item format or instructional intent. The study provides empirical evidence of the current capabilities of these models across different levels of pedagogical intervention and proposes a functional classification for their integration into ELE classrooms.