ChatGPT for language learning: assessing teacher candidates’ skills and perceptions using the Technology Acceptance Model (TAM)

Artificial Intelligence (AI) is projected to have a significant impact on education and language learning in the forthcoming years. Large Language Models (LLMs) such as ChatGPT offer language skill improvement and personalized feedback, but concerns exist over ethics, reliability, and over-reliance on technology. This mixed-methods study, based on convenience sampling (n = 58), examines EFL teacher candidates' critical skills and perceptions of using ChatGPT for language learning. Participants completed four collaborative writing tasks: text production and text refinement in Spanish (L1) and English (L2). Then, they compared and evaluated the results included in an anonymised document alongside ChatGPT-generated versions. Text analysis including SC (Sentence Count), ASL (Average Sentence Length), and VOCD (Vocabulary-Size-Dependent), was used to compare all of the written samples (human and machine). Statistical data from pre–post-delayed surveys, including a section on the Technology Acceptance Model (TAM), and focus group interviews were analyzed. The study employed ChatGPT-4 for generating the machine-written samples. The results showed no relationship between language nativeness (L1, L2) and participants' critical skills in distinguishing human versus machine-generated output. However, a bias was observed in the misattribution of machine-generated texts based on different parameters (SC and ASL). The findings indicate positive perceptions regarding Perceived Usefulness (PU) and Behavioural Intention (BI) but concerns on accuracy and reliability were raised. The originality of this research lies in its comprehensive examination of EFL teacher candidates' critical skills and perceptions of ChatGPT, shedding light on the implications and potential of using AI technology in language learning and teacher education programmes.