Lecturer (Faculty of Informatics), corpus-informed evaluation of AI-generated EFL listening materials
Designed corpus-based comparisons between human-authored EFL listening texts and outputs from LLMs including ChatGPT, Gemini, and Claude. Evaluated linguistic features such as discourse markers, formulaic n-grams, lexical range, and conversational authenticity to assess how model-generated language reproduces authentic spoken discourse. Used findings to inform educational design and model evaluation considerations for generative AI in language learning contexts. • Human-authored vs. LLM-generated listening-text feature comparison • Authenticity and pedagogical usefulness evaluation • Linguistic feature annotation/assessment guidance for evaluation criteria • Research outputs informing EFL materials development and responsible AI use