The more the merrier? Examining the effects of a conversational agent on EFL learners’ speaking in three conditions

The present study innovatively investigates how a generative AI conversational agent (TalkFriend) impacts EFL university students’ oral English proficiency, considering both cognitive and affective dimensions. Forty-five students were randomly assigned to individual or paired learning (with ‘Lead’ and ‘Assisting’ roles interacting with TalkFriend). Employing a multi-modal approach, we uniquely integrated EEG brainwave data with oral tests, questionnaires, and interviews. Findings revealed significant overall proficiency gains. Notably, paired learning fostered superior improvements in communicative confidence and fluency compared to individual learning, which primarily saw fluency gains. Lead learners in paired settings also exhibited markedly higher learning interest, a factor significantly correlating with their neural activity (EEG). Pronunciation accuracy appeared to develop independently. Interpreted through Vygotsky's Zone of Proximal Development (ZPD), these findings inform a proposed four-quadrant ‘emotional ZPD’ conceptual model, highlighting the crucial interplay of cognitive, affective, and social support (from both AI and peers). Our research offers critical neurocognitive and socio-interactional insights for optimizing AI tools in language education.