Selective Effects of an AI-Based Speaking Tool on Korean EFL University Students’ Speaking Proficiency and Self-Efficacy

다국어 초록 (Multilingual Abstract)

This study examined the selective effects of an AI-based speaking tool on Korean EFL university students’ speaking proficiency and self-efficacy using a quasi-experimental mixed-methods design. Eighty-one intermediate-level learners participated in a 10-week intervention (experimental group: n = 41; control group: n = 40). The experimental group used Speak, a mobile application integrating automatic speech recognition (ASR) and AI-mediated speaking interaction, while the control group completed workbook-based speaking preparation. Speaking proficiency was assessed through task-based speaking tests measuring fluency, grammatical accuracy, lexical complexity, and pronunciation, while self-efficacy was measured using an adapted 5-point Likert-scale questionnaire. Reflective journals and semi-structured interviews were also analyzed to explain learners’ perceptions and the divergence between fluency gains and non-significant affective outcomes. Quantitative results showed significant improvement only in fluency, with no significant gains in grammatical accuracy, lexical complexity, pronunciation, or self-efficacy. Qualitative findings indicated that repeated speaking practice and reduced anxiety supported fluency development, whereas limited metalinguistic feedback and reduced authenticity of interaction constrained broader linguistic and affective gains. The findings suggest that AI-based speaking tools are effective for enhancing fluency through repeated oral rehearsal but provide limited support for interaction-dependent speaking development and efficacy construction. The study highlights the importance of integrating AI-mediated speaking practice with teacher feedback and communicative interaction.