Human–AI Collaborative Oral Practice in University EFL: A Quasi-Experimental Study

Abstract

Limited opportunities for oral practice, delayed feedback, and speaking anxiety can constrain oral development in university English as a Foreign Language (EFL) programme. This study evaluated a 16-week human–AI collaborative instructional model in two intact university classes, with 42 students in the intervention class and 40 in the comparison class. The model combined AI-supported, repeatable oral rehearsal and form-focused diagnostics with teacher-led task framing, intercultural interpretation, and assessment. Outcomes were examined for listening–speaking performance, intercultural communicative competence (ICC), learning confidence, and speaking anxiety. Baseline comparisons indicated no significant differences between classes. Post-intervention analyses, controlling for corresponding pretest scores, showed statistically significant between-class differences across all four outcomes at the Bonferroni-adjusted threshold. The intervention class showed greater gains in listening–speaking performance, ICC, and learning confidence and a larger reduction in speaking anxiety than the comparison class. Supplementary interviews with 24 students indicated that private repetition reduced perceived evaluative pressure, immediate pronunciation and fluency feedback supported subsequent attempts, and teacher-led discussion was central to interpreting culturally situated meanings. Because instructional condition was confounded with class membership, the findings should be interpreted as differences associated with this implementation rather than definitive causal effects. Overall, the results support further investigation of human–AI instructional arrangements that expand oral practice while retaining teachers’ interpretive and assessment roles.