Abstract
Learners in higher education possess diverse goals of English language learning, ranging from academic communication to occupational language use, yet conventional classroom instruction often struggles to provide individualised practice opportunities. This article introduces an AI-driven English speaking practice programme that integrates a multimodal spoken dialogue system, automated scoring techniques, and explainable artificial intelligence (XAI). The programme is designed to identify learners’ strengths and weaknesses in lexical use and to provide paraphrased expressions generated by large language models as individualised diagnostic feedback. An experimental study with Japanese university students demonstrated that task repetition (i.e., speaking practice) alone led to short-term performance gains but failed to retain them, whereas the integration of diagnostic feedback improved students’ lexical use in a new task (i.e., transfer effects). Finally, this paper discusses the benefits and limitations of using a spoken dialogue system in university English educational settings and considers the roles of teachers in AI-mediated learning environments.
