Vahid Aryadoust

Spoken Dialogue Systems for Language Learning and Assessment: Opportunities, Evidence, and Challenges

Recent advances in artificial intelligence have renewed interest in the use of spoken dialogue systems (SDS) for language learning and assessment. This keynote reviews the evolution of SDS in language education, from early rule-based systems to recent architectures powered by large language models (LLMs). While earlier systems were limited in flexibility and interactional range, LLM-based dialogue systems can now support more natural and adaptive spoken interaction.

The presentation examines how these systems are being applied in language education, with particular attention to their use in formative speaking assessment. Research findings suggest that well-designed SDS can provide useful feedback on learner performance and generate scores that reach reliability levels comparable to those of trained human raters. These capabilities create new opportunities for scalable speaking practice and assessment. At the same time, the keynote addresses key limitations and ethical considerations, including issues of speech recognition accuracy and validity of automated scoring. The talk also introduces emerging platforms that allow educators and researchers to develop SDS with relatively limited technical expertise.

Overall, the keynote argues that, despite current limitations, spoken dialogue systems offer promising tools for supporting speaking development and formative assessment in future-oriented English language teaching.