Reducing Speaking Anxiety through AI Interaction: ChatGPT-Assisted English-Speaking Practice and Language Proficiency Development in Chinese Universities

Abstract

This study aims to explore the correlation between ChatGPT-supported English-speaking practice, speaking anxiety, and students' self-perception of their English-speaking ability improvement at Chinese universities. It also examines speaking anxiety statistically mediates the relationship between ChatGPT-assisted practice and proficiency development and if perceived interaction quality moderates the relationship. This study is a sequential explanatory mixed method design. The quantitative section presents the results of a cross-sectional survey of 412 undergraduate students in six public universities in China. The analysis is conducted according to PLS-SEM reporting logic including reliability, convergent validity, discriminant validity, predictive relevance and bootstrapped path estimates. ChatGPT-assisted English-speaking practice has a positive relationship with the self-reported English-speaking proficiency development and a negative relationship with speaking anxiety. The negative correlation between speaking anxiety and proficiency development and the significant mediating path between ChatGPT-assisted practice and proficiency development suggests that speaking anxiety hinders the utility of ChatGPT to assist in proficiency development. The perceived interaction quality also moderates the practice to proficiency relationship. This paper is an original article that contributes to the existing research on ChatGPT-assisted language learning by incorporating an interaction-based perspective of ChatGPT use, an affective mechanism and a contextual moderator in the Chinese university context. Because the quantitative data are cross-sectional and self-reported, the results are interpreted as evidence of substantial relationships instead of causal relationships.