Associations Between Generative AI–Based Pronunciation Feedback and Willingness to Communicate in English: The Mediating Role of English Pronunciation Self-Efficacy

Synthesis: Lu et al. (2026) examined, through the lens of Social Cognitive Theory, whether Chinese university EFL learners' perceptions of Generative AI-based pronunciation feedback relate to their willingness to communicate (WTC) in English, with English pronunciation Self-Efficacy as a hypothesized mediator. Using a cross-sectional survey of 1,701 learners, covariance-based structural equation modeling, and bias-corrected bootstrapping, they found that positive perceptions of GenAI pronunciation feedback were significantly associated with greater WTC in English, and that pronunciation self-efficacy partially mediated this relationship — the indirect effect accounted for 69.9% of the total effect while the direct effect remained significant.

Key Findings

• Positive perceptions of generative-AI-based pronunciation feedback were significantly associated with greater willingness to communicate in English among Chinese EFL learners.
• English pronunciation self-efficacy partially mediated the association: the indirect effect accounted for 69.9% of the total effect, while the direct effect of GenAI feedback perception on WTC remained significant.
• The results suggest pronunciation self-efficacy is an important but not exclusive mechanism — GenAI feedback may also relate to WTC through other routes such as reduced communication anxiety and enhanced psychological safety.
• The study conceptualizes GenAI feedback as a technology-mediated environmental influence and pronunciation self-efficacy as a personal cognitive factor within Bandura's Social Cognitive Theory, extending the framework to the Chinese EFL context and to intelligent feedback technologies.

Study Design & Method

The authors used a cross-sectional survey design with a convenience sample of 1,701 Chinese university EFL learners. Covariance-based structural equation modeling (CB-SEM) and bias-corrected bootstrapping were used to test the hypothesized relationships and the mediating effect of pronunciation self-efficacy on the perception-of-GenAI-feedback → WTC path. The analysis distinguished the indirect effect (through self-efficacy) from the direct effect of feedback perception on willingness to communicate, and quantified the indirect effect's share of the total effect.

What this means for practice

• Instructors. Add GenAI pronunciation feedback as a low-stakes supplement to classroom speaking practice: positive perceptions of that feedback were associated with greater willingness to communicate in English (total β = 0.574).
• Instructors. Use the tools to build confidence, not only accuracy: the path through English pronunciation Self-Efficacy carried about 69.9% of the total effect (indirect β = 0.401) against a modest direct effect (β = 0.173), and self-efficacy strongly predicted willingness to communicate (β = 0.661).
• Designers. Design feedback that encourages as well as corrects — learners' perceptions of the feedback (β = 0.607 on self-efficacy) rather than its measured accuracy drove the association — so favor encouraging language, actionable guidance, and adaptation to proficiency.
• Instructors. Keep teacher and peer feedback in place: self-efficacy mediated only part of the association, so routes such as reduced communication anxiety and psychological safety still matter alongside the technology.
• Instructors. Target the learners who avoid speaking, since private, immediate, and repeatable practice suits students whose pronunciation anxiety keeps them silent in class.

Limitations

• The cross-sectional design precludes causal inference.
• The convenience sample of Chinese university EFL learners bounds generalizability to other populations and contexts.
• All constructs (perceptions of feedback, pronunciation self-efficacy, WTC) are self-report measures collected at a single time point, so the mediating role of self-efficacy is inferred from covariance rather than manipulated.
• The study focuses on perceptions of GenAI feedback rather than objective feedback quality or its actual behavioral effects on speaking performance.

Citation

Lu, Y., Yang, Y., Cui, T., Yang, Z., Cai, Y., & Jing, B. (2026). Associations between generative AI–based pronunciation feedback and willingness to communicate in English: The mediating role of English pronunciation self-efficacy.