Why teachers marginalize pronunciation instruction? A new framework to predict and explain EFL teachers’ instructional practice using a machine learning pipeline

Pronunciation is central to oral proficiency because it determines intelligibility and interactional efficiency, yet it remains structurally marginalized in many foreign language classrooms. This study introduces a new predictive-explainable framework for pronunciation pedagogy research, reframing teachers’ pronunciation teaching not as a set of isolated correlates but as an ecologically determined outcome that can be predicted from interacting constraints and explained in an auditable way. Data were collected from 221 in-service Chinese teachers of English as a Foreign Language via a structured questionnaire measuring self-reported pronunciation-teaching use and 20 cognition/context items. An exploratory factor analysis supported a four-factor structure (curricular/systemic, resource, ideological/perceptual, and training/self-efficacy) showing strong psychometric adequacy. Using these factor scores, we trained a machine learning regression model with explainable measures to quantify predictor contributions. The model achieved strong predictive performance, indicating that multi-domain constraints jointly capture substantial variance in pronunciation teaching. Explainability results revealed a clear importance gradient: curricular/systemic factors dominated prediction, followed by training/self-efficacy and resources, with ideological/perceptual orientations contributing least once other domains were modeled. The findings suggest that the marginalization of pronunciation instruction is primarily curriculum-driven rather than ideologically motivated. This manifests the need for curricular reform, supported by high-transfer professional development and feasible resource provisions. Finally, the explainable machine-learning outputs could be used formatively to inform policy and support teacher learning.