AI-BASED LANGUAGE EDUCATION PLATFORMS: A SYSTEMATIC ANALYSIS OF EDTECH TOOLS FOR ENGLISH PROFICIENCY

Abstract

This study addresses a practical problem in higher education and enterprise deployments: institutions are adopting AI-based language learning platforms without clear, quantified evidence of which features most strongly relate to measurable English proficiency. The purpose is to estimate feature–outcome relationships in authentic settings. Using a quantitative, cross-sectional, case-based design, we analyze multi-institution data from cloud-hosted, enterprise-grade platforms (writing focused, speaking focused, and integrated skills). The sample comprises linked survey, telemetry, and assessment records, with proficiency indicators mapped to CEFR bands or rubric scores. Key variables include engagement intensity, feedback specificity, feedback immediacy, adaptivity breadth, and spacing quality, with controls for baseline proficiency, prior exposure, study time, device access, and demographics. The analysis plan includes descriptive statistics, zero-order correlations, multiple linear regression with case fixed effects and heteroskedasticity-robust errors, moderation by motivation, subgroup contrasts by platform modality, and spline checks for nonlinearity. Headline findings indicate that feedback specificity shows the largest unique association with proficiency (about 0.17 SD per 1 SD increase), followed by engagement (≈0.13), spacing quality (≈0.10), feedback immediacy (≈0.09), and adaptivity breadth (≈0.08). Motivation positively moderates the engagement–proficiency link, speaking cases benefit most from fast feedback and spaced practice, and writing cases benefit most from granular, itemized feedback. Returns to weekly minutes flatten beyond the upper quintile, suggesting calibrated rather than maximal practice. Implications for policy and practice include engineering actionable feedback at scale, setting latency service targets, exposing spacing indices, and aligning dashboards with revision uptake and error convergence to guide instruction and procurement.