Working Memory in Technology-Enhanced Language Learning: A Systematic Review from Interactive to AI-Mediated Contexts

Abstract

This systematic review conducts a historical–comparative analysis of working memory (WM) across two eras of language learning technology. It synthesizes 31 empirical studies, including 27 from the pre-AI "Interactive Era" (2010–2024) and 4 from the emerging "AI-Mediated Era" (2024–2025), supplemented by 10 contextual systematic reviews and theoretical papers. Traditional interactive studies show that CALL, online platforms, and multimedia environments provide multimodal support, adaptive feedback, collaboration, and flexible pacing, yet frequently induce cognitive overload, distractions, and unequal outcomes linked to individual WM capacity, which is treated as a fixed learner constraint. The AI-mediated cluster reveals a qualitative shift: AI-assisted writing reduces lower-level authorial encoding demands while increasing central-executive demands for critical evaluation, prompt management, and integrative synthesis; biometric-adaptive reading systems preemptively regulate cognitive load while enhancing comprehension; and AI-orchestrated VR-AR vocabulary instruction yields large gains only within empirically bounded multimodal channel limits. AI-mediated data-driven learning further offloads corpus search, freeing WM resources for pattern noticing and internalization. Across these strands, AI appears capable of compensating for less mature or lower WM capacity by dynamically regulating task demands. However, direct WM assessment is virtually absent from AI intervention studies, which rely instead on cognitive load scales and process indicators. This measurement gap limits causal inference about whether AI primarily reduces task demands, improves functional WM utilization, or strengthens WM capacity itself. Guided by three research questions addressing (a) design guidelines for WM-sensitive interactive instruction, (b) WM × AI affordances interactions, and (c) boundary conditions and unintended consequences of AI-mediated support, the review synthesizes these patterns and calls for future AI-mediated research to incorporate validated WM measures, implement aptitude–treatment interaction designs, and establish evidence-based boundaries for multimodal adaptivity in diverse EFL and ESL contexts worldwide.