ENGFORTHAI+: Mitigating Morphological Omission with Static Visual Scaffolding and Adaptive Content Generation

Abstract

While recent advancements in L1-tuned Automatic Speech Recognition (ASR) have significantly improved the diagnosis of segmental pronunciation errors, L2 learners continue to exhibit persistent morphological deficits. Specifically, Thai EFL learners frequently omit inflectional markers due to the high cognitive load of real-time morpho-syntactic planning. This study introduces “Static Visual Scaffolding” (SVS), a novel intervention designed to bridge the gap between declarative knowledge and procedural speech production using dynamic, AI-generated passage sentences. Unlike traditional Intelligent Tutoring Systems that provide reactive, post-turn feedback, the SVS framework acts as a “Heads-Up Display” (HUD) for the learner. As the learner reads contextually unique AI-generated scripts, the system utilizes low-latency streaming ASR to anticipate and mitigate grammatical spoken errors. The HUD projects subtle visual cues (e.g., a spectral highlight on specific morphemes) before the learner articulates the target verb, effectively preempting habitual errors. We evaluated this approach through a randomized controlled trial involving 30 Thai university students engaging in the oral reconstruction of AI-generated passages. Results indicate that the SVS group achieved a statistically significant reduction (ηp 2 =0.41) in grammatical spoken errors events compared to a control group receiving standard post-hoc corrections. Furthermore, analysis suggests that combining static cueing with infinite, AI-driven content reduces extraneous cognitive load, allowing learners to maintain fluency while attending to form. These findings propose a paradigm shift in Computer-Assisted Language Learning (CALL): moving from error detection to anticipatory cognitive offloading during complex speech tasks.