Limited study time constrains university EFL vocabulary learning, so efficiency should be evaluated alongside accuracy. A web-based multimodal environment was developed that uses a large language model for contextualized drills and tutoring, text-to-speech for pronunciation and listening rehearsal, and an interactive 3D mastery view for self-regulated tracking. Vocabulary knowledge is modeled as a discrete mastery state (m = 0–5), updated after each attempt, and an adaptive scheduler allocates practice across mastery strata. Learning ROI is defined as newly mastered words per hour and computed from logged study time and mastery transitions. In a three-month deployment (N = 171), learners achieved a mean ROI of 9.8 words/hour, about 60% higher than conventional estimates (5–6 words/hour); high-adherence users reached 17–21 words/hour. End-of-trial surprise review results indicated retention above 85%. For CET-4, the platform cohort obtained the highest mean score (457.66) and pass rate (74.24%) compared with Baicizhan (442.22; 64.81%) and traditional instruction (428.60; 53.70%). The results provide quantitative support for the hypothesis that multimodal personalization improves time-based vocabulary gains and their durability.
You et al. (Tue,) studied this question.