Empowerment-AI-Enhanced System for Initiating and Sustaining Vocabulary Learning in EFL Learners

This study proposes an adaptive English learning support system designed to close the gap between the vocabulary that learners can recognize and understand and the vocabulary they can confidently use in practical communication, with a focus on Japanese learners of English as a Foreign Language (EFL). The system generates a personalized vocabulary dataset by analyzing learners’ performance on recognition tests based on the New General Service List (NGSL) and the New Dolch List (NDL). To assess productive vocabulary, learners engage in translation tasks. Their unconstrained and vocabulary-limited outputs—generated by a large language model (LLM) limited to their personalized vocabulary—are compared to reference translations from DeepL using BLEU scores. The results reveal a clear discrepancy between learners’ perceived vocabulary knowledge and their actual ability to apply it contextually, highlighting both the benefits and limitations of constrained expression in adaptive learning. While the system currently does not employ physical agents, it draws on the principle of task excitation, which suggests that initiating small actions can increase motivation. The system is designed to present tasks at opportune moments to encourage learners to begin studying. This motivational framework aligns with the concept of Empowerment AI, which seeks to support the initiation of learning behavior by reducing psychological resistance. By integrating motivational design with personalized, LLM-based support, the system promotes both the initiation and persistence of learning, contributing to more autonomous and sustainable vocabulary development.