Synergistic Learning: An Architectural Blueprint for an Adaptive English Tutor using NLP and Deep Reinforcement Learning

Vergara E.E.P.; Chocce S.E.C.; Silva P.D.R. · Proceedings - 2025 5th International Conference on Information Systems and Software Technologies, ICI2ST 2025 (2025) · DOI: 10.1109/ici2st68465.2025.00026

Resumen

Personalization of language learning represents a key challenge in the development of intelligent educational systems. While English proficiency is increasingly important in academic and professional contexts, current technological solutions often present limitations both in the accurate assessment of language skills and in the dynamic adaptation of the difficulty of the proposed activities. In this paper, we present the design of a hybrid architecture that integrates two main components: (i) a multitask natural language processing (NLP) model based on RoBERTa, capable of automatically assessing the user's written proficiency in six dimensions aligned to the CEFR framework; and (ii) a deep reinforcement learning (DQN) agent combined with an ELO scoring system, oriented to recommend adaptive exercises based on the observed performance. This synergy between analytical assessment and dynamic adaptation allows not only to assign exercises with an appropriate level from the first access, but also to adjust the progression in an individualized way as the user progresses. Finally, an experimental validation plan is detailed that contributes to the exploration of new strategies for intelligent learning of English as a foreign language, laying the groundwork for future research in educational personalization.

Instituciones: Universidad Peruana de Ciencias Aplicadas