Cognitive Load Assessment and Optimization in College Japanese Classrooms Assisted by Lightweight Multilingual Models

Abstract

With the gradual adoption of multilingual large language models in foreign language classrooms, the cognitive load that learners experience when using intelligent tools has attracted increasing attention. Taking college Japanese-major classrooms as the empirical setting, this study builds an intelligent foreign language teaching assistant system based on a lightweight distilled multilingual model, collects learners’ subjective cognitive load data through questionnaires, and evaluates the changes in cognitive load while learners use AI for Japanese language output practice. The results show that the lightweight multilingual model effectively reduces part of learners’ memory load, but problems such as inaccurate model feedback introduce additional cognitive consumption. Accordingly, improvement countermeasures are proposed from two levels: model parameter optimization and adjustment of teaching interaction modes. This study provides a reference for the development and application of multilingual intelligent teaching tools for foreign language education.