Proposals and Methods for Foreign Language Learning Using Machine Translation and Large Language Model

Abstract

In this paper, we propose a new learning model that utilizes machine translation and large language models. While English education has traditionally been conducted through the relationship between English teachers and learners, replacing English teachers with machine translation and large language models may offer the potential to provide an equally or even more efficient and high-quality learning environment. The authors have developed a browser-based service to experience this educational environment for Japanese. To experience a new learning model that is high quality and efficient, we have implemented DeepL, a machine translation service that can translate with high accuracy, and ChatGPT, which uses a large language model that can generate natural sentences and adapt to a variety of tasks interactively. By combining these advanced services, it is now possible to provide explanations of the English translations and to evaluate the essays. This newly developed service is currently being experimentally used in English classes at a Japanese university. Interviews with users who used it revealed that they were easily exposed to English above their level. In other words, the results suggest that this proposed model can provide a better environment for English utilization than teachers. The developed service is available to anyone at the following URL. Transable: https://transable.net