A Transformer-GAN Based Approach for Evaluating and Improving High School Students' English Translation Skills

Abstract

This research presents a Transformer-GAN-based model to assess and enhance English translation skills among high school students. The blend of the Transformer architecture, capable enough to give a deep contextual understanding, has been merged with Generative Adversarial Networks (GAN) to enhance the fluency and accuracy of translations. This model has been tested on actual data collected from 320 students in Basra, Iraq, during the 2024-2025 academic year. The resulting dataset after cleaning, preprocessing, and feature extraction using Natural Language Processing techniques, TF-IDF for example, was then used to train and validate the model. Experimental results indicate that the proposed method is a significant improvement over traditional techniques, including BiLSTM + Fuzzy Inference System (FIS), in terms of accuracy (89.7%), precision (88.2%), recall (87.5%), and F1 Score (87.8%). The model could also well diag-nose, reduce, and remove common translation error types, such as grammatical mistakes, contextual incongruities, and cultural misunderstandings, while promoting student motivation toward learning achievement. As a whole, the Transformer GAN model provides scalable, intelligent, and effective solutions for the modern era of translation teaching, giving real-time feedback in adaptive evaluation. Application in the classroom practice has great potential for blending translation education with AI-assisted learning.