Optimizing Software Engineering English Translation using A Transformer Model with Enhanced Grey Wolf Optimization

The translation of documents related to software engineering (SE), is often very difficult be-cause of the uniqueness of vocabulary used in this area, the very strict syntax used when de-veloping software and many contextual dependencies when translating this type of docu-ment. In this paper we present an adaptive transformer-based neural machine translation (NMT) model customized with Enhanced Grey Wolf Optimization (EGWO) to provide better translation results in the software engineering field. The adaptive nature of our model allows for fine-tuning of hyper-parameters such as learning rate, dropout rate and number of attention heads to improve convergence and generalization from training to unseen data. We conducted several experiments to evaluate our model using established benchmark datasets including WMT, OPUS and UM-Corpus; results indicate that our EGWO transformer-based model produced an accuracy rating of 94.2%, precision rating of 93.5%, recall rating of 92.8% and significantly outperformed both LSTM and standard transformer-based models with respect to BLEU scores. Our approach presents an economical and scalable solution to highquality translation of software localization and domain-specific machine translation.