Improving Genetic Algorithm-Based Automatic Machine Translation Models with Gated Recurrent Units and Bilingual Evaluation Understudy

To address the low performance and limitations of traditional machine translation models, this study proposes an automatic machine translation model based on an improved Genetic Algorithm with Elite Strategy. The model uses Gated Recurrent Units to optimize the initialization of the population. Bilingual Evaluation Understudy is employed to evaluate complete translations, and experiments validate the proposed approach. The study’s simulation experiments showed that the improved genetic algorithm achieved Chinese-English mutual translation similarity rates of 88.15% -91.98% and 89.39 % -95.32 %, respectively. After further enhancements, the proposed model reached similarities of $\mathbf{9 0. 1 5 \% - 9 7. 7 8 \%}$ and $\mathbf{9 1. 3 8 \% - 9 8. 0 7 \%}$. In contrast, other models’ similarities ranged from $\mathbf{7 5. 0 0 \% - 8 9. 0 6 \%}$. Real-world performance tests revealed that the proposed algorithm scored 7.66-9.08 for accuracy and 8.55-8.83 for fluency, while the refined model achieved 9.01-9.28 for accuracy and 8.87-9.10 for fluency. Other models averaged $\mathbf{7. 5 6}$ for accuracy and $\mathbf{7. 5 0}$ for fluency. In summary, the proposed algorithms and models generate high-quality translations, enhancing the applicability of machine translation technology across various fields.