Construction of a Recurrent Neural Network Machine Translation Framework Based on Data Enhancement and Attention Mechanisms
Junting Huang · 2025
In order to improve the semantic modeling ability and translation quality of recurrent neural networks in machine translation tasks, a neural translation framework that incorporates data enhancement and multi-level attention mechanisms is constructed. The research design includes text-level semantic perturbation, language model-based context generation strategy and gated recalibration with multi-scale context fusion in the attention enhancement module, which systematically improves the model’s generalization performance and semantic alignment accuracy in low-resource environments. Comparison experiments on the WMT14 English-German dataset show that the constructed model outperforms the baseline model in BLEU, perplexity and syntactic coverage, especially in long sentence translation and complex syntactic structure processing, which verifies the effectiveness of the structural optimization and strategy integration.