This paper presents a novel approach to English translation by proposing a dual-channel Transformer model based on cross-layer semantic fusion. The model leverages dual processing paths and semantic integration to improve translation fluency and accuracy. Existing Transformer-based translation methods often suffer from semantic redundancy, weak feature interaction, and limited depth in cross-layer representation, leading to suboptimal translation quality. To address these limitations, we introduce the Dual-Channel Transformer with Cross-Layer Attention Fusion (DCT-CLAF), which utilizes two separate encoding channels and a cross-layer attention mechanism to enhance semantic representation and context comprehension. The dual channels independently capture shallow and deep semantic features, while the cross-layer fusion integrates these representations to improve alignment and translation accuracy. The proposed method is applied to English translation tasks across benchmark datasets to evaluate its effectiveness. Experimental results demonstrate that DCT-CLAF significantly outperforms traditional Transformer models in BLEU score, semantic coherence, and contextual accuracy. This confirms the model’s potential in delivering more natural and precise translations in real-world applications. The proposed method achieves a translation accuracy of 91.3%, semantic adequacy of 89.6%, syntactic fluency of 9, polysemy resolution score of 81.8%, low-resource adaptability of 83%, and model inference time of 200.4 ms.
Hongdan Wang (Wed,) studied this question.