A Cultural-Aware Transformer Model for Enhanced English Neural Machine Translation

Machine translation has advanced rapidly with the rise of neural networks, yet it continues to face challenges when dealing with culturally rich content-especially in English, where idioms, regional expressions, humor, and tone significantly influence meaning. Traditional systems, including rule-based methods and early neural models such as RNNs and LSTMs, often rely on direct word-to-word translation. As a result, they fail to convey the intended message accurately in cross-cultural contexts, leading to misinterpretations and reduced translation effectiveness. To overcome these limitations, this research proposes a Cultural-Aware Transformer with Contextual Embedding (CAT-CE) - a novel neural machine translation framework designed to better understand and process cultural nuances in English. The model enhances the standard Transformer architecture by integrating a contextual embedding layer that identifies and represents cultural elements such as idiomatic phrases, emotional tone, and language formality. This enables the system to preserve both semantic meaning and cultural context throughout the translation process. Experimental evaluations show that the proposed CAT-CE model achieves a BLEU score of 34.7, indicating higher accuracy in translating culturally loaded sentences, and provides improved contextual coherence compared to baseline Transformer and RNN-based models. The results highlight CAT-CE's potential to significantly enhance the quality and reliability of culturally aware machine translation systems.