Crossing Linguistic Barriers: A Hybrid Attention Framework for Chinese-Arabic Machine Translation

This study proposes an innovative method for Chinese-Arabic machine translation(zh-ar_MT), integrating adaptive local attention mechanisms(ALAM) with dynamic global attention mechanisms(DGAM) to effectively address differences in structure and grammar between the two languages. The local attention mechanism focuses on precisely capturing the relationships between neighboring words to deepen the understanding of local syntactic structures and fine-grained semantics, while specially optimizing for the word order and morphological differences in Chinese and Arabic. The global attention mechanism is used to capture long-distance dependencies within the entire sentence, providing the model with a comprehensive understanding of the context. By combining the outputs of these two mechanisms and further processing through a self-attention network, the model effectively coordinates local details with global contextual information. Compared to traditional Transformer methods, the proposed model achieved an increase of 2.25 (27.88-30.13) and 4.56 (31.19-35.75) BLEU points respectively in Chinese to Arabic (zh-ar) and Arabic to Chinese (ar-zh) translation tasks. This study provides a new perspective in the field of Chinese-Arabic machine translation and offers valuable reference for translating other language pairs with significant structural and grammatical differences.