THEORETICAL FOUNDATIONS OF PARALLEL CORPUS-BASED NEURAL MACHINE TRANSLATION

This article offers a paragraph-parallel corpora-based approach to Uzbek English neuro-machine translation, providing theoretical basics that illuminate the approach. Also, literature analysis paragraph-parallel corpora with attention mechanisms, together with application translation quality in progress, efficiency is emphasized. The proposed approach establishes a hierarchical attention layer as an additional context layer in the encoder-decoder architecture. The literature analysis reveals that previous studies have demonstrated that considering context at the document or paragraph level can significantly enhance translation quality. The proposed hierarchical model theoretically integrates context information at the word and sentence levels, providing accurate and consistent translation results. Experimental results have shown that the model provides a deeper understanding of the context and provides higher-quality translations compared to traditional NMT approaches. For example, in tests, the proposed model was found to increase the BLEU indicator by several percent compared to traditional methods. Theoretical conclusions from the model design show that precise alignment of text sections helps to maintain referential relationships. This approach contributes to the logical coherence and content coherence of translation by consistently taking context into account in machine translation systems.