Architecture of modern neural translation systems and polysemous words: comparative analysis of foreign and domestic systems

neural machine translation tools are continually evolving, demonstrating both technological and methodological advances. This article presents a comparative analysis of state-of-the-art architectures of leading domestic and international translation systems focusing on lexical disambiguation. It examines the implementation of modern neural models, optimization of their structural components, and improvements in translation accuracy on complex examples involving polysemous words. Special attention is given to differences in contextual processing and approaches to enhancing source text interpretation quality while minimizing errors. The analysis identifies key strengths and weaknesses of each architecture type, enabling recommendations for the future development of automatic translators. This article targets researchers, specialists, and practitioners working in machine translation and natural language processing.

neural machine translation, word disambiguation in machine translation, architecture of modern neural translation services, context