Addressing Lexical Ambiguity in English–Uzbek Neural Machine Translation

Abstract

One of the most difficult phenomena in the Machine Translation realm, and especially concerning English to Uzbek machine translation is polysemy — its meaning multiple in it. In this study, we explore the effect of polysemantic words on translation quality in terms of lexical ambiguity — contextual dependence and semantic disambiguation strategies. Employing a dataset encompassing twenty purposefully chosen sentences embedding words with high polysemy, the study explores the efficacy of modern Neural Machine Translation systems like Google Translate and DeepL. The results show that, as MT algorithms and neural network architectures have advanced, the same systematic deficiencies in resolving semantic ambiguity continue to exist for current systems. In particular, MT outputs are prone to statistical meanings over context-appropriate ones, resulting in errors specific domain interpretation, correct figurative language rendition, and homonym disambiguation. The study further proves that traditional methods including microglossarization and lexical choice codes, though partially useful, cannot simulate the human cognitive mechanism behind polysemy resolution as it depends on contextual correlation or associative networks. Enhancing translation accuracy and ensuring semantic correspondence between source and target texts also relies heavily upon lexical transformation methods—spanning specification, generalisation, differentiation, modulation and lexica compensation. In the end, working with polysemy serves not only to improve English-to-Uzbek machine-generated translation in terms of precision and naturalness, but also adds to the body of knowledge in a field that is concerned about complexity in linguistics more broadly, as well as cross-linguistic semantic relationship and computational perspective toward human-like language processing.