THE IMPACT OF ARTIFICIAL INTELLIGENCE ON THE QUALITY OF ENGLISH–UZBEK TRANSLATION

This paper examines the qualitative and quantitative impact of artificial intelligence (AI) and modern neural machine translation (NMT) paradigms on English–Uzbek translation performance. Utilizing both automatic metrics—such as Bilingual Evaluation Understudy (BLEU), Translation Edit Rate (TER), and Crosslingual Optimized Metric for Evaluation with Translation (COMET)—and human error annotation through the Multidimensional Quality Metrics (MQM) framework, this study evaluates the capabilities of leading commercial engines and open-weight Large Language Models (LLMs). The research focuses on the linguistic challenges inherent in translating between an analytical, isolating language with Subject-Verb-Object (SVO) order (English) and an agglutinative, synthetic language with Subject-Object-Verb (SOV) order (Uzbek). Empirical results indicate that while AI models have significantly reduced error rates in general news and conversational domains, low-resource dataset constraints, subword tokenization artifacts, and complex morphosyntactic alignments continue to produce systematic errors in specialized legal, medical, and literary registers.