Abstract
At present, in the field of medicine, high-quality and accurate translation from English into Kazakh is one of the key challenges in ensuring the accessibility and safety of medical information. This scientific study investigates the accuracy and effectiveness of machine translation systems widely used in practice, such as Google Translate and Yandex Translate, when applied to medical texts. The main objective of the study is to explore methods for achieving semantically and stylistically correct translation of medical terminology and complex sentences from English into Kazakh. For this purpose, a specialized corpus consisting of 102,374 sentences was compiled from international medical articles, clinical studies, and drug descriptions. The corpus was processed using the MarianNMT neural machine translation system and translated into Kazakh. For light post-editing of the translation results, the transformer-based Kaz-RoBERTa model was employed, while full post-editing was carried out using one of the large language models (LLMs), namely GPT-4.1, whose adaptability to medical texts was also examined. Translation quality was evaluated using the BLEU, TER, and METEOR metrics. The translations obtained after the initial MarianNMT machine translation were compared with the results after post-editing using the Kaz-RoBERTa and GPT-4.1 models. The analysis showed that translations processed with the Kaz-RoBERTa model achieved an 9% improvement over the baseline MarianNMT translations, while the use of the GPT-4.1 model resulted in a 23% improvement.