2026-07-12T06:42:40.436Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:2:12081 2026-07-05T09:35:55Z 2

Comparative Analysis of Serbian–to–English Translation: A Study of LLM–Based System vs University Students’ Performance 2026 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/2/12081 https://doi.fil.bg.ac.rs/pdf/journals/analiff/2026-1/analiff-2026-38-1-7.pdf Е. Марковић V. Bošković This paper offers a comparative analysis of Serbian-to-English translation performed by university students from the Department of English Language at Singidunum University and by AI-assisted translation systems based on large language models (LLMs), represented here by ChatGPT. The goal of this research is to evaluate the accuracy and overall quality of translation by comparing the performance of a large language model with that of a human translator in an academic setting and, thus, to illuminate the complementary roles of AI translators and human expertise in the translation process. The data for this study were collected by assigning identical translation tasks to both ChatGPT and a group of second-year university students. The research examines 50 texts to evaluate the performance of both translation approaches, resulting in a quantitative and qualitative analysis. Quantitative metrics include error rates and translation accuracy and are complemented by qualitative assessments of contextual appropriateness, idiomatic expressions, cultural references, and stylistic nuances. Preliminary findings indicate distinct strengths and limitations in both LLM and human translations. ChatGPT demonstrates grammatical precision and consistency across translations, while revealing weaknesses in cultural sensitivity, contextual understanding, and context-specific language usage. Conversely, student translations reflect a stronger grasp of cultural and contextual elements, including idiomatic expressions and stylistic flexibility. Nonetheless, variability in grammatical accuracy, structural coherence, and occasional lack of terminological precision may affect the overall quality of student output. These findings provide insights into advancing translation pedagogy, ultimately contributing to the evolving discourse in translation studies and linguistics. article https://doi.fil.bg.ac.rs/volume.php?pt=journals&issue=analiff-2026-38-1&i=7 38 1 139 158 0522-8468 Anali Filoloskog fakulteta