Verb Structure in Modern Eastern Armenian: A Comparison between Human and AI-Generated Translation

This thesis examines the treatment of verb forms in human and machine translations of George Orwell’s 1984 into Modern Eastern Armenian. While neural network–based translation tools, including large language models, have achieved notable progress in fluency and adequacy, their output often diverges from human translations in morphosyntactic patterns. The study is based on a bilingual corpus consisting of the English original text of George Orwell’s 1984, its published Armenian human translation, and machine-generated Armenian translations. A comparative analysis identifies systematic differences in verbal morphology and syntax between human and machine outputs. Alongside the structural analysis, an acceptability questionnaire of the translations was conducted with native speakers of Modern Eastern Armenian. Participants evaluated human- and machine-translated sentences on a five-point scale and provided qualitative feedback on authenticity and grammaticality. The findings contribute to the description of the Modern Eastern Armenian verbal system and offer insights into the capacity and limits of neural machine translation in a morphologically rich, low-resource language. By integrating corpus analysis with native speaker judgments, the thesis highlights both the strengths and shortcomings of machine-generated translations and suggests directions for more linguistically informed evaluation.