PRESERVING STYLE AND RHYTHM IN PERSIAN-TO-AZERBAIJANI POETRY TRANSLATION WITH A LARGE LANGUAGE MODEL: A PROMPT-BASED PILOT STUDY OF OMAR KHAYYAM'S RUBAIYAT

The translation of poetry is a particularly demanding test for automated language technologies because semantic accuracy, imagery, tone, rhythm and formal organization must be negotiated at the same time. This pilot study examines how prompt design changes Persian-to-Azerbaijani translations of six quatrains from Omar Khayyam's Rubaiyat, using ChatGPT (GPT-5.6 Sol) under three controlled prompting conditions: literal-semantic, poetic-natural and form-aware. The source texts are taken from the 1942 Foroughi-Ghani edition, and all outputs were generated in a single session on 6 October 2026. The translations are compared qualitatively according to four dimensions: semantic fidelity, preservation of imagery and metaphor, poetic naturalness in Azerbaijani, and preservation of formal-rhythmic organization. The analysis shows a stable trade-off: instructions emphasizing accuracy tend to produce semantically cautious but prose-like lines, while prompts emphasizing poetic form improve cadence, parallelism and end-rhyme at the cost of lexical additions, interpretive shifts and occasional stylistic artificiality. The findings suggest that prompting can redistribute translation priorities but does not remove the central tension between meaning and form, supporting the use of large language models as assistive rather than autonomous agents in poetry translation.