DISCOURSE BOUNDARIES IN ENGLISH–UZBEK AI PROMPTS: DISTINGUISHING INSTRUCTIONS FROM QUOTED MATERIAL

This paper examines how instruction-bearing expressions function when they occur inside material submitted to a generative AI system for translation, editing or linguistic analysis. Drawing on selected examples from an English–Uzbek prompt dissertation, the analysis distinguishes grammatical form from the authority to direct the current task. Particular attention is given to the interpretation of quoted operations, participant roles and output constraints. The discussion argues that translation must preserve both the meaning of an embedded instruction and its relationship to the enclosing request. A framework based on discourse function, addressee assignment and constraint scope is proposed for subsequent bilingual evaluation. The findings are qualitative and do not constitute evidence of measured model performance.