WHEN A COMMAND IS ONLY A QUOTATION: DISCOURSE SCOPE AND INSTRUCTIONAL AUTHORITY IN ENGLISH-UZBEK AI PROMPTS

Abstract

Prompts used for translation, editing and linguistic analysis frequently contain sentences that themselves express instructions. Their presence creates a problem that cannot be resolved by identifying imperative verbs alone: a command may constitute the current task or form part of the material on which that task operates. This article examines the distinction through a qualitative analysis of three instruction- bearing passages selected from an unpublished dissertation on English and Uzbek generative AI prompts. The manuscript examples are analysed in their documented explanatory contexts; additional English-Uzbek formulations are explicitly identified as analytical constructions. The procedure distinguishes grammatical form, discourse function, represented addressee and the scope of output constraints. The analysis identifies three potential misinterpretations: treating a quoted operation as the current task, applying an embedded constraint to the enclosing response, and assigning a quoted participant role to the current system. It also shows that translation may preserve the directive meaning of a sentence without authorising its execution. The article proposes a discourse-based annotation framework for examining these distinctions. Its contribution is interpretive and methodological; no claim is made about measured model performance.