PROMPT AS TEXT: TOWARDS A RHETORIC AND POETICS OF HUMAN–AI CO-COMPOSITION

Abstract This article argues that the prompt—the linguistic utterance through which a human initiates and directs a generative AI system—should be understood not merely as a technical input to be optimised, but as an emerging literary-rhetorical text in its own right, and therefore a legitimate and pressing object of humanistic inquiry. Since the early 2020s, prompting has become one of the most frequently performed acts of written composition in ordinary life, yet the vocabulary available to describe it has been supplied almost entirely by computer science and industry discourse, both of which treat the prompt as a lever for extracting reliable machine output rather than as an utterance with its own history, form, and stakes. This article contends that such a vocabulary is radically insufficient, and that the humanities possess the conceptual resources—developed over centuries of thinking about authorship, address, genre, and interpretation—to theorise prompting with a precision that purely instrumental accounts cannot achieve. The argument proceeds by placing prompting into conversation with several bodies of literary and rhetorical theory. It begins with the mid-twentieth-century debates over authorship associated with Roland Barthes and Michel Foucault, arguing that prompting intensifies rather than resolves the questions those debates raised: if meaning was already understood to be completed by an interpreting reader rather than fixed by an originating author, prompting introduces a second, non-human interlocutor who intervenes during composition itself, displacing and redistributing authorial intention before a finished text exists at all. The human prompter is repositioned as something closer to a librettist, a director, or a translator than to a sole originating author, retaining a directing role while relinquishing exclusive control over phrasing, structure, and even conceptual elaboration. The article then turns to Mikhail Bakhtin's account of dialogism to explain the temporal and structural shape of sustained prompting. Where a single prompt met by a single output might be dismissed as an isolated and rather thin speech act, most substantive prompting in practice takes the form of an extended, iterative exchange in which the prompter responds to what the system produces, refines instructions, supplies new constraints, and adjusts tone or direction in light of the system's previous response. This turn-by-turn structure externalises, in an unusually literal way, the dialogic quality that Bakhtin located within the novel's absorption of multiple voices, and suggests that the proper unit of literary-critical analysis is not the isolated prompt but the exchange as a whole, with its characteristic rhythms of proposal, response, correction, and refinement—rhythms that also resonate, the article suggests, with call-and-response traditions in oral and performative literatures, including the antiphonal structures of Bengali kirtan and pala-gaan.摘要 本文主张，提示词（prompt）——即人类发起并引导生成式AI（generative AI）系统的言语表达——不应仅被视作有待优化的技术输入，而应被视为一种新兴的独立文学修辞文本，因此成为人文研究中合法且亟待探讨的对象。自21世纪20年代初以来，提示词操作已成为日常生活中最常见的书面创作行为之一，但用以描述该行为的词汇体系几乎完全来自计算机科学与产业界的话语，二者均将提示词视为获取可靠机器输出的操控工具，而非拥有自身历史、形式与利害关联的言语表达。本文认为此类词汇体系存在根本性缺陷，而人文领域拥有历经数百年发展而来的概念资源——包括对作者身份、言说对象、文体类型与阐释方式的思考——能够以纯粹工具性阐释无法企及的精准度，对提示词操作展开理论化研究。 本文的论证通过将提示词操作与多套文学与修辞理论展开对话得以推进。开篇以20世纪中期罗兰·巴特（Roland Barthes）与米歇尔·福柯（Michel Foucault）围绕作者权展开的论争为切入点，指出提示词操作非但未解决，反而激化了这些论争提出的问题：倘若学界早已达成共识，认为意义并非由原创作者固定，而是由阐释性读者完成，那么提示词操作则引入了第二位非人类对话者，该对话者会在创作过程中介入，在成品文本诞生前便消解并重新分配作者意图。人类提示发起者的身份也应被重新定位：相较于单一的原创作者，其更接近歌剧脚本作者、导演或译者，虽仍保有引导角色，却放弃了对措辞、结构乃至概念阐释的独占控制权。 随后本文转向米哈伊尔·巴赫金（Mikhail Bakhtin）的对话理论，用以阐释持续提示词操作的时间维度与结构形态。单次提示词搭配单次输出的情形或许会被视作孤立且单薄的言语行为，但现实中绝大多数实质性提示词操作均采取延伸性的迭代交互形式：提示发起者会根据系统生成的内容做出回应，细化指令、提供新的约束条件，并依据系统此前的输出调整语气与方向。这种逐轮交互的结构以异乎寻常具象的方式，将巴赫金所提出的、存在于小说对多重声音吸纳过程中的对话性特质具象化，同时表明文学批评分析的基本单元不应是孤立的提示词，而应是完整的交互过程——后者拥有提议、回应、修正与优化的特有节奏。本文还指出，此类节奏也与口头及表演性文学中的呼应对答传统相契合，包括孟加拉语克尔坦（kirtan）与帕拉甘（pala-gaan）的对唱结构。