Enunciation is the act through which abstract language becomes concrete speech—through which a sentence in the dictionary becomes an utterance directed by a speaker to a listener in a specific context. De Certeau adopted this concept from Émile Benveniste to distinguish between language as a system of signs and language as a practice of communication. The same words mean differently depending on who says them, to whom, when, and why. Tone, emphasis, context, relational history—all shape meaning in ways that semantics alone cannot capture. AI-generated text is language without enunciation. It is syntactically and semantically competent but has not been said by anyone. The practitioner who uses AI output must supply enunciation—must transform the model's generated code into spoken language by directing it at a specific audience with a specific intention. This operation is invisible to observers but determines whether the work communicates or merely occupies space on the page.
Benveniste's distinction between langue (the abstract language system) and parole (the concrete speech act) provided the linguistic foundation for de Certeau's broader theory of practice. Language exists in two irreducible modes: as a structure studied by linguists and as an activity performed by speakers. The structure enables the activity but does not determine it. The dictionary defines 'home,' but the word's meaning when spoken—'I'm going home,' 'This is my home,' 'You're not welcome in my home'—depends on the enunciation. Context, intonation, and relational positioning transform the abstract definition into lived meaning. De Certeau recognized that this transformation is not merely linguistic but practical: it is the operation through which every abstract system becomes a lived reality.
The practitioner who works with AI must perform enunciation at every level. At the word level: replacing the model's preferred term with the term that captures her specific meaning. At the sentence level: adjusting tone, rhythm, emphasis to match the work's purpose. At the structural level: directing the model's generic organization toward the particular argument she is making. And at the intentional level: deciding not just what the text says but what she means by it—what purpose it serves, which audience it addresses, which aspects of her understanding it communicates. Without these operations, the text remains langue—competent, coherent, and experientially empty.
The difficulty of enunciation in AI-assisted work is that the model's outputs look like they have already been said. The prose is fluent. The structure is clear. The arguments are presented with confidence. But no one has directed this language at anyone. It is optimized for a statistical average of readers the model has never met. The practitioner must supply the direction—must transform the for-everyone into the for-someone, the generic into the specific, the code into the practice. This is the hardest and most important creative operation available to the AI-assisted builder, and it is the operation that separates work that lives from work that merely passes.
De Certeau developed the enunciation framework in his readings of Benveniste's Problems in General Linguistics (1966–74) and in his mid-1970s essays on language, writing, and speech. The concept received its fullest treatment in The Practice of Everyday Life, where de Certeau extended it from spoken language to all forms of practice—walking is the enunciation of the urban system, cooking is the enunciation of the recipe, reading is the enunciation of the text.
Enunciation transforms system into practice. Language becomes speech when someone says it to someone. The abstract becomes concrete through the act of directed communication.
AI generates langue; the builder supplies parole. The model produces syntactically correct, semantically coherent text that no one has said. The practitioner directs it, means it, makes it hers.
The same text means differently depending on the saying. Context, intention, relational positioning shape meaning in ways that semantics cannot capture.
Enunciation is invisible but constitutive. Observers see the text; they cannot see the practitioner's act of directing it. Yet the act determines whether the text communicates or lies inert.