The Utterance (Bakhtin)
In Bakhtin's framework, the utterance is the real unit of communication, as opposed to the sentence (the abstract unit of linguistics). A sentence is a grammatical structure existing outside time and social context; an utterance is what someone actually says, saturated with the speaker's intention, the listener's anticipated response, the history of how these words have been used before, and the social situation that makes this particular speech act meaningful. Utterances are inherently dialogic: they respond to prior utterances and anticipate replies. They are also inherently evaluative: every utterance takes a position, even when it appears neutral. Bakhtin's insistence on the utterance as the primary object of analysis was a challenge to structuralist linguistics, which studied language as a system of abstract relations. He argued that language lives not in dictionaries or grammars but in utterances — in actual, historically located, socially embedded acts of speech.
In the AI Story
Bakhtin's theory of the utterance received its fullest development in 'The Problem of Speech Genres' (written 1952–1953, published posthumously 1979). He argued that utterances are shaped by speech genres — relatively stable types of utterances that correspond to different spheres of human activity. The scientific paper, the legal brief, the dinner conversation, the love letter — each is a genre with its own conventions, its own relationship to truth, its own way of addressing and anticipating response. Learning a speech genre means internalizing not just vocabulary and grammar but the social logic of a domain: what can be said, what goes without saying, what must be said carefully, what is better left unsaid.
The AI prompt is an utterance in Bakhtin's sense: it responds to the builder's prior interactions with the tool, to the problem the builder is trying to solve, to the genre conventions of prompting (which are emerging in real time as communities of practice share techniques). The machine's response is also an utterance: shaped by the training data (prior utterances from millions of speakers), by the prompt (the immediate dialogic context), and by the parameters governing generation (temperature, top-p sampling, system prompts). The exchange is a dialogue of utterances, not a mechanical input-output process. The meaning emerges from the responsiveness, the multi-directionality, the saturation with prior and anticipated speech acts that neither party controls individually.
The Orange Pill documents moments when the dialogic character of the utterance became phenomenologically salient: when Claude's response reshaped Segal's understanding of his own question, when a prompt Segal thought was clear turned out to carry assumptions the machine interpreted differently, when the iterative refinement of an idea through twenty exchanges produced an insight that belonged to neither participant but to the dialogue itself. These are not anomalies but the norm: every utterance is a node in a vast web of responsive relations, and the AI makes this web visible by responding with a breadth of contextual awareness no human interlocutor could match.
The prescriptive implication: training in AI collaboration should teach utterance-awareness — the capacity to recognize that every prompt is shaped by prior interactions, that every response carries traces of the training corpus, that meaning emerges from the dialogue rather than being contained in either prompt or response. This is not a technical skill (though it has technical dimensions) but a rhetorical and philosophical one: understanding communication as fundamentally responsive, fundamentally social, fundamentally inexhaustible by any single speaker's intention.
Origin
Bakhtin began thinking about the utterance in the 1920s but developed the concept systematically only in the 1950s, in essays that would be published after his death. The theory was partly a response to Saussurean structuralism, which treated language as a synchronic system of differences, and partly a development of Voloshinov's (possibly Bakhtin's own) earlier critique of abstract linguistics in Marxism and the Philosophy of Language (1929).
The concept has been foundational to pragmatics, conversation analysis, and discourse studies since the 1980s. Its application to human-AI dialogue is a 2020s extension, though the framework fits with surprising precision.
Key Ideas
The utterance is the real unit of communication. Not the abstract sentence but the actual speech act in context.
Utterances are inherently responsive. Every act of speech replies to prior utterances and anticipates future responses.
Speech genres organize utterances. Stable patterns of address, tone, and expectation correspond to different social domains.
Prompts and responses are utterances. AI dialogue is a genuine exchange of socially embedded speech acts, not mechanical input-output.
Meaning emerges from responsiveness. The significance of an utterance lives in its dialogic relations, not in any speaker's isolated intention.
Appears in the Orange Pill Cycle
Further reading
- Mikhail Bakhtin, 'The Problem of Speech Genres,' in Speech Genres and Other Late Essays (1986)
- V.N. Voloshinov, Marxism and the Philosophy of Language (1929/1973)
- J.L. Austin, How to Do Things with Words (1962)
- John Searle, Speech Acts (1969)
- Herbert Clark, Using Language (1996)