Skip to main content
computer-science law linguistics political-science

​
Pragmatic implicature

​
Description

Pragmatic implicature is meaning that the speaker conveys beyond what they literally say, where the listener derives the additional content from context plus the assumption that the speaker is cooperating — being relevantly informative, truthful, and clear (Grice’s maxims). The literal content alone is too sparse to carry the full message; the cooperative-frame assumption licenses the listener to infer the rest. The diagnostic question — “can the speaker cancel the inferred meaning by adding ‘I don’t mean to imply X’ without contradicting their literal claim?” — separates implicature from entailment. If “I don’t mean to imply X” creates a contradiction, X is entailed. If it merely retracts a reasonable inference, X was implicature. “Some of the students passed” implicates not all passed (cancellable: “in fact all of them did”); “all of the students passed” entails some did (not cancellable without contradiction). The structural feature is the gap between what was literally said and what was meant — and the inferential bridge built across the gap by the cooperative frame. Lose the frame (e.g., a hostile interrogation, a literal-minded interlocutor, a cross-cultural mismatch on norms) and the bridge collapses. The implicature was real for the speaker, but the listener can no longer recover it. The cross-domain reach comes from how widely the structure recurs. Anywhere a sender and receiver share a cooperative frame, the sender can compress the message by relying on the receiver to fill in. Anywhere the frame breaks, communication that worked yesterday produces confusion today.

​
Triggers

User-initiated: User describes a situation where the gap between what was said and what was meant is doing work, or where a literal reading would miss the point. Vocabulary cues: “implicature,” “subtext,” “reading between the lines,” “implied,” “diplomatic language,” “what they really meant,” “indirect communication.” Agent-initiated: Agent notices that the surface content of a message is sparse but the surrounding context licenses additional inference. Candidate inference: “what’s the implicature here, and does the cooperative frame still hold?” Also fires when the agent observes a communication breakdown that looks like literal-vs-meant confusion. Situation-shape signals: Polite refusals, diplomatic communication, AI prompt engineering, UI affordance design, sarcasm and irony, indirect requests, hint-driven negotiation, cross-cultural communication, legal-vs-conversational language contrast.

​
Exclusions

  • Strictly literal / formal contexts — mathematical proofs, formal logic, machine-readable protocols, contract law. These domains intentionally minimize implicature because cancellability would destabilize the artifact. Reading implicature into “the function returns int” is a category error.
  • Loss of the cooperative frame — adversarial questioning, hostile cross-examination, deliberate misreading, jailbreaks of AI safety. When the listener has incentive to misinterpret or the speaker has incentive to mislead, implicature stops being the right concept; deception or coercion takes its place.
  • Cross-cultural mismatches on norms — a polite refusal that reliably reads as “no” in one culture can read as “soft yes pending logistics” in another. The implicature was real for the speaker but doesn’t compose with the listener’s frame. The structure is misfire, not implicature working correctly.
  • Sender-receiver mismatch on common ground — implicature requires the sender’s model of what the receiver knows to be approximately correct. A prompt that implicates “use library X” only works if the LLM has heard of library X; a UI that implicates “primary action” only works for users who’ve seen the convention. The implicature concept doesn’t carry when shared common ground is absent.

​
Structure

Internal structure of pragmatic-implicature: a table of its component slots and the concepts that fill them.

​
Relationships

Relationship neighborhood of pragmatic-implicature: a graph of the concepts it connects to and the concepts it is a part of.
  • chain-of-thought — opposite polarity on the reasoning-visibility axis. Together they form the catalog’s spectrum from “expose all inference” to “compress all inference into shared frame.” Many design decisions in prompt engineering and UI copy are choices about where on this spectrum to sit.
  • kayfabe — the social-structure container in which implicature thrives. Performative meetings, ritualized adversarial proceedings, and diplomatic protocol all communicate through implicature licensed by the shared performance frame.
  • endow — implicature is one of the primary delivery mechanisms for endowment when explicit declaration would be heavy-handed. “What do you think we should do?” endows the recipient with deliberation authority via implicature.
  • deixis — sibling concept on the context-dependence axis. Deixis depends on context for reference; implicature depends on context for inference. Both fail when the supporting frame fails.
  • doctrine — Grice’s maxims themselves are doctrines (named rules + triggering conditions + protected failures). The Cooperative Principle is the meta-doctrine; the four maxims are its specializations.

​
Examples

Grice's canonical example · linguistics

A: “How is C getting on in his new job at the bank?” B: “Oh quite well, I think; he likes his colleagues, and he hasn’t been to prison yet.” The literal content is innocuous; the implicature (C is the sort of person who might end up in prison) is licensed by the maxim of relation — B’s mention of prison must be relevant.

AI system prompts · computer-science

“You are a helpful assistant” is sparse literal content. The implicature is enormous: don’t lie, don’t refuse trivially, behave as if you have the role-knowledge, etc. Production prompts run heavily on implicature, which is also why they fail when the cooperative frame fails (jailbreaks attack the frame).
# TODO: revisit this implicates the writer left work incomplete and a future maintainer should pick it up. def calculate_total_with_tax(items, region) implicates that without region the calculation would be wrong. None of this is in the literal signature; the cooperative frame is “code conveys intent to its future readers.”
Sperber and Wilson’s relevance theory recast Gricean implicature on a cognitive foundation. In place of Grice’s four maxims plus a Cooperative Principle, they posit a single principle of relevance: every communicative act carries a presumption of its own optimal relevance, where relevance is the ratio of contextual effects (inferences produced) to processing cost (effort to derive them). The listener interprets utterances by seeking the interpretation that maximizes relevance under that principle — the inference the speaker most plausibly intended given the cost of conveying it that way. The framework subsumes Grice’s maxims as derivable consequences of relevance-seeking and explains both successful implicature and predictable misfires (when the speaker miscalibrates the listener’s contextual resources or processing cost).Inference: Relevance theory’s contextual-effects-over-processing-cost calculus generalizes beyond conversation. Any communication channel where the receiver actively infers meaning from sparse signal — UI affordances, API design, code comments, prompt engineering, diplomatic protocol — is subject to the same calculus. A signal that yields high contextual effects but requires costly inference (cryptic shorthand) and a signal that requires cheap inference but yields low contextual effects (over-verbose obvious statement) both score poorly. The design move is to maximize the ratio: yield the inferences the receiver needs at the lowest plausible processing cost given what the receiver already knows. Most “they should have understood that from context” complaints are misjudgments of the receiver’s available context, not of their inferential capacity.
“We are deeply concerned by recent developments” carries implicature that the diplomatic frame makes interpretable: depending on the surrounding context, this can mean anything from “we noticed” to “we are preparing sanctions.” The cooperative frame here is the institution of diplomatic communication; participants reliably decode the implicature.
Grice’s Cooperative Principle decomposes the pragmatic competence behind implicature into four maxims hearers assume speakers are observing: quantity (be as informative as required, no more), quality (do not say what you believe to be false), relation (be relevant), and manner (be perspicuous). Implicatures arise when a hearer notices that a speaker has apparently violated a maxim and reasons about why — concluding the speaker meant more than they literally said. The famous letter-of-recommendation example: a professor writes only that the student “has excellent handwriting and is always punctual.” The literal content is faintly positive; the implicature, derived from the conspicuous absence of substantive praise (an apparent violation of quantity), is sharply negative. Levinson’s Pragmatics and Sperber and Wilson’s relevance theory refine the framework while preserving the core move.Inference: Implicature’s load-bearing property for the catalog is cancellability — the speaker can append “…and is also a brilliant philosopher” and the negative reading dissolves. That distinguishes implicature from entailment, which can’t be cancelled without contradiction. The cancellability test is the diagnostic the catalog should reach for when distinguishing communicated-by-implication from communicated-by-assertion. The pattern recurs far beyond face-to-face speech. System prompts implicate norms by what they include and omit. Diplomatic and corporate communication routes much of its content through implicature precisely because the cancellability property gives deniability. UI affordances implicate function through placement and styling; code comments implicate intent without asserting it. The portable diagnostic is the question “would a cooperative communicator have said this, in this context, without intending more than the literal content?”
Brown and Levinson’s Politeness argues that face-threatening acts — requests, refusals, criticisms, impositions — are systematically softened across languages by indirection that conveys the speaker’s meaning via implicature rather than literal force. “Could you possibly close the window?” literally asks about ability but implicates a request; “I was wondering whether you might be free Tuesday” implicates an invitation without explicit imposition. The cross-linguistic comparison documents the same family of strategies (hedging, indirectness, conventional politeness formulas) appearing in unrelated language families, suggesting the indirectness-via-implicature pattern is a structural response to a universal social problem (managing face under the pressure of unavoidable face-threatening acts) rather than an idiosyncratic feature of any one language.Inference: The same indirectness-via-implicature pattern is a load-bearing design move outside language — in UI design (a “consider deleting?” affordance versus a “delete” button), in management feedback (questions that license self-correction versus direct critique), and in AI safety prompting (instructions that implicate desired behavior rather than literal refusals). The universality finding implies the design move is not stylistic but structural: implicature-licensed indirectness exists wherever direct force would damage the relationship or the frame more than the underlying message is worth. The diagnostic is asking whether the relationship cost of literal force would exceed the implicature’s cost in inferential bandwidth — when yes, route through implicature; when no, route directly.
“I’ll have to check my calendar” almost always implicates “no” in a context where “no” would be socially expensive to state directly. The cooperative frame is politeness; both parties decode the implicature, and explicit retraction (“I really mean I’m not free, not that I need to check”) feels like a register violation.
implicature with reversed polarity. The literal content is the opposite of what’s meant; the cooperative frame plus prosody / context flips the inferred meaning. Loss of the frame (e.g., reading sarcasm in text without tone) produces misreadings systematically.
Levinson’s Pragmatics is the standard textbook treatment of the field and the most-cited reference for the structure of conversational implicature. It systematizes Grice’s original cooperative-principle framework, distinguishes generalized conversational implicatures (default inferences that arise unless context cancels them — “some” implicates “not all”) from particularized conversational implicatures (inferences that depend on specific local context), works through the cancellability tests that distinguish implicature from entailment and presupposition, and surveys the formalisms (deixis, presupposition, speech-act theory, conversation analysis) that surround the central implicature machinery.Inference: The generalized/particularized distinction is the most under-recognized practical move in the textbook. Generalized implicatures are the ones that travel cleanly across contexts and that LLM prompts, API contracts, and UI conventions can quietly rely on. Particularized implicatures depend on specific local context and break when the context shifts (across cultures, across user populations, across software versions). When designing a communicative artifact intended to scale across contexts, the diagnostic is to identify which implicatures the design is relying on and check whether each one is generalized (safe) or particularized (fragile). Many “this worked in testing but failed in production” surprises are particularized implicatures that didn’t survive the context shift.
a button’s placement in the top-right of a dialog implicates “primary action”; the size, color, and position carry meaning the literal label doesn’t. Users decode the implicature near-automatically; UI designers who break the cooperative frame (by putting Cancel where OK is expected) produce errors disproportionate to the literal-content change.