SEO says relevance a thousand times a day. Relevance theory says what it is.

Dan Sperber is a French cognitive scientist and anthropologist who developed relevance theory with Deirdre Wilson. Their book Relevance: Communication and Cognition appeared in 1986. It moved the study of meaning from codes to inference.

Deirdre Wilson is a British linguist at University College London and the co-founder of relevance theory. She developed its account of explicature, irony and lexical meaning over four decades. Much of pragmatics today is written in her terms.

Paul Grice showed that people communicate by recognising intentions, and he described conversational maxims such as be relevant and be as informative as required. Relevance theory grew out of his work. It replaced his maxims with a single principle of relevance.

Also discussed at https://plato.stanford.edu/entries/implicature/

An utterance is an act of communication that comes with a built-in promise: that it is worth the listener's attention. The words encode only part of what is meant. The listener infers the rest.

Ostensive-inferential communication is communication in which a speaker openly shows an intention to inform and the listener infers what is meant. Showing and inferring are the two halves. A search result is a showing, and the reader does the inferring.

The cognitive principle of relevance states that human cognition tends to be geared to the maximisation of relevance. Attention goes where the most can be gained for the least effort. A reader scanning a result page follows this principle without knowing it.

The communicative principle of relevance states that every act of open communication conveys a presumption of its own optimal relevance. Whoever addresses someone implicitly claims to be worth the effort. A title in a result list makes that claim.

Processing effort is the mental work needed to understand an input and draw conclusions from it. Other things being equal, more effort means less relevance. A page that makes the reader work hard for its point lowers its own relevance.

If working out "not all" cost nothing, both readings would become available at the same moment. The inferred reading arrives 140 ms later, and 120 ms later than the same meaning stated with "only".

A cognitive effect is a worthwhile change in a person's picture of the world: a new conclusion, a strengthened belief, a corrected error. The greater the effect, the greater the relevance. An answer that changes nothing in the reader's mind is irrelevant, however well it matches the query.

An explicature is what a speaker explicitly communicates, the literal meaning developed with context into a full thought. It is already more than the words encode. A two-word query becomes a complete question only through this development.

Also discussed at https://plato.stanford.edu/entries/pragmatics/

An implicature is an assumption a speaker communicates without saying it, which the listener has to infer. In relevance theory it follows from the search for relevance. What a query implies can matter more than what it says.

An informative intention is the intention to make certain assumptions manifest to an audience. It is what a speaker wants the listener to take in. A page has one, whether its author thought about it or not.

A communicative intention is the intention to make the informative intention itself recognised. The listener sees that someone is trying to tell them something. This openness separates communication from mere information.

The code model describes communication as encoding a message into a signal and decoding it with a shared code. Relevance theory rejects it as incomplete, because the code delivers only evidence of what is meant. Keyword matching is a code model of search.

Described in a Google patent: https://patents.google.com/patent/US9940367B1/en

The presumption of optimal relevance is the assumption that an utterance is relevant enough to be worth processing and the most relevant one the speaker could and wanted to produce. It tells the listener where to stop interpreting. A reader stops at the first reading that pays off.

Relevance theory treats metaphor as a loose use of language, a point on a continuum from literal to looser uses. The listener takes from the word what is relevant and drops the rest. No special mechanism is needed.

Also discussed at https://plato.stanford.edu/entries/metaphor/

Relevance theory explains irony as echoic: the speaker echoes a thought or an utterance and expresses a dissociative attitude towards it, such as mockery or rejection. Great weather, said in the rain, echoes an expectation. The attitude is the point.

Hyperbole is treated as another loose use, an exaggeration the listener adjusts to what is relevant. I have told you a thousand times communicates frequency and annoyance. The number is not meant to be counted.

Stephen Levinson criticised relevance theory as too reductive and argued for default inferences that arise without context, the generalised conversational implicatures. The debate shaped both positions. Search systems work with defaults all the time.

The measurements and the finer machinery. The usual treatment names the principles, explicature and implicature. It leaves out how effort is measured in the eyes, the pupils and the brain, how word meanings are adjusted on the spot, how listeners guard against being misled, and the rival traditions the theory argues with. For search, this is where relevance stops being a slogan and becomes something that can be tested.

A higher-level explicature embeds the explicit content in a description of the speaker's attitude or speech act, such as she is asking whether or he regrets that. It captures how something is said. A query can carry doubt or urgency at this level.

A weak implicature is one of many possible conclusions a speaker encourages without committing to any single one. Poetic and suggestive language works this way. The reader carries more of the responsibility for what is taken.

A strong implicature is a conclusion the speaker clearly intends and that the listener must draw to make sense of the utterance. The speaker is responsible for it. Most answers to direct questions work through strong implicatures.

Procedural meaning is meaning that tells the listener how to process the content, carried by words such as but, so or after all. Such words encode instructions for inference. They guide the reader through a text.

Conceptual meaning is meaning that contributes a concept to the thought, as nouns and most verbs do. Relevance theory separates it from procedural meaning. The difference shows why small connecting words change how a page is understood.

Mutual manifestness is the relevance-theoretic replacement for mutual knowledge: an assumption is mutually manifest when both parties could represent it and know that the other could. It asks for less than certain shared knowledge. It matches what a searcher and a page can realistically share.

A cognitive environment is the set of assumptions that are manifest to a person at a given moment. Context is drawn from it. Two searchers typing the same query bring two different cognitive environments.

An ad hoc concept is a word meaning adjusted on the spot, narrowed or broadened for the occasion. Robyn Carston developed the idea. Cheap in cheap flights means something else than in cheap perfume.

Loose use is the use of a word with a meaning looser than its strict sense, as in the room is empty for a room with a chair in it. Approximation, hyperbole and metaphor lie on one continuum. The literal reading is rarely the intended one.

Also discussed at https://plato.stanford.edu/entries/metaphor/

A poetic effect is the relevance an utterance gains from a wide range of weak implicatures. No single conclusion carries it. Memorable writing often works this way.

A non-propositional effect is an impression, a feeling or a sensory quality an utterance creates without a statable conclusion. Tone of voice and style produce such effects. Readers take them from a page as much as they take facts.

The relevance-theoretic comprehension procedure says: follow a path of least effort in computing effects, test interpretations in order of accessibility, and stop when your expectation of relevance is satisfied. It describes how people actually settle on a meaning. It is also a description of how people read a result page.

Also discussed at https://plato.stanford.edu/entries/pragmatics/

Robyn Carston, a linguist at University College London, developed the relevance-theoretic account of explicature and lexical pragmatics, above all in Thoughts and Utterances from 2002. Her work on ad hoc concepts shapes the field. Much of what relevance theory says about word meaning is hers.

Diane Blakemore introduced the distinction between conceptual and procedural meaning in relevance theory, in Semantic Constraints on Relevance from 1987. She showed how discourse connectives constrain interpretation. The concept of procedural meaning is hers.

Free enrichment is the pragmatic addition of material to the explicit content that no word in the sentence demands, as in I have eaten, understood as today. It is optional and driven by relevance. Queries are enriched this way by every reader.

Saturation is the filling of a slot that the words themselves require, such as who I refers to or what ready is ready for. It is obligatory. It separates what a sentence needs from what a reader adds.

An unarticulated constituent is an element of what is meant that no word expresses, like the place in it is raining, as John Perry described. The thought is complete, the sentence is not. Queries are full of such unspoken parts.

Also discussed at https://plato.stanford.edu/entries/pragmatics/

Described in a Google patent: https://patents.google.com/patent/US8200694B1/en

A generalised conversational implicature is an inference that arises by default in most contexts, such as some suggesting not all. Stephen Levinson built his theory of presumptive meanings on it. Relevance theory treats the same inferences as context-driven.

Also discussed at https://plato.stanford.edu/entries/implicature/

If adults read "some" only by its literal meaning, they would accept the giraffe sentence as readily as the children. Fewer than half of them accept it.

A particularised conversational implicature arises only in a specific context. Is there a petrol station near here, asked while holding a can, implies that the car has run dry. The implicature disappears in another situation.

A scalar implicature is the inference from a weaker term on a scale to the denial of a stronger one, as from some to not all. Experiments show that people draw it at very different rates for different scales. It is the best-measured implicature of all.

Also discussed at https://plato.stanford.edu/entries/implicature/

If the stronger meaning were ruled out by the word alone, every weaker word would produce the same rate. The rates run from nearly everyone to almost no one.

Truth-conditional pragmatics, associated with François Recanati, holds that pragmatic processes help decide what a sentence says, beyond what it implies. Relevance theory shares this view. Meaning is settled with context from the start.

Also discussed at https://plato.stanford.edu/entries/metaphor/

Linguistic underdetermination is the fact that the encoded meaning of a sentence never fully determines the thought it expresses. Every utterance needs inference to be understood. A query underdetermines the need behind it in the same way.

Epistemic vigilance is the set of mechanisms by which people check whether to believe what they are told, described by Dan Sperber and colleagues in 2010. Understanding and believing are separate steps. A page can be understood perfectly and still not be believed.

Theory of mind is the capacity to attribute beliefs and intentions to others. Inferential communication depends on it. Understanding what someone means requires imagining what they have in mind.

Metarepresentation is the ability to represent a representation, such as a thought about a thought or an utterance about an utterance. Irony, quotation and attributed speech rely on it. So does every judgement about what a writer intended.

Billy Clark is a linguist whose textbook Relevance Theory, published by Cambridge University Press in 2013, is a standard introduction. He applied the theory to stylistics and to teaching. His work made the theory usable beyond specialists.

Tim Wharton is a linguist who extended relevance theory to non-verbal communication in Pragmatics and Non-Verbal Communication from 2009. Gestures, facial expressions and tone of voice fall within its scope. Communication is wider than words.

Eye-tracking records where and how long the eyes rest while reading. Longer fixations and regressions indicate greater processing effort. It turns the effort that relevance theory speaks of into milliseconds.

Pupillometry measures the dilation of the pupil, which increases with mental effort. It is an involuntary signal of cognitive load. Effort that a reader would never report shows in the eye.

An event-related potential is the brain's electrical response to a specific event, recorded at the scalp and averaged over many trials. It shows processing with millisecond precision. Language researchers use it to see interpretation as it happens.

The N400 is a negative brain wave peaking about 400 milliseconds after a word, larger when the word is hard to integrate into its context. Marta Kutas and Steven Hillyard reported it in 1980. It is a measure of how well meaning fits expectation.

The P600 is a positive brain wave around 600 milliseconds after a word, associated with reanalysis of structure or interpretation. Lee Osterhout and Phillip Holcomb described it in 1992. It marks the moment a reader has to reconsider.

Electroencephalography records electrical brain activity through electrodes on the scalp. Event-related potentials are derived from it. It is the method that lets claims about processing effort be tested.

Bayesian inference updates beliefs in proportion to new evidence. Computational models of pragmatics, such as the rational speech act model by Michael Frank and Noah Goodman, treat interpretation as such an update. They formalise what relevance theory describes.

Surprisal is the negative log probability of a word given its context, a measure from information theory. The less predictable a word, the higher its surprisal and the longer people take to read it. It gives processing effort a number.

A relevant result is one that gives the reader the greatest cognitive effect for the least effort, in their situation. Keyword match is one piece of evidence for that. Pages that answer the question behind the query with little effort are relevant in the sense the reader's mind uses.

Say the point early, remove effort that produces no effect, and give the reader a conclusion they did not have. Address what the query implies as well as what it says, and earn belief as well as attention. A page that asks back what the reader needs lowers the effort of finding the relevant part, which is the idea of Supervised Search.