A query does something. Speech act theory says what, and when it succeeds.

J. L. Austin, an Oxford philosopher, founded speech act theory with How to Do Things with Words, published in 1962 from his lectures. He showed that many utterances perform actions, such as promising or naming. He also asked when such acts fail.

John Searle, an American philosopher, systematised speech act theory in Speech Acts in 1969 and proposed five classes of illocutionary acts in the 1970s. He stated the rules that make an act count as what it is. Most later work uses his framework.

Ludwig Wittgenstein argued in Philosophical Investigations in 1953 that the meaning of a word lies in its use within language games. The idea prepared the ground for speech act theory. Meaning became a matter of what words do.

A locutionary act is the act of uttering a sentence with a certain sense and reference, in Austin's terms. It is saying something meaningful. It is the level a keyword index sees.

An illocutionary act is the act performed in saying something, such as asking, warning or promising. The same words can perform different illocutionary acts. A query is an illocutionary act before it is a list of terms.

A perlocutionary act is the effect an utterance produces on the listener, such as persuading, alarming or reassuring. It depends on the listener's reaction. A page aims at a perlocutionary effect whether it names it or not.

Illocutionary force is the kind of act an utterance performs, distinguishing an assertion from a question, a request or a promise. It is part of meaning. Two queries with the same words can carry different forces.

Propositional content is the state of affairs a speech act is about, independent of its force. You will close the door and will you close the door share content and differ in force. A system that reads only content misses the act.

Performative utterances accomplish an action by being uttered, such as I promise or I name this ship. They are neither true nor false in the ordinary way. They show most clearly that language acts.

Language games, in Wittgenstein's sense, are rule-governed activities in which words have their use, such as giving orders, describing or joking. Meaning depends on the game being played. Search is one such game with its own rules.

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

Logical positivism treated meaningful language mainly as statements that can be verified as true or false. Speech act theory showed that much of language does something else. Questions, requests and promises are meaningful without being true or false.

Illocutionary logic, developed by John Searle and Daniel Vanderveken in 1985, is a formal theory of how illocutionary force combines with propositional content. It describes which acts follow from which. It gives the theory a precise form.

Assertives commit the speaker to the truth of a proposition, as in stating, claiming or concluding. Most page content is assertive. Assertions can be checked for truth.

If conversation were mainly a way of stating facts, statements would make up most utterances. They make up about a third.

Directives attempt to get the listener to do something, as in requests, commands and questions. Most queries are directives addressed to a system. An answer is the fulfilment of a directive.

Commissives commit the speaker to a future action, as in promising, offering or threatening. Sign-ups, bookings and pledges contain them. A page that makes an offer performs a commissive.

Expressives express a psychological state, as in thanking, apologising, complaining or congratulating. Many queries carry an expressive layer of worry or frustration. A factual answer can miss it entirely.

Declaratives change a state of affairs by being performed, as in resigning, firing or cancelling, when made by someone with the authority to do so. Online, submitting a cancellation or an application works this way. The act creates the fact.

Pragmatic force is the function an utterance has in its context, as intended by the speaker. It includes illocutionary force and the context that shapes it. It is the part of meaning that makes the same words do different things.

The conditions of success, the listener and the machines. The usual treatment lists Austin's and Searle's acts and classes. It leaves out the conditions under which an act succeeds or misfires, the listener's uptake, the markers that signal force, the extensions to commitment and validity, and the annotation standards and recognition systems that bring speech acts into software. For search, those are the parts that decide whether a query was understood as the act it was.

Felicity conditions are the conditions an act must meet to succeed: the right content, the right circumstances, the sincere state of mind and the essential point of the act. A request is felicitous only if the listener can fulfil it and the speaker wants it. An answer that ignores them answers nothing.

An indirect speech act performs one act by way of another, as when can you pass the salt works as a request. John Searle analysed it in 1975. Queries phrased as questions often perform requests, and statements often perform questions.

If listeners answered the literal question, every yes-no form would draw a yes or a no. For the conventional form, nobody said yes or no and almost everyone simply gave the time.

An illocutionary force indicating device is a marker that signals which act is performed, such as word order, mood, intonation, punctuation or a performative verb. A question mark and a please are such devices. They are the visible trace of force in a query.

Described in a Google patent: https://patents.google.com/patent/US10592540B2/en

Direction of fit describes whether words are meant to match the world, as in assertions, or the world is meant to come to match the words, as in requests and promises. The idea goes back to Elizabeth Anscombe and was developed by Searle. A query for information and a query for action have opposite directions of fit.

Also discussed at https://plato.stanford.edu/entries/speech-acts/

Illocutionary uptake is the listener's recognition of the act being performed, which Austin considered necessary for many acts to succeed. A warning not recognised as a warning has not warned. A system that misreads the act has not taken it up.

Also discussed at https://plato.stanford.edu/entries/speech-acts/

The sincerity condition requires that the speaker has the psychological state the act expresses, such as believing an assertion or intending a promise. Violating it produces lies and false promises. It separates honest acts from insincere ones.

Also discussed at https://plato.stanford.edu/entries/speech-acts/

The preparatory condition covers what must be the case before an act can be performed, such as the listener being able to do what is requested or the speaker having authority to declare. Without it the act is void. A request to a system that cannot fulfil it fails here.

A constative, in Austin's early terms, is an utterance that states something true or false. He contrasted constatives with performatives. The contrast opened the question of what else language does.

A misfire, in Austin's terms, is an act that fails to come off because a procedural condition was not met, such as a marriage performed by someone without authority. It differs from an abuse, where the act comes off but is insincere. Failed queries are often misfires.

Also discussed at https://plato.stanford.edu/entries/speech-acts/

Verdictives, one of Austin's five classes, are acts of giving a verdict or an assessment, as by a judge, an umpire or a reviewer. Reviews and ratings online are verdictives. They carry a judgement with consequences.

Exercitives, another of Austin's classes, are acts of exercising power or rights, such as appointing, voting or warning. They depend on the position of the speaker. Institutional decisions online take this form.

Behabitives, in Austin's classification, are acts expressing attitudes and social behaviour, such as apologising, thanking and congratulating. They correspond roughly to Searle's expressives. They are the social side of what people do with words.

Illocutionary silencing is the prevention of a speaker's act from taking effect even when the words are said, as Rae Langton argued in 1993: a refusal that is not recognised as a refusal. Power decides whose acts receive uptake. It shows that success depends on the listener as much as the speaker.

Also discussed at https://plato.stanford.edu/entries/speech-acts/

A performative contradiction occurs when what someone says conflicts with the presuppositions of saying it, as in arguing that argument is impossible. Karl-Otto Apel and Jürgen Habermas used the concept. A page that denies what its own form claims commits one.

Validity claims, in Jürgen Habermas's theory of communicative action, are the claims every speech act raises: to truth, to normative rightness and to sincerity. Listeners can accept or challenge each one. A page raises all three with every sentence.

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

The conversational score, a notion from David Lewis in 1979, is the running record of what has been established in a conversation, including presuppositions and permissions. Each act changes the score. Dialogue systems need to keep it.

Described in a Google patent: https://patents.google.com/patent/US12148421B2/en

A macro-speech act, a concept from Teun van Dijk, is the overall act a whole text or discourse performs, such as a warning or a recommendation spread over many sentences. Single sentences serve it. A page as a whole performs a macro-speech act.

Interlanguage pragmatics studies how learners of a language perform and understand speech acts, often shaped by their first language and culture. Requests and apologies differ widely across cultures. International pages meet readers who perform acts differently.

Pragmatic competence is the ability to perform and interpret speech acts appropriately in context. It is separate from grammatical competence. Writing that is correct in form can still be pragmatically wrong.

Illocutionary commitment is the obligation a speaker takes on by performing an act, such as standing by an assertion or keeping a promise. Charles Hamblin modelled such commitments as commitment stores in dialogue. Pages accumulate commitments with every claim.

Dialogue act annotation labels each utterance in a dialogue with the act it performs, using schemes such as DAMSL. It turns speech act theory into data for machine learning. Systems learn to recognise acts from such labels.

ISO 24617-2, published in 2012, is the international standard for annotating dialogue acts, developed with Harry Bunt as a leading contributor. It defines communicative functions and dimensions. It gives speech act analysis a shared, machine-readable form.

Speech act recognition is the automatic identification of the act an utterance performs. Voice assistants and chat systems depend on it to decide whether to answer, act or ask back. Errors here send a request down the wrong path.

Also discussed at https://arxiv.org/abs/cs/0006023

If recognising what an utterance does were as easy as reading its words, the model would reach the human level. It stays well below it.

The Dialogue Act Markup Language, DiAML, is the XML representation defined in ISO 24617-2 for encoding dialogue act annotations. It makes annotations exchangeable between systems. It is the technical bridge between theory and software.

Customers perform acts when they search: they ask, request, complain, compare and commit. A page that answers only the words and misses the act loses the customer at the moment the act fails. Conversion is the successful completion of a speech act.

Recognise the act behind the query, meet its conditions of success, and make the page's own acts clear: what it asserts, what it offers and what it promises. Where the act is unclear, ask back. That is the idea of Supervised Search: the page secures uptake before it answers.