Speech act

• 1. Introduction

  ›

  A speech act is what happens when saying something and doing something turn out to be the same event. Say "I resign," "I apologise," or "You're fired" in the right circumstances, and you have not described a resignation, an apology, or a dismissal; you have carried one out. That simple observation, that language is not just a vehicle for stating facts but a form of rule-governed social behavior, upended a long tradition in the philosophy of language that treated statements as either true or false and little else. If speaking can itself be an act of resigning or promising, what exactly is being performed, and by what rules does it succeed or fail? How do listeners ever figure out that a question is secretly a request in disguise? And how did an idea born in a set of philosophy lectures end up shaping how software agents talk to each other and how courts read the law?

• 2. Naming The Act

  ›

  J. L. Austin delivered the lectures that would define the field at Harvard in 1955, later published posthumously as How to Do Things with Words. He opened by taking aim at a long-standing assumption in analytic philosophy: that the business of a "statement" is simply to state some fact, true or false, and nothing more. Austin pointed to a class of utterances he called performatives, where saying the words is already doing the thing they name, as evidence that this picture was incomplete.

  His student John R. Searle carried the project further in Speech Acts: An Essay in the Philosophy of Language, published in 1969, building explicit rules for how illocutionary acts are performed and proposing influential taxonomies of their types. Searle's later work connected the theory to intentionality and social ontology, and between Austin's initial framework and Searle's systematization, speech-act theory became a central tool across philosophy of language, pragmatics, communication studies, legal theory, and artificial intelligence.

• 3. Older Than Austin

  ›

  Long before Austin's Harvard lectures, the Scottish philosopher Thomas Reid had already drawn a line between purely "solitary" mental operations and what he called essentially "social" acts of mind, such as asking, commanding, and promising, acts that require being expressed to another person and that generate obligations between speaker and hearer. In 1913, the phenomenologist Adolf Reinach published an essay, Die apriorischen Grundlagen des bürgerlichen Rechtes, that worked out a detailed ontology of "social acts," including promising, claiming, and commanding, each one capable of creating new normative relations once successfully performed.

  Parallel developments were underway in linguistics itself. The Slovene philologist Stanislav Škrabec analyzed performative uses of tense and aspect, examining utterances where, in effect, saying so makes it so, such as judicial pronouncements. The psychologist and linguist Karl Bühler used the terms Sprechhandlung, or "speech action," and Theorie der Sprechakte, or "theory of speech acts," in his 1930s work on the functions of language, most notably in his 1934 book Sprachtheorie. Even earlier, in his lectures on the philosophy of right, Georg Wilhelm Friedrich Hegel described speaking as a form of acting, not merely speaking, tying utterance directly to the transformation of social reality. The later philosophy of Ludwig Wittgenstein pushed in a related direction, urging that the meaning of an expression be found in its use within a practice rather than in the abstract, and describing speaking a language as playing a "language-game," a shift toward "meaning-as-use" that later surveys treat as part of the backdrop to Austin and Searle's work.

• 4. Three Layers Of An Utterance

  ›

  Austin's framework splits an ordinary utterance into three layers. The locutionary act is simply producing a meaningful expression, with a particular sound, grammar, and conventional meaning. The illocutionary act is what the speaker does in saying it, whether that is asserting, warning, requesting, or promising, and it carries what Austin called illocutionary force. The perlocutionary act is what the utterance accomplishes afterward in an audience, such as persuading, amusing, or alarming them; if someone says "I'm hungry" and the listener responds by making food, that response counts as the perlocutionary effect of the original line. Later theorists added a further category, the metalocutionary act, which organizes or comments on the discourse itself rather than advancing its subject, such as marking a topic boundary or signaling the start of a turn.

  The illocutionary act sits at the center of the theory, because it is the level that captures the basic social action a speaker is performing. Austin's own examples ranged across stating, warning, ordering, and promising, each governed by "felicity conditions" that have to hold for the act to succeed, such as a speaker actually having the authority to pronounce a couple married. Searle treated "speech act" and "illocutionary act" as nearly interchangeable and built a systematic taxonomy of basic types, including assertives, directives, commissives, expressives, and declarations, distinguished in part by the beliefs, desires, and intentions each one expresses. Where Austin emphasized the conventional procedures behind an illocutionary act, Searle and those who followed him leaned more heavily on the mental states and inferences involved in recognizing one.

• 5. When Saying Makes It So

  ›

  "I promise to pay you back," "I sentence you to ten years' imprisonment," and "I hereby name this ship Sojourner" do not describe a promise, a sentence, or a naming; each one constitutes the act, provided the surrounding conditions are met. Austin argued that these performative utterances cannot properly be judged true or false at all, only "happy" or "unhappy," depending on whether the speaker held the right authority, the procedure was carried out correctly, and the participants held the appropriate intentions.

  Austin sorted failures of felicity into two categories: misfires, where the purported act does not come off at all because the procedure was never properly invoked, no marriage contracted and no bet laid despite the words being spoken; and abuses, where the act is successfully performed but done insincerely, as when a speaker promises without any intention of following through. Searle later generalized this into a set of constitutive rules attached to each type of illocutionary act. For promising specifically, he identified four conditions: a propositional content condition on what may be promised, preparatory conditions about the relationship between speaker and hearer, a sincerity condition requiring genuine intent to follow through, and an essential condition under which the utterance counts as taking on an obligation. Violating the content or preparatory conditions produces a misfire; violating sincerity produces an abuse, an insincere but still binding promise.

• 6. Reading Between The Lines

  ›

  Ask "Can you close the window?" and almost no one answers with a report on their own physical ability; the question functions instead as a request. The declarative "It's cold in here" can carry out the same job, prompting someone to shut a window or turn up the heat without any command ever being spoken aloud. Indirect speech acts like these are especially common in requests and refusals: answer "Would you like to meet for coffee?" with "I have class," and the reply reads as a polite rejection, not a factual statement about a schedule. English has entire conventionalized forms built for this, such as "Could you...?" or "Would you mind...?", designed to soften the imposition built into a direct command. Fieldwork on Jordanian shops and streets has even documented vendors using music and scent, alongside verbal cues, to invite a purchase without issuing an explicit command at all.

  Searle proposed that hearers work out an indirect illocution through a chain of inference grounded in Grice's Cooperative Principle, reasoning from the literal words and the surrounding context toward the conclusion that a request or refusal is actually underway. Later work in pragmatics and politeness theory has questioned whether one single inferential pattern can explain every case. Everyday language is full of utterances doing two jobs at once: "I hereby appoint you as chairman" both confers a new status and performs the act that grants it; an announcement asking passengers to extinguish cigarettes and return their tray tables both describes a rule and issues a command to follow it; and "Well, would you listen to that?" functions simultaneously as a request for attention and an exclamation of disbelief.

• 7. Far Beyond Philosophy

  ›

  In 1975, the researcher John Dore proposed that children's earliest utterances could be sorted into nine primitive speech acts, including labelling, requesting, greeting, and protesting, giving language acquisition researchers a framework for tracking how a child learns to communicate before mastering full grammar.

  Computer scientists picked up the theory to model how machines and people, or machines and other machines, coordinate. In 1991, researchers Morelli, Bronzino, and Goethe built a computational speech-act model of human-computer conversations for medical decision support, typing each exchange as a request, assertion, or confirmation. Terry Winograd and Fernando Flores went further in their book Understanding Computers and Cognition, treating everyday office coordination as networks of conversations built from requests, offers, promises, and declarations, tracked through a state-transition diagram; that "conversation for action" model went on to shape workflow management and business-process design.

  Outside computing, the Copenhagen School in political science treats securitization itself as a felicitous speech act, where a speaker, often a politician, frames an issue in a way meant to move an audience toward action, regardless of whether the underlying claim is true. In legal theory, laws themselves are read as speech acts that issue commands to citizens, and a death threat is treated as a criminal speech act that falls outside free-speech protection precisely because of what it does rather than merely what it states. In economic sociology, researcher Nicolas Brisset has argued that economic models themselves function as speech acts, performing actions across scientific, academic, and political fields at once, an approach that in 2016 was extended to finance under the label "financial logos," describing how mathematical risk models actively frame financial decision-making rather than simply describing it.

• 8. Rules, Logic And Deception

  ›

  No formalization of speech-act theory has ever won general agreement. In 1985, John Searle and D. Vandervecken tried to lay the groundwork for an illocutionary logic capable of capturing the theory in formal terms. Others took different routes: Per Martin-Löf worked on treating assertion inside intuitionistic type theory, while Carlo Dalla Pozza proposed a formal pragmatics that connects propositional content, handled with classical semantics, to illocutionary force, handled with intuitionistic semantics. More recently, researchers in artificial intelligence have proposed a Bayesian approach to formalizing speech acts, while the main working applications of the theory in practice remain concentrated in human-computer interaction, particularly chatbots.

  Designers of communication protocols for distributed and multi-agent systems have leaned on Searle's distinction between two kinds of rule: regulative rules, which constrain a behavior that could exist without them, such as traffic regulations, and constitutive rules, which actually define an activity, the way the rules of a game define what counts as a legal move within it. Software agents in these systems exchange messages carrying performatives such as inform, request, or query, formal labels drawn directly from Searle's categories and built into languages including KQML and the FIPA Agent Communication Language, treated as constitutive rules that create commitments and institutional facts rather than merely limiting what messages can be sent. The computer scientist Munindar P. Singh has argued that the "mentalist" semantics built into languages like these, which interpret messages in terms of an agent's beliefs and intentions, are poorly suited to open systems, and has pushed instead for a social semantics built on publicly observable commitments between agents rather than assumptions about their internal states.

  The same framework has found a use catching people in lies. Speech-act profiling has been applied to detect deception in real-time computer-mediated conversation, by analyzing the mix of speech-act types, assertions, questions, commitments, that appear across a chat or instant-messaging exchange, on the theory that a deceptive speaker's pattern of acts differs measurably from a truthful one's.