Speech Act Theory

In The You On AI Field Guide

The theory emerged from ordinary language philosophy's insistence that meaning is use. Austin's 1955 Harvard lectures, published posthumously as How to Do Things with Words (1962), catalogued the variety of actions language performs: promising, warning, betting, marrying, naming ships. Searle's contribution was to formalize the structure—identifying the felicity conditions (sincerity, authority, conventional procedure, appropriate context) that distinguish successful from unsuccessful speech acts. A promise made without intention to keep it is insincere; a priest's 'I pronounce you married' spoken by someone without authority does not constitute marriage. The illocutionary force depends not just on words but on the speaker's standing, intentions, and the social framework recognizing the utterance as binding.

Winograd and Flores's application to computing identified a mismatch at the heart of conversational AI. Human utterances to machines are genuine speech acts—requests carrying intentional weight, descriptions shaping commitments. Machine responses have the syntactic structure of commitments ('I'll restructure the database schema') without illocutionary substance—the AI does not commit to anything, does not intend anything, does not undertake obligations. It produces tokens shaped by statistical patterns of how commitments are linguistically expressed. When humans interpret machine outputs as genuine commitments, treating statistical reliability as intentional undertaking, the collaboration inherits risks invisible to surface performance. The gap between form and force is the gap between a tool and a partner.

Origin

Austin's framework grew from his 1950s Oxford ordinary language philosophy seminars, challenging the logical positivists' assumption that language's primary function is stating facts that are true or false. He demonstrated that vast territories of language—questions, commands, promises, apologies—are not truth-evaluable at all; they are performances whose success depends on social conditions, not correspondence to reality. Searle, Austin's student at Oxford, systematized the insights into a formal theory that became foundational to philosophy of language, pragmatics, and—through Winograd and Flores—human-computer interaction design.

Key Ideas

Locutionary, illocutionary, perlocutionary. Three dimensions of utterance: producing meaningful words, performing social action through those words, and producing effects on the listener—only the first is purely linguistic.

Felicity conditions. Speech acts succeed when speaker possesses appropriate authority, intends what the utterance expresses, follows conventional procedure, and operates in a context recognizing the act as binding.

AI's missing illocutionary force. Language models produce tokens with the syntactic structure of promises, explanations, commitments—but lack the intentional and social substance making those structures genuine actions.

Design for commitment support. Technology should make the structure of organizational speech acts—requests, promises, assessments—visible and manageable, supporting coordination without replacing judgment about what to commit to.