Skip to content

Speech Act Theory Maps to Entity Primitives

Research into whether linguistic categories of utterances map to knowledge graph primitives. Finding: Searle’s 5 illocutionary act categories map almost perfectly to AgentOS’s illocutionary primitives, validating the architecture from a completely independent theoretical framework.


English grammar classifies sentences by their function:

TypeFunctionPunctuationExample
DeclarativeMakes a statement / claims truthPeriod”The meeting is at 3pm.”
InterrogativeAsks a question / seeks informationQuestion mark”When is the meeting?”
ImperativeGives a command / requests actionPeriod or exclamation”Schedule the meeting.”
ExclamatoryExpresses strong feelingExclamation mark”What a great meeting!”

These are surface-level categories. Speech act theory goes deeper.


J.L. Austin (1962, How to Do Things with Words) discovered that every utterance operates at three simultaneous levels:

LevelWhat it isExample: “I’ll be there at 3pm”
LocutionaryThe literal content (what you say)The words and their meaning
IllocutionaryThe act performed (what you’re doing)Making a promise / commitment
PerlocutionaryThe effect achieved (what happens)The listener now expects you

The illocutionary level is the key one — it classifies what the utterance DOES in the world, not just what it says.

Searle’s 5 Categories of Illocutionary Acts

Section titled “Searle’s 5 Categories of Illocutionary Acts”

John Searle (1976) classified all possible illocutionary acts into 5 exhaustive categories based on their “direction of fit” between words and the world:

CategoryDirection of FitWhat it doesDefinition
AssertiveWord → WorldClaims something is trueSpeaker asserts a proposition about how the world IS. The words must match reality.
DirectiveWorld → WordRequests/demands action or informationSpeaker tries to get the hearer to do something. Reality must change to match the words.
CommissiveWorld → WordCommits to a future actionSpeaker commits themselves to a course of action. Reality must change (the speaker will act).
DeclarativeBoth directionsCreates reality by utteranceThe act of saying it MAKES it so. “I now pronounce you married.” “You’re fired.”
ExpressiveNeitherExpresses psychological stateSpeaker expresses feelings about a state of affairs. No truth conditions.

The direction of fit is the deep insight:

  • Assertives: words must match reality (descriptive)
  • Directives & Commissives: reality must match words (prescriptive)
  • Declarations: saying it makes it so (performative — both directions at once)
  • Expressives: neither — feelings just ARE

The mapping between Searle’s categories and AgentOS’s entity primitives is remarkably clean:

Searle CategoryDirection of FitAgentOS PrimitiveHow it maps
AssertiveWord → WorldstoryClaims about reality: “This happened,” “This is true”
Directive (action)World → Wordoutcome (task)Demands reality change: “Do this,” “Make this happen”
Directive (info-seeking)World → WordquestionDemands information: “What is X?”, “When did Y?”
CommissiveWorld → WordoutcomeCommits to future state: “I will do X,” “We aim for Y”
DeclarativeBothrightCreates reality by utterance: “You’re fired,” “This license grants…”
ExpressiveNeither???Feelings and reactions — possibly annotations, not first-class entities

The existing ontological primitives (work, actor, event, list) and the illocutionary primitives (story, outcome, question, right) form two orthogonal axes:

AxisWhat it capturesPrimitivesNature
OntologicalThe nouns — things that EXISTwork, actor, event, listWhat IS in the world
IllocutionaryThe verbs — what people DOstory, outcome, question, rightWhat people DO about the world

This gives a principled answer to “why these primitives and not others?” — because they span both axes of human knowledge: what exists and what we do about it.

Searle’s 5th category (expressives) doesn’t clearly map to an entity primitive. Expressives include:

  • “Thank you” (gratitude)
  • “I’m sorry” (apology)
  • “Congratulations” (celebration)
  • Emoji reactions, likes, upvotes

These might be best modeled as annotations on other entities (reactions to posts, acknowledgments of events) rather than first-class primitives. A “like” is an expressive act ON a work, not an independent entity.


The fact that a knowledge graph architecture independently arrived at categories that map to a 70-year-old linguistic framework is strong validation:

  1. Story = Assertive: Both are about claiming truth. A story asserts “this happened.” An assertive speech act claims “this is the case.”

  2. Outcome/Task = Directive + Commissive: Both are about changing reality. A task directs “do this.” A goal commits “we will achieve this.” Both have the same direction of fit — reality must change to match the words.

  3. Question = Directive (info-seeking): Searle classified questions as a subtype of directives — they direct the hearer to provide information. This validates question as its own primitive (a directive that seeks information rather than action).

  4. Right = Declarative: Both create reality by utterance. A license declares “you may do X.” A marriage certificate declares “you are married.” The act of declaring IS the creation.

  5. Work, Actor, Event, List: These have NO counterpart in speech act theory — correctly so. They describe what EXISTS (ontology), not what people DO (illocution). The two axes are genuinely orthogonal.


5. The Direction of Fit as a Design Principle

Section titled “5. The Direction of Fit as a Design Principle”

The “direction of fit” gives a clean test for classifying new entity types:

DirectionTestIf yes, it’s…
Word → World”Does this claim something about reality?”Story / assertive
World → Word”Does this demand reality change?”Outcome / directive+commissive
Both”Does saying this make it so?”Right / declarative
Neither”Does this express a feeling?”Annotation / expressive
N/A”Does this just exist?”Ontological primitive

This research led to the “5+5” (later 5+4, then evolved further) primitive architecture:

Ontological primitives (what EXISTS):

  • work — things people create
  • actor — things that can do things
  • event — things that happen at a time
  • list — organizational grouping
  • record — structured data entries

Illocutionary primitives (what people DO):

  • story — claims about reality (assertive)
  • outcome — desired states of the world (directive + commissive)
  • question — frames seeking resolution (directive, info-seeking)
  • right — reality created by declaration (declarative)

The expressives gap was resolved by treating reactions/expressions as relationship data (annotations) rather than entities.