Hockett (1960) identified 13 design features of language in Scientific American. While animal communication systems share some features, four are particularly developed in or characteristic of human language (though some, like displacement, appear in limited forms in other species — e.g., bee-dancing communicates about distant food sources): (1) Displacement — referring to things not present in space or time (past events, future commitments, hypothetical scenarios), (2) Productivity — creating and understanding utterances never before produced (novel governance structures, unprecedented policies), (3) Cultural transmission — language is learned, not innate (governance vocabulary is acquired, not genetically encoded), (4) Duality of patterning — meaningless elements (phonemes) combine into meaningful elements (morphemes/words) that further combine into infinite structures. Hauser, Chomsky & Fitch (2002) in Science proposed that recursion — the capacity to generate infinite structures from finite elements — may be the only uniquely human component of the faculty of language (FLN), distinguishing the faculty of language in the narrow sense from the broader faculty (FLB) that includes sensory-motor and conceptual-intentional systems shared with other species. Deacon (1997) argued that human brains and language co-evolved — language is not merely a tool the brain uses but a force that shaped brain architecture. Tomasello (2008) grounded language origins in shared intentionality — the uniquely human capacity for cooperative communication requires understanding others as intentional agents with shareable goals.

The architectural implication: language provides an open generative system — finite elements, infinite compositions. Animal communication is closed — fixed repertoire of signals for fixed situations. Organizational governance requires the open system because governance structures are not fixed — they are created, modified, contested, and replaced through linguistic acts. A wolf pack's hierarchy is genetically channeled. A corporation's governance structure is linguistically constituted.

Austin (1962) established in How to Do Things with Words that language is not merely descriptive. Performative utterances — "I hereby appoint you," "The board resolves that," "This contract binds the parties to" — do not describe a pre-existing reality; they bring a new reality into existence. Austin distinguished three acts performed in any utterance: the locutionary act (producing a meaningful expression), the illocutionary act (the conventional force — asserting, promising, ordering, declaring), and the perlocutionary act (the actual effect on the hearer). The illocutionary dimension — the force that makes a statement a commitment, a question a request, a declaration a constitutive act — is the mechanism through which institutional reality is constructed.

Searle (1969, 1995, 2010) formalized this through his institutional facts framework. In The Construction of Social Reality (1995), Searle established the formula "X counts as Y in context C": a piece of paper (X) counts as money (Y) in the United States (C). Institutional facts — money, property, marriage, corporations, legal authority — exist only because of collective acceptance through language. They differ from "brute facts" (physical facts independent of human cognition) in that they require linguistic constitution. In Making the Social World (2010), Searle argued that all institutional reality derives from "Status Function Declarations" — linguistic acts that assign deontic powers (rights, obligations, authority) to entities. Constitutive rules — rules that do not merely regulate an existing activity but create the very possibility of that activity (the rules of chess don't regulate a pre-existing game; they constitute it) — are the building blocks of all institutional reality.

The governance implication is direct: organizational governance is entirely constituted through speech acts. A policy IS a declaration, not a description of one. An authority delegation IS a performative act, not a report of one. A commitment IS a promise, not documentation of one. Without the linguistic act, the governance structure does not exist. This is not a metaphorical claim — it is an ontological one. Governance structures have no existence independent of the linguistic acts that constitute them.

Winograd & Flores (1986) in Understanding Computers and Cognition applied Austin and Searle's speech act theory directly to organizational computing. Their central insight: organizations coordinate through "conversations for action" — structured exchanges of speech acts (requests, commitments, assertions, declarations) that constitute organizational work, not merely describe it. The network of speech acts — requests followed by commitments followed by assertions of completion followed by declarations of satisfaction — constitutes the organizational coordination process. Winograd & Flores argued that computer systems should be designed not as information processors but as tools for coordinating linguistic action — supporting the speech acts through which organizational reality is constructed and maintained.

This is the most direct intellectual ancestor of governance infrastructure design. The conversations-for-action model — where organizational coordination is constituted by structured speech acts, not merely documented by information systems — provides the theoretical foundation for any governance substrate that treats governance as constitutive (creating organizational reality) rather than descriptive (recording what happened).

LeCun (2022) proposes the JEPA (Joint Embedding Predictive Architecture) framework for autonomous machine intelligence, arguing that world models should be learned from sensory experience rather than text. LeCun's position treats language as downstream of the perceptual substrate — useful for reasoning but not constitutive of the architecture itself. The position is clear: animals build world models without text; therefore text is not necessary for world models; therefore training on text is a "crutch" compared to learning from sensory experience.

The paradox: LeCun's own architecture includes a Configurator module that "modulates the behavior of all other modules" by setting goals, configuring predictions, and adjusting attention. In organizational domains, these functions — goal-setting, prediction-configuring, attention-adjusting — are performed through linguistic acts (strategic plans, policy declarations, reporting requirements). The Configurator that LeCun describes is, in organizational domains, a linguistically constituted module. He is describing a world model component that requires the linguistic infrastructure he dismisses.

The resolution is not that LeCun is wrong but that his argument has a domain boundary. For physical-world prediction (catching a ball, navigating terrain, recognizing objects), language is indeed unnecessary — animals demonstrate this. For organizational governance (setting policy, delegating authority, constituting constraints, declaring purpose), language is constitutive — the governance structures do not exist without the linguistic acts that create them. The animal-substrate argument fails for organizational domains because animal governance is fixed (genetically channeled pack hierarchies, colony roles) while human organizational governance is mutable (constituted and changed through speech acts). The mutable-institutional-reality argument permanently resolves the "animals have world models without language" objection for organizational domains.

Vygotsky (1934/1986) argued in Thought and Language that speech and thought are initially separate systems that merge around age three: thought becomes verbal, and speech becomes representational. Language development follows three stages: social speech (communication), private speech (self-directed talk), and inner speech (silent internal thought). Inner speech — highly condensed, symbolic, self-directed — is not merely the silent version of external speech but a distinct cognitive tool for planning, organizing, and problem-solving. Vygotsky's claim: language is not just a vehicle for expressing thought; thought comes into existence through language.

Carruthers (2002) extended this in Behavioral and Brain Sciences, arguing that language serves cognitive functions beyond communication: it enables explicit reasoning, conscious planning, and metacognition. The cognitive function of language is distinct from its communicative function — we think in language, not just speak in language.

The governance implication: organizational decision-making — weighing alternatives, considering constraints, formulating intent, committing to courses of action — is linguistically structured cognition. Decision-makers deliberate in language. They formulate plans in language. They communicate rationale in language. The internal cognitive process and the external governance act are both linguistic. This means that governance infrastructure must be linguistically compatible — it must operate at the linguistic level where governance cognition occurs.

Clark & Chalmers (1998) argued in Analysis that cognitive processes can extend beyond the biological brain into the environment. Their "parity principle": if a part of the world functions as a process which, were it done in the head, we would recognize as cognitive, then that part of the world IS part of the cognitive process. The Otto-Inga thought experiment: Otto (Alzheimer's patient) uses a notebook as external memory; Inga uses biological memory. If the notebook functions as memory (reliable, accessible, endorsed), it IS Otto's memory — the cognitive process extends beyond the brain.

Hutchins (1995) demonstrated in Cognition in the Wild that cognition in real organizational settings (ship navigation) is distributed across individuals and artifacts — cognitive processes are properties of systems, not just brains. Ong (1982) showed in Orality and Literacy that writing technology restructured human consciousness — literacy didn't just record oral thought but created new cognitive capacities (abstract categorization, logical reasoning, historical consciousness).

The LLM implication: if writing extended cognitive memory (Ong), and if external artifacts can be genuine parts of cognitive processes (Clark & Chalmers), and if organizational cognition is distributed across people and tools (Hutchins), then LLMs are the natural next extension of linguistic cognitive processing. LLMs extend the human capacity for linguistic pattern processing — composition, analogy, cross-domain connection, reformulation — beyond individual cognitive bandwidth. They are not AI replacing human cognition; they are linguistic infrastructure amplifying it, in the same lineage as writing, printing, and computing.

Three categories emerge from mapping language's constitutive role against world model ingredients:

Irreducible (cannot exist without linguistic constitution in organizational domains):<br>- Relational governance structure: The relationships between organizational roles, authority chains, and accountability structures are constituted through declarations, appointments, charters, and contracts. Without linguistic acts, the relational structure does not exist. (Searle's status function declarations; Winograd & Flores's conversations for action.)<br>- Normative constraints: Organizational rules, policies, and prohibitions are constituted through linguistic acts. A regulation IS a linguistic declaration with deontic force. Without the linguistic act, the constraint has no existence. (Searle's constitutive rules; Austin's performatives.)<br>- Declared organizational purpose: Mission statements, strategic objectives, and organizational goals are constituted through linguistic acts. Purpose in organizational domains is declared, not observed. (Searle's status function declarations.)

Beneficial but not irreducible (function better with linguistic support but can exist without it):<br>- Entity identity: Naming and classification are linguistic, but entities can exist pre-linguistically.<br>- Transition logic: Rules of change are often linguistically encoded but can be observed from behavioral patterns.<br>- Observation: Directed attention is linguistically guided but perception is pre-linguistic.<br>- Uncertainty: Risk assessment and uncertainty quantification use language but the phenomenon is pre-linguistic.

Unnecessary (exist independently of language):<br>- Distinctions: Boundary-drawing can be perceptual, not linguistic.<br>- Time: Temporal ordering is pre-linguistic.<br>- Memory-as-storage: Retention occurs in neural and physical substrates without language.

This three-category mapping is the sprint's core deliverable. It resolves the binary (language models vs. world models) by showing that language is architecturally irreducible for some world model ingredients and unnecessary for others — the answer depends on the domain and the ingredient.

The strongest objection to language irreducibility is evolutionary: animals build world models without language; therefore language is not necessary for world models; therefore dismissing language models as "world models" is justified. The mutable-institutional-reality argument resolves this:

Animals have governance. Wolf packs have hierarchy. Bee colonies have role differentiation. Primate groups have dominance structures. But animal governance is fixed — embedded in instinct, genetically channeled, not constituted through declaration. No animal can change its governance structure by declaring new rules. Only humans can constitute, modify, contest, and replace governance structures through speech acts. This is not a difference of degree (humans are smarter about governance) but of kind (human governance is linguistically constituted; animal governance is not).

Therefore: the evolutionary priority objection is correct for physical-world prediction (animals demonstrate world models without language) and incorrect for organizational governance (mutable institutional reality requires language). The field's binary — "language models are not world models" — is true for the physical domain LeCun targets and false for the organizational domain where governance structures are linguistically constituted.