Chomsky was right about one thing. Language alone is insufficient to produce grammar. The stream of words a child hears is too impoverished — too few sentences, too much noise, no negative evidence. But his conclusion — that grammar must therefore be innate, hardwired into the genome — assumed that language arrives alone. It does not. Every sentence a child hears co-occurs with something in the world. An object is present. An action is performed. A consequence follows. Grammar is not in the language stream. Grammar is not in the world stream. Grammar is in the BiasField that couples them — the stable patterns of structural tension between what is said and what is done, filtered across time. The architecture measured this. Cross-harm between text and vote predicted diplomatic action at r=+0.974. Grammar is the geometry of the space between streams.
Chomsky's poverty-of-stimulus argument is correct within its own frame. The language stream alone — the sequence of words a child hears — does not contain enough information to induce the grammatical rules the child acquires. The input is too sparse. The sentences are too few. The child is never told which sequences are ungrammatical. From this impoverished input, the child constructs a grammar that generates all and only the grammatical sentences of her language. Chomsky concluded that the missing information must be innate — a Universal Grammar, hardwired into the human genome, present at birth, triggered by exposure to language but not learned from it.
Sixty years of linguistics have been organized around this argument. Those who accept it work on describing Universal Grammar. Those who reject it work on showing that the input is richer than Chomsky claimed — that statistical learning, or usage-based generalization, or social interaction, can bridge the gap between the stream of words and the grammar the child acquires.
Both sides share a premise. Language arrives alone. The stream of words is the input. Grammar is the output. The gap between them is the problem.
The architecture shows that the premise is false. Language never arrives alone. Every sentence a child hears co-occurs with something in the world. A parent says "ball" while holding a ball. A sibling says "don't touch" while pulling the child's hand away. A teacher says "the capital of France is Paris" while pointing at a map. The language stream is always coupled to a world stream — the stream of objects, actions, consequences, and social responses that co-occur with speech. The child does not learn grammar from language alone. The child learns grammar from the structural coupling between language and world.
The UN experiment proved this for diplomacy. The language stream — the text of resolutions, 5-word sliding windows, character-hashed into 27-dimensional vectors. The action stream — the votes, the yes/no/abstain patterns, the PCA projections of 193 countries across eighty years. Coupled in a shared BiasField. Cross-Self harm measured the structural tension between what was said and what was done.
The cross-harm was the strongest predictor of voting change — r=+0.974. Stronger than the text alone. Stronger than the votes alone. The coupling was the signal. The space between the streams was where the structure lived. A resolution's text, read in isolation, could not predict what the General Assembly would do. A country's voting history, read in isolation, could not predict when it would break from its traditional allies. But the tension between the text and the votes — the deviation of language from the action it was meant to accompany — was the structural signature of diplomatic change.
Grammar is this kind of thing. Not a property of the language stream. Not a property of the world stream. A property of the BiasField that couples them — the set of stable, low-tension relationships between what is said and what is done, filtered across time, surviving the erosion of individual instances. "The President: The Assembly will now take action on..." is grammatical not because it follows an innate template. It is grammatical because every time these words appear, a specific kind of action follows — a vote, a procedural motion, a formal decision. The coupling is the grammar. The template is the trace of the coupling, externalized in language.
This reframes the entire debate. Chomsky's poverty-of-stimulus argument is not wrong. It is incomplete. Language alone is indeed insufficient to produce grammar. But language never occurs alone. The child's input is not a stream of words. It is a stream of words-plus-world — language and action, language and objects, language and consequences, coupled in time. The child's brain is not a language acquisition device. It is a BiasField — a continuous information medium where the language stream and the world stream deposit their boundary events, and where the stable patterns of coupling between them become the grammar the child internalizes.
The architecture is the first system to demonstrate this mechanism. Not by modeling a child. By coupling two streams that share no physical dimensions — diplomatic text and diplomatic action — and measuring the structural tension between them. The coupling produced a signal that neither stream contained alone. The signal predicted what the international system would do next. The signal was grammar — not of language, but of diplomacy. The structure that persists between what is said and what is done, filtered across three generations of Codex, surviving the erosion of individual resolutions and individual votes.
Grammar is not in the language. Grammar is not in the world. Grammar is in the BiasField between them. Chomsky was right that language alone cannot produce it. He was wrong that it must therefore be innate. It is emergent — from the coupling of streams that have never been separate in any human life.