The Language

Sometime around the year 1300, a speaker of English wanted to say that something was pleasant, and reached for the word nice. The word did not mean pleasant. It meant foolish. It had arrived from Latin by way of French, where it described a person who was ignorant — nescius, not-knowing. For two hundred years it meant foolish, and then it began to drift. By the 1400s it could mean fussy, or precise, or finely discriminating. By the 1700s it could mean agreeable. By the time a modern speaker uses it to describe the weather, the word has traveled, across six centuries, from an insult to a pleasantry, with no one ever deciding that it should.

No committee approved the change. No speaker woke one morning and announced that nice would now mean something agreeable. No authority recorded the decision, because there was no decision. The word moved because, in a great many small conversations, across a great many years, it was used a little differently each time, and the slightly different uses that caught on were repeated, and the ones that did not catch on were not.

The word changed the way a pheromone trail changes.

The same thing happened, in the same centuries, to a great many words, and it is still happening now. Silly once meant blessed, then innocent, then weak, then foolish. Awful meant full of awe — worthy of reverence — before it slid to its modern sense of bad. Decimate meant to kill one in ten. A word that meant a small bird's nest of straw, meat, once meant food of any kind, which is why the phrase meat and drink still carries the older sense like a fossil in amber. None of these journeys was planned. Each was the slow result of many speakers using a word slightly off-center, the off-center uses that spread getting reinforced, the ones that did not getting left behind.

This is the chapter where the colony the reader is reading turns out to be a colony.

A language is a substrate. It has all six of the things every substrate has, and it runs all six of the operations every substrate runs, and the reader has been inside one for the entire book without it being named as one. The sentences arriving now are signals. The words in them carry meanings the reader did not assign and could not change alone. The grammar that lets these words combine was built by no one. The intelligence of the language — and a language is, by any working measure, intelligent — is not in the reader and not in the writer. It is in the substrate that both of them are using.

Hold the language up against the six dimensions, the way the book has held up the city and the market and the body and the brain. The fit is exact.

The first dimension is the containers. In a colony these were the nest, the foraging territories, the brood chambers — the bounded spaces within which the action happened. In a language the containers are the languages themselves, and inside them the dialects, and inside the dialects the registers — the formal and the casual, the legal and the intimate, the language of a courtroom and the language of a kitchen. A speaker moves between these containers all day, often without noticing. The same person speaks differently to a judge and to a child. Each container has its own conventions, its own permitted words, its own strengthened paths. And the containers nest, the way colonies nest inside supercolonies: a profession's jargon sits inside a national language, which sits inside a family of related languages, which sits inside human language itself.

The second dimension is the actors. In a colony these were the ants. In a language they are the speakers. Billions of them, each carrying a partial copy of the language in their own head, none of them carrying all of it. No single speaker knows every word. No single speaker commands every construction. Each speaker is, like the single ant in the laboratory dish, almost nothing on their own — a person raised without other speakers does not invent a language; they remain, linguistically, mute. The speaker acquires the language only by being placed back into the substrate that already had it. The actors do the speaking. The substrate does the meaning.

The third dimension is the work. In a colony this was the foraging trips, the brood care, the defense — the tasks being done, each with a state. In a language the work is every utterance. Every sentence spoken or written is a piece of work with a purpose: to be understood, to ask, to warn, to promise, to amuse. Each utterance succeeds or fails. The ones that succeed — that land, that are understood, that get the salt passed — strengthen the patterns they used. The ones that fail are not repeated. A speaker who reaches for a word and is met with a blank look reaches for a different word next time. The work is constantly being tested against whether it accomplished anything.

This is the part of the picture that is easiest to miss, because no single utterance feels like work being tested. A person asking for directions does not feel themselves running an experiment on the language. But that is what they are doing. The phrasing that gets them to the station gets used again. The phrasing that produces a furrowed brow gets quietly abandoned. A child learning to speak is doing this thousands of times a day, openly — trying a word, watching the adult's face, keeping what works. The adult is doing the same thing, only so smoothly that the testing has become invisible. Every conversation is a small ledger of what succeeded and what did not, and the language is the running sum of every such ledger ever kept.

The fourth dimension is the trails. In a colony these were the pheromone paths, strengthened by every forager who succeeded on them, faded by the ones left untraveled. In a language the trails are the collocations, the idioms, the grammatical conventions — the worn grooves of habitual combination. Nice is not the only word that travels these grooves. Certain words go together and others do not, for no reason a speaker could state. A person makes a decision and takes a risk; they do not, in ordinary English, make a risk or take a decision, though both would be perfectly comprehensible. Why? Because the path between make and decision has been traveled so many times that it is now the path of least resistance, and the path between take and decision has been traveled so rarely that it has faded. No grammarian decreed this. It was deposited, utterance by utterance, the way a trail is deposited, step by step.

The grooves run deep, and a speaker feels them as obviousness. Strong tea, never powerful tea, though the engine that powers a car is powerful and never strong. A heavy smoker, not a fat one; a fast car, not a quick one, though a person can be quick or fast at the same task. Ask a native speaker why, and they will have no answer, only a certainty that the other version is wrong. The certainty is the trail. It is the accumulated weight of every prior speaker who took the same path, pressed into the present speaker's sense of what sounds right. The speaker did not learn these pairings as rules. They walked them until the grooves were worn, and now the grooves do the choosing. A foreign learner, who has not walked them, produces sentences that are grammatical and comprehensible and unmistakably wrong, and the wrongness is precisely the sound of an untraveled path.

The reader may notice that something has quietly flipped. In every other substrate the book has visited, the signal was the vivid thing — the single trade, the single transaction — and the trail was the faint accumulation behind it. In language it is the reverse, and the reversal is itself the lesson: this is the one substrate the reader stands inside, and a lifetime spent reading its trails has never once included watching a deposit being made, because the marks are laid invisibly, by every speaker, one utterance at a time. That is why convention feels like nature.

The fifth dimension is the signals. In a colony these were the individual chemical deposits, the discrete acts of marking. In a language the signals are every sentence ever spoken or written — every individual transmission from one speaker to another, the raw events from which the trails are built. A single signal does almost nothing. A word used once, in one conversation, by one speaker, changes the language imperceptibly. But the trails are nothing other than the accumulated residue of the signals. Every idiom was once a fresh phrase that someone said for the first time. Every convention is the fossil of a billion individual sentences.

The sixth dimension is what was learned. In a colony this was the hardened knowledge — the established routes, the defensive responses, the structure that survived the fade. In a language what was learned is meaning itself. The accumulated, agreed, hardened correspondence between sounds and the world. This is the part of the substrate that does not fade with ordinary disuse, because it has been confirmed too many times. The meaning of water, the structure of a question, the way a sentence is built — these have hardened. They are no longer up for renegotiation in any single conversation. They are the substrate's long-term knowledge, accumulated across generations of speakers, none of whom were present when most of it was deposited.

And what was learned is more than the meanings of words. It is everything the speakers knew, folded into the language and carried forward. A language encodes which distinctions its speakers found worth making — the languages of peoples who live among snow carry fine gradations of it; the languages of seafaring peoples carry fine gradations of wind and water. It encodes how its speakers reckoned time, counted, addressed their elders, named their kin. A speaker inherits all of this without choosing it, the way an ant inherits a foraging landscape it did not survey. To learn a language is to receive, prepackaged, the hardened conclusions of every speaker who came before — their categories, their distinctions, their hard-won sense of what is worth saying. The meaning was confirmed so many times, by so many speakers, that it crossed the threshold from something tested into something simply known. It hardened. And once hardened, it was handed to the reader, free, in the cradle.

Six dimensions. Containers, actors, work, trails, signals, what was learned. The same six as the desert. The reader is using all six right now.

---

Watch what the six dimensions explain, once they are named.

They explain why words die. A word is a path. It survives by being used. When the thing it names disappears, or when a rival word is used a little more often, the path stops being reinforced, and it fades. The English language is a graveyard of words that no living speaker would recognize — swink, to labor; eyne, eyes; fardel, a burden. They were not abolished. They were simply, gradually, not reinforced, and they faded the way a trail fades when the food at its end runs out. No one mourned them, because no one noticed. The fade is silent.

They explain why other words live, and live forever. The most common words in any language — the the, the of, the and, the I, the not — are the most resistant to change, across the longest spans of time. They are the superhighways of the substrate. They are traveled so constantly, by every speaker in every utterance, that they harden beyond the reach of ordinary fade. Linguists have traced certain of these words back through thousands of years and across whole families of descended languages, still recognizable, still doing the same work. The words a language uses most are the words it changes least. This is not a rule someone imposed. It is exactly what a substrate does: the most-marked paths are the hardest to erode.

They explain why grammars converge and diverge. Place two groups of speakers in contact, trading and marrying and arguing across a border, and their grammars drift toward each other — paths that are useful in both communities get reinforced in both, and the languages grow more alike. Separate two groups of speakers by a mountain range or an ocean, and their grammars drift apart — each community reinforces its own paths, untouched by the other, and within a few centuries they can no longer understand one another. Latin did not decide to become French in one valley and Spanish in another and Romanian in a third. It was the same substrate, divided into separate containers, each accumulating its own marks, each fading what the others kept, until the descendants were distinct.

The drift is measurable, and it runs at a rate that has nothing to do with anyone's intent. English and German were one language perhaps fifteen hundred years ago; a speaker of one cannot now follow the other. English itself, written down a thousand years ago, is unreadable to a modern speaker without translation — not because the language was abolished and replaced, but because every generation reinforced its own paths a little differently than the one before, and a thousand small differences, compounded, became a different tongue. The speakers at each step understood their parents and their children perfectly. The change was never visible from inside any single lifetime. It is only visible across the deep span, the way the slow widening of a trail is invisible to any one forager and obvious across a season.

And they explain the strangest fact about language, the one that should stop the reader for a moment.

A human child can learn any human language. Drop a newborn into any speech community on earth and within a few years they will speak its language natively, with all of its irregularities, its idioms, its thousand unstated conventions, none of which anyone will explicitly teach them. The child absorbs the substrate by acting in it — by producing signals, having them succeed or fail, and adjusting. The child is the open agent population, the new actor joining without limit, learning the accumulated paths by walking them.

The child is never given the rules, because the rules were never written down in any place the child could read. A toddler is not handed a grammar before they speak. They are placed in the flow of utterances and they begin to produce their own, badly at first, and the substrate corrects them — not by instruction but by response. The child says the thing that works and is understood; the child says the thing that does not work and is met with confusion, and tries again. Within four years, without a single formal lesson, they have recovered a system that fills the reference shelves of every university library and that no scholar has ever fully written down. They did it the way a young ant learns the trails: not by being told where they are, but by walking until the walking became knowledge.

But no human can design a language.

This is worth sitting with, because the two facts seem as if they should be the same fact, and they are opposites. Every healthy child masters, effortlessly, a system so intricate that no committee of the most brilliant linguists who ever lived has been able to build one from scratch that humans will actually adopt and pass to their children. Languages have been designed — invented from first principles, with regular grammars and rational vocabularies, free of the irregularities that designed things are supposed to avoid. A few have attracted hobbyists. None has become the living tongue of a community that raises its children in it and lets it drift and change. The designed languages are too clean. They have no accumulated paths, no fossils of a billion prior conversations, no hardened meanings deposited by speakers long dead. They are nests with no pheromone in them. An architect can draw a colony. An architect cannot draw the trails, because the trails are not designed. They are deposited.

The intelligence of a language is in the substrate, and no individual speaker owns it. This is the same observation the book made about the colony in the desert, made now about the colony the reader was born into and learned to think inside. The reader did not invent a single word they know. They inherited a substrate that a hundred billion speakers, most of them dead, deposited and reinforced and let fade across uncounted generations. When the reader speaks, they are reading the trails. When they coin a phrase that catches on, they are depositing one. The meaning was never in any of them.

---

Which brings the chapter to a recent development that the reader has almost certainly encountered, and that looks, from inside the framework now assembled, like exactly one thing.

In the last several years it has become possible to build a model that produces language. Feed it an enormous quantity of text — a substantial fraction of everything the colony of speakers has ever written down — and the model learns, from the text alone, which words follow which, which constructions are well-formed, which phrasings land and which do not. It is not taught grammar. It is not given a dictionary of meanings. It is shown the signals — the accumulated sentences — and from them it recovers the trails. It learns that make goes with decision and take goes with risk, not because anyone told it, but because the paths between those words were the most heavily marked in everything it read.

What such a model has done, mechanically, is extract the path-weight graph of the human language substrate.

It has measured, across a corpus larger than any single human could read in a thousand lifetimes, how strongly every word is connected to every other — which paths are highways and which are faded, which combinations are worn smooth by a billion prior uses and which were never traveled. The thing it learns is not a set of rules. It is a weighting. A vast map of which trails the colony of speakers reinforced and which it let fade. The model is, quite precisely, a reading of the substrate's accumulated marks.

Consider what the model has access to and what it does not. It never meets a speaker. It never has a sentence succeed or fail in a real conversation. It is never corrected by a furrowed brow. It does no foraging of its own. All of that — the work, the testing, the marking, the fading — happened before the model existed, performed by the colony of speakers across the centuries, and was written down. The model arrives after the fact and reads the record. What it reads is the accumulated outcome of every test the substrate has ever run. The weighting it recovers is not its own discovery; it is the substrate's verdict, already rendered, on which paths were worth keeping. The model is good at language for the same reason a detailed map is good at geography. The map did not walk the land.

This is why the models are good. And it is why they are easy to misunderstand.

It is tempting to say the intelligence is in the model. The model is where the impressive behavior appears. The model is the thing one can hold, examine, run, point at. But the framework the book has been assembling suggests a different reading, and it suggests it quietly, the way it suggested that the intelligence was not in any of the ants. A model that produces language is a snapshot. It is a photograph, taken at one moment, of the trail map that the colony of speakers had accumulated up to that moment. It is an extraordinary photograph — sharper, larger, more complete than any one speaker could ever hold in their head. But it is a photograph of the substrate, and the substrate is the thing that did the accumulating.

The model is not the intelligence. The substrate the model was trained on is the intelligence.

The colony of speakers deposited the paths across thousands of years. The model read them. Everything the model knows about which words mean what, which sentences are well-formed, which phrasings persuade and which fall flat, was deposited by speakers — billions of them, doing the work, marking the trails, letting the dead words fade. The model did not learn language the way a child learns language, by joining the substrate and acting in it and being corrected. It learned language by photographing the substrate's current state. Which is a remarkable thing to be able to do. But a photograph of a colony is not a colony. It does not forage. It does not adapt. It does not deposit new trails or let old ones fade. It holds, frozen, what the living substrate held at the instant the picture was taken.

This observation does not diminish the model. It locates it. The model sits in the same place the single ant sits — an agent of astonishing capability, holding a partial copy of something it did not create. The difference is only that this ant has memorized more of the trail map than any ant ever could. The intelligence it displays is real. It is just not the model's own. It belongs, as it always has, to the substrate — to the hundred billion speakers whose accumulated signals the model learned to read.

There is a quiet test that makes the point. Train a model on the accumulated text of a language, and it speaks that language with all the fluency the substrate accumulated. Train it on a language with few speakers and little written record — a language whose substrate is thin, whose trails were walked by thousands rather than billions — and the model is correspondingly thin. The model's capability tracks the richness of the substrate it photographed, not anything about the model itself. The same architecture, pointed at a deep substrate, is brilliant. Pointed at a shallow one, it is poor. The intelligence was never a property of the photographing. It was always a property of the thing photographed.

The reader will meet this observation again, when the book reaches its third movement and turns from the substrates that already exist to the one being built. For now it is enough to have noticed where the language lives.

Not in the speaker. Not in the model. In the substrate the colony of speakers has been depositing, marking, and letting fade, one sentence at a time, since before anyone alive was born — the same substrate the reader has been standing inside since the first word of this book, reading the trails, never once owning the meaning.