Writing Was Born Answering a Question

The story goes that in 1899 a scholar in Beijing fell ill, sent out for medicine, and found the beginning of Chinese writing in his prescription.

Wang Yirong had been Chancellor of the Imperial Academy three times over and was one of the empire's authorities on the inscriptions found on ancient bronze and stone — which is what makes the rest of the story possible, since he was among the few men alive who could have looked at what he was sent and known it. The apothecary sent him what apothecaries had been sending people for centuries: dragon bones. Old bones, dug up by farmers, ground to powder, taken for fever. Wang looked at his and saw marks on them. Not decoration. Characters — archaic, but recognizably the ancestors of the ones he had spent his life reading.

It is a very good story and it is probably too good. The illness in particular tends to arrive in later tellings and to change its name along the way, and Wang himself never published a word about the bones — the account comes to us through other men, after his death. There is a competing claim that puts the recognition a year earlier and in other hands.

What survives the doubt is duller and stranger. In the closing years of the nineteenth century, inscribed bones were reaching collectors in Beijing. Wang and his friend Liu E were among the first to understand that the marks were writing. Liu published the first printed collection in 1903. And the dealers, who had every commercial reason to lie about their source, were eventually traced to a village near Anyang in Henan, which turned out to be the last capital of the Shang.

And whatever the truth about the fever, this part is not in dispute. The bones really had been sold as medicine. For some unknown stretch of time, some unknown quantity of the oldest documents in East Asia was bought by the ounce and swallowed.

Here is what was written on them.

Not laws. Not poetry. Not a king's account of his own greatness, which is what a good deal of early writing elsewhere turns out to be. The oracle bones of Anyang carry questions.

A caution belongs here, immediately, because the obvious inference is one specialists have spent twenty years warning against. These are the oldest Chinese writings anyone can read. They are not necessarily the oldest there were.

The reason is sitting among the characters on the bones. One of them is the word for a document, and it is a picture: a bundle of bamboo strips, bound with a cord. The Shang were writing on bamboo. Bamboo rots. So what has come down to us is not what the Shang wrote — it is the fraction of what they wrote that happened to be cut into shell and bone, which are among the few things that survive three thousand years in the ground.

Robert Bagley has put the objection about as bluntly as it can be put: the belief that early Chinese writing was confined to the ritual setting where we first meet it has, he says, no basis. The script at Anyang is already mature. Mature systems are the end of a process, not the beginning of one, and the beginning is presumably on the bamboo, in the granary accounts, gone. Keightley answered him, and the two of them ended up agreeing that the system came together quickly under the late Shang state while continuing to disagree about what it came together for.

Which means the title of this essay is a claim I cannot fully make good on, and it would be cheap to let it stand without saying so.

The procedure was standardized and it was industrial. An ox shoulder blade or the underside of a turtle's shell was cleaned, sawn flat, and drilled with rows of hollows on the reverse. A hot point was applied to a hollow; the bone cracked on the other side with a sound the language may have taken its word from; and the crack was read. Then a scribe cut the record into the bone itself.

The record has parts. A preface: the date, and the name of the diviner. The charge — the matter put to the bone. On fewer of them, a prognostication, usually attributed to the king. And on a rare few, the showpieces Keightley called display inscriptions, a fourth element that stops you when you come to it.

A verification. What actually happened.

The most famous concerns a royal consort and whether her childbearing would be good. The verification records that after thirty-one days she gave birth, and that it was not good, because it was a girl.

Tens of thousands of pieces have come out of Anyang, and they are not the leavings of a few anxious kings. They are the working archive of a state department, staffed by named professionals, running a documented procedure at scale over generations, with outcomes set down beside forecasts.

And here the archive does something stranger than error-checking. The king, in his own records, is never wrong. The prognostications are hedged into branches, and prognostication and verification alike were carved after the event. This was not a check on the king. It was an institution with the form of accountability and none of the substance — not a Bronze Age failing, but a recognisable object, three thousand years old.

We do not usually let ourselves see this, because we have a filing category for it. These are beliefs — a word we mostly keep for other people's reasoning, and which does quiet work in stopping their machinery from looking like machinery.

But run the check anywhere and the same thing comes back. Not a mood. A specification.

In Yorubaland, Ifá is not a practice anyone picks up. It is a formal system of two hundred and fifty-six figures — sixteen principal odù, each carrying sixteen of its own — and every figure indexes a body of verse. Nobody holds all of it: the corpus runs to hundreds of verses per figure and keeps growing, so a diviner commands a working subset — deep in some figures, thin in others — and goes on learning until he dies. Wándé Abímbọ́lá, an initiated babaláwo and also the scholar who did most to make the corpus legible to outsiders, describes a training any conservatory would recognize: a master, years of it, a repertoire that is never finished. The diviner casts palm nuts or a divining chain, reads the figure that lands, and works from the verses it carries. UNESCO put the system on its heritage lists in the last twenty years, which is worth saying plainly for what it is — not the preservation of a memory but the recognition of a discipline practised now, by living people, this week, and taught at an institute in Ọ̀yọ́ founded to teach it.

One detail is worth pausing on, because a modern engineer would recognize it. At the opening of a consultation the client does not say the question aloud. He cups a coin to cover his lips, whispers the problem to it as voicelessly as he can, and sets it down; before the first cast the diviner touches his chain to the coin and asks Ifá whether he heard what was said to it.

Then the babaláwo casts and recites — not the verse that fits, but every verse he holds for that figure. The client listens for the one that bears on his own trouble and stops him there. The matching is the client's work.

Abimbola gives the reason plainly: the silence is there to keep bias out of the priest's mind when he comes to interpret. Bascom's informants at Ifẹ̀ put it harder — it was wrong for the diviner even to know, because knowing would tempt him to fake a verse. It is a norm, not a rule: a trusting client may ask outright.

A constraint on the interpreter, built in by the people it constrains, is not something a system arrives at by accident.

Tacitus, writing about the Germanic peoples around the end of the first century, gives a procedure so specific it reads like a manual.

A branch is cut from a fruit-bearing tree and divided into small slips. The slips are marked, and the marks are distinguished from one another, which is the detail that turns a handful of twigs into a system. They are scattered at random over a white cloth. Then, for a matter concerning the community, the priest of the tribe lifts them; for a private matter, the father of the household does it himself. He prays first. He looks up at the sky. And he lifts three, one at a time, without looking at what his hand is taking.

Then come the two clauses that most retellings stop just short of.

If the lots forbid the thing, there is no second consultation on the same question that day. And if they permit it, that is not yet the answer — confirmation by auspices is still required before anyone acts.

A procedure with a no-retry rule and a mandatory second check is not a mood. Somebody thought about how this could go wrong.

Tacitus has to be handled with gloves. Nothing attests that he set foot in Germania; he is working from Caesar, from a lost book of the elder Pliny, and from informants along the frontier. There is a reading of the Germania in which the free and uncorrupted German is a stick for beating a Rome the author finds soft — though it has never won the field, and it sits badly beside the same book's Germans, who are idle, drunk, and given to gambling away their own liberty on a last throw.

The text's later career is worse than its origins. German humanists took it up, then nationalism, and finally the Nazi state as evidence of an ancestral racial purity; in 1943 Himmler sent men into Italy to seize the manuscript, who failed only because its anti-fascist owners had hidden it in a chest. Momigliano called it one of the most dangerous books ever written.

None of which makes the procedure implausible. The practice is beyond doubt: Caesar describes Germanic lot-casting a century and a half earlier, too early to be copying anyone, and in one passage the lots are consulted three times over a captive, on whether to burn him at once or keep him. The number three is in both writers. The particular procedure, though, rests on Tacitus alone.

The right posture toward Tacitus is the one you would take toward any interested witness: useful, precise, not to be trusted about why.

In the Guatemalan highlands the instrument is a calendar, and the calendar is arithmetic.

The count of two hundred and sixty days — thirteen numbers turning against twenty named days — is the longest-running timekeeping system in the Americas. Its earliest securely dated instance is a day called 7 Deer, painted on a wall at San Bartolo in Guatemala somewhere between three hundred and two hundred years before the common era. It is still kept, and not by a handful of holdouts: by trained specialists in dozens of highland communities, initiated into a lineage and a system of shrines.

Barbara Tedlock's account of Momostenango is the one usually cited in English, and how she came by it is worth telling, because the polite version is false. She and her husband had been photographing people at the shrines without asking. She fell seriously ill. A daykeeper named Andrés Xiloj divined about the illness and, without being told what they had done, said they had been entering shrines without keeping the days, and that both of them would die. They left for the capital. Then they came back and asked what could be done, and the answer was an apprenticeship: two hundred and sixty days of it, the length of the count itself, with prohibitions attached that had to be kept exactly. At that point, she wrote, they stopped being ethnographers asking questions and became apprentices.

A word about how this essay knows what it knows. Its two closest accounts come from initiated insiders — a Yoruba practitioner who is also a professor, an anthropologist who trained as a daykeeper — which is exactly why they are good, and exactly the standing objection to them. Reviewing Tedlock in 1983, Benjamin Colby complained the initiation was asserted rather than evidenced: no transcripts, no account of how she reached her own readings. She answered in print the next year. Insider access buys depth and costs verifiability, and nothing gets both.

And the continuity is not a happy accident of isolation. It is the outcome of a fight. On the twelfth of July 1562, at Maní, the Franciscan Diego de Landa — not yet a bishop, then merely the head of his order in Yucatán — gathered the books of the Yucatec Maya and burned them. He wrote afterwards that the people had regretted it to an amazing degree, and that it had caused them much affliction. That is his own sentence, and the horror in it is that he noticed and wrote it down anyway. Of all the hieroglyphic books of the Maya, four are known to survive. The fourth was authenticated only in 2018 and turns out to be the oldest surviving book in the Americas.

But consider what Landa did and did not accomplish. He destroyed the books. He did not destroy the count — because the count was never in the books. It was in people.

It still is.

And then the one closest to us, which we mostly do not see because it is still in the newspaper. Astrology in the West was not a fringe. For most of two thousand years it was the establishment — taught in the universities, consulted by physicians, retained at courts, argued over by theologians precisely because it mattered. Bologna paid lecturers in it for centuries. One was burned for heresy in 1327; another, much later, had the young Copernicus for an assistant.

Peter Struck's study of the classical world shows philosophers treating the subject as a serious problem rather than a lapse in seriousness — and his reviewer Jennifer Larson objected on exactly the ground this essay stands on: the craft in technical divination cannot be folded into a faculty of insight, because it runs on large bodies of learned knowledge drilled over years. That is an objection to Struck. It is also a description of everything above.

Every culture on every continent built instruments for it.

Here is the thing that ought to be strange, and mostly is not remarked on.

None of these people were comparing notes.

Bronze Age Anyang and the Yoruba city-states and the German forests and the K'iche' highlands did not exchange designs. And yet lay the procedures side by side and the same parts keep turning up, in the same arrangement, like a machine that several unconnected workshops independently converged on.

There is a randomizer, and it is always something the practitioner does not control. Heat in a bone. Palm nuts. Strips thrown on cloth. The turning of a day-count nobody can hurry.

There is a bounded answer space, and it is always finite and enumerated. Two hundred and fifty-six figures. Sixty-four. Twenty day-names against thirteen numbers. A crack that goes one way or the other.

There is a corpus attached to each figure, and it is always large, memorized, and old — the part that takes years, and the part that cannot be improvised.

There is a designated interpreter, and who it will be is fixed before anyone asks — the priest for the community's business, the head of the household for the household's, the babalawo after a decade of memorization. Never simply whoever happened to feel like reading it.

There is, remarkably often, something done to keep the interpreter from steering. The Yoruba diviner casts before he is told the question. The Germanic officiant does not look at the slip he lifts. The daykeeper who diagnosed Barbara Tedlock was told nothing. Traditions with no contact between them each decided the reader should be denied something, and built the denial in.

There is a required form for the question, and this is the part the modern reader skims past. You cannot cast on a vague unease. On Keightley's reading — contested, and still argued over — the Shang charge is characteristically put as a proposition to be tested rather than as an open wondering. The Yoruba client must arrive with something definite enough to whisper. The procedure will not start until the asking has been made sharp.

And there is a public procedure, performed the same way each time, which everyone present can see was performed the same way each time.

That is not a belief system. That is a design pattern, arrived at independently, on four continents, by people who could not have told each other about it.

So what was the machine for.

That is contested, argued in the journals now, and deserves its own transmission rather than a paragraph here. The tempting answer — that these were early psychology and everyone secretly knew it — is condescending and unsupported. These people were not being ironic.

But one property of the design requires no theory at all, because it is a mechanical consequence of the parts.

Every one of these instruments makes it impossible to proceed until the question has been made exact.

That is what the bounded answer space is for, and the required form of the charge. A finite set of outcomes cannot take a vague input; the machine will not accept it. Before any of the arguable things happen — before the crack, before the verse, before the interpretation — the person asking has had to do something difficult: work out what they are actually asking.

Anyone who has watched someone circle a decision for a month knows that this is not the easy part. It is nearly the whole part.

Then, mostly, the instruments went.

The comfortable version is that they were disproved and set aside, and in places that is roughly true. Better machinery arrived for weather, harvest, illness, and it worked, and people used it. Nobody should want a hot point on a shoulder blade making the call on an infection.

But it was not everywhere a peaceful retirement. A great many of these systems did not fade: they were burned, banned, criminalized and preached against, by people who knew exactly what they were destroying and destroyed it for that reason. Four surviving codices is not what attrition looks like. It is what a policy looks like.

What is left is an asymmetry. We have inherited the greatest answering machinery ever built. It is magnificent, and nobody who has used it honestly wants it back. Ask it anything and it will tell you, at any hour, in any language, with a fluency that would have looked like sorcery to every civilization named above.

And we have kept almost nothing for the other half.

There is no discipline left whose whole function is to slow a person down at the point of asking. No trained figure whose job is to refuse the vague question. No procedure that will not begin until you have said, precisely, what you want to know.

We have the corpus and the interpreter and the astonishing speed. What fell away was the part that made you formulate the charge.

So the narrower claim is the one to make, and it is the more interesting anyway.

The oldest writing anyone can still read in East Asia is a record of people asking. That is not a metaphor; it is a description of a corpus of bones, catalogued and published, whose contents are the charges of a Bronze Age state and, on a rare few, a note of what happened next.

Perhaps that is where writing started. Perhaps it only tells us which writing somebody thought worth keeping. But look at what the second version says. Of everything the Shang were setting down — the accounts, the whole administrative traffic of a state, on strips long since rotted into the soil of Henan — what they cut into the imperishable material was the questions.

Somebody heated a bone to find out whether it was going to rain, and then, when it did not rain, went back and carved that down too.

For a while after that, the bones were medicine. Villagers gathered them out of the fields and sold them to apothecaries by the catty, and because buyers wanted clean bone, the inscribed ones mostly had the writing scraped off first. The questions were not lost by accident. They were removed as a blemish on the goods.