Seeing Ghosts and Running About

SERIES · Herbal Histories. Herbal Histories takes one claim about a drug plant — the kind that circulates in a hundred articles without a citation — and goes to the document it is supposed to rest on. Twenty-five issues, each built on primary sources read in the original: statutes, papyri, newspaper columns, parliamentary papers and manuscripts in Dutch, German, Spanish, Portuguese, Latin, Nahuatl, Sanskrit, Arabic, Chinese and Egyptian. Translations are the author's own and are marked as such. When a claim survives the document the issue says so; when it does not, the failure is the story.

The oldest object in the world that carries the most-quoted sentence in the history of Chinese cannabis was printed in Edo in 1854.

It is the last leaf of the first of three slim volumes, bound in the Japanese fashion with thread through four holes, its cover a flat vermilion. The blocks were cut for a private studio belonging to a physician named Mori Risshi, who came from Fukuyama and spent much of his working life on the textual criticism of Chinese medical classics. The leaf carries the end of the class of drugs called grains. Sesame, then hemp, then a colophon reading Bencao jing, volume one.

The hemp entry runs to thirty-eight characters:

The translation is mine, here and throughout, unless I say otherwise.

That is the sentence. It appears in almost every account of cannabis in China, and it is almost always introduced with a date: the Shennong bencao jing, the Divine Farmer’s classic of materia medica, compiled by the mythical emperor Shennong around 2700 BCE. The world’s oldest pharmacopoeia. The first written record of cannabis as a psychoactive drug.

Set the date against the object. Between the emperor and the leaf lie about four and a half thousand years, and the leaf is the near end of it.

Open the same volume at the front and Mori explains himself, in a preface facing a title leaf stamped with the red seals of the men who owned the book after him.

醫經從來，鮮有古本。……舊經之文竟併合于諸家書中，無復專本之能傳于後矣。……一朱墨之相錯，文字譌脫，亦復在所不免。

Medical classics have seldom come down to us in ancient copies. … The wording of the old classic ended up merged into the books of various authors, and no separate copy survived to be handed on. … With vermilion and black interleaved, corrupt and dropped characters were not to be avoided either.

He is describing a book that stopped existing — not destroyed in a fire or buried in a tomb, but absorbed. Somebody quoted it, and then somebody quoted the quotation, and at some point nobody was keeping the original any more because everything in it was already somewhere else.

The absorption can be dated fairly closely, and the dating undoes the ascription without any help from modern scholarship.

Start with Zheng Xuan, who died in 200 CE and who annotated the Rites of Zhou. The passage concerns the office of the jiyi, the physician for illnesses, whose charge is to nourish the sicknesses of the people 以五味、五穀、五藥 — with the five flavours, the five grains and the five drugs. Zheng glosses the five grains as 麻黍稷麥豆, hemp and broomcorn millet and foxtail millet and wheat and beans, and then adds a remark about compounding: 其治合之齊，則存乎神農、子儀之術云 — as for the proportions in which they are combined, that lies in the arts of Shennong and Ziyi.

So by the end of the second century “Shennong” is already the name of a body of pharmacological technique. It is not yet the name of a book.

Six centuries later the Tang commentator Jia Gongyan works over the same passage and reaches for the catalogues. He finds an entry in the Zhongjing bu, an early library register, for a Ziyi bencao jing in one scroll, and he notes the thing that matters: 並不說神農 — and it does not say it comes from Shennong.

That single clause is the whole problem in five characters, and it was written down by a Chinese classicist around the middle of the seventh century.

What happened between Zheng Xuan and Jia Gongyan is that somebody wrote the book. The consensus of specialists is that the compilation is a work of the Han, roughly the first or second century of our era, anonymous and layered — assembled out of material from several transmission lineages rather than composed by anyone. Joseph Needham and Lu Gwei-djen, in the botany volume of Science and Civilisation in China, discuss the text at pages 235 to 243 and describe precisely the situation Mori describes: a work that survived only through quotation in later pharmacopoeias, and that was reconstituted from those quotations by later editors. They name three of the reconstructors — Lu Fu in 1616, Mori Tateyuki in 1845, Liu Fu in 1942.

Mori Tateyuki is Mori Risshi. Needham dates his work to 1845; the volume in front of us was printed in 1854. Both dates are probably right, and the space between them — the years in which a scholar's manuscript becomes a printed object other people can own — is a small version of the thing this article is about.

The evidence for the layering is in the text itself, and it is unusually direct. Around the third century a physician named Wu Pu compiled a materia medica of which fragments survive in quotation, and his entries look like this:

Shennong is one authority among several, disagreeing with the others about a basic property of the drug. He is not the author of the book. He is a label on a column of a table — the name of one school's opinion, set beside Qibo's and Leigong's and Bianque's, in the way a modern pharmacopoeia might record that two laboratories reported different results.

That is what the word does in the title. Shennong bencao jing names the stratum, not the man.

Here is the mechanism that made reconstruction possible at all, and it is a matter of ink.

When Tao Hongjing produced his collected commentary around the year 500, he did two things at once: he preserved the older classic and he added a second body of material, the Mingyi bielu, the separate records of eminent physicians. To keep them apart he wrote the classic in red and the additions in black. The distinction was carried forward for five hundred years in manuscript. When the Song began printing pharmacopoeias, red was not available, and the printers solved it by reversing the type — the classic's characters were cut white against a black ground, so that the eye could still separate the two layers on a monochrome page.

You can watch a Song scholar using it. Su Song, compiling the illustrated materia medica of 1061, is trying to work out whether máfén is the flower of hemp or its fruit, and he reaches for the ink as evidence:

The vermilion characters say. He is citing a typographic convention as though it were a witness, because that is exactly what it was: the only surviving evidence of where the older stratum ended.

Now the ink starts to fail. Copyists mix the colours. Printers make errors. Editors disagree about which lines were red to begin with. This is the corruption Mori names in his preface, and it is what the reconstructors set out to reverse.

The most influential attempt was made in 1799 by Sun Xingyan, a formidable Qing philologist, working with his nephew Sun Fengyi. Their text was published in Sun Fengyi's collectanea, the Wenjingtang congshu, and it is the version that circulates most widely today — the one behind most modern printings and most digital editions.

The method is stated openly in the preface Shao Jinhan wrote for it:

Two moves. Sort by typography, then correct the typography by rule.

Sun spells the rule out himself, in a note attached to the very first drug in the book. Cinnabar, in the Daguan printing, is given as growing "in the mountain valleys of Fuling", and the whole phrase is set in black — which should mean it belongs to the later stratum. Sun overrules the page:

All that follows is treated on this model. He is not transcribing the evidence. He is emending it, systematically, on a hypothesis about how place-names accumulate.

Turn to the hemp entry and you can watch the rule bite. In the Song pharmacopoeia the hemp seed is said to grow 生太山川谷 — in the mountain valleys of Mount Tai. Mount Tai stands in the county of Fenggao, and Fenggao is on Sun's list. So in Sun's reconstruction the classic reads 生川谷, in river valleys, with nothing about Mount Tai; and Mount Tai is demoted to the Mingyi bielu, where Sun's apparatus duly records it — 生太山 — as somebody else's addition.

The hemp entry in the standard edition of the Shennong bencao jing is missing a mountain because a scholar in 1799 decided that mountains got added later.

He may well have been right. That is not the point. The point is that the sentence people quote as the voice of a Bronze Age emperor has, demonstrably, been through the hands of an editor with a theory.

And the editors do not agree with one another. Compare the two most reputable reconstructions, fifty-five years apart:

Sun Xingyan and Sun Fengyi, 1799: 麻賁。味辛，平。主五勞七傷，利五臟，下血，寒氣。多食，令人見鬼狂走。久服，通神明、輕身。一名麻勃。

Mori Risshi, 1854: 麻蕡，一名麻勃。味辛平。生川谷。治七傷，利五藏，下血寒氣。多食令人見鬼狂走。久服通神明輕身。

Sun has the five taxations and the seven injuries; Mori has only the seven injuries. Sun puts the alias at the end; Mori puts it at the head. Sun's classic does not say where the plant grows; Mori's does. And they write the drug's name with different characters — Sun with 賁, following the Shuowen dictionary, Mori with the 蕡 that everyone else uses.

Neither man had a manuscript. Neither was being careless. They simply made different decisions about the same pile of quotations, and the modern citation, which names no editor at all, silently picks one.

The other thing everybody knows about this entry is that máfén means the flowering tops — the resinous female inflorescence, the part that would in fact be psychoactive. That reading is what allows the sentence to be a report of intoxication.

It has been contested continuously for fourteen hundred years, and the man who reconstructed the standard text came down against it.

The classical philology is not really in doubt. The Erya, the oldest Chinese lexicon, glosses 蕡 as 枲實, the fruit of the hemp plant. The Yili, in the chapter on mourning garments, defines the coarsest grade of mourning band as 麻之有蕡者, hemp that has fén on it — that is, the female plant, the one that bears. Zheng Zhong, the elder Han commentator, states it flatly: 麻實曰蕡, the hemp fruit is called fén. And there is a line in the Huainanzi that settles what ordinary people understood by the word: 胡人見蕡，不知其可以為布 — the Hu see fén and do not know it can be made into cloth — on which the commentator Gao You notes, 蕡，麻實也, fén is the hemp fruit.

Against that stands Tao Hongjing, around 500, who wrote 麻蕡即牡麻，牡麻則無實 — máfén is the male hemp, and the male hemp bears no fruit. He then glossed the alias mábó as the fluffy stuff on top of the flowers, and left hemp seed as a separate entry.

In 659 the Tang imperial revision took him apart:

That last question is the sharpest instrument in this whole dispute, and it is an argument from filing. The entry does not sit among the herbs. It sits in the class called 米穀, rice and grains, in the superior grade. Whatever the compilers thought máfén was, they thought it was the sort of thing that belongs with sesame and millet.

By 1061 Su Song had read all of them and could not choose. He noted that hemp flowers are not a food, which favours the Tang view; noted that the vermilion characters give máfén and hemp seed different flavours, which makes them look like two substances; noted that a Tang pharmacological treatise separately describes hemp flower as bitter; and ended with a shrug that has never been improved on:

Five hundred years after that, Li Shizhen lost patience with the lot of them. In the Bencao gangmu, under a heading meaning "correcting errors", he wrote:

The most influential pharmacologist in Chinese history, working in the sixteenth century, declared the received text of this entry corrupt and rewrote it. His own conclusion about the word was that máfén 此當是麻子連殼者 — this must be hemp seed with the husk still on.

So the sequence runs: the oldest lexica say fruit; Tao says male plant; the Tang says fruit and calls Tao wrong; Su Song asks whether there are three substances; Li Shizhen says the text is corrupt and means husked seed.

And in 1799, at the bottom of his own hemp entry, having assembled the reconstruction that everyone now quotes, Sun Xingyan wrote his verdict:

The editor of the standard text of the Shennong bencao jing held that the word in the famous sentence means the seed.

Which leaves the sentence itself, and the sentence turns out not to be about hemp.

Open Sun's reconstruction at the herb section and read the entry for yúnshí, the seed of a thorny climbing legume:

Now the entry for làngdàngzǐ, henbane:

And hemp:

The same couplet, three times, in one book. Excess produces ghost-sight and frantic movement; sustained use produces lightness of body and communication with the spirits. It is a template. The compilers reach for it when they want to place a drug in a particular category — potent, double-edged, and belonging to the borderland between medicine and the arts of transcendence — and they apply it to plants with nothing pharmacological in common. Henbane is loaded with tropane alkaloids and will produce delirium. The legume will not.

This does not read as anybody's report of what happened when they ate too much hemp. It reads as a classificatory formula, and once you have seen it in three places you cannot unsee it in the fourth.

Two further features of the book confirm the register. Kuángzǒu, running about frantically, is not only an effect in the Bencao jing; it is a disease. Dragon's teeth are prescribed for fits, madness and kuángzǒu in children and adults. So are the hooves and hair of the six domestic animals, alongside ghost infixion and gǔ poisoning. And ghosts are everywhere: the collection is full of drugs that kill spirit-things, expel malignant ghosts, and treat conditions whose names are built out of the word. The preface itself instructs the physician to treat ghost infixion and gǔ poisoning with toxic drugs. In this system ghosts are causes of illness, and seeing them is a symptom with a place in the nosology.

The clearest proof that 見鬼 belongs to a technical register rather than a recreational one is that people went looking for it deliberately. In the Tang, Meng Shen's dietetic materia medica gives a recipe, preserved in the Song pharmacopoeia and reprinted by Li Shizhen directly beneath the famous sentence:

A hundred days, facing the sun, in a compound of three drugs. Whatever that is, it is not a description of getting high; it is a technique, with a schedule and a ritual orientation, and hemp seed is one ingredient of three.

The same is true of the other place hemp touches the visionary. Tao Hongjing noted that adepts compound mábó with ginseng 令逆知未來事, to make a man know future events in advance. The Bencao gangmu preserves the formula from an older prescription collection — hemp flower gathered on the seventh day of the seventh month, two ounces of ginseng, powdered and steamed, a spoon-tip at bedtime — and describes what it is for. It is a remedy for forgetfulness. Then Li Shizhen adds his own judgement, and it is the driest line in the whole entry:

A working pharmacologist in the 1590s, looking at an exaggerated claim about hemp's power over the mind, and calmly marking it down.

None of this makes hemp marginal. It makes it ordinary, which is a different thing and easier to lose.

Hemp is one of the five grains. Zheng Xuan's gloss puts it first in the list, and the physician for illnesses in the Rites of Zhou is charged with using the five grains as part of treatment. It is filed among the grains in every major pharmacopoeia from Tao Hongjing to Li Shizhen. Su Gong's argument in 659 works precisely because that filing was uncontroversial: nobody needed persuading that hemp belonged with food.

It was also the cloth. The Huainanzi line about the Hu who see hemp fruit and do not know it can be woven only makes sense in a world where everyone the author expected to read him did know. Mourning garments were graded by hemp: the coarsest band was made from the female plant, and the ritual codes specify which. When Su Song describes hemp cultivation in 1061 he describes farmers choosing speckled seed to get a heavier set, and stripping bark for cloth. The Bencao gangmu records the practice of pulling the male plants once they have shed pollen, which is both a piece of agronomy and, incidentally, a demonstration that Chinese farmers understood the plant's dioecy perfectly well.

The medicine, when you look at what was actually prescribed, is overwhelmingly the seed. Hemp seed as a laxative, in Zhang Ji's pill for constrained spleen. Hemp seed boiled into gruel for the elderly, for difficult urination, for post-partum bleeding. Hemp seed oil for hair. The root decocted for retained placenta. In the whole vast prescription apparatus that Li Shizhen assembles under great hemp, the flowering tops appear rarely, and when they do they are treating wind conditions, menstrual blockage and wounds, not producing visions.

There is a Ming prescription for intermittent fever using hemp leaves that ends 其狀如醉，醒即愈 — the patient appears drunk; on waking he is cured. That is about as close as this literature comes to describing intoxication as a therapeutic effect, and it is a Ming addendum, not a Han classic.

The economic plant and the visionary plant are not the same story, and the popular accounts fuse them because a fibre crop makes a poor opening paragraph.

Which brings us to the surgeon.

The claim is repeated everywhere: Hua Tuo, the physician of the late second century, performed abdominal surgery under general anaesthesia using a preparation called máfèisàn, cannabis-boil powder, whose active ingredient was cannabis. The first general anaesthetic in history, sixteen hundred years before ether.

Here is the entire early evidence. Chen Shou's Records of the Three States, compiled around 289 CE, in the chapter on men of technique:

And Fan Ye's History of the Later Han, completed around 445 CE:

That is all of it. Both sources name the powder. Neither gives a single ingredient. There is no recipe for máfèisàn in any early source — not in the histories, not in the pharmacopoeias, not in the prescription collections. Every cannabis identification is a modern conjecture reasoning backwards from the character 麻.

The character will not carry it. Hemp is one thing 麻 means. The other, in medical prose, has to do with numbness — but that sense is well attested only much later, and the compound 麻沸 was doing different work in Hua Tuo's own century. Zhang Ji's Treatise on Cold Damage, the great clinical classic of about the same generation, uses it repeatedly, and it is not a drug:

The two ingredients are rhubarb and coptis. Máfèi tāng is what they are steeped in. It is water — water at a particular stage of boiling, when the bubbles come up thick. So in the surviving medical Chinese of Hua Tuo's lifetime, 麻沸 describes the state of hot water, and the name máfèisàn is at least as easily read as a powder taken with such water as a powder made of hemp.

There is one more piece of evidence, and it is the quietest and the most damaging. Wu Pu — the man whose materia medica preserves the oldest surviving stratum of opinion about hemp, the man who lists Shennong's view against Qibo's and Leigong's — was Hua Tuo's own student. The histories say so. His hemp entry survives in quotation. It records flavours, toxicity, incompatibilities, and a warning that the leaves will kill you.

It says nothing whatever about surgery, or sleep, or the absence of pain.

So: what can actually be said.

A Han compilation of the first or second century, anonymous, assembled out of several lineages of opinion, one of which was labelled with the name of a culture hero. In it, hemp is filed among the grains, and its entry carries a formula that the same book applies to at least two other plants, marking the drug as potent and belonging to the borderland where medicine touches the arts of transcendence. What part of the plant the entry means was already disputed by 500, was ruled on by the Tang in 659, was called three possible things by Su Song in 1061, was declared corrupt by Li Shizhen in 1590, and was decided against the flowering-tops reading by the very man who assembled the modern text in 1799.

The book itself stopped existing as a separate object somewhere between the sixth and the eighth centuries. What we have is a reconstruction, and there is more than one of them, and they differ in the wording of the sentence in question.

That last leaf in Edo was printed in 1854, three years before the Indian Rebellion and eleven years after O'Shaughnessy's hemp paper was reprinted in London. Mori set the blocks, ruled the empty columns at the end, and closed the volume. Somewhere in the century and a half after that, the date attached to the sentence migrated backwards past every one of the men who made it — past Mori and Sun and Li Shizhen and Su Song and Su Gong and Tao Hongjing and Wu Pu — and came to rest on an emperor who did not write it, in a millennium from which no Chinese writing survives.

The compression has a direction. It always ends with fewer people in it.

Translations. All renderings from Chinese in this article are mine. Where a phrase is disputed I have given the Chinese so a reader can check me.

Texts used, and in which state. The Shennong bencao jing is quoted from two named reconstructions, never as an undifferentiated "original": Sun Xingyan and Sun Fengyi's of 1799, and Mori Risshi's as printed in 1854. Their divergences at the hemp entry, at the entry for cinnabar, and in Sun's editorial notes were established by reading the two texts against each other directly. A third widely circulated received text, which reads 主治七傷 and attaches the alias to hemp seed, was also consulted; it agrees with Mori against Sun on the seven injuries and with neither on the placement of 生川谷. The parallel entries for yúnshí and làngdàngzǐ that establish the ghost formula were checked in both Sun's text and the third received text, and read the same in each.

The Zhenglei bencao is quoted from the Siku quanshu recension. That copy is a hand-written eighteenth-century transcription and does not reproduce the reversed-type distinction between the classic and the Mingyi bielu; my account of that typography rests on Su Song's own reference to "the vermilion characters", on Shao Jinhan's description of "the black and white characters of the Daguan bencao", and on Sun's instruction to move a phrase from black to white. I did not inspect a Song or Ming printed copy showing the reversal, and no image of one appears here. That is the largest outstanding gap in this article's evidence.

The Bencao gangmu is quoted from the Jinling first edition, dated by the National Diet Library to 1590, both from its text and from the digitised leaves reproduced as Figs. 3 and 4. The Taiping yulan is quoted from the Siku recension of juan 995, where the characters for hemp and for ghost are rendered as unmapped variant glyphs; my reading of that passage is therefore an inference from context, though a secure one, and the differences I report between it and Sun's text — the absence of the five taxations, "brings down blood qi" for "blood and cold qi", the reversed order of the closing clause, and the presence of Mount Tai — do not depend on the two doubtful glyphs. A Sibu congkan witness appears to read the line without the ghost character; I could not retrieve that page and do not rely on it.

The Hua Tuo passages are quoted in full from the Sanguozhi and the Hou Hanshu, and the negative claim is a direct observation: neither text gives an ingredient. I searched both biographies. The máfèi tāng parallel is from the Shanghan lun, where the compound unambiguously names a liquid the drugs are steeped in. That establishes what 麻沸 could mean in period medical prose; it does not prove what máfèisàn contained, and nothing here does. The composition is unrecorded.

Specialist scholarship. Joseph Needham and Lu Gwei-djen, Science and Civilisation in China, vol. VI part 1, Botany (Cambridge: Cambridge University Press, 1986), treat the Shennong bencao jing at pp. 235–243 and describe it as surviving only through quotation and reconstituted from those quotations, naming Lu Fu (1616), Mori Tateyuki (1845) and Liu Fu (1942). Needham's hemp discussion, at pp. 170–176 of the same volume, is taxonomic and is not cited here for anything pharmacological; a full-text search of the volume found no occurrence of máfén. Paul U. Unschuld, Medicine in China: A History of Pharmaceutics (Berkeley: University of California Press, 1986), and Miranda Brown, The Art of Medicine in Early China (Cambridge: Cambridge University Press, 2015), are the standard treatments of the pharmacopoeial tradition and of the retrospective construction of medical ancestry respectively; I did not re-verify their pagination in preparing this piece and have therefore loaded no specific claim onto either. The argument above rests on the primary witnesses, which in this instance are unusually forthcoming: the Qing and Edo editors documented their own methods in their own prefaces.

Hui-Lin Li's "An Archaeological and Historical Account of Cannabis in China", Economic Botany 28:4 (1974), is the source from which much of the repeated dating descends. I have treated it as an object in the history of this claim rather than as an authority, and I have not relied on it. Chen Yinke's 1930 argument that Hua Tuo's surgical legend derives from Indian Buddhist narrative is relevant and would strengthen section VI; I could not verify its citation directly and have therefore left it out rather than repeat it.

Not established here. Whether the Han compilers thought máfén was the flower, the fruit, or something else. What máfèisàn contained. Whether anyone in Han China used hemp for intoxication. The archaeological record of hemp fibre and textile in early China, which is real and substantial and which I could not document to this series' standard within the time available; readers should not take its absence here as evidence against it.

Not claimed. That the Shennong bencao jing is late, or fraudulent, or worthless. It is a Han text, it is early, and it is the foundation of a pharmacological tradition. What it is not is a book by an emperor from 2700 BCE, and the sentence about ghosts is not a report about cannabis.

The point of separating these columns is that popular accounts collapse them into one date.

Translations here are the author’s own and are marked as such. Sources are reproduced rather than linked. Claims that failed are printed with the verdict on them.

Subscribe — complete in the inbox.