The Voynich manuscript
In 1912 a book dealer bought a manuscript nobody could read. It had already defeated the scholars at the court of a Holy Roman Emperor, and Athanasius Kircher, the Jesuit who claimed he could read Egyptian hieroglyphs. It went on to defeat the American codebreakers who broke Japan's wartime ciphers, and the computers that came after them. This is the whole story: the book, the suspects, the failures, and the paper that says we have been turning its pages the wrong way round.
Prologue · Look
Before the theories, before the numbers, look at the thing. Almost everything that makes this book strange is visible on folio 1r, if you know where to look.
Open the book and this is what you see. A page of calfskin, worn at the edges, holed by insects, with four paragraphs of neat, fluent writing in an alphabet that exists nowhere else. The letters have the easy loops of someone who has written them a thousand times. The lines run left to right, the paragraphs have a straight left edge and a ragged right one, and the page looks, in every way but one, like a thousand other 15th-century books. The one way is that nobody on earth can read a word of it.66
Tap a number, or its note, to zoom in on that spot. Tap again to see the whole page.
Folio 1r. Beinecke MS 408, Yale, public domain.52
Here is what this page is not. It is not a cipher anyone has broken, though the best in the world have tried. It is not a known language in disguise, though a dozen have been proposed. It is not, as far as the evidence goes, a fake. And it is not alone: 200-odd pages follow, with plants, stars, naked women in green water, and a fold-out map of nine circles.
What follows walks through all of it, in the order a careful reader would want: the physical book first, then where it has been, then the pictures, then the text and the strange statistics that make it unlike any language, then who wrote it and this year's claim that its pages were never meant to be bound this way, then the century of people who said they had solved it. The last chapter is a ledger: what is established, what is argued, and what is guessed.
I · The object
Forget the legends for a moment. Start with the thing itself: calfskin, iron-gall ink, a few cheap pigments, and about 240 pages in a limp vellum cover that is not even its own.
The Voynich manuscript lives in a vault at Yale's Beinecke Rare Book and Manuscript Library, where its catalogue entry calls it, flatly, "Cipher manuscript", shelfmark Beinecke MS 408.3 It is not large. Each leaf measures about 225 by 160 millimetres, roughly the size of a paperback, and the whole block is about five centimetres thick.35
Somebody numbered its leaves 1 to 116, probably in the 16th century. Only 102 of them survive. Fourteen have been lost, some clearly cut out, leaving stubs in the binding.45 Ten of the surviving leaves are fold-outs; one, the famous "rosettes" sheet, unfolds to six panels.4 That is why nobody can agree on a page count: Yale's press office says 234, others say around 240, and it depends on how you count the folds.18
The leaves are calfskin, settled in 2014 when a team at the University of York tested the protein in the skin.5 Yale reckons it took 14 or 15 calves.18 In 2009, for an Austrian television documentary, a laboratory in Illinois examined the ink and paint. The writing and the outlines of the drawings use one ordinary iron-gall ink. The blue is ground azurite, the greens are copper compounds, the red-brown is red ochre, and the binder is probably gum arabic. Nothing in it is out of place for the 15th century.859 There is, oddly, almost no yellow: René Zandbergen, who has studied the book for three decades, thinks a plant-based yellow has simply faded away.39
The cover is a later goat-skin wrapper from the 18th or 19th century. Insect holes in the first and last leaves show the book once had wooden boards.443
The same documentary paid for the test that changed everything. Greg Hodgins at the University of Arizona cut four slivers of parchment, each about a millimetre by six, from four different leaves, and dated them by radiocarbon. All four agreed. The calves died between 1404 and 1438, with 95% probability.67
Two cautions, both from the people who did the work. First, the date is for the skin, not the ink: carbon dating cannot say when someone wrote on it, only that the parchment came first.6 Second, the materials and handwriting are all consistent with the early 15th century, so the simplest reading is that the book was made then.820 But a determined sceptic can still say: someone could have found blank old vellum later. Keep that in mind when we get to the hoax theories. (The Beinecke's own web page about its star exhibit, for what it is worth, still carries its pre-2009 guess of the late 15th or 16th century.58)
The date does more than fix an age. It is a test that every proposed author has to pass. One famous name fails it outright; the rest have to explain how they came by century-old blank calfskin.
The book falls into sections that scholars named long ago from the pictures, not from any reading of the text. Each tile is a page, in the order the book is bound today, with the lost leaves left as gaps and the panels of fold-out sheets marked. Hover over or tap the tiles. Then switch to see two things you cannot see by flicking through the real book: the two statistical "languages" that Prescott Currier, a US Navy codebreaker, found in 1976,21 and the five scribes that Lisa Fagin Davis identified by their handwriting in 2020.20 Notice how the languages and the scribes line up with each other, and with the sections.
II · The journey
The book was made around 1420. The first person we can prove owned it signed his name around 1610. What happened in between, nobody knows.
Most of what the public "knows" about the manuscript's past comes from one paragraph in one letter. In August 1665 a Prague physician, Johannes Marcus Marci, sent the book to Athanasius Kircher, the Jesuit polymath in Rome who had claimed to read Egyptian hieroglyphs. Marci's covering letter says that a friend, now dead, had told him the book once belonged to Emperor Rudolf II, who paid its bearer 600 ducats and believed it to be the work of Roger Bacon.911 That is the whole of the evidence for Rudolf. No inventory of his famous collection lists it.13 Yale's catalogue still states the Rudolf story as fact; Zandbergen, more carefully, calls the informant credible but the claim unproven.313
The letter is itself a small mystery. Yale's catalogue dates it "1666 [or 1665?]"; the last digit is cramped. But a second letter, written to Kircher in January 1666 by another Prague scholar, already asks whether he has made any progress with the book, so 1665 it must be.1112 Kircher, who had an answer for everything, never wrote a word about it.
The calves whose skins became the book are alive. The parchment is made, and (most likely) soon written on.6 Measured
Nothing. Not a catalogue entry, not a mention, not a copy. Someone, somewhere, kept it.
Court records show Rudolf II buying books from an Augsburg physician, Karl Widemann, for 600 Rhenish gulden. The sum echoes Marci's 600 ducats. The record does not name the book. Hypothesis9
Jacobus Horčický de Tepenec, Rudolf's chemist and distiller, writes his name at the foot of the first page. He was granted the title "de Tepenec" in October 1608 and died in September 1622, so the signature must have been written between those dates. Someone later rubbed it out. It is the earliest proof of ownership. Documented1519
Georg Baresch, a Prague alchemist, writes to Kircher about "a certain old book" he cannot read, full of "pictures of herbs, of which there are a great many in the codex, and of varied images, stars and other things". He had already sent Kircher hand-copied samples. This letter, found by Zandbergen in the Jesuit archives in the 1990s, is the first certain record of the manuscript. Documented10
The book sits in Jesuit hands in Rome. When the Italian state seized church property in 1873, the Jesuits hid their private manuscripts. Exactly where is disputed. Tradition says the Villa Mondragone, a Jesuit college at Frascati in the hills south-east of Rome; Zandbergen's 2023 reading of the archives says another Jesuit house, the Villa Torlonia at Castel Gandolfo, a few miles away.914
Voynich tells an audience in Philadelphia that the books "had lain buried in the chests in which I found them in an ancient castle in Southern Europe". On the same stage, the philosopher William Romaine Newbold announces that he has read it. The author, he says, was the 13th-century friar Roger Bacon, writing in a microscopic cipher and describing what he had seen through a microscope, sperm cells included. Refuted 19311417
Kraus gives it to Yale. It becomes Beinecke MS 408. Documented3
Yale University Press publishes a full-size facsimile. The Beinecke says the manuscript's pages draw about half of all traffic to its online image viewer, and that its curator gets at least three emails a week from people who have cracked it.18 The next year a Spanish publisher begins selling 898 exact replicas, stains and wormholes included, at about €7,000 each, and a symphony named for the book has its premiere in New Haven. Documented5354
Lisa Fagin Davis publishes multispectral images taken in 2014. On the first page, three faint columns of letters appear beside the text: a Latin alphabet, a column of Voynich letters, and a second Latin alphabet shifted by one. Somebody was testing, letter by letter, whether the script was a simple substitute for the Latin alphabet. Davis argues from the handwriting that it was Marci himself, in the 1660s, before he sent the book to Rome. It did not work then either. Documented19
Colin Layfield and Lisa Fagin Davis publish "Singulion Structure and the Voynich Manuscript". Chapter V, below. New, contested2
Voynich is a character a novelist would reject as too much. A Polish revolutionary jailed by the Russians, exiled to Siberia, escaped in 1890 and reached London that October with nothing.15 Within a decade he had a rare-book shop on Soho Square and a reputation for finding things nobody else could. He married Ethel Boole, whose novel about an Italian revolutionary sold millions in the Soviet Union.16
He never said where he found the book. "I am bound by promises made to those from whom I acquired it," he wrote in 1917.14 His widow, Ethel, left a sealed letter to be opened after her death. Opened in 1960, it said the books had come from the Jesuits at Frascati, in the hills outside Rome.9 He also, around 1914, dabbed chemicals on the erased signature to try to read it, and left stains that hid it for a century.19
Did he fake it? Chapter VI takes that up. The short answer is that the carbon date, the medieval ink, the wormholes that run through the writing and the 17th-century letters describing the same book make that very hard to sustain.6810
III · The pictures
Almost every page is illustrated. The drawings are the only part of the book anyone can "read", and they raise as many questions as the text.
About half the book is plants: one per page, usually, with root, stem, leaf and flower drawn in ink and washed with green, blue or brown, and paragraphs of text wrapped around them.36 They look like a herbal, the standard medieval reference book of medicinal plants. The trouble is that almost none of them can be identified. Some seem to be composites, the root of one plant on the stem of another.36 A few have plausible matches: the plant on folio 9v has been called a wild pansy since the 1930s, and Zandbergen tentatively reads folio 2v as a water lily.4546 Folio 33v was called a sunflower by a botanist in 1944. Sunflowers reached Europe from the Americas after 1492, so that would make the book at least 60 years younger than its parchment. The carbon date says otherwise, and so do most botanists.45
Beinecke MS 408, Yale, public domain.52
After the plants come circular diagrams of suns, moons and stars, including a cluster that looks very like the Pleiades.36 Then a zodiac, and a strange one. It starts with Pisces rather than Aries. Aries and Taurus are each drawn twice, on two pages, with 15 figures apiece rather than 30. Capricorn and Aquarius are missing, almost certainly cut out with folio 74.542 Around each sign stand rings of small naked women, each holding a star, each with a label of Voynich text: 30 to a sign, which looks a lot like the 30 degrees of a zodiac sign.40 In the centre of each, a later hand has written the month in a Romance language: "mars", "aberil", "may", "octembre". Zandbergen reads it as northern French; others have argued for Occitan.6641
Signs, folios, figure counts and the month names follow Zandbergen's page descriptions.404142 Images: Beinecke MS 408, Yale, public domain.52
Sagittarius is the page that scholars keep coming back to, because it is drawn not as a centaur but as a man with a crossbow. The weapon's shape and the man's clothes are the sort of thing an art historian can date. A 2003 study of the crossbow put it in the first half of the 14th century; Zandbergen finds its closest parallels in German manuscripts of 1400 to 1500.5739 Either way, the picture and the carbon date are not far apart.
Then the section that gets the manuscript into the tabloids. Across twenty pages, naked women stand, sit and wade in pools of green and blue, connected by pipes, tubes and channels that run from one drawing to the next and sometimes off the page altogether.37 Scholars have called it "biological" or "balneological", the study of baths. Medieval books about hot springs, such as De balneis Puteolanis, a 13th-century poem on the baths near Naples, show bathers in pools fed by pipes, and the resemblance has been noticed often.36 Others see the tubes as anatomy, or allegory. Nobody knows. A small animal in the margin of folio 80v has been called an armadillo, which would be awkward, since armadillos live only in the Americas; it has also been called a pangolin and a lamb.61 Nobody has published a firm count of the figures. A tally by one researcher on the main Voynich discussion forum puts the human figures in the whole book above 500, of whom about 50 are clothed and perhaps 50 to 60 are men.68
Folio 78r. Beinecke MS 408, Yale, public domain.52
One detail in this section turns out to matter enormously. On the back of folio 78, two channels of water leave a tub and run into the gutter of the book. They reappear on the front of folio 81. Between them, as the book is bound, lie four pages of something else.37 Here are the two pages side by side, as they are not in the book. Hold that thought until Chapter V.
The largest sheet in the book unfolds to six panels and carries the strangest drawing: nine circles in a three-by-three grid, joined by causeways and tubes, with a castle, towers, a thing like a volcano or a cone of scales, and what may be the sun and moon.3860 Is it a map? A cosmos? A plan of a city? The one hard clue is architectural. The castle's battlements have swallow-tail merlons, the V-notched tops that belong to northern Italy and the Alps in the 15th century, though they also turn up in southern Germany, Switzerland and Austria.3839 It is one reason Zandbergen's best guess for the book's origin is "Alpine". Drag, pinch or use the buttons to explore.
Folios 85v to 86r. Beinecke MS 408, Yale, public domain.52
Two more sections close the book. In the "pharmaceutical" pages, rows of plant parts, roots and leaves, are laid out beside ornate jars, as if a herbal had been cut up and sorted: 57 rows of herbs and 241 fragments, by one count.36 Then come the "recipes", 23 pages of nothing but text, broken into more than 300 short paragraphs, each marked with a star in the margin, 324 stars in all.543 If the book really is a book of remedies, this is where you would expect the instructions.
And on the very last page, folio 116v, four lines in a different script altogether: a mix of Latin letters, German-looking words and a few Voynich letters that nobody has read either. The first word is widely read as "poxleber", an old German word for goat's liver, and the last line seems to end "so nim geis milch", "so take goat's milk", a reading proposed in the 1930s. Crosses between the words look like those in short charms. A remedy? A pen trial? A later owner's note? The 2024 multispectral scans made it clearer and no more legible.4319
IV · The script
The letters are easy to copy and impossible to read. The strange part is not that we cannot translate the text. It is that, statistically, it does things no known writing does.
The script runs left to right and top to bottom, in paragraphs with a straight left edge, as if written by someone fluent in it. Corrections are so rare that Voynich's secretary, Anne Nill, after years with the book, doubted it contained any.66 Scholars disagree on how many distinct letters there are, because they disagree on where one letter ends and the next begins. Two Yale linguists, Claire Bowern and Luke Lindemann, count 21 or 22 common signs plus about as many rare ones; Lisa Fagin Davis, whose handwriting study comes up in Chapter V, puts the basic set at around thirty.2320
To talk about it at all, researchers transliterate it: each sign gets a Latin letter. The standard system is the European Voynich Alphabet, or EVA, from the 1990s.47 It is important to say what EVA is not. The letter "a" in EVA does not mean the sign sounds like "a". It is a label, chosen because the sign looks a bit like an "a". When you see "daiin" written out, that is a transliteration, not a reading.
Here are the nineteen signs that do most of the work, grouped by how they behave: a few vowel-like letters, the two "bench" shapes, the strokes that pile up before a word's end, the letters that close a word, the "q" that almost only opens one, and the four tall "gallows". Tap a letter. The shapes are from a font drawn by the pathologist Gabriel Landini, who made it from the manuscript's own hand in the 1990s.49
Here is how folio 1r begins, in Landini's font and in EVA. Hover over a word to match the two.47
Type EVA letters (a to z, and spaces) and see them in the script. This is a party trick, not a translation: it only shows what the signs look like.
For this page we took the standard transliteration of the whole manuscript, about 34,000 words of running text in some 7,400 distinct forms, and ran the same simple counts on it and on three ordinary texts: an English novel, an Italian poem and a Latin history.4748 These are the findings that every theory has to explain, and most of them were first noticed by the American codebreakers who worked on the book after the Second World War.2221
Voynichese words are all much the same length. Two out of three have four, five or six letters, and words of ten letters or more barely exist. English and Latin have long tails of long words. This narrow, hump-shaped distribution is one of the oldest puzzles, and one of the arguments that the "words" are not words at all, or are built from parts according to rules.2322
This one is the heart of the problem. If you know one letter, how hard is it to guess the next? In English, Italian or Latin, not very: roughly three bits of uncertainty. In Voynichese, about two, a huge gap. The next letter is far too easy to predict. "q" is followed by "o" 98% of the time; "i" runs are followed by "n"; "d" likes "y".4723 A language that predictable would carry very little information per letter, which is why some researchers suspect a cipher that pads the text, a shorthand, a constructed language, or nonsense.2326 The Hebrew-by-computer claim of 2016, one of the ledger entries in Chapter VI, never explained this.35 Bowern and Lindemann tested the obvious fixes, such as dropping vowels or using shorthand, and found none of them pushes a real language this low; the only thing that does is to merge sounds that the language keeps apart.23
In the three texts we tested, a word repeats itself straight away less than once in 2,000 words. In Voynichese it happens about once in 120, and near-repeats, where only one letter changes, are commoner still: "qokedy qokedy", "chedy qokedy shedy".47 How damning that is depends on what you compare it with. Bowern and Lindemann, using samples from hundreds of languages, found the Voynich rate high but "within the realm of plausibility". Torsten Timm and Andreas Schinner, who argue the text is meaningless, replied that those samples were badly cleaned.2326 In 2020 those two built a model in which a scribe writes each new word by copying a word nearby and tweaking it, and showed it reproduces this and many other statistics. Their conclusion: the text may be meaningless, generated rather than written.26 Others read the same facts as a sign of inflected words, or a verbose cipher.23
In any real text the commonest word is about twice as common as the second, three times the third, and so on. The result is a straight line on a log-log plot, called Zipf's law. Voynichese obeys it. In 2013 two physicists went further and showed that Voynich words cluster by section, the way vocabulary clusters by topic in real books: "herbal" words stay in the herbal pages.24 The hoax camp replies that cheap generating processes can mimic both.2625
Prescott Currier's 1976 discovery, the "two languages", is visible in a single chart. The commonest word in the herbal pages, "daiin", drops to a third of its frequency in the bathing and star pages, and whole families of words ("chedy", "shedy", "qokeedy") that dominate language B are all but absent from language A.21 Currier himself was careful: he did not claim two real languages, only two statistically different kinds of text.21 They might be two dialects, two subjects, two cipher keys, or two scribes with different habits. Lisa Fagin Davis's handwriting study, in the next chapter, found that the scribes and the languages line up closely, which makes the last of those harder to dismiss.20
Finally, the oddest thing of all. In an ordinary text, a line break is an accident of the page; the words do not know it is there. In the Voynich manuscript they do. The first word of a line starts with "p", "t" or "y" far more often than any other word does, and starts with "o" or the bench letters far less. Some 85% of paragraphs open with a gallows letter, which has led Bowern and Lindemann to suggest the gallows there are not part of the word at all, just a mark that a paragraph is starting.2322 Whatever produced this text was thinking in lines, which is not how anyone writes language, but is how someone might encipher it, decorate it, or make it up.
Analysts since the 1960s have described Voynich words as if they were built from slots: an optional prefix, a start, a middle, and an ending.2223 This toy follows a simplified version of those rules, with the slot contents weighted roughly by how common they are. Press the button and it will tell you whether the word you built actually occurs in the manuscript.
V · The hands
The "mad genius in an attic" theory needs one author. The handwriting says there were five. And a paper published this summer says the sheets they wrote were never meant to be nested into a book at all.
Lisa Fagin Davis is executive director of the Medieval Academy of America and a palaeographer, a specialist in old handwriting. In 2020 she did to the Voynich manuscript what palaeographers do to any medieval book: she compared the shapes of individual letters across every page. She found five scribes. Scribe 1 wrote 113 of the 227 pages, including most of the herbal. Scribe 2 wrote the whole bathing section and a second batch of herbal pages. Scribes 3, 4 and 5 took the rest. And the scribes track Currier's languages: scribe 1's pages are language A, scribe 2's are language B.20
That matters for every theory. A hoax needs five conspirators, or one forger faking five hands. A cipher needs five people who all knew the system. A natural language needs a community of people who could write it. "The 'lone genius' idea is kind of out of the door," Davis told the Guardian this week.1
Then, in July 2026, after presenting the method at a Yale linguistics seminar the previous autumn,56 Davis and Colin Layfield, a computer scientist at the University of Malta, published a paper in the journal Digital Medievalist with a dull title and a startling claim: "Singulion Structure and the Voynich Manuscript".2 To follow it you need one piece of bookbinding.
A medieval book is made of sheets folded once, called bifolia. Normally a scribe nests several folded sheets inside one another to make a gathering, or quire, sews through the fold, and stacks the quires. A quire of four sheets is a quaternion; one of five is a quinion; and a "quire" of a single sheet is a singulion.2 The arrangement decides which pages end up next to which. Try it.
Layfield and Davis argue that the Voynich sheets were written as singulions, each one a little four-page booklet complete in itself, and only later nested into quires by someone who did not know the order, and bound.2 They have three kinds of evidence.
They then tried to work out an original order. For the recipes section, their best reordering raised the similarity between facing pages by 20%.2
It does not decode anything. The authors say so. What it does is explain why the text "flows" so badly from page to page, and it changes the picture of who made the book: not one author in one sitting but a group, producing booklets, "perhaps a set of booklets through which information was passed down from one generation to the next", as Davis put it to the Guardian.1 She has said the only other medieval manuscripts she knows that were kept as single sheets are north African Qur'ans made as teaching booklets; that comparison is hers in interview and is not in the paper, which uses two Latin manuscripts as its controls.12 Add the women, the baths and the possible plants, and she arrives at a guess she is careful to label: "it starts to feel like a book of women's folk knowledge that's being preserved. But you cannot prove that yet."1
Two caveats are worth keeping in view. The similarity measure was designed for texts you can read; applying it to text where you cannot even be sure what a word is, is a leap, and researchers on the hoax side have already said so.2 And the claim that the book is misbound rests partly on an absence: the test found no significant difference between the two halves of a sheet and the two sides of a leaf, which is what the theory predicts, but an absence is weaker evidence than a presence. It will stand or fall on whether other methods find the same thing.2 The authors' own closing hope is modest: that "linguists and cryptologists will revisit past work in the light of this hypothesis".2
"The irony is that whoever mis-bound this manuscript still did us a huge favour by ensuring its survival."Lisa Fagin Davis, to the Guardian, October 20261
VI · The claims
Roger Bacon's microscope. A Cathar suicide manual. Hebrew by computer. Proto-Romance in two weeks. Every few years someone cracks the Voynich manuscript, and every time the same things go wrong.
Before the ledger, a test. A genuine decipherment has to do four things, and most claims fail the first: it must give each sign a fixed value, it must produce grammatical text in a known language, it must work on passages the decoder did not choose, and it must explain the statistics in Chapter IV. Newbold's method, the first great failure, produced Latin only because the final step was to anagram the letters, which can produce anything.17 Nearly a century later the "proto-Romance" reading of 2019 failed for the same reason in a new form: its sign values changed from word to word.31
A Pennsylvania philosophy professor claimed each letter hid tiny shorthand strokes which, anagrammed, gave Latin: Bacon had seen sperm cells through a microscope in the 1200s. In 1931 John Manly showed the "strokes" were cracks in the dried ink and that the anagram step could yield any text at all.1735
A Rochester lawyer's simple substitution plus heavy abbreviation. The Latin it produced was unacceptable to Latinists and no other researcher took it up.35
A Yale cancer geneticist announced in the journal Science that the author was Anthony Askham, a 16th-century English astrologer, and the text concerned women's diseases. The method depended on his own transliteration and the output made no sense.6435
The man who broke Japan's Purple cipher ran two study groups on the manuscript, in 1944–46 and 1962–63, and produced the first machine transcription. His own verdict he sealed as an anagram in a 1959 footnote, opened after his death: the book was "an early attempt to construct an artificial or universal language of the a priori type". Nobody has found such a language from the 1400s.5122
A Navy cryptanalyst showed the text splits into two statistically different halves, now called Currier A and B, and that more than one person wrote it. Still the backbone of all text analysis.21
Not a solution but the survey every later researcher starts from, written inside the National Security Agency and declassified decades later.22
A Yale philosopher mapped the signs to digits and the digits to letters, read a few plant labels, and concluded the book was a 16th-century fake made to sell to the emperor. The readings did not convince, and the carbon date later ruled out the century.6535
The text was a pidgin of Flemish, Old French and High German describing a rite of ritual death among the Cathars, the heretical sect of medieval southern France, and a cult of the Egyptian goddess Isis. Linguists demolished the grammar; historians pointed out the "Cathar" practices contradict everything known about the Cathars.35
A Keele computer scientist showed that a table of syllables and a card with holes cut in it, a Cardan grille, could generate Voynich-like gibberish fast. It proved such text is cheap to make, not that this text was made that way; and the grille was only described in 1550, a century after the parchment.2562
Two botanists matched the plants to those of 16th-century Mexico and proposed that the book was made there after the Spanish conquest. That requires the book to have been written on vellum a century old, and ignores the 15th-century handwriting, quire marks and binding.3334
A Bedfordshire linguist proposed sound values for 14 signs by matching labels to plant names, as Champollion did with royal names when he cracked Egyptian hieroglyphs. He called it "partial". Critics found the letter values inconsistent and the plant identifications circular. Bax died in 2017 without taking it further.3235
Two Alberta computer scientists built a method for guessing the language behind an unknown script and ran it on the Voynich. It said Hebrew, assuming the words were anagrammed. The press said a computer had cracked the book; the authors said the opposite, and the paper itself warns the result could be an artefact of the anagramming. It does not explain the predictability of the letters shown in Chapter IV.272835
A television writer announced in the Times Literary Supplement that the signs were Latin shorthand and the book a compilation of remedies for women. Within days Lisa Fagin Davis noted the two lines he had "translated" were not grammatical Latin and the women's-health idea was decades old. "Frankly, I'm a little surprised the TLS published it."29
A Bristol research associate published in Romance Studies that the text was a lost ancestor of the Romance languages, written by Dominican nuns. The university's press release went out on 15 May 2019 and was withdrawn the next day: "concerns have been raised about the validity of this research from academics in the fields of linguistics and medieval studies". Davis: "'proto-Romance language' is not a thing."3031
Schinner had argued in 2007 that the text's statistics look like the output of a random process.63 The 2020 model, in which the scribe writes each word by copying one nearby and altering it, reproduces Zipf's law, the repeats, and much else. The strongest current version of the "it means nothing" hypothesis; the natural-language camp disputes it.26
Handwriting analysis across every page finds five hands, tracking Currier's two languages. The current baseline.20
Two Yale linguists reviewed everything known about the text's structure and argued it is best treated as a natural language in an unusual encoding, while conceding "the contents remain undecipherable". Hoax researchers replied in print.23
Forty-two volunteers were asked to write meaningless text. On 41 of 42 statistical measures the Voynich text looked like their gibberish; on 38 of 42 it looked like meaningful text. The one thing the volunteers could not reproduce was the Voynich habit of putting particular letters only at particular places in a word. The authors do not claim the book is gibberish, only that the question is open.67
A science journalist designed a cipher using dice, playing cards and substitution tables that turns Latin or Italian into text with Voynich-like statistics, using only 15th-century tools. He does not claim it is the real method, only that a meaningful cipher of the period could produce such text. A rebuttal followed in 2026.5055
See Chapter V. Not a decipherment; a claim about how the book was made. Critics question whether text-similarity methods built for readable text can be trusted on this one.2
It is the obvious question, and a serious one: he was a dealer who stood to gain, and he was cagey about where it came from. Against it: the parchment is 15th-century, the ink and pigments are medieval, wormholes run through the writing, the sewing is old, and three 17th-century letters from Prague describe an unreadable book of herbs and stars that matches this one.6810 The multispectral scans found no erased earlier text anywhere, so it is not a palimpsest.19 To forge it, Voynich would have needed a stock of blank 1420s calfskin, five handwriting styles and a 1639 letter in a Jesuit archive in Rome. Zandbergen calls that an "unacceptable coincidence".35 That is not quite the same as impossible, which is why the theory never entirely dies.
VII · What we know
The honest state of play, in three columns. Most popular accounts blur them. This page has tried not to.
If you want to look for yourself, Yale has put every page online at full resolution, free to use.352 The standard transliteration, the analyses, and three decades of patient catalogue work are on René Zandbergen's site, which is where most of this page's facts come from.5 And the Beinecke's curator gets three emails a week from people who have solved it.18 Read Chapter IV before you send the fourth.
VIII · Sources and method
Numbers in the text link here. Where sources disagree, the text says so. The statistics in Chapter IV were computed for this page; the method and code are on the About page.
