IT HAS been called the most mysterious manuscript in the world. Each page is
filled with strange illustrations of medicinal plants, astrological diagrams and
naked women. The whole book is bursting with long-forgotten secrets.
There鈥檚 just one snag. Its 234 pages are written in an unknown script
representing an unknown tongue, and despite nearly a century of study by the
world鈥檚 best cryptanalysts, academics and hobbyists, no one has verifiably
deciphered a single word. What has become known as the Voynich manuscript could
be the ravings of a madman, a made-up language or an elaborate hoax to dupe a
16th-century monarch out of a small fortune in gold. Nobody knows.
鈥淚t鈥檚 maddening,鈥� says Jacques Guy, a retired French-born linguist who
studied the manuscript as a hobby for more than 10 years. 鈥淵ou would think by
now we鈥檇 have figured it out.鈥� Guy is one of a band of hobbyists who are still
trying to crack the Voynich manuscript, armed only with computers and a bunch of
kooky ideas.
Advertisement
Statistical techniques borrowed from the Human Genome Project have thrown up
some fascinating clues to the nature of the text. And now that the most accurate
ever transcript of the manuscript is nearly complete, these researchers may have
the best chance yet of finally solving the puzzle. If they succeed, it won鈥檛 be
before time.
The American collector Wilifrid M. Voynich came across the book at the Villa
Mondragone in Frascati, Italy, back in 1912. Tucked inside the tome was a letter
dated 1665 or 1666 from the rector at the University of Prague, asking a
well-known scholar to have a go at cracking the cipher. According to the letter,
the Holy Roman Emperor Rudolf II of Bohemia, who reigned from 1576 to 1612,
bought the manuscript for 600 ducats鈥攁bout three-and-a-half kilograms of
gold. The letter also hinted that Roger Bacon, the 13th-century Franciscan monk
and mystic who dabbled in science and cryptography, might have penned the
manuscript, though this is highly uncertain. The book is now held at Yale
University in the Beinecke Rare Book and Manuscript Library.
But the intriguing history is only half the appeal. The Voynich itself is a
decidedly strange object. It鈥檚 a small, thick book about the size of a hardback
novel, illustrated with bizarre pictures. Forget the sumptuously illuminated
manuscripts you鈥檝e seen in museums: the Voynich鈥檚 art is about as sophisticated
as that of a 12-year-old鈥攅ven if the subject matter is somewhat more
advanced.
The botanical section features crudely drawn flowers and plants, unlike
anything found in nature. Accuracy does not appear to be the artist鈥檚 goal,
though some of the plants look vaguely like peppers and sunflowers. Then there
are the women who appear in the biological section鈥斺�漬ymphs鈥� as Voynich
devotees call them. Short and squat with pot bellies and small, pointed breasts,
these females wade through ghoulish green baths fed by pipes resembling arteries
and Fallopian tubes. The astrological section is perhaps the most carefully
drawn, featuring zodiacal circles and drawings of the heavens, including what
appears to be the Andromeda Galaxy.
But even more mysterious than the artwork is the text. The characters look
teasingly familiar: some resemble Roman characters, Arabic numerals and Latin
abbreviations. Elaborate 鈥済allows鈥� characters decorate many beginnings of lines,
while an enigmatic swirl like the number nine can be found at the end of many of
the words. Yet the document鈥檚 meaning remains infuriatingly obscure.
When Voynich brought the manuscript to the US, he invited cryptographers to
try their luck at decoding it. The first to claim success was William Newbold,
at the University of Pennsylvania who announced in 1919 that the manuscript was
a copy of Bacon鈥檚 lab notes. Newbold鈥檚 translation implied that Bacon had access
to telescopes and microscopes, instruments not thought to exist in the 13th
century.
Failed Attempt
Eventually Newbold鈥檚 interpretation was discredited. He died muddled and
frustrated, but refusing to admit defeat. Since Newbold, many cryptanalysts and
crackpots have claimed victory over the Voynich, but all of them suffered the
same problem: their solution did not apply to all the text in the manuscript.
Bizarre theories about the manuscript still crop up now and again, such as the
idea that it is written in vowel-deficient Ukrainian, or that it is a rare
document describing the Cathar movement.
Whatever the truth, it seems the text was encoded using a method so arcane as
to render it impossible to decrypt, even with today鈥檚 computers. Cryptanalysts
at the US National Security Agency had a shot at it in the 1960s and 1970s. They
transcribed it into a machine-readable form and ran a few statistical tests but
even the NSA couldn鈥檛 solve it.
Then, in the late 1970s, Yale scholar Robert Brumbaugh said he鈥檇 worked it
out: the manuscript was a hoax designed to fleece Emperor Rudolf II of his
riches. Only a few sections of the text were decipherable, said Brumbaugh,
presumably to convince the discerning buyer that the item was genuine. Since the
Voynich bears little likeness to the sumptuous illustrated manuscripts of its
time, and because it was written on the kind of cheap paper used to wrap fish,
the hoax theory is a popular one鈥攅xcept with Voynich devotees like
British-based pathologist Gabriel Landini of the University of Birmingham. 鈥淭hat
sort of attitude is not very productive,鈥� he says.
After all, there are perfectly good reasons aside from hoaxing why the
Voynich should be hard to crack. Finding a true solution means either working
out a cipher or finding a code book鈥攁 sort of dictionary in which each
Voynich word has an equivalent word in Latin, English or some other language. If
the manuscript was coded with a Voynich dictionary, then unpicking the text may
prove almost impossible unless a tattered copy of the code book turns up
somewhere.
Prospects are better if the book was written using a cipher鈥攁n
algorithm for replacing the ordinary letters in a text with Voynich characters.
The ciphers normally used by 15th or 16th-century cryptanalysts were not all
that high-tech. Most consisted of simple substitutions, in which a different
character replaced each letter of the alphabet in the message. This kind of
cipher is easy to solve. You tabulate the frequencies of the characters in the
enciphered text, and then match the distribution with that of the base language
to find out which letter each character represents.
Much harder to crack are ciphers that use any of several characters to
replace a given letter. Leon Battista Alberti published this concept in the
1460s, so the Voynich could be written in one of these if it dates from a later
period. It鈥檚 also possible to crack a polyalphabetic cipher by comparing letter
frequencies, but far harder than for a simple cipher.
The Voynich presents its own special difficulties as well. How do you compare
letter frequencies when you don鈥檛 know what language it uses? Worse, researchers
know little about the nature of the Voynich alphabet itself. Even handwriting in
English can be hard to read鈥攖wo cursive e鈥檚 in a row can look confusingly
like a cursive u. And in the flowery and variable Voynich handwriting, many of
the letters flow together. Some characters, such as the gallows letters, may
simply be paragraph markers. Or perhaps each Voynich word is a letter, or maybe
the spaces between words are placed at random to confuse the reader.
It may seem hopeless, but researchers have still managed to glean some basic
facts. The manuscript is about 250,000 words long, containing about 40,000
different words. There are between 23 and 30 characters, and curiously, none
behaves like a number. The manuscript reads from left to right, and most
Voynichese words are about six characters long. They show less variation in
length than those of English, Latin and most other Indo-European languages and
there are a lot of repeats鈥攗p to four consecutive repetitions of a word is
common, as are strings of words that vary only by one character.
In the past 10 years, since the inception of a Voynich e-mail circle to share
ideas and findings, researchers have been using statistical methods to search
the manuscript for clues about what it could be hiding. For all its strangeness,
the Voynich does indeed display certain consistent statistical properties that
are hallmarks of natural languages. So the manuscript is unlikely to be the
random writings of a madman or a fraud.
One way to analyse the manuscript is by studying its entropy鈥攁 measure
of how densely information is packed into the characters or words. A
high-entropy language contains lots of variation in the character order, word
order, or both. There are different kinds of entropy but one of the simplest can
be calculated by asking, given any character, how much uncertainty there is
about what the next character might be. For example, in English we know that a q
is almost always followed by a u, so there is low uncertainty and thus low
entropy for that particular letter.
In 1976, Yale physicist William Bennett found that this kind of entropy for
Voynichese was low compared with Latin, English and other European languages.
There are low-entropy languages from Polynesia, but it seems unlikely that
Hawaiian islanders were wandering around Renaissance Italy.
This low entropy rules out a simple or polyalphabetic cipher, since the
former leaves entropy unchanged, and the other increases it. Although there are
some kinds of cipher that can decrease entropy, the entropy data seems to
suggest that Voynich is a code-book text, or possibly a made-up language.
A much stranger theory was put forward a few years ago by Jorge Stolfi, a
Brazilian computer-graphics researcher at the University of Campinas in Brazil.
By running a computer program that analysed the placement of each character
within words, he found that certain letters seem to appear only in certain parts
of the words鈥攊ndicating a prefix, middle and suffix to each word.
He conjectured that Voynichese might be a version of Chinese, written down
phonetically by Chinese visitors who accompanied Italian explorers home from
their travels. Chinese words are monosyllabic, and each one has three phonetic
components. Thus the prefix might signify the tonal quality of the word, while
the middle determines the consonant and the suffix the vowel.
Stolfi made a pizza bet with other Voynich aficionados that his Chinese
theory was correct, but has since backed down. The others aren鈥檛 ready to accept
Stolfi鈥檚 defeat just yet, however. 鈥淚 tried to buy Landini a pizza but he said
the idea is not dead until someone actually deciphers the manuscript,鈥� he
says.
Meanwhile, Stolfi has refined his computer model, and last year published a
new analysis that finds a 鈥渃rust, mantle and core鈥� to the words. This consistent
structure makes it unlikely that Voynichese is random gibberish or a
polyalphabetic cipher. His findings do seem compatible with a dictionary-like
translation, or a non-European language with largely monosyllabic words.
Stolfi鈥檚 current theory, inspired by the rigid word structure, is that the
Voynich 鈥渨ords鈥� are actually numerals. Many numbering systems place restrictions
on the order in which different symbols appear鈥攁 six-character Roman
numeral, for example, would probably start with M, D or C, and end with I, V or
X. So could the Voynich manuscript actually be a list of numbers? The fact that
they are mostly of a similar length suggests another level of complexity, and
Stolfi thinks you might need a code book to translate the numbers into
meanings.
Hidden Information
Several other researchers are continuing to explore the idea that the
manuscript is enciphered. New studies suggest that its entropy may not be as low
as has been thought. Rene Zandbergen, who works as a systems analyst in
Darmstadt, Germany, found last year that whereas the first and second characters
in each Voynich word do indeed have low entropy, the third and subsequent
letters carry more information. This means it could yet be an enciphered
language after all, and suggests Arabic or Oriental origin.
Meanwhile, Landini has studied the text with a technique called spectral
analysis. It鈥檚 a method normally used to search for patterns within fluctuating
strings of data such as DNA bases or musical notes. The likelihood of finding a
character repeated after a certain period鈥攅ight letters later,
say鈥攊s plotted against the period. For random inputs such as white noise,
the resulting 鈥減ower spectrum鈥� is a flat line. But for signals with some
underlying pattern, the power spectrum deviates from a straight line鈥攅ven
if the pattern is very subtle.
Landini published his results in October (Cryptologia, vol 25, p
275). He found that the patterns in the Voynich manuscript matched those of
natural languages. For example, a natural period of about 5.9 characters was
evident, which coincides with the average word length in the manuscript. This
seems to confirm that the spaces do indeed separate real words, rather than
being placed arbitrarily to confuse the reader.
So what鈥檚 the next line of attack? Perhaps most important for now is getting
clean data for the statisticians to work on. The microfilm versions available
from Yale for $40 can be difficult to read, increasing the confusion over
what is and what isn鈥檛 a letter, so Voynich enthusiasts have started creating a
high-fidelity digital transcription. Zandbergen and Landini have collected the
transcriptions made by the US National Security Agency researchers and others
over the years. They鈥檝e been poring over each copy to resolve discrepancies and
come up with a definitive version. They hope eventually to persuade the Beinecke
Library to issue a full-colour version on CD-ROM. 鈥淚t is like the Human Genome
Project for the Voynich,鈥� says Landini. 鈥淥nce we have a complete copy, we鈥檒l be
able to repeat all the things that have been done before on better-quality
诲补迟补.鈥�
Luck and inspiration will still play an important part, however, says Jim
Reeds, a professional cryptanalyst at AT&T Labs in Florham Park, New Jersey.
Reeds is known for deciphering another 15th-century work, a book by Johannes
Trithemius, a German abbot and avid cryptographer. 鈥淪upercomputers don鈥檛 crack
codes by themselves,鈥� says Reeds. 鈥淭hey need people to come up with some sort of
辫濒补苍.鈥�
But Jacques Guy says he鈥檚 tired of the chase, although he still regularly
posts updates on the Voynich e-mail circle. 鈥淭he Voynich manuscript is a litmus
test of our knowledge of language,鈥� he says. 鈥淎nd the result of the test is
pretty dreadful.鈥� Guy reckons it鈥檚 time to pursue a hobby that might actually
bring a real pay-off鈥攑erhaps decrypting the Easter Island tablets. 鈥淎t
least I know they鈥檙e not a hoax.鈥�