Surely hoping to emulate the stunning success of sideburns and urban beards in recent years, Gordon Rugg is now apparently trying to revive his old papers on the Voynich Manuscript, along with the fame on the world stage that they brought him before.

He has therefore recently co-authored a paper in Cryptologia – Gordon Rugg & Gavin Taylor (2016): Hoaxing statistical features of the Voynich Manuscript, Cryptologia, DOI: 10.1080/01611194.2016.1206753 – which I’m perfectly happy to cite, simply because I immediately append my opinion of it, both then and now: that it is specious quasi-academic nonsense that only an idiot would be convinced by. And any academic referee who read the paper and thought it sensible is an idiot too: sorry, Cryptologia, but it’s just plain true.

Rugg once again argues – just as he did 12 years ago – that the Voynich Manuscript must surely have been hoaxed using a set of tables and grilles (broadly similar to Cardan grilles, a mainstay of popular books glossing 16th century cryptography) to ‘randomly’ select word-fragments from those tables, while yielding the visual appearance of the ‘Currier Languages’, specifically the Voynich Manuscript’s two main ‘dialects’ (or, as Currier himself would have preferred to say to avoid being misunderstood, ‘statistical groupings’).

Because these tables and grilles allow people to quickly generate hoaxed text mimicking the structure and statistics of Voynichese, he and his co-author Gavin Taylor triumphantly conclude (much as Rugg did before):

“The main unusual qualitative and quantitative features of the Voynich Manuscript are therefore explicable as products of a low-technology hoax, with no need to invoke an undiscovered new type of code and/or the presence of meaningful text in the manuscript.”

In my opinion, this was a dud argument in 2004, and – given all we have learned about the Voynich Manuscript in the decade and more since – it’s an even bigger dud in 2016. Specifically, I think there are Four Big Reasons why this is so:

Reason #1: Rugg’s History Doesn’t Work

Given that nobody used a Cardan grille before Girolamo Cardano (1501-1576) invented it in 1550, Rugg’s requirement that his putative Voynich hoaxer’s “low-technology” mechanism uses a sophisticated Cardan grille variant necessitates a post-1550 date.

But opposing that is (a) the radiocarbon dating of the vellum to the first half of the 15th century, (b) the mid-15th century ‘humanistic’ handwriting that is used on every page, (c) the 15th century handwriting used for the quire numbers, (d) the 15th century handwriting used for the back page, and (e) numerous Art History arguments pointing to a 15th century origin (which I get bored of reprising, and of defending against Diane O’Donovan’s endless sniping).

So, to shore up his wonky historical timeline, Rugg has to start by saying that the Voynich Manuscript is not only a hoax, but also an extraordinarily sophisticated late-16th century literary forgery, where all these distinctive 15th century features were codicologically layered on top of one another (and using century-old vellum) in order that the finished hoax artefact resemble some unknown kind of 15th century herbal manuscript.

In 2004, we already knew enough to say that this made no sense and was manifestly wrong (I certainly did so, even if nobody else did): but by 2016, this side of Rugg’s claim alone shouldn’t stand up for even a New York second.

So… does his 2016 paper fix this problem in any obvious way? No, sorry, it doesn’t. (Italian playing cards, really? I don’t think so.)

Reason #2: Digital Mimicry Is Insufficient

Unlike the recent herds of Bax-inspired historical linguists roaming wild across the arid Voynichese plains, a-hunting for dry tufts of linguistic tumbleweed lodged in the statistical cracks to feed upon, Rugg initially constructed his clever tables ex nihilo: for a long time, he considered the problem of Voynichese as a purely forward construction issue. That is, all he was trying to do was to mimic the statistics of Voynichese: his claim was therefore not that he could reproduce Voynichese, but that his tables and grilles could produce something that resembled Voynichese (if you didn’t look too closely).

This was, of course, an extremely lame ta-da to be passing off as any kind of über-theory. And so, after a great deal of prodding, he then went on to claim that it should be possible to work backwards from the Voynich Manuscript to try to reconstruct the tables that were used locally. But – to the best of my knowledge – he has retrofitted not even a single paragraph’s worth of tables and grilles in all those years, let alone an entire bookful. (It turns out that Voynichese is much less regular and well-formed than it at first looks.)

Rugg then back-pedalled once again, saying that all he was trying to do was to prove the possibility that a mechanism along these lines could conceivably have been used to generate the Voynich Manuscript.

Yes, and the Voynich Manuscript could conceivably have been found in the middle of a giant golden egg, laid by a space turkey on the Pope’s lap. “Conceivability” isn’t a particularly useful metric, let’s say.

Reason #3: Rugg’s Computer Science Doesn’t Work

At its core, Rugg’s idea of using tables to generate the ghostly immanence of historical signal is a kind of anachronistic computer game hack (and I speak as someone who wrote computer games for 20 years). Beyond the comforting surroundings of his basic word-model, he adapts each and every exception case (and Voynichese has plenty of these: paragraph-initial, line-initial, line-final, A, B, Pharma-A, Bio-B, labelese, etc) with layer upon layer of yet further improvised explanatory hacks.

But even if you – somewhat trustingly – accept that these multi-layered CompSci hacks will collectively coordinate with each other to do the overall job Rugg claims they will, they still all fall foul of the basic problem: that prior to computers, nobody used tables to generate text in such a futilely complicated manner.

Don’t get me wrong, using tables to simulate cleverness is a great hack (and Rugg understands completely that a Cardan grille is nothing more than an indirection method for selecting a subset of a two-dimensional table), and one that sat at the heart of countless late-1980s and 1990s computer games (the Bitmap Brothers were particular masters of this art).

But it’s at heart a great modern hack, not a 16th, 17th, 18th, or even 19th century hack.

Reason #4: Rugg’s Arguments Don’t Work

Even though the preceding three reasons are each gnarly enough to throw their own Herculean spanner into Rugg’s works, this fourth reason is about a problem with the entire structure of his argument.

Rugg claims that his solution of Voynich Manuscript verifies his “Verifier Method”, the approach he claimed to have used to crack it (and on top of which he has built his career). But all he has actually proved is his ability to retrofit a single bad solution to it that is, though not historically or practically credible, conceivably true. This is, in other words, an extraordinarily weak conclusion to be drawing from a hugely rich and complicated dataset, comprising not only the Voynichese text but also all the physical evidence and provenance information we have.

I’ll happily admit that he has produced a possible solution to the Voynich Manuscript’s mystery: but this has come at the cost of discarding any vestige of historical or practical likelihood, an aspect which was just about visible in 2004 but which should be glaringly obvious in 2016.

And if that’s what the poster child for his Verifier Method looks like, I shudder to think what the rest of it looks like.

21 thoughts on “Gordon Rugg, “The Man Who Cracked The Mystery Of The Voynich Manuscript”, cracks it once again (NOT)…

  1. “sniping” usually requires an object.

    I don’t “snipe” at the manuscript. I explain its imagery.

    This I am qualified to do. That it contains evidence or older, earlier and non-European culture and content is a conclusion I am entitled to draw from the evidence offered by the primary document.

    This post does no justice to your readers; to your own intelligence; to me or to the editors of Cryptologia.

    When did your interest in a small, fairly unremarkable-looking fifteenth century manuscript become a matter of “Be on my theory team or die?”

    Your energy and intellectual curiosity – your ability to see other ‘takes’ on the manuscript – initially brought you showers of comments. Now, not even those who think well of you are immune from your “team spirit” attitude. Not so becoming.

    In any case, I’m taking a break from the madness of the “believe me or you’re for it” mentality. Who cares whether the bloody thing proves, in the end, to have been made in northern Italy, England or southern France, or Bohemia – or whether the text is or isn’t in cipher, or generated or not, or using the sort of grille we only know of through Cardin’s exposition.

    If understanding the manuscript becomes irrelevant, and eight years’ research and explanation of the imagery, is cultural traces, indications of non-Latin origin and so forth weigh less with you than obedience to your ‘authority’ then who he hell are you? Not the chap whose blog I began reading, and of whom I’ve continued to speak well and to recommend since 2008.

    Time to take a break.

  2. – and this may surprise you. I have a real lack of patience with people unable to distinguish between form and content. They would fall into a giggling heap about a typo in Herodotus, or Einstein’s poor math – get over it. I meant to write as I did and to write ‘Cardin’.

    Cheery-bye

  3. Here’s the high-level problem with Rugg’s claims that I have:

    If I reach into the metaphorical urn from Intro to Probability and Statistics 201 and pull out a single sample from an unknown distribution, then in the absence of evidence to the contrary the reasonable assumption is that this lone sample — shivering alone and afraid in a world it didn’t make — is roughly modal. ‘Cause that’s what “modal” _means_. As a result, the reasonable assumption is that, whatever the Voynich Mss. text is (enciphered plaintext, generated pseudo-language, whatever), the Mss text is a fairly average example of text generated using that method.

    As an aside, this is why claims of a glyph-level mono-alphabetic cipher are so unpersuasive. A mono-alphabetic cipher doesn’t really transform the underlying plaintext in any meaningful way, so all the statistical weirdness of the Voynich text has to be pushed back into the purported plaintext because there is nowhere else for it to go. The result is a plaintext that sounds contrived or nonsensical.

    In the case of Rugg’s theory, the way this problem manifests is in the claim that “highly structured tables” are needed to generate Voynich-like text (https://hydeandrugg.wordpress.com/2013/06/23/the-voynich-manuscript-emergent-complexity-in-hoaxed-texts/). In other words, the properties of the Mss. text are not natural consequences of the grille-based text generation mechanism, but have been pushed into the limited class of “highly structured” tables (out of the universe of tables that could have been chosen from) chosen from by the Mss. author(s). Rugg pushes the statistical weirdness into the specific tables chosen, rather than its arising organically from the grille method.

    Rugg has made some attempt to explain (for example) the binomial word length distribution as a natural outcome of the grill method (https://hydeandrugg.wordpress.com/2013/09/03/hoaxing-the-voynich-manuscript-part-6-planning-the-word-structure/#more-419), although it’s somewhat odd that after making the slightly hand-wavey argument he makes in that post he doesn’t back it up with an actual word length distribution from one of the grille-based texts he’s generated.

    P.S. — On the historical side, add glyphs that appear to be inspired by Arabic numeral forms that are pre-1500.

  4. Anton Alipov on September 15, 2016 at 7:07 am said:

    There are some tiny things speaking against the hoax theory, whether grille-based or not. For example, two most frequent “Voynich stars” (labeled objects of f68r1 and f68r2) – namely, otol and odaiin – both occur in f1r, and both in the same paragraph. Any hoax theory will necessarily need to call such things coincidental.

  5. Hi Nick,

    I thought that after his complete failure to provide anything remotely similar to the VM – using his LEGO method – Gord would give it a final R.I.P.
    However, he apparently had another “inspirational coffee break” as he once called it).

    Unfortunately, it must have been anther decaf since his gibberish ( no, I do not mean his article but his product 🙂 is in reality no gibberish at all, only since it is an organized puzzle-game.

    Provided he still did not postulate which grammatical rules (observed in the VM) is the rying to emulate in his grill, all he can get is the not-so-well-done steak.

  6. James R. Pannozzi on September 15, 2016 at 1:18 pm said:

    Nick’s comments are cogent, incisive and damning and pulls the rug out (sorry, couldn’t resist) from under Gordon’s collection of pseudo-scientific rationalizations.

    After spending 32 years in the world of software development and software engineering, it occurs to me that if Rugg’s theory were even half correct, he could and should be able to provide, in this age of the future in which everyone has relatively high power computers sitting on their desktops, an EXACT algorithm, exactly reproducing the Grille selections done by a mere human of several hundred years ago and ending up with at least some genuine Voynich “text”. That does not mean a “sorta like this” algorithm. Try using, for example, Koza’s genetic algorithm approach described in his books, in Lisp. After some hours or days of searching, it should be able to come up with with something. Or, try Dykstra’s method of weakest pre-conditions if you’re in the mood for some serious pseudo computer “science”. Of course, we all know it won’t but let Rugg find out the hard way, it might prove instructive.

    Instead, Rugg cobbles together a collection of exceptions and corrections which might very well produce something looking LIKE the Voynich without being it.
    Rugg has a computer, so go ahead – program yourself a genetic algorithm or neural network and have it set to work searching for Grille operations which exactly produce even a page of Voynich exactly. Then publish in Cryptologia, not before.

  7. Well, gentlemen, I’m sorry to see that you all are still stuck in ‘codiology’ mode. There is absolutely zero coding in any folio of the so-called Voynich manuscript (now better known as Boenicke manuscript 408).
    Until you get ‘a round tuit’ (a round clip-on button which has printed on it : “tuit” ) you will never be able to intuit or translate the written material of B-408 ,

    So, Nick and friends, until you can begin translating the Spanish/Latin and Nahuatl translation of the Spanish/Latin — you are stuck in a relentless circle of codiology.

    bd

  8. Gordon Rugg gives evidence for a bell curve token length distribution being generated by his method. He doesn’t explicitly state that he is describing tokens rather than a word list but, in context, he is.

    See Verifier, Voynich and Accidental Complexity. Posted on June 22, 2013

    This is a fundamental error. The VMs tokens do not graph to such a curve. The token length distribution of the VMs is within the range of known texts. A bellish curve is generated by the length distribution of *distinct* words (word list or lexicon or types). This is partly due to lumping words from different parts of the VMs. If non-symmetrical curves, skewed and humped differently, are added together, the product will be more symmetrical than the individual curves. For this additive curve to be significant, the creator of the VMs would have to have:
    1. produced a word list
    2. selected words for each section
    3. arranged the words in a text
    Even so, there are known texts with word length distributions that are fairly close to the combined VMs word list. At least one has shorter words.

    I agree with Mr. Pannozzi but I don’t think complicated software is necessary. With software to mimic Rugg’s hand scheme, it should be easy to produce 10,000 or 100,000 tokens. All that has been presented, as far as I can discover, is 1949. This is “software which implements some of the features of the manual version”. How difficult is it to write a program to duplicate the hand method? I am no programmer but I would guess it might take about 30 minutes to an hour.

    Within that limit (1949 tokens), the number of unique words in the generated text is abnormally low. The edit distance between the words is abnormally low; much lower even than in Quire 13, which is the lowest in the VMs — which, in turn, is lower than in ordinary text.

    Gordon Rugg has reversed Stolfi’s discovery. There’s nothing wrong with that if you don’t object to overfitting data. As far as bundling it with claims that impress people who are ignorant of the subject, that happens in other fields. Causes resentment and life goes on.

  9. Nick, your deliciously caustic demolishment of Rugg’s stuff brought tears to my eyes.

  10. Julian: it’s nothing I haven’t blogged or told him before, though perhaps not all at the same time. 🙂

  11. Jan: as I recall, the technically correct term to describe the output of his table-and-grilles process is ‘Ruggish’.

  12. Dear Prof. Panozzi :

    If you would like to get back to translating the “Voynich” manuscript, you might like to pick up a copy of “Nahuatl as Written” (Nahuatl Studies Series Number 6″):
    Series Editor: James Lockhart
    Associate Series Editor Rebecca Horn
    Publishers: Stanford University Press and
    UCLA Latin American Center Publications

    bd

  13. Nikolaj on October 12, 2016 at 5:10 pm said:

    Good day!
    I don’t agree c Gordon Rugg.
    The Voynich manuscript is not written with letters and characters denoting letters of the alphabet one of the ancient languages. Moreover, in the text there are 2 levels of encryption, which virtually eliminates the possibility of computer translation even after replacing digits with letters.
    I picked up the key, which in the first section I could read the following words: hemp, wearing hemp; food, food (sheet 20 at the numbering on the Internet); to clean (gut), knowledge, perhaps the desire, to drink, sweet beverage (nectar), maturation (maturity), to consider, to believe (sheet 107); to drink; six; flourishing; increasing; intense; peas; sweet drink, nectar, etc. Is just the short words, 2-3 sign. To translate words with more than 2-3 characters requires knowledge of this ancient language. The fact that some signs correspond to two letters. Thus, for example, a word consisting of three characters can fit up to six letters of which three. In the end, you need six characters to define the semantic word of three letters. Of course, without knowledge of this language make it very difficult even with a dictionary.
    If you are interested, I am ready to send more detailed information, including scans of pages showing the translated words.
    Nicholas.

  14. Andrew on April 25, 2018 at 1:19 pm said:

    I agree with one of the previous commentators that Gordon is really just “bundling [previous work] with claims that impress people who are ignorant of the subject”.

    And in this day and age, where information can be retrieved at the touch of a button. It’s not working.

    Brilliant piece Nick.

  15. Mark Knowles on May 25, 2018 at 2:17 pm said:

    I have been giving some thought to what could be termed the hoax spectrum or noise spectrum. We can think of 2 extremes:

    1) The Voynich is a hoax and all the text is meaningless.

    2) The Voynich is not a hoax and every character gives meaning and none are superfluous.

    Then are states between these two.

    Rangimg from:

    3) All the text is meaningless except for one word on page???

    4) All the text is meaningful except for one word on page???

    Then between these we have greater or less percentages of the manuscript that are noise or null or meaningless.

    E.g. 50% of the text is just a filter and so unnecessary or null
    50% of the text is meaningful

    So I guess the question is where on the spectrum is the Voynich actually lies. Could Gordon Rugg be part right and part wrong? i.e. Could the null part of the text by generated by a procedure and the other meaningful? I am inclined to doubt this.

    However for some time my suspicion is that a significant part of the text is null due to the level of repetition amongst other things, though I certainly don’t believe it is a full hoax. This is a suspicion without strong evidence to support it.

    In fact in a theoretical sense I would think it impossible to prove the Voynich is a hoax as there could always be a very small part of it that is not.

  16. J.K. Petersen on May 26, 2018 at 12:28 am said:

    Rugg may be partly right…

    The text may be synthetically generated, but I don’t think Voynich text would result by the method he describes. There’s more order to the composition than one would get using a Cardan grille of the sort he demonstrated.

    .
    I don’t think synthetic text or even nonsense text necessarily makes it a hoax. Even nonsense or synthetic text might have an underlying reason (I can think of a couple).

    .
    In a theoretical sense, I don’t think it would be impossible to prove the VMS were a hoax (if that were the case) even if part of it were not.

    There are many documents where someone perpetrating a hoax started with something genuine and altered it for his/her own purposes. Whether it can be proved depends on many things, including the proportion of genuine/nongenuine, the time period elapsing between them, the skill of the fraudster, etc. I don’t think one can generalize about this—each instance would a unique case.

  17. Peter M on May 26, 2018 at 7:50 am said:

    @all
    Where is the difference between the VM and the reality.
    Then just have a look here ……

    https://www.facebook.com/photo.php?fbid=2047149155507728&set=gm.1599847693458359&type=3&theater&ifg=1

    I wish all of you a nice weekend. 🙂

  18. The reality is that if the VM was the product of any attempt to pass it off as something that it was not, with the intention to deceive, then I guess only then would it pass muster for a hoax. It should not matter in the least, the composite nature of its material form, whether or not some elements related to a specific historical period ie. old velum parchment decorated with diagrams and charts etc., of another period perhaps. From my understanding, the so called VM may never have been passed off as anything that might be construed as a deliberate false utterance, so nobody would have ever had a case to answer for fraudulent false pretence, if indeed it was ever deemed to be a hoax.

  19. Shouldn’t the layman expect to see some correlation between the five main subject criteria, bearing in mind that the montonous unknown dialogue does not seem to change pace or form throughout the various categories. Also the subjects do not appear to follow any particular format, such as one might hope for in an almanac of a type common to the period suggested. The only reason I can make for the diminutive size of the manuscri, if it were to be authentic, is that it was quite possibly designed to be easily carried in the field. Apart from that, the frollicking nymphs don’t lend much weight to any logical theme in the way they are represented and of course the part six recipes are mere speculation from what we’re led to believe.

  20. Mark Knowles on June 25, 2018 at 5:27 pm said:

    Your criticisms of Rugg’s arguments are ones I agree with.

    To me his biggest assumption is his starting point that “if the Voynich contains a real cipher then it would have been deciphered by now.” The idea that two hundred pages of a medieval cipher should be relatively straightforward for modern cryptographers to decipher appears to me to be flawed. It seems that he is overestimating the capabilities of modern cryptography. As I have said elsewhere it appears that devising a cipher which is very hard to break is actually relatively easy.

  21. Two topics..Two Carden Grilles..Two Flash Gordons..Two snake oil salesman..Two days..Too bad..

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Post navigation