Category Archives: Voynich Manuscript

Two long(-ish) form Voynich manuscript articles emerged recently, one in the Jewish magazine The Tablet “Tablet Magazine”, the other in the New Yorker’s online blog section. These tell us quite a lot – though not really about the Voynich Manuscript itself, but rather about how the Voynich Manuscript is now perceived.

The first article, by Batya Ungar-Sargon, is called Cracking the Voynich Code: The quixotic quest to read meaning in the patterns of a bizarre manuscript that has bedeviled scholars for years.

Her basic take is that “the Voynich Manuscript has become a beacon for a secular community of quasi-Talmudic scholars whose interpretive ingenuity and stamina have few parallels“, so her piece is built around interviews with several of them (including the “patient, tireless” Gordon Rugg, and the “deeply humble” Rich SantaColoma). [She also talked with me on the phone for an hour, but perhaps I didn’t fit her template 😉 ].

Taken as a whole, fitting her article into a primarily Jewish-interest magazine was always going to be a bit of stretch: William Friedman was Jewish, sure, but that’s a small piece of material to make a full-length dress out of. I can’t help but wonder whether Batya’s ambition is to write long-form pieces for the New Yorker, and that this was a try-out for her portfolio. She clearly writes well, but I don’t think her journalistic instincts are yet fully honed – her article, in my opinion, is still more ‘relating’ than the literary reportage to which she aspires.

The second article – The Unread: The Mystery of the Voynich Manuscript – by Reed Johnson is, coincidentally enough, from the New Yorker blog section. As such, it’s a kind of New Yorker long-form take on a blog post, i.e. longer than a normal blog post, but quite a lot shorter than a typical New Yorker article (I used to subscribe to it, though how I ever found enough time to read each issue I don’t know 🙂 ).

This isn’t Ungar-Sargon-style journalism, but is instead Reed’s telling the story of how he came to waste three years (only three years? Pshaw!) on the Voynich Manuscript – basically, while trying to write his own “Dan Brown–style thriller”, having nearly completed his “M.F.A. in fiction at the University of Virginia” in 2010. He tries to introduce a little light drama into his account (Did he crack the Voynich? Did he finish his book?) but with enough of a wink to astute readers that they know the resolutions long before the end.

Unlike Batya, Reed is not an observer looking in on the Voynich research world from the outside, but is instead an active participant in what he calls the “often fractious” Voynich mailing list. His feeling about this is that “If crowds have any wisdom, soon we should see the fruits of a more recent deciphering project: Internet crowdsourcing“. And yet, he also wonders whether it would be a disappointment for the Voynich Manuscript to be decrypted – that, “no matter how thrilling such a text might be, it [would] remain a disappointment for being closed off, completed — for being, in the end, no longer a mystery“.

My own conclusion is that the Voynich mailing list has become more part of the problem than part of the solution: and that the extraordinarily productive collaboration its early days saw was more down to the small number and high calibre of the participants (Jim Reeds, Jim Gillogly, Jacques Guy, etc), most of whom left the list long ago. Really, the collective wisdom of the crowd very much depends on the crowd you happen to be dealing with: though Reed stops short of showing his hand in this regard, so we end up knowing what happened but not his thoughts or feelings about it. Perhaps the whole card game hasn’t yet concluded for him.

What’s nice about these two articles is that, for all their differences, they are both good examples of clear-headed contemporary writing about the Voynich Manuscript, far from the lurid wodges of mystery-soaked ahistorical fragments I frequently used to see. Indeed, both give an account of the Voynich’s history that is broadly correct, something which simply never happened even a decade ago: perhaps the radiocarbon dating has helped validate the Voynich as a “proper” subject.

And yet… it’s as if something (or someone) is missing from the whole party. The Voynich Manuscript has had many of the best codebreakers of the age (the Friedmans, Manly, Tiltman, etc) examine it closely: yet as these articles show, a lot of contemporary discourse still revolves around the – frankly rather foolish and shallow, I think – postmodernist cipher/hoax tension as exemplified by Gordon Rugg and Rich SantaColoma.

To my mind, it’s as if something really important is missing from the whole conceptual landscape of how the Voynich is perceived, that everyone is somehow in the wrong kind of doubt. We’ve collectively travelled a really long way forward, for sure, but the ideas and insights gained on that journey have all been zapped by a kind of “motivated learning” paralysis, where debate is held in a stasis between powerful epistemological agendas.

It often feels as though, myself excepted (and who listens to what I say, ha!), the Voynich-as-a-genuine-historical-artefact point of view has no champion. I genuinely tire of the way people continually generate possible alternative histories for it, when I’m just about the only person trying to reconstruct the mainstream history they’re so busy fighting against.

I want to ask those “theorists”: why do you find the idea that the Voynich Manuscript was made basically when its radiocarbon dating says so dreadfully upsetting? Why do you invest so much time and effort into identifying outlandish alternatives that might possibly be made to work (with a few well-chosen tweaks to the mainstream historical timeline)? Do you not see that, by kicking back so hard against a straightforward historical account that hasn’t even been written yet, you are yourself holding everything back? Can you not see that by doing this you have become part of the problem, not part of the solution?

That is the Voynich Manuscript debate that’s missing, the elephant in the room that nobody wants to talk about. But nobody is writing that particular article, and I’m not sure anyone ever will… and perhaps we’re all worse off for that silence.

The news rattling the bars of the Voynich research cage loudest right now is surely the publication of a paper by Marcelo Montemurro and Damián H. Zanette called Keywords and Co-Occurrence Patterns in the Voynich Manuscript: An Information-Theoretic Analysis, deftly summarized in New Scientist as New signs of language surface in mystery Voynich text.

M&Z’s abstract brings out a lot of what they were trying to do – and also points exactly to their mistakes.

“Here we analyse the long-range structure of the manuscript using methods from information theory. We show that the Voynich manuscript presents a complex organization in the distribution of words that is compatible with those found in real language sequences. We are also able to extract some of the most significant semantic word-networks in the text. These results together with some previously known statistical features of the Voynich manuscript, give support to the presence of a genuine message inside the book.“

Central Assumption: the authors implicitly hypothesize that they can get meaningful results for long-range comparisons because Voynichese is homogeneous across all its sections.

…The Problem: this assumption is false (or very nearly so), because there are significant macro-level differences in the way the language in different sections works (Currier A, Currier B, labels) as well as many mid-level differences (Herbal-A, Q13-ese, etc).

Central Conclusion: the authors believe that their language-centric statistical machinery has identified “The thirty most informative words in the Voynich manuscript”.

…The Problem: I’m pretty sure that the authors have in fact very probably identified arguably the thirty least informative words in the Voynich Manuscript. (That may be an independently useful result, but it’s probably not really what they were hoping for.)

I’ll explain.

Voynichese is extremely predictable at a letter-level: it has many rigid letter-level adjacency rules (‘4’ is almost always followed by ‘o’, etc) and position rules (4o- is consistently word-initial, -89 is consistently word-final, etc) and a high level of letter-context predictability.

Yet at the same time, it also has a very large dictionary relative to its text size. I often criticize Gordon Rugg for suggesting historically incorrect Cardan grille-like tables (i.e. they’re a century too late for the Voynich’s construction dating) and for inappropriately back-projecting his modern CompSci mindset onto the early Renaissance (i.e. it’s 500+ years too early for the kind of table-driven hackery he proposes). However, he is absolutely right that a reconstructed Voynichese “dictionary” would, to a modern computer scientist’s eyes, look very much as if it had been generated or permuted by some means.

The paradox is therefore that these two apparently opposite aspects of Voynichese are able to coexist: how on earth can we reconcile its letter rigidity & predictability with its wild word variability?

I think the key to resolving this is to grasp that there is some kind of generative or confounding principle at work within a rigidly predictable framework. That is, that even though there are lots of rules, these rules act as a kind of “container” for semantic or cryptographic variability to exist within.

Hence I believe that Montemurro’s statistical machinery is identifying “words” that fall within the container layer rather than in the confounded content layer. Hence these are arguably the thirty least informative words in the Voynich Manuscript.

It’s a hard point to understand, let alone accept: the confounding trick (some kind of transposition cipher? some kind of paper cipher machinery? some kind of cipher wheel?) driving Voynichese’s inherent variability remains as profoundly unreachable now as it has been for over 500 years.

My apologies to Montemurro and Zanette, but the central challenge we face isn’t to find new language-based statistical tests to apply to the Voynichese corpus, however clever they may be. Rather, it is to find ways of resolving the Voynich Manuscript’s central paradox: how is it that Voynichese is both letter-rigid and word-variable at the same time?

Incidentally, M & Z conclude in their paper that results point to a semantic link between the Recipe and Astro sections, and between the Herbal and Pharma sections. Actually, had they been more aware of the codicology analyses that have been done, they would have seen that their results are consistent with the writing phase order.

In fact, there are many indications that what I call Voynichese’s ‘container’ layer above evolved during the writing, with the most obvious evolution being between Currier A and Currier B. I suspect that what their statistical machinery has imperfectly captured is therefore simply a snapshot into the evolution of the container layer, and not anything ‘semantic’ as such.

In short, the aspect of Voynichese that is most nearly homogeneous across all its sections is its “container” layer: so what Montemurro and Zanette have done is make long-range comparisons between evolutions of the container layer. Currently, my best guess is that these are likely to be almost entirely composed of cipher system meta-tokens (shorthand tokens, transposition cipher placeholders, etc) rather than the semantic contents, which appear instead to have been confounded by some means.

So, rather than finding a “genuine message” (as New Scientist put it), perhaps they have instead found a “genuine container” for the message? This may prove to be a very useful result in its own right, but it’s probably not the smoking gun linguistic proof they were hoping to use to discredit Rugg’s tables.

When I was young, I often used to play Scrabble with my grandmother Win on my way home from school. (By which I mean her maisonette was on my route home, not that we played Scrabble on the bus.) Which probably helps account for the deep-rooted enjoyment I still get from weird and wonderful words, many decades later.

From way back then, my favourite English word has always been “svelte” (though “tergiversate” was nipping at its heels for a couple of weeks last year). The reason I particularly like svelte is that it’s (I’m struggling to describe) ‘productively onomatopoeic’, in that the slow ‘l’-sound in the middle makes it feels elegant (indeed svelte) on the tongue. Really, it’s a word with an unusual (but nicely matching) mouth feel, one that manages to stand out from a dictionary sized pack. With getting too synaesthetic on you, to me it’s a kind of David Gower four of a word, a left-handed ping that’s over the boundary before the fielders even notice it’s gone. Something can’t be half-svelte, it’s either got it or it hasn’t.

Svelte also brings right to the fore the mad ragtag heterogeneity of English, the arbitrary coupling together of chance encounters over the millennia. To some it sounds Svedish Swedish (or perhaps a piece of stray Elvish?) but it’s actually a French word (svelte), from an Italian (svelto, “stretched out”), from Vulgar Latin (ex + vellere, i.e. to stretch + out).

(You might therefore suspect that it shares some kind of origin with “vellum” which is also stretched out, but the latter has its roots in “veal”, i.e. young calves: hence vellum is properly fine calfskin.)

Languages are like that: for all their modern apologists, academies, and syntactic niceties, they’re at heart accidental rather than designed. Esperanto and all the other modern conlangs are all very well, but a good part of the charm of real-world languages is the way stray and mongrel words hop in to fill the semantic gaps that inevitably open up as culture mutates and evolves. English obviously needed a word that expressed presence of svelteness in an object, why else would svelte have succeeded and persisted otherwise?

But (and isn’t there always a but in Cipher Mysteries)… where’s all that in the Voynich Manuscript’s language? Even if William Friedman was completely and utterly wrong about the Voynich’s being an artificial constructed language (which he was), I really can see exactly why he thought & believed that. For Voynichese words show such a strong family resemblance – a strongly interlinked productive grammar, if you will – that it almost precludes anything else. Whatever Voynichese is, there is definitely an artificiality to it, or at least an abundance of artifice. I suspect that anyone trying to map Voynichese onto a direct language base will almost inevitably find (to their eventual embarrassment) that it’s just too artificial to be workable: and that’s pretty much what Elizebeth Friedman concluded too.

So here’s your Voynich paradox for the day. I’m sure that there can be no “svelte” in the Voynichese ‘language’ as we see it, because the overwhelming majority of its words arise from a compact productive grammar quite unlike that of a real, heterogeneous, messy, accidental, historic language: and yet the look of Voynichese so resembles a language that it’s hard not to feel as though you’re perpetually a mini-dictionary away from just reading it.

Of course, for me the resolution of this paradox comes down to a well-chosen bunch of steganographic tricks (such as verbose cipher, shorthand, etc) that serve to conceal the plaintext in a misleading form… but you will no doubt have your own theories about how to slice through such a Gordian knot. 🙂

A correspondent has asked me to summarize the evidence I’ve found in the Voynich Manuscript suggesting bifolio reordering. As long-term Cipher Mysteries readers will know, I laid much of this out in my 2006 book The Curse of the Voynich: but a lot has also emerged in the years since.

A practical starting point here is my long-standing page on the Voynich Manuscript’s codicology. This points to evidence for a whole sequence of fairly direct codicological conclusions:-
(1) The Folio Numbers Are Not Necessarily Correct
(2) The Bifolios Are Not Necessarily The Right Way Up
(3) The Quire Numbers Are Not Necessarily Correct
(4) The Bindings Are Not Necessarily Correct
(5) The Quires Are Not Necessarily In The Correct Order
(6) The Quire Contents Are Not Necessarily Correct
(7) The Paints And Colours Used Are Not Necessarily Original

To this, I’d add some other evidence:-

(8) The quire numbers and the folio numbers are not (quite) consistent.

As John Grove pointed out back in 2002, the Voynich Manuscript’s Q9 (“Quire 9”) was rebound along a different fold after the quire numbers were added but before the folio numbers were added, leaving the Q9 quire number in the wrong place (i.e. not on the back page of its quire). And, as Glen Claston later pointed out, broadly the same thing happened for Q14 (the nine-rosette page): once Q14’s first binding fold collapsed or was damaged, the large multi-folio page was then restitched along a different, less obviously damaged fold, again leaving the Q14 quire number in the wrong place.

Of course, given that most of the (15th century) quire numbers look roughly a century older than the (16th century) folio numbers, a bit of rebinding between the two phases is perhaps to be expected. But all the same, this inconsistency should alert us to the fact that the bifolios were actively being worked on between the quiration and foliation.

(9) Some of the quire numbers’ downstrokes continue within the wrong quires.

I found two clear examples of this (Curse p.18): (a) the downstroke of the ‘9’ in “29” (i.e. ‘secund-us’) continues at the bottom of a page in Q6; and (b) the downstroke of the ‘5’ in “5t9” (i.e. ‘quin-t-us’) continues at the bottom of a page in Q3. In both cases (and particularly in the first of the two), it seems likely that at the time the quire numbers were added, the herbal bifolios were in quite a substantially order from the order we are presented with several centuries later.

(10) Vellum tears with parallel orientation may indicate that those bifolios came from the same tanned skin.

The examples I found (Curse pp.54-56) were on the f16-f9 bifolio and the f10-f15 bifolio, as well as on the f38-f35 bifolio and f36-f37 bifolio. The fact that the bifolios were still immediately adjacent in both cases loosely implies that the basic idea of codicological continuity during construction (i.e. that adjacent bifolios individual sections were probably folded and cut down from larger sheets of vellum) may well be sound. It also suggests that the f28-f29 bifolio (which has a stitched vellum tear) may be out of sequence.

(11) Currier “Herbal A” and Currier “Herbal B” bifolios seem jumbled up.

Back in 1976 or so, US Army cryptanalyst Prescott Currier noted two apparently distinct “dialects” of the ‘Voynichese’ language: he called these “A” and “B”, and pointed out a whole set of curious properties that helps you distinguish them from each other. Moreover, any given bifolio has only “A” or “B” writing on it: this broadly supports the idea that these correspond to two broadly separate writing phases, rather than two separate writers writing in parallel.

(12) The three sunflower-like drawings look to have been separated.

f33v, f40v and f50r all contain pictures of similar sunflower-like plants, and are all “Herbal B” pages: this reinforces the idea that the Herbal B pages should be looked at as a quite distinct content collection from the Herbal A pages. I’d add that the Herbal B “plants” seem far more artificial to me than Herbal A “plants”, some (but not all) of which actually resemble real plants (e.g. water lily, pansies, etc).

(13) Q13 and Q20 may have originally each been two smaller quires that were later merged.

There is a whole heap of content analysis that supports the idea that what we now call “Q13” was originally a ‘Q13a’ and a ‘Q13b’ (as proposed by Glen Claston in 2009) and that what we call “Q20” was originally a ‘Q20a’ and a ‘Q20b’ (as proposed by me in 2010).

This is not so very far from the observation [(3) above] that, given that the jars in the pharma section seem to progress from the end of Q19 to the start of Q17, Q19 originally preceded Q17. In short, we can’t be at all sure that the quire arrangement we see now matches the original quire arrangement or order – quires may well have been formed of smaller original quires that were merged (for whatever reason) before the quire numbers were added.

(14) The two pages with “chicken scratch” marginalia may well have originally been adjacent.

I suggested this in my 2012 Voynich Centenary Conference presentation “Between Vellum and Prague”, which tried to reconstruct how the Voynich Manuscript’s quires had moved around between its original ‘alpha’ state and the final (foliated) state. I think it would therefore be interesting to find out whether f66v (in Q8) and f86v3 (in Q14) were from the same vellum skin.

(15) Some bifolios that were (probably) central to a quire are now not.

f84v-f78r should clearly have been the centre of a Q13 quire (the pictures join across the bifolio’s central fold): but I should add that Rene Zandbergen also pointed out in 2010 that f18v-f23r may well have been the centrefold page-pair of a quire; and that I also suggested much the same of f33v and f40r in 2006 (Curse, p.70), though for a different reason.

In the 1970s, Captain Prescott Currier noted that the Voynich Manuscript’s text seemed to contain two separate ‘languages’ (“A” and “B”), each containing sub-languages that varied yet further. For him, the language differences were primarily statistical rather than linguistic: to tell what we now call ‘Currier A pages’ & ‘Currier B pages’ apart, he observed that (using the EVA transcription):-

“(a) Final ‘dy’ is very high in Language ‘B’; almost non-existent in Language ‘A.’
(b) The symbol groups ‘chol’ and ‘chor’ are very high in ‘A’ and often occur repeated; low in ‘B’.
(c) The symbol groups ‘chain’ and ‘chaiin’ rarely occur in ‘B’; medium frequency in ‘A.’
(d) Initial ‘chot’ high in ‘A’; rare in ‘B.’
(e) Initial ‘cTh’ very high in ‘A’; very low in ‘B.’
(f) ‘Unattached’ finals scattered throughout Language ‘B’ texts in considerable profusion; generally much less noticeable in Language ‘A.’“

Similarly, he thought that the writing seemed to have been done by at least two hands (specifically, a larger, rounded hand he called “1” [mainly on A pages], and a more cramped, tighter hand he called “2” [mainly on B pages]), which he was convinced were those of at least two different people. He also pointed out that certain Voynichese letters appeared to have a very position-dependent behaviour, and that a line of text seems to be a functional unit in some way.

I would argue that Currier’s work has been arguably the single most influential piece of Voynich research of the last few decades, because it in effect erects a pragmatic conceptual framework for working with the Voynich Manuscript that every researcher who follows should strongly bear in mind, if not actually use.

So when I get told about so-called Voynich ‘research’ that treats the entire Voynich Manuscript as a uniformly homogenous linguistic entity (i.e. ignoring Currier completely), I give a little sigh of muted exasperation and move swiftly on. This is simply because Currier’s languages are to the Voynich manuscript what the Gillogly strings are to the Beale Ciphers: hence any claimed explanation or decryption that fails to account for Currier’s raw (yet actually rather unexpected, if you think about it) set of statistical observations is simply doomed to failure, period, even 40 years on.

I think it’s important to point out that Currier wasn’t some technical-minded Army codebreaker doing a bit of Voynich moonlighting: having graduated in Philology (with a focus on Romance Languages) from the University of Washington, he was surely perfectly placed to contribute a balanced analytical insight into the elusive internal structure of Voynichese. Hence I think the real reason that Currier’s work has been so influential in the field is that he really cared about what he was doing, and that he wanted to make a constructive, positive difference to Voynich studies.

Sadly, Currier’s insights failed to inspire a community-wide statistical assault on the Voynich Manuscript: researchers trundled on with their existing ad hoc studies, perpetually reinventing wheels – the big red revolution bus never arrived at the Voynich stop. The only obvious difference was that at least a few sensible people (Rene Zandbergen, Mark Perakh, etc) did manage to do statistical tests on A and B pages separately, which is a start, I guess… but only a start.

But because Currier restricted his work to statistical observations, he never built his framework up into the kind of thing Annales historians call a problematique, i.e. a fully rounded research question that drives future research forward. It’s all very well cleverly spotting the presence of different languages (some people prefer to say “dialects”, but it’s an open question) within the text, but that does beg some rather big questions, so-called “elephants in the room” that everyone can see but nobody talks about:-

* Why are the different languages fragmented across the document, often mixed up within a single quire?
* What gives rise to all the variation both within Currier A and within Currier B?
* Why was there a need for multiple languages at all? Why not just stick with Currier A?

Fast forward to 2013, and I think we can answer at least one of these questions, and provide reasonable (if tentative) answers to the other two.

Firstly: the simple reason that the Voynich languages are in disarray appears to be that the bifolios themselves are in disarray. I and others have uncovered numerous different codicological artefacts that strongly suggest the initial gatherings were disrupted, bound, rebound, and indeed misbound; and there is even specific evidence that quite a few bifolios ended up reversed relative to their original facing direction (i.e. folded back to front across the central crease).

Essentially, as the bifolios themselves were scrambled, so too were the languages: which is why A bifolios and B bifolios appear juxtaposed within individual bound quires. Yet given that there are large homogenous stretches of A and B bifolios, it seems likely that the scrambling wasn’t absolute – while I don’t think the Voynich bifolios were ever blown down a street in the wind, I do believe that what we see arose from a combination of planned shuffling (e.g. moving the large multi-panel bifolios towards the back and binding them there) and unplanned shuffling (binding breaking on some quires, spilling the bifolios onto the floor).

Secondly: I strongly believe that the structural and palaeographic differences between A pages and B pages tells a strong story of two major writing phases (let’s call them the “A phase” and the “B phase”). Identifying different composition phases through close reading is the kind of thing that modern historians do all the time, so this isn’t a fundamentally new approach: the only nuance here is that rather than close textual analysis (for Critical Reading) or art technique de-layering (for Art History), we’re instead looking at a cryptanalytical close reading. But then again, isn’t that what Currier was hoping for?

Note that I’m not speculating here about why there were two writing phases: at this point it’s enough just to identify them and give them names. But it does point to some interesting questions about why there should be both Herbal A pages and Herbal B pages, and what the difference between them might turn out to be (a topic upon which I’ve previously speculated more than enough for one lifetime, some would say).

Thirdly: within each of the A & B writing phases, I believe that the variations in the statistics will turn out to have arisen because of cryptographic evolution during each phase. By this, I mean that the core cryptographic system in use at the outset of each phase evolved during the various writing phases, as the author(s) finessed the system to work around specific cryptographic challenges encountered along the way, and so ending up a very different beast at the close.

I suspect that this will prove to be a set of “ratchet” effects, in that once changes were made to the system they would probably tend to stay in place until they in turn were replaced or finessed. I therefore believe that the cryptanalytical challenge we face is working out the evolutionary curves that the A system and the B system traced out – quantifying and then mapping them as if the system driving them were a probabilistic Markov state machine, its configuration relentlessly evolving as the text flows from page to page to page.

As to the specifics of how we should do this, you’ll have to wait for the next post…