Hitherto, browsing through the Voynich manuscript always was a tedious affair, with people either using Beinecke‘s fairly awkward library tool, or perusing the image gallery they had set up at home for themselves. Either way, discussing matters regarding the VM has always been inconvenient when referring to details beyond a mere folio number: One had to refer to “the weirdo character on the second line of f213r”, or such, to communicate special points of interest. Of course, this is prone to misunderstandings. The fact that the VM mailing list doesn’t allow attachments helped to compound the difficulties.

Showing a detail of f77r with the URL 'http://www.jasondavies.com/voynich/#f77r/0.144/0.467/5.00'

Now, Jason has taken matters in his hands and has set up a website with a tool called The Voyager which contains the high-res scans of the Voynich, and a simple interface to navigate through the various images and zoom in and out of them, much like Google Maps. So far so good, but the real hoot of it all is the fact that the current settings for viewing an image are stored in the URL of said page. This means that pointing out a particular detail of any of the folios to someone else on the web is as simple as moving to the detail in question, copying the URL from the address line of the browser, and sending this URL to your other party who will simply paste it in their address field — ta-dah!

Thanks, Jason, for a simple, elegant and powerful little piece of software! This really rocks.

For example, to a high degree of certainty, we will have gotten our transcription wrong somewhere; not in the sense of writing down some individual error, but of consistently misidentifying letters as being different or seperate where they are truly the same or compound, or vice versa. But as long as we don’t know whether “iiin” is one letter or four, whether “ch” is one letter, two different letters or just “cc” in disguise, and as long as we can’t even know for sure whether there’s one, two, or four different gallows, all our estimates about character frequencies and word lengths are on very shaky ground. And hence, all our results.

So, while I’m all for statistical tests (After all, it’s all we’ve got, right?), and while I’m wishing for a a “universal statisticator” which would spit out the essential statistical parameters for various ciphertext candidates, I recommend taking such results with a grain of salt.

Take for example something as simple as a monoalphabetic substitution cipher.*) If someone employed that cipher and used the VM alphabet for the ciphertext, we’d probably despair in something even as simple as that, because our prime tool in that case, viz. statistical frequency analysis, would fail for sure as long as we fed it with a transciption which misidentified the ciphertext character set.

Likewise, wordlength distributions etc. are all prone to be distorted due to systematic transcription errors. So, while I agree that statistics can give us valuably hints, I’d regard them as circumstantial evidence, but not as hard facts, and if one test does give results which don’t put a theory in accordance with the VM, I wouldn’t dismiss the theory immediately, if the rest of the story looked good.

As for Zipf’s law, which is so often quoted in the context of the VM — I’ve got my hesitations about that in particular. IIUC, we don’t really know what it means if a distributions does or doesn’t obey Zipf’s law, and since random texts and even the sizes of cities can follow Zipf’s law, I wouldn’t assign too much importance to this test.

So, my advice is, stay tuned, but don’t stand with baited breath.

*) I know, the VM is not a monoalphabetic substitution.

Slightly edited from a post to the Voynich Mailing List.

When one follows the developments around the VM, research into it seems to be like a stream, where new ideas and theories appear somewhere upstream in the distance to come closer, and let themselves be examined more fully while they drift by you, before they follow the water downstream and finally vanish in the dusk of forgetableness. Rarely one of them will leave so much as a beacon behind.

One repeated pattern that seems to crop up time after time, and goes mostly unnoticed by even the old-time and seasoned Voynicheros, is the more or less explicit assumption that one VM letter is the equivalent to one plaintext letter, and that one VM word corresponds to one plaintext word.

Now, I can see where this would come from: Of course it’s natural to assume, it’s the most simple and straightforward way to do it (Which, in itself ought to be a warning sign: If the VM was enciphered “simply and straightforward”, it would have been solved long ago…), it’s the way ciphers were done in period, and it lends itself readily to easy analysis.

Unfortunately, in all probability this is not the way the VM was cooked up. Let’s look at a few of the arguments against this case:

• The main body of the VM consists of a rather limited character set. Depending on your choice of transcription, you’ll arrive at somewhat around 17 different characters, which make up the better part of the VM. Admittedly, there is a large number of “minor” characters, but these are fairly rare. Now, assuming that the VM has a European medieval background, the character sets which are conceivable for the plaintext are latin, greek, cyrillic and perhaps arab or hebrew letters. But unfortunately all of these writing systems sport more than 20 different characters, plus possibly numerals, interpunction and other special characters. The VM character set simply appears to be too small to accomodate a complete alphabet.
• Words show a very strong internal structure. VM words are far from random, au contraire: While nobody has as yet been able to completely determine which rules are underlying their composition, it’s pretty obvious that there are rules which are followed fairly coherently. To my knowledge, there are no European languages and only very few at all in the world which have such a determined set of rules for word composition, whether they are being written in their original alphabet or in that of the VM. What is worse: If in addition to switching to a new alphabet the author has employed an enciphering system (which he obviously has, because if the VM was a simple substitution cipher, we would have solved it), this would contribute to reducing the word structure and make it appear more random. (The only schemes which might increase the word regularity would be something like a transposition cipher with alphabetical sorting, which is conceivable, but wouldn’t exactly render the task of deciphering easier even for the intended audience.)
• One could argue that the VM author dropped eg the vowels from the plaintext before enciphering, hence reducing the number of different VM characters required. But this in turn would have reduced the word regularity. Bear in mind that the EVA transcription allows the better part of the VM to be read out aloud, which shows that there is in the VM ciphertext a structure retained which assigns to the VM letters functions comparable (but probably not identical) to “consonants” and “vowels”.
• All of the above work together to render an information content in the VM of roughly 2 characters/ciphertext word; obviously far too little for any natural languages.
• No words in the VM have been identified which seem to have the function of particles in natural languages, like articles, “and”, “or” and similar.
• On a more subtle level, there also appear to be things going on with decreasing average word length towards the end of a line, which don’t sit well with a 1:1 correspondence.

So, all in all this is why I’m convinced that the VM is something different than a simple substition or even a transposition cipher. Something else is going on, and the enciphered information is categorized in letters and grouped in words, but these ciphertext letters are not plaintext letters, and the ciphertext words are not plaintext words.

Strokes, anyone?

I’ve received some comments that people were having problems understanding what the whole Stroke Theory is really about. Admittedly, I’ve contributed fairly to this confusion by letting the theory develop and reporting the “increments”, but never having provided a concise overview of the status quo.

I want to make amends for that and have written a short PDF document which outlines the basic tenets of the ST and explains some features and consequences of it. I will keep this document updated, so that it will always reflect the current state of affairs, and you don’t need to wade through pages of history.

The actual works of the enciphering system according to the ST is only three pages, the rest are essentially discussions of special properties and background information.

Here goes.

And right now I’m hacking away at a little tool which should settle the question of the Stroke Theory once and forever.

But it could be a little while until it’s sophisticated enough, up and running…

“If at first you don’t succeed, use a bigger hammer.” The sad thing is that while one supposed side effect of this would be to make the project and VM research more public (in the vein of the SETI project), and hence to attract more people to it, I’m afraid a bungling approach like what is currently considered would have the opposite effect of exposing us to ridicule.

(Not to mention the fact that people seem to think that using a distributed approach would invalidate all previously gathered statistical information about the VM, like that it certainly is not a simple substitution, and that with a high degree of confidence one VM word is not equivalent to one plaintext word.)

So, everybody feel free to subscribe to the list (if you haven’t already done so). The list still is the central hub for information and “research” regarding the VM.

Thanks for the good job, Rich!

To avoid leaving the pour soul in the dark, the way I understood this quote is that in the course of scientific research, many a “beautiful” (elegant, simple, powerful) hypothesis is developed when one digs into a topic. Unfortunately, it happens quite so often that later experimental findings “rear their ugly head” by giving proof that the beautiful theory is beautiful, but false. And since we are not in the art department, truthfulness takes precedence over beauty.

In the case of the VM (admittedly not exactly a “science”), beautiful theories about its encryption get constantly slayed by statistical or systematic evidence to the contrary. Unfortunately, there is a tendency for people to twist the facts (or outrightly ignore them), rather than giving up the perceived beauty of their approach.

I had written a little tool to analyse the VM ciphertext and decompose it into the hypothetical “syllables” of the Stroke Theory, where each ciphertext “syllable” would represent one plaintext letter. This tool worked by constantly modifying the hypothetical syllable set and retaining those modifications which lead to an overall increase of “coverable text” (= ciphertext words that could be composed from the syllable set). The tool got “saturated” (ie, further changes would increase the overall coverage no more), when the syllable set could compose 66% and 74% of the ciphertext by volume, for Currier A and B, resp. This was interesting, but by no means convincing.

A different approach used a second tool which would synthesize ciphertext from inoccuous plaintext, according to the rules of the Stroke Theory. This test was simply on a qualitative basis, to see whether the ciphertext rendered this way would look anything like the VM, and, in my humble opinion it actually did.

It took me about a year to figure out that one could combine the two approaches, namely letting the analytical tool work on the results of the synthetical tool. In a perfect world, namely if the analytical tool worked correctly, this chase of one’s own tail should result in a 100% coverage of the plaintext.

Of course it didn’t. For instance, the plaintext used — the German 15th century “Weinbuch” — contained a few characters (like arab digits or umlaute) which couldn’t be properly transcribed in the synthesis and which could thus not be recovered.

Still, the result was that about 68% of the plaintext words by volume could be synthesized before the analysis program showed signs of stalling. This teaches us two things:

1. It’s — pardon my french — fucking close to the results for the VM, and
2. This can’t be due to special unencrypted characters alone.

Upon closer inspection, I found the culprit indeed. Namely, I had set the minimum syllable length in the analyzer to 2 strokes, and this is of course stupid. Letters like “I”, “l”, “o” will hardly require more than one stroke (and indeed required only one in my synthesis). Thus, these letters were effectively indecipherable, and this may well account for a good deal of the lost coverage.

This error of course would also hold true for the VM. (There is no reason why the VM author should have chosen to use a minimum of two strokes per plaintext letter.) And the fact that this programming error led to almost the same amount of lost coverage in the VM as in the synthesized text could be a hint that the same effects are in play, and that I’m thus on the right track by chasing my own tail…