Replicating The Voynich Manuscript


Q: If the manuscript is a hoax, isn't it needlessly elaborate? Rudolph was pretty undiscriminating.
A: One possible explanation is that the manuscript was intended to fool Dee, not Rudolph. In this scenario, the manuscript would be sold by a third party (e.g. Puccius) to Dee, who was a dealer in rare books; Dee would buy it in good faith, and then sell it on to Rudolph. Dee was no fool, and was familiar with cryptography, so a hoax would need to be good to get past him. The tricky part of hoaxing a manuscript with tables and grilles is getting the method right; after that, producing a long manuscript is not much harder than producing a short one, and a long manuscript could probably be sold for more money.

Q: Can you reconstruct the original tables and grilles, and then re-create a page of the manuscript?
A: It's important to remember that there's a random element in the table and grille method, as well as a deterministic element. I mentioned this in the Cryptologia article, but should probably have emphasised it more. If you simply move the grille across the table in a systematic manner, then this produces regularities in the output text even when you switch to a different grille on the same table. For instance, if a row of the table contains the prefixes "qo, ol, qo, dy, qo" in the first five prefix cells, then every grille that goes along that row will produce five words beginning with "qo, ol, qo, dy, qo" respectively - they'll just be combined with different midfixes and suffixes. This effect is pretty obvious, and a key part of using tables and grilles to produce a hoax is that you need to break up such sequences (for instance, by skipping randomly up and down the table at intervals). This means that if you use the same table and the same grille on three separate occasions, you'll get different output each time. Laura Aylward has produced some good examples of this with her software, using a simple randomising algorithm - I'm happy to provide soft copy to anyone who wants it.

I've done some initial work on trying to reconstruct table fragments from the manuscript, and the results so far are most consistent with a grille being moved across a table with a fairly high degree of randomness (long story; I'll post details once I've worked it through further). If that's the case, then even if someone did recreate the tables and grilles, that would still not be anywhere near enough to reproduce a given page of the manuscript.

Q: Doesn't the output from the table and grille method simply look like Voynichese, rather than recreating Voynichese?
A: If the manuscript is a hoax produced using tables and grilles as described above, then the specific words in it are largely the product of chance, and the same hoaxer with the same tables and the same grilles would have produced a different text if they started over again. In that scenario, if you recreated the tables and grilles, then you would probably produce quite a lot of words which don't occur in the manuscript, and probably fail to produce quite a lot of words which do occur in it; in other words, whether or not you produce the same text as in the manuscript is not a key issue. Within this scenario, then Voynichese itself simply looks like Voynichese; this scenario postulates that Voynichese is a meaningless, semi-random assemblage of word fragments, without any external validity.

The situation would clearly be different if Voynichese had more of the features of a normal human language, such as Italian or Vietnamese; in that case, there would be more grounds for suspecting the existence of a discrete vocabulary which had external validity (in the sense that, for instance, different corpora of text would contain a high proportion of the same words). In that case, combining word fragments in a semi-random manner would be inadvisable. However, since Voynichese appears to be very different from normal languages (e.g. its lack of syntax, and its very constrained morphemic structure), and since there is no corpus of other material in authentic Voynichese, this appears to be a somewhat hypothetical point.

Q: Hasn't your work been rebutted?
A: There have been various objections to it, but I'm not convinced by any I've seen so far. The statistical objections are based on data from less than one percent of the possible permutations; the other objections appear to be based on misunderstandings. There's a brief list of alleged rebuttals here, with my responses, and a more detailed list here, with more detailed responses.

Q: Does it matter whether or not you used sixteenth-century techniques?
A: Yes. The table and grille method involves using a modified Cardan grille; the Cardan grille was not invented until 1550. This clashes with the apparent age of the manuscript the few datable features suggest a date between 1470 and 1520. This is consistent with the manuscript having been produced in the second half of the sixteenth century, and being made to look as if it dates from an earlier period. This does not prove that the manuscript is a hoax, but is consistent with its being a hoax. As far as I know, nobody has proposed a method which can produce text similar to Voynichese using techniques dating from 1470-1520.

Q: Why bother?
A: Codes are at the heart of modern security systems. When I started work on this project, there was a real possibility that some Renaissance genius had invented a type of code which our best codebreakers couldn’t crack. That was too tantalising a possibility to ignore.

Q: Why did you get interested?
A: I’m interested in difficult research questions, particularly ones where there’s a chance that everyone’s been making the wrong assumptions, or asking the wrong questions. This looked like a good challenge which I could do as a hobby.

Q: If the manuscript is a hoax, does it mean that all the previous research was a waste of time?
A: Quite the opposite. I couldn’t have found this solution without having previous people’s work to build on. Stolfi’s work and Philip Neal’s work were particularly important starting places, for example, and we all depended on the machine-readable versions created by the various transcription groups.

Q: Why would anyone bother to hoax something like this?
A: There are various possibilities: a student wanting to trick the experts, an alchemist wanting to impress possible clients with a book of secrets. The most likely explanation, though, is money: in 1586, Rudolph II paid an enormous amount of money for the manuscript, equivalent to tens of thousands of pounds. That would be well worth three or four months preparing the hoax.

Q: Who was the hoaxer?

A: The short answer is that it was probably Edward Kelley. The long answer is that the method I’ve found depends on a modification of a technique first described in the 1550s. It also depends on combining this technique with another approach which could either derive from Kabbalistic scholarship or, more probably, from Kelley’s work with Dee. The balance of probability is that it was someone working between 1550 and 1586, who was familiar with Kelley’s work, and who was capable of forgeries. Kelley is the best match for that description.

Q: What are you going to do next?
A: I’m planning to apply this approach to other difficult research problems, probably in medicine or in technology.

© Site design: Stephanie Dale 2003
Last updated: 14th December 2004 by ST