Hoaxing the Voynich Manuscript, part 7: Producing the text

By Gordon Rugg

The six previous articles in this series looked at the component parts of a hoax. This article shows how those components can be put together, to produce  the text for a large document consisting of meaningless gibberish. This process is much the same regardless of which script you use for that gibberish, and regardless of which illustrations you use. The script and illustration issues are discussed in article 8, which I’ve already published

There are a few key points about this hoaxing process that are absolutely central to understanding why it gives new insights into Voynich Manuscript research. These points are:

  • This process  isn’t random.
  • This process isn’t deterministic – there isn’t an algorithm that would let a future researcher reproduce the text within a given page using the same table and grille.
  • This process produces numerous complex statistical regularities in the output text as completely unintended side-effects of a very simple production process.

This method is fast and easy to use. You can generate meaningless gibberish text as fast as you can write it down. I’ve produced quasi-copies of various pages from the Voynich Manuscript, where I’ve copied the original illustration, and generated the appropriate amount of meaningless gibberish text to match the amount of text in the original page. It consistently took about an hour and a half per page. More time spent on text within a page was balanced by less time spent on illustration and vice-versa, so each page took about the same time regardless of whether it was mainly text, mainly picture or a mixture.

At that rate, one person working alone could produce a document as long as the Voynich Manuscript (about 240 pages) in under ten weeks.

Here’s how the method works.

The starting point is a large table of meaningless gibberish. Here’s what a table populated with Voynichese syllables looks like.

Slide01Image copyleft Hyde & Rugg, 2014

The table has a structure, as follows. The table is divided into sets of three columns. Each of those sets contains one column that contains prefixes, another column that contains roots, and another column that contains suffixes. The image below shows the structure for a small section of a table. I’ve used colour coding to show the sets of columns, and the initials P, R and S to show which of those columns contain prefixes, roots and suffixes respectively.

Slide03Image copyleft Hyde & Rugg, 2014

The easiest way to fill the table in is one category at a time – for instance, doing all the prefix columns first, then doing all the roots, and then doing the suffixes. I’ve tried other ways, and this way is the least error-prone.

Here’s what the table section above looks like with some of the cells populated with syllables. I’ve used the EVA transcription of some common Voynichese syllables, to make them easier to read. Their relative frequencies are similar to the frequencies in Voynichese, but I haven’t tried to replicate the real frequencies exactly, because I wanted to construct a table fragment that demonstrated how some key points work.

Slide04Image copyleft Hyde & Rugg, 2014

The next figure shows the same section of table, completed.

Slide05Image copyleft Hyde & Rugg, 2014

Some of the cells are empty. That’s deliberate. It replicates the way that many Voynichese words don’t have prefixes, or don’t have suffixes, or don’t have roots, or some combination of all three.

That, incidentally, is one of the many odd features of Voynichese; quite a few of its words consist only of a prefix and a suffix, with no root in between. I don’t know of any real language that does this.

Selecting syllables

The next stage involves taking a grille, i.e. a piece of card with three holes cut in it, like this. I’ve shown the same grid pattern on the grille that I used on the table; this makes it easier to keep track of what’s happening.

green table

Image copyleft Hyde & Rugg, 2014

To generate some text, you start by putting the grille over the table, so that the holes in the grille show you the contents of three cells – a prefix cell, a root cell, and a suffix cell. For the illustration below, I’ve started with the grille at the top left of the table. I’ve shown the grille as grey and faintly transparent, so you can see the underlying table. The left hand side of the grille sticks off the end of the table, and isn’t shown in the figure.

Slide06Image copyleft Hyde & Rugg, 2014

This reveals the word ochedy, which is a perfectly respectable Voynichese word. You can now write this word down as the first word on your hoaxed page.

Moving the grille for the next word

This is the step that has most often been misunderstood. You now need to move the grille across the table to produce the next word. However, you can’t simply move it three cells to the right horizontally.

If you simply move the grille horizontally across each time, then you’ll end up with a regular sequence of syllables in successive words.

Suppose, for example that you use Grille 1 to produce a line of words from the row that starts with the rare syllable oqo (the third row down from the top). The first four words will start with the prefixes oqo olo y or respectively. I’ve indicated that with highlight on the figure below.

Slide07Image copyleft Hyde & Rugg, 2014

Now suppose that you use Grille 2, which has a different pattern of holes, to produce a line of words from the same row. You’ll get different roots and suffixes, but your first four words will still begin with the prefixes oqo olo y or respectively.

That’s a glaringly obvious give-away to anyone with any knowledge of code breaking. They’ll spot that sequence very quickly, and swiftly use it to unravel the structure of your hoax.

Simply using a regular sequence of moves up or down whenever you move across doesn’t help. Because that sequence of moves is regular, you’d still end up with a regular sequence of syllables. It would simply be a different regular sequence of syllables from the one you’d get if you moved the grille straight across horizontally.

What you have to do instead is introduce an element of unpredictability into the horizontal move. So, for example, you might move three cells across and one row up to produce the next word, and then move three more cells across and two rows down to produce the word after it. The key thing is not to have any regular pattern in those vertical moves that would produce an identifiable regular sequence of syllables.

This is one place where the very repetitive nature of Voynichese works on the side of the hoaxer. It’s easy to spot sequences of syllables that involve rare syllables, but it’s much more difficult for a code breaker to spot sequences that involve very common syllables.

The next figure shows this principle in action. The grille has been moved across three columns, and down one row, to break up the pattern. It shows the word qo, with no root or suffix. That’s your second word.

Slide08Image copyleft Hyde & Rugg, 2014

You now repeat the process, with another move three cells across, and an arbitrary number of cells up or down.

Slide09Image copyleft Hyde & Rugg, 2014

Handling repetition

This third move shows you another word consisting of only a prefix. It’s the word qo again.

So what do you do? You now have the same word occurring twice in a row. One option is simply to write it down and keep going. Another option is to move the grille further up or down so that you get a different word, and write that word instead of the second qo.

The “move the grille till you get a different word” option has obvious attractions. However, if you’re tired and bored after a couple of hours of generating material that you know to be meaningless gibberish, then there’s a fairly reasonable argument for the “just write it down and keep going” option.

Anyone familiar with Latin (which would include any educated western European in the fifteenth or sixteenth century) would know that Latin contains a fair number of reduplications and near-reduplications, where the same syllable is repeated for grammatical reasons. An example is Latin quisquis, meaning whoever, which is a reduplication of the more common quis, meaning who.

The amount of repetition in the Voynich Manuscript is well beyond what occurs in any known real language, but scholars at the time had no way of knowing what was plausible for an unknown language, and what was implausible. So, even if someone picked up on the amount of repetition, they would have no way of knowing whether it was implausibly high or not.

Scholars in the fifteenth and sixteenth centuries might speak a fair number of languages, but that’s a very, very different proposition from knowing the core findings of comparative linguistics and historical linguistics. Linguistics as an academic discipline isn’t about learning to speak lots of languages; it’s about the underlying principles of language in general. Linguistics students learn about the general principles of how languages operate, and learn about the rare features that only crop up in a few languages, as well as the features that occur in all known languages. A handful of languages, for instance, use whispered vowels; a lot of languages use tone to differentiate two or more words that consist of the same consonants and vowels; all languages have rules about word order.

So, returning to the production process, you choose your option and you write down whichever word that gives you, and you continue the process across the table till you reach the end of the table or the end of the line.

Handling shortfalls

If you reach the end of the line before you reach the end of the table, then the next step is easy; you just move your grille back to the first three columns of the table, a bit further down, and repeat the process. If your table is about forty rows deep, that’s usually more than enough to generate a page of text without having to change grilles.

Sometimes, though, you reach the end of the table before you reach the end of the line that you’re writing on the page. The grille quite often shows short words, and sometimes it shows empty cells for all three syllables. That can easily produce a batch of output text that isn’t long enough to fill a line on the page.

Again, you have various options. One is to use the grille in some way to generate more words. Another is just to make some words up out of your head, and write them down. If you’re just dealing with a few words, then this is easy to do, because Voynichese is so repetitive. It’s nowhere near so easy to do this for more than a few lines, because inventing text in your head is tiring.

That’s one big advantage of using table and grille; if you’re aiming to produce over two hundred pages of output, then the table and grille is much less mental effort.

The figure below shows what happens when you reach the bottom of the table. You take a different grille, with a different pattern of holes in it, and you start again at the top of the table, generating a new set of text from the same table. This time, we get the word okedy as the first word of the next batch of text.

blue grille second lineImage copyleft Hyde & Rugg, 2014

Further tables

There’s a limit to the number of different patterns of holes that you can use; if the holes are too many rows apart, then the grilles become tiring to use. The precise limit will depend on the size of the cells in your table, and the size of the table, but whatever the precise number, it won’t be enough to produce over two hundred pages from one single table. You’ll eventually need to produce a different table, which involves lots of scope for human error; I’ve blogged about the implications of that in various articles on this site.

Another major advantage of using table and grille is that you can have one or more assistants producing text for you at the same time using different tables and grilles, to speed up the production process and to save you from having to do the work. If the different  tables use roughly the same set of syllables in the same proportions as each other, then you’d expect the output text from the different tables to look pretty similar. However, you’d be wrong. The output is significantly affected by things like how regularly you populate the cells with the syllables. Also, when you’re working down your list of syllables that need to go into your new table, it’s horribly easy to overlook one syllable completely, so that the output from your new table doesn’t contain any examples of that syllable.

That would make sense of why the Voynich Manuscript contains at least two “dialects” – these could reflect the use of two master tables, and probably of other tables derived from those masters.

Conclusion and some further thoughts

So, that’s an overview of how to produce the text for a hoax. The basic principle is simple; you just move a grille across a table, writing down the words that it shows, and remembering to move the grille up or down in some non-systematic way each time you move it across three columns to show the next word.

It’s a simple process, but it produces a lot of accidental complexities.

For instance, in the table section above, the rare syllable oqo will usually only appear at the start of a line. However, if you reach the end of the table before you reach the end of the page of output text, and you generate the remaining text by moving your grille back to the start of the table, then you’ll produce a few cases where the syllable oqo also occurs near the end of a line, probably without even noticing that you’ve done it. From the viewpoint of a modern analyst, that would be a puzzling pattern – why does oqo usually occur at the start of a line, and sometimes occur near the end of a line, but never occur in the middle of a line?

Which is exactly the sort of pattern that occurs for some syllables in Voynichese. It’s a pattern that would be hard to produce via a schizophrenic rant or automatic writing in a trance or by making up text in your head, which is one of the reasons that the hoax theory was not seriously considered until the table and grille theory came along, providing an explanation for how such complex regularities could emerge as an accidental side-effect of a simple process. That, however, is another story, and one about which I’ve already blogged at length…


All images above are copyleft Hyde & Rugg, unless otherwise stated. You’re welcome to use the copyleft images for any non-commercial purpose, including lectures, provided that you state that they’re copyleft Hyde & Rugg.

Other articles in this series:









8 thoughts on “Hoaxing the Voynich Manuscript, part 7: Producing the text

  1. Pingback: Voynich articles overview | hyde and rugg

  2. What’s the difference between “random” and “non-systematic”? Even though your ‘hoax’ argument now seems to turn on this specific distinction, you haven’t (as far as I know) really defined how to tell them apart.

    A computer science definition would be fine.

    • I’m using “random” in the sense of “truly random” as opposed to e.g. “haphazard” or “arbitrary” or other non-deterministic processes. This has been a key part of the table and grille model from the outset, and was a key issue in Schinner’s choice of statistical test.

      I’ll put up a short article about this, since it’s an interesting point.

      • I look forward to the clarification. To my own compsci eyes, “truly random” is something of a straw man, in that it is so unbelievably hard to achieve in normal practice that it has almost no use as a practical attribute.

  3. Pingback: One hundred Hyde & Rugg articles, and the Verifier framework | hyde and rugg

  4. Pingback: 150 posts and counting | hyde and rugg

  5. When you “Get To The End” of your “Grill”, (where you say) make A_nother Grill & keep going etc… Just ROTATE IT (simple). [Flip It Over]…
    Top/Bottom, Bottom\Top etc. Double you system.


    • Another procedure to ‘keep going without writing out another table’, would be to permute the rows of the table.

      An easy way to do this would be to cut the table into several horizontal strips, each containing an arbitrary number of rows, and then reorder the strips, as and when needed.

      Such an economical scheme might, perhaps, appeal to a ‘lazy’ hoaxer (assuming this kind of permutation was available at the time of writing).

      4occ 🙂

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s