The Voynich Manuscript: Emergent complexity in hoaxed texts

By Gordon Rugg

This article is a short summary of an article by Laura Aylward and myself, originally published online in 2004. The full version is here:

Our article is about producing meaningless gibberish text using the table and grille method, with a view to producing text similar to that in the Voynich Manuscript. We found a variety of complex side-effects from various ways of using the table and grille method, which would affect the statistical properties of the output.

The next section of this article consists of direct quotes from the Aylward and Rugg article, interspersed with my notes.

I’ve shown the quotes as boxed images, to distinguish them from the notes.

Here’s an extract from the summary section.


A key point here is that we’re saying that some unusual properties will arise in text produced using tables and grilles, as unplanned emergent properties. I’ve discussed elsewhere on this blog how it’s possible for complex statistical properties to emerge accidentally from simple processes.

One issue that particularly interested us was tables that had been filled in very systematically, as described below. These turned out to produce some very interesting outputs.

structure and characters

This is saying, among other things, that you’re likely to get statistical differences between words produced using different parts of the same table, with the precise differences depending on how the table was filled in. You might, for example, produce longer words towards the start of a line of the gibberish text that you produce than you produce at the end. This could arise completely unintentionally; it’s perfectly possible that a hoaxer might produce this effect without noticing it.

This can affect the statistical properties of the text produced, both within a single line, and between different lines, depending on whether the table was filled in column by column or row by row.

The same effect would operate at a longer range, across entire multi-page sections of the text produced, if each section was produced using a different table that had been filled in with a different degree of structure. This is described below.


It has long been recognised that the text of the Voynich Manuscript can be divided into versions of two “dialects”. It’s not a clear-cut distinction; some sections show elements of both “dialects”. In another article on this blog, about visualising textual structures in the Voynich Manuscript, I show how the syllable distributions across sections of the Voynich Manuscript are more different than the difference between samples of text in German and English.

The sections of the Voynich Manuscript vary noticeably in the length and the repetitiveness of the words within them. We found that this type of variation in length and repetitiveness arose as an emergent property when using highly structured tables to generate meaningless gibberish text, as shown below.

Note: in all the following examples, the software generate ten words in each line, for consistency. Some words are “blank” as shown by two consecutive full stops. This occurs when the grille reveals only empty cells, which happened quite often.

Note: in all the following examples, the tables were filled in using the list of common Voynichese syllables and syllable frequencies, including blanks, compiled by Stolfi. Only the degree of structure in the tables was varied.


The example above showed sudden variations within the text produced by a single table. There are also variations across tables, which tend to reflect the degree of structure in each table, as shown below.

table examples

The tables with a medium degree of structure tended to produce longer words, words that were more consistent in length, and fewer “empty” words where the grille revealed only blank syllables. The tables with a high degree of structure tended to produce shorter words, more variation in word length, and more “empty” words.

This would be reflected in the entropy values, etc, that could be calculated for each text. Since we could systematically manipulate the repetitiveness, length, consistency of word length, etc, it was reasonably likely that we could eventually produce text with the same statistical properties as Voynichese via trial and error if we so wished.

Our conclusions about using tables with a high degree of structure are as follows.

text properties

Note: This refers specifically to text produced using highly structured tables. Tables with very low structure will produce text with very different features, usually the converse of the ones listed above.

In conclusion, this article describes variations and emergent properties in text produced using different varieties of tables and grilles, and shows that these emergent properties can produce features that are very similar to the features in the text of the Voynich Manuscript.

The full article is here:


6 thoughts on “The Voynich Manuscript: Emergent complexity in hoaxed texts

  1. Pingback: The Voynich Manuscript: Non-random word sequences as a byproduct of hoaxing | hyde and rugg

  2. Pingback: Our main posts: An overview by topic | hyde and rugg

  3. Pingback: Voynich articles overview | hyde and rugg

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

  5. Pingback: The Voynich Bombe | The Voynich Bomb

  6. Pingback: The Rugg and Taylor “Cryptologia” article on the Voynich Manuscript | hyde and rugg

Leave a Reply

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

You are commenting using your 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