v4j

Voynich for Java (v4j) library

Project maintained by mzattera Hosted on GitHub Pages — Theme by mattgraham

Note 012 - Mind your Words

Last updated Oct. 11th, 2026.

This note refers to release v.15.0.0 of v4j; links to classes and files refer to this release; files might have been changed, deleted or moved in the current master branch. In addition, some of this note content might have become obsolete in more recent versions of the library.

Working notes are not providing detailed description of algorithms and classes used; for this, please refer to the library code and JavaDoc.

Unless differently stated, this note uses the Slot alphabet.

Please refer to the home page for a set of definitions that might be relevant for this working note.

« Home

Abstract

Over the decades, numerous researchers have identified highly unusual, structured orthographic patterns within the text of the Voynich:

  1. Models such as the Core/Mantle/Crust paradigm or the Slot Alphabet, which suggest rigid constraints on character ordering within word types.

  2. The anomalous frequency of characters in distinct textual loci

    First line of paragraphs:

    a. Gallows (except for ‘k’ and ‘K’) tend to appear more frequently in first line of paragraphs, preferably in the first token of the line. a. There is also a preference for ‘S’ to appear in first line of paragraphs. a. Some “endings” (e.g. ‘E’, ‘B’, and ‘n’) avoids the first line of paragraphs.

    First token in a line:

    a. Some characters, like ‘t’, ‘p’, ‘s’, ‘y’, and ‘d’ are over-represented at line start. a. Others characters, like ‘k’, ‘C’, ‘S’, ‘a’, and ‘r’ are under-represented.

    Last token in a line:

    a. ‘m’ is over represented, conversely, ‘l’ and ‘r’ are under-represented. a. For some clusters, ‘d’, ‘o’, ‘n’, and ‘y’ shows a significant deviation in their distribution.

  3. The presence of suffixes added to tokens that appear at beginning of lines.

  4. Changes in average token length relative to line position, where tokens tend to be longer at the beginning of a line or in first line of a paragraph and markedly shorter in the second position.

In this note, I want to explore the hypothesis that there must be some word types that appear only at some position of the text, which are responsible for the above patterns.

Methodology

For this note, the majority version of the Voynich was used; only the cluster text in running paragraphs (IVTFF locus type = P0 or P1) is considered, tokens containing unreadable characters were ignored.

In the first test, I divided tokens in buckets, one per position in the line, so all tokens appearing at the beginning of a line are in the first bucket, those appearing in second position in a line go into the second bucket, and so on.

I then calculate, for each bucket, the percentage of word types that appear significantly more often in a bucket, compared to the totality of other buckets.

The experiment has been conducted separately for each cluster in both directions (from beginning and end of line){1}, results are shown below{2}. For comparison, the experiment was also executed on a scrambled version of the text where tokens where shuffled at random.

Percentage of words appearing more often at a given position in line

Clearly there are word types that prefer to appear only at first or last position in a line. Whether an unusual percentage of word types tend to appear more frequently in second position of a line is more dubious, but it also seems to be the case.

A similar experiment has been conducted by bucketing tokens based on which line in a paragraph they appear; below the results. The spike at the end in PA is probably due to the fact this cluster contains paragraphs with few lines, so less words are in lines after the fifth, introducing volatility.

Percentage of words appearing more often at a given line in a paragraph

This also confirms that some words prefer to appear at beginning of line.

Conclusions

Our hypothesis seems to be confirmed: there are some word types that prefer to appear in first line of a paragraph, in first (and second?) or last position in a line. These special words might be responsible for the behaviors we see in the text that are summarized at the beginning of this note.

We can define a population of “interesting words” as those appearing preferably in a given position in the text:

  1. First line of paragraphs (in blue in the below illustration)

  2. First token in a line (in green in the below illustration)

  3. Second token in a line (in purple in the below illustration)

  4. Last token in a line (in red in the below illustration)

and contrast this population with the “standard population”, defined as the set of words that do not appear in any of the above positions (circled in yellow in the below illustration).

Example paragraph showing different area of the text

For example, in Herbal-A section (HA cluster) there are 3,302 tokens and 1,006 word types appearing outside specific loci (“standard population”). Considering the first line of paragraphs as special locus for the experiment, the “sample” will contain 1,293 tokens and 768 word types. The word “Sol” appears 46 times in the standard population (frequency=1.4%), assuming it will have same frequency all over the text, it is expected to appear in the sample 18 times. However, it is found in sample 29 times (frequency=2.2%); this deviation is enough to pass the binomial test (p=0.01) and mark “Sol” as an interesting word.

Here you can see the Voynich text with interesting words highlighted.

Notes

{1} Class InterestingWords was used for this purpose.

{2} The file Interesting Words.xlsx in this folder contains detailed results of the analysis, including diagrams.

« Home

Copyright Massimiliano Zattera.

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.