Writing a fake Voynich by rules: copies, habits and the link between words

[quimqu]

Some more data abou my skribe that might be of interest. I compared the relationship between last letter of a word and next word. Here is a comparin table:

Text	Whole word → next word	Last letter → next word	Last letter → next first letter	Share that is contact
Skribe (no meaning)	0.044	0.120	0.118	98%
Voynich (ZL3b)	0.078	0.127	0.124	98%
Latin, Ambrose (4th c.)	0.225	0.202	0.019	9%
English, Nashe (1594)	0.307	0.204	0.050	25%
Catalan, Tirant lo Blanc (1490)	0.725	0.456	0.104	23%
German, Simplicissimus (1668)	0.370	0.233	0.047	20%
French, La reine Margot	0.712	0.447	0.093	21%
English, Culpeper herbal (1652)	0.763	0.432	0.093	22%
English, Dorian Gray	0.702	0.398	0.077	19%
Manx	0.786	0.389	0.074	19%
Vietnamese (syllables)	0.951	0.281	0.040	14%
Chinese (pinyin syllables)	1.453	0.255	0.024	9%

Bits above chance (words shuffled within windows of 8), 30,000 words per text.

In every language I tried, the end of a word tells you something about the next word, but almost nothing about the letter it starts with: only 9 to 25% of that information is about the next first letter. The rest is grammar: a Latin ending like -us or -am announces what kind of word comes next, not how it is spelled. In the Voynich it’s the other way round: 98% of what the last glyph tells about the next word is just its first glyph. That’s exactly what my skribe does, and the sribe has no grammar at all, only a rule that the way a word starts depends on how the previous one ended.