Liaisons in the Voynich text

[quimqu]

Edit 05/11: figures rechecked from scratch. The table was computed with a mismatch (glyph words counted with one filter, bare words with another) and some rows didn't match the glyphs they name; corrected below, together with points 1, 2 and 4.

I have been running more statistics on this, and four things came out that I find interesting.

1. The forbidden pairs work across the space. Inside a word there are pairs of glyphs that are very rare (oo, ya, dk...). When two words meet and would make one of those pairs (with a space in between), a glyph is often put to separate them, and that glyph makes both new contacts legal: it can stand after the word before and before the word after.

2. The scribe doesn't avoid the clash, he fixes it. If I take the connectors off and look at which word follows which, the clashes are as frequent as if the words were shuffled (27% in both cases). He never picks a different word to avoid the problem. The word comes as it comes, and the fix is done right there with the pen. But not after writing the word. The scribe knows beforehand that there must be a connector (we would see a considerable shrinkage of the space).

3. It is always the second word that adapts. 9 fixes out of 10 are done at the start of the second word, almost none at the end of the first. But the connectors don't all work the same way. o (or y) before a gallows is added: a gallows word gets it after a word ending in n, r, s, d, m nearly always (95%), and the o is common even without a clash (70%). q is the other way round: for the same words (okal, okedy, oky...), the form with q is the normal one in running text, 70% at the start of a line where nothing comes before, 70% after a word ending in o or y, and it is dropped after a word ending in n, r, s, d, m (28%), after a drawing gap (19%) and in labels (3%). So o is something you add to fix a contact, and q is something you take away. And how much q a page keeps is still a habit of the page and of the word.

4. Labels and restarts after a drawing behave as if a word ending in n or r was there: q is rare (2% in labels, 11% after a drawing, close to the 20% after n, r, s, d, m), and gallows words carry y, o or qo nearly always (94-99%). Ordinary line starts don't: qo is at 48% there, halfway between the two interior cases, and it changes with the section; and gallows words carry y, o or qo 75% of the time, less than anywhere inside the line (83% after o, y, l; 94% after n, r, d, s, m).

Here is which word starts get fixed and which don't ("how often" is the words with the glyph out of the same words with or without it):

word before ends with	next word starts with	fixed with	how often
o, y, e	a	d	89%
n, r, d, s, m	k, t	o or y	95%
o, y	o	q	63%
n, r, d, s	l	o	79%
anything	r, e	nothing	not avoided either

r and e are left alone: after the glyphs that can't stand before them inside a word, r- and e-words come as often as chance would give. y- and ch-words are a bit rarer there (about 0.7 of chance), and I don't see a clear glyph that fixes them. Why, I don't know.

So the space separates words but the glyph grammar runs through it. Labels and restarts after a drawing start in a fixed state; ordinary line starts don't. Whether this is sounds running on in speech or just a writing procedure that looks one glyph back, I still can't say.

Comments welcome, as always.