The following warnings occurred:
Warning [2] Undefined variable $can_link - Line: 6 - File: showthread.php(1658) : eval()'d code PHP 8.5.12 (Linux)
File Line Function
/inc/class_error.php 157 errorHandler->error
/showthread.php(1658) : eval()'d code 6 errorHandler->error_callback
/showthread.php 1658 eval




Liaisons in the Voynich text
#50
hansolothe3rd wrote:If I merge the alternants in Currier B, only about 18% of the last-glyph -> next-word information goes away (0.25 -> 0.20 bits per word). That's twice what a random merge of the same size removes, so it's real, but most of the coupling is somewhere else.

Thanks, this is very good work, and it fits what I have.

I think I can say where the other 80% lives: it is still at the start of the next word, just not only in qo/o and y/gallows.

I split the next word in three parts, each one given the one before: its first glyph, its ending (last two glyphs) and the word itself. Then I measure how much knowing the previous word helps to predict each part (gain over unigram, nats per word, out of sample, word pairs inside the line).

totalfirst glyphendingthe word itself
Voynich (ZL)0.1790.149 (83%)0.0210.002
Latin0.2290.051 (22%)0.157 (69%)0.025
Italian0.8100.215 (26%)0.432 (53%)0.152
my skribe (no meaning)0.1200.1100.0050.000

In the languages, the previous word mostly tells how the next one will END (grammar, mostly through short function words). In the Voynich it tells how the next one will START, and almost nothing about which word it is. Your qo/o and y/gallows alternants are part of that start, but there are other families too: after -y the next word goes to qok-, after -n to ch-/sh-/o-, after -r/-s to a-. I checked it with your kind of measure (information of the last glyph of the previous word about the next word, pairs inside the line, minus the same with the words shuffled inside the line). My numbers are smaller than yours (0.096 bits in B), I think because of the baseline, so look at the shares:

Currier B, merge the next word bytypesinformation removedrandom merge of the same size
alternants only (qo/o, y+gallows/gallows)5357 → 495638%0%
the whole onset (q, o, and an initial d/s/l/r/y: qokeedy, okeedy, ykeedy → keedy; daiin → aiin)5357 → 376679%0%

So most of it goes when the whole onset goes. In Currier A my estimator gives almost nothing to begin with (0.004 bits), so I can't say anything there.

Two small things to add:
- the whole previous word says a bit more about the next start than its last glyph alone (A +0.010, B +0.016 nats). It comes from the family of the word, not from the glyph before the last one.
- frequent words are more tied to the glyph before them than rare words, already the first time they appear on a page. The public skribe does this now (SCRIBE.md §16).

And about the single scale: I tried a cipher where the allowed glyph pairs and the junction rule are built into the tables. It respects every pair and has the Voynich junction, but it only passes 20-31 of my 88 tests. Pairs are needed but far from enough, the word has its own whole-word shape. That fits your #49.
[+] 1 user says Thank You to quimqu for this post
Reply


Messages In This Thread
Liaisons in the Voynich text - by quimqu - 03-11-2026, 10:06 AM
RE: Liaisons in the Voynich text - by nablator - 03-11-2026, 10:48 AM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 11:07 AM
RE: Liaisons in the Voynich text - by ololololo - 03-11-2026, 11:25 AM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 11:45 AM
RE: Liaisons in the Voynich text - by Bluetoes101 - 03-11-2026, 12:20 PM
RE: Liaisons in the Voynich text - by Bernd - 03-11-2026, 12:27 PM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 12:58 PM
RE: Liaisons in the Voynich text - by tavie - 03-11-2026, 02:34 PM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 03:45 PM
RE: Liaisons in the Voynich text - by nablator - 03-11-2026, 03:15 PM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 03:52 PM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 04:36 PM
RE: Liaisons in the Voynich text - by nablator - 03-11-2026, 05:35 PM
RE: Liaisons in the Voynich text - by tavie - 03-11-2026, 10:24 PM
RE: Liaisons in the Voynich text - by quimqu - 03-11-2026, 11:16 PM
RE: Liaisons in the Voynich text - by todihk - 04-11-2026, 12:15 PM
RE: Liaisons in the Voynich text - by ololololo - 04-11-2026, 12:59 PM
RE: Liaisons in the Voynich text - by quimqu - 04-11-2026, 09:28 PM
RE: Liaisons in the Voynich text - by todihk - 05-11-2026, 10:03 AM
RE: Liaisons in the Voynich text - by quimqu - 05-11-2026, 10:13 AM
RE: Liaisons in the Voynich text - by tavie - 04-11-2026, 11:16 PM
RE: Liaisons in the Voynich text - by quimqu - 04-11-2026, 11:33 PM
RE: Liaisons in the Voynich text - by JoJo_Jost - 05-11-2026, 06:41 AM
RE: Liaisons in the Voynich text - by quimqu - 05-11-2026, 10:32 AM
RE: Liaisons in the Voynich text - by Torsten - 05-11-2026, 02:26 PM
RE: Liaisons in the Voynich text - by quimqu - 06-11-2026, 09:09 PM
RE: Liaisons in the Voynich text - by nablator - 07-11-2026, 03:55 PM
RE: Liaisons in the Voynich text - by tavie - 06-11-2026, 09:25 PM
RE: Liaisons in the Voynich text - by todihk - 07-11-2026, 08:33 PM
RE: Liaisons in the Voynich text - by nablator - 07-11-2026, 09:37 PM
RE: Liaisons in the Voynich text - by quimqu - 07-11-2026, 10:42 PM
RE: Liaisons in the Voynich text - by quimqu - 07-11-2026, 11:15 PM
RE: Liaisons in the Voynich text - by nablator - 07-11-2026, 11:30 PM
RE: Liaisons in the Voynich text - by quimqu - 07-11-2026, 11:59 PM
RE: Liaisons in the Voynich text - by nablator - 08-11-2026, 12:08 AM
RE: Liaisons in the Voynich text - by quimqu - 08-11-2026, 11:31 AM
RE: Liaisons in the Voynich text - by todihk - 08-11-2026, 01:07 AM
RE: Liaisons in the Voynich text - by Mauro - 08-11-2026, 01:08 PM
RE: Liaisons in the Voynich text - by ololololo - 08-11-2026, 11:50 AM
RE: Liaisons in the Voynich text - by hbaraiki - 09-11-2026, 03:46 PM
RE: Liaisons in the Voynich text - by RadioFM - 13-11-2026, 04:54 AM
RE: Liaisons in the Voynich text - by nablator - 15-11-2026, 10:44 PM
RE: Liaisons in the Voynich text - by Mauro - 15-11-2026, 10:15 PM
RE: Liaisons in the Voynich text - by quimqu - 10 hours ago

Forum Jump:


Users browsing this thread: 2 Guest(s)