11 hours ago
Quick follow-up, and thank you both.
nablator, got it, thank you! I'll line your space classes up against my ink ruler and post here whatever comes out, agree or not.
Mauro, in information terms, kind of yes. Once you know how the previous word ends, the y-/qo- choice is mostly predictable, so it doesn't carry much on its own. But it's not random filler either. When I merged the alternant pairs (qo/o, y+gallows/gallows), Currier B lost about 18% of its word-boundary information, roughly twice what merging random glyph pairs costs. So to me it acts more like a spelling or liaison rule than like nulls sprinkled in to throw a reader off. Whether it meant anything to the scribe is a different question.
And an honest update on the "one scale" idea from my #45. I ran it, with the tests written down before running:
1. The unrepaired clash does get extra room in ink: a bare o- after -y gets about 0.06 box heights more space (p 0.002). That's quimqu's and nablator's result again, on a different ruler.
2. But it's no more than the general junction rule already predicts (p 0.06, not shown). So no extra repair-for-space trade-off on top of it.
3. The inside-word pair score does track ink spacing in Currier B (looked at, not sealed), but it didn't carry over to Currier A when tested sealed. And an earlier round of mine already showed inside-word counts predict which words follow each other worse than nothing.
So the single-scale picture mostly didn't hold up. Better to say so here than leave it hanging. Next for me is nablator's spaces against the ink gaps.
nablator, got it, thank you! I'll line your space classes up against my ink ruler and post here whatever comes out, agree or not.
Mauro, in information terms, kind of yes. Once you know how the previous word ends, the y-/qo- choice is mostly predictable, so it doesn't carry much on its own. But it's not random filler either. When I merged the alternant pairs (qo/o, y+gallows/gallows), Currier B lost about 18% of its word-boundary information, roughly twice what merging random glyph pairs costs. So to me it acts more like a spelling or liaison rule than like nulls sprinkled in to throw a reader off. Whether it meant anything to the scribe is a different question.
And an honest update on the "one scale" idea from my #45. I ran it, with the tests written down before running:
1. The unrepaired clash does get extra room in ink: a bare o- after -y gets about 0.06 box heights more space (p 0.002). That's quimqu's and nablator's result again, on a different ruler.
2. But it's no more than the general junction rule already predicts (p 0.06, not shown). So no extra repair-for-space trade-off on top of it.
3. The inside-word pair score does track ink spacing in Currier B (looked at, not sealed), but it didn't carry over to Currier A when tested sealed. And an earlier round of mine already showed inside-word counts predict which words follow each other worse than nothing.
So the single-scale picture mostly didn't hold up. Better to say so here than leave it hanging. Next for me is nablator's spaces against the ink gaps.


![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)