What are Voynichese words?

[vosreth]

It is only the meaningful hypothesist who is troubled by this: "Words and spelling and text should be uniform everywhere, in paragraphs, labels, radials, circulars, but it is clearly not so. Why, why?"

Perhaps I should step back from the "meaningful or meaningless" framing, because I think it conflates several rather different things.

Consider: a medieval logician writes qokeedy to mark determinate supposition, daiin for distributed. Perfectly systematic. But unless you know what terms those markers attach to, the content can't be deciphered, is lost, and therefore meaningless. Or another analogy; instructions for navigating a memory palace: "At the third arch, place the object." Grammatical, rule-governed, but completely meaningless to anyone else than one person.

So "meaningless" could mean for example:
a) Random noise
b) Systematic notation whose referents we lack
c) Deliberate deception

These different interpretations predict different statistics.

And what properties would such a system need in order to reproduce the observed boundary and continuation effects?

I'm not sure this is the right question, I believe there are many totally different ways to create these patterns. This is I think the approach @magnesium took with the Naibbe cipher - recreating some of the statistical properties of Voynichese in a plausible way. Does this bring us closer to understanding how the Voynich Manuscript was created? I don't know.

One thing that is almost certain is that if Voynichese is a cipher, it's a one to many cipher, one that allows encoding the same plaintext in many possible ways.

Wouldn't that imply that these properties have to be deliberately implemented in the encoding mechanism unless they arise from the plaintext?
For me it seems more likely that the text is encoded via many-to-one cipher, for example all nouns collapsing to generic o.

I ran one of Greshko's reference Naibbe ciphertexts through similar analysis. Naibbe encrypts letter-by-letter with random table selection, implementing the Zattera slot grammar for word-internal structure. Crucially, these reference texts encrypt real prose: Dante's Divina Commedia, Pliny's Natural History, Grosseteste's De sphaera, and a medieval alchemical herbal. Meaningful Latin and Italian, not gibberish.

Results (naibbe_Cleaned_52_07_10_word_lines.txt from You are not allowed to view links. Register or Login to view.):

Glyph-position within words:
- q at word-start: VMS 98.8%, Naibbe 100%
- q→o: VMS 97.6%, Naibbe 98.9%
Both match. That's the slot grammar doing its job.

Cross-word transitions:
- VMS q: 24.4% After y-final, 8.2% after l-final
- VMS ch/sh: 17.4% After y-final, 32.0% after l-final
- Naibbe q: 19.1% After y-final, 20.2% after l-final
- Naibbe ch/sh: 17.4% after y-final, 17.4% after l-final
The ch/sh:q ratio after y-final is 0.71 in VMS, 0.91 in Naibbe. After l-final it's 3.88 in VMS, 0.86 in Naibbe.

VMS shows a fivefold shift depending on what came before. Naibbe is flat as Norfolk. Why? Naibbe doesn't know what it just wrote. Each word is encrypted fresh. The randomness washes everything out. And here's the rub: meaningful plaintext doesn't help. Dante's tercets and Pliny's botanical observations go in, flat distributions come out. The content doesn't shine through.

The e-gradient:
Same story. After ch/sh contexts:
- Naibbe: e 52% → ee 31% → eee 42% (no gradient; eee actually rises)
- VMS: e 65% → ee 27% → eee 10% (monotonic decrease)

What this suggests:
The VMS wasn't generated by purely local random encryption. It shows cross-word dependencies and systematic gradients that don't emerge from Naibbe-style processing, even with structured plaintext underneath. This doesn't tell us what the VMS is. But it rules out option (a). Whatever system produced it tracked context somehow. Whether that's a sophisticated cipher, a compositional system, or something else remains gloriously open.

I can see a possible explanation for the differences between paragraph text and non-paragraph text ( labels, radials, circulars ). And it is once again consistent with the hypothesis that the manuscript is meaningless.

Labels, radial text, circular text would have been written after the drawings were completed, when a page was nearly completed. Radials and circulars would also have needed the writer to turn the page.

The writer, perhaps because drawing was not his forte or because the need for turning broke the momentum of his work, may have been in a different mindset when the time came to do the labels, radials, circulars. Perhaps he just did not have the same motivation for the astrological charts, and turned to the comfort of easy-to-write words. It is a psychological trick. When we are annoyed we do things differently.

Paragraph text, however, was less complicated. The writer just sat and wrote, line after line, a continuous stream without stopping, a whole page in one uninterrupted sitting, not having to shift and turn.

On scribe psychology: Naibbe shows what "local choices without context tracking" looks like. Flat. The VMS patterns hold across sections and hands. That's hard to attribute to mood, unless the mood was uncommonly methodical. If you subscribe to multiple scribes, this becomes even more uncommonly methodical.