Harvard physicist's take on the VMS

[RadioFM]

But any encoding that required use of a physical device would make writing way too slow.  That is a general objection to the gibberish theory: it requires a complicated method, rigorously followed through the whole book -- but then why would the Author go to such trouble?

And I agree! I don't think wheels or any complex device was involved in the generation of the text, I prefer self-citation to that. Again, despite the limited amount of text used for the paper, Gaskell & Bowern's results providence evidence for the gibberish case as a natural tendency of some individuals; we have actual empirical data that is consistent with the gibberish hypothesis, and I can't stress "empirical" enough.

That may be true if we count each connected glyph as one letter.  But the various word models strongly suggest that some pairs of glyphs functioned as distinct letters.  Like che, ke, in, iin, and maybe combinations with o/a/y too.   Then the alphabet can easily have more than 23 letters.

Yes, word models such as yours describe glyph affinity, and when you start grouping combinations like those you mentioned and parsing them as single characters (thus increasing the alphabet), the text naturally shortens. The main problem with this approach is that there are some many instances of ambiguous parsing, that there's no way to systematically group characters together and get sensible word-to-word ("phrase-level") syntax. We have instances of chol, ol, lol, dy, dchdy, choldy. How do you group characters in dcholdy? dch-ol-dy? d-choldy? d-chol-dy? dch-oldy? The same parsing problem is present throughout the entire ms. There's just no way to parse it unambiguously without making all sorts of unfounded assumptions and rules as-you-go. And this is something that all solvers struggle with, no matter their target language. The glyph-to-glyph rules, although mostly consistent, are still riddled with so many weird exceptions that it doesn't make sense a single (or few) simple encryption/transliteration schemes were put in place and systematically followed throughout. There's so much unexpected flexibility despite bulk of the text being explained by word models like yours or EMS's.

My recollection is that the next-character entropy of Voynichese is lower at the start of the word, but remains more or less constant along it; whereas in European languages it drops along the word.  Thus the entropy per word of Voynichese turns out to be the same as that of European languages -- about 10 bits.  Is this wrong?

You're right, but I think you're mixing two observations that don't necessarily correlate one another. The next-character entropy of Voynichese is lower at the start/prefix of the vord and then rises, and the per-word conditional entropy is somewhat comparable to European languages (I don't know about East Asian langauges), but I'm not sure if one of those two metrics induce the other. Higher-order conditional character entropies (3rd and higher order) of Voynichese seem comparable to Germanic, Latin and Romance languages back when I computed them, too.

I don't see how we can have any expectation about the amount of text that it should contain.

I believe trying to match conditional character entropies can give a fair estimate in that regard, assuming the usual medieval cipher and language (Ger/Lat/Rom), unless we're dealing with say, a codebook or a system whereby single glyphs can act on the semantic (e.g. okeedy = yes, qokeedy = yesn't).

Most Voynichese glyphs are constructed from pairs of simple pen strokes: a left half which is either i or e, combined with one of 5-6 different strokes at right in almost all possible ways.(...) Thus the similarity of some glyphs to Latin letters is a not-at-all-unlikely coincidence. 

Oh come on! They even let a -que abbreviaton slip in there! There's no way it's all 100% pure stochastic coincidence it just so happens to look like Latin letterforms lmao.

Anyway, the way those glyphs are use in Voynichese implies that their phonetic values is not at all like those of the matching Latin letters.  Taking these at face value is a common mistake of beginner voynichologists.

Why would a European trying to transliterate an East Asian language devise a glyph system, clearly inspired by letterforms of his time/place yet for some reason phonetically not at all related, is beyond me. I find the gibberish or cipher hypotheses much more plausible, especially since some of the glyphs are also seen on ciphers from ca. 14-15th century, perhaps before.