08-01-2026, 11:05 PM
It's an interesting idea, but there are couple of properties of the VMS that don't jibe well with this thread's exact implementation of the cipher:
1. The token and word type length distributions. This particular sort of cipher does not alter plaintext word lengths, so the distributions of word lengths in the plaintext would carry over to the ciphertext. The quasi-binomial distributions within the VMS, centered on lengths of ~5-6 glyphs (depending on how one defines a single "glyph"), are unusual relative to typical Latin text.
2. Unaltered word beginnings and endings. As outlined at the beginning of this thread, this sort of cipher does not rebracket / respace the plaintext in any way that would obscure the beginnings or endings of the plaintext words. As a result, the words themselves would be essentially unaltered in how they statistically relate to each other. Word frequencies and skewed pair frequencies would strongly suggest the grammatical functions of—and the specific identities of—especially common words, from which the substitution scheme could be unraveled. That hasn't happened for the VMS.
But there are ways around this. As I did in the Naibbe cipher, you can strip the plaintext of its original spacing and then respace it—either randomly, as I did, or through some deterministic method—such that the respaced chunks are 1-4 letters long and that the spaces don't necessarily correspond to plaintext word endings and beginnings. In addition, multiple cipher-focused studies of the VMS have suggested verbose encodings (where "verbose" means that a single Latin alphabet letter maps to one distinct string or multiple distinct strings of Voynichese glyphs). This 2022 study specifically suggests a steganographic cipher in which one letter is often represented as a string of 1-3 Voynichese glyphs: You are not allowed to view links. Register or Login to view.
1. The token and word type length distributions. This particular sort of cipher does not alter plaintext word lengths, so the distributions of word lengths in the plaintext would carry over to the ciphertext. The quasi-binomial distributions within the VMS, centered on lengths of ~5-6 glyphs (depending on how one defines a single "glyph"), are unusual relative to typical Latin text.
2. Unaltered word beginnings and endings. As outlined at the beginning of this thread, this sort of cipher does not rebracket / respace the plaintext in any way that would obscure the beginnings or endings of the plaintext words. As a result, the words themselves would be essentially unaltered in how they statistically relate to each other. Word frequencies and skewed pair frequencies would strongly suggest the grammatical functions of—and the specific identities of—especially common words, from which the substitution scheme could be unraveled. That hasn't happened for the VMS.
But there are ways around this. As I did in the Naibbe cipher, you can strip the plaintext of its original spacing and then respace it—either randomly, as I did, or through some deterministic method—such that the respaced chunks are 1-4 letters long and that the spaces don't necessarily correspond to plaintext word endings and beginnings. In addition, multiple cipher-focused studies of the VMS have suggested verbose encodings (where "verbose" means that a single Latin alphabet letter maps to one distinct string or multiple distinct strings of Voynichese glyphs). This 2022 study specifically suggests a steganographic cipher in which one letter is often represented as a string of 1-3 Voynichese glyphs: You are not allowed to view links. Register or Login to view.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)