The Naibbe cipher

[magnesium]

An excellent and thought-provoking approach! I am pleased to see that my own work on the manuscript’s text appears to have contributed, in part, to the inspiration behind it – my sincere thanks for the kind acknowledgement. 

 

Thank you for your work! Your 2022 conference paper was a major source of inspiration.

Although I have so far only been able to engage with the paper on a preliminary level, it strikes me as a promising contribution to the field.
At first glance, the chronological placement of the proposed method seems more plausible than that of my own PIII hypothesis. At the same time, however, the encryption mechanism appears significantly more complex.

Part of the motivation for the cipher's design was to try and find a more efficient way of generating more word types, while remaining broadly consistent with 15th-century homophonic substitution ciphers. Considering Voynich B, for example, we have ~23000 tokens to come up with >4500 unique word types. Generating and/or decrypting such a wide variety of word types using a PIII-style codebook requires a very large codebook. If there are separate ways of encrypting unigrams (as whole word types) and longer n-grams (using various lists of affixes), we preserve many of the appealing statistical properties of having individual tokens map to individual plaintext letters, while also dramatically shrinking the requisite size of the lookup tables.

I would expect that any attempt at decipherment using the complex codebooks will be extremely demanding. I nevertheless hope to explore the idea further during the upcoming days. 

I encourage you, and everyone, to try your hand at decrypting Naibbe ciphertext using the preprint's Tables 10-12. The end of the main paper includes a one-line decryption exercise, and Figure 4 consists of the first several hundred letters of Julius Caesar's De bello gallico. As a reminder, decryption proceeds as follows:

1. Check a given token against the unigram word types (Table 10). If there's a match, decrypt the token as a unigram. This will go much more smoothly than you might think because 70-80 unigram word types account for >40% of all tokens, so you'll see a lot of repeat appearances, and only a small number of possible prefixes are used to make unigram word types.
   
2. If the token doesn't match the unigram word types in Table 10, the token encrypts a bigram, where the token's prefix encrypts one letter and the token's suffix encrypts the other. Most prefixes are made using Table 8; most suffixes are made using Table 9. As you get a feel for the cipher, the prefix-suffix breakpoint becomes much easier to spot, in no small part because the prefixes and suffixes are meant to mimic intuitive Voynichese "prefixes" and "suffixes." For example, if a word type contains an e glyph, that e and everything to the right of it is a suffix. Similarly, if you spot a bench-and-gallows glyph (e.g., cth), that glyph and everything to the left of it is the prefix.
   
3. Once you have the bigram's prefix and suffix, you'll look them up in either Table 11 (prefixes) or Table 12 (suffixes).