From "decryption" to "translation"

[geoffreycaveney]

It wasn't really meant as a riddle. Also, the second example is 'closer' in a way, since the inverse of the operation also allows compression of real Voynichese.

However, this is *not* how the real MS text was generated. At best it could give a hint into the direction.

The reduction in entropy with respect to the Italian text is the result of  a verbose substitution.
The word patterns arise by chopping up the original text in syllables (sort of), in combination with the verbose substitution. A syllable in this case is defined by :
- a consonant cluster followed by a vowel cluster
I chose Italian because very many words end in a vowel, so the above definition of syllables largely (but not completely) keeps word boundaries intact.

The conditional entropy of the compressed Voynich text goes up to a staggering 3.1
The problem is with the single character entropy, which is 'only' 3.7.
It is not just the value that is still too low. It is the frequency distribution as a whole that is wrong, and this is not easy to fix.

This is very interesting. Especially the inverse operation generating compressed Voynich text with 3.1 conditional entropy. But I recognize the other problems that remain.

By the way, with any language, for words that end in consonants you can simply add a "dummy" or default vowel at the end of all such words, and treat the resulting syllable like any other syllable. In fact, when languages with final consonants are written in syllabary scripts, this is effectively what the script does to them. (Although when Mycenaean Greek was written in the Linear B script, it rather tended to omit the final consonant from the written representation entirely.)

However, I believe we must also consider the possibility of the Voynich MS scribe using a process that does *not* allow us to perform a direct inverse operation. For example -- and this was also true with Linear B -- using the same symbol to express two or more actual distinct phonemes in the language. (Linear B did this for /r/ and /l/, for /p/ and /ph/, and for /k/, /kh/, and /g/, for example.) The type of process you are using can never be used to turn Linear B back into Mycenaean Greek, because Linear B simply did not distinguish between Greek "korus, "gorus", and "khorus", for example.

I think it would be very interesting to do entropy and character pair distribution analysis on the Linear B texts. I would not be surprised if, due to the simplifications of Greek described above, they have statistics similar to Hawaiian and Voynichese. Actually no, the fact that syllables like "ko", "po", "to", "ro" are all different symbols, and do not share a common "o" character, would increase the entropy, etc. But converting Linear B into an alphabetic transcription may then likely produce text with statistics like Hawaiian and Voynichese.

---

Pilcrows and capitula were frequently in line with the main text, not just at the beginnings of paragraphs.


Also, a pilcrow sometimes looks like a letter (especially in the middle ages). The Capitula symbol was a "C". Sometimes it had a vertical stroke through it, but not always. Sometimes the only way you could distinguish it from the letter C was because it was in a different color. In other words, the C shape could be both a pilcrow and a letter. Some pilcrows look like t without the loops.

I've posted two blogs with examples:

You are not allowed to view links. Register or Login to view.
You are not allowed to view links. Register or Login to view.


In the second blog, there is a pic that shows how pilcrows sometimes stretch over more than one letter, just as some of the VMS gallows glyphs sometimes stretch over more than one glyph.


Here's an example of pilcrows that are in line with the main text:

Thank you for sharing JKP, these examples are very interesting. That second text you posted is incredibly difficult to read.