I’ve got an idea but I’m not sure if its worth pursing further or not
its basically from Jorge Stolfi’s earlier work which hints at a numerical cipher
You are not allowed to view links. Register or Login to view.
suppose we take this as the case, even though repeated charecters pose a challenge we can dismiss them as an attempt to throw the decoder off.
this would imply that each word maps onto a number which maps onto the positional rank of a word in a lexicon/dictionary.
there’s plenty of challenges with this approach but solvable to a degree
first, Voynichese has roughly 13-20 distinct glyphs, so there are far more glyphs than bit positions. I quote claude here “The scheme isn’t base 20. It’s binary, with positions as digits, so the real constraint is that you need at least as many positions as the length of the longest code, and glyph classes beyond that have no role ”
Second is how to map out the characters/letters to numbers non arbitrarily. here I thought perhaps zipfs law can come in handy.
let’s suppose the key-book is a latin lexicon / dictionary. this would imply that the most frequent word in voynichese would map onto the most frequent word in that target language. And by locating its rank position in the dictionary key book we can decode the number of the voynichese word and so forth until we have a good mapping between characters and numbers. this also has its challenges but for the sake of brevity I’ll omit it.
another limitation is if the author themselves assigned the position ranks to the lexicon words and did that somewhat randomly but thats a bit far fetched. a more likely scenario is them making a lexicon of words that maps onto the vms but that would be similar to the above scenario.
also i don’t know much about 15th century dictionaries you’d need to assume that this dictionary is readily available or something. also its somewhat difficult to find the target language.
but generally speaking has this been tried before. do you think its fruitful or is it a dead-end?
Edit:
interesting observation from claude
“The lexicon-size issue. Stolfi’s toy code gives exactly binomial(9) only if the word list is about 512 words, since 9 bit positions numbers 512 items. With more words you need more positions, and the mean length grows (about half the positions plus one for the marker). Roughly 8,000 distinct word types, which I’m recalling and haven’t checked, would need about 13 positions and a mean near 7.5, but the observed mean is 5.5. So the distinct-word fit needs either few types or a different mechanism.”
its basically from Jorge Stolfi’s earlier work which hints at a numerical cipher
You are not allowed to view links. Register or Login to view.
suppose we take this as the case, even though repeated charecters pose a challenge we can dismiss them as an attempt to throw the decoder off.
this would imply that each word maps onto a number which maps onto the positional rank of a word in a lexicon/dictionary.
there’s plenty of challenges with this approach but solvable to a degree
first, Voynichese has roughly 13-20 distinct glyphs, so there are far more glyphs than bit positions. I quote claude here “The scheme isn’t base 20. It’s binary, with positions as digits, so the real constraint is that you need at least as many positions as the length of the longest code, and glyph classes beyond that have no role ”
Second is how to map out the characters/letters to numbers non arbitrarily. here I thought perhaps zipfs law can come in handy.
let’s suppose the key-book is a latin lexicon / dictionary. this would imply that the most frequent word in voynichese would map onto the most frequent word in that target language. And by locating its rank position in the dictionary key book we can decode the number of the voynichese word and so forth until we have a good mapping between characters and numbers. this also has its challenges but for the sake of brevity I’ll omit it.
another limitation is if the author themselves assigned the position ranks to the lexicon words and did that somewhat randomly but thats a bit far fetched. a more likely scenario is them making a lexicon of words that maps onto the vms but that would be similar to the above scenario.
also i don’t know much about 15th century dictionaries you’d need to assume that this dictionary is readily available or something. also its somewhat difficult to find the target language.
but generally speaking has this been tried before. do you think its fruitful or is it a dead-end?
Edit:
interesting observation from claude
“The lexicon-size issue. Stolfi’s toy code gives exactly binomial(9) only if the word list is about 512 words, since 9 bit positions numbers 512 items. With more words you need more positions, and the mean length grows (about half the positions plus one for the marker). Roughly 8,000 distinct word types, which I’m recalling and haven’t checked, would need about 13 positions and a mean near 7.5, but the observed mean is 5.5. So the distinct-word fit needs either few types or a different mechanism.”

