What if every other word of the 340 is mirrored backwards? Meaning certain stretches of characters or rows are fully flipped as the characters are horizontally transposed (ie, ">T<H" becomes "H>T<").
This could explain:
[list=1]
[/*:m:3nexsngx]
[/*:m:3nexsngx]
[/*:m:3nexsngx]
[/*:m:3nexsngx]
[/*:m:3nexsngx]
[/*:m:3nexsngx]
There are definitely lots of promising mirroring patterns to test out. It would be wonderful to automate a search for bigram frequency improvements and sequential homophone fidelity increases as different portions of rows or "words" are reversed, similar to how searches have been done across different periods. I was able to quickly find repeating bigram totals of 31, 37, and even 46 with three of the first ten mirroring patterns I tested out by hand. Mirroring the entire bottom-left quadrant seems especially promising. Can anyone find a better pattern?
One could generate n-grams from text where every other words is mirrored or whatever you want. I am pretty sure I tried that a while ago, glurk once brought up a similar idea. AZdecrypt supports batching of n-grams if needed.
Unrelated to this, I did try a large 6-ngram file with half backwards words. Unfortunately, that file can’t solve the 340 if my mirroring hypothesis is correct.
I’m proposing that the symbols themselves are horizontally flipped, not just a rearrangement to their horizontal order.
The z340 alphabet seems to be missing several backwards symbols. But perhaps they’re only missing from our current point of view? Mirroring portions of the text can increase (or decrease) the alphabet by up to 8 symbols.
What if all characters had an individual mirror, including characters that look the same both ways? For instance, a "forward" T, could represent a different plain text letter than a "reverse" T. Just like a forward R most likely is different from a reverse R.
In order to correctly decode the cipher, you would have to know if the column (or row) was mirrored.
If every other row were mirrored, the adjacent "+" symbols could represent different letter’s. I counted up the characters that could fall into this category a while back. I would have to look for my notes to see the exact number, but there were quite a few.
I’m proposing that the symbols themselves are horizontally flipped, not just a rearrangement to their horizontal order.
Oh I see. If not more than 20 to 25% of the symbols are mirrored then the "substitution + sparse polyalphabetism" and 6 or 7-grams should be able to get it.
I wonder if your hypothesis can be checked with statistics. Perhaps the cycles may react more favorable while mirroring symbols. Or put a hill-climber to work that scores cycles with symbol mirroring operations. The results could be tested against a random set of symbols such that the symbol "p" not becomes the symbol "q" but another random symbol such as "W". Then "p" can only become "W" and vice versa. Not sure if I am making sense.
Hey beijinghouse,
Can you show what you did to get 46 bigrams?
To get 46 repeating bigrams, I resized the 340 into a 5 x 68 grid. Then I mirrored every grouping of 5 that didn’t have a bigram. Then I unmirrored a handful of these swaps in the 1st half back because I could see they would match new bigrams formed in the 2nd half.
I was using the webtoy transcription since I could mostly case swap to produce mirroring. T/t, <>, and a handful of numbers still need special handling.
