I have what I believe to be a Vigenere or similar polyalphabetic cipher (IC = 0.0459). However, the ciphertext is further encoded by 'Latin-like' letters, meaning I don't have the ciphertext alphabet (but have some idea of the plaintext translations. For example, the words for 'Water' and 'Ice' (using Japanese 'Mizu' and 'Kori' respectively) translate to the following symbol sets:
Where symbols are assigned Latin letters in order of appearance. There are plenty of these 1-word strings to draw from, but extremely few actual lines of text. I applied the Kasiski examination to the one actual string that I could find (shown below) using dCode.fr's page, and got the most likely keyword length of 6 (with 3 and 9 as close contenders).
ABCDE FGHIGJK LBEDG HIM NEJDM OGJPDCQR EQH SEBHK NBCRM LKH TBURUPGC VEDQHIGH FJ WGJCQK REM QGHBE AGCRDMHE YMQUG
(Note: ABCD here are not the same as ABCD above!)
My next step if this were a much longer body of text would be to divide it into 6 interleaved Caesar alphabets and use frequency analysis, but that doesn't even get me 20 letters per alphabet to work with, and almost no repeat letters. The body of nouns, though, is vastly larger - assuming the six-letter keyword is the same for every noun and begins at the same position, is it possible to encode two different letters to get the same ciphertext result? If not, what are some other suggested methods of attack, or ciphers that would allow this to occur?