| Summary: | In historical cryptology, cleartext represents text written in a known language ina cipher (a hand-written manuscript aiming at hiding the content of a message).Cleartext can give us an historical interpretation and contextualisation of themanuscript and could help researchers in cryptanalysis, but to these days thereis still no research on how to automatically detect cleartext and identifying itslanguage. In this paper, we investigate to what extent we can automaticallydistinguish cleartext from ciphertext in transcribed historical ciphers and towhat extent we are able to identify its language. We took a rule-based approachand run 7 different models using historical language models on ciphertextsprovided by the DECRYPT-Project. Our results show that using unigrams andbigrams on a word-level combined with 3-grams, 4-grams and 5-grams on acharacter-level is the best approach to tackle cleartext detection.
|
|---|
