Anyway what will be easy to remember for a human will be sequences exhibiting high ngrams correlation, which is efficiently modeled by markov chains. Wondering if studies have been done from this perspective on password strength, and how chains generated from leaked password / passphrase collections would deviate from ones generated from common language.
