Different estimates will put it at different amounts, but as a general rule of thumb it is safe to say an English sentence has one bit of entropy per character. One NIST estimator only levels out to 1 additional bit per additional character after twenty character + special/lower/numeric/upper , with more more entropy credited to characters prior to twenty and with a bonus for each additional character type. zxcvbn takes a lot of things into consideration, for example zxcvbn would have less entropy than i29fks because it is a spatial pattern on the keyboard, I think it also takes into account syntactic structure of English sentences as well. this paper (acl.ldc.upenn.edu/J/J92/J92-1002.pdf) suggests an upper bound of 1.75 bits per character in English although it does mention that they think improvements are possible this site puts it at 1-2 bits per character: http://unixhelp.ed.ac.uk/CGI/man-cgi?ssh-keygen+1  I guess with all sources considered, it is probably the most accurate bet to say that English contains slightly over 1 bit per character, but a conservative estimate and best rule of thumb is to not attribute more than one bit of entropy per byte of English IMO, even though in reality the entropy is probably slightly more than that.