I think it's that in the public domain books there can sometimes be so many errors from straight OCR'ing-without-checking (which is why I usually download from MobileRead - they've most often been created by people whose love extends to thorough scanning for errors. Thanks!).
I had this thought...imagine doing a simple OCR and upload of
Finnegans Wake. (Aside from verifying against another text) How would know they were errors of OCR?
Cheers,
Marc