Jake Anderson vapourforge.com> writes: > My understanding of the system is that they have one known word (ie > dictionary) and one unknown word (from book) > person enters known word for auth, and that gains them access, unknown > word if it agrees with 3 other people is then used for the text. I am not following you here. At point 0 a program cuts a piece of book text image from a larger page and makes it into a Captcha (warps it etc). It may first try to OCR it and end up with 5 guesses for a word. Then it will serve the next 5 Captcha requests using the warped picture and one of the 5 words each. That gives the turks 20% success chances at the first try but none of them know it and neither does the computer that prepared the question. So, maybe the computer accepts all 5 possible answers as correct for that Captcha and keeps serving it again and again to random users until a statistically significant set of answers allows it to choose 'which' of the 5 is most likely correct. Then the current Captcha is classified as 'solved' and the program moves on the the next piece of text. I can see no other (unattended) way to do it right. And this may open the door on spammers etc using an OCR program to arrive to one or several of the 'right' 5 OCR interpretations of a Captcha and thus defeat the Captcha's original intent (that of Turing test). There must be a catch. Peter P. -- http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist