On Dec 26th Alan Grimes said. >> !+ d03$n'7 vv0rk b3cuz $uch 4 $!st3m c4n'+ r34d m! 31337 +3x+.
It doesn't work because such a system can't read my ????? text The dialect you just wrote in is a fairly easy substitution cipher for humans to break in their heads because the replacement values share similar visual or phonetic characteristics to the characters they are substituting for. Since ascii values are used in patterns which NLP programs match to, and hence lose those visual and phonetic characteristics it is not unreasonable that the program would not be able to read this unless either it's patterns were encoded to accept these simple replacements values as valid substitutions or a cryptographic module was front ended to break the cipher. Either option has a very low payback in terms of work required right now. My matching system does use Levenschtein distance and dictionaries sorted by most frequently used words to do intelligent substitutions for incorrectly spelled words so that in you above example Words like $uch and m! at least would be resolved correctly. If the above Hackerese dialect becomes a popular judge ruse in the Turing competitions I imagine I could add the popular hacker substitutions into my patterns fairly easily using a tool like SED, but performance and readability of the patterns would tend to suffer. By the way I consider most 11 and 12 year olds fairly intelligent and the few I asked were not able to make the translation of this sentence into English. I also was unable to correctly translate 31337 even in context so I guess unless the 31337 was what the cryptographers refer to as a null value inserted to make decryption harder I would fail the Turing test here as well! ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
