Pages, some stolen, some original

▼

Tuesday, August 31, 2010

Words?


Supposedly ReCaptcha works by putting up two distorted words, one of which it knows how to spell. If you get that one right, then it assumes you also spelled the other one right, and it stores this information away for use by an OCR (Optical Character Recognition) program so it can use your spelling to recognize this word.

However, today's puzzle only compounds the ignorance. The instructions say "Type the two words", but onces & indrect are not words (okay, onces might be a word, as in multiple occurences of once, but that's going to be a rare case). I suspect they are typos for ounces and indirect, which is something OCR is going to have to learn to deal with.

Update February 2017 replaced missing picture.
at

3 comments:

  1. Ole Phat StuAugust 31, 2010 at 9:42:00 PM PDT

    Actually it is better to have non-words. Why?

    Well, any OCR SW used by a hacker to read the letter-sequences would then most likely do a dictionary lookup to confirm its guess. With a non-word, the dictionary lookup confirmation would fail and the OCR would try a different guess or give up.

    PS: Recently one of the test 'words' I encountered was 'nukulah' : O tempora, o mores :-(

    PPS: Why do americans spell 'whore' as 'ho/hoe'?

    ReplyDelete
  2. Chuck PergielAugust 31, 2010 at 11:10:00 PM PDT

    "Ho" is what the rappers use.

    ReplyDelete
  3. Ole Phat StuSeptember 1, 2010 at 10:14:00 PM PDT

    Hmm, why do you use a double P in that word? ;-)

    ReplyDelete