No, that's not how it works. An automated system is passed the pages that have been scanned, and it performs recognition on those pages. Any words that it has a low confidence threshold on are automatically sent to Recaptcha so that they can be crowdsourced. The only point at which a human being will look at the word is if the crowdsourcing doesn't reach a consensus on what the word is (in this case, the word will almost certainly qualify).
And this means you don't have to get the scanned word completely right. If your guess is somewhere in the right ballpark with other people's guesses, you pass.
no subject
so if the majority who get that bit of weirdness type in Cthulu it will show up as Cthulu in the book thats being scanned and OCR'd.
no subject
But you'd think that they'd make sure that the scanned bit was actually a word, and not something untypable.
*I remember reading about some internet group (I assume anonymous) gaming a bunch of recaptchas that some time was using, so they could win a poll.
Edit: Aha! http://musicmachinery.com/2009/04/27/moot-wins-time-inc-loses/
no subject
no subject