The TL;DR is that recaptcha uses one word that it knows as the solution and the second is one that recaptcha is determining the answer for. Recaptcha is currently trying to turn print archives of newspapers into normal digital text. So it takes unknown words and has users try to solve it. It then stores the results and starts assigning confidence values to the unknown words. So if you get a answer with a relatively high level of confidence it’ll let you through. If you get a brand spanking new word, it’ll let you type just about anything.