I saw something similar to this the other day. Instead of the random string of characters, the captcha was a scanned in word from an old manuscript that some library was trying to digitise.
They are basically using OCR to store all of the books in the library, but of course OCR is prone to drastically misreading the words and it requires human input to correct these mistakes. These errors are made even worse by the fact that many of the books and manuscripts are in a very poor condition. By releasing the words as captchas, they can get thousands of words identified every day, making the process much easier for them. By having two words each time (one known and one unknown) plus giving the same set to multiple users, they can reduce the chance of errors in the system.
The added bonuses are that the variety of writing styles and damage on the manuscripts makes the words that much harder for bots to identify, and with the captchas being actual words it's correspondingly easier for humans to identify (assuming they can read and spell
).