Hi!
Recaptcha is nice but it contains some internalization and localization fails.
1) The "image" captchas have an ethnocentric northamerican bias.
Example: I see the text "One way only" inside a white square. Recaptcha asks me to mark all traffic signs.
Despite of that sign not being a traffic sign under the 8.1IC traffic sign standard used in Spain,
I have to mark the "One way only" text as a traffic sign (or I would fail the captcha), thus "training" captcha to interpret that american-only signal is a traffic signal for Spanish people, also.
2) Despite of translation, the "image" captchas shape reality just as English language does.
Different languages shape reality in different ways.
See this example:
Recaptcha shows me a hamburger sandwich, a subway-like sandwich, a french bread sandwich and a english bread sandwich.
Then, it asks me (in Spanish): "Seleccione todas las imágenes de sandwich").
In "castillian" Spanish (es-ES), we have "sandwich" (english bread sandwich), "hamburguesa" (hamburger sandwich) and "bocadillo" (subway-like sandwich or french bread sandwich).
But if I click only the image that represents what we call "sandwich" in Spain, I won't pass the captcha. Nor do I if I click the images that represents a "bocadillo".
I have to translate the Spanish question back to English to pass the captcha by clicking the images that the English language calls "a sandwich".
3) Audio captchas still in English.
This is a major problem. Audio captchas are still in English.
This prevents blind people from around the world to pass the audio captchas.
And, when these capchas are used by government sites, it could have legal consequences (as an example,
I could sue some Spanish goverment sites for using reCaptcha, since in Spain any action or decision took by government employees that constitues discrimination against handicapped people is a major fault).