I’m going to contribute to an answer for the question in your title line. Google and Facebook both have “crowdsourced” translations going on, that is to say, they allow humans to translate small components of the site. Once you have a certain number of people saying “word” means “other word” in other language, and combined with any actual proofing, this is a rather accurate and low cost way of automated translations. Of course, with each submission, data is created. Eventually, the Googlebot will be able to preemptively translate a page, or take a good stab at it, and let that be corrected by native speakers, so that the next time it learns from its mistakes… I think this type of collective and predictive intelligence will be seen much more often in the future across all sorts of fields.