It's a lesson in the difference between translation and localization.
In this case, Google's approach confuses the latter for the former, because its corpus contains parallel texts that have not only been translated (into a different language), but also localized (into a different locale). For example, it may be extrapolating incorrectly from a French document and an English document with the following phrases in the same location:
FR Cliquez ici pour la version anglaise.
EN Click here for the French version.
This sign is another good example of how parallel texts are not always translations:
In your Japanese airport example I think they're intentionally not the same phrase. "Welcome back" to the Japanese people and "Welcome to Japan" to the foreigners.
Nice. It looks like it's confusing the different meanings of "English": "a language originating in England" vs. "native tongue".
If you're translating something like "Give it to me in plain English", you'd want the second meaning. Likewise, "It's Greek to me" doesn't hold the quite same weight when translated literally into Greek.
Could be. If you tell it to translate to Spanish, it says "Inglés" (which is incorrect, btw, since it shouldn't be capitalized) or "francés", depending on whether you leave the period. But, if you tell it to translate into German, it says "französisch" either way.