I find how well that works is directly related to how closely related English is to the language in question. The further you get from its Germanic and Latin roots, the proportionally less sense a rough translation makes.
Some of that is figures of speech, some of it is colloquialisms, some of it is just weird grammar. For example, Japanese tends to omit the subject, which is incredibly confusing when directly translated if you aren't expecting it.
Germany newspapers apparently use weird grammar conventions that makes translation engines have fits. Google Translate is pretty good on german conversations, anything the EU publishes, and literature, but is really terrible at newspapers.
The translation of the above article is pretty good because a number of people have looked at it and made corrections, but the average German Newspaper article translation is nearly incomprehensible.
> Google Translate is pretty good on german conversations, anything the EU publishes, and literature, but is really terrible at newspapers.
AFAIK Google translate is trained by using corpora of documents that have known-good translations. I could imagine that they are trained using EU documents, as these would usually be available in multiple languages. That would explain why performance on stuff published by the EU is so good...
Compared to Japanese, "I have in Berlin been" almost sounds natural.
In Japanese, you'd say "Berlin ni ita", or translating rather literally, "Berlin in was". You'd omit the "I", which can lead to some interesting misunderstandings.
Needless to say, Google translate borders on useless here.
Some of that is figures of speech, some of it is colloquialisms, some of it is just weird grammar. For example, Japanese tends to omit the subject, which is incredibly confusing when directly translated if you aren't expecting it.