Native German speakers do not usually forget umlauts (since they are separate keys) and so Google translate won't have the necessary training data to robustly handle German without umlauts. The output appears as gibberish because the language model can't distinguish the input from gibberish. That becomes more obvious if you intentionally input nonsense: http://languagelog.ldc.upenn.edu/nll/?p=35120
True. However, I claim Google translate is a lot of times used by non-conversant level German speakers, who don't have the right layout. I would expect Google translate to be trained somewhat adversarially. Example:
"Ich werde Bücher lesen" translates just fine. If we skip the Umlaut Google translate "forgets" the meaning of books. If we change the tense to present everything translates just fine.
