Can anybody help me explain situation:
i send input data to translate script (for example - cat);
and i receive output data from translate script (katze katze katze).
Why multiple words output? Any idea.
This is a common behavior of NMT systems when mainly trained on complete sentences. You should consider adding shorter phrases and terminologies in your training data.
That has been our experience :-). In our production set-up we direct one-word & two-word segments to a non-NMT server.
We too. But i want know math part this situation )