Neural Machine Translation English to German


i try to work with this workflow.

Is there any way in current workflow after certain number of rows to skip German translation and let the tool to make prediction on its own?

I just need to translate the text from English to German (and vice versa) with the help of NLP (Natural language processing). Maybe this workflow is not suitable for this purpose?

Thank you!

Hello Katerina,

the trained model in this example can translate short sentences from English to German.

There is a newer version for Neural Machine Translation on the EXAMPLES Server, which has in addition also a deployment workflow. It shoes one example how the trained network can be used for new sentences.*WI2o2pt4hStUwRrZ*DVUXE5Ho8frog33t

The network structure we used in this example is pretty easy. We are planning to publish an updated version, which uses a bidirectional LSTM, soon. This new version leads to better translations.

For translations from German to English you would have to swap the role of the English and German sentences and adapt the flow variables. Please let me know if you need help with these steps.



Hello Kathrin,

thank you very much for your answer!

Can you help me, plaese with these steps and write what I need for it. I lern KNIME and I need to translate with Natural language processing. Do you have maybe the article where I can lern how to make NATURAL language processing workflow?

Thank you,

Hi Kathrin

Thank you for your feedback! Looks like Deployment workflow is exectly I’m looking for.
Have a small questin. Could I use as an input only single short sentences or several lines/sentences would also work?
At least I see on user interface for String Input control there are options to set Multi-line Editor width and Multi-line Editor height. But even after increasing this values I can enter only single line to Default Value text box.
May be different control should be used for this purpose?

Thank you !

Hi Katerina,

in this article we explain how NMT can be performed:

We trained the network on short sequences. Therefore the network probably won’t be able to translate longer sequences and I wouldn’t allow multi line input.