Test data from different corpus

Support
1

Hi,

can I train NMT with data from one corpus and test it using data form another corpus?

Sevilay

2

Hi,

Yes, you can.

1 Like
3

hi,
thank you for respond, I need parallel corpus to work with and I have used parallel data from this site http://opus.nlpl.eu/ , but unfortunately there are a lot of repeated sentences, I was using kazakh-turkish pair, is there any other site can I depend on?

Sevilay

4

Hi,
which model should I choose to predicate my test data as I see there are more than one model such as demo-model_step_50000.pt…etc.

Sevilay

5

I’m not aware of any site for this language pair.

If you don’t want to search too far, just select the last one. Otherwise, look at the training logs for the model which has the best validation score.

1 Like