Omnios (Spain) interview question

How do you automatically evaluate the performance of an NLP model?