Even though it's been quite while ago, still looking forward to make the default CRF training inside the MLWordTagger to be faster or at least the training progress can be shown to prevent canceling it because it's like running forever.

I'm also have quite similar problems when I tried to combine all of the sentences into one single tokens and labels which is different than .JSON file that is seen on the MLWordTagger example. However, when I tried to separate it into multiple samples / sentences, the default CRF training algorithm is seems to running indefinitely.