Probably the final version I'll release. The only potential improvement I can see is removing some of the lower quality training data.

Many different tags used in training, not required.