My case:

With a very large dataset of totally unnormalized data, semantic folding is a silver bullet.