The new Spanish models trained on LDC data contain a few bogus POS tags.
I call them "bogus" because they only show up a few times in the corpus, and only seem to confuse the models.
Task: ensure the list of unique POS tags in all models matches those described in our Spanish docs.