More discussion in this PR and #132.

Basically, we need a policy for differentiating between OccupancyType and SubaddressType, since our use of the two tags differs from the URISA standard. From my interpretation of the existing training data, I'm inclined to say that SubaddressType divides an address into a smaller unit (e.g. an office suite, a mailroom, a floor) while OccupancyType defines the endpoint of an address (e.g. an apartment, a room).

This definition introduces some interesting ambiguities: some common tokens (room, floor) could be OccupancyType or SubaddressType based on their context. For example, suite would normally be an OccupancyType, but in the context of suite 30 room 406 it would become a SubaddressType.