Three ideas seem to have gotten combined:
The urge to classify things according to "what they are."
A need to express the semantic characteristics of things, by specifying which attributes and relationships and names are relevant and valid for them. The easiest paradigm: certain rules and constraints apply to certain classes of things.
A tradition of data description, based on record types.
So, for example, an Account can be a Savings Account or a Checking Account. Kent's first point is that we would recognize both Savings and Checking Accounts as Accounts. His second point is that Savings Account may have distinct attributes and relationships from Checking Account, and vice versa. And his third point is that ...