The term "equivalence clusters" in the paper seems more appropriate than "categorization".
"Categorization" to me suggests language use; once a cluster is defined by constraints it is given a language key as a shorthand.
That key creation to tag clusters doesn't seem to be what's described here. Just the clustering is described. And that innate clustering seems obvious once already experienced in life. Sensory data makes it obvious an apple is not an orange.
More accurate to say physical reality has "categorization baked into physical constants."
