the class of substances, but the latter might be an ontological category whereas the
former is not an ontological category.
Westerhoff (2005: 27–8) seems to think that appealing to the notion of depend-
ence helps eliminate gerrymandered sets from counting as ontological categories. Say
that one kind K1 is dependent on another K2 just in case, necessarily, if K2 has no
members, then K1 has no members. How does this notion of dependence help? All
universals are universals-or-substances. And nothing could be a universal without
being a universal-or-substance. So the class of universals is empty if the class of
universals-or-substances is empty. Souniversal-or-substanceis a more general kind
thanuniversaldespite being gerrymandered.
If we are to identify categories with classes of entities, the classes in question must
benaturalclasses as opposed to gerrymandered or“merely disjunctive”classes.^25
A putative ontological category must“carve reality at the joints.”Westerhoff (2005: 27)
hesitates to directly appeal to the notion of naturalness because he worries that the
notion is obscure. The notion of a natural kind (or property) might be somewhat
obscure but it is clear that the notion is theoretically useful. The notion of naturalness
has been employed in plausible accounts of core philosophical concepts, such as
intrinsicandextrinsic properties,causationandlaws of nature, andmeaningand
reference.^26 Why not then appeal to naturalness directly and say that ontological
categories are highly generalnaturalclasses, kinds, or properties? This definition
correctly implies thatuniversal-or-substanceis not an ontological category (since it is
a merely disjunctive class), but allows thatuniversalmight be. An appeal to some-
thing like naturalness, fundamentality, or structure is in play in Paul’s (2013) account
of categories, for example.
However, this construal of ontological categories suffers from what Westerhoff
(2002: 288; 2005: 35–8) callsthe cut-off point problem. There is a partial ordering of
natural kinds (ordered by themore general thanrelation or by themore natural than
relation) whose topmost node or nodes are occupied by ontological categories. But
since ontological categories can be nested, these terminal categories need not be the
only ontological categories. How far down the ordering can one go before one ceases
to have ontological categories? It might be thatelectronis a paradigm of a natural
kind, but it is probably not an ontological category.
A more sophisticated version of this view is one that holds that ontological
categories are those properties that provide the best overall balance of generality
and naturalness. On this view, ontological categories needn’t be maximally natural
kinds, since those categories that best preserve generality and naturalness might fail
(^25) For the reason discussed earlier, we should not identify ontological categories with classes or
collections of any sort. However, as we will see in section 4.3, the notion of naturalness is appealed to
when formulating the account of ontological categories defended here. 26
See, for example, Merrill (1980) and Lewis (1983a), 1984, and 1986) for a discussion of these jobs and
how the notion of naturalness is employed to perform them.