Thanks for copying me in on this.
As a latecomer to the discussion (is there a posted list of the Top 40?),
I may be missing something important in the distinction between relations
and unary predicates, but McCarthy's stress on relations reminds me of
Peirce's position that a system of type labels (such as the Sowa ontology)
is compatible with relational terms only (which Peirce called "copulants"
or "distributives"), and that it necessarily excludes designative and
descriptive modalities of the (immediate) object as it is presented in a
sign.
As mentioned earlier, a defining feature of the Sowa ontology is that it
applies the three Peirce-inspired categories directly to concepts rather
than mediately through classes of sign modalities. This abridgement is
understandable, as Peirce's own typology of signs was never fully
completed and exists in its last form only in a (partially unmailed)
correspondence. With the exception of the icon-index-symbol and
[mark]-token-type trichotomies, it has not entered into the logical canon.
Moreover, Peirce's system of ten classes, each trichotomized, yields 3 to
the 10th power combinations, a very large number of semiotic
discriminations.
But many of these combinations are not possible or not interesting. The
combination of types (or "famisigns") with signs that represent their
immediate objects by describing or designating them Peirce held to be "not
possible." The only sign-modality of the
descriptive-designative-distributive trichotomy compatible with types is
the distributive, so named because it represents a continuing and general
relation applicable to all members. Peirce's discussion is available in
the Harvard edition of the Collected Papers, Vol 8, chapter 8, sections
346-65.
If we accept this theoretical result-- that types must be distributive--
or even the approach to sign-modalities underlying it, it will influence
what we can expect of a computationally applied "ontology", and will also
affect how we treat the varied artifacts we are inspecting (the EDR nodes,
the WordNet upper level, the CyC and Pangloss upper levels, the Sowa
crystal).
If Peirce is correct, the more formally we define ontology, the more we
shall be driven to the notion that it can include nothing but relations,
or rule-like conditions.
Lee
On Mon, 4 May 1998, John F. Sowa wrote:
> I agree that an ontology must include more than unary predicates.
> That was C. S. Peirce's essential point, which I have not always
> been able to articulate clearly (and unfortunately, neither was he).
>
> John Sowa
>