I was thinking, as Karen suggests, that an AP would specify, say, that the
"range" of dct:subject and similar properties is the VES GeoNames. An AP
for a museum community might specify AAT as the VES; another AP might
specify the VES as a union of GeoNames and AAT. Interoperability and
inter-KOS relationships are established by mappings between KOSs, not
hard-wired into a set of sub-properties.


The case of literals is very interesting. It is tempting to disambiguate
the literal "China" by using properties like "has as subject (place)", "has
as subject (person)" (the cover of Sunfighter by Kantner/Slick), "has as
subject (ceramics)", "has as subject (housekeeping)", etc. But that shifts
the issue from values in a KOS to properties in a metadata schema. That is,
the knowledge is organized via a KOS in some circumstances, and by a schema
in others.


Subject instance triples with literal objects are plain messy - but they
are probably going to be in the majority in the triple soup through
generation by social networking sites using uncontrolled so-called
folksonomies. And we can't expect folks to choose which specific-subject
property they're gonna use, or enter a term appropriate to a pre-set
property, or even be aware of the issues. (And some of those folks are,
sadly, professional librarians ...)


I think management through KOS is probably better than through schema
properties; presumably it is easier to apply machine-mediated quality
control by ensuring that the object of a "has as subject" property is from
a named KOS than it is to determine that "dog" is not an appropriate object
for a "has as subject (place)" property.


I think the divided world Karen foresees is inevitable. These issues have
been around for a long time, and I guess RDF/Semantic Web/linked data
technologies are not going to provide a better means of resolving them.


+++ for FKOS! Surely some work is going on somewhere towards this?


Are we seeing the emergence of APs for KOSs? It strikes me that FKOS is a
named-graph pattern similar to what is being discussed in DC-Architecture
about the DC Abstract Model and APs [1]. Other patterns are suggested by
the FRSAD analysis of subject categorization. Can the SoDC-CL proposed by
Alistair Miles [2] cover FKOS and other patterns?


Cheers


Gordon



[1]
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1202&L=dc-architecture&P=31326
[2]
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind1202&L=dc-architecture&P=30886



On 27 February 2012 at 20:42 Karen Coyle <[log in to unmask]> wrote:

> On 2/27/12 10:53 AM, [log in to unmask] wrote:
> >There is no requirement for a
> > specific-subject (sub-)property such as frbrer:"has as subject (place)"
> > (and its 10 companions, one for each Group entity), because this can be
> > represented by an Application Profile Vocabulary Encoding Scheme or
KOS,
> > for example LCSH, DDC, local SKOS vocabulary, etc.
>
> I don't see how an AP resolves this, that is, how an AP overcomes the
> lack of a "subject (place)". I do think that to some extent vocabularies
> can help if the values are represented by URIs from vocabularies that
> specify a subject "type." If your value is an entry from GeoNames, or is
> a geographical subject from LCSH, then you probably have what you need
> to clarify that the subject is a place. But dct:subject can have
> literals as values, and for those there is no distinction. What one
> might end up with is a metadata world where those distinctions between
> types are available only for some RDF-defined vocabularies but not for
> literals. Actually, that seems to be what we have today for dct:subject.
>
> We should also note that we don't yet have a way to describe a
> vocabulary that has facets. In part that is what was attempted with MADS
> in RDF, but unfortunately that ontology is forced to replicate the whole
> of the MARC Authorities record, so it's a bit messy. I think it would be
> interesting to postulate a FKOS - faceted knowledge organization system
> - language.
>