Re: Issue-57

On Mon, 2011-06-13 at 13:51 -0400, Jonathan Rees wrote:
> On Mon, Jun 13, 2011 at 11:30 AM, Jeni Tennison <jeni@jenitennison.com> wrote:
> > Hi,
> >
> > For those who don't follow it, there's a thread on httpRange-14 / Issue-57 at the moment on the linked data mailing list. A good example message is:
> >
> >  http://lists.w3.org/Archives/Public/public-lod/2011Jun/0186.html
> >
> > where Richard says:
> >
> >  Being useful trumps making semantic sense. The web succeeded *because*
> >  it conflates name and address. The web of data will succeed *because*
> >  it conflates a thing and a web page about the thing.
> >
> >  <http://richard.cyganiak.de/>
> >    a foaf:Document;
> >    dc:title "Richard Cyganiak's homepage";
> >    a foaf:Person;
> >    foaf:name "Richard Cyganiak";
> >    owl:sameAs <http://twitter.com/cygri>;
> >    .
> >
> > I don't think that this is covered by any of the scenarios in Jonathan's document at:
> 
> My *intent* was that this situation would be covered by
>   4.4 Coerce an information resource to what it defines its URI to name
> but I guess the section name and description are not adequately evocative.
> 
> I actually don't object to this attitude too much; if people want to
> opt out of machine inference who am I to say otherwise. They
> themselves say they don't care about it (i.e. inference is best done
> in some subtle and complicated way by people, not in a simple and
> stupid way by computers). Should they decide to care in the future,
> they can generate new RDF. It might be a lost opportunity and an
> interoperability risk, but so it goes - the horse has been led to
> water and it finds the water unpalatable.

I do not think that is a fair characterization.  Richard's example is
*not* opting out of machine inference.  It is merely opting out of
certain inferences that *some* applications need but others do *not*
need.  And that is as it *should* be, as it is not possible to cater to
*all* applications.

The subtle mistake that is being made repeatedly here is in assuming
that someone's data is *wrong* (or socially irresponsible) if it
conflates two things that we humans find useful to distinguish, such as
people versus web pages -- *even* if the class of applications for which
that data is intended have no need to make such a distinction!  This is
myth #4 in "Resource Identity and Semantic Extensions: Making Sense of
Ambiguity":
http://dbooth.org/2010/ambiguity/paper.html#myth4

> 
> >  http://www.w3.org/2001/tag/awwsw/issue57/20110531/
> >
> > In particular, I don't think the kind of 'punning' that we talked
> about (where different properties treat the given resource as being
> different kinds of thing) copes with the rdf:type property (shortened
> to 'a' in the Turtle) having two different values.
> 
> Correct. I think that in addition to expanding all the properties,
> you'd also have to expand all the types, so that 'information resource
> defining its URI to be a person' is a subclass of Person and so on. I
> can add this to the presentation.
> 
> > Similarly, it's really unclear in the above example whether the
> > owl:sameAs relates to the Person or the Document (until you find a 
> > description of <http://twitter.com/cygri>, which of course might be
> > a resource that is both a Document and a Person itself).
> 
> I think it has to be the Document, at the level of sameAs, since
> otherwise you get false equations between documents that describe the
> same thing. 

No, you only get false equations if you *also* try to assert that
foaf:Document is disjoint with foaf:Person.  *Your* applications may
need to make that distinction, but that is an issue with the
*combination* of your application and Richard's data.  It is not an
issue with Richard's data per se: his data is perfectly fine for other
applications that have no need for that distinction.

> That assumes an entailment regime that can lead you to
> such conclusions... RDFS entailment can't, so as long as you stay away
> from interoperation with OWL content, sameAs doesn't mean anything in
> particular, and there's no risk of inconsistency.
> 
> This community ought to be perfectly happy with me saying the URI
> refers to the information resource, since otherwise they'd be
> admitting that the assignment *does* matter!
> 
> The thing to do is to get this community involved in specifying how to
> write metadata, e.g. embedded license declarations for RDF files that
> are about information resources, 

The licensing example is a *very* good one, as it aptly demonstrates the
problem that this ambiguity creates for *that* class of applications.
But I think it would be a mistake to suggest that *all* socially
responsible data should support that class of applications.



-- 
David Booth, Ph.D.
http://dbooth.org/

Opinions expressed herein are those of the author and do not necessarily
reflect those of his employer.

Received on Tuesday, 14 June 2011 00:39:00 UTC