owl:sameAs and identity [was Re: blog: semantic dissonance in uniprot]

On Wed, 2009-03-25 at 09:25 -0700, Mark Wilkinson wrote:
> [ . . . ]
> Another predicate is needed that is less "rigourous" - owl:kindOfLike  :-)

I disagree.  I think it is more a matter of getting used to the fact
that terms (URIs in the case of RDF) are almost always ambiguous -- they
admit multiple interpretations.  And though one RDF graph may restrict
the set of possible interpretations in one way, another RDF graph may
restrict the set of possible interpretations in another way, such that
there are no interpretations that are consistent with the merge of those
RDF graphs.  

For example, suppose we have an RDF graph G1:

	@prefix : <http://example#> .
	@prefix owl: <...> .
	:apple1 owl:sameAs :apple2 .

and an RDF graph G2:

	@prefix : <http://example#> .
	@prefix owl: <...> .
	:apple1 owl:sameAs :apple3 .

And suppose that there are (somehow) definitions provided
for :apple1, :apple2 and :apple3 that somehow restrict the possible
interpretations as follows:

	:apple1 either maps to a red or a green apple.
	:apple2 either maps to a red or a pink apple.
	:apple3 either maps to a green or a yellow apple.

Then in any interpretation 
http://www.w3.org/TR/rdf-mt/#interp 
that is consistent with G1, both :apple1 and :apple2 must denote (or
"map to") a red apple.   Similarly, in any interpretation that is
consistent with G2, both :apple1 and :apple3 must denote a green apple.
But for the merge of G1 and G2, there is no consistent interpretation
possible.  

This does not mean that owl:sameAs was the wrong thing to assert.  It
just means that G1 and G2 made different choices about the
interpretations that they wished to permit.  The identities
of :apple1, :apple2 and :apple3 were ambiguous by their definitions, as
will be true in almost every case in the real world.  (See, for example,
http://www.ibiblio.org/hhalpin/irw2006/presentations/HayesSlides.pdf .)
In essence, G1 and G2 made different assumptions about the ambiguous
identities of :apple1, :apple2 and :apple3.  But since those assumptions
were both (individually) within the range of ambiguity permitted by the
definitions of :apple1, :apple2 and :apple3, the authors of G1 and G2
were free to make that choice.  Incidentally, this is precisely the
reason for distinguishing between "core assertions" and "ancillary
assertions" in
http://dbooth.org/2007/uri-decl/ 

In short, instead of thinking of resource identity as a fixed mapping
from a URI to a particular thing, one can think of it as a mapping from
a URI to a cloud of possible interpretations.  Different RDF graphs will
further restrict the set of possible interpretations to different
subsets of that cloud.  Only by admitting the exact same range of
ambiguity will two URIs be interchangeable in all RDF graphs.  But that
is *not* what owl:sameAs asserts.  It merely asserts that in all
possible interpretations for a *given* graph, the two URIs denote the
same resource.  (I'm paraphrasing, of course.  The RDF Semantics section
on interpretations does quite a good job of explaining the idea:
http://www.w3.org/TR/rdf-mt/#interp .)  But for a given graph, there may
still be many consistent interpretations.  And the set of
interpretations that are consistent with G1 may be different from (and
may not intersect with) the set of interpretations that are consistent
with G2.  This is okay.  It's not a mistake.  It just means that we need
to get used to it.

David Booth

Received on Thursday, 26 March 2009 03:36:01 UTC