Jyoti,
Thanks for the reference. I agree the paper makes some good points.
A promising alternative to deal with legacy graph-based formalisms
with hierarchies and otherwise sparse axiomatization (of which SNOMED
is one example) and still provide "upward mobility" to OWL is the
RDFS(FA) proposal (http://dl-web.man.ac.uk/rdfsfa/). I think this
website is a "must-read" for those of us interested in this sort of
thing, and a more powerful alternative target language for such legacy
vocabs than the other alternative I mentioned of just using skos
descriptors.
(btw Vipul: RDFS(FA) is also extremely relevant to a parallel thread
going on here about "multi-layered KR", as it is designed to separate
instance modeling from class modeling in a way that is much more akin
to UML than is possible in RDFS).
John
On Jul 24, 2008, at 2:00 PM, Jyotishman Pathak wrote:
Hello John,
I think these are very interesting ideas.
While we are at the topic of representing SNOMED in OWL-DL (I am
assuming you are referring to OWL-DL 1.0), I would like to point to
a paper by Kent Spackman published in 2007: http://www.webont.org/owled/2007/PapersPDF/submission_26.pdf
In the paper, Kent gives insights on how OWL-DL 1.0 may not be
adequate to model SNOMED. So, while I agree that "It is far less
expressive than OWL" (SNOMED falls in EL++ logic), there are certain
elements in SNOMED as of now that cannot be expressed by OWL-DL 1.0.
Cheers,
- Jyoti
On Thu, Jul 24, 2008 at 12:35 PM, John Madden <[EMAIL PROTECTED]>
wrote:
Hey Chime,
Thanks for coming up with this project task proposal relating to
conversion of legacy terminologies to OWL/RDF, it's very exciting.
I really like your idea of picking a specific subdomain, like drug
terminology, and using that to test out the pitfalls/possibilities.
(Actually, I think very domain-specific ontolgies have, as a rule,
the strongest likelihood of short-term practical utility.)
We (Mary and I) have an ongoing project involving cancer reporting
for public health where I've always dreamed of producing an OWL/RDF
adaptation of content culled from a variety of sources including
SNOMED CT, and also others. Unlike the NIH cancer ontology which
includes a lot of biosicence related content, we'd focus exclusively
on supporting routine clinical aspects of cancer care.
I'd love to make this a use case. It does involve modeling some
"utility" classes and relations (like Patient, Physician, etc.) but
I'd like to move that stuff out into some more generic project.
(As Vipul knows, modeling that stuff always involves taking
appropriate cognizance of constitutencies at HL7, CDISC, CaBIG, etc.
etc. -- although I think it would be grand just to demonstrate to
the world that it is possible to do this without stepping on
anyone's toes through the judicious use of imports, sameAs/
differentFrom, equivalentClass, seeAlso, etc).
John
P.S. w.r.t SNOMED CT, to just clarify the point I in the call today
about whether there would ever be an OWL-SNOMED: the expressivity of
the DL underlying SNOMED is roughly on a par with that of RDFS. It
is far less expressive than OWL. I could therefore imagine an RDF-
SNOMED, but not an OWL-SNOMED.
Anyway, unlike RDF/S, SNOMED has never had a published formal
semantics, and certainly not a model-theoretic one like RDF/OWL's.
(Indeed, the absence of an explicit model-theoretic semantics makes
the claim that SNOMED is DL-based at all pretty fuzzy-wuzzy. I'd
maintain this is so even though, in practice, each SNOMED CT release
has to classify on the ontylog reasoner.) Hence my remark earlier
that it might be safest just to consider SNOMED concepts, if
represented as resources, as being instances of skos:Concept.