Jeff,
Despite appearances to the contrary, we actually classify this using OWL
DL. What is representing on this site is an ontology that was developed
with a classifier that predates OWL by some years, but has been mapped
into OWL DL. What isn't obvious, however is:
(a) We map almost all of the assertions in the FactBox into annotation
properties, as names, synonyms, textual descriptions, etc. aren't the
sort of things provide value to a DL classifier. In OWL 1.0, this
involves a bit if trickery as it doesn't allow you to subclass
owl:AnotationProperty. I believe that this shortcoming has been fixed
in OWL 1.1...
(b) The "Kind" entity that you mention below is not intended to be a
second-level category. It is, instead, an indirect artifact of the
classifier that was used to create the ontology. The classifier was an
ELH variant that allowed role value maps and, as such, didn't allow role
domain and range restrictions. The "Kinds" were a partial work-around,
as they were declared to be disjoint.- See: "F. W. Hartel, S. d.
Coronado, R. Dionne, G. Fragoso, and J. Golbeck. Modeling a description
logic vocabulary for cancer research. Journal of Biomedical Informatics,
38(2):114-129, 2005" for all the gory details. In any case, we will be
eliminating these from the contents in the not too distant future.
The information that does get mapped into the DL is that in the
associations section. Using your example below, you will see:
Parent: Gene (primitive)
Every instance of Antigen_Gene Gene_Plays_Role_In_Process at least one
instance of Immune_Response.
If you look at the source for the page, you will find that these are
created by two templates:
{{LexWiki_Parent|1=BGT_Gene(B16785)|view=Defined|def=partial}}
{{LexWiki_Association|1=BGT_Gene_Plays_Role_In_Process(R37)|2=BGT_Immune
_Response(B18103)|qual=some|view=Defined}}
Which contain sufficient information to produce:
1) Antigen_Gene is a subclass of Gene
2) This information was explicitly stated (vs inferred by a reasoner)
3) Antigen gene is not fully define (e.g. this is a subclass vs. an
equivalent class)
4) Antigen_gene some plays_role_in_process immune_response
At the moment, we aren't transmitting all of this information to the
classifier via RDF - we were waiting for the "nary relations" to settle
down and will be incorporating it shortly. We get around this
limitation by assuming:
1) assuming all subclass relationships are "subclass" vs. "equivalent
class"
2) assuming all objectProperties use the existential quantifier (vs.
universal)
3) not transmitting inferred associations (which is correct.)
That said, I wholeheartedly agree with your argument about second-order
classes. Fischer, in
http://www.ipsi.fraunhofer.de/orion/pubFulltexts/Fischer_1998.pdf
presents a pretty convincing argument for them. Some examples of this
sort of things can be found deeper within the BiomedGT resource. One
example (albeit not very good, as it has no real business being in this
resource) are individual countries, although they probably just belong
on the instance level. Another, which is much more interesting involves
the relationships between genes and alleles. A particular gene is a
fairly abstract thing, representing a closely related set of nucleotide
sequenes that "typically" appear in a known relative location on a
particular chromosone and who have a relatively constant pattern or
motif. An "allele", which is an exact sequence of nucleotides that
occurs at an exact (?) location, is an *instance* of a particular gene.
Despite this, "allele" doesn't describe an individual in any sense of
the word, as there can be anywhere from 5% to 99% of the human
population that manifests this particular allele and, even if you do
narrow it down to a single individual, there are a billions of cells in
the individual's body that have an instance of that allele.
This relationship is not currently represented in the BiomedGT resource,
but a new model will be forthcoming in the next year or so.
Cheers,
Harold Solbrig
-----Original Message-----
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of
Jeff Thompson
Sent: Friday, March 07, 2008 1:24 AM
To: Semantic MediaWiki devs
Subject: Re: [SMW-devel] Classes vs. Categories
Thanks for the links. On biomedgt, I see that the Category inheritance
is used to make a class heirarchy:
http://biomedgt.org/index.php?title=Special:CategoryTree&target=BGT_TopT
hing(%40)&mode=0
Notice that each category not only is the subject of a subClassOf
relation, but also many other relations that describe the class, which
is not allowed by OWL DL. See the rich fact box for BGT Antigen Gene:
http://biomedgt.org/index.php/Category:BGT_Antigen_Gene(B54432)
The category BGT Antigen Gene is in the category of BGT Gene as a
subclass of it:
http://biomedgt.org/index.php/Category:BGT_Gene(B16785)
But notice that BGT Gene is in the category BGT Gene Kind. This is not
a subclass relation, but rather the class BGT Gene is an instance of the
second-order class BGT Gene Kind. Indeed, the RDF export wrongly uses
subClassOf:
<owl:Class
rdf:about="http://BiomedGT.org/index.php/Special:URIResolver/Category:BG
T_Gene-28B16785-29">
<rdfs:subClassOf
rdf:resource="http://BiomedGT.org/index.php/Special:URIResolver/Category
:BGT_Gene_Kind-28B179-29"/>
</owl:Class>
So once again, even in this semantically aware application, category is
improperly used for both "subClassOf" and "instance of", and many
properties other than subClassOf are applied to a class, so we need OWL
Full anyway. This is a case for not relying on the wiki category
system, and needing explicit "subClassOf" properties.
Harold Solbrig wrote:
> See http://biomedgt.org/ or http://www.wiktolog.org/agrowiki/ (the
> latter is slightly out of date at the moment, but it includes an idea
> of what might be done with basic OWL/Dublin Core and the like). These
> examples are slightly different cases, however, as are importing
> classes
> *into* the wiki rather than creating them. As we are developing
> ontologies and classification schemes, there is very little use of
> "instances".
>
> It is our hope that ontology developers will be able use these
> resources (or at least the first one) to comment on and propose
> changes to the ontology contents. We export these proposals in a
> Protege-OWL editor through the RDF export mechanism, although we have
> to do a goodly amount of transformation to accomplish this. Ideally,
> we would like to reach the point where we can generate real OWL
> through the RDF when applicable.
>
> As an aside, we have to tweak the auto-completion code in
> SemanticForms to enumerate Categories rather than just articles. As
> no one else seems to have this requirement, we are assuming that our
> use case is somewhat non-standard.
>
> The Mayo Clinic is also developing a similar mechanism to curate the
> contents of the International Classification of Diseases version 10
> (ICD-10), but this isn't publicly available at this time.
>
> Harold Solbrig
> Apelon, Inc
>
> -----Original Message-----
> From: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED] On Behalf Of
> Jeff Thompson
> Sent: Wednesday, March 05, 2008 12:24 PM
> To: Semantic MediaWiki devs
> Subject: Re: [SMW-devel] Classes vs. Categories
>
> Jon Lang wrote:
>> Jeff Thompson wrote:
>>> Good points about the difference between OWL DL and OWL Full.
>>> So if you only want to export OWL DL, what would you do with a page
>>> like the President or Dog or Wine pages on Wikipedia, which are
> pages about a class.
>> Place articles about classes in the Category namespace.
>
> This is indeed the logical answer. I asked the question trying to be
> provocative since I haven't seen a wiki where main articles like Dog
> are put in the Category namespace.
> Have you seen this in practice?
>
>
>
> ----------------------------------------------------------------------
> --
> -
> This SF.net email is sponsored by: Microsoft Defy all challenges.
> Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Semediawiki-devel mailing list
> Semediawiki-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/semediawiki-devel
>
> ----------------------------------------------------------------------
> --- This SF.net email is sponsored by: Microsoft Defy all challenges.
> Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Semediawiki-devel mailing list
> Semediawiki-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/semediawiki-devel
>
>
>
------------------------------------------------------------------------
-
This SF.net email is sponsored by: Microsoft Defy all challenges.
Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Semediawiki-devel mailing list
Semediawiki-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/semediawiki-devel
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Semediawiki-devel mailing list
Semediawiki-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/semediawiki-devel
