W3C home > Mailing lists > Public > semantic-web@w3.org > April 2005

Re: Inheritance

From: Frank Manola <fmanola@acm.org>
Date: Wed, 06 Apr 2005 17:10:50 -0400
Message-ID: <4254505A.8080406@acm.org>
To: Hans Teijgeler <hans.teijgeler@quicknet.nl>
CC: 'Jeen Broekstra' <jeen@aduna.biz>, semantic-web@w3.org

Hans Teijgeler wrote:
> Hi Frank,
> I think that the real problem lies in the fact that the XML and 
> RDF worlds are lacking a proper underlying data model.
> In the process industries we have worked for 15 years on a generic, 5NF, 
> data-driven data model <http://www.infowebml.ws/ECM4.5/ECM4.5.html>that 
> is advanced enough to model the lifecycle information of facilities like 
> a refinery or oil well. "Lifecycle" for an oil well easily means 100 
> years, so we needed something solid and not the gizmo-of-the-day. We 
> first model the information and then find a way to exchange and 
> integrate it in what is the leading technology at that moment.

I'm not going to argue about whether XML has a proper data model or not 
(not my problem!), but as far as RDF is concerned, it depends on what 
you mean by "data model".  I think RDF is a data model roughly in the 
sense that the (database) relational model is a data model (think of 
classes as unary relations and properties as binary relations, with 
URIrefs or literals as the values).  Like the relational model (and, in 
fact, like first order logic), RDF doesn't build a whole lot of 
additional meaning into its basic concepts, such as some of the specific 
details you've described below.  That's not to say that those might not 
be useful for a lot of applications, but other applications might (and, 
in fact, do) want to look at the world somewhat differently.  RDF tries 
to be neutral in that respect.  That's not being "gizmo-of-the-day", 
it's (for one thing) recognizing that there are a lot of different 
"modeling religions" in the world.  There's certainly nothing stopping 
you folks from adding specialized meaning to the kinds of classes you 
define (as long as you don't expect someone not subscribing to your 
particular modeling approach to understand it).  Some OWL concepts, for 
example, specialize RDF concepts, and many people prefer to start from 
OWL and build on those specializations.

> For us your triple  "ex:Volkswagen dc:creator ex:Frank" cannot be true 
> at all, because a *class* in our terminology is:/ "//A *class* is a 
> *thing* that is an understanding of the nature of things and that 
> divides things into those which are members of the class and those which 
> are not according to one or more criteria."./ It is an abstraction that 
> has always existed and will always exist, and therefore cannot be 
> created. It may not have members and it may not yet have been 
> discovered, but it was there starting with the Big Bang. So the *class* 
> "Volkwagen" was there already before there were even human beings. Quite 
> a paradigm shift, I guess! The *physical_object* "myVolkswagen" was 
> created by the VW factory somewhere in this world, and it meets the 
> criteria for membership of the *class* "Volkswagen". (Note: "/A 
> *physical_object* is a *possible_individual* that is a distribution of 
> matter, energy, or both/." AND: "/A *possible_individual* is a *thing* 
> that exists in space and time/.")
> Another thing that I read in the RDF Primer and that I read here again 
> keeps amazing me. That is the assumed possibility that it cannot be 
> discriminated whether a reference is to a thing or to its description. 
> In my world the record about a thing has a UID and the description of 
> the thing, being an instance of a totally different entity data type, 
> has a different UID. No confusion possible, ever.

I don't think this last is exactly accurate.  As far as RDF is concerned,

ex:Volkswagen dc:creator ex:Frank

describes the subject of the statement, i.e., ex:Volkswagen.  But also 
as far as RDF is concerned, ex:Volkswagen is a pure identifier (in this 
case, presumably of a class of cars).  The problem occurs when you then 
go *outside* RDF (using URIs for retrieval in the Web is outside RDF) 
and expect automatic consistency in what the URI refers to;  different 
"refer to" mechanisms are involved.  Certainly I can specify separate 
URIs in RDF to refer to a thing and a record of a thing.  What those 
URIs return if you plug them into a browser is a separate issue (so 
would it be if you used those URIs as keys in the columns of a 
relational database, and then did selections in the database based on 
those keys, versus looking them up in a browser).

> So returning to your example: "ex:Volkswagen dc:creator ex:Frank" would 
> read in our lingo (the format is free-format):
> *class_of_inanimate_physical_object*"Volkwagen"  is_represented_by a  
> *class_of_information_representation*"Volkswagen catalog"
> *class_of_information_representation*"Volkswagen catalog" is_defined_by  
> *physical_object*"Frank" 
> About your remark on XML Namespace: we don't use them other than for 
> references to schemas. The standard ISO 15926-7 that I wrote uses XML 
> Schema in a rather unusual manner, where we make reference to 
> identifiers that are not necessarily inside the same XML document 
> (sounds RDF-ish!). Click here 
> <http://www.infowebml.ws/topics/07-mapping-to-rdf/mapping%20to%20rdf.htm> 
> if you want to see how I tried to map this to RDF.

That's fine, and certainly some people use the convention that RDF URIs 
should enable retrieval of their definitions.  It's just that RDF 
doesn't *mandate* that convention.  There's room for additional 
development work in this area (which is in fact going on).

> The strange experience with the XML Schema forum was that I have never 
> understood what the end products of the, undoubtedly serious, efforts of 
> the other subscribers was. And to some extent I have that experience 
> somewhat already with the semantic web scene. My colleague Onno Paap of 
> Fluor Corp. complained to me that he was unable to find out (for RDF and 
> OWL) where "the rubber meets the road". But we keep trying.
> Thanks for your help!
> Hans

De nada.

Received on Wednesday, 6 April 2005 21:01:25 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 1 March 2016 07:41:45 UTC