Re: [Linking-open-data] watchdog.net and LOD best practices

Hi Matthias,

> But are "data objects" not just another layer of abstraction? RDF/OWL 
> gives us the possibility to use URIs and ontologies to describe the 
> things we are interested in (districts, people, protein sequences) 
> without taking detours through thinking in terms of documents, data 
> sets, conceptualisation and other abstractions. When we say that 
> <http://example1.org/john_doe> owl:sameAs 
> <http://example2.org/john_doe>, this is a statement about an object in 
> reality, and not a data object (and, of course, not about a document 
> either). Maybe this is just a matter of semantics, and with "data 
> objects" you were actually referring to these objects in reality. For 
> me, the term "data" is strongly associated with artefacts in our 
> computer systems, and not objects in reality.
>
I don't think I agree here. In my sense, all resources "described" in 
RDF are representatitons. There are no thing such as "non-information 
resource". Even if the description of a "resource" in the URI rfc say 
that it can refers to actual "things", we only have their RDF 
representations. So, I do think that subjects and objects resource in 
triples are representations of some things, and not the actual things.

But, we have to keep in mind that the Web is a of document (information 
resource, web resource, call it what you want) and a way to communicate 
these documents. Like in a RDF world, on the web, you can't refers to 
"non-information resource", so you can't say that a URI is the URI that 
defines some "thing". There is no way to infer that a URL is *not* an 
information resource (so, a non-information resource).

About sameAs, see bellow.

>> We have to do a better job of explaining what Object Identity is about
>> in the context of the Linekd Data Web, especially as this is an area of
>> computer science that predates the Web.
>
> I think foundational ontologies such as DOLCE [1] and BFO [2] could be 
> of great utility for clarifying how identities of objects and classes 
> are defined. I guess that such 'strict' approaches to ontology 
> engineering are currently not very popular in the LOD community, but I 
> think that their value will soon be recognized when the LOD dataset 
> cloud grows -- and more and more questions about the actual meaning of 
> 'sameAs' and 'equivalentClass' arise. At least, this was the case with 
> the complex RDF/OWL datasets we are dealing with in the life science 
> and health care community [3].

Well, what is a Class and what is an Individual of a Class? The line is 
gray, and and someone's class will be another person's individual. There 
are certainly best practices such as DOLCE and BFO, but these are not 
sacred books, and different usecases will have an hard time on them.

How to use equivalentClass and sameAs? I think this is quite simple 
(hahaha... well...). Basically:


equivalentClass: say that all individuals of a class A belongs to a 
class B, and all the individual of a class B belongs to a class A. 
However, this doesn't mean that A and B have the same semantic! (check 
the OWL doc for more information).

sameAs: can *only* relates two *individuals*. It means that the two 
individuals are exactly the same thing (including their semantic).

So you can just use sameAs on a Class in OWL/Full when the class is not 
only a class but also an Individual of another class.

So, as you can notice: what is a Class and what is an individual becomes 
even more fuzzy in OWL/Full since a class can *also* be an individual of 
another class :)

But since we are in a Web of Representations, there will be errors and 
inconsistencies (there are on the Web, and there will be on the Semantic 
Web)...

And this is normal: my representation of the World, is not the same as 
your Representation of the World.


In a case or another.... the discussion is not close (check on the TAG 
mailing list for example ahahah :) )



Take care,


Fred

Received on Thursday, 17 April 2008 12:15:06 UTC