Re: Disjointedness of FRBR classes

Quoting Tom Baker <tbaker@tbaker.de>:


> Please remind me who exactly is saying that the WEMI classes are formally
> disjoint.  There are several formalized expressions of FRBR in circulation.
> Which one(s) do you mean here and what is the status of that expression
> according to IFLA (or JSC, or anyone else)?  I was under the impression that
> the RDF expressions were all still just drafts, hence subject to possible
> revision...

Tom,

If you look at the FRBR properties in the OMR, you see things like:

- is created by (corporate body)
- is created by (person)

etc. It's really quite instructive to look at the properties (all 216  
of them!):
   http://metadataregistry.org/schemaprop/list/schema_id/5.html

Which tells me that FRBR limits "is created by" to only FRBR-defined  
entities as objects. Otherwise, you could have an "is created by" that  
is open to other entities, but you do not. Every property is assigned  
solely to a single class for domain and a single class for range. The  
properties in the RDA namespace in the OMR that are not constrained to  
a single FRBR entity are considered by JSC to be outside of RDA and  
not RDA compliant. (In fact, we've talked about moving them to a  
different namespace to make this clear.)

The thing is that the IFLA/FRBR documentation doesn't use the term  
"disjoint" because it is not in their vocabulary, and I don't see any  
documentation for the registered versions besides what is in the  
registry.

There is the page on the LLD wiki which set up a conference call we had:
   http://www.w3.org/2005/Incubator/lld/wiki/Library_standards_and_linked_data
which talks about disjointedness and constraints, but this is an  
informal write-up provided by Gordon. You may recall that the folks on  
that call pushed back against this rigidity.

I have no idea the extent to which members of the IFLA FRBR review  
group understand the implications of a formal model with disjoint  
classes and properties assigned to them. The RDA folks seem surprised  
when some of us demonstrate to them various ways that they've painted  
themselves into a corner. This tells me that they simply do not  
understand the implication of formal models, and were assuming that  
you could "fudge" if you needed to. At the same time, they argue for  
precision and rigidity.

(Note, there is a "FRBRer Open Model which is defined as "Based on the  
current text, Feb 2009. This element set contains properties from the  
FRBRer model, from which all property domains and ranges have been  
removed." This, however, is not being used as the basis for RDA and I  
don't know really if we know what IFLA intends by this. Again, there  
isn't any documentation that I can find that explains the registered  
vocabularies. Gordon possibly could tell us.)

The FRBR *documentation* (not a formal model) appears to be more  
lenient than this: it admits that some definitions, like the  
definition of Work, could be open to interpretation. RDA, however, has  
taken a very non-lenient view. There are no properties that are not  
assigned to one and only one FRBR entity, and any deviation from that  
is deemed "wrong."

We keep going round and round on this, but I have the feeling we  
aren't talking to the right people. We really need to have this  
conversation with the IFLA and RDA development groups, rather than  
talking one-on-one. Some of us have been advocating this for a while,  
but the groups are reluctant. This is one of those areas that should  
come under: "Collaboration with the Semantic Web community."

kc

>
>> These properties exist because FRBR is *so* rigid. ...
> ...
>> If bibo:Book or bibo:Article or dct:BibliographicResource are
>> inherently disjoint with FRBR (since they do not constrain you from
>> violating FRBR rules), the ov:commonThing properties let you express
>> FRBR relationships on these resources without making your reasoner
>> implode in a puff of logic.
>
> What an image...!! :-)  Seriously, has anyone suggested that the disjointed
> nature of these classes be re-assessed in light, say, of the principle of
> minimal semantic commitment?  Is it too late for such a discussion?
>
> Tom
>
> --
> Tom Baker <tom@tombaker.org>
>
>



-- 
Karen Coyle
kcoyle@kcoyle.net http://kcoyle.net
ph: 1-510-540-7596
m: 1-510-435-8234
skype: kcoylenet

Received on Sunday, 23 October 2011 21:19:54 UTC