On 7 Dec 2011, at 00:38, Alexander Johannesen wrote: > Hiya, > > Karen Coyle <li...@kcoyle.net> wrote: >> I wonder how easy it will be to >> manage a metadata scheme that has cherry-picked from existing ones, so >> something like: >> >> dc:title >> bibo:chapter >> foaf:depiction > > Yes, you're right in pointing out this as a problem. And my answer is; > it's complicated. My previous "rant" on this list was about data > models*, and dangnabbit if this isn't related as well. > > What your example is doing is pointing out a new model based on bits > of other models. This works fine, for the most part, when the concepts > are simple; simple to understand, simple to extend. Often you'll find > that what used to be unclear has grown clear over time (as more and > more have used FOAF, you'll find some things are more used and better > understood, while other parts of it fade into 'we don't really use > that anymore') > > But when things get complicated, it *can* render your model unusable. > Mixed data models can be good, but can also lead directly to meta data > hell. For example ; > > dc:title > foaf:title > > Ouch. Although not a biggie, I see this kind of discrepancy all the > time, so the argument against mixed models is of course that the power > of definition lies with you rather than some third-party that might > change their mind (albeit rare) or have similar terms that differ > (more often). > > I personally would say that the library world should define RDA as you > need it to be, and worry less about reuse at this stage unless you > know for sure that the external models do bibliographic meta data > well. >
I agree this is a risk, and I suspect there is a further risk around simply the feeling of 'ownership' by the community - perhaps it is easier to feel ownership over an entire ontoloy than an 'application profile' of somekind. It maybe that mapping is the solution to this, but if this is really going to work I suspect it needs to be done from the very start - otherwise it is just another crosswalk, and we'll get varying views on how much one thing maps to another (but perhaps that's OK - I'm not looking for perfection) That said, I believe we need absolutely to be aiming for a world in which we work with mixed ontologies - no matter what we do other, relevant, data sources will use FOAF, Bibo etc.. I'm convinced that this gives us the opportunity to stop treating what are very mixed materials in a single way, while still exploiting common properties. For example Musical materials are really not well catered for in MARC, and we know there are real issues with applying FRBR to them - and I see the implementation of RDF/Linked Data as an opportunity to tackle this issue by adopting alternative ontologies where it makes sense, while still assigning common properties (dc:title) where this makes sense. > HOWEVER! > > When we're done talking about ontologies and vocabularies, we need to > talk about identifiers, and there I would swing the other way and let > reuse govern, because it is when you reuse an identifier you start > thinking about what that identifiers means to *both* parties. Or, put > differently ; > > It's remarkably easier to get this right if the identifier is a > number, rather than some word. And for that reason I'd say reuse > identifiers (subject proxies) as they are easier to get right and > bring a lot of benefits, but not ontologies (model proxies) as they > can be very difficult to get right and don't necessarily give you what > you want. Agreed :)