Pat, see my responses inline below.
On Wednesday, June 22, 2011 at 1:13 AM, Pat Hayes wrote:
> On Jun 21, 2011, at 11:24 PM, Chime Ogbuji wrote:
> > The Relations Ontology is very central to many of the recently developed biomedical ontologies and (speaking only for myself - since is the realm of my day job) is sort of a foundation for many of them.
> Well, this is an aside from the theme of this thread, but if this is true then these ontologies have far worse problems than a mere change of identifiers. But leaving that aside, please tell me how having unreadable and meaningless identifiers is going to help with the (admittedly very real) problems which arise when identifiers are changed?
It won't, but having a change with such an impact in a mature and widely deployed ontology for mostly stylistic consideration (IMO) doesn't help either.
> > The impact of such a change would be quite large. Certainly, the ontologies I develop and the ones I use would have to undergo what I would consider a substantive revision.
>
> How would this be any easier if it was a change of meaningless identifiers rather than a change of meaningful ones? And don't tell me that meaningless identifiers never change.
>
I'm not saying that meaningful identifiers would solve the ontology evolution problem. My concerns are in regards to the immediate effect of what I think is an unnecessary change and the impact on readability. The former is certainly a one time thing, but the latter is a long running problem (until our tools catch up with the additional burden imposed on them).
> Seems to me that the central issue is that important 'foundational' ontologies should simply **not change**: that any change should be a new version, with new identifiers. This is very much what Tim B-L was talking about with his doctrine of 'cool URIs'.
>
I would tend to agree with this approach.
> Ironically, the use of SW technology - specifically, URIs and XML namespaces - is an elegant way to handle this. To revize something like the RO, leave it alone and change the root URI for the new 'version', while keeping the identifier extensions unchanged. Then a single change in the header of an XML file will be enough to switch all the RO names from their old to their new meanings. (OK, I know this is a simplification, but you get the idea.)
>
Yes, I do and I think mechanisms like this can go a long way in managing the impact of inevitable change to evolving, foundational ontologies.
--
Chime Ogbuji
Sent with Sparrow