Thank you, Sally! I have been beating the drums on a related point (the difference between data objects singlularly associated with particular organisations, people or projects -- like specimens and datasets; vs. data objects likely to be shared by many -- like taxon names & concepts, reference citations, and named localities) for years now. Your observation of other practical differences relating to the impact of GUID duplication further emphasizes the distinction.
Perhaps the two break-out sections at the workshop should be split according to "GUIDs for data objects not likely to be duplicated across multiple providers" (existing "specimens" group), vs. "GUIDs likely to be shared by multiple providers" (existing taxon names/concepts group).
I would imagine that the former group would lean toward provider-assigned GUIDs and emphasis on fast/easy implementation; whereas the latter group would lean towards centralized GUID assignment (or at least coordination of GUID assignment) and emphasis on developing software tools to cross-walk multiple independent datasets with broadly overlapping content of shared data objects, in order to minimize redundant GUID assignment to shared objects.
Aloha, Rich
-----Original Message----- From: Taxonomic Databases Working Group GUID Project [mailto:TDWG-GUID@LISTSERV.NHM.KU.EDU]On Behalf Of Sally Hinchcliffe Sent: Sunday, January 29, 2006 11:13 PM To: TDWG-GUID@LISTSERV.NHM.KU.EDU Subject: How GUIDs will be used
Something I was thinking about over the weekend (I really must get a life)
I was just reading the California Digital Library paper on ARK identifiers (I think it got circulated many moons ago, yes I am just getting caught up on my 'homework' now...) and the following assertion stood out: (page 3)
'[W]hat we're looking for are persistent actionable identifiers, where an actionable identifier is one that widely available tools such as web browsers can use to convert a simple "click" into access to the object ...'
This strikes me as only half true. For specimens, yes, most of the use of GUIDs will be ultimately to allow a user to get their hands on either the specimen itself or an electronic object sufficiently informative about the specimen (e.g. a picture or a grid ref) that they can do something scientific with it. In that case it mostly doesn't matter if one specimen has multiple different ids (due to aggregations, splittings, derivatives, sampling etc.) as long as all of those ids lead back to the same thing.
But for the more abstract things like names, concepts and maybe even descriptive terms, I can imagine the guids themselves becoming widely used without there ever (or very rarely) being a reference made back to the source of the id because there's not much more _there_ (other than confirmation that no mistakes have been made in transmitting the id). The real benefit of _these_ GUIDs is in being able to make machine-computable statements about x being the same as or different to or composed of y. In this case, multiple guids referring to the same thing does become a problem because each duplication dilutes the amount of information carried by that GUID as to whether x truly is different from y, and reduces the chance that two of the same things will carry the same id.
Does this make sense? It suggests that the delegation model for specimens (e.g. just let each collection issue its own ids for anything it likes) might have to differ from the one for names, concepts or other largely abstract objects (where we might want to explicitly divvy out the domains to particular nomenclators, concept banks and so on)
Sally *** Sally Hinchcliffe *** Computer section, Royal Botanic Gardens, Kew *** tel: +44 (0)20 8332 5708 *** S.Hinchcliffe@rbgkew.org.uk