Monday, July 27, 2009

Thoughts on FRSAD

I don't usually publish my individual comments on things sent out for review within our community, but I've decided to make an exception for the FRSAD report. I'm actively working with a FRBR implementation (and trying to take in as much of FRAD as we can), and anything I can do to help push FRSAD (FRSAR? what's in a name? ha - there's got to be a FRAD joke in there somewhere...) to be useful to the work I'm doing I see as a good thing. So here are the comments I sent in through official channels.

-----------------

In short, I think good work has been done here but it doesn't meet my needs as someone working diligently (and actively implementing FRBR and FRAD) to re-imagine discovery systems in libraries.

While I am a great believer in the power of user studies to inform metadata models, I believe inappropriate conclusions have been drawn here. It doesn't surprise me at all that users had trouble sorting actual subjects into categories such as concept, object, event, place. But that doesn't mean our models shouldn’t make that distinction. Users wouldn't be able to distinguish between Work/Expression/Manifestation/Item, either, but those are still useful entities for us to use underlying our systems.

The draft report rightly notes that the concept/object/event/place division is only one way of looking at it, that other divisions such as those outlined by Ranganathan and the framework (which seems to be basically abandoned?). But that's the very essence of a *model* - to pick one of many possible representations and go with it, in order to achieve a purpose. The fact that competing interpretations are possible is not a rationale for abandoning selecting one that can advance the purpose of the model (even taken together with user studies showing users don’t gravitate to any one specific division). By choosing concept/object/event/place (or Ranganathan's model, or , or any other option) we can delve deeper into the modeling we need to do and provide a way forward for our discovery systems. By refusing to do so, we don't advance our case the way we must.

The thema/nomen structure outlined here is very useful. However, I believe strongly the report should not stop here. Going further is often stated here as "implementation dependent" but I think there is a great deal of room for the conceptual model to grow without venturing into actual implementations. Certainly FRBR and FRAD take that approach.

In general, the thema/nomen structure could apply to any attribute or relationship under vocabulary control. There is great (and unfortunately here unexplored) potential for this model to apply beyond simply aboutness. Limiting it in this way I believe is a disservice to those of us who are attempting to use these models to reinvent discovery systems.

I'm concerned about the significant lack of cohesion between the FRBR, FRAD, and FRSAR reports. They show their nature of independently generated by different groups with different interests over a long span of time. This limitation definitely needs to be overcome if these reports are to be useful as a whole for the community. Each could be used on its own, but we need a more coherent group. In fact, the thema/nomen structure in the FRSAD draft isn't really all that different than the (whatever entity)/name structure presented in FRAD. Much greater cohesion of the three reports could be made - what's written here seems to ignore FRAD in particular. I believe this is a missed opportunity. I think the most significant mismatch between the three reports is where they draw the boundary for how far a "conceptual model" should go.

On a higher level note, the report reads more as an academic paper outlining alternative options rather than providing a straightforward definition of the conceptual model. I respect the background work done here, and believe it needs to be done. There's a lot of room for papers like that in this environment; however, this report series needs to serve practitioners better and stick closer to the model.

On a more practical note, in the report the Getty AAT is often referred to by example. Yet most of the facets in the AAT bring out the "isness" (which in the introduction is explicitly described as out of scope) rather than "ofness" or "aboutness". For example, on p. 45, #7 under "select," "ale glass" in AAT is intended to be used for works of art that ARE ale glasses, not works (presumably textual) that are ABOUT ale glasses. This internal inconsistency is a serious flaw in the report.

I'm certainly not one to promote precoordinated vocabularies, but they exist in library metadata and we must deal with them. It's unclear to me from this report how these fit into the model proposed.