Tuesday, January 02, 2007

FRBR OO - Not?

Posted on the FRBR blog was a link to an article by Allen Renear and Yunseon Choi
Allen H. Renear and Yunseon Choi: Modeling Our Understanding, Understanding Our Models: The Case of Inheritance in FRBR (95 KB PDF). In Grove, Andrew, Eds. Proceedings 69th Annual Meeting of the American Society for Information Science and Technology (ASIST) 43. Here’s the abstract:
They argue against seeing FRBR as having inheritance between the Group 1 entities because only the Item entity is concrete, the others are abstract.

The argument is simple: FRBR describes works as abstract and items as concrete. If all properties of “higher” entities are inherited by “lower” entities then items inherit the property of being abstract, and therefore items will be both abstract and concrete. But nothing is both abstract and concrete - therefore there is no unlimited general property inheritance in FRBR.
They make their point using a symbol set that isn't part of my vocabulary, so I'm taking on faith that they've proven this adequately. I have to say that I tend to consider all aspects of metadata to be abstract in nature, since it is a representation of something else, so their argument doesn't quite work for me.

This brings up for me, however, some larger issues, such as: Do we need a bibliographic concept that we can describe as a formal model? The FRBR model doesn't appear to survive formal analysis (see citations in the article), but does that really matter? I'm not a great fan of formality (at least not compared to some other folks), but it worries me that we are embracing a concept that we may not all understand in the same way. I have twice seen references to the "Work" entity as being "the idea." This strikes me as being horribly wrong, but without something a bit more (pardon the expression) concrete to go on, I don't see how we are going to come out with a definition that we can all agree on. And if we need to jigger the FRBR model a bit to make it work better, what's the mechanism for doing so?


Anonymous said...

I think the only point of a model like FRBR is to wprk toward a situation where we all DO have 'the same idea'. I think 'we' have not, historically--at least not sufficiently enough of the 'same idea' for 21st century inter-operability concerns. I think developing this 'same idea' is the only reason for the FRBR model effort--even if the specifics of the FRBR model aren't exactly right, it is the only effort currently existing to do that. (Well, calling it an 'effort' may be a bit of a stretch, since it isn't really being actively 'developed', but that's another story...)

So if we don't _currently_ all have sufficiently 'the same idea', that's not neccesarily a problem with FRBR, it could be because FRBR hasn't 'caught on' yet. OR, it could be because there's something lacking in FRBR. Hard to say.

Then there's the question of what is "sufficient enough" of 'the same idea'? That's tricky too.

But I agree that I find the particular argument mentioned here rather unconvincing. I feel like they are evaluating FRBR by philosophical criteria that aren't really that relevant to the goals of FRBR. On the other hand, I agree in general that FRBR is probably NOT unambiguous and consistent enough to help us work toward having sufficiently 'the same idea'. FRBR shouldnt' be thought of as 'done', it's a first draft---the problem is that no work is being done to advance it (which would require people starting to figure out how to USE it, to see where the problems are).

Jonathan Rochkind

Allen Renear said...

Just a note to clarify that while the particular argument cited is decisive against general unqualified inheritance in FRBR, it is not intended as support for our principal critique of more plausible positions, and we present it mostly as a first step in clarifying the possibilities. As we say in the paper: "It may be objected that this argument is sound but irrelevant as the only properties ever at issue were the attributes ... explicitly specified in FRBR" (which excludes arbitary properties such as "being abstract" or "being a work").

The more plausible position pro inheritance (but still wrong we think) is that it is the attributes (all or some) explicitly identified in FRBR that are inherited by so-called "lower level" entities. We take this position seriously and discuss it at length, but argue that:

(i) it isn't implied by the FRBR model; and

(ii) it isn't needed for any of the reasons most likely to be advanced for holding it: collocation and denormalization.

The first claim can be established by inspecting the ER diagram and scrutinizing the text. The second claim we support by showing how collocation and denormalization are both be achieved naturally within the FRBR model without inheritance.

In addition we also offer a positive argument (from category error) to the conclusion that inheritance of properties such as typeface and language (likely inheritance candidates, on our reading of FRBR expositions), ought not to be allowed.

-- allen