I was glad to see Joe Kiegel initiate this discussion and am responding now not so much to put a point of view as to give the thread a bump. The existence of the RDA vocabularies has been the elephant in the room in discussions about BIBFRAME and this attempt to compare their characteristics is very welcome.
However, I'm not sure it's clear yet how we need to frame that comparison. Joe points out, for instance, that BIBFRAME provides for administrative metadata while RDA does not; but the current BIBFRAME provisions come directly from MARC and are arguably not going to be well suited to RDF data.
The most obvious contrast between BIBFRAME and RDA/RDF is that the latter is a dedicated vehicle for a content standard. Given the difficulties we've already experienced trying to shoehorn RDA into MARC, that would appear to be a difference worth heeding. But we're also told that vocabularies should be designed for easy reuse by other communities, and I've heard that advanced as an argument for adopting a simpler core vocabulary. (As I understand it, RDA/RDF's "unconstrained" properties were introduced partly in response to that same demand.)
I suspect a lot of people, if they're like me, are going to want help knowing how to think through questions such as these. Karen's remark about needing to "re-iterate back from our attempts to create a viable RDF version of library data to the cataloging rules, and create at least a subset of the rules that can support a viable data format" strikes me as a useful starting point for discussing the relationship of cataloguing rules to the vocabularies they're to be expressed in, and not only in connection with identifiers but also generally. 


Chew Chiat Naun

Cornell University Library

[log in to unmask]

(607) 254 8031


On 1 April 2015 at 12:52, Karen Coyle <[log in to unmask]> wrote:
Diane, if you look at a bunch of RDF vocabularies, you will find that the ratio of classes to properties is often around 1/3 - 1/5, and some have even more classes than they do properties, with extensive class/subclass relationships. Some of these latter perhaps exhibit "class-itis," as Dean Allemang in his design books calls it, such as FaBiO which has about 300 properties and 750 classes. RDA has 10 classes and over 900 properties, and I believe (although I'd have to download and re-count the various RDA vocabularies) no sub-classes. BIBFRAME has 104 classes with 51 subclass relationships and 544 properties (in a version I have on my hard drive). So I'd say that the 1/90 ratio of classes to properties in RDA is pretty flat, along with the 0 sub-class to class relationship, if that's the case.

My guess is that as it matures, RDA will develop a deeper structure because that will facilitate some data processing functions. You may recall a discussion here a few months ago where developers wanted to implement bibliographic forms as subclasses rather than properties because they find them easier to work with. I can't attest to that, but my experience over other vocabularies shows that classes are heavily used, and I'm assuming that is for good reason.


On 4/1/15 9:07 AM, Diane Hillmann wrote:

I'm mystified by your assertion that RDA "has virtually no class relationships--it's essentially a flat data space".  RDA's classes are based on the FRBR model--the antithesis of 'flat data space'.

It's my understanding that there is more than one way to skin the search beast, with classes being only one. Title in RDA has many subproperties for the specific kinds of titles: variant titles, parallel titles, series titles, etc., and it's my understanding that those relationships, too, can define the boundaries of searches.

How to accommodate strings and identifiers is certainly an issue, though still an open one for RDA. I agree that blank notes is not an attractive solution.


On Sat, Mar 28, 2015 at 10:05 AM, Karen Coyle <[log in to unmask]> wrote:
Joseph, thanks for doing a comparison. Note that BF has about 400 properties, while RDA has nearly a thousand, so it is true that RDA is more detailed that BF. However, RDA has virtually no class relationships -- it's essentially a flat data space. This will have implications for the use of RDA in actual systems, since class relationships help you do things like "search all properties in the title class."

On 3/27/15 8:04 AM, Joseph Kiegel wrote:

Under the influence of MARC, BIBFRAME has a large set of properties for identifiers while RDA is limited.

The RDA rules often allow either strings or identifiers. RDA in RDF is essentially silent in most cases on whether the value for a property is expected to be a string or an identifier, and therefore it can presumably be either. This, however, is highly problematic when working with RDF data. In general, it's never good to not know what kind of data to expect for a field in your metadata -- it complicates input interfaces and the programs that use the data. However, if you want to have the possibility in your data to accommodate both strings and identifiers, you are kind of forced to create different properties for those different choices, which would mean nearly doubling the number of RDA properties. Although I find the use of blank nodes in BF to be a complicating factor, I assume that in many cases those blank nodes are there as a way around this string-vs-identifier problem, allowing each statement to point to a blank node that can have either or both.

To me this is evidence that we need to re-iterate back from our attempts to create a viable RDF version of library data to the cataloging rules, and create at least a subset of the rules that can support a viable data format with clearly defined data values for each property. The "string or identifier" in the rules just isn't workable in a data format.


RDA is not yet able to express subject relationships (RDA chapters 33-37) and BIBFRAME has a mechanism for this.

Holdings Information:
Although not fully elaborated, BIBFRAME has properties for holdings information while RDA has almost nothing.

RDA is richer than BIBFRAME

RDA provides properties for all parts of series statements, while BIBFRAME has a single property:  series.

RDA has more properties for specific types of notes.  While BIBFRAME has note properties, the term "note" in a property name may mean simply that its range is a literal, e.g. findingAidNote, musicMediumNote.

Technical Details of a Resource:
RDA has a large number of properties for technical details of resources such as polarity, playingSpeed, fileSize, etc.  It is not clear how BIBFRAME handles this type of information.

Inverse Properties:
RDA provides inverse properties (e.g. animator and animatorOf) while BIBFRAME lacks them.

Karen Coyle
[log in to unmask] http://kcoyle.net
m: +1-510-435-8234
skype: kcoylenet/+1-510-984-3600

Karen Coyle
[log in to unmask] http://kcoyle.net
m: +1-510-435-8234
skype: kcoylenet/+1-510-984-3600