And the other issues below:
The role of a BIBFRAME Agent with respect to a Work/Instance is expressed either by (a) a URI; or (b) a string, when there is no known URI for that role.  The paper finds fault: In method (a), with the perceived restriction of role URIs to LC relators. With method (b), in general.
The first is simply a misperception. There is no such restriction; the URI can come from any appropriate role vocabulary.

That's great, of course :) Thank you for clarifying.  

The issue I have with it isn't the use of LC relators, it's that existing relationships can't be used, as the object has to be a bf:Authority, rather than a real person. However, given the discussion in the authority thread, that seems like it's understood -- no need to belabor the point further.

As for “method (b)” it is simply impractical to insist that every role have a URI, particularly since during early BIBFRAME implementation much of the BIBFRAME data will be converted from MARC records.  

And, I think for the first time, I'm not sure that I agree here.  If we can mint URIs for resources that are generated from MARC records, as per the examples on the site, then individual libraries should be equally able to mint their own relationship URIs.  They might not be semantically useful, but neither are the strings in Relators.

See also the global vs locally defined discussion -- if the relationship is defined, described and maintained ... that seems great!

If a MARC record represents a person’s role as “ed.”, whoever does the conversion might not be able to find an appropriate URI for that role. The string should not simply be discarded; it should be retained if for no other reason than that a human end-user might be able to make sense of it.  

This is somewhat more convincing, but the same solution applies -- automatically create a relationship and include in rdfs:comment or similar a note that the original relator was "ed."

As a compromise it would be good to treat Relator in the same way as the other similar classes: give them a node that starts off blank and records the string, but can be swapped out in the future for a real URI.  It should also have properties like those of Identifier, such as Assigner, Qualifier, Status, and Scheme.  This would allow future reconciliation efforts to distinguish "Ed." from Stanford with "Ed." from Harvard, if the two institutions were to use them in internally consistent but different ways.

Predicate Proliferation 
This section points out that there are perhaps many more properties in BIBFRAME than necessary, and many could be eliminated.  This is a legitimate concern, and BIBFRAME is and will be undergoing vocabulary refinement, informed by the current BIBFRAME testbed activity.

I'm happy to provide a better initial list of these to fuel the discussion, if that would be useful. 

Record vs.  Graph
Based on the examples, we believe this section draws almost entirely on a mistake in the Bibframe vocabulary.  All of the properties mentioned in this section are currently associated directly with Work and Instance resources.  They should instead have a domain of bf:DescriptionAdminInfo We regret this mistake.

Ahh, that is much clearer. Thanks! 

A related question... what's the range of bf:derivedFrom?  Is it also bf:DescriptionAdminInfo, or is there a missing "bf:OriginalRecord" class or similar?


Model Inconsistency
The issues raised here do deserve closer inspection.  Rob speculates that the reason why bf:Event and bf:Provider are not of class bf:Authority is because there are not library authority lists for events and providers.  That isn’t the reason; one has nothing to do with the other.  Anyway, bf:Event and bf:Provider are currently undergoing re-thinking.

Okay, happy to hear that!  Event and Provider are the most obvious cases, if it would be useful I can try and generate a more comprehensive list?


Rob Sanderson
Technology Collaboration Facilitator
Digital Library Systems and Services
Stanford, CA 94305