Hi Karen, all,
Very good points! Indeed repeating all RDA elements in BIBFRAME seems a
bit redundant, and using RDA classes/properties directly in BIBFRAME
wouldn't seem like an extension. But there's a difference between the
vocabularies you mention (e.g. MADS/RDF) and using RDA elements, since
while the former has been built in to BIBFRAME to some extent (LOC
wisely decided not to reinvent everything), the latter is completely
unspecified since the RDA model is, when you get to details, quite
different from BIBFRAME.
Here are some thoughts on why using the RDA-in-RDF vocabulary is a bit
difficult currently, and why BIBFRAME-with-RDA might be a better choice
for some scenarios:
While RDA-in-RDF is a full RDF model, it's not really published as such.
The RDA registry is structured as a set of element sets (classes,
properties and values). This is not what is generally expected from a
data model (ontology). It's very difficult to understand the model based
on what's available from the RDA registry, as it's just a set of
disconnected lists of elements. There are a few examples [1] but they
are not really explained. There are also R-balls available [2], but they
are not explained either. Those are about the only examples there are
since, to my knowledge, almost nobody has published RDA-in-RDF data (two
counterexamples are mentioned on the RDA registry site [3]). While the
BIBFRAME documentation could be improved a lot, it does a much better
job of explaining the model and how to apply it in practice than the RDA
registry site.
The "language-agnostic", opaque naming of RDA elements doesn't really
help. Quick, what does rdam:P30088 stand for? What about this triple:
"ex:E1 rdae:P20001 rdaco:1020"? I know this has been debated to death
before within the RDA community, but I think that a data model that
doesn't use any human-understandable mnemonics is doomed to fail. A
triple where neither the subject, property nor object is immediately
recognized by a human being is pretty difficult to work with. This is
like catalogers talking in MARC field codes, just much worse, since
there are many more elements in RDA, they are not generally visible in
any UIs (instead RDA toolkit numbering is often used), and the
identifiers intentionally have very little structure - MARC codes at
least tend to follow some patterns such as x00 for people.
To my knowledge, there is just one generally available tool that can
work with RDA-in-RDF: RIMMF. I've tried it (in a very nice Kiviathon
workshop!) and it does a fairly good job of converting legacy MARC
records to the RDA model, and also gives a rather intuitive display of
the model (though personally I'd structure the UI a bit differently,
perhaps using nested forms instead of overlapping windows). But it's not
open source so its inner workings (including the conversion specs)
cannot be inspected nor built upon. It's clearly a prototype and anyone
intending to do real cataloging in pure RDA (not the MARC flavor) has to
implement similar functionality from scratch. And since very few
institutions currently have real RDA data that would actually follow the
WEMI model, bootstrapping the ecosystem of RDA-in-RDF data is pretty
difficult.
I don't mean to criticize RDA in general (although it surely could be).
I think it's just obvious that not much resources have been invested in
the RDA-in-RDF model, the main focus is elsewhere. The consequence is
that the RDA-in-RDF model is unlikely to ever be applied widely. Some of
the properties (especially the unconstrained ones) are used in a few
places, but not the full model. I think we were the first, or at least
one of the first, to apply the RDA-in-RDF constrained properties for
corporate names when publishing our corporate names dataset [4] - and
the RDA agent model is a lot simpler than the WEMI-based bibliographic
model.
In contrast, starting with the current BIBFRAME model, which at least
has been properly specified as a data model and has several tools
available, and then adding RDA elements as necessary for the finer
details, makes a lot of sense. One can convert existing MARC records to
BIBFRAME using marc2bibframe2 (or one of the other conversion tools that
exist or are being built, like bib2lod or the Casalini Libri products)
and then expand the model, cherry-picking elements from RDA as
necessary. But ideally there would be some guidance available on how to
do this, with examples from actual data sets that have been expressed
using this hybrid model. Maybe this would be a task for the larger
BIBFRAME community?
-Osma
[1] http://www.rdaregistry.info/Examples/
[2] http://rballs.info/
[3] http://www.rdaregistry.info/rgDatasets/
[4] http://finto.fi/cn/
Karen Coyle kirjoitti 29.10.2017 klo 00:31:
> On 10/25/17 12:23 PM, Joseph Kiegel wrote:
>> This is interesting work but it doesn't meet my goals. I want to see
>> full support of RDA built into BIBFRAME, not as extensions.
>
> I suspect that "full support of RDA" could only be an implementation of
> RDA in RDF, which is the vocabulary integrated into the RDA Toolkit. It
> seems redundant to repeat all of that in BIBFRAME since it already
> exists. (I have no idea why work has been done on BIBFRAME and not
> RDA... that's beyond my ken.)
>
> Using elements from another vocabulary isn't what I would consider to be
> an extension, and BIBFRAME already uses a good handful of vocabularies
> including rdf, rdfs, madsrdf:
>
> @prefix bf: <http://id.loc.gov/ontologies/bibframe/> .
> @prefix bflc: <http://id.loc.gov/ontologies/bflc/> .
> @prefix madsrdf: <http://www.loc.gov/mads/rdf/v1#> .
> @prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
> @prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#> .
> @prefix xml: <http://www.w3.org/XML/1998/namespace> .
> @prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
> @prefix zs: <http://docs.oasis-open.org/ns/search-ws/sruResponse> .
>
> I think, though, that your question mainly speaks to "why BIBFRAME and
> not RDA?"
>
> kc
>
> I don't
>> want to get "pretty far" toward expression of relationship
>> designators, I want full, transparent and easy expression of the full
>> set of designators. Trying to infer relationships from higher level
>> designators and genre codes is fraught when a work has multiple
>> relationships.
>>
>> Class proliferation is an unavoidable fact of BIBFRAME, given the
>> design decision to avoid proliferation of properties. The
>> bibliographic world is complex and we need either properties or
>> classes to express the intellectual distinctions we make in our
>> cataloging code. If we are not going to use properties, then let it
>> be classes.
>>
>>
>>
>> -----Original Message----- From: Bibliographic Framework Transition
>> Initiative Forum [mailto:[log in to unmask]] On Behalf Of
>> Steven Michael Folsom Sent: Monday, October 23, 2017 10:55 AM To:
>> [log in to unmask] Subject: Re: [BIBFRAME] Resource
>> relationships
>>
>> Hi Karen,
>>
>> Absolutely; I think we’re describing the same observations. Select
>> reuse of RDAU with the understanding that (as you said), the “same
>> semantics may be inferred from the triple rather than being encoded
>> in the property” was one of our goals.
>>
>> This document describes our thinking from almost a year ago now,
>> https://wiki.duraspace.org/display/LD4P/bibliotek-o?preview=/79795231/83237330/bibliotek-o_pattern_relations_201612.pdf.
>>
>> It definitely involves using deeper parts of the RDA property
>> hierarchies. Quoting from the doc:
>>
>> “In the case of derivative and equivalence relationships this paper
>> recommends the use of the unconstrained RDA properties, where RDA
>> provides for a set of more granular subproperties. However, we
>> recommend bf:references/bf:referencedBy over rdau:P60848 (“has
>> referential relationship with”). The latter property is symmetric,
>> and we see clear use cases (such as tracing a path of intellectual
>> influence) where symmetry is not desirable. We will use BF 2.0
>> properties except for relationships that we can express with
>> rdau:P60250 (“is derivative”) and rdau:P60191 (“has equivalent”) (and
>> their subproperties) and proposed properties for relating events and
>> works.Accompanying, sequential, and whole/part relationships will be
>> addressed in other discussion papers and/or future work.”
>>
>> A couple properties slipped in to our proposal that I would hope we
>> move away from (like “adapted as choreography” should be replaced
>> with the more general “adapted as”), but otherwise I think it still
>> holds up. As you said, through testing these decisions can be tuned.
>>
>> Thanks, Steven
>>
>>
>> On 10/22/17, 1:13 PM, "Bibliographic Framework Transition Initiative
>> Forum on behalf of Karen Coyle" <[log in to unmask] on behalf
>> of [log in to unmask]> wrote:
>>
>> thanks, Steven. This is along the lines of what I was suggesting. I
>> note in the LD4L document[1] that you list the RDAU relationship
>> properties and sub-properties, e.g.
>>
>> rdau:P60250 Label: is derivative URI:
>> http://rdaregistry.info/Elements/u/P60250 Definition: Relates a
>> resource to a resource that is a modification of a source resource.
>> Subproperties: ●rdau:P60115 "is modified by variation as"
>> ●rdau:P60120 "is remade as" ●rdau:P60121 "is set to music as"
>> ●rdau:P60177 "is abstracted in" ●rdau:P60178 "is indexed in"
>> ●rdau:P60180 "is adapted as choreography" etc.
>>
>> Is LD4L using the full sub-property group? (Or potentially, could
>> it?) If so, that could go a long way to making more of a match
>> between RDA and BF in Joe's list, which only includes the higher
>> level relationships.
>>
>> Or, when you say: sticking to the general properties, do you mean
>> only the highest level, e.g. "is derivative"? Does your comment
>> about including resource types in properties as an "anti-pattern"
>> include properties like "is set to music as"? I think in a sense the
>> rdau property name is reflecting a range definition, and assumes
>> perhaps a point of validation for matching the relationship and the
>> range. Is LD4L taking a strict approach on that?
>>
>> kc [1]
>> https://wiki.duraspace.org/display/LD4P/bibliotek-o?preview=/79795231/83237330/bibliotek-o_pattern_relations_201612.pdf
>>
>> On 10/20/17 12:30 PM, Steven Michael Folsom wrote:
>>> In the LD4L analysis and adoption of BF2 we chose simply to extend
>>> BF2 using select RDAU properties. [1]
>>>
>>> RDAU isn’t perfect, but sticking to the general properties would
>>> get us pretty far. Our biggest reason for only using *select* RDAU
>>> properties is proliferation caused by the anti-pattern of minting
>>> new properties to account for the type of things being related; we
>>> don’t need “is opera adaptation of” if we have the more general “is
>>> adaptation of” and an Opera class/genre term.
>>>
>>> Similarly, using relationships to create new classes (as described
>>> in Joe’s bf:ChoreographicAdaption example below) causes unnecessary
>>> Class proliferation; each class would then require parallel classes
>>> for all the different types of relationships between resources.
>>>
>>> The other option being floated below (using bflc:Relationship)
>>> seems like over engineering for a problem that could be solved with
>>> just reusing the parts of RDAU that are well designed.
>>>
>>> Is this under consideration?
>>>
>>> Respectfully, Steven
>>>
>>> [1] For more information on the LD4L analysis of BIBFRAME, see: -
>>> https://bibliotek-o.org/overview/overview.html -
>>> https://wiki.duraspace.org/display/LD4P/bibliotek-o
>>>
>>
>> -- Karen Coyle [log in to unmask] http://kcoyle.net m:
>> +1-510-435-8234 skype: kcoylenet/+1-510-984-3600
>>
>>
>
--
Osma Suominen
D.Sc. (Tech), Information Systems Specialist
National Library of Finland
P.O. Box 26 (Kaikukatu 4)
00014 HELSINGIN YLIOPISTO
Tel. +358 50 3199529
[log in to unmask]
http://www.nationallibrary.fi
|