I think there's a fundamental difference between MODS and DCTERMS that
make this nearly impossible. I've sometimes described this as the
difference between "metadata as record format" (MARC, oai_dc, MODS, etc)
and "metadata as vocabulary" (DCTERMS, DCAM, & RDF Vocabs in general).
These aren't incompatible, but the semweb and dc communities haven't
quite figured out how to define a record format from an unending
collection of vocabulary elements. DC Application and Description Set
Profiles and the W3C's Named Graphs are, IMO, steps in this direction.
"Converting" MODS to DCTERMS doesn't make much sense in this context
though. MARC/MODS are record formats, so their metadata comes in the
form of complete sentences. The trouble is that the grammar for those
sentences is either: free text people-grammar -or- derived from many
different sources (code-lists, ISBD, MARC, AACR#, and now RDA).
Translating this by taking the words and mapping them to a set of terms
without thinking about the grammar gives results like running text
through babblefish. It might be readable, to an extent, but it's
certainly not round-trip-able and it may or may not make sense.
Just building a record format from the full catalog of DCTERMS doesn't
make much sense either. DCTERMS is just a list of words that DC thinks
might be useful in resource descriptions. That's part of the reason DC
never made an oai_dcq. Without an application's context, it would be of
little value. Plus, DCTERMS is added to from time to time, and even the
namespace isn't the full set of DCAM compatible properties. For example,
DC endorses the MARC Relator Terms defined as RDF properties:
http://dublincore.org/usage/documents/relators/
More importantly, the folks who are looking for a set of classes and
properties for RDF-ized library metadata are turning to many places
beyond the DCTERMS namespace: FOAF, Bibliontology, SKOS, and many
others, including growing their own.
RDF's grammar comes from the RDF Data Model, and DC's comes from DCAM as
well as directly from RDF. The process that Karen Coyle describes is
really the only way forward in making a good faith effort to "put" MARC
(the bibliographic data) onto the Semantic Web.
Best,
-Corey
MJ Suhonos wrote:
So, back to my statement, let me re-state it as:
"dcterms is so terribly lossy that it would be a shame to reduce MARC21 bib data to
it."
Right � sorry, I think I did understand your original point as meaning this,
but both you and Eric reiterate a fine point about the endless confusion
between MARC-as-data-format and MARC-as-semantic-model.
I still stand by my point, though, in asking *why* it's a shame to reduce it
(and introduce loss). Let me try to clarify below.
Some of those elements may seem to be overkill ("Alternative Chronological
Designation of First Issue or Part of Sequence"), but the fact is that someone
somewhere in cataloger land has found a use for it.
Yes, even to me as a librarian but not a cataloguer, many (most?) of these
elements seem like overkill. I have no doubt there is an edge-case for having
this fine level of descriptive detail, but I wonder:
a) what proportion of records have this level of description
b) what kind of (or how much) user access justifies the effort in creating and
preserving it
My general rule has always been to retain the most detailed level of
granularity that you can, because in indexing, display, etc. you can easily
mush elements together, but once you've put them together it's devilish to get
them back apart again.
Absolutely, and from that perspective I understand considering loss as
something to be avoided at all costs. But I also wonder what degree of
diminishing returns we see in cataloguing practices that are reflected in our
devotion to this rule, and thus our descriptive standards.
So, to clarify my intent further: I'm looking to come up with something that is
based on the 80/20 (or perhaps 90/10) rule � that is, losing the top 10-20% of
detail in exchange for a vastly simpler (and thus easier to work with) data
model. Isn't that what MODS and DCTERMS do, roughly?
Obviously this will not work as a canonical record that preserves all of the
human effort that has gone into cataloguing, but if it makes that 80-90% of
metadata (still a huge number) easily available, that seems like a huge step
forward in terms of increasing access to me.
The non-library computer types don't appreciate the value of human-aided
systematic description.
I think I appreciate the value of human-based description pretty well. My concern is that the
"mind prison" that we attribute to MARC and its intricacies may actually be a symptom of
catering to a million edge cases of "someone somewhere in cataloger land" rather than
focusing on working for the bulk of use cases.
But I realize this now sounds like an NGC4lib thread, and for that I apologize. :-) So,
to keep it pragmatic: it sounds to me that people think doing something as
"basic" as getting millions of records out of binary MARC format into something
as lossy and unrefined as DCTERMS to expose them isn't a worthwhile effort?
MJ
NB: When Karen Coyle, Eric Morgan, and Roy Tennant all reply to your thread
within half an hour of each other, you know you've hit the big time. Time to
retire young I think.
--
Corey A Harper
Metadata Services Librarian
New York University Libraries
20 Cooper Square, 3rd Floor
New York, NY 10003-7112
212.998.2479
[email protected]
