On Dec 5, 2013, at 8:55 AM, Ross Singer <[log in to unmask]> wrote:

> Eric, I'm having a hard time figuring out exactly what you're hoping to get.
> 
> Going from MARC to RDF was my great white whale for years while Talis' main
> business interests involved both of those (although not archival
> collections).  Anything that will remodel MARC to (decent) RDF is going be:
> 
>   - Non-trivial to install
>   - Non-trivial to use
>   - Slow
>   - Require massive amounts of memory/disk space
> 
> Choose any two.
> 
> Frankly, I don't see how you can generate RDF that anybody would want to
> use from XSLT: where would your URIs come from?  What, exactly, are you
> modeling?
> 
> I guess, to me, it would be a lot more helpful for you to take an archival
> MARC record, and, by hand, build an RDF graph from it, then figure out your
> mappings.  I just don't see any way to make it "easy-to-use", at least, not
> until you have an agreed upon model to map to.


Ross, good questions. I’m hoping to articulate and implement a simple and functional method for exposing EAD and MARC metadata as linked data. “Simple and functional” are the operative words; I’m not necessarily looking for “fast”, “best” nor “perfect”. I am trying to articulate something that requires the least amount of infrastructure and technical expertise.

Reasonable RDF through XSLT? Good point. I like the use of XSLT because it does not require very much technical infrastructure — just ubiquitous XSLT processors like Saxon or xsltproc. I have identified two or three stylesheets transforming MARCXML/MODS into RDF/XML.

  1. The first comes from the Library of Congress and uses Dublin
     Core as its ontology, but the resulting RDF has no URIs and
     the Dublin Core is not good enough, even for my tastes. [1]

  2. The second also comes from the Library of Congress, and it
     uses a richer, more standard ontology, but I can’t get it to
     work. All I get as output is a plain text file. I must be
     doing something wrong. [2]

  3. The found the third stylesheet buried the MARC/MODS RDFizer.
     The sheet uses XSLT 1.0 which is good for my xsltproc-like
     tools. I get output, which is better than Sheet #2. The
     ontology is a bit MIT-specific, but it is one heck of a lot
     richer than Sheet #1. Moreover, the RDF includes URIs. [3, 4]

In none of these cases will the ontology be best nor perfect, but for right now I don’t care. The ontology is good enough. Heck, the ontologies don’t even come close to the ontology I get when transforming my EAD to RDF using the Archives Hub stylesheet. [5] I just want to expose the content as linked data. Somebody else — the community — can come behind to improve the stylesheets and their ontologies. 

Where will I get the URIs from? I will get them by combining some sort of unique code (like an OCLC symbol) or namespace with the value of the MARC records' 001 fields.

Here is an elaboration of my original recipe for making MARC metadata accessible via linked data:

  1. obtain a set of MARC records
  2. parse out a record from the set
  3. convert it to MARCXML
  4. transform MARCXML into HTML
  5. transform MARCXML into RDF (probably through MODS first)
  6. save HTML and RDF to disc
  7. update a mapping file / data structure denoting where things are located
  7. go to Step #2 for each record in the set
  8. use the mapping to create a set of site map files
  9. use the mapping to support HTTP content negotiation
 10. create an index.html file allowing humans to browse the collection as well as point robots to the RDF
 11. for extra credit, import all the RDF into a triple store and provide access via SPARQL

I think I can do the same thing with EAD files. Moreover, I think I an do this with a small number of (Perl) scripts easily readable by others enabling them to implement the scripts in a programming language of their choice. Once I get this far metadata experts can improve the ontologies, and computer scientists can improve the infrastructure. In the meantime the linked data can be harvested for the good purposes link data was articulated.

It is in my head. It really is. All I need is the time, focus, and energy to implement it. On my mark. Get set. Go.


[1] MARC21slim2RDFDC.xsl - http://www.loc.gov/standards/marcxml/xslt/MARC21slim2RDFDC.xsl
[2] modsrdf.xsl - http://www.loc.gov/standards/mods/modsrdf/xsl-files/modsrdf.xsl
[3] mods2rdf.xslt - http://infomotions.com/tmp/mods2rdf.xslt
[4] MARC/MODS RDFizer - http://simile.mit.edu/wiki/MARC/MODS_RDFizer
[5] ead2rdf.xsl - http://data.archiveshub.ac.uk/xslt/ead2rdf.xsl

— 
Eric Lease Morgan