« Donations to Creative Commons and Wikimedia | Main | Ed Barker joins the Eduserv Foundation »

February 15, 2007

More ruminations on compoundness and complexity (and metadata)

This is a somewhat belated post that I started a few days ago, but put to one side while we concentrated on reading through the pile of Eduserv Research Grant proposals.

A couple of weeks ago I attended the workshop on describing complex objects that Andy referred to, and at which he gave a presentation (I was in the happy position of being able to sit in the back row and nod enthusiastically).

The programme featured presentations on three fairly widely used "packaging formats": MPEG-21 DIDL (by Frances Knudson of Los Alamos National Laboratory), METS (by Markus Enders of Goettingen State and University Library) and IMS Content Packaging (by Sheila Macneill of CETIS and University of Strathclyde).

The programme also included a presentation by Wilbert Kraan of CETIS on an IEEE LTSC project called RAMLET (Resource Aggregation Model for Learning, Education and Training), which has developed an ontology that can be used as the basis for mapping between instances of different "packaging formats".

Andy's presentation was the last of the five, and, leaving aside the DC-specific aspects, I thought probably his key point was that metadata is at the heart of what we call "content packaging" - metadata that describes certain specific characteristics of resources in order to allow applications to perform certain specific functions, certainly, but ultimately a key part of a "package" is some set of "statements" about some resources - and more specifically about relationships between resources.

So, when I create the content of a <structMap> element in a METS instance or of an <organisation> element in an IMS CP instance, I'm describing relationships between resources. (I'm consciously not commenting further on DIDL here as I'm much less familiar with the specification and after Frances' presentation I feel I need to go away and read up a bit more before making (probably quite misguided) comments about it!)  To take a very simple example, if I create a <structMap> like (rough outline for illustration purposes only - I don't promise that this is a complete/valid METS XML fragment!):

   <div label="My paper">
      <div label="My section 1">
          <fptr fileid="file001" />
      <div label="My section 2">
          <fptr fileid="file002" />
      <div label="My section 3">
          <fptr fileid="file003" />

or an IMS CP <organization> like (caveats as above!):

    <item identifier="item1">
       <title>My paper</title>
       <item identifier="item2" identifierref="file0001">
         <title>My section 1</title>
       <item identifier="item3" identifierref="file0002">
         <title>My section 2</title>
       <item identifier="item4" identifierref="file0003">
         <title>My section 3</title>

then in each case I'm "saying" that one resource (titled "My paper") is composed of a sequence of component resources titled "My section 1", "My section 2" and "My section 3". Elsewhere in the METS or IMS CP document I provide URIs of those resources. OK, it's a bit more complicated than that, but for the purposes of this argument, I'll stick to a simple case. And I could "say" exactly the same thing by constructing a Dublin Core metadata description set or an RDF graph, e.g. using the Turtle syntax for RDF:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix dc: <http://purl.org/dc/elements/1.1/> .
@prefix ex: <http://example.org/terms/> .

_:resource1 dc:title "My paper" ;
            ex:hasOrganisation [ a rdf:Seq ;
                                 rdf:_1 <http://example.org/docs/1> ;
                                 rdf:_2 <http://example.org/docs/2> ;
                                 rdf:_3 <http://example.org/docs/3> ] .

<http://example.org/docs/1> dc:title "My section 1" .
<http://example.org/docs/2> dc:title "My section 2" .
<http://example.org/docs/3> dc:title "My section 3" .

And I could probably do something similar using various other metadata specifications that allow me to describe relationships between things. I'm conscious that I'm over-simplifying somewhat, and METS and IMS CP provide other features that go beyond describing relationships, particularly in terms of describing how to embed representations of resources within an instance, but nevertheless I think Andy's point is a good one: a description of relationships between resources is a form of metadata. (Footnote: Sheila doesn't sound completely convinced!)

The other key point emerging from Andy's presentation, which he also highlighted in his earlier post, is that resources are of different types and relationships between resources are of different types, and he proposed a distinction between "compond objects" and "complex objects" on the basis of the different categories of relationship being described.

It seems to me that METS and IMS CP are fundamentally about describing what I think of as structural relationships - Andy's "compound object" case - : when I construct a METS structMap or an IMS CP organization, I'm "saying" resource W has components resources X, Y and Z. Further, I think METS and IMS CP support a specific subset of structural relationships i.e. they deal essentially with (ordered?) tree structures, where a "parent" resource has as components a sequence of "child" resources.

And (rather more tentatively!) I'd venture that the types of resources with which METS and IMS CP are concerned are, more or less, what the Web Architecture categorises (albeit somewhat vaguely!) as "information resources" i.e.

We do not limit the scope of what might be a resource. The term "resource" is used in a general sense for whatever might be identified by a URI. It is conventional on the hypertext Web to describe Web pages, images, product catalogs, etc. as "resources". The distinguishing characteristic of these resources is that all of their essential characteristics can be conveyed in a message. We identify this set as "information resources."

This document is an example of an information resource. It consists of words and punctuation symbols and graphics and other artifacts that can be encoded, with varying degrees of fidelity, into a sequence of bits. There is nothing about the essential information content of this document that cannot in principle be transfered in a message. In the case of this document, the message payload is the representation of this document.

But Andy went on to consider the example of the ePrints DC Application Profile, which is concerned with the description of resources of several different types, at least some of which - agents, for example - are not "information resources", and the description of various types of relationship which are not structural, e.g. relationships like is-created-by, is-published-by, and so on. While it is quite possible to describe relationships of these types between resources using statements in a Dublin Core metadata description set or using RDF, it seems to me I can not describe such relationships using a METS structMap or an IMS CP organization.

The point I'm trying to make here is not that I think Dublin Core is "better" than METS or IMS CP, but rather that, in order to make decisions about which specifications we use in this area, it's important to understand what each of the "packaging formats" allows us to "say" about "things in the world". From this viewpoint, the syntactic structure of, say, a METS XML instance is of less interest than what information such a document allows us to convey about resources and the relationships between them, i.e. what models underpin those formats - not models of the packaging instance itself (which I think is what is described by e.g. the IMS Content Packaging Information Model) but of the resources described or referred to within that instance. 

Such considerations will be important in the context of the OAI ORE initiative: for example, if an existing "packaging format" is used to serialise the ORE model, then it becomes critical that we understand fully any model inherent in that format - any built-in assumptions about the nature of the resources referenced or described, and the nature of any relationships between resources that are expressed within the format - , and that we ensure that any such serialisation accurately reflects the ORE model.


TrackBack URL for this entry:

Listed below are links to weblogs that reference More ruminations on compoundness and complexity (and metadata):


You're right that you can't describe relationships such as "published by" or "created by" using the structural map in METS, or at least, using *only* the structural map. In retrospect, the choice of the term 'structural map' was ill-advised on my part, since it carries an obvious implication that the structural map section of a METS document is the entirety of the structural metadata involved (relationship metadata?), when in fact, the entire METS document constitutes structural metadata, including the other major sections of METS, and the linking mechanisms between the structural map, the file section and the other forms of metadata in a METS document (descMD, adminMD, behaviorMD, etc.).

We intentionally avoided getting into the 'is created by', 'is published by' sort of relationships in METS, since others were already hard at work developing their own metadata specifications for that type of information (DC, IMS Metadata, MODS, etc.). Our only concern in METS was to allow those other forms of metadata to be linked to a resource(s), and to provide a functional characterization of the nature of the metadata (whether it was primarily descriptive, technical, rights or provenance metadata). IMS CP and MPEG-21 DIDL have the same basic operating model, but forego the functional characterization that METS employs. MPEG-21 DIDL doesn't care if you're linking a resource with a DC record or a LOM record, and will happily accommodate either or both, just like IMS-CP and METS.

So, yes, METS is fundamentally about structural relationships, but not solely about the types of structural relationships existing between the component content files of an object; the structural relationships between content and other forms of description are an equally vital part of METS, and it wouldn't be of any real use if it didn't support those links.

Frankly, I don't think any of the information packaging standards prevalent in the field provide particularly good mechanisms for serializing the others. MPEG-21 DIDL wasn't designed to handle the sort of 'onion layer' typification of files-containing-files-containing-bitstreams that METS added to address the management needs of digital preservationists, and METS wasn't designed to handle the sort of negotiations over content delivery that MPEG-21 DIDL supports. All of the standards are flexible enough so that someone clever could probably get any one of them to serialize the others, but it would be a kludge, and an ugly one at that. If people really want a serialization syntax that adequately addresses all of the packaging standards and their various different takes on the relationships that exist between content and metadata, well, that's why RDF was invented, and the work that RAMLET is doing in trying to develop an OWL ontology of packaging relationships is probably the best way to go. But everyone should be aware that just as crosswalks between descriptive standards are not always 'lossless', crosswalks between various structural standards via an RDF ontology aren't necessarily going to be lossless either. Different packaging languages express the needs of the communities that developed them, and the needs of the educational technology community are not exactly aligned with the digital library community, which are not aligned 100% with the needs of the commercial interests concerned with selling multimedia over networks.

I've long since given up hope that there's going to be a single, unified packaging standard. And it should be fairly obvious to all concerned at this point that whatever packaging standard a community settles on for their internal use will not be sole packaging standard that they must deal with. Educational systems that deal with IMS-CP are also going to need to access library content that will not be in IMS-CP, and content from a variety of commercial sources that is also highly unlikely to be in IMS-CP. No digital library in the world is going to have the luxury of dealing solely with METS, since we have to continually absorb content from a variety of sources that won't be using that packaging standard; in point of fact, one of the factors influencing METS' design was the assumption that it would *not* be the only packaging standard, or necessarily the most useful one in all cases, but that it would provide a 'least common denominator' form of packaging that could be applied to other, more sophisticated, application-specific forms of structural metadata (e.g., TEI, MXF, etc.).

For OAI ORE, I think the implications of all this are that they can take one of two approaches in their approach to modeling/serializing. They can employ the approach we took in METS, aiming for a truly least common denominator, highly abstract packaging standard, accepting that it will necessarily serve as a wrapper around more complex packaging that serves the needs of particular communities, or they can try to develop a complete ontology of structural relationships between content and metadata that subsumes all existing packaging standards and rely on RDF to encode it. The use cases they settle on I think will fundamentally decide which approach they take, so I think they need to be very careful about being inclusive in that process.

The comments to this entry are closed.



eFoundations is powered by TypePad