Learning objects are largely misunderstood, in part due to the complex and related theories used to describe them. These learning object theories have some common points: learning objects are composed of some media (usually, but not always digital) and some metadata used to describe and locate the media.

Learning Object = Media + Metadata

This should be rather straightforward, except that in implementation, there is often (almost always) a strong bias toward the metadata side of the equation.

Learning objects are often described using one of several metaphors, ranging from "learning objects as lego bricks" to "learning objects as atoms or molecules". In practice, however, the metaphor used to describe learning objects is largely irrelevant. We wind up using software designed with "learning objects as metadata" as the primary metaphor. The various contexts and uses of resources become lost in noise, and many practicing teachers have given up on "learning objects" since they aren't seen as practical, meaningful, or even useful.

Instead of focusing on the metaphor, we first need to come to an effective means of modelling learning objects. Are they simply assets, such as images? Are they something more than that? What is the threshold that separates mere assets from "learning objects?" Confusion may be caused by a misunderstanding (or at least an underestimation) of the fluid nature of learning, as the threshold will vary by individual, and by context.

What is needed is a way of describing learning objects which will enable this variation of usefulness in response to context. A monotlithic IMS LOM model of learning objects makes this difficult.

One example, which I have used before to describe the difficulties of monolithic metadata descriptions, is an image of a seashell.

Sample Seashell Image

This image will mean different things to different people, each with their own context. A layman may see a sea urchin shell. A biologist may see the skeletal remain (test) of a Diadema antillarum. A chemist may see a rich source of calcium. An architect may see a robust structural design. Each of these contexts are valid, and complementary. And difficult to describe in a single IMS LOM record, if for no other reason than the contexts all come from different individuals who may not know each other, and certainly aren't available to collaborate on the creation of a single IMS LOM.

Currently, each individual would have to create their own complete IMS LOM, largely duplicating effort, and adding their own context into /lom/classification/keyword/langstring, or wherever they feel it fits best. This isn't very realistic, as it is hard enough to get a single LOM properly filled out to describe a single resource.

Current "Metadata-Centric" Model of Learning Objects

So, what is this "metadata-centric" model that seems so prevalent? It's a natural and logical result of building software to meet the various "eLearning" specifications, such as IMS LOM, CanCore, and Dublin Core. They make it easy to fall into this pattern of placing the metadata at the centre of the universe. It's also easy for developers to do this, as metadata is tangible. Building a database query is quite analagous to dealing with metadata - you have fields, default values, attributes, etc...

As such, there is nothing "wrong" with this model. In fact, it has served quite well, providing us with a concrete set of needs which can be addressed readily by writing code. This is what developers do best, after all.

This approach to learning objects results in a simple model, and as such, is relatively straightforward to implement. Each resource is represented in a database by a single metadata document, in a strict 1:1 relationship.

Metadata Centric Learning Object

Problems with Metadata-Centric Model

While this model is conceptually simple, and easy (although non-trivial) to implement, it enforces a strict top-down approach to learning objects. Each learning object has a single resource, and a single "owner" - the person who filled out the IMS LOM. Again, this makes it easier to implement, but limits functionality - collaboration becomes more difficult, and an authoratative approach to the metadata follows. It becomes less feasible for individuals other than the original metadata author to add context to a resource.

Since the original metadata author "owns" the resource, it becomes difficult to truly reuse a resource. It is easier to simply duplicate the resource and create a new (and separate, and unrelated) piece of metadata. While this is entirely possible, it takes away the real power of reusability - each reuse should add something to the resource, not separate it from previous uses. This approach would lead to multiple unrelated copies of identical resources, and reuse through recontextualization isn't much easier than starting from scratch.

Resource-Centric Model of Learning Objects

A resource-centric model takes the emphasis away from any individual piece of metadata, and shifts it back to the resource. At the same time, it returns power to the individual, allowing them to add personalized metadata to describe how a resource is appropriate to their own context. One way to think of this is as "metadata soup." Instead of atomic metadata documents, we end up with a fluid collection of personalized metadata, which can be viewed as an aggregate to fully describe a resource in whatever context is available. And, if a context isn't available, feel free to add your own, and rest assured that it will be absorbed into the metadata soup which the resource is floating in.

The various contexts, provided by any number of different individuals, are all available, and are all equally valid. It becomes possible to view a resource from the perspective of a concept map, visualizing the relationships between contexts, uses, and perhaps similar resources with their own contexts and uses.

It becomes relatively trivial, then, to relate to the seashell as any or all of an infinite number of contexts, described by an infinite number of indivdiuals. This changes things to a more grassroots approach, giving the individuals the power to use, reuse, and contextualize resources at will.

The equation changes, to emphasize the media or resource, rather than the metadata, and we wind up with something that can be much more than the sum of metadata and media.

Learning Object > media + sum(metadata)

Resource as the focus (media, image, video, ...)

Resource-Centric Learning Object

metadata falls into supporting role

Breaks the 1:1 relationship, can be 1:many

bottom-up approach - can easily just add more metadata related to the resource to contextualize the resource for your purpose

Multiple related metadata sets allow multiple simultaneous descriptions of a resource - IMS, DC, IMS CP, MPEG7, ... as well as multiple contexts for each resource

Multiple sources of metadata means anyone can provide metadata to describe any resource

Challenges with the Resource-Centric Model

Resolution of identifiers (GUIDs) to relate various metadata descriptions to resources

Use of URI for resource as physically resolvable GUID?


Although one goal of a resoure-centric model is to enable individuals to recontextualize resources, it will not solve the issues of the "reusability paradox" or the "myth of reusability". At best, it can provide a means to effectively manage quality resources, and in practice, may be used as a way to reduce workload on content producers, rather than making every individual a metadata publisher.