Here's how I tend to think about learning objects. Learning objects are the data contained within content packages. Pretty specific but it works for me and here's why. When I talk about e-learning materials I think of three classes of data (see figure).
Raw assets often though not always exist outside of a specific learning context, that is to say whatever context they do have is usually inherent in the asset. For example in an x-ray of the chest the context is the data contained within the x-ray itself, the pixels of the x-ray. It is the act of adding context by creating metadata (context for how the asset relates other data including a learning outcome for instance), and assembling raw assets into larger aggregations (e.g. adding a text narrative to an x-ray) that creates a meaningful piece of learning material. Ironically in this scheme it's raw assets that are the building blocks of learning materials rather than in previous analogies where learning objects have been likened to, for example, Lego bricks. Raw assets are given a learning context and assembled, or aggregated to use the common parlance into learning objects, the stuff learners use to learn from. Learning objects get moved around between learning environments as content packages.
At one extreme, most people are fairly comfortable with what constitutes a raw asset (itself an assembly of data, pixels in an image for example but that's taking too reductionist an approach) and at the other extreme content packages are well understood, not least because there are specifications that describe what they are and how they're structured. Between these two lie the learning objects. Sometimes it's difficult to conceptualize learning objects if you've never worked with tools that make it completely unambiguous what they are. There are tools available that understand this simple metaphor and function well to create one class of data from the other (for recommendations of what I think are the best tools get in touch). Perhaps my particular way of looking at this problem will be more familiar to people who are used to learning content management systems and are therefore comfortable with the progression of complexity from raw asset to content package. Others may find this approach too simplistic and instead prefer to think of the increase in contextualization from raw asset to context package. Either way it amounts to just about the same thing. But perhaps there's one of the underlying problems with the way people think about learning objects. There is sometimes an uneasy overlap in ways in which educationalists looks at e-learning materials and the way education technologists do. What works as an analogy for some doesn't always work for others.
But of this there can be no doubt, the materials or 'objects' that sit midway within the opposing continua of organisation and reusability are the learning objects. If we don't like the name then let's find another, but the space remains.