The Reusability Paradox

In which it is demonstrated that the automated assembly of
certain types of learning objects is not possible

The Reusability, Collaboration, and Learning Troupe at Utah State University



Purpose
Like many other fields, the so-called "learning object" community has frequently used metaphors to communicate with those outside it. In order to make them easier to grasp, learning objects and their behavior have been likened to LEGOs, Lincoln Logs, and a number of other children’s construction pastimes. These analogies continue to serve their intended purpose of giving those new to the field an easy way of understanding what learning objects theorists are trying to do: create small pieces of instruction (LEGOs) that can be assembled (stacked together) into some larger learning-facilitating structure (castle or spaceship). Unfortunately the metaphor seems to have taken on a life of its own. Instead of serving as a quick and dirty introduction to an area of work, this overly simplistic way of talking seems to have become the method of expression of choice for those working at the very edge of Instructional Technology – even when speaking to each other. This point was driven home recently at a conference of a professional educational technology organization, where the LEGO metaphor was used in every presentation on learning objects, and even those on related topics such as metadata.

The problem with this trend is manifest in the degree to which the LEGO metaphor confines and controls the way people think about learning objects. Wiley (2000) discusses a number of problems with the LEGO metaphor and recommends an alternative. Here we consider only one of the difficulties previously identified with the LEGO metaphor. 

  • Any LEGO block is combinable with any other LEGO block
The implicit assumption propagated by the metaphor is that any learning object should be combinable with any other learning object. Because learning objects so designed are technically interoperable, the reasoning goes, computers can perform the labor-intensive work of combination. This ambition is typified in the IEEE’s Learning Technology Standards Committee’s Learning Objects Metadata Working Group 1997 purpose statement, which includes the following point. 
To enable computer agents to automatically and dynamically compose personalized lessons for an individual learner (LOM, 2001).
While we would not put the instructionally meaningful assembly of any two learning objects outside the realm of human capability, our purpose in this paper is to demonstrate that the instructional use or “combination” of certain types of learning objects (“types” here being variations of the objects’ grain sizes) cannot be automated. This single revelation will have a variety of meaningful implications.
 

How to Use This Article
In the following discussion learning objects of two types are assumed to exist: “small” and “large.” In practice, the designators “small” and “large” represent ends of a continuum on which all learning objects may be measured. While more types of learning objects exist than “small” and “large,” the differences between these two remain when additional types of objects are admitted.

It is recommended that the reader skip the list of definitions, read the description of the example learning object, and get right to the argument. The definitions may be referred to at any point in the argument through links provided inline.
 

Definitions

  1. learning object - a digital resource that can be reused to facilitate learning (back)
  2. small object - a single learning object uncombined with any other (e.g., a single JPEG) (back)
  3. large object - many learning objects combined to make a bigger, aggregate learning object (e.g., a webpage including a text file, several images, and an animation) (back)
  4. a learning object's internal context – the elements (e.g., other learning objects) juxtaposed (e.g., spatially or temporally) within a learning object (back)
  5. a learning object's external context – the elements (e.g., other learning objects) against which a learning object is juxtaposed (e.g., spatially or temporally) to facilitate learning (back)
  6. instructional use of a learning object – the automated or by-hand placing of a learning object within an external context (back)
  7. instructional fit – the degree to which the instructional use of  a learning object, as opposed to other variables, facilitates learning (e.g., the Pythagorean theorem would not fit well in a second grade math lesson) (back)
  8. learning object user – a system or human that makes instructional use of a learning object (back)
  9. metadata – descriptive information about properties of a learning object (back)
  10. learning object discovery – the process by which a user locates a candidate (for use) learning object (back)
  11. objective metadata – properties of a learning object to which meaningfully falsifiable values can be assigned, such as the learning object’s author, file size, or mime type (back)
  12. subjective metadata – properties of a learning object to which meaningfully falsifiable values cannot be assigned, such as the learning object’s meaning or usefulness (back)
  13. instructional architecture – a known configuration of external contexts (e.g., instructional templates which learning objects may be “plugged into” in order to facilitate learning) (back)
Example Learning Object
As an example, we will consider a webpage containing an art history lesson composed of an image of the Mona Lisa, an image of Da Vinci, text describing the history of Da Vinci and the Mona Lisa, and an animation of Da Vinci's face being overlaid on the Mona Lisa (see Figure 1). The webpage, complete with graphics, is an example of a large object. An individual picture, such as the image of the Mona Lisa, is an example of a small object.
 

Argument
Proposition 1.1:  A learning object has no external context independent of its instructional use.
Rationale: External context has been defined as the juxtaposition of a learning object against other elements (e.g., other learning objects). When an object is not in use (i.e., when the object alone, as it exists in a digital library) there is no juxtaposition, and therefore no external context.

Proposition 1.2: The number of external contexts in which a learning object will instructionally fit is a function of the internal context of the learning object.
Rationale: The example case learning object (an art history website, which is a large object) is usable in an art history curriculum (and perhaps in some meta-domains such as website design). This is because the component learning objects have been instructionally used specifically to facilitate learning in (i.e., to fit into) the domain of art history. A component learning object, such as the image of the Mona Lisa, fits in these and additional external contexts, because the specificity of the art history domain is in its external context, and is solely a function of its instructional use. Independent of that use, the learning object will fit units on popular culture, attitude, or in the creation of a collage.

Proposition 1.3: A large object has a greater internal context than a small object.
Rationale: Two or more small objects can be contained in a large object. Because the internal context of the large object consists of the internal contexts of its components, the large object will have a greater internal context than any of its components.

Proposition 1.4: Large objects fit into fewer external contexts than small objects.
Rationale: Follows from Propositions 1.2 and 1.3.

Proposition 1.5: Metadata facilitates the discovery of learning objects.
Corollary 1.5.1: Metadata facilitates the instructional use of learning objects.
Rationale: Because many learning objects are non-textual, they cannot be discovered via full-text searching. metadata provide a way for these learning objects to be discovered or located. A learning object cannot be used unless it is known to the user.

Proposition 1.6: Metadata about the internal context of large objects is more valuable to users of a learning object than metadata about the learning object's previous external contexts.
Rationale: A large object has an internal context sufficient to restrict its use to a closed set of learning (i.e., external) contexts (Proposition 1.4). Before a learning object can be used instructionally the possible externals contexts of use must be identified, and a decision must be made regarding the instructional fit of a learning object into the target external context. Fit can only be assessed by examining the internal context of the learning object and comparing it to the target external context, making metadata regarding the internal context of the learning object necessary to its use (assuming that users will not examine every learning object individually and will rely on metadata to support learning object discovery).

Proposition 1.7: Metadata about the external context of small objects is more valuable to users of a learning object than metadata about the learning object's internal context.
Rationale: Small objects are by definition uncombined, single elements. While small objects exhibit some juxtaposition of internal elements (e.g., the foreground and background in a photograph), this internal context is much less significant than that of a large object, meaning that the possible external contexts of use of a small object are significantly greater in number than those of a large object. Since the internal context of a small object does not eliminate it from use in many external contexts (as the large object's internal context does), metadata regarding the internal context of a small object provides less support to users making use decisions regarding the small object. However, examples of the manner in which others users have used the small object may provide valuable use data that supports small object use decisions by learning object users.

Proposition 1.8: The potential for instructional use of different types of learning objects will be maximized by different types of metadata.
Rationale: Follows from Propositions 1.6 and 1.7.

Proposition 1.9: The value of objective metadata in facilitating learning object discovery is stable across learning object types, be they small or large.
Corollary 1.9.1: A stable set of objective metadata should be captured for each learning object.
Rationale: Proposition 1.8 states that different types of metadata must be used to maximize the potential for use of different types of learning objects. Propositions 1.6 and 1.7 demonstrated that the specific metadata needed to facilitate discovery (and therefore instructional use, Corollary 1.5.1) relate to the internal and external contexts of the learning object. Because the interpretation of context is a subjective matter, the differences in necessary metadata are differences in necessary subjective metadata, meaning that the value of objective metadata is the same for all learning object types.

Proposition 1.10: Subjective metadata for small objects should focus on capturing the external contexts of use of the small object.
Rationale: Follows from Propositions 1.6, 1.7, and 1.8.

Proposition 1.11: Subjective metadata for large objects should focus on capturing the internal context of the large object.
Rationale: Follows from Propositions 1.6, 1.7, and 1.8.

Proposition 1.12: The instructional use of large objects can be automated.
Corollary 1.12.1: Large objects are best suited to use by automated users (e.g., computer systems).
Rationale: The internal context of a large object significantly limits the external contexts into which it will instructionally fit (Proposition 1.4). This limitation of possible external contexts of use can be combined with an instructional architecture (i.e., a known configuration of external contexts) to facilitate the automation of the placing of large objects into external contexts in which they will fit. (See Wiley (1999) for a description of a simple instructional architecture which concretely demonstrates the substance of this Proposition).

Proposition 1.13: The instructional use of small objects cannot be automated.
Corollary 1.13.1: Small objects are best suited to use by human users.
Rationale: The internal context of a small object constrains the number of external contexts into which it could fit much less than the internal context of a large object does (Proposition 1.4). This necessitates the use of additional decision support data to select one of several potentially fitting learning objects, that is, it forces instructional fit decisions to rely on data other than that expressed in metadata. Deprived of decision support data, an automated system is incapable of reliably using small objects.

Proposition 1.14: Different types of learning objects are best suited to instructional use by different types of learning object users.
Rationale: Follows from Propositions 1.12 and 1.13.
 

Discussion
The purpose of learning objects and their reality seem to be at odds with one another. On the one hand, the smaller designers create their learning objects, the more reusable those objects will be. On the other hand, the smaller learning objects are, the more likely it is that only humans will be able to assemble them into meaningful instruction. From the traditional instruction point of view, the higher-level reusability of small objects does not scale well to large numbers of students (i.e., it requires teachers or instructional designers to intervene), meaning that the supposed economic advantage of reusable learning objects has evaporated.

This result is quite significant. It means, for example, that corporations and others who wish to create systems of automated learning object assembly must use large objects, possibly foregoing their previous assumptions about the size of learning objects with which they can work. It also means that the large objects whose assembly can be automated are quite un-reusable, at least compared to smaller objects.

Either way, it would seem that there are only two options: throw out the learning objects notion altogether, or encourage the development and use of only large objects, settling for their limited reusability. There is, however, another option.

The only quantity certain to scale with large numbers of students is the number of students. If a more constructivist view of learning is admitted, small, highly reusable objects can be brought to bear on instructional problems without suffering from scalability issues. This could be accomplished by creating learning environments in which learners interact directly with the small objects, manipulating and combining them to construct meaning for themselves. Computer Supported Intentional Learning Environments (Scardamalia, et al., 1989), Open-ended Learning Environments (Hannafin, et al, 1999), and other computer-based constructivist environments provide models of ways in which these small objects might be used by learners.
 

Conclusion
As we have demonstrated, the method learning object proponents have evangelized as facilitating reusability of instructional resources may in fact make them more expensive to use than traditional resources. We have demonstrated that the automated combination of certain types of learning objects can in fact be automated. However, it would appear that the least desirable relationship possible exists between the potential for learning object reuse and the ease with which that reuse can be automated: the more reusable a learning object is, the harder its use is to automate.  Identically, the less reusable a learning object is, the easier its use is to automate. This discovery is depressing, indeed.

However, as is often the case, this disappointment has pointed toward something we may have never considered otherwise: the student-directed constructivist use of small learning objects.
 

References

  • Hannafin, M.J., Land, S., & Oliver, K. (1999). Open learning environments: Foundations and models. In C. Reigeluth (Ed.), Instructional Design Theories And Models (Vol. II). Mahway, NJ: Erlbaum.
  • LOM. (2001). Draft standard for learning object metadata. Retrieved April 2, 2001, from the World Wide Web:  http://ltsc.ieee.org/wg12/LOM_WD6_without_tracking.doc
  • Scardamalia, M., Bereiter, C., McLean, R.S., Swallow, J., & Woodruff, E. (1989). Computer supported intentional learning environments. Journal of Educational Computing Research, 5, 51-68.


Comments to David Wiley.