Memolab has been built to illustrate the qualities of a good learning environment. The pedagogical domain of MEMOLAB is the acquisition of the methodology in experimental psychology, i.e. experimental design.
MEMOLAB targets the acquisition of the methodology of experimental design in psychology. Nowadays, this methodology is taught in most universities by presenting methodological principles and by discussing experiments from the literature. Students have no real design experience until they have to create an experience with real subjects for their own thesis. At that stage, any mistake in the experimental plan becomes very costly. This strategy for training psychologists could be compared to asking future pilots to fly real aircraft immediately after theory courses, skipping the flight simulator stage. MEMOLAB is an artificial lab where future psychologists can experience (some aspects of) experimental design, without facing the risk of spoiling their subjects. In Memolab, the experimental subject domain is the study of human memory. Basically, the learner's activities form a cycle: design an experiment, run it, and analyze the output data. Here is an example of an experiment that the learner can design: two groups of subjects will study a list of 20 words, and then later on try to remember these words. Between the encoding and the recall, the first group will wait during 10 seconds, while the second group will wait for 40 seconds. The comparison of their recall performance indicates the effect of the delay length. This experiment is illustrated by Figure 1. It illustrates how the learner designs an experiment within the first lab (we will see later that MEMOLAB includes several labs). An experiment is constructed by positioning a set of events on a workbench. The vertical axis represents the time dimension. The activities of a particular group of subjects are vertically aligned: In our example, they first encode some material, then there is some delay and then some attempt to recall what they studied before. When an experiment has been fully designed, the learner may ask to simulate the results. The working details of the simulation are described in Appendix I: "The Simulation" (page 21).
Figure 1: The learner builds an experiment (at level 1) by assembling `events' on the workbench. An event defines the task (read, listen, recall, wait,...) of a group of subjects using some material (e.g. a list of 12 words). Panes on both sides of the workbench provide libraries where the learner builds and select the components that he assembles on the workbench.
Any psychological or pedagogical theory investigates learning from a specific viewpoint. For instance, some of them pay attention to the effect of practice, but neglect the importance of understanding. Some concentrate on discovery, and forget about imitation. However, when one observes an actual learning process, it becomes clear that learning does not result from a single activity but from the integration of multiple activities, based on multiple sources of information. A learning environment must takes this complexity into account by offering a wide range of learning resources that will accommodate the variety of individual learning styles and learning needs. Therefore MEMOLAB includes several tools:
The richness of a learning environment is not only a function of the number of tools but of the extent to which those tools are integrated. We made this integration possible by using a homogeneous object-oriented internal representation scheme. Let us illustrate this with the simulation. The results of an experiment are computed by analogy with those of similar experiments (see Appendix I: "The Simulation" (page 21)). MEMOLAB explains how the results have been computed referring to these experiments (`Simulation Trace'). As illustrated by Figure 3, the learner may double-click on one of these references and jump directly into the hypertext node where the referenced experiment is described. The hypertext is also connected with the rule-based agents (see section 4.4. on page 11 and section 5. on page 15). Most current research on hypertext systems has so far been concerned with the features of the hypertext itself. The originality of our approach is to have considered the place of the hypertext within a complex system, to situate the information search within a context that makes this information useful. The architecture of our hypertext is described in Appendix II: "The Integrated Hypertext" (page 26).
- The make-material tool
A difficult aspect in designing an experiment is to select carefully the items that the subjects will have to remember. A list of words has several features that the novice experimenter is not aware of, but which affect memory performance: the length of words (number of syllables), the frequency of words in our everyday language, the phonological or semantic homogeneity of a list of words. Figure 2 shows the "Analysis" tool of the make-material window.
Figure 2: The Analysis of chosen items in the make-material tool
- The hypertext `Methodology'
The very idea of a learning environment is that the acquisition of procedural knowledge requires active problem solving (by contrast to questions-based courseware that focuses on declarative knowledge). However, the weakness of microworld designers has been to underestimate the role of domain specific knowledge. MEMOLAB corrects this bias by giving access to declarative knowledge (theories, laws, facts,...) about the methodology of experimentation and theories on human memory. The hypertext structure enables the learner to find information at the level of granularity he is interested in. Our hypertext facility includes the features now common to most similar applications (trace of read nodes, web view, book view, footnotes,...). ETOILE's hypertext module is described in Appendix II: "The Integrated Hypertext" (page 26).
Figure 3: Integration of the hypertext and the simulation: the learner clicks on an experiment referred in the simulation trace and gets the hypertext node where this experiment is described.
- The hypertext `Encyclopaedia of Memory'
Designing an experiment on human memory requires both methodological skills and some knowledge specific to the domain of memory. For instance, learners must know the concept of "independent variable", but they must also know that `word frequency' is an interesting independent variable linked to `word length'. MEMOLAB includes an hypertext that provides theoretical and experimental data on human memory.
- The simulation & data analysis tool.
Research on the use of simulation in educational research (Roschelle,1988; Fredriksen & White, 1988) has shown that the simulation method itself is important. The role of a simulation is not simply to produce realistic results. The simulation process must be carried out in such a way that the learner can understand the results. The simulation built in Memolab produces a commented trace of the simulation process that enable the learner to understand how these results have been computed. Furthermore, a set of statistical results is computed and the learner can inspect the generated data. The simulation tool is described in Appendix I: "The Simulation" (page 21).
If the richness of a learning environment is a quality, its complexity may reduce learning. The designer must structure his learning environment in a way that the learner may benefit from this richness. The `structured environment' idea illustrates a fundamental difference of perspective between learning environments and traditional courseware: the meaning of `teaching' shifts from asking questions and providing feedback (although this is not excluded) towards structuring the environment in a way that provides optimal learning conditions.
The structure we have adopted is a sequence of increasingly complex microworlds (or problem solving situations, or laboratories for Memolab). This sequence leads to a "final" microworld that should be as close as possible to the real context in which the learner will have to apply the acquired skills. For instance, the final lab in Memolab uses the language that is commonly used by the community of experimenters in their exchanges (papers, conferences,.). However such a microworld is probably too complex to be mastered directly by a naive learner (as in the real world). It should be preceded by simpler microworlds. In the first one, the learner will have to solve simple problems, considering only some basic parameters. In the next one, he will manipulate more powerful operators and to use additional resources available in order to cope with more complex problems.
The architecture of MEMOLAB includes three microworlds, referred hereafter as `levels'. Each level concerns a different kind of experiment and uses a partially different interface:
By using a sequence of microworlds, one must address the issue of the learner's transition between microworlds. To smooth this transition, we designed the interface in such a way that these microworlds partially overlap. Each microworld uses a different set of commands for creating and for modifying the components of an experiment (events, sequences or plans). When an experiment has been completed at level N, the system translates this experiment into the representation that it would have at level N+1. The future representation scheme is therefore systematically associated with the actual scheme mastered by the learner. After some time, the learner is invited to express himself directly with the new language (N+1). This principle has been called the `language shift' (Dillenbourg, 1992) and is described more completely in Appendix IV: "The Design Metaphors" (page 33).
- At level 1, an experiment compares n groups of subjects along one dimension. The interface emphasizes the concrete, elementary and chronological aspects of an experiment.
- At level 2, the learner describes an experiment as a set of sequences. A sequence is a generic treatment that can be replicated several times with equivalent materials. The learner designs experiments that test the interaction between two parameters.
- At level 3, the interface reifies the logical structure of the experiment as a plan.
We used the image of the pyramid in order to emphasize the hierarchical relationship between operators that are used in successive microworlds. This notion of hierarchical integration is based on the neo-piagetian theory of Robbie Case (1985). In brief, this theory states that the qualitative improvement of children's cognitive structures is impaired by quantitative limitations in controlling elementary structures. To by-pass this limitation, children restructure their knowledge into higher order schemes that recombine patterns of previously existing sub-schemes. The relationship between Case's theory and MEMOLAB is described in Appendix IV: "The Design Metaphors" (page 33).
The pyramid metaphor illustrates our conception of system design, as we described it initially (see section 2. "Research Methodology" on page 3). In itself, the pyramid is not a very original image. It simply helps the designer to `view' his own system within a theoretical framework, for instance to think of the relationship between two corresponding system operators at different levels as the relationship between two schemas in Case's theory.
Learning does not only result from solving problems or using tools, but also from interacting with agents about these on-going activities. Even if the environment is rich and structured, some learners may need help while they solve problems. Some learners need to be pushed forward, otherwise they repeatedly use the same inefficient methods. An agent can of course be human, e.g. a tutor who monitors the learner's work or some peer learner sharing a desk. Regarding computational agents, we distinguish two types of agents:
This discrimination between two classes of agents has been essential in separating pedagogical knowledge from domain-specific knowledge and therefore to extract from MEMOLAB the components that constitute the toolbox. This architecture will be described in the next section.
- Agents that conduct domain specific interactions (i.e. help to build an experiment). These agents can be domain experts or a computerized collaborative learner (Dillenbourg & Self, 1992b). One can also have several experts that solve the same problem with different viewpoints (Moyse & Elsom-Cook, 1992). The interaction between the expert and the human learner concerns the problem to be solved.
- Agents that select and monitor learning activities or help the learner to manage his own learning process. In MEMOLAB, we discriminated two kinds of pedagogical agents, the tutors and the coach who is a kind of "tutor in chief". The various tutors have the same function but fulfill it differently according to their style.