How do we learn? This is a question with multiple, multifaceted answers. Indeed, the question underlies many TDLC research projects. And in cognitive science, it is often fruitful to focus not on the normal operation of a cognitive system, but rather its failure. Thus, one recently completed TDLC project centered on failure – specifically, why students fail to learn in educational settings.
It is a fact that students may grasp important concepts being taught, but their mastery of the information is tenuous and rapidly lost over time. So if memory retention could be improved, there would be significant aplication in such settings as foreign language and early math instruction. With computer models and experimental studies, PIs Hal Pashler (UC San Diego) and Mike Mozer (University of Colorado) tackled the practical issue of achieving durability of learning by concentrating on two factors that have long been known to influence memory retention:
- The benefits of testing during study – testing with feedback following mastery of material often results in better and more durable learning than re-studying the same material.
- The spacing effect – when students study material over multiple sessions that are appropriately distributed in time, long term memory retention is generally improved.
The relationship between spacing of study sessions and subsequent recall of material is complex and has a strong dependence on timing. So the researchers conducted studies to determine what sorts of materials would benefit from both testing and spacing and the quantitative nature of the benefits. To accomplish this, students were given two opportunities to study material, separated by a particular intersession interval (ISI) and then given a final test following a final retention interval (RI).
The researchers developed a computational model, the Multiscale Context Model (MCM), that characterizes the optimal ISI as a function of the student, the material, and the RI. Although many explanations have been proposed for the spacing effect, and although other computational models exist, MCM is the first that is truly predictive.
To make a prediction, MCM is given easy-to-collect data, such as the “forgetting function” that characterizes the fading of memory after a single study. MCM then predicts recall following multiple study sessions. Dr. Mozer explains, “The ISIs and RIs were explored on an educationally relevant time scale – with RIs as long as a year – and we found that we could empirically identify an optimal ISI, one not too short and not too long, for a given RI, as well as for a specific student and specific material to be mastered.” Numerical optimization procedures can also be employed to determine optimal spacing across multiple sessions, which would not be feasible to determine through human experimentation.
This study and its improved theoretical framework could well have practical impacts on classroom education into the future, and in fact has led to a new project, exploring together with Javier Movellan, how to achieve maximum retention for minimum study time.
by Carolan Gladden