Schmidt, R. A. (2003). Motor schema theory after 27 years: Reflections and implications for a new theory. Research Quality for Exercise and Sport, 74, 366-375.
Abstract: The schema theory for discrete motor skill learning (Schmidt, 1975), originally published in 1975, has generated considerable interest and received strong challenges over its lifetime. In this paper, I focus on the findings generated since 1975 that bear on the theory and highlight those that produce difficulties for it and will be motivators for differing theoretical viewpoints in the future. At the same time, I examine other lines of evidence that seem to bolster the original lines of thinking. Finally, I provide some suggestions for a much needed new generation of motor learning theory, pointing out particular features.from the schema theory that could be included and suggesting gaps and omissions that will need additional data and theorizing in future attempts.
What defines motor schema theory?
Generalized motor programs. Feedback from the periphery is too slow to account for many skilled actions, so central planning is necessary. Hence the generalized motor program. To account for variability, these central programs can be modified to take different rate and force values.
Two compartments of memory. We employ recall memory when we select the type of movement that will meet our goals in a given context. We employ recognition memory when we evaluate the movement we actually employed against our stored representation of that movement to assess of error.
The schema concept. Schemata are the relations between movement parameters and the outcomes of those movements. Motor learning, then, becomes the the development of schemata.
Solutions to storage and novelty problems. Schema allow for efficient storage (in the fact of, ostensibly, a limited cognitive warehouse) and for diversity of action.
What aspects of the theory should we keep for future theories?
Motor programs. When we look at EMG recordings of blocked vs. unblocked movement, much of the muscle action is identical. This sequencing does not seem like it could be explained by a dynamical systems approach–the pattern can’t just come together by muscle action, because the muscles aren’t really acting. A central plan seems more likely.
Generalized motor program. We see dissociations between a general movement pattern and at least one type of parameter (rate), supporting the contention that we learn a general program and can scale it. However, the initial contention that force was a second parameter that could simply be linearly scaled to make GMPs generalizable does not seem to be the case (e.g. walking under a load involves scaling of the extensor muscles’ force, but no such scaling is necessary for the flexor muscles).
Limitations of the theory
The motor schema concept does not easily address the motor equivalence problem nor does it address issues of hierarchical control (e.g. in writing where the wrist’s function is different from, and in response to, the elbow’s activity). The theory has had little to say about where generalized motor programs originate. GMPs may arise in situations (random practice with reduction in feedback), that encourage a reconstruction of action. Learning GMPs and learning parameters are dissociated because the latter is slowed under similar conditions.
Let’s say I notice a disparity between the movement I actually employ, and the motor plan that I selected from my repertoire for the situation.
Q1) If the movement failed to achieve the goal, do I…
- assume that the essence of the plan was correct, but that my execution was sloppy? or
- assume that I executed an action that fit within the limits of the plan, but that the plan didn’t allow for success in these circumstances.
With a) I might conclude that some parameters of my action were outside the bounds that demarcate a productive action? If so, we might expect the representation to narrow, to constrain future motor activity to a more restricted set of movement parameters.
With b), though, it might be that failure leads us to believe that the parameters in the original motor plan were, in fact, too narrow, and should be broadened to improve the plan’s productivity in the world. So we’d update the motor plan to be less restrictive than before.
Q2) Assuming these are both reasonable responses, how does the cognitive architecture choose in which way to update the schema? (This would seem to be an assimilation/accomodation question.)
Q3) Should we think of this in Bayesian terms, with the system holding on to motor plans that have a good history of surviving tests (unless there’s a preponderance of evidence to the contrary), and more likely to change motor plans with a shorter/poorer track record (even with relatively paltry evidence to the contrary)?