In evaluating the role of the basal ganglia and cerebellum in skill learning, it is useful to compare these two prominent subcortical structures in terms of both anatomy and physiology. Although the neural circuits of each are unique, there are correspondences between the features of the cerebellum and basal ganglia that warrant a brief description. First, both form loop-like circuits in which inputs from the cortex are processed and then relayed back to the cortex via the thalamus. Second, both have inhibitory projections to their output targets: The internal capsule ofthe globus pallidus inhibits the thalamic nuclei and the Purkinje cells of the cerebellar cortex inhibit the cerebellar nuclei. Finally, both the cerebellum and the basal ganglia appear to use a divergent-convergent architecture in which input signals are distributed across a vast range of neural networks before recom-bining into a more compact, topographic organization. In the basal ganglia, this occurs as small cortical regions project to many loci within the striatum, which in turn project reconverging output to focal regions with the globus pallidus. In the cerebellum, inputs from mossy fibers are distributed to and reintegrated by thousands of Purkinje cells. One possible function of this design is to allow for inputs to form associations with information from many sources.
Given these shared principles, it is important to consider some differences between the two structures. First, unlike the cerebellum, the basal ganglia do not have any direct efferent or afferent connections to the spinal cord and comparatively few connections to the brain stem. In this manner, the basal ganglia are positioned to modulate the cortical selection and instantiation of actions, whereas the cerebellum is likely involved in both the planning and execution of movements.
Second, the two subcortical systems seem to use feedback in a differential manner. As described previously, the climbing fiber inputs to the cerebellum are generally considered to be the source of an error signal that is minimized as learning progresses. In contrast to the prevalence of error signals in theories of cerebellar learning, models of basal ganglia have emphasized reward systems based on the dopaminer-gic pathways. Thus, whereas the cerebellum is assumed to learn based on negative feedback from error signals, the basal ganglia appear to learn using positive feedback from reward signals.
In trying to understand how the computational roles of the two structures may be distinguished, theorists have emphasized the role of cerebellum in movements that require precise coordination and timing: Eye-blink conditioning, VOR, and multijoint movements all necessitate that motor commands that are exactly timed in relation to environmental events or changes in the positions of other body parts. In contrast, the basal ganglia are generally associated with learning tasks in which a particular action or series of actions must be performed in a novel context. For example, in the SRT task, the simple finger press responses are easily made and benefits are likely accrued when the correct one can be anticipated.
Along these lines, Daniel Willingham has argued that the basal ganglia perform motor learning computations that are distinct from those tapped by sensor-imotor integration tasks. Willingham evaluated performance of individuals with Huntington's disease across a variety of motor learning tasks and found that the learning impairments appear limited to a subset of the tasks, including rotary pursuit, the SRT task, and a tracking task in which the target followed a fixed pattern. The patients showed normal learning on mirror tracing and on a tracking task in which the target moved randomly. Noting that the Huntington's patients' deficits appeared to be restricted cases in which the movements were predictable, Willingham proposed that the basal ganglia are critical for learning sequences of open-loop responses.
In this sense, the basal ganglia can be thought of as operating on higher level representations that involve selecting a particular goal. In contrast, the cerebellum has been shown to be required when fine-tuning is required in the motor programs responsible for accomplishing a set goal. These generalizations suggest that the basal ganglia are critical for SRT learning, whereas the cerebellum is involved in learning tasks that require precise movements and the on-line coordination of effectors with sensory input. That is, the basal ganglia support "knowing what to do'' and the cerebellum underlies "knowing how to do'' motor tasks. Computational models have generally been consistent with this view. The successive inhibitory processing stages of the basal ganglia, from striatum to globus pallidus and then from pallidus to thalamus, have been shown to offer the unique feature of operating as a winner-take-all process, a mechanism for selecting the appropriate output given a particular context. In contrast, the inhibitory output from the cerebellar cortex is well suited for appropriately tuning the output from the cerebellar nuclei.
Features of this proposed distinction are likely to prove useful, but in some cases it is almost certainly an oversimplification. For instance, this framework cannot easily account for the fact that patients with cerebellar lesions are most impaired on SRT learning. We would expect that the patients could learn the pattern, but that they would have difficulty in producing the coordinated sequence of actions. Perhaps the cerebellum does not distinguish between error signals associated with poor movements and those associated with meeting the task requirements (e.g., pressing the correct button).
Contrary to earlier views of brain function, the functional domains of both the cerebellum and basal ganglia are no longer restricted to motor control and motor learning. Both structures project to various association areas of the cerebral cortex, suggesting a role in higher level cognition. Moreover, imaging studies consistently reveal metabolic changes in these areas that cannot be accounted for in terms of the motor requirements of the tasks, and neuropsycholo-gical studies have found that lesions in both the basal ganglia and cerebellum can disrupt performance on a wide range of cognitive tasks. The manner in which these subcortical structures contribute to motor skill acquisition may prove useful in understanding how they influence learning more generally.
Was this article helpful?