UCSF

Even well-learned motor skills must be performed differently depending on context (e.g., the skills involved in riding a mountain bike and road bike will generally differ). At the same time, learning must take advantage of the similarities between similar skills performed in different contexts by transferring, or generalizing, learning gained in one context to performance of similar skills in other contexts (e.g., it would be useful if learning first how to ride a road bike makes it easier to then learn how to ride a mountain bike). The flexibility of motor skills, therefore, depends on the ability for the nervous system to adaptively balance the generalization and specificity of learned modifications.

This adaptive balance is perhaps best illustrated in sequential motor skills, such as speech or dance (indeed “sequential” applies to a wide range of skills). These skills depend on the reuse of individual gestures in multiple sequential contexts (e.g., a single phoneme in different words). Yet optimal performance requires that a given gesture be modified appropriately depending on the sequence in which it occurs - this “coarticulation” is thought to enable the smooth and rapid production of skills. A diversity of experimental studies on humans have revealed that learned modification to a given gesture tends to generalize when the same gesture is used in other contexts; however, there is an additional capacity to learn highly context-specific modifications to individual gestures if such learning is the optimal way to respond in a given sensory environment. How this adaptive balance between generalization and specificity is implemented in neural mechanisms is largely unclear.

In this dissertation I report on experiments describing the neural mechanisms enabling generalization and specificity of vocal learning in birdsong. Bengalese finch song consists of variable sequences of discrete vocalizations called “syllables.” I first showed that at the behavioral level, Bengalese finches balance generalization and specificity of learned modifications to syllables in a manner that looks remarkably similar to the balance previously demonstrated for humans in similar motor adaptation experiments. In particular, when birds are instructed to modify a syllable in one sequential context, learning generalizes across contexts; however, if unique instruction is provided in different contexts, learning is highly-specific for each context, to an extent unexpected given the original propensity to generalize.

I then used localized inactivation of a cortical-basal ganglia circuit specialized for song to find that this balance between generalization and specificity reflects a hierarchical organization of neural substrates. Primary motor circuitry [the “motor pathway” (MP)] encodes a core syllable representation that contributes to generalization, while context- specific input from cortical-basal ganglia circuitry [the “anterior forebrain pathway” (AFP)] biases this representation to enable context-specific learning.

Finally, I performed neural recording experiments with the goal of further understanding how, in terms of changes to neural activity, this context-specific pitch bias is implemented. By analyzing the correlation between spiking activity in LMAN (the output of the AFP) and RA (the primary motor circuitry within the MP crucial for encoding syllable acoustic structure) during singing and learning, we found evidence suggesting the presence of a premotor signal conveyed from LMAN to RA, generated during learning, which acts to bias pitch through biasing of motor activity in RA. This biasing signal may be the outcome of the integration of signals encoding context, performance, and feedback in the AFP.

Taken together, these results (1) establish Bengalese finch song as a model system to study the flexibility of motor skill learning, (2) localize two key behavioral components of flexibility - generalization and specificity - to two different circuits, and (3) provides empirical support for the neural mechanisms by which these two circuits interact to adaptively balance generalization and specificity of learning.

Beyond birdsong, these findings may suggest broader principles regarding the neural mechanisms of flexibility in the learning and execution of motor skills, the contributions of cortical-basal ganglia circuitry to such flexibility, and the neural mechanisms involved in the control and adaptation of sequenced motor skills.