The analogy is not just a coincidence: as it was said before, the activity of pre-
frontal cortex as a control mechanism together with the MD network cannot
be understood without appealing to the feedback it receives from subcortical
structures. The fact that the prefrontal cortex and striatum are connected in a
loop would indicate a mutual modulation that has the capacity to flexibly and
emergently alter the flow of processing in the brain without the need for a
prime mover or homunculus.
The process of generating a procedure would be based on repeated exposure
to the same activation patterns throughout the course of various trials, which
would generate a chunk. In experiments by Fujii and Grayb iel (2003, 2005) it
is sho wn that after training monkeys to make complex sequences of saccadic
eye movements, some neurons in dorsolateral prefrontal cortex and striatum
start to show increased activation at the boundaries of those sequences, signaling
that a chunk has been completed. The process of forming a chunk is correlated
with decreased activation in the prefrontal cortex, whose sustainment capabilities
are not needed anymore to connect the dots that would constitute an action
sequence. The process of acquiring a chunk can therefore be understood as a
transfer between the declarative and the procedural memory components, or as
a gradual decrease in cognitive control and MD network activation. Buschman
and Mi ller (2014) propose a model in which the development of procedures
would reduce cognitive load by decreasing the amount of time in which the
MD network needs to be sustained to correctly solve cognitive tasks. Within
this model, fast associations of a stimulus-response nature would be handled by
the faster, more primitive basal ganglia, while the development of abstract cat-
egories and generalizations would require the slower processing of the prefrontal
cortex, with repeated exposures that guarantee a link between the different
associations.
Chunking is a widespread phenomenon in animals, which show differing
degrees of flexibility in the process of retrieving and modifying chunks. Usually,
once a behavior has been proceduralized, the learning process is considered as
finished and, in many animals, behavior crystallizes. Crystallization has been
thoroughly studied in the domain of vocal learning birds. Some species of birds
have males that learn their courtship and territorial songs by imitating a tutor’s
song during a period of babbling known as the subsong stage (Jarvis 2007).
Du ring this period, the song patterns that are produced are very variable, with
a degree of accuracy that seems to go back and forth until it starts to stabilize
and the bird acquires its song. In some species like the zebra finch, this song
gets fixed for life and is considered to be crystallized. However, in other species,
such as the canary, crystallized songs are renewed for each breeding season.
Finally, other species of birds such as the lyrebird are open-ended learners, as
they can use their vocal components to acquire new vocalizations throughout
life (Petkov and Jarvi s 2012). These distinctions seem to depend on species-
specific striatal capacities to regulate plasticity. In humans, unchangeable or
crystallized action patterns are rare, save perhaps for cases of mental illness like
OC-spectrum disorders (Graybiel 2008), and the acquisition of skilled behavior
Language and working memory 113