answers and establish linking hypotheses between linguistics and working mem-
ory, we need first to find a neurocognitive model that can extract them from
the specificity of the multicomponent model and translate them into a more
neutral code. The next three subsections will try to further develop these con-
cepts by adapting them to what we know about how processing works at the
level of the brain.
1.2 Working memory as an emergent state
The multicomponent model of WM is a top-down approach, meaning that it
is a model that was designed to account for behavioral performance in a series
of experiments. Based on what we know about how the brain functions, it is
also possible to propose bottom-up approaches to WM. The bottom-up cogni-
tive model that will be introduced here, Baar sās (1988) Global Workspace theory,
was not originally designed to explain WM functions per se, but to provide a
functionalist, brain-based interpretation of the phenomenon of conscious aware-
ness. In spite of that, since both models deal with how processing functions,
recent cognitive and AI models based on the global workspace already incor-
porate a full account of WM functions (e.g. Frank lin et al. 2012), so it is worth
dedicating a few paragraphs to understand how the concepts that we have
extracted from the multicomponent model can start to approach the neural
substrate, and how the exercise of reconstructing them from the bottom up
can significantly change them.
According to Global Workspace theory, cognition and behavior result from
the ever-changing, complex interaction between localized processors that are in
charge of representing the world. These processors range from low-level visual
feature detectors to more complex systems that integrate them, producing
conceptual structures dedicated to specific domains of cognition, such as the
detection of objects, the animate vs. inanimate distinction or basic counting
mechanisms (Spelk e and Kinzler 2007). Faced with the myriad of stimuli that
constantly bombard us, these processors will activate in parallel, trying to match
an input with the specialized function that they can perform best, producing
cascades of activation that will necessarily yield multiple, conflicting interpreta-
tions of the world. An important question in cognitive science, known as the
frame problem, is how these localized processors get prioritized and integrated
into a coherent whole to create a mind that can navigate the environment in
an efficient, flexible, and appropriate manner. In short, how does a modular
mind become central cognition (Fodor 1987, Den nett 2006)?
Global workspace theory (Baars 1988, 2007; cf. also Dehae ne and Changeux
2011) describes the way in which this happens as an emergent property of the
architecture of the brain, which is designed so that localized processors can
communicate through a virtual space with limited bandwidth, known as the
global workspace. The global workspace is used by localized processors to form
various long-distance coalitions whenever their interpretations can complement
each other, getting stronger by isolating competing coalitions until these are
106 Gonzalo Castillo