Séminaire du 13/04/2017 à 10h30 Ansgar Endress, professor at the City University, London




An intermediate state between domain-generality and domain-specificity

Cognition might be characterized by “domain-general” mechanisms that apply
across domains, or by domain-specific mechanisms that are specific to
different domains. I draw on research on statistical learning and the
detection identity relations (e.g., the ability to notice that the last
two syllables in “dubaba” are identical) to suggest that some mechanisms
are available in many domains but not in many others, and that individual
abilities to use a given mechanism are relatively uncorrelated across
domains. As such, these mechanisms can be neither domain-specific (as
they exist in multiple domains), nor domain-general (as they don’t exist
in all domains, and do not correlate across domains).

I use these results to argue for the existence of intermediate states
between domain-generality and domain-specificity that I call
“domain-bound.” Domain-bound mechanisms might exist in multiple domains
(and thus are not domain-specific), but not in other domains (and thus are
not domain-general). I argue that similar phenomena might arise in two of
the most prototypical domain-general mechanisms: attention, and at least
in some species, inhibition, raising the question of whether fully
domain-general mechanisms exist, and whether those mechanisms that appear
to be domain-general might not be domain-bound.

I provide a straightforward evolutionary explanation of domain-bound
mechanisms. Just as molecular and morphological traits, cognitive
mechanisms might become duplicated over the course of evolution, resulting
in independent, localized copies in different domains. As a result, the
specificity of a domain might reflect which elemental computations are
available within this domain, and how they can be combined.