(Un)Masking Selection: from innate to learned signalling and beyond

Simon Kirby & Terrence Deacon

University of Edinburgh & University of California Berkeley

simon@ling.ed.ac.uk

One of the unusual features of human language is the degree to which it
involves learning.  We acquire an enormous amount of information about
our native language through early learning, and invest a lot of
resources to its acquisition.  In return we get a communication system
that is uniquely supple.  When combined with symbolic reference, a
learned system can be rapidly adapted to new uses as our communicative
needs change.  In addition, learned systems of reference give rise to
cultural evolution.  We have argued elsewhere that this in turn is
responsible for the fundamental components of syntactic structure, the
other hallmark of human language. 

We can think of the emergence of learning as a major transition in the
evolutionary history of human language - a necessary transition from a
largely innate simple signalling system to one that enables subsequent
evolutionary development.  We are left with the question: What caused
this transition?

Recent studies by Kazuo Okanoya of a domesticated species of finch
provide us with a model for the emergence of learned behaviour on a very
rapid evolutionary timescale.  The finch has been bred in captivity for
250 years and has a song that is supported to a large degree by early
learning.  Surprisingly, the feral progenitor of the finch has a far
less flexible, more innately pre-specified song.  Learning has evolved
despite only a short period of domestication.  What makes this really
remarkable is that the bird was bred for plumage, not song. 

In this paper we present a computational model inspired by the finch
data showing how learning can replace innate coding of behaviour in the
absence of direct selection.  We argue that learned behaviour actually
emerges when selection is "masked" - in other words, when a feature
previously under selection pressure is exposed to genetic drift. 

These results suggest that the emergence of learning in the evolutionary
history of human language resulted from an alteration in the fitness
landscape that reduced selection pressure on an our innate signalling
system.  As features of the genotype are masked from selection and
relinquish control, other previously ineffectual factors can come to
influence behaviour.  In the finch example, these unmasked influences
probably included auditory experience and a variety of
social/environmental biases.  Thus, de-differentiation can be the first
step toward more distributed and flexible behaviour. 

We argue that these twin processes of masking and unmasking may have had
a critical role in the evolution of our unique biological capacity for
language acquisition.