Philosopher and Cognitive Scientist; Professor of Logic and Metaphysics, University of Edinburgh, UK; Author: Surfing Uncertainty: Prediction, Action, and the Embodied Mind.

The Iterated Learning Explanation of Structured Language (Language As An Adaptive System: Humans As Hosts)

My choice is the iterated learning explanation of structured language. This is one of those beautiful explanations that turns things on their head exposing their workings and origins in a brand new way. It suggests powerful alternatives to views that depict human brains as heavily adapted to the learning of human-like languages. Instead, it depicts human-like languages as heavily adapted to the shape of the learning devices housed in human brains.

The core idea is that language is itself a kind of adaptive system that alters its forms and structures so as to become increasingly easily learnable by the host agents (us). This general idea first appears (as far as I am aware) in Terry Deacon's 1997 book, The Symnolic Species. It has been pursued in depth by computationally-minded linguists such as Simon Kirby, Morten Christiansen and others. Much of that work involved computer simulations, but in 2008 Kirby et al published a paper in Proceedings of the National Academy of Sciences augmenting those proofs in principle with a laboratory demonstration using human subjects.

In these experiments subjects were taught a simple artificial language made up of string/meaning pairs, and then tested on that language. Some of the test items were meanings that had featured in training, while others were new meanings. Then comes the trick. A 'new generation' of subjects is then trained using not the original items but the data from the previous generation. The language is thus forced through a kind of generational bottleneck, such that one generation's choices (including errors and alterations) provide the next generation's data. What the experimenters robustly found (echoing the earlier simulation results) was that languages subject to this kind of cumulative cultural evolution became increasingly easy to learn, exhibiting growing regularities of construction and inflection. This is because the languages alter and morph in ways that are a better and better fit with the basic biases of the subjects (the hosts). In other words, the languages adapt to become easier to learn by the kinds of agent that are there to learn them. They do this because learners expectations and biases affect both how well they recall actual training items, and how they behave when presented with novel ones.

Language thus behaves a bit like an organism adapting to an environmental niche.

We are that niche.