Inttranews Special Report: Andrew Wedel, assistant professor of linguistics, University of Arizona
One of the most fascinating - and
enduring - questions in linguistics is how language
gets its structure: is this structure genetically
determined, and innate, or does emerge over time under
the influence of physical and social constraints on
its use? The issue is not just an academic one: it
has ramifications in fields as seemingly wide apart
as primatology and artificial intelligence.
1. How did the concept of self-organisation come about?
Our understanding of the world has benefited greatly
from the reductionist strategy that models systems
as the simple sum of their parts, but it has long
been recognized that under some circumstances, the
parts may interact in complex ways that simple reductionism
cannot illuminate. For example, ripples a streambed
arise when flowing water causes sand grains to bump
against one another. The little piles of sand change
the way the water flows, which changes the way the
sand grains move, and so on. The interesting point
here is that one cannot easily pinpoint causes for
the specific ripples that develop, because causation
is diffusely distributed over the countless tiny interactions
between the elements of the system. Furthermore, because
of the feedback loops involved, imperceptibly tiny
differences in the starting state can have enormous
consequences in the development of the system, making
it essentially impossible to predict in any specific
way how the ripples will form, just by looking at
the sand grains at the start.
This is a very general set of requirements, and as
a consequence, it is likely that self-organisation
is not rare, but is rather a ubiquitous source of
structure in the universe. Language, in particular,
is rich in the kinds of interaction that support self-organisation,
so it makes sense to ask what aspects of language
structure may be emergent.
The concept in its modern form seems to have taken
shape over time from the 1940’s through the
1970’s. Concise reviews of the idea and its
history can be found under the headings of ‘self-organization’
and ‘emergence’ in Wikipedia. It is of
course still a topic of very intense research.
To start with, each of two computers are given a
set of randomized sound-meaning categories, i.e, words.
In each round, one of the computers says all of its
words to the other computer, which tries to match
the sound with one of its own word-categories. If
a computer recognizes a word, it is stored in memory
and may influence that computer’s future pronunciation
of that word. This results in a positive feedback
loop between perception and pronunciation. The computers
also have a tendency, like people, to mumble, which
in the context of the feedback loop puts pressure
on the words to become more and more alike over time.
It is my considered opinion that making such a black-and-white distinction obscures more than it reveals. Given that every feature of behavior relies in some way on a mapping over time from genotype to phenotype in the context of an environment, nothing is ever entirely one or the other (see Barbara Scholz, Nature 415: 739 for brief discussion of this issue). What I have shown is that a linguistic system of contrastive sounds does not have to be directly specified in the genes, but can rather be an emergent consequence of more basic, innate categorization processes interacting with communicative behavior.
6. What are the implications of your research?
The dominant, ‘universal grammar’ model
of linguistics holds that much of the structure of
language is defined by a highly specified ‘instinct’
for language. My work, and that of many others (see
the that of e.g., Juliette Blevins, Bart de Boer,
Joan Bybee, Jeffrey Elman, James Hurford, Simon Kirby,
Bjorn Lindblom, Pierre-Ives Oudeyer, Janet Pierrehumbert,
Luc Steels and many more), suggests instead that many
recurring features of language structure may emerge
from the interaction of more basic cognitive behaviors
with each other and with the external environment
over generations of language use and transmission.
This recognition has opened up a vast, exciting new
We have always had a tendency to characterize things we do not understand in mystical terms, so it makes sense that some people would like to see ‘self-organization’ in this way. However, it is important to remember that the same, garden-variety laws of cause and effect are still operative in self-organization-driven emergence of structure. What is different is our ability to clearly identify the causal chain leading from one state to another. Two factors can make this difficult in self-organizing systems: 1) the fact that causality is widely distributed among many different elements, as well as being distributed over time; 2) the fact that feedback loops can amplify initial differences that are too small for us to identify. These together can make the development of structure seem ‘magical’, but in fact it is not.
Please indicate below any reviews, references
or links you would like us to cite:
A more detailed explanation of the work described
in this interview can be found at: