Evolution of Language: An Agent-Based Simulation
Mark Dras
Macquarie University
Language Technology Group
Tuesday 30 October at 11am
Abstract
Languages change continuously, in terms of their phonology,
morphology
and syntax: they can mutate in isolation, split off and diverge,
or be
affected by other languages. This happens because languages are
not
just cognitive phenomena but are conditioned by social interactions,
making them dynamical systems in the same way as biological and
physical phenomena. In all of these systems there are typical
patterns of change: change starts off slowly, accelerates, then
slows
again, and this takes place over several generations. There are
only
a few previous attempts to model language change, and these either
impose the form of the change onto the model or give empirically
inaccurate results.
We take a particular language phenomenon, vowel harmony in Turkic
languages, and model that using an agent-based simulation implemented
using a set of Swarm libraries in Objective C. In the simulation
the
agents are endowed with low-level preferences and linguistic and
social behaviors. We show that empirically-observed patterns of
change are an emergent property of the model, in that they arise
naturally, but without prior specification, from the interactions
of
the agents.
As part of this, at a broader level, we've become interested in
questions raised by the notion of agent-based simulation: What can
be
said about the methodology of developing agent-based models? How
do
they compare with mathematical models? And what can these simulations
"prove" when used for science? This talk will suggest
some answers
(or at least proto-answers) to these.
Short resume
Mark Dras got his PhD, in computer science, from Macquarie
University
in 1999. He has just finished a two-year postdoc at the Institute
for
Research in Cognitive Science at the University of Pennsylvania,
and
is now back at Macquarie. His research interests centre around models
of language transformations, and include formal and mathematical
properties of synchronous grammars, machine translation, and language
evolution.
Back to HAIL Home Page
|