Language, Games and Evolution

Gerhard Jäger and Robert van Rooij

[revised and updated June 19, 2004 ]


Summary: With this course we pursue three goals:
  • to give in introduction into basics of Game Theory for linguists and philosophers of language
  • to review existing application of GT to linguistic problems, especially in the realm of pragmatics, and
  • to give an overview over the evolutionary reinterpretation of GT ("Evolutionary Game Theory") and its potential applications in typology and pragmatics.

Game Theory: We will start with a brief introduction into the basic concepts of GT. It will explain the notions of players, strategies and utilities. We will then cover the notions of a mixed strategy, strategy domination, and Nash equilibria. The underlying assumptions of classical GT are sometimes debatable for linguistic applications. The notion of bounded rationality was designed to solve related problems in economics. We will review equilibrium concepts based on this view, especially Pareto efficiency and risk dominance.

Previous Work on GT and Pragmatics: David Lewis introduced signaling games as a model of communication. It serves to explain why and how conventional meanings can be associated with natural language expressions. Unfortunately enough, this framework has largely been ignored in linguistics and philosophy of language (but see van Rooij 2002 for an application to a pragmatic phenomenon). It is of central importance for the topic of the course as a unifying framework and will be discussed in some detail. Especially the distinction of pooling vs. separating equilibria is of immediate relevance. Much work in formal pragmatics rests on the assumption that sender and receiver in communication have to choose between several productive or interpretive options which have differential utility for the purposes of communication. The similarity to the game theoretic setup has been noticed at various places, even though no generally accepted framework has emerged so far. We will give an application-oriented overview over the most important GT approaches to pragmatics.

Evolutionary Game Theory: This is a branch of GT that originates from biology but is applicable to social sciences as well. In this realm, it is based on the notion of adaptive learning, which introduces a dynamic component. We will discuss various learning concepts that are relevant for linguistic applications. The classical equilibrium concepts are replaced by various notions of evolutionary stability in the dynamic setting. The concepts of stochastic stability and stability under correlation are especially relevant here. We will present applications of these notions to pragmatics (the emergence of Horn strategies in signalling games) and typology (evolutionary stability of argument marking systems). Last but not least, we will relate the EGT approach to other applications of evolutionary models to linguistic problems. Most noteworthy here are the mathematical models of Martin Nowak and his co-workers, and the computational simulations by Jim Hurford, Simon Kirby and their group in Edinburgh.


Downloadable Reader:


Here are some experimental programs for simulating evolutionary games.