CULTURAL EVOLUTION RENDERS LINGUISTIC NATIVISM IMPLAUSIBLE BILL THOMPSON, KENNY SMITH, SIMON KIRBY Language Evolution & Computation Research Unit, The University of Edinburgh, 3 Charles St., Edinburgh, EH8 9AD, UK Bill@ling.ed.ac.uk, Kenny@ling.ed.ac.uk, Simon@ling.ed.ac.uk A central issue of debate in linguistics is the nativist hypothesis: the proposal that universally observed features of languages can be explained by a universally shared and biologically determined cognitive substrate that strongly and transpar- ently predicts these linguistic features (Chomsky, 1965). An evolutionary stance appears to provide support for this hypothesis (Pinker & Bloom, 1990). However, language is underpinned by social learning and cultural transmission alongside bi- ological evolution, and the interactions between these three systems are relatively poorly understood. We set out a general model of these interactions that leads to a surprising conclusion: cultural transmission renders the biological evolution of strong domain-specific innate constraints unlikely, but we nonetheless witness strong universal tendencies in the cultural systems of these simulated populations. In this talk, we present the results of a simple implementation of this model. The details of our approach are as follows. Our formalisation has a basis in three components: a rational framework for language acquisition, an iterated learning model of cultural transmission, and a representation of biological evolution. We assume language learners must iden- tify which of two arbitrarily distinct language types, t 0 and t 1 , is most likely to have generated the linguistic output of a learner from the previous generation: an observed set of b utterances, d. Each language type is associated with a diagnostic utterance class u, such that P (u y |t y )=1 - ǫ and P (u y |t x=y )= ǫ. Learners apply Bayes’ rule to determine P (t y |d) on the basis of this likelihood function and a prior expectation for t y given by a prior distribution over language types, which is characterised by the parameter p, where P (t 1 )= p and P (t 0 )=1 - p. A learner’s prior expectations are given by an inherited genome, which is simply a set of n genes, where each gene in a genome g n = {g 1 ,g 2 ,...,g n } is a member of the set G = {0, 1}. We treat the prior as a polygenic quantitative trait deter- mined by the additive effects of individual genes, so that p = 1 n ∑ n i=1 g i . Genetic transmission is subject to point mutation at rate µ, and under selection: each in- dividual reproduces with a probability proportional to the accuracy with which it