The Adaptive Mechanisms for Language Henry Brighton & Simon Kirby University of Edinburgh henryb@ling.ed.ac.uk In addition to uncovering the origins of human language, a central goal for evolutionary linguistics is to explain why language has the structure it does. The defining characteristic of evolutionary explanations is that they appeal to the mechanisms underlying the dynamics of the feature to be explained. In addition, evolutionary explanations typically rely on the notion of adaptation. In other words, we identify features of linguistic structure that fit some functional need and then seek mechanisms that can explain this fit. Two such mechanisms are discussed at length in the literature: genetic transmission, and linguistic transmission. Natural selection operating on genetic transmission is a hugely successful explanation for adaptive structure. Its application in evolutionary linguistics was championed by Pinker and Bloom (Pinker & Bloom, 1990), and has been formalised in a number of recent papers by Nowak and colleagues. In Nowak et al's "language dynamical equation", for example, the average communicative success of a population determines its reproductive fitness (e.g., Nowak et el, 2001). Linguistic transmission is also commonly taken to be an evolutionary mechanism that can explain adaptive structure. For a language to survive over time it must be repeatedly used and learned by generations of linguistic agents. Models of linguistic evolution aim to relate properties of agents, such as learning bias, to observed features of linguistic structure. The central idea is that repeated learning and use will impose adaptive pressures, for example, it is claimed that recursive compositionality is language's response to learners limited exposure to data. The question we pose here is: can we develop a general theory of adaptation resulting from linguistic transmission? We will focus on two obstacles to the development of a general theory. Firstly, we must refine our understanding of computational models that we have previously used (e.g., Kirby 2002). The problem with models of linguistic evolution (aka. iterated learning) thus far is that there is little formal understanding of the process, with existing models varying substantially. The result is that general statements are hard to justify. An alternative approach is to look at idealised mathematical models of the transmission process (e.g., Niyogi & Berwick 1997). Thus far, however, these models have not been used to understand the origins of adaptive structure. Secondly, we need to better understand what the fundamental elements of a general theory should be. It is tempting to use biological evolution as a model and seek out "replicators" and "selection mechanisms", but is this justified? How confident should we be that such analogies can be made? Research into computational simulations has promised to inform the debate concerning the origins and evolution of language, but thus far research into this area has not been formally consolidated. In this talk we will identify the key issues in need of resolving and focus on recent research into answering the questions. References * Nowak, M. A., N.L Kormarova and P. Niyogi. (2001). Evolution of Universal Grammar. Science. 291: 114-117 * Pinker, S. and P. Bloom. (1990). Natural language and Natural Selection. Behavioural and Brain Sciences. 13: 707-784 * Kirby, S. (2002). Natural Language from Artificial Life. Artificial Life. 8(2):185--215. * Niyogi, P. and R. C. Berwick. (1997). Evolutionary Consequences of Language Learning. Linguistics and Philosophy. 20:697-719