The emergence and development of structural systematicity in language: an experimental study

The main objective of this project is to investigate the dynamic structure of signal-meaning mappings in language. We will empirically measure how systematicity (the regularity in the mappings between signals and their meanings) in linguistic structure emerges and evolves as a result of pressures imposed on language through its usage and cultural transmission. This project will form the first large-scale experimental study into the evolution of linguistic structure using human language learners; we will design and implement a series of experiments using a novel paradigm of co-operative language learning, with the following objectives:
1. To investigate whether languages evolve to become more systematic, or whether there is a relationship between the level of systematicity of a language and the degree to which its structure persists through cycles of use and transmission; to measure the interactions between systematicity and frequency.

2. To quantify the effects of the social and communicative context in which a language is learnt and used on its systematicity; to distinguish the differential effects of generational turnover and usage within a single generation.
3. To investigate the directionality of symbolic associations between signals and meanings; to determine whether signals adapt to meanings to a larger extent than meanings adapt to signals, or vice versa.

4. To characterise the dynamics of cumulative adaptation in linguistic structure in terms of the co-adaptations between signals and meanings.
More broadly, the main theoretical objective of the research is to establish new connections between evolutionary linguistics and the neighbouring disciplines of psychology and cultural evolution, bringing empirical evidence from psycholinguistic experiments to bear on the suitability of cumulative cultural evolution as a powerful explanatory mechanism for the emergence and evolution of linguistic structure. The project's final objective is to develop, and to release to the wider research community, the experimental framework used in this project, so that it will be able to be used for future research into related questions.

Our species possesses an ability that is unique in nature: we communicate meanings to one another that have never previously been spoken. This feat is possible by virtue of structural properties that are present in all languages. In particular, the way signals are formed systematically relates to their corresponding meanings. For example, we understand the meaning of a sentence because we understand the meanings of its component words, and the way they are put together.
How did language come to have exhibit this systematicity of structure? Since it clearly bestows on our species significant adaptive benefits, a reasonable suggestion is that it arose as an innate component of our cognitive apparatus through biological evolution by natural selection. However, recently a number of scholars have emphasised the importance of cultural evolution in our species’ origins and development. Could this provide an alternative explanation for systematicity in language?
In this project we have demonstrated that systematicity does indeed arise through the process of cultural evolution. By recreating the transmission of languages in the experiment lab, we examined the process of language evolution in controlled circumstances. We found that systematic structure evolves out of randomness in certain circumstances and not others. When language is acquired by naive learners and used for communication, structure emerges.
Finally, by running computer simulations of cultural and biological evolution, we have shown that this new process of linguistic evolution actually inhibits the emergence of language-specific genes in our species, casting further doubt on the idea that language structure is innate.

Our central finding is that systematicity in language is a cultural adaptation arising from the influence of naive learners on the transmission of language in a communicative context. We first developed some of the analytical techniques for language evolution experiments (Cornish et al 2009, Language Learning) and then greatly extended the methodology by adding a communicative context, thereby avoiding the need for the previous experiments’ artificial interventions to maintain the expressivity of the evolving language (Scott-Phillips & Kirby 2010, Trends in Cognitive Sciences). In addition, we are now able to contrast horizontal and vertical cultural transmission, and by doing so show that it is transmission to naive learners that is critical for language to evolve a systematic (i.e. compositional) mapping between meanings and signals. This is a surprise, since the standard assumption in the cultural evolution literature is that there are few significant differences between horizontal and vertical transmission (Kirby et al in preparation for Proceedings of the Royal Society B). In addition to examining the effect of transmission, our experiment allows us to look at the effect of negotiation and the communicative nature of the task (Tamariz et al in preparation for Cognition). We also investigated the effects of the level of systematicity in a language on how well it is learned and reproduced (Smith et al 2010, EVOLANG proceedings & in preparation for Cognitive Science)
Our demonstration of the crucial role for cultural transmission has significant implications for the nativist hypothesis, and we have taken advantage of the grant being situated within the Language Evolution and Computation Research Unit to collaborate with other researchers and identify the effect of interaction between cultural and biological evolution using a Bayesian model. The striking result is that the existence of cultural transmission of language means that there are no conditions under which strong domain-specific innate constraints will evolve (Kirby et al, in preparation for Nature).
Our experiments are predicated on the assumption that learners have the ability to infer the meanings reliably in an interaction. To support this conclusion, we have collaborated with other researchers in the research unit to explore cross-situational learning of meaning-signal pairs using both experiments and computational models which we reported in the journal Cognitive Science (Smith et al, 2011 Cognitive Science; Blythe et al, 2010 Cognitive Science).
A central theme in this project has been the effect of a presumed weak bias towards systematicity that operates especially in naive learners. Accordingly, we are interested in the developmental profile of this bias, and how it might influence both the output of child learners and potentially the structure of child-directed speech. A complete investigation of this awaits a follow-up ESRC application by Kirby and Scarabela later this year that will focus particularly on the effect of diminutives cross-linguistically on systematicity. However, we were able to lay the foundations for future research with our preferential-looking experiments. Preliminary results suggest that the systematicity bias is indeed detectible in experiments with children as young as 3;6.