Kirby, S.
Perhaps the most striking difference between human language and the vast range of communication systems in the rest of nature is its systematic structure. Each utterance in a language stands within a network of interdependencies with other possible utterances, whereas a typical alarm call repertoire for example, is best thought of as a collection of independent items. The systematicity of language is what enables its widespread productivity, both by the combinatorial reorganisation of meaningless elements, and compositional reorganisation of meaningful elements. A key research challenge for an evolutionary approach to language is therefore the origins of this unique systematic structure.
I will survey experimental evidence that systematicity emerges as an inevitable consequence of the way in which language persists from one generation to the next by cultural transmission. This cultural transmission is the result of a process we call ?iterated learning?. The product of one individual's learning is the target for a subsequent learner, creating a chain of transmission in which behaviour can evolve as a result of learning biases. The same phenomenon can be observed across a wide range of experiments; previously independent items come to exhibit systematic interdependencies.
Most previous iterated learning experiments have involved participants being exposed to a set of data in a training phase, which is then recalled in a subsequent testing phase (and in turn forms the training data for the next participant). In this talk, I will introduce a more recent approach, ?immediate imitation iterated learning? which interleaves training and testing as participants imitate item by item. This allows us to observe systematicity arising as an effect of implicit statistical learning across trials. I will show how this technique can be applied to questions about the origins of sequential structure in both language and music, and the wider comparative landscape of language evolution.