To join the meeting, click https://uva-live.zoom.us/j/83932324002.
Abstract
Human language is striking in the way it exhibits systematic structure at all levels of description. Words are recombined to make phrases whose meanings are composed from the meanings of their parts. In addition, words themselves are created from the recombination of meaningless sub-lexical units. This duality of patterning is ubiquitous to language, but largely absent from the many other forms of communication in the natural world. Communication elsewhere in nature is typically made up of individual instances of direct pairings between meaning and signal, without the intervening levels of compositional and combinatorial structure. The story of the evolution of language is the story of a transition from item-based to system-based signalling.
What might drive this evolutionary shift from item to system? In this talk, I will set out an argument that cultural evolution by iterated learning is the engine that creates this kind of systematic structure. I will present two unpublished studies, inspired by data from emerging sign languages, that demonstrate how iterated learning can create combinatoriality and compositionality. In the first, I will present a simulation model which suggests that social structure can influence the rate at which combinatorial structure emerges. This provides an explanation for why Israeli Sign Language and Al-Sayyid Bedouin Sign Language differ in their degree of sub-lexical structure. In the second, I will present the results of a large scale online cultural evolution experiment modelling the iterated learning of gestural descriptions. The results show a transition from simultaneous to segmented descriptions of events. This mirrors a transition found across subsequent cohorts of signers in Nicaraguan Sign Language.
The cultural evolution of sets of behaviours appears inevitably to give rise to systematicity where there was once just a collection of individual items. This is ultimately due to a universal inductive bias for simplicity operating in learning, and amplified by culture. Systems are simpler, even if the behaviours they produce are more complex. This is why languages look the way they do.