Ravignani, A. 1, 2 , Delgado, T. 1 & Kirby, S. 1
1 Language Evolution and Computation Research Unit, School of Philosophy, Psychology and Language Sciences, University of Edinburgh
2 Artificial Intelligence Lab, Vrije Universiteit Brussel
Humans are well-versed at perceiving and learning sequences. When confronted with language or other culturally-transmitted systems in the lab, humans introduce and amplify structural regularities making the systems easier to learn. Is this regularization process language-specific? We tested participants in a non-linguistic iterated learning task. Participants were asked to reproduce sets of drumming sequences, with recalled rhythmic patterns becoming training sets for new generations of participants in a transmission chain. Here, a number of regularities developed over experimental generations: (i) within transmission chains, time between the start of one drum hit and the start of the next hit became categorically distributed from ranges that were initially continuous and uniformly distributed, leading to drumming sequences composed of few alternating inter-onset durations; (ii) drumming patterns became easier to learn, shown by measuring imitation fidelity; (iii) drumming sequences became more systematic over time: i.e the structure of each sequence in a recalled set provides expectations about the structure of other sequences. Finally, the evolution of structural regularities in this duration-based task strikingly resembles previous results in a visual sequencing task. We conclude that structural regularities do not only emerge when semantics is involved or when learning a set of explicitly language-like behaviours.