Proceedings of the Annual Meeting of the Cognitive Science Society

Natural languages partition meanings into labelled categoriesin different ways, but this variation is constrained: languagesappear to achieve a near-optimal trade-off between simplicityand informativeness. Across 3 artificial language learning ex-periments, we verify that objectively simpler kinship systemsare easier for human participants to learn, and also show thatthe errors which occur during learning tend to increase sim-plicity while reducing informativeness. This latter result sug-gests that pressures for simplicity and informativeness operatethrough different mechanisms: learning favours simplicity, butthe pressure for informativeness must be enforced elsewhere,e.g. during language use in communicative interaction.