Extension of iterated learning model based on real-world experiment

Abstract

Some language acquisition studies have used the iterated learning model (ILM). The ILM has been proposed as a hypothetical model of language evolution, and thus the validity of the language acquisition model embedded in the ILM framework remains unclear. In this study, we examine the learning ability for language acquisition using an iterated learning experiment based on ILM in the real world. We introduce the Levenshtein distance for evaluating the similarity of language transmissions made between human participants and between ILM agents. In the real-world experiment, participants extract matching parts, which is a new learning ability that ILM agents do not have. The results show that the language similarity of ILM’s agents remains roughly constant. On the other hand, the language similarity of participants indicates a decreasing trend through transmission generations. We assume that this decreasing trend is caused by the additional process of extracting matching parts. Consequently, we introduce this process to the original ILM to construct an extended model.