Modelling emergence with exemplar-based iterated learning
Viktor Trón

The idea behind emergentist models is that language structure is shaped by use. Iterated learning focuses on the  learner as the main bottleneck in language transmission and queries to what extent the biases of cognition explain structural generalizations in a particular linguistic domain.

This talk describes a general dynamic exemplar-based model applied to a restricted domain of Hungarian morphology. The agent performing iterated learning is an exemplar-based model of linguistic categorization. Simulations in this model start from a dataset and let it undergo a series of transformations as it is learned and reproduced iteratively. Assessing the dynamics of this process helps us test hypotheses about directions of change and about the link between model properties and regularity in the data.