ABSTRACT
This chapter examines recent progress in understanding the relationship between naturalness and learnability of concepts within the conceptual spaces framework. According to a recent proposal, natural concepts are represented by cells of optimally partitioned similarity spaces, where optimality is understood in terms of a number of criteria. Here, we focus on the learnability criterion, which posits that natural concepts should be more easily and reliably learnable than non-natural ones. We review two computational studies that investigate this relationship using different modeling approaches. Both studies show a significant advantage in learnability for natural concepts over non-natural ones. These findings provide support for the learnability criterion as a key feature distinguishing natural from non-natural concepts. We also discuss the limitations of the current research, including its focus on individual learning. We mention future directions, particularly the potential of a new type of agent-based model to investigate social aspects of concept acquisition in relation to the natural/non-natural distinction.
