Modeling individual mental representations underlying similarity judgments

Perception is inherently subjective and the way people categorize or judge the world around them can differ. While recent studies have shown how to uncover features of mental representations underlying similarity judgments, they generally ignore individual differences. In this paper, we propose a computational framework based on the ”odd-one-out” task to identify a shared set of features that describe similarities between stimuli, while allowing for the fact that different people may prioritize these features differently. We show how prior knowledge about a stimulus domain can be incorporated and tested and how to uncover new features in an exploratory manner. We demonstrate the framework in a study on accent perception, where we reveal that people’s first language can shape how they perceive continuous and categorical aspects of accents. We highlight the value of this framework for iteratively refining our understanding of how individuals mentally represent the stimuli in their environment.