Similarity
Robert L. Goldstone, Ji Yun Son · Oxford University Press eBooks · 2012
Abstract Humans and other animals perceive and act on the basis of similarities among things because similarities are usually informative. Similar things usually behave similarly, and because we can grasp these similarities, we can organize and predict the things in our world. Four major classes of models have been proposed for how humans assess similarities. In geometric models, entities are represented by their positions in a multidimensional space, and similarity is based on the proximity of entities in this space. In featural models, entities are described by their features, and the similarity of entities is an increasing function of their shared features and/or a decreasing function of their unique features. In alignment-based models, the similarity between two structured entities is calculated by placing the elements of their structures into correspondence. In transformational models, the similarity between two entities is conceptualized as the number of transformations required to transform one entity into the other. We discuss issues related to the flexibility and constraints of similarity, and how similarity grounds other cognitive processes.