Similarity as transformation

We propose that similarity is determined by the transformation distance between representations: entities which are perceived to be similar have representations which are readily transformed into one another, whereas transforming between dissimilar entities requires many transformations. We present three experiments that indicate that similarity is strongly influenced by transformation distance. These data present a challenge for featural or spatial accounts of similarity. Finally, we introduce a family of transformation-based accounts of similarity, called 'Representational Distortion', as a specific example of a transformational approach to similarity. (C) 2002 Elsevier Science B.V. All rights reserved