MODELING PERCEPTUAL SIMILARITY AS ANALOGY RESOLVES THE PARADOX OF DIFFERENCE DETECTION

There is a paradoxical dissociation be- tween recognizing that two stimuli are differ- ent and recognizing how they are different. We show that this dissociation can be captured by modeling perceptual similarity as a species of analogical processes. Using SME to model comparison, we show that the dissociation arises naturally from different stages in the analogical mapping process. Rather than rely- ing on hand-coded input representations, our model uses an automatic, incremental encod- ing process to generate representations from the same stimuli as given to human partici- pants.