Categorical Perception -
Advanced Review Categorical perception Robert L. Goldstone��� and Andrew T. Hendrickson Categorical perception (CP) is the phenomenon by which the categories possessed by an observer influences the observers��� perception. Experimentally, CP is revealed when an observer���s ability to make perceptual discriminations between things is better when those things belong to different categories rather than the same category, controlling for the physical difference between the things. We consider several core questions related to CP: Is it caused by innate and/or learned categories, how early in the information processing stream do categories influence perception, and what is the relation between ongoing linguistic processing and CP? CP for both speech and visual entities are surveyed, as are computational and mathematical models of CP. CP is an important phenomenon in cognitive science because it represents an essential adaptation of perception to support categorizations that an organism needs to make. Sensory signals that could be linearly related to physical qualities are warped in a nonlinear manner, transforming analog inputs into quasi-digital, quasi-symbolic encodings. ��� 2009 John Wiley & Sons, Ltd. WIREs Cogn Sci 2010 1 69���78 Wseven hen we look at a rainbow, we tend to see about distinct bands of color, even though we know from physics that the dominant wavelength of light that meets one���s eye changes smoothly from the top to bottom of the rainbow. Although the rainbow presents itself to us with a continuous and full range of visible wavelengths of light, we tend to see it in terms of distinct colors such as red, yellow, blue, and violet. This effect is a striking example of categorical perception (CP). According to this phenomenon, we tend to perceive our world in terms of the categories that we have formed. Our perceptions are warped such that differences between objects that belong in different categories are accentuated, and differences between objects that fall into the same category are deemphasized. This is allegorically illustrated in Figure 1. CP is an important phenomenon in cognitive sci- ence because it involves the interplay between humans��� higher-level conceptual systems and their lower-level perceptual systems. Traditional information flow dia- grams in cognitive science typically draw a clean division between perceptual and conceptual systems with information moving only from perception to the conceptual system the frequency of CP effects indi- cates permeability and bidirectional influence between ���Correspondence to: rgoldsto@indiana.edu Department of Psychology and Brain Sciences, Indiana University, Bloomington, Indiana 47405, USA DOI: 10.1002/wcs.26 these systems. We humans do not simply base our categories on the outputs of perceptual systems inde- pendent of feedback. Instead, our perceptual systems become customized to the task-useful categories that we acquire, slowly at the evolutionary timescale or quickly at the timescale of individual learning. Another reason why CP is theoretically impor- tant is that offers a potential account for how the apparently symbolic activity of high-level cognition can be grounded in perception and action.1 A basic feature of human symbolic thought is that people form equivalence classes. In the classical notion of an equivalence class, distinguishable stimuli come to be treated as the same thing once they have been placed in the same category.2 This kind of equivalence is too strong when it comes to human concepts because even when we place two objects into the same category, we do not treat them as the same thing for all purposes and objects can be placed in different categories when in different contexts. Some researchers have stressed the intrinsic variability of human concepts���variabil- ity that makes it unlikely that a concept has the same sense or meaning each time it is used.3,4 Still, it is impressive the extent to which perceptually dissimilar things can be treated equivalently given the appropri- ate conceptualization. To the biologist armed with a strong mammal concept, even whales and dogs may be treated as equivalent in contexts related to bio- chemistry, child rearing, and thermoregulation. Even Volume 1, January/February 2010 ��� 2009 John Wiley & Sons, Ltd. 69
Advanced Review wires.wiley.com/cogsci FIGURE 1 | An illustration of categorical perception. When an observer looks at objects (chickens) that fall into two or more categories (coops), differences among objects that fall into different categories are exaggerated, and differences among objects that fall into the same category are minimized. Conceived by Robert Goldstone, Made perceptual by Joe Lee. sea lions may possess equivalence classes, as Schuster- man et al.5 have argued that these animals show free substitution between two entities once they have been associated together. CP provides a mechanism for the origin of these (near-) equivalence classes. By CP, our perceptual systems transform relatively linear sensory signals into relatively nonlinear internal representations. The extreme case of this kind of nonlinear transformation is a step function by which increases to a sensory signal have no effect on perception until the signal reaches a certain threshold. At that threshold, perception changes qualitatively and suddenly. During the flat portion of the staircase function, different input signals have equivalent effects. Hence, CP can provide us with equivalence classes, the beginning of proto- symbolic thought. Why would we, or mother nature, want to build cognitive systems with equivalence classes? One reason is that they are relatively impervious to superficial similarities. Once one has formed a concept that treats all skunks as equivalent for some purposes, irrelevant variations among skunks can be greatly deemphasized. People may never be able to transcend superficial appearances when categorizing objects,6 nor is it clear that they would want to.7 Still, one of the most powerful aspects of concepts is their ability to make superficially different things alike.8 If one has the concept ���Things to remove from a burning house���, even children and jewelry become similar.9 Across modalities, the spoken phonemes /d/ /o/ /g/, the French word ���chien���, the written word ���dog���, and a picture of a dog can all trigger one���s concept of dog,10 and although they may trigger slightly different representations, much of the core information will be the same. Equivalence classes are particularly useful when we need to make connections between things that have different apparent forms. Equivalence classes are particularly useful when we need to make connections between things that have different apparent forms. CP is the first stage of this process of responding to the essential, rather than superficial, aspect of an entity. It is the same reason why most current electronics are digital: To provide tolerance to superficial variation in voltage signals that are irrelevant to the critical information. It may well be that current computers are too brittle because they throw away too much analog variation in their pursuit of discrete symbols. Still, it is worth remembering that the informational system benefiting from the most years of ���research and development���, provided by evolution is the genetic code of life itself, which closely approximates a digital code consisting of nucleotides and codons. Complex cellular machinery is dedicated to assuring that the code is relatively inert, and is protected from many contextual influences.11 It is reasonable to think that our cognitive system benefits from the same strategy of developing (quasi-)reusable codes. CP IN SPEECH As operationalized in psychology, CP is said to be present when people more reliably distinguish physically different stimuli when the stimuli come from different categories than when they come from the same category.12 The effect was originally established with speech phoneme categories. For example, Liberman et al.13 generated a continuum of equally spaced consonant-vowel syllables with endpoints reliably identified as /be/ and /ge/, as shown in Figure 2 (top left graph) by varying the second formant transition.14 There is a point (around stimulus value 4) where there is a relatively rapid decrease in the probability of observers hearing the sound as a /be/ to hearing it as /de/. At a later point, around values 9 and 10, observers rapidly shift from /de/ to /ge/ identifications. In addition to giving participants an identification task, participants were also given an ABX discrimination task. In this task, observers listened to three sounds���A followed by B followed by X���and indicated whether X was identical to A or B. Observers performed the task more accurately when syllables A and B belonged to different phonemic categories, as indicated by their identification probabilities, than when they 70 ��� 2009 John Wiley & Sons, Ltd. Volume 1, January/February 2010