Cognitive Psychology: Sensation and Perception

Recent experimental work is beginning to explain how we see shapes and how we organize the mental representations that help us recognize familiar objects rapidly. According to one theory, object recognition involves a three-stage process. First, we segment an object into parts. This "parsing" tends to take place at points of concavity where the outline of the object curves inward. Second, we identify these parts as arrangements of simple three-dimensional shapes: cylinders, cones, blocks, and wedges. These primitive shapes or "geons" can be distinguished readily from one another regardless of the viewing angle. Images of even highly complex natural objects can be built up from a remarkably small set of geons; in fact, only 36 geons are needed to distinguish among all objects. In the third stage, according to the theory, we recognize familiar objects by virtue of their unique arrangement of geons.

Consistent with the theory, experiments on object recognition show, for example, that when asked to view something new, our eye movements tend to focus on points of concavity. Moreover, we can identify objects even when much visual information is missing so long as critical points of concavity are still represented. From these points people can infer the component geons, and from knowledge of the geon arrangement can infer what the objects are. When the concavities signaling geons are masked, objects are unidentifiable from their edges and vertices alone.

Perceiving Words in Reading

Reading written language presents its own shape-recognition challenges. Words are composed of letters, which in turn are composed of elementary features: lines, curves, and angles. From these features, how do we arrive at words? In one traditional view, the visual system first analyzes words into these elementary features, then recognizes certain combinations of features as meaningful letters, and then combinations of letters as meaningful words. Recent research has revealed, however, that reading is probably accomplished in a quite different way.

Consider, for example, a simple experimental task in which a subject must decide whether the last of a string of letters flashed quickly on a screen is a particular one, such as a "K." It turns out that subjects are faster when the string spells out a meaningful word such as WORK than when it makes up a word-like sequence such as SORK or a random-letter string such as CGHK. People process letters more effectively in the context of words than of nonwords. This is known as the "word superiority effect." If the traditional view were correct, letter perception would not be affected by whether the string was a word.

Many scientists now view visual perception in terms of information processing based on interactive activation Researchers assume that there is feedback between levels of processing. For example, if a single vertical line is found at the lowest (feature-analysis) level, this information is sent to the next (letter-recognition) level, where it activates representations of letters (such as T and N) that contain such a line. That information is then sent back to the feature-analysis level, prompting it to be especially sensitive to other features associated with the candidate letters, such as right angles (for T) or acute angles (for N).