So far, this chapter has focused on bottom-up processing. Bottom-up processing empha- sizes the importance of the stimulus in object recognition. Specifically, the physical stim- uli from the environment are registered on the sensory receptors. This information is then passed on to higher, more sophisticated levels in the perceptual system (Gordon, 2004).
For example, a moment ago, the sensory receptors in the retina of my left eye regis- tered information about a shape that streaked passed my window. This information included characteristics such as the object’s shape (vaguely oval, with a long, narrow structure on one end), its color (pale brown, with two narrow, darker stripes), and its speed (rapid). The arrival of this information started the object-recognition process. This information started from the most basic (or bottom) level, and it worked its way up until it reached the more sophisticated cognitive processes beyond the primary visual cortex. The combination of sim- ple, bottom-level features allows us to recognize more complex, whole objects.
The very first part of visual processing may be bottom-up (Palmer, 2002). How- ever, an instant later, the second process begins. This second process in object recogni- tion is top-down processing. Top-down processing emphasizes how a person’s concepts and higher-level mental processes influence object recognition. Specifically, our con- cepts, expectations, and memory help in identifying objects. We expect certain shapes to be found in certain locations, and we expect to encounter these shapes because of past experiences. These expectations help us recognize objects very rapidly. In other words, our expectations at the higher (or top) level of visual processing will work their way down and guide our early processing of the visual stimulus (Donderi, 2006; Gregory, 2004a). Here is a likely explanation for the perceptual experience that I just described. My house is located in a woodland setting. An occasional deer, wild turkey, or coyote may wander past my window, but chipmunks and squirrels are much more common. My top-down processing (in terms of my expectations and memory) combined together with specific information about the stimulus from bottom-up processing. As a result, I quickly concluded that the blurry shape must have been a chipmunk.
Top-down processing is especially strong when stimuli are incomplete or ambigu- ous. Top-down processing is also strong when a stimulus is registered for just a fraction of a second (Groome, 1999).
Top-Down Processing and Visual Object Recognition 45
How does top-down processing operate in vision? Some researchers propose that specific structures along the route between the retina and the visual cortex may play a role. These structures may store information about the relative likelihood of seeing various visual stimuli in a specific context (Kersten et al., 2004).
Cognitive psychologists propose that both bottom-up and top-down processing are necessary to explain the complexities of object recognition (Riddoch & Humphreys, 2001). We cannot ask whether perceivers interpret just the whole or just the parts, because both processes are required. For example, you recognize a coffee cup because of two almost simultaneous processes: (1) Bottom-up processing forces you to register the com- ponent features, such as the curve of the cup’s handle; and (2) the context of a coffee shop encourages you to recognize the handle on the cup more quickly, because of top-down processing. Let’s now consider how this top-down processing facilitates reading.
Top-Down Processing and Reading
Before you read further, try Demonstration 2.3. As you can see, the same shape—an ambiguous letter—is sometimes perceived as an H and sometimes as an A. In this demonstration, you began to identify the whole word THE, and your tentative knowl- edge of that word helped to identify the second letter as an H. Similarly, your knowl- edge of the words MAN and RAN helped you identify that same ambiguous letter as an A in this different context.
Researchers have demonstrated that top-down processing can influence our abil- ity to recognize a variety of objects (e.g., Gregory, 2004a; Henderson & Hollingworth, 2003; Hollingworth & Henderson, 2004; Kersten et al., 2004; Riddoch & Humphreys, 2001; Tanaka & Curran, 2001). Most of the research on this topic examines how con- text helps us recognize letters of the alphabet during reading.
Psychologists who study reading have realized for decades that a theory of recog- nition must include some factor (or factors) other than the information in the stimu- lus. When you read, suppose that you do identify each letter by analyzing its features. In addition, suppose that each letter contains four distinctive features, a conservative guess. Taking into account the number of letters in an average word—and the average reading rate—this would mean that you would need to analyze about 5,000 features every minute. This estimate is ridiculously high; your perceptual processes couldn’t handle that kind of workload!
Demonstration 2.3
Context and Pattern Recognition
Furthermore, we can still manage to read a sentence, even if some of the middle let- ters in a word have been rearranged. For example, Rayner and his colleagues (2006) found that college students could read normal sentences at the rate of about 255 words per minute. They could still read jumbled sentences such as, “The boy cuold not slove the pro- belm so he aksed for help.” However, their reading rate dropped to 227 words per minute. One of the most widely demonstrated phenomena in the research on recognition is the word superiority effect. According to the word superiority effect, we can iden- tify a single letter more accurately and more rapidly when it appears in a meaningful word than when it appears alone by itself or else in a meaningless string of unrelated letters (Palmer, 2002).
For example, Reicher (1969) demonstrated that people’s recognition accuracy was significantly higher when a letter appeared in a word such as work, rather than in a non- word such as orwk. Since then, dozens of studies have confirmed the importance of top- down processing in letter recognition (e.g., Grainger & Jacobs, 2005; Jordan & Bevan, 1994; Palmer, 1999; Williams et al., 2006). For example, the letter s is quickly recognized in the word island, even though the s is not pronounced in this word (Krueger, 1992).
Researchers have also shown that the context of a sentence facilitates the recogni- tion of a word in a sentence. For example, people easily recognize the word juice in the sentence, “Mary drank her orange juice” (Forster, 1981; Stanovich & West, 1981, 1983). Let’s discuss a classic study that explored this word-in-a-sentence effect. Rueckl and Oden (1986) demonstrated that both the features of the stimulus and the nature of the context influence word recognition. That is, both bottom-up and top-down pro- cessing operate in a coordinated fashion. These researchers used stimuli that were either letters or letter-like characters. For example, one set of stimuli consisted of a perfectly formed letter n, a perfectly formed letter r, and three symbols that were intermediate between those two letters. Notice these stimuli arranged along the bottom of Figure 2.6. In each case, the letter pattern was embedded in the letter sequence “bea-s.” As a result, the study included five stimuli that ranged between “beans” and “bears.” (In other words, this variable tested the effects of bottom-up processing.)
The nature of the context was also varied by using the sentence frame, “The _____ raised (bears/beans) to supplement his income.” The researchers constructed four sen- tences by filling the blank with a carefully selected term: “lion tamer,” “zookeeper,” “botanist,” and “dairy farmer.” You’ll notice that a lion tamer and a zookeeper are more likely to raise bears, whereas the botanist and the dairy farmer are more likely to raise beans. Other similar ambiguous letters and sentence frames were also constructed, each using four different nouns or noun phrases. (In other words, this variable tested the effects of top-down processing.)
Figure 2.6 shows the results. As you can see, people were definitely more likely to choose the “bears” response when the line segment on the right side of the letter was short, rather than long: The features of the stimulus are extremely important because word recognition operates in a bottom-up fashion. However, you’ll also notice that people were somewhat more likely to choose the “bears” response in the lion tamer and zookeeper sentences than in the botanist and dairy farmer sentences: The context is important because word recognition also operates in a top-down fashion. Specifi- cally, our knowledge about the world leads us to expect that lion tamers and zookeep- ers would be more likely to raise bears than beans.
Top-Down Processing and Visual Object Recognition 47
Think about how these context effects can influence your reading speed. The pre- vious letters in a word help you identify the remaining letters more quickly. Further- more, the other words in a sentence help you identify the individual words more quickly. Without context to help you read faster, you might still be reading the introduction to this chapter!
FIGURE 2.6
The Influence of Stimulus Features and Sentence Context on Word Identification.
Source: Based on Rueckl & Oden, 1986
10% 0 20% 30% 40% 50% 60% 70% 80% 90% 100%
Stimulus feature (used in the word bea-s)
Percent bears response (rather than beans ) Botanist and dairy farmer Lion tamer and
zookeeper
n
n n
n n