Concepts and Knowledge

Cognitive Psychology: Theory, Process, and Methodology - Dawn M. McBride, J. Cooper Cutting 2019


Concepts and Knowledge

Questions to Consider

· What is a concept?

· How are concepts mentally represented?

· How are concepts and knowledge organized?

· What do we use concepts for?

Introduction: Game Night

Game night is always a big hit at our house. Sometimes we play a variation of the old $10,000 Pyramid TV game show. In our version of the game, one person names exemplars and the rest of the players have to figure out what category the exemplars are from (e.g., Charlie says, “bird, airplane, baseball, bat.” Isabel calls out, “things that fly”). Points are awarded for the number of exemplars and for guessing the correct category. Sometimes the categories are fairly straightforward (e.g., robin, sparrow, hawk, cardinal, penguin, ostrich: “birds”), but other times they are more difficult (e.g., wallet, photo album, laptop computer, cell phone, jewelry, vintage vinyl record collection, guitar: “things you would take from your burning house”). Other popular hits on game night include Bridge (the classic card game), Beyond Balderdash, Ticket to Ride, Munchkin, and Apples to Apples. When I announce, “Kids, let’s do a game night tonight. Go pick out a game,” the kids run off to the playroom and return (after some heated discussion sometimes) with the night’s game. How do they know what to count as a game? Part of the answer lies in the fact that they’ve had a lot of experience with games (my son even watches an Internet show that showcases a different “tabletop game” each episode). Based on these experiences they are able to recognize and select things that fit their mental concept of “game.”

This chapter is about the pieces of our mental world: concepts and knowledge. The material is closely related to discussions of semantic memory in Chapter 5 and of language meaning in Chapter 9. In his book The Big Book of Concepts, Gregory Murphy (2002, p. 1) opens with “Concepts are the glue that holds our mental world together.” Concepts are our mental representations of categories of things in the world. Being able to recognize and group things into mental categories is an extremely important cognitive ability. It allows us to identify what something is, what the properties of that thing are, and how we can behave with the thing (e.g., Should we eat it? Will it hurt us? What can we use it for?). For example, imagine that you were handed the object in Photo 10.1. You may not know what to do with it. However, if somebody told you it was a fruit (it is a pitaya or dragon fruit), then you would probably assume that it is something you can eat, that it probably has seeds, and that it may be sweet. You would base these assumptions on what you know about the concept of fruit. Furthermore, if somebody were to show you another one, you would be able to recognize it as a fruit, with the properties of fruit, without needing to be told these things again. Indeed, without our ability to categorize like this, we would have to identify and learn the properties of things anew each and every time we encountered an object.

We begin this chapter with the classical view of concepts as definitions and then review the theoretical and empirical problems with this view. Then we describe three alternative views. Following this, we describe how concepts are used for categorization, organized into larger structures of knowledge, combined together, and used to make inductive inferences.

What Are Concepts?

The Classical Approach: Concepts as Definitions

If somebody were to ask you, “What is a square?” you may come up with something like “a closed four-sided figure, composed of four straight lines of equal length, joined at ninety-degree angles.” This definition of square works quite well. The set of features are necessary (identifying the features something must have to be a square) and sufficient (if something has all of these features, then it must be a square) for identifying members of the category. The advantage of this approach is that using a definition is a relatively easy way to identify whether an object is or is not a member of the category. All one needs to do is to match the features of the object with the features listed in the definition. This view of concepts as definitions with lists of necessary and sufficient properties can be traced back to early Greek philosophers (e.g., Plato and Aristotle) and was generally assumed until the mid-twentieth century. However, in the last century philosophers and psychologists began identifying problems with this view.

Photo 10.1 If you were handed this object, what would you think that it is? What could you do with it?

Image

©iStockphoto.com/v777999

Theoretical Problems With Definitions as Concepts

Let’s change our example. Suppose that somebody asked you, “What is a game?” As with our square example, you would probably try to think up a definition of what a game is. However, in contrast with the square example, you would probably find it difficult to come up with a single definition that adequately captures everything you categorize as a game. Philosopher Ludwig Wittgenstein (1953), shown in Photo 10.2, used the concept of “game” as part of a theoretical argument against the definition approach to concepts. He argued that it may not be possible to identify a list of necessary and sufficient features for many categories, in particular “real-world” categories. Consider what is common to board games. Now extend that to card games, ball games, and to the Olympic Games. What features are common to them all? Here are some possible features of games: have competition (winners and losers), have an aspect of luck and/or skill, provide fun or amusement. But consider a child throwing a ball against the garage and catching it. Is she playing a game? If the answer is yes, then who are the winners and losers? Wittgenstein (1953) argued that the category members shared a family resemblance. That is, it is usually easy to see that children look like their parents, although it may be difficult to pinpoint the precise set of features they share (see Photo 10.3). Family resemblance points not to a single set of defining features but rather to members of categories connected by overlapping sets of features. In this approach, concepts are not defined by necessary and sufficient features but rather connected by a series of overlapping similarities across features. Consider another of Wittgenstein’s examples: the concept of “chair.” Look at the objects presented in Photo 10.4. Most people would agree that they are all examples of their concept of chair. However, while it may be easy to agree about how best to categorize these objects, it is not as easy to agree on a common definition of what a chair is. Give it a try. Write down what you believe are the necessary and sufficient features of a chair. Compare your definition with those of other students in your class. Chances are you won’t find the same level of agreement about the features as you did with the categorization of the pictures. Wittgenstein’s theoretical arguments are generally viewed as strong evidence against the classical definition approach of concepts.

Family resemblance: things belonging to a category are related by virtue of having a set of overlapping similar set of features

Photo 10.2 Austrian philosopher Ludwig Wittgenstein

Image

Photo Researchers/Science Source/Getty Images

Empirical Problems With Definitions as Concepts

In addition to Wittgenstein’s theoretical arguments, many empirical findings suggest that the classical view of concepts as definitions is incorrect. One characteristic of the definition approach is that it determines whether something is part of a category, but once something is determined to be a category member it does not make distinctions between category members. However, McCloskey and Glucksberg (1978) demonstrated that category boundaries are not always so clear-cut. They presented their participants with pairs of words. The second word was a category name. The participants’ task was to quickly judge whether the first word was a member of that category (e.g., dog-mammal, participants should indicate yes). Their results indicated that for some items, this task was easy: Items were either clear members (e.g., chair-furniture, yes) or clear nonmembers (cucumber-furniture, no). However, some items were much more difficult (e.g., bookcase-furniture; curtains-furniture). For these items, there was disagreement across participants (with some responding yes and others no) as well as within participants across different testing sessions (for some items they changed their minds 22 percent of the time). The data suggest that we do not treat all members of a category equally. Instead we behave as if some members of a category are “better” than others. For example, take a minute and write down all of the birds you can think of. Chances are that birds like “robin,” “blue jay,” and “sparrow” are category members you wrote down early in your list. But consider birds like “ostrich” and “penguin.” Where did these birds fall on your list (if they made it at all)? These members are usually considered much less “typical” than birds like “robin” and “sparrow.”

Photo 10.3 A mother and her two daughters. Notice the family resemblance between the three.

Image

Caroline Woodham/Alamy Stock Photo

Rosch and Mervis (1975) presented participants with twenty members of six categories (see Table 10.1 for three examples) and asked participants to rate the typicality of each member. A separate group of participants was asked to list attributes of each of the members. Some attributes were listed more frequently than others. Exemplars that had more of these frequent attributes were considered more typical members of the category. Rosch and Mervis interpreted these findings as support for Wittgenstein’s family resemblance view. They argued that concepts are overlapping networks of attributes. Typicality of members within a category depends on how they compare to an abstract combination of the most frequent attributes. So, typical category members have many frequent attributes (i.e., features common to many category members) and very few attributes that are frequent in other categories. This theory is discussed in greater detail later in the chapter.

Photo 10.4 Examples of the concept “chair.”

Image

©iStockphoto.com/MarkSwallow

©iStockphoto.com/jgfoto

©iStockphoto.com/Akhilesh

©iStockphoto.com/Enma_Ai

Image

The typicality effect is among the most common empirical findings in cognitive psychology and has been found using a wide range of methodologies beyond rating tasks. For example, Rips, Shoben, and Smith (1973) used a speeded category verification task in which they presented participants with sentences like “A robin is a bird” or “An elephant is a bird.” Participants had to respond with “True” or “False” as quickly as they could. They found that responses were much faster for typical members of a category (e.g., “A robin is a bird”) than for atypical members (e.g., “A chicken is a bird”). As in our demonstration earlier, Mervis, Catlin, and Rosch (1976) showed that typical items are produced first when prompted to produce category members. Typical items are usually learned first (e.g., Meints, Plunkett, & Harris, 1999) too. When mentioning two category members together, the more typical member is usually mentioned first (e.g., “robins and penguins” rather than “penguins and robins”; Kelly, Bock, & Keil, 1986). Garrod and Sanford (1977) demonstrated that reading time of an anaphor is faster (“the vehicle” in “the vehicle narrowly missed the pedestrian”) if the antecedent it refers to is a typical category member (e.g., “the car” versus “the bus” in “the bus/car came roaring around the corner”).

Typicality effect: a result where more common members of a category show a processing advantage

Stop and Think

· 10.1. What are necessary and sufficient properties of a concept?

· 10.2. What are the major theoretical and empirical arguments against concepts as definitions

Further support for typicality effects comes from patients with semantic dementia who show progressive impairment of conceptual knowledge. Mayberry, Sage, and Lambon Ralph (2011) demonstrated that the impairments are constrained by concept typicality. They asked patients to match words and pictures to categories. Their patients made more errors on atypical items than on typical items. Additionally, typicality effects are not restricted solely within categories. Barsalou (1985) found typicality effects for exemplars outside of categories. For example, a chair is considered a better nonmember of the category “bird” than a butterfly is, further demonstrating that category membership is not an all-or-none process.

While the definition approach may hold some intuitive appeal and work for some artificial categories (e.g., “square”), it has generally fallen out of favor as a processing model. Its main failings include the absence of clear necessary and sufficient features, no clear categorical boundaries, and typicality effects for category and noncategory members. The section that follows briefly discusses alternative theoretical approaches proposed to explain how we mentally represent concepts.

Alternative Approaches to Concepts

Many theoretical approaches to concepts have been proposed to replace the classical definitional approach. This section briefly reviews several of these approaches. Keep in mind that within each of these approaches are many individual theories, each with their own specific details (much like the categories they have been constructed to explain).

Prototype Approach

The prototype approach grew primarily from the theoretical and empirical work initially developed by Eleanor Rosch (Rosch & Mervis, 1975; sometimes this approach is referred to as the family resemblance or probabilistic approach). This approach views concepts as abstract representations (prototypes) that summarize the common and distinctive attributes of the members of the category that comprise the concept (e.g., Hampton, 1979; Smith, Rips, & Shoben, 1974). The prototype of a category is essentially a weighted average of the important features of its members. Important features are those shared by most of the members (common) and not by members of other categories (distinctive). Category membership is determined by virtue of the similarity of the object’s attributes to the prototype’s attributes.

Prototype approach: the idea that concepts are represented based on a typical (common) instance of that concept

Think back to our opening story about all of the things my family considers games. Bridge involves two pairs of people competing against each other at cards to reach at least 100 points first. It consists of multiple rounds of hands, with each hand consisting of a bidding stage and a playing stage. Beyond Balderdash consists of a group of people making up potential definitions of an obscure word, the basic plot corresponding to an obscure movie title, or things that happened on a particular date. These made-up things are read aloud, along with the actual answer, and players vote for the one they believe is the actual answer. Points are awarded for getting the correct response or having other players vote for the response you wrote. Ticket to Ride involves building railroads along different routes across a board with a map on it. Longer routes are awarded more points. Players randomly select cards with target routes (e.g., Los Angeles to Miami) that the player is awarded extra points for achieving. At the end of the game, the player with the most points wins. Table 10.2 presents feature lists for five potential members of the concept of “game.” Looking over these examples, one might abstract a prototype for games as things that have a system of rules and use cards where players compete for points and the highest point getter is the winner. Now let’s suppose we encounter Yahtzee for the first time. Would it fit into our concept of “game”? According to the prototype approach, we would compare the features of Yahtzee to our prototype features. While Yahtzee doesn’t include the use of cards, it does share the other features with our prototype. Now consider playing catch (throwing a baseball between two individuals). Would we consider that a game? Maybe not, since there does not seem to be much overlap of features. Suppose that we had two pairs of people tossing the ball, with each team counting the number of successful catches and declaring the team with the highest count the winners. Now the scenario overlaps more with the features of our prototype, and our judgment of it may change to include it as part of the concept.

Image

Figure 10.1 Recalled Examples of Animals Similar to Robins and Ostriches

Image

However, not all researchers accept the idea of a single abstract representation that spans an entire concept. Instead, they propose an approach grounded in the belief that categorization of new objects is based on specific memories of past examples, rather than something like a single prototype.

Exemplar Approach

The exemplar approach (e.g., Medin & Schaffer, 1978) proposes that concepts consist of separate representations of experienced examples of the category. In other words, categorization of an object is accomplished by comparing it to all of your memories of similar things. Staying with our games example, suppose you open up a present and it is a colorful box containing some dice, a deck of cards, some plastic tokens, a board, and a set of rules. These contents may bring to mind specific experiences you have had that involve objects with similar features (e.g., Monopoly, Candy Land, Risk) so that you compare the memory of these objects with the new one in front of you and determine that it is a board game. The major difference between this approach and the prototype approach is that comparisons are being made to memories of actual experiences rather than an abstraction of those experiences.

Exemplar approach: the idea that concepts are represented based on exemplars of the category that one has experienced previously

So how does this approach explain the typicality results? Recall that the most typical items of a concept are those that are similar to many other members of the concept. So on average, the more typical of a concept an object is, the more similar it will be to recalled members of that concept. The less typical of the concept an object is, the fewer members of that concept that will be recalled. Additionally, the object may have many features similar to members of other concepts, resulting in the retrieval of memories of noncategory objects. For example, suppose you see a robin for the first time. It may bring to mind memories of many birds (e.g., sparrows, cardinals, woodpeckers, and blue jays). The high similarity of features between robins and these remembered birds results in “robin” being interpreted as a typical member of the concept “bird” (see Figure 10.1). However, suppose you saw an ostrich. Ostriches don’t share many features with most common birds. They aren’t small and don’t fly or hang out in trees. Instead they are big, with a long neck and long legs, and have feathers that look like fur. If you have encountered an emu before, an ostrich will probably come to mind, but not many other birds are likely to (maybe swans and geese). You might even think of other large animals like alpacas. As a result of these recalled memories, ostriches are considered much less birdlike than the robin.

Figure 10.2 Examples of Stimuli Like Those Used in the Allen and Brooks (1991) Study

Image

A lot of research has attempted to distinguish between the exemplar and prototype approaches. Much of this work has used experimental paradigms in which participants are taught new artificial concepts and then tested with novel examples. The advantage of using artificial concepts is that researchers can tightly control the features involved and can examine how the concepts are initially acquired. For example, Allen and Brooks (1991) presented participants with cartoon animals having different environmental background contexts (e.g., desert or forest scene). They systematically manipulated features of the cartoon characters. Participants had to learn a rule to categorize the cartoon characters into either “diggers” (who dig holes to live in) or “builders” (who build homes from materials in their environment). Half of the participants were explicitly told the rule; the others were not told the rule. Of the five features manipulated, three were relevant to the categorization (i.e., leg length, angularity of body type, spotted or not) and two were not (i.e., number of feet and length of neck). Figure 10.2 presents some examples of these artificial stimuli. Participants were trained to learn the categorization rule using eight exemplars. Following the learning phase, participants were then tested with new examples. The researchers could vary the similarity of the new test items to the exemplars used in the learning phase by manipulating the nonrelevant features (number of feet and length of neck as well as the environmental context). The researchers could create new items that were either “good” or “bad” matches. Bad matches were created by keeping irrelevant features constant (same background, number of feet, and neck length) but changing one of the critical features resulting in the item being a member of the other category (see Figure 10.3). Good matches were created by changing a feature that did not change the category. Participants were slower and made more errors categorizing “bad matches” than “good matches” (see Figure 10.4). This finding suggests that participants were relying on the similarities to the specific learned exemplars rather than relying on an abstraction like a prototype.

Mack, Preston, and Love (2013) compared computational models of the exemplar and prototype approaches with fMRI scans of people’s brains as they performed a categorization task. Prior to scanning, participants were taught to categorize novel objects into two categories. Following this, they then categorized old and new objects while in the fMRI scanner. Both the exemplar and prototype models accounted well for the behavioral data. The researchers then used the two models to compute the representational match between the test objects and the different representations (exemplars versus prototypes). They then compared these representational matches with the brain response data. Their results indicated that the exemplar model provided a better prediction than the prototype model for most of their participants.

While most of the results from experiments using artificial concepts favor the exemplar approach over the prototype approach (see Murphy, 2002, for a review), it is important to recognize a potential limitation of such research. The conceptual structures used in these artificial concepts are very simple relative to naturally occurring concepts. So for naturalistic concepts like games or birds, we can develop much richer prototypes and have more exemplars with which to make comparisons. Indeed, evidence suggests that we may use both approaches, depending on context. Malt (1989) used pictures of real animals in a priming task that allowed her to investigate whether exemplars or prototypes were activated during a categorization task. Across a series of experiments, her results suggested that we may use both exemplar and prototype representations to make categorical decisions.

Figure 10.3 Examples of “Good” and “Bad” Matches Used in the Allen and Brooks (1991) Study

Image

Figure 10.4 Results of the Allen and Brooks (1991) Study

Image

In many respects, the prototype and exemplar approaches are similar. In both, concept learning and categorization involve identifying features and making comparisons to either an abstract prototype or other recalled exemplars. However, both approaches place a heavy emphasis on observable features and also largely ignore the role of prior knowledge in learning and using concepts.

Concepts Based on World Knowledge Approach

Barsalou (1985) examined the typicality of taxonomic concepts like those used in Rosch and Mervis’s (1975) study along with a set of goal-derived concepts (e.g., birthday presents, foods not to eat on a diet, things to take from your house if it is on fire). Goal-derived concepts are categories of things grouped together, not because of shared observable features but rather how well their members satisfy a particular purpose. Barsalou measured three variables: central tendency (essentially a measure of family resemblance), frequency of instantiation (how often an item was considered a member of a category), and how well an item satisfied the goal (which Barsalou called the “ideal”). His results indicated that all three variables were important to determining an item’s typicality. Because the exemplar and prototype approaches depend on observable features, the finding of an abstract feature like goal directedness (the ideal) is problematic for these approaches. Results like these have led some theorists to develop an approach in which conceptual structures are part of a larger system of general knowledge.

In Chapters 7 and 9 we introduced the concepts of schemata and scripts as representations of knowledge. Cohen and Murphy (1984) argued that prototypes are better represented as schemata than as unstructured lists. For example, rather than representing a bird as an unstructured list of weighted features, a schema for “bird” would be a structured set of dimensions (often called slots) that can be specified with particular values. Our schema for birds may include dimensions for physical characteristics like “outer skin: feathers”; “number of legs: two”; “mouth type: beak”; “movement: flies, walks, swims.” Furthermore, the dimensions may be connected such that they may restrict the values they can take. For example, number of legs and movement might be connected such that if the object has no legs, then movement can’t take “walks” as a value. This approach represents a move toward richer conceptual representations incorporating broader pieces of general knowledge.

Murphy and Medin (1985) argued that similarity-based theories of concepts fall short of adequately describing why concepts are coherent or meaningful because they don’t take into account our theories of how the world works. Consider our concept of “bird” again. The properties “has wings,” “is covered in feathers,” “lives in nests,” and “can fly” are related to each other. Lists of features may capture the fact that these features often co-occur, but the theory approach goes beyond simple correlation. Our knowledge about the world provides a reason that explains the co-occurrence of these features: lightweight feathers and wings allow birds to fly, which in turn allows them to nest in trees high above many predators. The causal relationships between these features are part of our general world knowledge, and their use as part of the conceptual process can explain how and why the features in our conceptual representations stick together. Similarly, knowledge may also play a role in explaining why an ostrich, which does not have the highly salient bird feature “can fly,” is still considered a member of the category if we consider that the reason it can’t fly is that it is too heavy for its wings to carry it aloft. The theories approach may also explain why some features are listed while others are not. For example, even though birds and airplanes are both often brightly colored, we would probably only list it as a feature for birds because it is not a particularly salient or important property of airplanes. In contrast, a feature like “has wings” is salient for both concepts and will likely be listed for both birds and airplanes.

Lin and Murphy (1997) examined the influence of knowledge within a categorization task. They had two groups of participants learn about an artificial tool (a “tuk”), like that shown in Figure 10.5, and were presented with a story about how the tool was used. The main experimental manipulation was in the functional importance of some of the features in the two stories. In Story A, Part 1 is critical to the functioning of the tool (used to capture the prey) but not in Story B (used to hang the tool for storage). The opposite is the case for Part 2 (it stores the pesticide for Story B; in Story A it protects the hunter’s hand). In the learning phase of the study, participants were shown exemplars of each category along with either Story A or Story B and asked to memorize what the category was about. During the categorization phase, participants were given single exemplars (see the right side of Figure 10.5) and asked to answer quickly as to whether the item was a tuk (they also asked participants to rate how typical they thought the items were of the category). Some of the exemplars lacked the critical functional feature, making them inconsistent with the story the participants were presented with during the learning phase. Across a series of experiments, with participants who were given Story A during learning, exemplars consistent with Story A were categorized as a “tuk” more often, rated as more typical, and categorized faster than those inconsistent with Story A (see Figure 10.6). These results clearly demonstrate the importance of general background causal knowledge for our conceptual system (Carey, 1985; Keil, 1989; Rips, 1989).

Figure 10.5 Examples of Stimuli Used in the Lin and Murphy (1997) Study

Image

Other Alternative Approaches to Concepts

The previous section briefly reviewed three current approaches to the question of how we represent conceptual knowledge. While these approaches have been widely adopted and investigated, they are not the only alternatives to the classical approach of concepts as definitions. The approaches already described have been developed largely within the representational theoretical framework of cognitive psychology. Other approaches have been proposed within different frameworks. For example, Barsalou (1999) proposed the perceptual symbols theory of conceptual representation that has its roots grounded within an embodied theoretical framework. This approach proposes that our conceptual system is largely perceptually based rather than based in amodal symbolic representations. In this approach, a concept like “apple” isn’t represented separately from our perception and actions. Instead, how we see apples, how apples smell and taste, and how it sounds and feels when we bite into an apple are all directly part of our represented concept of apple. A number of models of the conceptual system have also been proposed within connectionist frameworks (e.g., Cree, McRae, & McNorgan, 1999; Rogers & McClelland, 2004; Smith & Minda, 2000) inspired by neural networks. Network models like these highlight a feature of concepts that we have not yet discussed in this chapter: Individual concepts are typically organized as part of larger knowledge structures. The next section reviews approaches proposed about how concepts are organized.

Figure 10.6 Results From the Lin and Murphy (1997) Study

Image

Stop and Think

· 10.3. What is a prototype? How are prototypes used to represent concepts?

· 10.4. What is an exemplar? How are exemplars used to represent concepts?

· 10.5. How are the prototype and exemplar approaches similar?

· 10.6. How might world knowledge impact conceptual representations?

Organizing Our Concepts

Conceptual Hierarchies

To this point we have dealt primarily with examples in which we are trying to decide whether something is a member of a particular isolated concept. However, our conceptual world rarely breaks down into such a circumscribed situation. Consider the activity being played in Photo 10.5. We can describe the picture as people playing a game, or a card game, or poker, or perhaps even a particular type of poker (e.g., five-card draw or Texas hold ’em). The point is that single objects or events are typically members of many different larger or smaller categories. Empirical studies have demonstrated that concepts are typically structured hierarchically. Based on work examining a broad range of cultures, Berlin (1992) argued that this is a universal feature of all natural world categories. Figure 10.7 shows a simplified conceptual hierarchy for games. Categories higher in the figure are referred to as superordinate to lower levels, while categories lower are referred to as subordinate to higher levels. The links between concepts represent “is a” relations, in the sense that “poker” is a member of “card games.” One of the features of this organization is that subordinate categories may inherit the properties of their superordinate categories. For example, if you learn something new about the category “card games,” you may be able to generalize this new knowledge to all of the subcategories of card games. This feature allows us to know a lot about something that we may never have actually encountered once we learn what category it belongs to. Furthermore, these relationships are assumed to be transitive. That is, if poker is a kind of card game, and Texas hold ’em is a kind of poker, then Texas hold ’em is a kind of card game with all of the properties of a card game.

Photo 10.5 People enjoying a game of poker.

Image

©iStockphoto.com/RyanJLane

Basic-Level Concepts

Consider the pictures in Photo 10.6. What would you call each thing? Most people will answer this question with “dog,” “flower,” and “car.” However, other reasonable answers could include “border collie,” “daisy,” and “Ford Thunderbird,” or “animal,” “plant,” and “vehicle.” Another common finding is that one level is typically privileged over other levels. These privileged levels are commonly referred to as basic-level concepts. Roger Brown (1958) observed that parents typically prefer to use these middle levels of the hierarchy of concepts when speaking to their children. Research has established a wide variety of basic-level effects: children learn basic categories and their names before other levels (e.g., Anglin, 1977), basic categories typically share common shapes and movements (e.g., Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976), they allow for faster categorization of pictures (e.g., Tanaka & Taylor, 1991), and they are used more frequently in free naming (e.g., Cruse, 1977).

Basic-level concept: level of concept hierarchy where common objects (e.g., dog) reside

Figure 10.7 Simplified Hierarchy of “Games” Concept

Image

Photo 10.6 Basic concepts activity.

Image

©iStockphoto.com/Anton_Sokolov

©iStockphoto.com/Imo

©iStockphoto.com/GlobalP

Rosch and her colleagues (e.g., Rosch, 1978; Rosch et al., 1976) argued that basic-level objects are those at which the category members share the highest number of features. This suggests that basic-level concepts are more informative than other levels (e.g., Markman & Wisniewski, 1997; Murphy & Brownell, 1985). Basic levels provide a lot of information about the categories and are also distinct from other concepts at a similar level in the hierarchy (Murphy & Lassaline, 1997). For example, consider the basic categories of cats and dogs. Knowing that something is a cat is very informative; you can infer many properties about it (e.g., it meows, chases mice, has whiskers, purrs). You also know that dogs and cats are distinct concepts (e.g., dogs bark, chase cats, have a wet nose). Superordinate concepts (e.g., mammals and reptiles) tend to be distinctive but not as informative. Subordinate concepts (e.g., spaniels and border collies) tend to be informative but not as distinctive.

Organizational Approaches

Stored-Network Approaches

How these hierarchies and basic levels are mentally represented is a matter of theoretical and empirical debate. One theoretical approach is that these hierarchies are stored in memory as networks of relationships. For example, consider the model of semantic memory (see Figure 10.8). This model, proposed by Collins and Quillian (1969), is a network of related concepts and their associated features. Links in the model correspond to different kinds of relationships. “Is a” links represent the hierarchical structure, while the “has,” “is,” and “can” links represent the properties associated with the concepts. Collins and Quillian proposed that when an object is categorized, “activation” (think of it as a kind of mental flow of information) spreads from that object’s corresponding concept node to other associated nodes. For example, to verify the statement “A canary is a bird,” activation would spread from the concepts “canary” and “bird.” If the spreading activations intersected, then the answer would be “yes.” A major advantage of this organization is one of cognitive economy. For example, features shared by all animals can be stored at the animal level and need not be stored at any of the subordinate levels (e.g., bird, fish, robin, or trout). In addition to efficient mental storage, this organization also allows for property inheritance and generalization of new objects. For example, upon learning that a horse is an animal, the concept “horse” would inherit the features associated with the concept “animal.” The model predicted distance effects: The more “is a” links that need to be traversed, the longer the verification times. Early research using a speeded property verification task (e.g., reply “true” or “false” as quickly as you can to the following statements: “A canary is red,” “A rose has roots”) supported the model’s predictions.

Superordinate concept: the level of concept hierarchy where general categories of the basic-level concepts (e.g., mammal) reside

Subordinate concept: the level of concept hierarchy where specific exemplars of a basic-level concept (e.g., husky) reside

However, further research yielded problems for the model. It was not able to account for typicality effects (e.g., Rips et al., 1973). The model had no mechanism to explain why some subordinates were considered better than others. Hampton (1982) found that the transitive inheritance of properties is sometimes violated. For example, while people agree with the two statements “A lamp is a kind of furniture” and “A car headlight is a kind of lamp,” they don’t verify the transitive combination “A car headlight is a kind of furniture.”

Consistent with the normal scientific process, a wide variety of other network models have been developed and tested in response to these (and other) limitations. For example, Collins and Loftus’s (1975) revised semantic network model proposed changes to how activation was spread and the addition of variable weighting on the connections between concepts (e.g., to capture typicality effects, the link between “robin” and “bird” would be stronger than the link between “penguin” and “bird”). Many other models have been proposed (e.g., Anderson, 1976; Anderson & Bower, 1973; McClelland & Rumelhart, 1981; Nelson, Kitto, Galea, McEvoy, & Bruza, 2013). In fact, the stored-network approach is among the most widely adopted and persistent theoretical approaches within cognitive psychology.

Cognitive economy: the idea that concept information is stored at the most efficient level of the hierarchy

Figure 10.8 Simplified Version of Collins and Quillian’s (1969) Taxonomic Hierarchy Model

Image

Feature Comparisons Approaches

An alternative to the stored network view is that hierarchical relationships are computed using reasoning processes rather than being directly stored in a semantic network. In this approach, deciding how concepts are related involves comparing features of the two concepts. In other words, if you were encountering a tapir for the first time and trying to decide whether it is an animal (see Photo 10.7), you would compare the features stored with the concept “animal” to the features of the tapir (a roughly pig-shaped herbaceous mammal found in regions of the Southern Hemisphere). Given that tapirs share features with animals (e.g., can move and have skin, eyes, ears, and mouth), you would make the inference that they are considered animals. Typicality effects would reflect the degree of overlapping similarity of features.

Photo 10.7 If seeing this object for the first time, what features would you consider to decide if it is an animal?

Image

©iStockphoto.com/wrangel

Neuroscience-Inspired Approaches

Recently, several other feature-based models have been proposed using a variety of frameworks. While the details of these accounts are complex and beyond the scope of our current review, they are similar in that none of the models includes explicit representation of hierarchical conceptual structure (and in some cases no direct representations of concepts). However, even without these relationships explicitly represented, the models can simulate the effects demonstrated in the research (e.g., Hampton, 1997; Murphy, Hampton, & Milovanovic, 2012).

Patients with semantic dementia suffer from the progressive impairment of their conceptual knowledge. Elizabeth Warrington (1975) described three patients who had impairments in their conceptual knowledge reflected in deteriorated vocabulary (both production and comprehension) and their knowledge about the properties of objects. These patients often had difficulty naming pictures and describing characteristics of common objects. Patterson, Nestor, and Rogers (2007) showed a picture of a zebra to a patient who replied that it was a horse and asked what the stripes were for. Neuroscientists have examined the pattern of deficits that patients exhibit (those with semantic dementia as well as other disorders) and proposed theories of how concepts are represented in our brains (e.g., Barsalou, 2010; Mahon & Caramazza, 2009).

Much of this work has focused on “where” in the brain concepts are located. In a review of the literature, Thompson-Schill (2003, p. 288) wrote, “The search for the neuroanatomical locus of semantic memory has simultaneously led us nowhere and everywhere.” It is widely believed that our conceptual knowledge is distributed across multiple regions of the brain, involving areas for both perception and action. Knowing about an orange may involve how it looks (round and orange), the taste, the smell, how to peel it, and how it can be split into sections. These features of an orange are probably represented in different brain regions. One of the central theoretical questions has been on the mechanism that ties all of these features together. Patterson et al. (2007) reviewed two sets of theories addressing this issue. One set of theories suggests that our concepts are directly represented within the connections between these sensorimotor areas. Other theories propose distinct areas of the brain (sometimes called convergence zones or hubs) that function to bind these features together such that there are conceptual representations distinct from sensory and motor areas. This is similar to the ideas described in Chapter 5 about how episodic memories are encoded and stored, where the relevant features (e.g., sensory features) are stored in the appropriate areas of the cortex (cortical areas specialized for that sense), and how features bind during encoding.

Figure 10.9 Two Theoretical Approaches Described in Patterson et al. (2007)

Image

Source: Patterson et al. (2007, figure 1).

Figure 10.9 represents these two views. Both approaches propose that concepts are represented in the network of connections (depicted by the orange lines) between the different cortical systems involved in representing objects. These areas correspond to regions responsible for the processing of sensory, motor, and linguistic information. The bottom half of the figure represents the approaches in which the conceptual system includes an area of convergence (shown in red in the bottom left part of the figure), where information from different cortical regions is bound together. Connections between the cortical regions and the convergence zone are depicted by the red lines.

Pobric, Jefferies, and Lambon Ralph (2010) used TMS (see Chapter 2 for a description of this technique) to test these different approaches. Using TMS, they were able to temporarily induce category-specific picture-naming deficits in normally unimpaired participants. They applied TMS to three brain regions: the anterior temporal lobe (ATL, which is thought to function as an amodal conceptual hub), the inferior parietal lobule (IPL, which is thought to be involved with processing concepts involving manipulable man-made objects), and the occipital pole (OCC, which served as a control condition). Stimulation of the ATL led to slowed naming across all types of concepts. In contrast, stimulation of the IPL generated category-specific slowed naming (only naming of highly manipulable objects was slowed). Stimulation of the OCC had no effects on picture naming. This pattern of results (see Figure 10.10) is consistent with the distributed plus hub approach.

Stop and Think

· 10.7. What are superordinate, basic, and subordinate levels of concepts?

· 10.8. What empirical evidence suggests that basic-level concepts are processed differently from other levels of concepts?

· 10.9. How do network models represent the hierarchical structure of concepts?

Summary of Conceptual Organization

Concepts are not isolated representations floating around in unstructured semantic memory spaces. Rather, concepts appear to operate in relatively stable and predictable organizational structures. While the structure of these systems is clearly related to each individual’s concepts and experience, there is remarkable similarity between individuals across languages and cultures. We group things together into similar categories and typically treat particular kinds of categories as basic for many tasks. How and why we represent these regularities is a matter of ongoing empirical and theoretical debate.

Figure 10.10 Results of Pobric et al.’s (2010) TMS Stimulation Study

Image

Using Concepts: Beyond Categorization

Up until this point, we have focused our review on research investigating how we use concepts to categorize the world around us. However, we use categories for other purposes as well. We use concepts to make predictions about the properties of new objects and categories. This process is called category induction. Our use of stereotypes to make predictions about people is based on social concepts. We can also combine concepts productively, which may result in the creation of new concepts. There are also individual differences in what we know and have experience with. So one might ask how the conceptual systems for experts may differ from those of novices. Explorations of how we use concepts beyond categorization processes have implications for theories of how they are represented in our cognitive systems. The final section of this chapter briefly reviews the research on some of these other conceptual processes.

Category Induction

Suppose that your neighbors call you up to ask you to take care of their cat while they are out of town for the weekend. Even if you have never seen their cat, you will have a general notion about what the cat looks like (e.g., furry, pointed ears, whiskers) and know that it may purr if you pet it and will need food and water while your neighbors are away. This ability to generalize from what we know about a category to a novel object is an example of category induction. It is one of the most important functions of our conceptual system.

Consider another example. If I told you there was a sickness going around the neighborhood and that another neighbor’s parrot was sick, would you be more worried about your pet parrot or your canary? What about your pet dog? Or your son? These instances of category induction involve inferring the properties of one category onto other categories.

Rips (1975) systematically examined how we make these kinds of inferences. He asked participants to imagine an island with eight species of animals on it (i.e., sparrows, robins, eagles, hawks, ducks, geese, ostriches, and bats). He then presented them with a statement about a given species of animal, like “all of the robins have a disease.” He then asked participants to rate the likelihood that another species on the island (i.e., the target category: bats or sparrows) would get the disease. Rips demonstrated that the likelihood of making the inference that another species would get the disease depended on two main factors. The typicality of the given species impacted whether participants made the inference that other species may get the disease. Estimates were higher if the diseased species in the initial statement was a typical member of the category (e.g., higher ratings if robins had the disease than ostriches). However, the typicality of the target category had no effect. The second important factor was the similarity between the given and target species. Ratings were also higher for robins and sparrows than for robins and bats.

Since this early study, a variety of other important characteristics have been shown to be important for category induction. Murphy and Ross (2005) demonstrated that how certain we are that something is a member of a category impacts the likelihood of making inductions (e.g., suppose that you aren’t sure whether a bat is a bird or not, but you are certain that a robin is a bird). Heit and Rubinstein (1994) found that the likelihood of the induction depends on how relevant it is to the kind of categories being compared. For example, you are more likely to make an inference about an anatomical property (e.g., the heart rate) than a behavioral property (e.g., migration patterns) between two species that are both mammals (e.g., a bear and a whale). However, if the two categories are instead related by virtue of their environment (a tuna and a whale both live in the sea), then you are more likely to make the induction about a behavioral property (e.g., migration patterns). In other words, our world knowledge about how properties are related to their categories has an impact on the inferences we make between categories. Similarly, Lassaline (1996) showed that the sharing of causal relationships of features between categories impacts inductions. For example, suppose that you were told that

1. A tenrec has a weak immune system, pale skin, and an acute sense of smell.

2. A spinosa has a weak immune system and pale skin.

Then you were asked about the likelihood that a spinosa has an acute sense of smell. The likelihood of making the induction increased if participants were also given a causal relationship between some of the features of the animals (e.g., for both animals a weak immune system causes their pale skins).

Stereotypes

We also make inductive inferences in our day-to-day social lives. Imagine that you are attending a local neighborhood mixer and are introduced to Steve. Steve seems friendly and outgoing, is dressed head to toe in black, and has spiked bleached-blond hair. He splits time discussing the different qualities of Fender and Gibson guitars and the novel he recently finished. If somebody were to ask you whether you thought Steve was a professional musician or a psychology professor, chances are pretty high you would answer that he is a musician. We often make decisions like these based on a person’s appearance, actions, and the context to classify the person into social categories. These processes appear very similar to those used when we examined the object in Photo 10.1. Once we recognize it as a type of fruit, we infer many properties, such as that it is edible and it may be sweet and have seeds.

Much of what we read or hear about in the news is about the negative consequences of using stereotypes, particularly with respect to targets of social stereotyping behaviors. Why do we use stereotypes when we make judgments and decisions about people? Social psychologists have adopted many of the conceptual theories discussed here when developing theories of how and why we use stereotypes. One common view is that using stereotypes is a fundamental and cognitively efficient way to interact in social contexts (Macrae, Milne, & Bodenhausen, 1994). It has been proposed that stereotypes are part of a two-stage process (e.g., Banaji & Greenwald, 1994; Devine, 1989). The first stage is an automatic activation of stereotypic knowledge within some kind of stored representation of knowledge. In other words, when we encounter other people, we quickly sort them into social categories based on their readily available features. In the encounter with Steve, we may initially categorize him as a musician based on his appearance and his interest in guitars. This initial stage may later be followed by a second, more controlled deliberate stage of processing (see Chapters 4 and 12 for more discussion of automatic and controlled processing). As we learn more about Steve (e.g., that he works at the local university and does research examining scientific reasoning), we are able to overcome the initial stereotyping processes and correctly categorize him as being a psychology professor (Macrae, Bodenhausen, & Milne, 1995).

A common assumption in these theories is that the stereotypic knowledge is learned and represented in the same way as the conceptual systems we have been discussing in this chapter. In their review, James Hilton and William von Hippel (1996, p. 240) define stereotypes as “beliefs about characteristics, attributes, and behaviors of members of certain groups. More than just beliefs about groups, they are also theories of how and why certain attributes go together.” Brewer, Dull, and Lui (1981) demonstrated that stereotypes of the elderly may be represented hierarchically (e.g., subordinates: grandmother, elder statesman, senior citizen) and that within this hierarchy most stereotypical behaviors appear to operate at a basic level, rather than at more general superordinate or subordinate levels. Findings like these support the notion that stereotype conceptual representations may operate in much the same way as our more general conceptual system.

Expertise

Look back at Figure 10.1 and name as many of the birds as you can. If you are like me, you may not feel like you know a lot about birds. However, you may know somebody who knows a lot about birds (e.g., the person who reads about birds, often goes on bird-watching vacations). How might being an expert about a particular domain impact our concepts and organization of concepts within that area of expertise?

Murphy and Wright (1984) compared feature lists generated for psychological disturbances (e.g., childhood emotional disorders) by groups, with levels of experience ranging from expert (e.g., practicing clinical psychologists) to novice (e.g., undergraduates). Their results indicated that experts have richer conceptual representations and higher levels of agreement in their feature lists for categories. Tanaka and Taylor (1991) examined the hierarchical conceptual structures for samples of bird and dog experts. They found that within their areas of expertise (e.g., the dog conceptual space for the dog experts), experts’ basic levels of categorization shifted to a lower level of the hierarchy (e.g., to a level that nonexperts typically considered subordinate). However, when those same experts were tested in a domain outside of their area (e.g., the bird conceptual space for dog experts), they considered the usual level of the hierarchy to be the basic level. Medin, Lynch, Coley, and Atran (1997) examined categorization and inductive reasoning in three types of tree experts (landscapers, taxonomists, and park maintenance workers). They found that the different group experts structured their conceptual systems differently. Landscapers tended to structure their categories along goal-derived purposes (e.g., how the trees are used), while taxonomists and maintenance workers structured their categories along scientific and folk-defined taxonomies, respectively. However, across types of experts, inductive reasoning suggested that the genus-level categories were treated as the basic level of their hierarchies.

Conceptual Combination

Think about an apple. What features would you list that make up the prototypical apple? Are the colors white or brown on your list? Probably not. Now consider the term “peeled apple.” Chances are that if you were to list the features of the combined concept it would include the feature of white (and maybe brown if you think about what happens to a peeled apple when exposed to the air for a few minutes). We opened this chapter with “Game night is always a big hit at our house.” If we consider word meanings as labels that represent concepts (see Chapter 9 for more discussion of word meaning and concepts), then we can think of a sentence as representing a large complex concept made up of the combination of smaller concepts. How do we combine individual concepts into larger, more complex concepts?

Most of the work on conceptual combination has focused on relatively small combinations (e.g., peeled apple, game night). Research suggests that the process is not simply the intersection of the two categories (e.g., the concept “game night” is not the things that are both in the categories of “game” and “night”). Smith and Osherson (1984) presented participants with pictures like those in Photo 10.8. They found that a picture of a red apple was judged to be more typical of the combined concept “red apple” than it was of either simple concept “apple” or “red things.” Interestingly, for atypical things like a picture of a brown apple, the effect is even larger. In this case the picture was rated as somewhat typical of “brown things,” not typical of “apples,” and very typical of “brown apples.”

Stop and Think

· 10.10. What is category induction? What factors have been shown to impact the processes of category induction?

· 10.11. How are stereotypes related to concepts?

· 10.12. How does expertise impact conceptual organization?

Standard prototype and exemplar models did not have mechanisms that could easily account for effects like these. Smith, Osherson, Rips, and Keane (1988) proposed a model in which concepts are represented as prototype schemata with dimensions and values (see our earlier discussion of Cohen & Murphy, 1984). While the Smith et al. (1988) model captured the early data well, further research has revealed limitations of the model. For example, sometimes it is difficult to predict which dimensions of the combined categories modify each other. For example, compare the changes to “apple” and “farmer” when modified in “organic apple” and “organic farmer.”

The Future of Research and Theory of Concepts

The review in this chapter reflects many of the central findings in the psychological investigation of concepts and knowledge. The theories presented represent the dominant approaches developed to explain the research. At this point you may be asking yourself, “Which approach is the correct one?” Unfortunately, there isn’t a simple answer to this seemingly easy question. None of the theoretical approaches can account for all of the data. In fact, the research reviewed in this chapter focused largely on relatively simple concepts of concrete objects. Researchers have also identified and explored other interesting aspects of concepts. Given what you now know about concepts, think about some of the following questions: How do children acquire concepts? Are there differences between natural and artificial categories? How are abstract concepts like “love” and “justice” and verb concepts like “run” represented? Where and how are concepts represented in the brain? Rather than abandon the approaches, researchers continue to develop and test new, more complex approaches.

Photo 10.8 Which of these apples would you consider to be more typical?

Image

Jupiterimages/Creatas/Thinkstock

Hemera Technologies/PhotoObjects.net/Thinkstock

Thinking About Research

As you read the following summary of a research study in psychology, think about the following questions:

1. What aspects of concepts are examined in this study?

2. What are the independent variables in this study?

3. What are the dependent variables in this study?

4. What alternative explanations can you come up with to explain the results of this study?

Study Reference

Sloutsky, V. M., Kloos, H., & Fisher, A. V. (2007). When looks are everything: Appearance similarity versus kind information in early induction. Psychological Science, 18, 179—185.

Purpose of the study: The authors examined the categorization and inductive processing of 4- and 5-year-olds. Of particular interest was whether young children base their categorical inductions on category membership or physical similarity. This summary describes only the first experiment of the research article.

Figure 10.11 Stimuli From the Sloutsky et al. (2007) Study

Image

Source: Sloutsky et al. (2007, figure 2).

Method of the study: The researchers presented children with pictures of artificial buglike animals (“ziblets” and “flurps,” see Figure 10.11). The animals were created by combining six attributes: body, tail, antennae, wings, buttons, and fingers. The two categories of animals were defined by the relationship between the number of buttons and fingers: one category had more fingers than buttons, the other fewer fingers than buttons. The children were told a story about getting a new pet from the store. The store had nice, friendly ziblets and wild, dangerous flurps. The children were told the rules about how to tell ziblets from flurps (by counting their fingers and buttons), along with examples of each type. During a learning phase, the children were presented with novel bugs and asked to categorize them as either ziblets or flurps and were provided feedback as to whether they were correct (along with a reminder of the distinguishing rules). This was then followed by categorization trials (like the learning trials but without feedback). Then the children were given an induction task, consisting of three animals. For one animal, they were told that it had a hidden property (e.g., it has thick blood), and they were then asked to select from the other two which one had the same hidden property. The researchers were able to construct stimuli so that they could directly compare category membership against similarity of appearance. An additional final categorization task was performed to ensure that the children had not forgotten the categorization rule.

Figure 10.12 Results From the Sloutsky et al. (2007) Study

Image

Source: Sloutsky et al. (2007, figure 3).

Results of the study: The results are presented in Figure 10.12. The children demonstrated clear use of category membership in the categorization task. In contrast, on the induction task, the children rarely used category membership when making their selections. Instead, the evidence suggests they used physical similarity to the item with the given hidden trait to make their inductive selection.

Conclusions of the study: The authors concluded that young children (4- and 5-year-olds) able to make categorization judgments with new categories do not use category membership to make inductive inferences.

Chapter Review

Summary

· What is a concept?

A concept is a mental representation of a category of things in the world. The conceptual representation is a mental organization of the knowledge we have about categories of things stored in our long-term memories.

· How are concepts mentally represented?

The chapter reviewed three main approaches. The classical approach of categories as definitions has generally been refuted on both theoretical and empirical grounds. The prototype approach is that concepts are represented as an abstract average of representative features of the items in a category. The exemplar approach is that concepts are based on similarities to retrieved memories of previously encountered category members. The knowledge-based approach suggests that conceptual representations must also include theories about how different features are related.

· How are concepts and knowledge organized?

Concepts appear to be organized hierarchically, with general superordinate groupings and more specific subordinate groupings. There is theoretical debate as to whether these hierarchical relationships are directly represented in long-term memory or computed through feature comparisons. Additionally, certain levels of the hierarchy are treated as basic-level concepts, showing preferred processing.

· What do we use concepts for?

Concepts may underlie most of our cognitive processes. We use them to categorize things, allowing what we already know about a concept to apply to new instances. Similarly, we can use concepts to make inferences about other similar concepts. We can also combine categories to productively create new and potentially more complex concepts.

Chapter Quiz

1. The classical approach to concepts is that they are mental representations of

1. the averaged features of all members of a category.

2. the collection of all retrieved memories of encounters with members of a category.

3. a definition consisting of necessary and sufficient features of all members of a category.

4. how and why features of category members are related to one another.

2. Rips et al. (1973) demonstrated that people verify “a robin is a bird” faster than “a chicken is a bird.” This is an example of

1. an exemplar effect.

2. a typicality effect.

3. a basic-level effect.

4. category induction.

3. Consider the concept of an apple. Match the concept label with its label within a conceptual hierarchy.

1. Basic level

2. Superordinate level

3. Subordinate level

§ ___ Golden delicious

§ ___ Fruit

§ ___ Apple

4. A schema representation for the concept “bird” consists of

1. an unordered list of common features.

2. a list of common features ordered in terms of their typicality.

3. a structured set of dimensions with particular values for the dimensions.

4. all of the recalled memories of past experiences with birds.

5. Imagine you read in the paper that a particular model of automobile had recently been recalled because of electrical issues. Based on the research on category induction, which of the following inferences would you most likely make?

1. That your car might develop electrical issues.

2. That your house might develop electrical issues.

3. That your truck might develop electrical issues.

4. That your car might develop mechanical issues.

6. Summarize the methods and conclusions from the Allen and Brooks (1991) study.

7. Summarize the methods and conclusions from the Lin and Murphy (1997) study.

8. Compare and contrast the exemplar and prototype views of concepts.

9. Why is the lack of transitive inheritance properties (Hampton, 1982) a problem for the Collins and Quillian (1969) model?

10. How does expertise in an area impact our conceptual representations?

11. How are stereotypes similar to other concept representations?

Key Terms

· Basic-level concept 271

· Cognitive economy 273

· Exemplar approach 264

· Family resemblance 259

· Prototype approach 262

· Subordinate concept 272

· Superordinate concept 272

· Typicality effect 261

Stop and Think Answers

· 10.1. What are necessary and sufficient properties of a concept?

Necessary and sufficient properties are those that define whether something is or is not a member of a category.

· 10.2. What are the major theoretical and empirical arguments against concepts as definitions?

Necessary and sufficient property definitions are generally difficult to derive for naturally occurring categories. Additionally, category membership does not appear to be all or none. Instead, some members of a category differ in how typical they are of the concept.

· 10.3. What is a prototype? How are prototypes used to represent concepts?

A prototype is an abstract average of representative features of the items in a category. Category membership is determined by comparing the features of an object with the features of the prototype representation.

· 10.4. What is an exemplar? How are exemplars used to represent concepts?

Exemplars are retrieved memories of previously encountered things. Category membership is determined by comparing the features of an object with the features of recalled exemplars of different categories.

· 10.5. How are the prototype and exemplar approaches similar?

Both the prototype and exemplar approaches rely on similarity comparisons of features between the current object and representations retrieved from memory.

· 10.6. How might world knowledge impact conceptual representations?

Theories about how features are related to one another have been shown to have an impact on how items are categorized and what kinds of categorical inferences are made.

· 10.7. What are superordinate, basic, and subordinate levels of concepts?

Evidence suggests that concepts are organized hierarchically, with general superordinate groupings and more specific subordinate groupings. Basic-level concepts are particular levels of the hierarchy shown to be processed preferentially. Members of basic-level concepts show high levels of similarity within their category and distinctiveness from things belonging to other concepts.

· 10.8. What empirical evidence suggests that basic-level concepts are processed differently from other levels of concepts?

There are many demonstrations of basic-level preferences. These include how children tend to learn basic-level category names early and how basic-level names allow for faster categorization and naming of pictures and show higher levels of category induction.

· 10.9. How do network models represent the hierarchical structure of concepts?

A common assumption in many theories of conceptual structure is that hierarchical relationships are directly stored as part of the concepts. For example, the concepts “robin” and “bird” would share an “is a” link. Alternatively, some approaches suggest that hierarchical relationships may be computed through feature comparisons. In other words, we can decide that a robin is a bird by virtue of the feature overlap between the concepts “bird” and “robin.”

· 10.10. What is category induction? What factors have been shown to impact the processes of category induction?

Our ability to generalize from what we know about a category to a novel object is an example of category induction. The typicality of category members and the feature similarity between the two concepts involved in the induction have been shown to impact the likelihood of making the induction.

· 10.11. How are stereotypes related to concepts?

Stereotypes have been characterized as conceptual representations of social categories. Like general concepts, stereotypes have been demonstrated to have hierarchical structure and basic levels.

· 10.12. How does expertise impact conceptual organization?

Experts have more experience and knowledge within particular domains. Evidence suggests that within their domains of expertise, experts may develop different hierarchical structures related to their experience and knowledge, as well as treat lower levels as their preferred basic level of representations.

Student Study Site

Image

edge.sagepub.com/mcbridecp2e

SAGE edge offers a robust online environment featuring an impressive array of free tools and resources for review, study, and further exploration, keeping both instructors and students on the cutting edge of teaching and learning.

Image

Image

Ianni Dimitrov Pictures/Alamy Stock Photo