Psychology Of Natural Concepts Research Paper

1. Concepts And Their Functions

A concept is a mental representation, and category refers to the set of items selected by a concept. The concept allows people to tell if a new item is a member of the category (classification). Thus, one can have the concept ‘dog’ and use that concept to know that a particular object belongs to the category of dogs.

Classification is an important function of concepts, but it is not the only one. Concepts are the representations that people use to think. They are necessary for many cognitive activities, such as for predictions, comprehension, reasoning, communication, and building new, more complex, concepts. In fact, it is these functions that make concepts so essential. For example, the concept ‘dog’ allows one to classify a new item as a dog. However, classifying something as a dog is not that useful unless one has lots of other knowledge about dogs that can be applied. If that knowledge allowed a prediction of whether the dog might bite, this would be important for knowing how to act. Similarly, one’s knowledge about dogs may allow one to understand why a dog chased a toy or to reason about the relation of dogs to other mammals. In addition, because people have much common knowledge and often refer to concepts by the same words, people can learn from others’ experiences. Finally, having a mental representation of dog means that one can combine it with other concepts to get new concepts, in what is called conceptual combinations, such as ‘couch dog.’

2. Conceptual Structure

A central question about natural concepts is their conceptual structure, how the information about each concept is organized and used for its various functions. To begin, what knowledge permits the classification of a new item?

2.1 The Classical View

The earliest answer, going back to at least Aristotle, is that conceptual knowledge includes a definition to identify new members. This classical view asserts that there is a set of simple features that define each concept—if an object has all of these features then it is a member of the category, otherwise it is not. For example, a square is a two-dimensional, four-sided, closed figure with equal angles and equal sides. All objects that have those features are squares, and any object that is missing one or more of these features is not a square. There are two main types of problems with this classical view as a theory of people’s conceptual structure. First, it fails to account for the fact that some members of a category are viewed as better, or more typical members, than other members of the category. Thus, ‘robin’ is a typical bird, whereas ‘ostrich’ is not. These effects are seen in both time to verify category membership and in ratings of typicality. If both birds meet the definition, then something else must be determining this goodness-of-category membership. Second, although it is not difficult to figure out the definition for concepts such as square, it is much more difficult to understand what the defining features are that allow us to classify an item as a member of a natural category such as birds. Clearly, many features are important to being a bird, but no-one has been able to characterize a set of observable features that would allow accurate classification of all birds without including some non-birds. Even bird-crucial features such as feathers are problematic, since a bird that lost its feathers, perhaps owing to disease, would still be a bird.

2.2 Prototypes And Family Resemblance

Natural concepts often have highly inter-related feature representations, even if there is no single set of features possessed by all members of the category. For example, birds have feathers and lay eggs, but in addition they have wings, beaks, tend to fly, and so on. Thus, the members of the category tend to have many similar features, even if none of them perfectly separates every bird from every non-bird. This idea led to the suggestion that the conceptual structure might consist of a summary representation, or prototype, that captures the most commonly occurring feature values for the category members (Posner and Keele 1970, Rosch 1978). To classify a new instance, one compares it with the prototypes and chooses the most similar prototype, with more important features given more weight. There are no necessary or sufficient features for membership, but prototypical features each add evidence that it is a member of that category. Thus, although an ostrich does not fly or live in trees, it can be classified as a bird because it has many of the other features in the bird prototype (such as has feathers, has wings, lays eggs, etc.).

This view addresses both the problems mentioned with the classical view. First, typicality is directly accounted for—more typical members have more of the prototypical features. For example, a robin might have all of the features, whereas an ostrich has only some. Second, there are no defining features, but rather a set of typical features that tend to be true of birds.

How are the members of the category related to the prototype? Although few members may have all the prototype features, all members will have some of the features. This type of structure is called family resemblance because it is similar to the feature distribution one would see with an extended family (Rosch and Mervis 1975; see Wittgenstein 1953 for an earlier treatment in philosophy). There will be some family features that occur often across the whole family (e.g., tall, red hair, blue eyes, small nose), but some family members will have more of them than others.

The prototype view has much intuitive appeal and accounts for many of the empirical findings. However, there are a set of problems that all relate to the idea that the prototype view is too simple—it does not take into account much of the other knowledge people have about the concepts, the category members, and the relation of this concept to other concepts. People often know not just the prototypical features, but the range of common feature values and whether there is a correlation among various features (such as size of birds and whether they sing). See Smith and Medin (1981) for a thorough examination of the classical and prototype views.

2.3 The Theory-Based View

The theory-based view argues that concepts must be integrated with other knowledge. First, the representation of concepts is not a simple list of prototypical features but includes the relations among features (such as correlations) and explanations as to how the features are related (for example, how wings, feathers, and hollow bones enable flying, or how flying enables having nests in trees). Second, the concepts are explicitly related to other concepts, because part of our understanding of any concept relies upon its relation to much else that we know. An understanding of a natural concept, such as ‘dog,’ requires one to relate it to what we know about living organisms, pets, digestive systems, and so on. The point of the theory-based view is that our conceptual knowledge requires embedding in these complex knowledge relations so that we can quickly and reliably use our knowledge for classification and other category functions; Medin (1989) explains this view in more detail and points out problems with other views that are dealt with by the theory-based idea.

Most importantly, this view helps understand a central issue in natural concepts, conceptual coherence—what makes categories members of a concept all go together? Instead of relying on similarity to the prototype, the theory-based view argues that there is a set of underlying (perhaps unobservable) features and explanations that lead to items being members of the same category. Thus, although birds have some similarities to one another, the real reason ‘birds’ is a coherent concept is because people believe birds to have some basic underlying commonality (perhaps genetic structure). The importance of this distinction is illustrated by porpoises—although they look like many fish, they are classified as mammals.

Despite the advantages of the theory-based view in bringing to bear much of our other knowledge, there is concern about the complexity of the view and that it has not been fully described yet. Many researchers believe that this complexity is necessary for integrating conceptual structure work with other work on concepts (some of which is described below), but also believe that the prototype view is a good, if simplified, description for many cases. There is also some question as to whether even all natural concepts have some underlying commonality that is necessary and sufficient for category membership (Malt 1994).

3. Organization Among Concepts

The focus has been on the structure of each concept, but the relations among concepts is an essential part of our knowledge. An important aspect of the relations among concepts also is the level of abstraction at which a particular item may be classified. For example, when people see an item, they are likely to classify it as an ‘apple’ not as a ‘fruit’ or a ‘Granny Smith’ apple. Rosch and her colleagues showed that there was one level, the basic le el, which is psychologically privileged. See Lassaline et al. (1992) for a review.

The evidence for this basic level is very strong. Not only do people choose to name objects at this level, they are faster to verify an item is at this level (an apple) than at either the more abstract (fruit) or more specific (Granny Smith apple) level. In addition, people often learn these concepts earlier and a variety of measures show it is the most abstract level at which objects share many common features.

The interpretation for the basic level is that it represents a compromise between two opposing goals of the categorizer: the goal to have the most information possible and the goal to have the categories be as discriminable (distinctive) as possible. The in-formativeness leads to more specific levels, but the distinctiveness favors more abstract levels. The basic level is a compromise—getting more specific (such as Granny Smith apple) does not add much information but does greatly lower the distinctiveness (since it must be distinguished from other apples). Getting more abstract (fruit) greatly reduces the in-formativeness.

Are there exceptions to the basic level? Yes, and they help to clarify how such a level may come about. Very atypical members may be classified more quickly at a more specific level. For example, a sports car is classified more quickly as a sports car than as a car. The claim is that the atypical items have many features that distinguish them from other members of the basic-level category (such as other cars), so the most informative and distinctive level may be at a more specific level. For natural concepts, think of an atypical bird such as penguin or ostrich. Second, the basic level seems to change with expertise. Dog trainers can classify the breeds they work with as quickly at the more specific level, such as collie, as at the basic level of dog. Again, distinctiveness appears to be an important part of the explanation—with extended experience, experts learn additional features distinguishing members of the basic level.

4. Uses Of Concepts Beyond Classification

Much of the research on natural concepts has focused on classification, which is why it has been the main focus of this research paper. However, as mentioned earlier, the importance of concepts is that they play a role in many cognitive activities. When people classify an object, they can access knowledge about the concept to help in various types of reasoning and understanding. Much of this work is reviewed by Medin and Atran (1999).

4.1 Inferences From A Concept

Even young children will use category membership to make inferences about a new item. Gelman and Markman (1986) showed children pictures of two animals (such as a flamingo and a bat) and taught the children novel properties about each animal. For example, they might learn that a flamingo feeds its baby mashed-up food and that a bat feeds its baby milk. They then were shown a picture of a blackbird that was perceptually similar to the bat, but was labeled as a bird. The children thought the blackbird would feed its baby mashed-up food, just like the flamingo. Categories provide a basis for inference.

Not only are categories important for inference when people know the category, but they are important for inference even when there is uncertainty about the category. Suppose you are in the woods and see an animal, which you think is probably a skunk, but could also be a small dog, squirrel, or raccoon. What should you infer about the probability it will bother you? One possibility is that you could estimate the probabilities that each of those animals would bother you and weight those estimates by the likelihood of each animal to obtain an overall probability. The second possibility is that you assume it is the most likely type of animal, in this case a skunk, and make your estimate based only on that category. Although the first way seems more rational, it comes at the cost of much increased effort, even in this simple case. In fact, under many circumstances, people choose the second approach, just relying on the most likely category.

This research shows the importance of conceptual knowledge in making inferences both when the category membership of an object is known and when it is uncertain.

4.2 Inference From Relations Among Concepts

Concepts are not represented separately from our other knowledge. As mentioned in Sect. 3, an important relation among concepts is the level of abstractness. The previous section examined the inferences made from a single category, whereas this section examines how inferences might be made from one category to other categories. For example, suppose you are told that a new Property X is true of robins and of bluejays. Do you think it would also be more likely to be true of sparrows or of geese? Most people answer sparrows—in general the more similar the first categories are to the last category, the stronger the inference. In addition to similarity, other factors affecting the strength of the inference are typicality and diversity. The more typical the first category, the stronger is the inference to the more abstract category —so if you know Property X is true of robins, you think it is likely to be true of all birds, compared with when you know the property is true of ostriches. Diversity refers to how varied the first categories are, or how much of the more abstract category they cover. For example, wolf and elephant are very different so if you know Property X is true of both of them, you are likely to believe it is true of all mammals. However, if you know Property X is true of two similar mammals, such as wolf and dog, then you are less likely to believe it is true of all mammals.

People use their knowledge of these hierarchies to make a variety of inferences. Some cross-cultural work points out similarities and differences between cultures that vary in their knowledge of biological categories. In particular, a Mayan culture in which the people spend much time with local mammal species shows much greater differentiation of the categories and also uses more knowledge of the ecology in making their inferences compared with American college students.

4.3 Conceptual Representations Are Affected By Use

Conceptual representations have large effects on a variety of conceptual uses, such as inferences, understanding, etc. In addition, it is important to remember that these conceptual representations are constructed partly in response to how they are used. This point is easiest to see in some work examining the organization of complex domains. When different types of tree experts sorted a large number of tree names, landscapers’ sortings were greatly influenced by landscaping properties of trees, such as their shadiness. Ross and Murphy (1999) had people sort a large number of food names. Although the dominant influence was standard food categories such as breads, fruits, and meats, there was a strong influence on food categories that arise from human activity, such as breakfast foods, snacks, and desserts. Hence one should not think of conceptual representation as mirroring the outside world. Our representations depend on the world and on what we do.

Bibliography:

1. Gelman S A, Markman E M 1986 Categories and induction in children. Cognition 23: 183–208
2. Lassaline M E, Wisniewski E J, Medin D L 1992 Basic levels in artificial and natural categories: Are all basic levels created equal? In: Burns B (ed.) Percepts, Concepts, and Categories. Elsevier, Amsterdam, pp. 327–78
3. Malt B C 1994 Water is not H O. Cognitive Psychology 27: 41–70
4. Medin D L 1989 Concepts and conceptual structure. American Psychologist 44: 1469–81
5. Medin D L, Atran S (eds.) 1999 Folk-biology. MIT Press, Boston
6. Posner M I, Keele S W 1970 Retention of abstract ideas. Journal of Experimental Psychology 83: 304–8
7. Rosch E 1978 Principles of categorization. In: Rosch E, Lloyd B (eds.) Cognition and Categorization. Erlbaum, Hillsdale, NJ, pp. 27–48
8. Rosch E, Mervis C B 1975 Family resemblances: Studies in the internal structure of categories. Cognitive Psychology 7: 573–605
9. Ross B H, Murphy G L 1999 Food for thought: Cross-classification and category organization in a complex real-world domain. Cognitive Psychology 38: 495–553
10. Smith E E, Medin D L 1981 Categories and Concepts. Harvard University Press, Cambridge, MA
11. Wittgenstein L 1953 Philosophical Investigations. Macmillan, New York