Figure 101

A sample of the sorts of materials used in Bruner et al.'s (1956) study of concept acquisition.

shapes) with different values (e.g., cross/square, one/three, plain/striped). From the experimenter's viewpoint, certain items in the array were instances of a rule; for example, the rule three, square shapes identifies items 20, 23, 26 as members of it and all other items as non-members. In one of their tasks, subjects were shown one example of the rule and had to discover the correct rule by asking the experimenter whether other items were instances of the rule.

Bruner et al. identified several different strategies used by subjects in these experiments that could be viewed as possible ways in which people might acquire concepts in everyday life. However, Bruner et al.'s work was carried out in a domain of fairly artificial categories. Can we expect people to operate similarly when making judgements about natural categories, involving the commonplace objects of everyday life? The short answer is "no" (although see Armstrong, Gleitman, & Gleitman, 1983).

Category judgements reflect typicality gradients

Natural categories do not seem to be as clear-cut as Bruner et al.'s artificial categories; some instances of the category are better examples of the concept than others. For example, Rosch (1973) asked people to rate the typicality of different members of a concept and found that some members were rated as being much more typical than others (see also Rips, Shoben, & Smith, 1973). A robin was considered to be a better example of a bird than a canary. Indeed, the category can be described in terms of a typicality gradient of its members; that is, an ordering of the members of the category by their relative typicality scores. Furthermore, this typicality gradient is a good predictor of the time subjects take to make verification judgements. That is, subjects take longer to verify statements involving less typical members (e.g., "A penguin is a bird") than statements involving more typical members (e.g., "A robin is a bird").

Categories do not have clear boundaries

Some categories are fuzzy, their boundaries are not clear-cut to the extent that some members can slip in and out of the category. That is, even though some highly typical instances are considered by most people to be category members and less typical instances are considered to be non-members of the category, between these two extremes people differ on whether an object is a member of the category and are also inconsistent in their judgements. That is, sometimes they think the object is a member of the category and other times they think it is not. McCloskey and Glucksberg (1978) found that their subjects were sure about saying that a chair was a member of the category furniture and that a cucumber was not a member of this category. But they disagreed with one another on whether book-ends were a member of the category furniture and differed in their own category judgements from one session to the next (see also Barsalou, 1987; Hampton, 1998; and the later section on concept instability).

Category judgements with hierarchies

Intuitively, another common category judgement we can make is about the hierarchical relationships between concepts, captured by questions like "Is a chicken a bird?" and "Is a chicken an animal?". Much empirical research has been directed at this question to determine the structure of conceptual hierarchies. One of the key questions has been how many levels of abstraction are used by the human conceptual system. It is easy to think that there may be many levels of abstraction. For example, our hierarchies might start with things, below which there are living and non-living things, below living things there might be animals and vegetables, below animals there could be warm-blooded and cold-blooded animals, below warm-blooded animals there could be land-animals and sea-animals, and so on. For the sake of cognitive economy, it is clear that people must have some efficient scheme for organising hierarchies of concepts.

As we shall see, many studies have revealed that people use about three levels of abstraction and that there is a marked "basic-level" at which categorisation is carried out. The idea of a basic level arose out of anthropological studies of biological and zoological categories (Berlin, 1972; Berlin, Breedlove, & Raven, 1973; Brown et al., 1976). Berlin (1972) noted that the classification of plants used by the Tzeltal Indians of Mexico corresponded to the categories at a particular level in the scientific taxonomy of plants. For instance, in the case of trees, the cultures studied by Berlin were more likely to have terms for a genus such as beech than for general, superordinate groupings (e.g., deciduous, coniferous) or for individual species (e.g., silver beech, copper beech). The reason Berlin gave for this basic level was that categories such as "beech" and "birch" were naturally distinctive and coherent groupings; that is, the species they include tend to have common patterns of attributes such as leaf shape, bark colour and so on. The basic level was the best level at which to summarise categories. More recent research by Atran (1998; Atran et al., 1999; Lopez et al., 1997) has suggested that these conceptual systems are invariant, for just these categories, across many cultures, suggesting that they may be core domains of human knowledge that have been naturally selected for by evolution.

In psychology, Elanor Rosch and her associates discovered much of the specific evidence on the basic level and the three levels of generality (Rosch et al., 1976a). They found that at the highest level of abstraction, the superordinate level, people have general designations for very general categories, like furniture. At the lowest level, the subordinate level, there are specific types of objects (e.g., my favourite armchair, a kitchen chair). In between these two extremes is the basic level. While we often talk about general categories (that furniture is expensive) and about specific concepts (my new Cadillac), we typically deal with objects at the intermediate, basic level (whether there are enough chairs and desks in the office). Rosch et al. (1976a) asked people to list all the attributes of items at each of the three levels (e.g., furniture, chair, easy chair) and discovered that very few attributes were listed for the superordinate categories (like furniture) and many attributes were listed for the categories at the other two levels. However, at the lowest level very similar attributes were listed for different categories (e.g., easy chair, living-room chair).

Rosch et al. (1976a) also found evidence that basic-level categories have special properties not shared by categories at other levels. First, the basic level is the one at which adults spontaneously name objects and is also the one that is usually acquired first by young children. Furthermore, the basic level is the most general level at which people use similar motor movements for interacting with category members; for instance, all chairs can be sat on in roughly the same way and this differs markedly from the way we interact with tables.

Category members at the basic level also have fairly similar overall shapes and so a mental image can capture the whole category. Finally, objects at the basic level are recognised more quickly than objects at the higher and lower levels. It seems that at the basic level there is maximal, within-category similarity relative to between-category similarity That is, categories that are similar are grouped together in a way that sharpens their differences from other categories.

Theoretically, these organisational properties are proposed to reflect a balance between the principles of informativeness and cognitive economy. The basic-level categories (like chair) are noted by a balance between informativeness (the number of attributes the concept conveys) and economy (a sort of summary of the important attributes that distinguish it from other categories). Informativeness is lacking at the highest level because few attributes are conveyed, and economy is missing at the lowest level because too many attributes are conveyed.

However, it is important to note that basic-level concepts do not always correspond to intermediate terms (e.g., chair in furniture-chair-armchair). In non-biological categories (like furniture) the intermediate term tends to correspond to the basic level. However, in biological categories the superordinate term tends to correspond to the basic level (e.g., "bird", in bird-sparrow-song-sparrow). This difference is seen as being a function of the amount of experience people have with members of biological categories. That is, one's experience with the instances of a category will lead to differences in one's basic level. So, ornithologists would be more likely to consider sparrow to be the basic level for the bird category because, given their expertise, this is the most distinctive level. Similarly, Berlin's findings with the Tzeltal probably reflects their expertise concerning the differences between trees (but see Atran, 1998; see also later section on neuropsychological evidence).

Using categories for prediction

It is only relatively recently that empirical research on categorisation turned to the arguably more ecologically valid task of prediction (e.g., Corter & Gluck, 1992; Heit, 1992; Lin & Murphy, 1997; Malt, Murphy, & Ross, 1994; Murphy & Ross, 1994; Ross & Murphy, 1996, 1999; Waxman & Markow, 1995; for earlier pieces see Markman, 1989; Rips, 1975; Smith & Medin, 1981). Murphy and Ross (1994) pointed out that categorisation by itself is not very useful; people do not classify things for the sake of classifying them, they classifying things to make predictions about those things. For instance, having decided that a certain object is a dog, you can predict that it might bite, a prediction that would not follow if one had classified it as a cat. This phenomenon is often called inductive inference from categories (see Chapter 15 for more on induction).

Heit (1992) examined how people make predictions from learned instances or from instances that were similar to learned instances (see also Anderson, 1991; Osherson et al., 1990). His subjects memorised a description of 30 individuals who had three potential traits (e.g., Larry is a Jet and liberal, Harry is a Shark and married, Ben is a Jet and unathletic; where Jets and Sharks are clubs; see also later section on similarity). The subjects learned only one trait of a given individual but were told that each individual had two other traits. They were then asked to guess the probability (on a scale of 0 to 100) that a given individual had a proposed trait (e.g., whether Larry was likely to be single). Heit's results showed that people could make one-step and two-step inferences about these unseen traits. In a one-step inference, they inferred a trait based on the similarity of the given individual to other individuals with similar features; so, if one was asked whether Larry was likely to be unathletic, and had been told that Ben and Bill, also members of the Jets, were unathletic, then you might infer that there was a high probability that Larry was unathletic. In the more complex two-step inferences, Larry might remind you for other reasons of Ben and

Bill who, in turn, might remind you of Harry and it is Harry's features that are used to make the inference about Larry This study shows something of the potential complexity of prediction from category instances.

Other work in this area has examined how predictions are made when there is some uncertainty about the classification of an object; for example, where a far-away object seems to be a dog and may therefore bite (Murphy & Ross, 1994; Ross & Murphy, 1996). A simple model might predict that the object has a feature based on how often that feature occurs in the category (i.e., the inference makes use of a single category); if in your experience 75% of the dogs you know bite, then the probability that a given dog will bite is .75. However, a more complex proposal suggests that if there is uncertainty about the classification then this probability would have to be modified in some way (see Anderson's, 1990, 1991, 1996, rational model). A further proposal made has been that the likelihood of being bitten should be modified by the likelihoods of biting one knows about for other animal categories (i.e., the inference makes use of multiple categories). In short, the inference is sensitive to the base rates for biting in the dog and other animal categories (see Chapter 17 on people's treatment of base rates in judgement and decision making). Overall, in an extended series of experiments Murphy and Ross found little evidence for the use of multiple categories in a prediction task, but found that people made use of a single category. Furthermore, this result was the case irrespective of the uncertainty of the initial classification.

The instability of concepts

It is commonly assumed in theories of concepts that the representations of concepts are relatively static, but Barsalou (1987, 1989) argues convincingly that this assumption may be unwarranted, that concepts are unstable. He points out that the way people represent a concept changes as a function of the context in which it appears. So, for example, when people read "frog" in isolation, "eaten by humans" typically remains inactive in memory. However, "eaten by humans" becomes active when reading about frogs in a French restaurant. Thus, concepts are unstable to the extent that different information is incorporated into the representation of a concept in different situations (see also Anderson & Ortony, 1975). It seems that only a subset of the knowledge about a category becomes active in a given context; what Barsalou (1982) calls context-dependent information.

Instability has also been found in the graded structure of category exemplars (see Barsalou, 1985, 1989). As we saw earlier, a category's graded structure is simply the ordering of its exemplars from most to least typical. For instance, in the bird category American subjects order the following instances as decreasing in typicality from robin to pigeon to parrot to ostrich. Instability shows itself in the rearrangement of this ordering as a function of the population, the individual, or context (see Barsalou, 1989). For example, even though Americans consider a robin to be more typical than a swan, they treat a swan as being more typical than robin when they are asked to take the viewpoint of the average Chinese citizen.

Furthermore, some categories are not well established in memory but seem to be formed on-the-fly (Barsalou, 1983). These, so-called ad hoc categories, are constructed by people to achieve certain goals. For example, if you wanted to sell off your unwanted possessions you might construct a category of "things to sell at a garage sale". Barsalou has shown that the associations between instances of these concepts and the concept itself are not well established in memory but can be constructed if required (for more recent work see Ross & Murphy, 1999 and Chapter 9).

Making Money With Garage Sales

Making Money With Garage Sales

Is your home bursting at the seams with stuff? Is every closet crammed so full that any one of them is a death trap waiting to be opened? Has it been years since the last time you parked the car in the garage? Never fear, help is on the way. You need to get rid of some of that stuff. Dont you dare call it junk. Remember, one man or womans trash is another ones treasure!

Get My Free Ebook


Post a comment