lecture 8 - concepts and categories Flashcards

Question

borderline cases

Answer 1

if membership in a category is ‘all-or-none’ then there should be no borderline cases: an item should either satisfy the definition of a category or it should not. - McCloskey and Glucksberg (1978) provided evidence that confirmed this intuition for a whole range of categories. They used a method of asking for categorization judgements. They asked their participants to respond either ‘Yes’ or ‘No’ to questions of category membership (such as ‘Is a robin a bird?’). Participants were also asked to rate the same instances for typicality. McCloskey and Glucksberg then considered the level of agreement that participants showed in their categorization judgements, both across individuals and within the same individuals over two times of testing. They found that participants readily agreed on highly typical and atypical items, yet disagreed over time and across individuals for some items of intermediate typicality. For example, people rated ‘chair’ as a highly typical item of ‘furniture’, and were consistent amongst themselves and over time in judging a chair to be an item of furniture. Similarly, with highly atypical items such as a ceiling they were consistent in judging this not to be an example of furniture.With items of intermediate typicality, such as bookends, they were much less consistent. McCloskey and Glucksberg thus gave empirical weight to the intuition that many categories have borderline cases. - How telling is this evidence? The classical view certainly implies that categories should have no borderline cases. However, it is at least possible that some of the instances that appear borderline are not genuinely indeterminate, unlike the case of colour categories. It might be that patterns of disagreement reveal a lack of knowledge. It is possible that an instance definitely belongs to one or another category (i.e. not a borderline), but uncertainty makes -the item appear borderline Another possibility is that inconsistency reflects perspective- dependence it is not obvious that McCloskey and Glucksberg’s examples actually did involve uncertainty or perspective- dependence. the compelling evidence for borderline cases seems to undermine the classical view.

Answer 2

- A further source of difficulties for the classical view has been the observation of intransitivity in categorization judgements. Transitivity is observed with many relationships: The relationship is ‘transitive’ because the last statement follows from the first two. Is categorization transitive? That is, if As are members of category B, and Bs are members of category C, does it follow that As are also members of category C? According to the classical view it does (and perhaps your intuition agrees). Th e classical view holds that membership in a category is all or none – if an instance falls into a category, it does so unequivocally. There can be no exceptions. So it should just follow, unequivocally, that rabbits must also be animals. Hampton (1982), however, showed that people’s categorization judgements are not in general consistent with transitivity. For example, he found that participants would agree that ‘car seats are a kind of chair’ and that ‘chairs are a kind of furniture’ but not agree that ‘car seats are a kind of furniture’. Similarly, people might agree that Big Ben is a clock and that clocks are furniture, but not that that Big Ben is an item of furniture. The fact that people strongly reject the transitive inference in these cases represents a real problem for the classical view.

Answer 3

- Wittgenstein (1953), in developing his account of language-games, considered the idea, as implied by the classical view, that there are common properties to all instances of the category of game - If Wittgenstein is right, then the classical view is simply mistaken. Whereas it contends that categories have common properties, Wittgenstein’s position is that most categories are like ‘game’ – when you look closely for common properties, you find none. - Wittgenstein suggests that most categories are really indefinable. Wittgenstein has not proved that natural categories cannot be defined, and so it is possible that someone might yet provide definitions. But the philosophers Kripke (1972) and Putnam (1975) undermined even that idea. They considered what would happen if something that was taken to be ‘definitional’ was later found to be wrong. Th e critical issue is what would be the implications of such a ‘discovery’. The conclusion that Kripke and Putnam draw is that we might be shown to be wrong about virtually any property that we happen to believe is true (or even defining) of a category. If so, then our beliefs about natural categories never really amount to definitions and the classical view must be mistaken.

Answer 4

concepts are organized around a measure of the central tendency of a category, otherwise known as the prototype. Sometimes the proto- type may correspond to an actual instance, but in general it is like a ‘best’ category member, formed by statistically aggregating over examples of the category one encounters. - Rosch, for instance, believed that it is a feature of the natural world that certain attributes or properties tend to correlate or cluster together, and it is these natural clusters of correlated attributes that prototypes describe. Th ese properties cluster together in a way that feathers, lips, gills, and an ability to swing through tree branches do not. Whether or not an instance is a category member then depends upon how similar it is to the prototype: an instance falls within the category if it achieves a certain criterion of similarity. If an instance is too dissimilar, if it mismatches on too many properties, then it falls outside the category. Th is account is a little like the classical view: both are committed to the idea that similarity explains categorization. For classical theory, instances fall within a category if they match each and every element of the category’s definition, and outside the category if they mismatch on any one. The critical difference is that for prototype theories an instance falls within a category even if it mismatches on a number of properties. - although prototype theories could be thought of as having merely relaxed the classical view’s criteria for category membership, the upshot is that prototype theory might be able to explain category membership for the many categories that resist definition Prototype theories have been formulated in different ways. One of the earlier formulations was provided by Smith et al. (1988), and this captures many of the qualities found in different versions. - First, there are multiple possible values for each attribute, capturing the fact that no one value is necessary for category membership – apples are typically, but not necessarily, red, for example. - Second, diagnosticities indicate the extent to which each attribute is important for deciding category membership. - Third, the values are weighted and these weights indicate the extent to which each value contributes to typicality; the highest weighted values are those of the prototype. - Categorization depends upon achieving a criterion similarity with the representation of the concept, one that depends on matching properties as before, but now diagnosticities and weights enter into the computation as well (though we don’t need to go into detail). - Much research continues to be devoted to making formal models of the prototype view and testing these against empirical data (e.g. Minda and Smith, 2011). - Prototype theories can readily explain the typicality effects discovered by Rosch and her co-workers. - 1. Instances that differ in typicality are assumed to differ in terms of the weighting of values on which they match the concept. For example, in Table 5.3, a difference in typicality between red and brown apples is reflected in a difference in the weighting for red and brown. - 2. Sentences such as ‘a robin is a bird’ are likely to be verified more quickly than ‘a penguin is a bird’ because, for high typicality instances, the criterial similarity required for verifying the sentence is likely to be achieved after matching just a few properties. This is because most attributes that match will have higher-weighted values, and so any criterion for category membership will be reached quickly. For low typicality instances like penguin, many attributes will mismatch or will have low weighted values, and so more matches will have to be made before the criterion is reached. - 3. Typicality is likely to correlate with how widely category members share attributes. This follows from the fact that the diagnosticities of attributes and weights of values themselves reflect the statistical distribution of those attributes and values. Th e more widely shared a value is, the greater is its weight. In Table 5.3, for example, ‘round shape’ receives a high weight indicating that many (many) more apples are round than square. Since high typicality instances tend to match on high weighted values, it follows that they will also possess properties that are widely shared. - However, despite prototype theory being able to accommodate many of the findings that undermined the classical view, difficulties have emerged.

Answer 5

- Armstrong et al. (1983) considered whether typicality effects occur for concepts that appear to be definitional. Their examples of definitional concepts included ‘female’, ‘plane geometric figure’, ‘odd number’, and ‘even number’ - Armstrong et al. believed that category membership for these concept is determined not by similarity to a prototype but by a definition: whether a number is even depends on whether dividing it by 2 yields an integer. Curiously, however, they found a range of robust typicality effects (summarized in Table 5.4), implying that even these apparently definitional concepts have an internal structure; these effects were also found using the sentence verification task. they argued that the existence of typicality effects should not be taken as conclusive evidence that category membership is determined by similarity to a prototype. They proposed instead a dual-process model, in which concepts possess a ‘core’ that is used when we judge category membership and a set of identification procedures that we use to identify instances of a category on particular occasions (often rapidly). Armstrong et al. suggested that the classical view might explain the concept’s core, while prototype theory explains identification procedures. Unfortunately, inasmuch as this proposal involves both theoretical approaches, it appears to inherit some of the problems faced by each.

Answer 6

typicality effects change with context Roth and Shoben (1983) showed that typicality effects are changed by linguistic context. Medin and Shoben (1988) also found that typicality judgements change with context. Medin and Shoben found that small metal spoons were more typical than large metal spoons, they found that large wooden spoons were more typical than small wooden spoons. So, the contribution to typicality made by the values large and small depended on whether one was thinking about metal spoons or wooden spoons. Prototype theories cannot easily explain such demonstrations of the instability of typicality. - First, the very idea of instability seems to be at odds with Rosch’s claim that prototypes correspond to stable clusters of correlated properties that reflect the structure of the natural world. - Second, in connection with Table 5.3, Roth and Shoben’s results suggest that the weightings of values and/or diagnosticities of attributes are themselves changeable. However, it is unclear what mechanism could be responsible for such changes. Third, Medin and Shoben’s results suggest that the contributions to typicality of diff erent properties (e.g. size and material made from) are mutually dependent.

Answer 7

- Researchers have tried to explain the meanings of phrases and larger linguistic units in terms of com- plex concepts, i.e. combinations of lexical concepts. The meaning of the phrase ‘red car’ would then be explained in terms of the combination of the constituent lexical concepts ‘red’ and ‘car’. How could concepts combine to yield the meaning of such a phrase? - If concepts are structured around prototypes, then perhaps they could combine through combining their prototypes. The difficulty, however, is that no one really knows how this might be done. Though many suggestions have been made, they all appear to fail for one reason or other. For example, one suggestion has been that the prototype for ‘red car’ is formed from the prototype for ‘red’ and the prototype for ‘car’ (the prototypical red car would therefore be a prototypical car that was prototypically red). Though this seems a sensible suggestion, and appears to give the right interpretation for ‘red car’, this could not work in general. Following the same reasoning, the prototypical ‘pet fish’ ought to be a prototypical fish that is also prototypically pet-like – perhaps something like a cuddly salmon. Th e real proto-typical ‘pet fish’ is of course more like a goldfish – neither a prototypical pet nor a prototypical fi sh. More problematic still for combining prototypes, the prototypical ‘stone lion’ ought to be something like a real lion made of stone, i.e. an impossible object. How could the prototypes for ‘stone’ (perhaps granite or limestone) and ‘lion’ (a real lion) combine to give the right interpretation (i.e. a stone statue of a lion)? - How would these prototypes combine to yield the required interpretation? Complex concepts continue to present real difficulties for most theories of concepts (see Fodor, 1998).

Answer 8

notion of similarity. concepts are instead representations of the individual members (or exemplars) of the category that we have experienced (Medin and Schaffer, 1978). - Brooks et al. (1991) examined clinicians’ ability to diagnose (or categorize) new examples of skin disorders. Those examples that were similar to previously studied cases were categorized more accurately than those that were dissimilar. The conclusion that Brooks et al. draw is that categorization is influenced both by explicit rules and by similarity to other exemplars. - A similar finding was reported by Allen and Brooks (1991) using artificial categories of cartoon pictures of animals. Even though participants were explicitly told the categorization rule, they were more accurate and faster in categorizing items similar to previous exemplars. - Allen and Brooks believed reliance on exemplar-level information was likely to happen only in certain circumstances, but Hahn et al. (2010) have argued that such effects are more pervasive. - Demonstrations such as these suggest that exemplar-level information is very difficult to ignore in categorization tasks, and that people categorize items through reference to two sources of information: summary, category-level information and information about exemplars. Just as with other similarity-based views, it has proven possible to develop sophisticated models of categorization that have had much success in accounting for empirical data (see, for example, Nosofsky, 2011)

Answer 9

problems these theories have encountered have led researchers first to question the importance of similarity, and second to propose that categorization involves much larger knowledge structures, called theories (or commonsense theories to distinguish them from scientific ones). The approach has become known as the ‘theory’-theory of concepts. - similarity-based accounts have achieved considerable success and remain popular. - Hampton (1998) conveys some sense of this. Using McCloskey and Glucksberg’s (1978) data (both typicality ratings and categorization judgements were collected), he examined whether the probability of an item being judged a category member could be predicted from its typicality (reflecting its similarity to a prototype). - Focusing on just the borderline cases, Hampton showed that typicality was a very good predictor, explaining somewhere between 46 per cent and 96 per cent of the variance in categorization probability. So, regardless of the difficulties facing similarity- based accounts, similarity (as measured by typicality) seems to be a good indicator of categorization. - Nonetheless, Hampton found other predictors of categorization probability (though none was as good a predictor as typicality). These included lack of familiarity; the extent to which an instance was judged ‘only technically speaking a member’ of a category (e.g. a dolphin is technically speaking a mammal but superficially appears more similar to fi sh); and the extent to which participants judged an instance was ‘technically speaking not a member’ (e.g. a bat is technically speaking not a bird despite superficially appearing more similar to birds than to mammals). - That these last two factors were predictors suggests that categorization draws upon deeper, more theoretical knowledge than just similarity alone. We now turn to some of the reasons why, in spite of these successes, many researchers have become dissatisfied with the notion of similarity.

Answer 10

- The philosopher Nelson Goodman identified a number of problems with similarity; indeed, he described it as ‘a pretender, an impostor, a quack’ (Goodman, 1972, p.437). - One concern is with whether similarity genuinely helps us to explain categorization. Aft er all, in prototype theories saying that an instance is similar to the prototype means that the two share some properties in common. But note that this further explication removes the notion of similarity: ‘is similar to’ becomes translated as ‘shares properties with’. So, what explains categorization is not similarity per se but the sharing of properties. - However, a further problem arises since there is no obvious limit to the number of properties any two objects may share. - Murphy and Medin (1985) ask us to consider the similarity of plums and lawnmowers. ‘You might say these have little in common, but of course both weigh less than 10,000 kg (and less than 10,001 kg, . . .), both did not exist 10,000,000 years ago (and 10,000,001 years ago, . . .), both cannot hear well, both can be dropped, both take up space, and so on’ (p.292). It seems that, depending on what counts as a relevant property, plums and lawnmowers could either be seen as very dissimilar, or very similar. So, for similarity, explicated in terms of shared properties, to provide meaningful explanations of categorization, we need to know what counts as a property. We need some way of declaring ‘lack of hearing ability’ as irrelevant in comparing plums and lawnmowers, for example. For Murphy and Medin (1985), observations such as these suggest that similarity is shorthand for something else that explains why categories hang together, or cohere.

Answer 11

In opposition to similarity-based views, Murphy and Medin argued that concepts are explanation-based,that there is some explanatory principle or theory that unites the category. For Murphy and Medin, relationships between attributes are evidence that our concepts are embedded in larger and broader knowledge structures. Sometimes these structures have been labelled commonsense theories, sometimes merely background knowledge. - Murphy and Medin speculate that many categorization judgements become automatized, particularly when members of the same category have relatively consistent perceptual properties. Under these conditions, the role of our underlying theories becomes obscured, and so we may (erroneously) conclude that categorization is determined by similarity. However, even in these cases, when novel instances emerge or where there is disagreement (with borderlines perhaps), we turn to our underlying theories. - What evidence is there that categorization is determined by theories as opposed to similarity? Rips (1989) asked his participants to consider triads of objects. Two objects belonged to distinct categories (e.g. a pizza and a quarter) and were chosen so that participants’ largest estimate of the size of one category (quarter) was smaller than their smallest estimate of the other (pizza). Rips then asked his participants to consider a third object, telling them only that it was of intermediate size (i.e. larger than the largest estimated size of a quarter and smaller than the smallest estimated size of a pizza). He asked which of the two other categories this third object was more likely to belong to, and which of the two it was most similar to. Th e two judgements dis- sociated: that is, participants judged the object more likely to be a pizza, but more similar to a quarter. - Rips also found dissociations when participants considered, for example, a bird-like creature that came (via mutation or maturation) to resemble an insect, and other dissociations between categorization and similarity have been demonstrated (e.g. Rips and Collins, 1993; Roberson et al., 1999). - Kroska and Goldstone (1996) showed their participants scenarios that described a putative emotion. Each scenario constituted a set of phrases such that one phrase was central to one emotion and other phrases were characteristic of a different emotion. For example, one scenario included the phrases ‘Threat of harm or death’, ‘Being accepted, belonging’, and ‘Experiencing highly pleasurable stimuli or sensations’. The first of these phrases was considered central to the emotion category ‘fear’. Th e remaining two phrases were considered characteristic of the emotion category ‘joy’. - Kroska and Goldstone found their participants tended to categorize this scenario as an instance of fear (i.e. a member of the category ‘fear’) but they also judged it to be more similar to an instance of joy. That is, judgements of category membership were influenced by properties considered central to a category, while judgements of similarity were influenced by characteristic properties. Again, these findings show that judgements of category membership can dissociate from judgements of similarity. It seems that there are deeper reasons for people’s categorizations – in the quarter example, perhaps they realized that pizzas can, in principle, be any size, whereas their commonsense theories of coins tell them they are produced to a regulation standard

Answer 12

- The theory-theory has proved an important and useful way of thinking about concepts. It has, for instance, reminded researchers of difficulties with the notion of similarity, and it has proved to be a useful peg on which to hang a range of disparate findings whose common theme is that categorization is influenced by deeper, causal knowledge of categories, as well as by knowledge of their superficial properties. - However, there are a number of difficulties with the theory-theory. Some of the findings taken to support the theory-theory are really demonstrations that similarity does not always explain categorization, and this does not necessarily imply that theories are what is needed. - Moreover, it is not clear what is meant by ‘theory’. Whereas similarity-based views could be made relatively precise (see Table 5.3 for instance), formalizing theory-theories seems much more difficult. Some researchers have tried to pin down what is meant by a commonsense theory via a comparison with scientific theories (see Gopnik, 1996). - However, other researchers believe such a comparison undermines the idea that commonsense theories are theories at all (see Gellatly, 1997). For example, Murphy (2000) argues that the background knowledge that influences concepts is too simplistic and mundane to be likened to a scientific theory. Indeed, he eschews the term ‘theory’ in favour of the more neutral ‘knowledge’. - A further difficulty with the theory-theory is that it is hard to imagine how combining theories could explain complex concepts. Scientific theories are notoriously difficult to combine. Indeed, for decades, theoretical physicists have struggled to combine theories of electricity, magnetism, and gravity into one unified theory. - So how can theories be combined so effortlessly in understanding phrases like ‘red car’ when they are so difficult to combine in general? - Even if we talk of combining knowledge rather than theories, we are still left with the difficult problem of working out which knowledge gets combined and the mechanism by which this is done. Given these problems, it is ironic that the theory-based view is motivated in part by difficulties with the notion of similarity. Arguably, it has supplanted this with the equally mysterious notion of a ‘theory’.

Answer 13

- Psychological essentialism is one attempt at more precisely formulating the view that categorization is influenced by deeper, explanatory principles. Medin (1989) and Medin and Ortony (1989) suggested that people believe and act as though category members have certain essential properties in common. That is, people categorize things according to their beliefs about essential properties. They may also believe that the essential properties constrain a category’s more superficial properties. For example, the essential properties of birds might be thought to involve their genetic make-up, properties that would constrain their appearance and behaviour. - Essential properties can be characterized as those properties that if an object did not possess them, it would not be that object. Th e essential properties of birds are properties that all birds necessarily possess; if something doesn’t possess them, then it isn’t a bird. Essential properties may seem rather like the defining properties of the classical view. However, there is one critical difference. According to psychological essentialism most people will not know what a category’s essential properties are, but will still believe that the category has some. We might speculate as to what the essential properties are – perhaps for biological categories they would be genetic properties – but, in general, our beliefs will be vague and may turn out to be incorrect. So psychological essentialism proposes that people’s concepts may contain a ‘place-holder’ for an essence – and the place-holder may even be empty, reflecting a lack of knowledge as to what the essential properties might be. Of course not everyone’s place-holder need be empty. Indeed, it is usually thought that discovering essential properties is a job for science. A metallurgist or chemist, perhaps, might uncover the essential properties of gold, just as a biologist might for birds. So, experts may have their place-holders partially or completely filled – they may know (or think they know) the essential properties. But these beliefs may turn out to be in error too, so the place-holder is presumably capable of revision. - Psychological essentialism is consistent with much of the evidence supporting the theory-theory. Much evidence supporting psychological essentialism specifically has come from studies of the development of categorization. For instance, Gelman and Wellman (1991) found that 4- and 5-year-old children believe the insides of objects to be more important than their outsides in determining category membership. For example, they asked children whether a dog would still be a dog if its outsides were removed, and also if its insides were removed. Children thought that instances would remain in the category if the outsides were removed, but not if their insides were removed. - According to Gelman and Wellman, children are being essentialist since they believe that something internal, something hidden and ‘inner’, is causally responsible for category membership. However, psychological essentialism has not gone unchallenged. Malt (1994) examined the concept of water. If people believe H2O to be the essence of water, then their categorization of liquids as water should be strongly influenced by the proportion of H2O those liquids contain. - However, Malt found that people’s categorizations were strongly influenced by the source of the water, its location, and its function. Indeed, pond water was thought to be ‘water’ but was judged to contain only 78.8 per cent H2O; tears were judged not to be ‘water’ but to contain 88.6 per cent H2O. So the belief in the presence or absence of H2O was not the only factor in deciding membership in the cate- gory ‘water’. Kripke (1972) and Putnam (1975). For example, Putnam argued that even if we discovered that all cats are robots controlled from Mars, they would still be cats. What we didn’t note there is that they used thought experiments such as this to support essentialism. - Braisby et al. (1996) subjected these to an empirical test. They asked participants to give categorization judgements in thought experiments such as Putnam’s robot cat. In one condition they were told ‘You have a female pet cat named Tibby. For many years people have assumed cats to be mammals. However, scientists have recently discovered that they are all, in fact, robots controlled from Mars. Upon close examination, you discover that Tibby too is a robot, just as the scientists suggest.’ Participants were then asked to indicate whether they thought that a series of statements were true or false. These included statements expressing essentialist intuitions (e.g. ‘Tibby is a cat, though we were wrong about her being a mammal’) and statements that expressed the contrary intuition (e.g. ‘Tibby is not a cat, though she is a robot controlled from Mars’). Only about half of the participants thought that the essentialist statements were true and the contrary ones false. Moreover, many participants seemed to give contradictory judgements: they either judged both statements to be true, or judged both to be false. - Braisby et al. argued that these findings did not support essentialism, but implied that concepts change their content according to context and perspective (see Braisby and Franks, 1997). - Jylkkä et al. (2009) empirically re-examined the Braisby et al. data and came to a diff erent conclusion from the original authors. Th ey suggested that people’s intuitions offer support for essentialism. However, their participants’ responses were actually quite varied, and the greater proportion of responses were not consistent with essentialism. One way of interpreting this debate is that both studies suggest essentialism may accurately characterize some people’s categorizations, but that an alternative account of concepts is required to explain categorization in general. - Hampton et al. (2007) re-examined some of the dissociations reported by Rips (1989) and mentioned in Section 2.4.2. They found that when key manipulations were presented as between-subject rather than within-subject, the proportions of essentialist responses fell, and overall the majority of participants adopted responses inconsistent with essentialism. They also found important individual differences, in that some participants consistently gave essentialist categorizations, whilst others gave more importance to appearance and behaviour. - Hampton et al. speculate whether such individual differences may reflect more general and stable cognitive styles. While the argument for essentialism was originally developed to explain natural concepts (e.g. cat, water), researchers have begun to consider whether the theory applies in other domains also. For example, Bloom (1996) proposed that artefacts such as torches are created with a specific purpose in mind, and that the creator’s intention constitutes the essence of the category. - However, Malt and Sloman (2007) disputed this account, demonstrating that people often describe artefacts according to their function or use. For example, their participants tended to describe a teakettle as a watering can if it was used as a watering can, even though it was created with the intention that it should heat water. (See also Bloom, 2007, for counterarguments.) - Essentialism has also been applied to social categories. For example, Haslam and Whelan (2008) have proposed that psychological essentialism plays a central role in the categorization of race and ethnicity, gender, mental disorder, personality, and sexual orientation. - Haslam et al. (2000) developed essentialist belief scales to examine essentialism thinking in relation to categories, and argued that this centred on two key dimensions: the extent to which a category was seen as a natural kind, and the extent to which the category was seen as coherent and homogeneous. - Haslam et al. (2002) have also discovered links between such dimensions and prejudice, and Haslam and Levy (2006) revealed a similar structure to people’s thought concerning homosexuality. However, such studies largely measure people’s beliefs about categories and not categorization per se. - Examining people’s categorization of sexual orientation, Braisby and Hodges (2009) failed to find signs of essentialism, with participants categorizing sexual orientation according to described behaviour and giving little weight to putative essential properties such as genetics. - There has also been mixed evidence concerning the role that expert opinion plays in categorization. Malt (1990) presented people with objects that they were told appeared ‘halfway’ between two categories (e.g. a tree halfway between an oak and a maple) and asked them to indicate how they would solve the dilemma of categorizing the object. She offered her participants three options. They could ‘ask an expert’, ‘call it which- ever you want’, or indicate that they could ‘tell which it is’ if they could only think about it long enough. For pairs of natural categories such as ‘robin-sparrow’ and ‘trout-bass’, 75 per cent of participants suggested they would ask an expert, whereas for pairs of artefact categories, such as ‘boat-ship’, 63 per cent of participants suggested it was possible to ‘call it whichever you want’. - This evidence suggests that people may be psychologically essentialist for natural categories, at least to some degree, because they recognize that experts may be in a better position to judge categorization when lay people cannot. However, the data overall are not conclusive. - Braisby (2001) examined the extent to which people modify their categorization judgements for genetically-modified biological categories when told the opinions of experts. For example, his participants might be asked to consider a genetically modified salmon, and were told either that expert biologists had judged that it was a salmon or that they had judged that it was not. He found that only around half of the participants changed their categorization judgements to conform to the judgements of the biologists. - Moreover, around a quarter of participants would change their categorization judgements to conform to those of shoppers (i.e. a group presumed not to be expert with respect to the category’s essential properties). - Braisby argued that only around a quarter of participants were modifying their categorization judgements because of the biologists’ expertise with the relevant essential properties, and so the majority of responses did not provide evidence for psychological essentialism. - Indeed, participants seemed to base their judgements on non-essential properties such as appearance and function (as well as genetic make-up). - Braisby and Hanlon (2010) also found that the extent to which people would defer to experts in categorization tasks was related to a measure of their cognitive style. - Witkin et al. (1962) coined the term ‘field dependence’ to refer to a distinction in cognitive style that has been variously labelled as analytic-holistic. - Braisby and Hanlon found that 89 per cent of field dependent participants deferred in their categorization, as opposed to just 23 per cent of fi eld independent participants. Such a result appears to support the suggestion made by Hampton et al. - Lastly, it should be noted that much of the evidence cited in support of psychological essentialism (e.g. Gelman and Wellman, 1991) only indirectly relates to beliefs in essential properties. Gelman and Wellman, for example, found that children thought that removing the outsides from something like a dog did not alter its category membership, but removing its insides did. - However, for these data to support essentialism, a further inference is required, one that relates insides to essences. In a similar vein, Strevens (2000) actually argues that the notion of essence or essential properties is not required to explain empirical data such as these. Of course, psychological essentialists have responded to some of these criticisms, so it seems fair to say that the arguments are not yet settled. However, some of the criticisms of other theories may also apply to psychological essentialism – how might it help us understand complex concepts, for example

Answer 14

The classical view falters because we cannot identify necessary and sufficient conditions of category membership for all but a very few concepts. Prototype theory has difficulties explaining context-sensitivity and complex concepts. Ultimately, both suff er for their use of the notion of similarity, which seems unable to explain categorization fully. Theory-based notions of concepts are imprecise and cannot obviously explain complex concepts. Lastly, psychological essentialism has received mixed empirical support, and much of the empirical evidence only indirectly relates to the notion of essences. we have probably learned more about the phenomena of categorization even as various theories have been found wanting. And, of course, adherents of those theories continue to introduce modifications in order to explain recalcitrant data. N

Answer 15

t categorization is not one single process. Maybe people categorize items in different ways in different circumstances. Indeed, discursive psychologists, whose approach we earlier bracketed off , might argue that categorization depends essentially on context, and that there is nothing common to all the cases that we call categorization. Were context to have such an unbridled influence we might expect categorization to appear unsystematic. Yet much of the evidence presented in this chapter points to the opposite – we have examined a wide range of empirical data that are highly robust. One way of reconciling the idea that people categorize things differently on different occasions with the idea that categorization is nonetheless systematic is to suggest that there are (a determinate number of) different kinds of categorization. it is conceivable that these could be usefully framed by the different theories of concepts. often we need to reach an agreement or adopt a convention as to whether something belongs to a category (e.g. whether a 16-year-old is a child or an adult). - Similarly, prototype theory may usefully explain categorization in circumstances where we need to categorize something rapidly, or perhaps under uncertainty, maybe when we are in a position to take into account only an object’s superficial properties. - Likewise, theory-based views may describe categorization when we are seeking a more reflective and considered judgement, perhaps when we are using categorization in order to explain something. And essentialism may usefully explain how we categorize when we wish to be consistent with expertise and a scientific knowledge of the world. - Machery (2005) makes a similar point. In his examination of complex concepts, he argues that during categorization people employ a range of information, including elements of commonsense theories, exemplar-level information, and prototypes. If this is right, then it would be simplistic to expect one of the theories we have examined to provide a comprehensive account of concepts and categorization. - Speculative though this possibility is, Smith and Sloman (1994) have provided some empirical evidence that there may be some truth to it. They sought to replicate some of the dissociations between similarity and categorization judgements obtained by Rips (1989) and described in Section 2.4.2. - Rips found that people judged an object intermediate in size between a quarter and a pizza to be more similar to a quarter, but more likely to be a pizza. Smith and Sloman obtained the same dissociation only when participants were required to think aloud whilst making their decisions and so articulate reasons for their judgements - Smith and Sloman interpreted this finding as pointing to two modes of categorization: a similarity-based mode of categorization and a rule-based mode. The implication is that people will either focus on similarity or on underlying rules or structure depending on how the categorization task is presented. When in similarity-based mode, categorization seems to conform to similarity-based accounts, such as prototype theories. When in rule-based mode, categorization seems to be more theory- or explanation-based. - Though this does not show that there are as many different ways of categorizing as there are theories of concepts, it does suggest that categorization may not be a single process. It is a possibility, therefore, that some of the different accounts of concepts may be implicitly concerned with different kinds of purpose in categorization, and ultimately with different kinds of categorization. - In a similar vein we can rethink the phenomena that are taken as evidence of the nature of concepts. Earlier we noted that concepts and words bear a complex relationship to one another, but much of the evidence we have so far reviewed has tended to equate the use of category words with categorization. However, while our use of category labels is certainly influenced by our beliefs about categorization, it is also influenced by language more generally. Indeed, we can label something with a category word yet not believe that it belongs to the category – describing a statue of a lion as a ‘lion’, for example, does not indicate that we think the statue really is a lion. - Malt et al. (1999) showed that the same is true for how we label containers, such as ‘box’, ‘bottle’, and ‘jar’. Th ey found that whether an item was called a ‘bottle’ depended not so much on how similar it was to a prototypical bottle, but whether there was something similar that was also called a ‘bottle’. In this way, for example, a shampoo container might get called a shampoo ‘bottle’ despite bearing little similarity to a prototypical bottle. So, whether we apply a category label (e.g. bottle) to an object depends in part on how that label has been used historically and only in part on whether we think that the object really belongs to the labelled category

Answer 16

extent to which different types of concept require a different theoretical treatment. Some categories like even number seem amenable to definition. For these well-defined categories, the classical view appears to give a good explanation of category membership, though it does not obviously explain how some even numbers are considered more typical than others. Perhaps this would require some- thing like Armstrong et al.’s dual-process account, and involve its attendant difficulties a modified classical view would provide a good explanation of these kinds of category - In a similar vein, prototype theories seem to work well for fuzzy categories, categories that seem to have genuine borderline cases. For these similarity to a prototype might provide the best explanation of category membership, since there is no prospect of defining these categories and people in general do not seem to have relevant commonsense theories - Theory-based and essentialist approaches are likely to be most successful for categories for which people have commonsense theories. Perhaps unsurprisingly, these include many categories for which scientific theories have also been developed, e.g. sparrow and gold. These are categories where it is relevant to develop a deeper, explanatory knowledge of the causal principles underlying the category. As we have seen, however, researchers have extended the claims of essentialism to include other kinds of category, such as artefact and social categories, e.g. introvert - While it remains only a possibility that different concepts may require different theoretical treatments, there is evidence nevertheless that different kinds of concepts may be realized in the brain in different ways (e.g. Warrington and Shallice, 1984; Strnad et al., 2011; though see also Tyler et al., 2003). - This evidence may yet challenge the view that one theory could accommodate all kinds of concept. Even if people were to conclude that different categories require different theoretical treatments, it would still be important to find some way of relating the different theories so we could understand in what sense they were all theories of concepts.

Answer 17

onsider whether all categorizers are the same. - Medin et al. (1997) recruited participants from three occupational groups with correspondingly different experience and knowledge of trees: maintenance workers, landscapers, and taxonomists. They then asked them to sort the names of 48 different kinds of tree into whatever groups made sense. - The taxonomists tended to reproduce a scientific way of sorting the trees; the maintenance workers produced a similar sorting, although they gave more emphasis to superficial characteristics (such as whether trees were broad-leaved). They also tended to include a ‘weed tree’ group that was not present in the taxonomists’ sorts, and which included trees that cause particular maintenance problems. Th e landscapers didn’t reproduce a scientifc taxonomy, but justified their sorts in terms of factors such as landscape utility, size, and aesthetic value. - Lynch et al. (2000) also showed how the typicality ratings of the same kinds of tree expert differed from those of novices. Typicality for the expert group reflected similarity to ideals – i.e. trees judged to be best examples of the category were not of average or prototypical height, but of extreme height; in contrast, the ratings of the novices were largely influenced by familiarity. - Studies such as these suggest that different people do not necessarily categorize things in the same way. The goals that a person has as well as the extent of their knowledge may influence the way they categorize and, by extension, be reflected in their concepts. - It is also possible that categorization may differ in more fundamental ways for different kinds of person. For example, there may be deeper individual differences, possibly rooted in cognitive style (see, for example, the discussion of essentialism above). - There is also evidence for disruption of semantic processing in schizophrenia (Chen et al., 1994), and accumulating evidence that categorization is altered in autism (Coutinho et al., 2010), Alzheimer’s dementia (Doughty et al., 2009), and schizotypy (Morgan et al.,2009). Morgan et al.’s results are of special interest since, in this and similar studies, the sample of (relatively) high schizotypes are drawn from the normal, healthy population, typically students, and therefore differences among categorizers cannot easily be dismissed as manifestations of disordered cognition.

Answer 18

category knowledge is multi layered, encompassing knowledge of the causal properties relevant to a category, knowledge relevant to explaining category membership and the properties of instances, knowledge of function, and knowledge of superficial properties useful for identification and judgements about appearance. - It also seems that we are capable of calling on different kinds of category knowledge on different occasions and for different purposes. While these observations are not inconsistent with a single theoretical treatment of concepts, they nonetheless raise the prospect that competing theories provide good explanations of somewhat different sets of phenomena, and so are not directly in contradiction. Though categorization presents substantial challenges for researchers, these are challenges for all cognitive psychologists.