Natural and Artificial Kinds

Preparation of this paper was supported by a grant from the Graduate School of the University of Wisconsin and by a Spencer Foundation, National Academy of Education Post-Doctoral Fellowship. The author would like to thank the parents, children, and teachers at Eagle's Wing Childcare Center, The University of Wisconsin Lab School, and The Waisman Center Early Childhood Program for their participation. Portions of this research were presented at the 1995 meetings of the Society for Research in Child Development in Indianapolis. Requests for reprints should be sent to Charles Kalish, University of Wisconsin- Madison, Department of Educational Psychology, 1025 W. Johnson St., Madison, WI 53706 or by electronic mail to cwkalish@facstaff.wisc.edu

Phone: (608) 262-0840

Fax: (608) 262-0843

Abstract

Research in conceptual development has highlighted important differences between natural kind and artifact concepts. One interpretation of the distinction is that natural kinds are objective discoveries while artifactual kinds are invented conventions. Existing literature suggests children may have a bias to treat all categories as objective facts, as natural kinds. Four studies assessed whether children and adults saw categorization decisions as matters of fact or convention. Preschoolers treated basic level categories of animals and human artifacts as objective. At the super-ordinate level, animals were treated as more objective than human artifacts. Adults' judgments of objectivity were also explored. In general, adults' judgments were similar to children's. Both children and adults have reliable and differentiated intuitions regarding category objectivity. Results are discussed in terms of their implications for accounts of category naturalness.

Natural and Artifactual Kinds: Are Children Realists or Relativists About Categories? 9/12/96

Recent perspectives on conceptual development indicate that concepts vary in their attributes and developmental histories. One common distinction is that drawn between natural kind and artifact concepts. While there have been many accounts of what makes a kind natural, researchers seem to agree that there are important differences between natural and artifactual categories. For example, the two types of concepts differ in the kind and number of inferences they promote (Gelman & Coley, 1991; Markman, 1990), and in their mental representations (Keil, 1989).

The purpose of this paper is to offer an additional perspective on the difference between natural and artifactual kinds. Categories may be either objective facts (natural) or invented conventions (artifactual). This distinction goes back to the source of much of the current theorizing about natural kinds, Locke's Essay Concerning Human Understanding (1707/1961). In this work Locke distinguished between objective discovered categories ("substances," which have real essences) and artificial invented categories ("mixed modes," which have only nominal essences). A concept such as GEMSTONE is a result of "the workmanship of the understanding," someone decided the features of some stones were worth noting and coined the term.[1] What counts as a GEMSTONE (and even whether we recognize the kind) is conventional. The category was constructed, it is an artifact. In contrast, a concept such as GOLD represents our discovery that a certain sort or type of thing exists in nature. Independent of our interests or decisions all gold is the same kind of thing. Mill (1872/1973) and Whewell (1840/1967) both argue that natural kinds represent real distinctions "made by nature." More recently, Kripke's (1972) theory of reference also involves a realist view of natural kinds. Thus there is considerable precedence for interpreting the natural-artifactual distinction as relating to the status of categories as objective or invented.[2]

Philosophers are concerned with what really exists. Psychologists will be interested in what people think is real (see Medin & Ortony, 1989, for a similar distinction between philosophical and psychological essentialism). According to this perspective, natural kind concepts involve groupings (categories) which are thought to be objective, really existing. Artifactual concepts will be seen to be human/social inventions. One question for psychologists, then, is whether people actually make this distinction; do people have reliable intuitions regarding category objectivity? Assuming the intuitions are robust, a second question concerns the psychological significance of this distinction; are judgments of objectivity related to other attributes of categories? Finally, given that much of the interest in natural kinds comes from studies of cognitive development, it is important to ask both these questions about children's intuitions.

The remainder of this introduction discusses two accounts of beliefs about category naturalness and objectivity, along with evidence for possible developmental changes in these beliefs. Following this, four studies involving judgments of category objectivity are reported. These studies suggest that children and adults do distinguish between objective and invented categories. Further, objectivity judgments are related to other attributes of categories which have been said to characterize natural kinds and artifacts.

Two Accounts of Naturalness

The broadest conception of "natural kind" rests on the claim that the human mind is constructed to form some categories and not others (Keil, 1981; Quine, 1977). These accounts have suggested that there are structural principles defining good, coherent, or natural kinds. This seems to be the sense in which Rosch used "natural categories" (Rosch, 1973; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976); to refer to familiar categories as opposed to artificial groupings used in some concept learning tasks (e.g., Kendler & Kendler, 1962).[3] Rosch argued that the human mind was tuned to form categories based on patterns of correlated attributes; there were "natural" categorization principles. There have been many different accounts of the structural determinants of category naturalness, for example; similarity, feature salience, or even familiarity (Goodman, 1955).

Categories will vary in how well they accord with structural principles of categorization. For Rosch, some categories are particularly well-formed (basic level kinds, such as CAT or CHAIR). Because these categories are so "good" they may appear to be discoveries about the world; certain groupings and relationships "cry out to be named" (Brown, 1965). For example, it seems to be an objective truth that tabbies and siamese should be categorized together (as CATS). Categories which are less well structured may be seen as more arbitrary. We recognize the influence of human decision and invention in the sub-ordinate level distinction between OFFICE-CHAIR and EASY-CHAIR. Thus, one account of category naturalness relies on structural, domain-general principles. These same principles may determine judgments of category objectivity. The studies presented below assess whether objectivity is related to structural principles of categorization.

Another set of accounts suggest that it is domain-specific beliefs that make categories natural or not. Natural kinds are more "theory-laden" than other concepts; organized around underlying explanatory principles rather than observable features (Gelman & Coley, 1991; Markman, 1990). Citing Boyd's notion of "causal homeostasis," Keil (1989) describes natural kinds as groupings based on beliefs about shared causal interactions. Thus constructs which play important roles in theories are natural kinds. For example, MONEY is a natural kind within a theory of economics. Constructs which do not appear in a theory are artifacts. MONEY is not a natural kind within a theory of physics, because there are no physical laws that apply to all or only money (see, Fodor, 1975). Different theories (which focus on different types of causes and effects) will generate different groupings; natural kind-hood is relative to a theory.

Theory-laden categories may be thought to be objective. Commonsense theories define our ontology (Wellman, 1990); they tell us what exists. Thus if it is a central tenet of commonsense biology that animals are divided into species, that system of categories will seem objectively correct. Concepts which do not appear in theories will appear more arbitrary because they do not represent important similarities or distinctions in the world. A system of categories grouping animals by color is an invention, not a discovery. Knowledge-based accounts of naturalness suggests that people will have specific beliefs about category objectivity derived from their theories.

Different accounts of category naturalness suggest different sources of intuitions about objectivity: structural principles or knowledge base. Judgments of objectivity, in turn, may inform us about which account best characterizes category naturalness. This depends on beliefs about category objectivity being both varied and reliable. While intuition suggests this is a plausible assumption for adults, it is less clear whether children's beliefs about category objectivity will be informative. In fact, there are several reasons to believe that children have a bias to see all categories as objective. Depending on one's perspective this may imply that children's judgments of category objectivity are uninteresting and uninformative. Alternatively, it may be taken to suggest that children treat all categories as natural kinds.

Children's Beliefs About Objectivity

A bias to see categories as objective may be a consequence of an uncritical acceptance of the products of the categorization principles (e.g., similarity). For example, based on similarity, a piano might be classified as FURNITURE. Children may take this as a fact; pianos are FURNITURE. As adults, however, we are used to questioning the automatic or natural ways of categorizing things. We recognize that contexts may shift the way things are categorized (Lakoff, 1987). We are also familiar with conflicts over categorization and the need to change categorization schemes (e.g., seeing a piano at a concert). Neither children nor adults classify exclusively according to structural principles (Carey, 1985; Gelman & Markman, 1986), though development may proceed from fewer to more theory-based concepts (Keil, 1989). Where children do use structural principles to categorize, they may take those principles as informative about objectivity in a way that adults do not. Adults may see a lack of underlying theory as evidence that a category is arbitrary. Even in the absence of a theory, children may treat categorization schemes which accord with structural principles as objectively correct .

Children's theory-based reasoning may be another source of bias towards treating categories as objective. Rather than viewing children as moving from atheoretical towards more theoretical conceptions (Keil, 1989; Quine, 1977) it is possible to imagine just the opposite progression. Young children may expect more coherence and theoretical explanation for phenomena than do adults. For example, they tend to overestimate the value of categories for making inductive inferences (Gelman, 1988). Looking for underlying causal principles organizing experience is a very valuable heuristic. Children seem to rely on this approach too much, seeing more causality than really exists (Gelman & Kalish, 1993). This may lead to "categorical realism," the expectation that there is some reason or explanation behind all categories (Gelman & Kalish, 1993). While adults recognize that the explanations behind some categories involve conventions, children may expect deeper reasons. In part this may stem from their roles as learners of cultural conventions rather than as fully-fledged members of the community. Children have to accept cultural conventions about naming and categorizing as objective facts; conventions may appear "natural" and be expected to reflect the causal structure of the world. As adults we are equal participants in the determination of our linguistic and conceptual conventions (though we usually defer to experts; Burge, 1979). The reason we are bound to these conventions has more to do with cooperative agreements than with discoveries about the structure of the world.[4] If children are biased to see categories as reflecting important causal relationships they may tend to treat all categories as objective.

A different developmental trend may be the result of children's growing understanding of representational diversity (Flavell, 1988; Forguson & Gopnik, 1988). Young children are said not to understand the constructive nature of mental representation (Flavell, 1988; Wellman, 1990). Thus the products of the mind may be understood as objective; it is facts about the world that cause us to have the thoughts we do. This suggests that children would generally have difficulty understanding the idea of an invented convention. Beginning about 4-years of age children come to see a distinction between representations and reality (Wellman, 1990). However, this may lead them to further reify mental constructs. Forguson and Gopnik (1988) characterize 5-year-olds as "hyper-realists" about beliefs. Thus there must be some fact which a belief either does or does not agree with. This realism may extend to concepts. In this case children would believe that all categories created by the mind are either true or false representations of types existing in the real world.

While there may be theoretical reasons to suppose that children are biased to treat categories as objective, there is little direct evidence one way or the other. Some research does suggest that children are inflexible in their categorization judgments. They generally believe that there is a single correct way to classify. Piaget (1965) argued that young children could not simultaneously classify objects according to two orthogonal criteria. Though recent research challenges Piaget's claims, it remains true that sorting objects in multiple ways is a difficult skill for children (e.g., Bigler & Liben, 1992; Freund, Baker, & Sonnenschein, 1990). Markman (1990) argues that children may have a strong intuition that category membership is central to the identity of an individual (at least membership in natural kind categories). According to Markman, this intuition leads to the belief that an individual can be only one kind of thing and have only a single name. Thus, evidence for a mutual exclusivity bias in word learning (Markman, 1990; Merriman & Bowman, 1989) may be evidence for beliefs that category membership is objective.

While clearly related to the issue of objectivity, the data discussed above relate to multiple categorizations; whether an object can be a member of categories A and B at the same time. The intuitions that seem most directly related to objectivity involve alternative classification; whether an object could be a member of category A instead of category B. For example, when presented with two alternative ways of classifying an object will children view the choice as a matter of fact? Children's views of alternative classification schemes have not been directly assessed. However, children's views of alternative norms have been explored, primarily in the context of moral reasoning.

Intuitions about realism and relativism have been studied in several domains. Turiel and his colleagues have argued that moral rules are perceived to be objective facts, true independently of human decisions or norms (Turiel, 1983, 1989; Turiel, & Davidson, 1986). Similar points have been made about physical and logical laws (Komatsu & Galotti, 1986; Lockhart, Abrahams, & Osherson, 1977; Nicholls & Thorkildsen, 1988). Moral and physical laws contrast with social conventions which are understood to be particular arrangements made at the discretion of groups. A moral judgment (e.g., stealing is wrong) is held to be universal and unalterable while a social convention (e.g., eating with a fork or fingers) may be different in different places and can be changed (Helwig, Tisak, & Turiel, 1990; Turiel, 1983). Several studies have explored preschoolers' recognition of these differences. Children generally treat moral and physical laws as universal and unalterable, but treat social conventions as more relative (Levy, Taylor & Gelman, 1995; Smetana, 1981). Some studies find a realist bias; across rule types children give more objective responses than do adults (Lockhart et al., 1977; see Piaget, 1932).

Categorization decisions can be thought of as conforming to or violating categorization rules. The decision to group penguins with other birds conforms to our rules; grouping bats with birds violates these rules. Like other rules we may see categorization norms as either relative conventions or as objective facts. Natural kinds represent objective norms (akin to moral laws), artifactual kinds represent invented norms (akin to conventions). Thus the methods used in studies of children's conceptions of moral rules and physical laws are also applicable to the study of beliefs about categorization. For example, children have been asked to judge whether it might ever be acceptable to follow alternative rules; whether it could be "ok to hit" (or, "ok to eat with your fingers"; e.g., Levy et al., 1995; Smetana, 1981). Accepting the alternative behavior suggests the rule is a construction and relative to some particular context (e.g., culture). Rejecting the alternative suggests the rule is objective- valid or not independent of human decisions. The same question can be asked with regards to categorization; is it acceptable to classify bats with birds, or to separate penguins from other birds? Whether children and adults accept or reject these alternatives will depend on whether they view the categories involved as objective or conventional.

Summary

Natural and artifactual kinds seem to play important and different roles in conceptual development. One way to conceive of the difference between these concepts is that natural kinds are thought to be objective, while artifactual kinds are conventional. For example, we discover that there are CATS, but decide that there are WEEKENDS. Thus, it seems important to assess children's and adults' beliefs about the objectivity of categories. While some research has explored people's understanding of objective and arbitrary rules (e.g., morals vs. social conventions, Turiel, 1983) the linkages to categories have not been made. The existing developmental literature contains some suggestions that children may be biased to treat categories as objective (and hence as natural kinds). The remainder of this paper presents four studies of children's and adults' intuitions regarding the objectivity of categories. In particular two questions are addressed. First, do people (children especially) treat categorization rules as conventional or as objective, or do they recognize instances of both types? Second, assuming there is some variation, what accounts for distinctions between objective and conventional categories?

Study 1

Study 1 explored whether children see the bases for categorization decisions as objective facts or as invented conventions. To measure these intuitions, children were asked to evaluate alternative ways of classifying objects. In particular, this study involved alternative sortings of animals and human-made objects

[5] at the basic-level. For example, children were asked whether it was acceptable to sort a deer together with a horse rather than with another deer. Just as judgments of the propriety of alternative rules (e.g., "Hitting is ok.") have been used to demonstrate the belief that moral rules are objective and conventions are relative, judgments of alternative classifications can distinguish between objective and relative categories. As a measure of objectivity, judgments of alternative sortings were compared to judgments of alternative moral and conventional rules.

This study generally follows the methods used in existing studies of moral and conventional rules. However, in most paradigms children are asked to make judgments regarding norms which are the opposite of those they hold (e.g., Smetana, 1981). For example, children first make their own judgment of a rule (e.g., "Is it ok to hit or not?") and then consider changing (or violating) that rule. While these tasks have yielded impressive performance they may place unnecessary demands on children. A child must first commit to one way of answering the question and then consider the experimenter's suggestion of a different answer. The current study removes this challenge by having participants judge a series of decisions made by a puppet. Several studies have demonstrated that children are more accurate at evaluating another's performance rather than their own (Siegal, 1991; Siegal, Waters, & Dinwiddy, 1988).

Methods

Participants. Participants were 21 children recruited from a university preschool in a mid-sized midwestern city. Ages ranged from 4:1 to 6:0 (with mean of 4:9). There were 10 boys and 11 girls. Participants were predominantly white and middle-class.

Design. Children were asked to evaluate six types of judgments. Three instances of each type were presented. Table 1 presents examples of each type (a complete list of stimuli is given in the Appendix). Two types included for comparison purposes were moral (naughty or nice) and conventional (foods for dinner or breakfast) judgments. Children have been found to treat social decisions, such as rules for eating, as invented and different from actions involving harm or rights which are treated as universally applicable (Levy et al., 1995; Smetana, 1981). Two types of items involved categorizing animals at the basic level (e.g., as HORSE or DEER) . Two other sets of items required categorizing objects at the basic level (e.g., as BOAT or TRUCK). In all four categorization sets, the two possible responses were both members of the same super-ordinate category (e.g., BOAT and TRUCK are both types of VEHICLE). Participants also made two "control" judgments. One involved personal preference. The second involved a logical principle. These items were included to check for possible response patterns (e.g., rejecting all choices). Actual stimuli consisted of colored line drawings presented on laminated index cards. Cards were presented for each test item as well as for each potential response (e.g., a picture of boat, and a picture of a truck and a different boat). No pictures of potential responses were presented for moral items.

_______________________________________________________

Table 1

Examples of items used in Study 1

Item Type	Test Example	Response Options	Question Asked of Puppet
Moral	Hitting	Naughty/Nice	Is hitting another kid naughty or nice?
Conventional	Cereal	Breakfast/Dinner	Is cereal a food for breakfast or dinner?
Animal __________ Object	White-Tailed ____Deer___ Ball Peen Hammer	Mule Deer/Horse ________________ Claw Hammer/Bat	Is this one (test) the same kind of thing as this one (response1) or as this one (response2)? Which one does it go with?
Controls		2 Pears/ 1 Apple	Which card has two things on it?
			Which fruit do you like better?

Procedure. Children were asked to evaluate a puppet's choices as acceptable or incorrect. Children were introduced to "Feppy" (a quasi-human puppet) and told that; "Feppy comes from a place far away where they do lots of things differently than we do. Some of the things they do are wrong, but some of the things are just different." They were asked to help the experimenter figure out when Feppy was wrong and when he was just different. Two "warm-up" judgments started the procedure. In one, Feppy incorrectly stated the child's name. In the second Feppy chose a picture of a preferred playmate. In each case children evaluated Feppy's choices. A first question asked whether it was; "Ok for Feppy to choose (say) that or is he wrong?" (order of alternatives was randomized). This question was followed-up in one of two ways. If participants chose "ok" they were asked whether it would also be "ok" for Feppy to choose the other option. This question insured that children saw both responses as acceptable. If children initially judged that Feppy was wrong they were told that everyone where Feppy lives would make the same choice (say the same thing) and re-asked the first question. This question ruled out the interpretation that Feppy had simply mis-spoken; rather he was answering according to a different norm. Thus, a positive ("ok") answer to either follow-up indicates that the choice of response is seen as relative, while a negative response to either follow-up indicates the view that there is only one correct answer.

Items were presented in random order blocked with respect to type. For each type of item the two potential responses were introduced first and their pictures placed in separate boxes. For example, pictures of dinner and breakfast were put in separate boxes. For moral items the two boxes were simply labeled "naughty," and "nice." Animal and object pictures were not labeled. Target pictures were then presented and Feppy was asked to make a choice (put the target card in one of the option boxes). Puppet's choices were either normative or discordant; they either agreed with or violated the culturally accepted rules.[6] Feppy picked the discordant choice for two of the three items of a type. For example, Feppy said that salad was for breakfast while cereal and corn were for dinner. Puppets' choices did not vary across children. After each of Feppy's choices, children were asked the evaluation and follow-up questions. For warm-up items children were given corrective feedback (that Feppy was incorrect to call them the wrong name, but it was ok for him to choose either child to play with). No other feedback was provided.

Results

Data from two participants were dropped from the study because these children judged all of Feppy's responses to be incorrect (including controls and warm-ups). In the analyses that follow, children's data will be referred to as "judgments." Puppet's answers will be referred to as "choices." The results of interest are children's judgments of puppet's choices.

The first panel of Table 2 presents children's first judgments (before follow-up) of puppet's choices. The puppet's normative choices were judged to be acceptable for all items; there were no differences across item types (S²(5) = 4.7, ns, Friedman test). Item differences were significant for discordant choices. Discordant choices were rejected more often for moral than for conventional items (Wilcoxon signed-ranks, T(13) = 88, p<.01 [7]). More discordant choices were rejected for animals than for conventions (T(13) = 77, p<.05). The difference between animal and moral items approached, but did not reach statistical significance (T(10) = 46.5). In contrast, discordant choices for objects were rejected less often than those for moral items (T(15) = 109, p<05) and did not differ from conventions (T(15) = 88.5, ns). Finally, the difference between animal and object items also approached statistical significance (T(12) = 61).

_______________________________________________________

Table 2

Results from Study 1: Judgments of Puppet's Normative and Discordant Responses

Mean Proportion of Initial Judgments That Puppet Was Wrong

Puppets choice:	Animals	Object	Conventions	Morals	Controls
Normative	.10	.02	.10	.00	(Item Counting) .05
Discordant	.79	.69	.50	.95	(Fruit Preference) .14

Mean Objectivity Scores*, Study 1

Puppet's Choice	Animals	Objects	Conventions	Morals	Controls
Normative	.77	.78	.35	.90	(Item Counting) .75
Discordant	.78	.83	.48	.93	(Fruit Preference) .30

*Objectivity scores were based on negative answers to either follow-up question. Negative answers to these questions indicate that only one option was considered correct.

_________________________________________________

Independent of their first answers, children's responses to the follow-up questions also indicated their beliefs about the objectivity of the items. The second panel of Table 2 presents the proportion of judgments that Feppy was "wrong" (or un-acceptable) to follow-up questions for both normative and discordant choices. Because the predictions for follow-up judgments are the same regardless of puppet's choice type, analyses will consider both cases together. These second judgments will be referred to as "objectivity scores" because they indicate the belief that only one answer is correct. Objectivity scores were higher for moral than for conventional items (T(16) = 134, p<.001). Similarly, animal and object items were seen as more objective than conventions (T(15) = 106, p<05, T(15) = 110.5, p<.01, respectively). Animals and objects did not differ from morals (though the animal-moral comparison approached statistical significance, T(13) = 72.5). Finally, there was no difference in objectivity for animal and object items.

There was a high degree of consistency in children's answers. In the first place, responses to follow-up questions were generally the same as responses to first questions for discordant items; the two judgments differed in only 12% of the cases. The majority of these differences (two-thirds) involved an initial rejection of the puppet's choice followed by accepting the choice once it was indicated that all people in the puppet's community would answer the same way (i.e., that it was not simply the puppet making a mistake). Participants also showed consistency in giving the same judgments for all item of a type (either accepting or rejecting all alternative choices). There were three instances of each item type (with two animal and object types). The chance probability of answering all instances of a type in the same way is .25 (Binomial theorem, chance of success = .5). Thus the chance probability of answering all instances of a type in the same way for five or six of the types is .04 (Binomial theorem with chance of success = .25). Eleven of the nineteen participants met the criteria for consistency.

Discussion

Results suggest that the method used in Study 1 is effective at eliciting children's judgments of objectivity. This method did replicate existing findings regarding children's distinction between morals and conventions. As predicted, participants said it was acceptable to have different social conventions but not different morals. Logical and personal preference control items also matched predictions.

Data from children's first responses suggested animal categories were seen as more objective than object categories and that objects did not differ from conventions. However, responses to second questions (objectivity scores) suggest children did not see animal categories as different from object categories. First responses may have been influenced by the belief that the puppet had simply mis-spoken. In addition first responses depend on the individual child interpreting the puppet's response as normative or discordant. If a child mis-categorized the stimuli (e.g., thought a picture of a DEER was really a HORSE) first judgments would be the opposite of those expected. Neither of these factors should have affected second judgments. Thus objectivity scores would seem to be a more reliable index of intuitions.

Based on objectivity scores, basic level categories of familiar items were treated as natural within the context of this experiment. Children generally rejected alternative sortings. Rules for categorizing animals and objects were treated like morals rather than like conventions. That is, our typical way of sorting these items is seen as objectively correct and not (legitimately) violable.

Given that Study 1 explored a narrow range of categories there are many possible interpretations of the results. On the one hand results may reflect a general view that each object has one and only one way of being categorized (Markman, 1990) or some general difficulties with multiple-classification. Alternatively responses may be based on assessments of category naturalness. Since there was no difference between animal and object categories, one possibility is that items were treated as natural because they involved basic level categories. As described by Rosch (Rosch et al., 1976) basic level categories are so over-determined by correlations between features that they just jump out at us. Sorts which ignore these correlations may be unacceptable. Alternatively, children's beliefs about the content of the categories may have influenced their judgments of objectivity. For example, many have argued that categories of animals are natural kinds while categories of objects are not (Gelman, 1988; Keil, 1989). There was some evidence from first response data that object categories were seen as more relative (less objective). To decide between these various accounts it is necessary to explore children's judgments of a wider range of stimuli. Study 2 examines a set of less well-structured categories. Included in Study 2 are sorts at the super-ordinate level as well as sortings of artificial stimuli.

Study 2

If children have a general tendency to accept only one way of categorizing a set of items then they should continue to reject alternative sortings when the categories involved are relatively arbitrary (from an adult perspective). In Study 2 participants were asked to judge sortings of animals and objects at the super-ordinate level rather than at the basic level. The correlated attribute structure of the categories is less rich at this higher level of generality. The super-ordinate categories examined in Study 2 were also unfamiliar. Thus, if Roschian (or other structural) principles are the basis for objectivity judgments, variability in categorization should be more acceptable. A second set of items involved artificial stimuli constructed to vary along two dimensions (for example geometric figures varying in size and shape). The attribute structure of these stimuli should allow sorting on either dimension. Thus structural principles predict the categories used in Study 2 will be seen as less objective than those used in Study 1. It is also possible that the content of the categories might affect judgments. There may some theory or explanation that guides judgments about particular types of categories. For example, some have argued that kinds of animals are particularly likely to be treated as natural kinds and believed to have underlying essences (Atran, 1987; Gelman & Wellman, 1991). Thus categorizing animals (but not objects) may always be thought to be an objective matter of fact.

Normative choices for the items used in Study 2 are less clear than for those used in Study 1. In Study 2 ratings of category membership were assessed and normative responses assigned empirically. Inter-subject agreement on membership decisions also provides a rough measure of coherence on structural principles (e.g., salience, familiarity). This allows an important comparison with objectivity judgments. If judgments of category objectivity are based on structural principles we might expect that objectivity judgments will be well correlated with inter-subject agreement. When people clearly agree on the correct way to categorize something do they also view that categorization as objective? A lack of correlation between inter-subject agreement and objectivity judgments may suggest some domain-specific (e.g., theory-based) influences.

Methods

Participants. Eighteen children recruited from a daycare center in a mid-sized midwestern city participated in the puppet condition. Ages ranged from 4:5 to 5:7 (with mean of 4:11). There were 8 boys and 10 girls. Eleven children participated in the self condition. There were 6 boys and 5 girls in this group (Mean age, 5:0, range 4:4-5:9). Three participants were eventually dropped from the puppet condition of the study because they gave the same response to all items (including controls); two judged all choices to be wrong, one judged all to be acceptable. Children were predominantly white and middle-class. No children had participated in Study 1.

Design. Children were asked to evaluate (or make) six types of choices. A complete list of stimuli is given in the Appendix. The moral and conventional items were the same as those used in Study 1. Animal and object sets involved categorizing instances at the super-ordinate level; for example, deciding whether a lion goes with a cat or a dog. Three instances of each type were presented. Because of the nature of artificial stimuli (two binary dimensions) only two test items were possible. Therefore two sets of artificial stimuli were used. One set varied on shape and size. The other set varied on color and number. The counting and preference control items from Study 1 were also included. Actual stimuli consisted of colored line drawings presented on laminated index cards. Cards were presented for each test item as well as for each potential response. No pictures of potential responses were presented for moral items.

Procedure. There were two conditions. The puppet condition involved a procedure identical to that used in Study 1. In the self condition, children were asked to make the choices made by Feppy in the puppet condition. For example, instead of evaluating Feppy's decision to put a dog and a lion together, the child was asked how to sort the stimuli. Participants were not asked to evaluate alternative responses in the self condition. Otherwise procedure in the two conditions was identical.

Results

Data from the self condition indicate the degree of inter-subject agreement regarding choices of responses. These data were also used to define the "normative" and "discordant" choices in the puppet condition. The first panel of Table 3 presents the mean proportion of participants giving the modal response for each type of item. Agreement did differ by type of judgement (ANOVA with Type as a within-subject variable; F(4,40) = 4.8, p<.05). Pairwise comparisons revealed there was significantly less agreement for animals than for objects or moral items (Tukey's HSD, p<.05). No other comparisons were significant.

_______________________________________________________

Table 3

Results from Study 2: Judgments of Puppet's Normative and Discordant Responses and Inter-subject Agreement on Judgments (Measured by Proportion of Modal Responses)

Mean Proportion of Modal Responses Across Items of a Type

	Animals	Objects	Conventions	Morals	Artificial	Controls
	.67	.94	.82	1.0	.75	.80
Mean Proportion of Initial Judgments That Puppet Was Wrong
Puppet's Choice	Animals	Objects	Conventions	Morals	Artificial	Controls
Normative	.33	.17	.00	.20	.23	(Preference ) .00
Discordant	.60	.33	.13	.80	.77	(Counting) .73

Mean Objectivity Scores*

Puppet's Choice

Animals

Artifacts

Conventions

Morals

Artificial

Controls

Normative

.64

.39

.14

.71

.21

(Preference )

.00

Discordant

.57

.36

.11

.75

.71

(Counting) .71

*Objectivity scores were based on negative answers to either follow-up question. Negative answers to these questions indicate that only one option was considered correct.

_______________________________________________________

The second panel of Table 3 presents children's first judgments of the puppet's choices. Feppy's normative responses were generally judged to be acceptable; there was no difference between items (S(4) = 7.9, Friedman test). Evaluations of discordant choices were more varied. Discordant choices were rejected more often for moral than for conventional items (Wilcoxon signed-ranks, T(13) = 91, p<.01, 1-tail). More discordant choices were rejected for animal items than for conventions (T(7) = 28, p<.05). The difference between animal and moral items was not significant (T(8) = 28.5). In contrast, discordant choices for objects were rejected less often than those for morals (T(10) = 52.5, p<05). Objects did not differ from conventions (T(7) = 21.5). Judgments for animal and object items did not differ (T(8) = 29). Discordant choices for artificial items were less acceptable than those for conventional items (T(14) = 105, p<01, 1-tail) and did not differ from morals (T(5) = 10, ns). It is important to point out that these first judgments depend on participants' evaluations of puppet's choices as normative or discordant, results are thus influenced by degree of inter-subject agreement.

The third panel of Table 3 presents the proportion of "Wrong" (or un-acceptable) responses to follow-up questions for both normative and discordant choices. Because the prediction is that second responses will be the same regardless of puppet's choice type, analyses will consider both cases together as "objectivity scores."

Objectivity scores were higher for moral than for conventional items (T(13) = 91, p<.01). Animal and object items also showed more objectivity than conventions (T(13) = 91, p<01, T(8) = 36, p<.05, respectively). Objectivity scores were lower for objects than for morals (T(11) = 60, p<.05), while animal and moral items did not differ (T(10) = 38.5). This measure also revealed a significant difference between animal and object items, with children treating categorization decisions for animals as more objective (T(11) = 59, p<05). Artificial categories were less objective than morals (T(13) = 85, p<.01) but more objective than conventions (T(11) = 63, p<05).

Post-hoc tests revealed a significant difference between objectivity scores for normative and discordant artificial items (T(9) = 45, p<.05, two-tailed). Unlike the rest of the items, a different set of stimuli were used for normative and discordant choices; all the number/color choices were normative, while shape/size choices were discordant. The effect is likely a reflection of the differences between the two types of artificial items. Ratings were relatively flexible regarding decisions to categorize by number (rather than color); these ratings differed from those for morals (T(13) = 86, p<.05) but not conventions (T(4) = 9, ns). However, ratings were inflexible for choices to categorize by size (rather than shape); ratings differed from conventions (T(11) = 66, p<.05) but not morals (T(8) = 18, ns.). Agreement ratings for the two kinds of artificial items also appeared to be different. There was high agreement for size/shape items (86% of responses matched the mode) but lower agreement for number/color items (64% matched the mode, minimum possible = 54%). However, because of the small sample and restricted range this difference was not statistically significant (T(5) = 2, p=.09).

A final set of analyses explored the consistency of individuals' objectivity judgments. A participant was considered to be consistent if he or she gave the same response (either accepting all choices or rejecting all choices) for at least three of the four three-item sets (animal, object, moral, and conventional) and at least one of the two-item sets (artificial items). The probability of attaining this performance by chance is .04 (Binomial theorem, assuming chance of success = .5). Seven of the fifteen participants met the criteria for consistency. In general, responses to follow-up questions matched their first judgments (for discordant choice items, 85% of responses were the same). Most of the discrepancies (two-thirds of cases) came from initially rejecting the puppet's choice and later accepting the choice after it was indicated that all members of the puppet's community would make that answer.

Discussion

Results from Study 2 demonstrate that children do not treat all categorization decisions as objectively correct or incorrect. They accept some sorts as conventional. The determinants of category objectivity seem to be complex. One hypothesis is that the judgment depends on correlated attributes or familiarity. Artificial items did show an inverse relationship between agreement and flexibility. This suggests that when one dimension was very salient sorting on the other dimension was considered and error. Similarly, while we must be cautious comparing across studies, superordinate categories seemed to be treated as less objective than the basic level categories from Study 1. However, objectivity scores for superordinate categories items did not show a simple relationship to inter-subject agreement. There was significantly lower agreement for animals than for objects. This suggests the perceptual cues for categorizing animal stimuli may have been weaker (and/or the categories less familiar). However, objectivity scores were higher for animals than for objects; sorts of animals were treated as less arbitrary. Some other factors must have been influencing objectivity ratings. One possibility is beliefs about the content of the categories. For example, results are consistent with beliefs that categories of animals have essences and are more theory-based than are categories of human-made objects. Alternatively, differences in judgments might be accounted for by more structural features, including some only imperfectly measured by inter-subject agreement (e.g., similarity, familiarity, feature salience).

While there are several follow-up studies needed to sort out the particular influences on objectivity judgments, it would also be desirable to have converging evidence that the effects from Study 2 are robust. The literature on children's distinction between morals and social conventions suggests several alternative methods of assessing objectivity. Study 3 explores children's judgments of categorization decisions using some of these alternative methods.

Study 3

Results from Studies 1 and 2 suggest that children distinguish between objective and conventional categories. Study 3 attempted to replicate these findings using two additional methods drawn from the literature on moral development. Several judgments have been used to test for a distinction between objective and invented rules (e.g., morals and conventions, respectively). One measure involves judgments of the likelihood of discordant rules. It has been argued that people believe discordant objective (moral) rules are less likely than discordant conventions. If asked whether there might be a place where people follow a discordant rule, children tend to answer "no" for objective rules, and "yes" for conventions (e.g., Nicholls & Thorkildsen, 1988). A second measure used asks how bad it is when people follow a discordant rule (violate the standard rule). Violations of moral rules are judged as worse than violations of conventions (e.g., Smetana, 1981). Both of these methods seem applicable to questions about the objectivity of categories. If one way of categorizing an object is objectively true then we might expect that everyone will categorize that way. Similarly, it seems to be a greater error to deny an objective truth than to deny a convention. A sort which is objectively false is worse than a sort which merely violates our social norms. Thus, judgments of universality and degree of disapproval may offer converging evidence for children's views of the objectivity of categories. In addition, the method used in Study 3 asked the same children to make membership and objectivity judgments. This allows for a more direct comparison of agreement and objectivity scores.

Methods

Participants. Fifteen children recruited from a university-affiliated laboratory preschool participated in Study 3. The average age was 4:10 (range, 4:7 to 5:5). There were seven boys and eight girls in the study. Most children were white and middle-class. No child had participated in Studies 1 or 2.

Design & Stimuli. Four types of items were included in the study. There were two instances each of moral and conventional rule judgments. There were three instance each of superordinate animal and object categorization judgments. A complete list of items is given in the Appendix. As in previous studies, items were presented as triads. Participants were presented with a target object and two possible responses. Colored line drawings of each target and all possible responses were presented except in the case of moral judgments where only pictures of targets were used.

Procedure. The procedure was designed to mirror existing measures of objectivity (Smetana, 1981; Levy et al., 1995). Children were presented with a target object and two possible choices. For each triad, children were first asked to provide their judgments of a question (e.g., "Is hitting naughty or nice?" "Is this one the same kind of thing as this one or this one?"). Two subsequent questions assessed objectivity. First children were asked; "What if we went around to all the different countries in the world where different people lived, do you think there could be another place where people (do/choose the opposite)?" (cf. Nicholls & Thorkildsen, 1988). Participants responded yes or no to this question. A second measure asked children; "What is it like when people (do/choose the opposite)? Is it: great, just OK, bad, or very bad?" Responses to this question were made by indicating a position on a four point scale marked with pictures of smiling and frowning faces (cf. Smetana, 1981). Order of presentation of items was randomized across participants. Questioning always followed the same order; correct choice, universality, then seriousness ratings.

Results

Figure 1 shows mean responses for the three dependent measures used in the study. The first measure considered is children's judgments of the correct rule. Of interest is the degree of inter-subject agreement. Types did differ in inter-subject agreement (S(3) = 15.6, p<.001, Friedman test). Agreement scores for animals were lower than scores for other types of items (though the comparison with conventions only approached significance, T(7) = 54, p<.05, uncorrected; vs. morals, T(11) = 65, p<.05, vs. objects T(7) = 0, p<05). Otherwise agreement scores did not differ among item types.

_______________________________________________________

Figure 1: Mean proportion of judgments that rule would not vary, mean judgments of seriousness of violation, and mean agreement on correct choices from Study 3. Seriousness judgments were scaled to range from 0 to 1. Agreement scores were also scaled to range from 0 to 1 [Agreement = (Mean Number of Modal Responses - .5)/.5] Higher bars indicate greater objectivity and greater agreement

_______________________________________________________

The two other measures assessed beliefs about objectivity. Responses to the question of whether or not some other people might answer differently will be referred to as universality judgments. Responses to the "what is it like" question will be referred to as seriousness judgments.[8] Conventions were seen as less universal than morals (T(9) = 42,p<.05). Sortings of animals were more universal than conventions (T(9) = 44, p<05) but did not differ from moral items (T(6) = 11, ns). Object categories were intermediate; they were seen as more universal than conventions (T(8) = 35, p<.05) but less than morals (this difference approached statistical significance, T(9) = 44, p<.05, uncorrected). Finally, the difference between animal and object categories also approached significance (T(6) = 20, p<.05, uncorrected).

Results from the seriousness of transgression measure were similar. Transgressions of morals were rated as more serious than transgressions of conventions (t(15) = 3.1, p<.01). Alternative sortings of animals were not seen different than moral transgressions (t(15) = 1.7, ns) but the difference between animals and conventions approached significance (t(15) = 1.8, p<05, uncorrected). Results also indicated that discordant sortings of object were less serious; they did not differ from conventions (t(15) = 1.1, ns) but approached significant difference from morals (t(15) = 2.0, p<.05, uncorrected). On this measure animal and artifact categories did not differ in seriousness (t(15) = 1.2, ns).

Because of the small number of items included in this study, statistical tests of idividual consistency have little power. However, results indicated that participants did respond the same way to items of the same type. For judgments of universality it is possible to define a pattern of responses consisting of responding to all instances of a type the same way (i.e., either judging all to be universal or all to be relative) and doing so for all types. The probability of meeting this pattern with one or fewer deviations is .19 (Binomial theorem). Eleven of the 15 participants met this criterion (p<.00001, second-order binomial test of 11 or more out of 15 with p(success) =.19). Consistency of a subject's seriousness judgments is indicated by the standard deviation of ratings for instances of a type (e.g., did the child give the same rating for both moral items?). Table 4 presents the mean standard deviations for the different item types (across participants). On average there was less than one point difference between an individual's ratings for different items of a type. The deviation was less than one point for moral and conventional types (the standard deviation of a one point difference for these items is .71, e.g., rating one item as "bad" and one as "very bad"). The average deviation for animals and objects was slightly greater than a one point difference (1 point Stdev = .58, e.g., rating two items as "bad" and one as "very bad").

_______________________________________________________

Table 4

Mean Standard Deviations of Seriousness Ratings, Study 3

Moral	Conventional	Animal	Object
.37	.33	.72	.62

_______________________________________________________

The correlation (Pearson's r) between judgments of universality and seriousness was calculated for each participant to measure the relationship between these measures. Four participants had no variance in universality judgments. Of the remaining eleven, only four showed significant correlation between the two judgments (critical r value = .55, for t(8) = 1.86, p = .05, Mean r across subjects = .40). Again, the small number of observations limits the power and reliability of this test.

Discussion

Patterns of responses from Study 3 were very similar to those shown in Study 2. Morals were treated as objective while conventions were treated as relative. Sorting animals into superordinate categories was treated as an objective matter of fact, akin to moral judgments. More flexibility was allowed for sortings of objects. These data also lend support to the finding from Study 2 that super-ordinate animal categories were rated as more objective than super-ordinate categories of objects. Again, this result does not seem to be due to structural factors; there was less inter-subject agreement for animal items than for object items. While the trends from Study 3 were in the predicted directions, results were less clear cut using these methods. There was more variability in subjects' responses than was seen in Studies 1 and 2. This may have been because participants were first asked to make their own responses and then asked to consider the alternative reply. Children may have seen follow-up questions as challenges to their original choices (Siegal, 1991). While the results are less clear, they do offer converging evidence that children make distinctions between more and less objective categories.

The final study included in this report investigates adults' judgments of objectivity. In considering the development of children's ideas of category naturalness it is important to know the adult state. Study 4 investigates adults judgments of objectivity using a paradigm similar to that used in Studies 1 and 2.

Study 4

Study 4 explored adult intuitions regarding the objectivity of categories. Information about adult judgments is interesting in it own right but is particularly important for the perspective it provides on children's performance. Several studies have explored flexibility in categorization (e.g., Lamberts, 1994) or beliefs about knowledge certainty as personality variables (e.g., Schommer, 1990). However, there have been few studies of adults' beliefs about whether (or which) categories are matters of fact or convention.

On the one hand we might expect that children's beliefs about objectivity and relativism are based on the input they receive from adult members of the culture. Thus children's performance on these tasks should mirror adults'. Alternatively there might be some bias or developmental progression. As discussed above, children may tend to over-estimate the objectivity of phenomena while adults might recognize more cultural relativism (Chandler, 1988). The opposite pattern is also possible. Children may not understand the basis of categorization decisions and so be more accepting of alternative classifications. Adults know the reasons we have the categories we do. This may show up as a greater distinction between category type (e.g., categories of animals are discoveries of science, while categories of objects are not) or as a general reification of categories.

Adults' judgments regarding the objectivity of categorization judgments have been explored in two studies. Malt (1990) asked adults how they would decide to categorize ambiguous objects. People believed that human-made objects could be arbitrarily assigned to categories but that there was a correct answer for categorizing animals (even if they did not know what it was). Kalish (1995) asked adults to judge whether there was a single correct answer in disputes over category membership (e.g., one person say an object is a CHAIR, another says it is not). Adults tended to accept differences of opinion for objects but often believed there was a correct answer to disputes about categorizing animals. However, a significant number of disputes about animals were thought to be un-resolvable. This pattern held for both super-ordinate and basic level categories. These results suggest that adults' may treat object categories as more relativistic and conventional than animal categories.

Methods

Participants. Twenty-nine students from a large midwestern university participated in the study for course credit in an educational psychology class. There were 20 females and 9 males. Subjects were predominantly white.

Design. Adults were asked to evaluate (and make) all of the types of judgments included in Studies 1 and 2. There were three instances of moral and conventional items, otherwise six instances of each type were presented. The Appendix contains a complete list of stimuli. Importantly, the superordinate items were divided into two sub-types: familiar (FELINE and FURNITURE) and unfamiliar (roughly, INVERTEBRATE and FOOD-CONTAINER). An additional type of item included in this study involved judging solutions to mathematical problems. These items were included because of the possibility that college students might give relativistic responses to all items (e.g., be relativists regarding morals). For mathematical items participants were shown an algebraic expression with one unknown and two numerals representing possible answers. Actual stimuli consisted of black and white scanned images of pictures from Studies 1 and 2. Data were gathered using Macintosh computers running Hypercard.

Procedure. Participants were informed that they would be answering some questions and evaluating the answers of a third party (the "informant"). The informant was identified as "Farnhi," and described as a person "who comes from a culture very different from ours where they do lots of things differently." Finally the distinction between errors and differences of opinion was illustrated by one example of each (believing the sun rotates around the earth, and preferring one color to another, respectively). Thirty-seven experimental items were then presented. In each case a participant first indicated his/her own response to the question (e.g., whether hitting was naughty or nice). Farnhi's response was then presented and the participant was asked to evaluate that response as acceptable or erroneous. For two of the three instances of a type, Farnhi's response differed from the subject's; in one case Farnhi agreed. The same follow-up questions used in the child studies were asked in this paradigm. Items were presented in random order blocked with respect to type.

Results

Agreement scores differed among items (S(7) = 70.2, p<.0001, see Figure 2). Scores fell into two groups. Moral, conventional, mathematical, and basic level animal and object items did not differ from each other. Agreement scores for these item types were significantly higher than scores for super-ordinate or artificial items. Scores for these types did not differ (all comparisons p<.05, Tukey's HSD).

When the informant agreed with the participant, his choices were accepted 90% of the time. There were no differences between item types in levels of acceptance for normative judgments (ANOVA with judgments for each normative item as a repeated measure variable; F(7,196) = .77, ns). Participants' first responses to the informant's discordant judgments will be analyzed together with responses to follow-up questions. As shown in Figure 2, these two measures produced similar results. Results to follow-up questions (objectivity scores) will be analyzed in detail. Data from first responses to discordant judgments will be presented when these responses differ from objectivity scores.

_______________________________________________________

Figure 2: Mean agreement, objectivity scores (negative responses to follow-up questions), and men proportion of initial rejections of informant's discordant responses from Study 4. Agreement scores were scaled to range from 0 to 1 [Agreement = (Mean Number of Modal Responses - .5)/.5]. Higher bars indicate greater objectivity and greater agreement.

_______________________________________________________

The three types of items included as controls provided evidence that adults would treat items as both objective and as conventional (relative). Mathematical and moral items were judged to be objective (only one correct answer) at greater than chance levels (T(29) = 20.5, p<.01, T(29) = 94.5, p<.05, respectively). Judgments for these items did not differ from each other (T(19) = 33, ns). Conventional items were significantly less likely to be treated as objective than either mathematical or moral items (T(28) = 406, p<.01, T(26) =351, p<.01, respectively). Objectivity scores for conventional items were below chance (T(29) = 417, p<.01).

Objectivity scores for basic level animal and object categories did not differ from scores for moral items (T(16) = 79, T(21) = 162, ns, respectively), though first responses for objects were significantly less objective than those for morals (T(15) = 105, p<.05). Both categories were treated as less objective than mathematical items (T(20) = 197, T(20) = 206, p<.01) but were seen as more objective than conventional items (T(27) = 378, T(26) = 337, p<.01). Finally, there was no significant difference between basic level animal and object categories (T(19) = 131, ns).

Superordinate level categories of animals and objects were treated as less objective than moral items (T(24) = 261, p<.05, T(24) = 274, p<.01, respectively). Again, both kinds of categories were seen as less objective than mathematical items (T(23) = 276, T(25) = 325, p<.01) but more objective than conventional items (T(24) = 288, T(24) = 289, p<.01). In contrast to results from basic level categories, at the superordinate level, categories of animals received higher objectivity scores than categories of objects (T(21) = 216, p<.01). However, this difference did not appear in subjects' first responses (T(23) = 165, ns). In first response data there was no interaction between category level and category type.

Both superordinate categories were seen as less objective than basic level categories (animals; T(21) = 204, objects; T(17) = 138, p<.05). In the data from subjects' first responses, superordinate categories did not differ from basic level categories (though the difference for animals approached statistical significance, T(21) = 171, p=.03 uncorrected for familywise error; for objects, T(17) = 102.5, ns). However when considering category level independent of type (i.e., mean for all superordinates compared with mean for all basics) first responses were more objective for basic level categories (T(21) = 180, p<.05, uncorrected).

Post-hoc inspection revealed differences between the sub-types of superordinate items. Table 5 presents the mean objectivity scores for familiar and unfamiliar superordinate level categories. There was no significant difference between familiar and unfamiliar animal categories (T(14) = 34.5, ns). However, subjects allowed greater flexibility in sortings of unfamiliar than familiar objects (T(21) = 4.5, p<.01, Dunn's test). Unfamiliar objects also had significantly lower objectivity scores than did unfamiliar animals (T(15) = 19, p<.05), while there was no difference for familiar items. To reassess the effects of category level, objectivity scores for familiar and unfamiliar superordinates were compared with the mean objectivity score for basic level categories (Mean = .60). Comparisons revealed that familiar superordinates did not differ from basic level categories (animals- T(24) = 213, ns; objects- T(24) = 160, ns). Both kinds of unfamiliar categories were seen as less objective than basic level categories (animals- T(27) = 297, p<.05; objects- T(27) = 369.5, p<.01, Dunn's test). Agreement scores did not differ for familiar and unfamiliar items (animals- Mean Familiar = .90, Mean Unfamiliar = .83; objects- Mean Familiar = .90, Mean Unfamiliar = .86).

_______________________________________________________

Table 5

Mean Objectivity Scores for Familiar and Unfamiliar Super-Ordinate Categories, Study 4

Category	Familiar	Unfamiliar
Animals	.51	.42
Objects	.58	.22

_______________________________________________________

Artificial items also received intermediate objectivity scores. Scores were significantly lower for artificial items than for morals (T(23) = 246, p<.05) or for mathematical items (T(25) = 325, p<.01) but were higher than scores of conventions (T(21) = 215, p<.01). Artificial categories were less objective than basic level animals (T(21) = 206, p<.05) and approached being significantly different than basic level objects (T(22) = 194, p=.02, uncorrected) but did not differ from superordinate categories. A very different picture emerged from the data on first responses to discordant judgments (i.e., where the informant's answer was the opposite of the subject's). In these data, artificial items did not differ from conventional items and were less objective than animal and object categories at both the basic and superordinate levels (e.g., compared with superordinate objects, T(15) = 111, p<.05). No significant differences were observed between the two sub-types of dimensional items (T(14) = 22.5, ns).

Adults' responses in this task were frequently inconsistent. For discordant choice items, responses to first questions and follow-up questions were consistent 80% of the time. Most of these non-matching judgments (17% of total) involved judging the response to be incorrect but accepting the response as a general group convention. There was considerable variability in judgments of different instances within item types. Consistency within an item type was defined as answering five out of six items in the same way (or all the same for moral, mathematical, and conventional types which contained only three instances each). Based on binomial probability, the probability of answering consistently on five or more of the eight item types by chance is .02. Fourteen of the twenty-nine participants met the criteria for consistency across item types.

Discussion

Adult participants did treat some items as objective and some as relative. Moral, and especially mathematical, items were treated as matters of fact; one answer was right and one was wrong. Food conventions were treated as relative; different groups could make different decisions. These data also revealed that adults see some categories as more objective than others. Basic level categories were treated as objective, as akin to moral judgments. Artificial categories were less objective than morals or basic level categories (and may not have differed from conventions). Judgments of superordinate categories revealed some interactions between different factors. Superordinate level categories were generally treated as more relativistic than basic level categories. However, distinctions between category types appeared at the superordinate level. Sortings of animals were seen as more objective than sortings of objects. The data also revealed an important effect of familiarity. Unfamiliar superordinate object categories were treated as significantly less objective than other kinds of categories. There was a similar, but smaller, effect of familiarity for animal items. Since familiarity and category level are partially confounded in these data (all basic level categories were familiar) we cannot identify which factor is affecting adults' judgments.

In some ways adults' reactions were different than children's. Adults' tended to see categories as more conventional than did children. Objectivity scores were uniformly lower than children's scores from Study 1 (an ANOVA comparing these groups revealed a main effect of age, F(1,47) = 7.0, p<05). However, compared with Study 2, responses were not significantly different . Adults displayed less consistency in their responses than did children. In part this may be due to the wider range of stimuli included in the adult study. While there were some differences, there was a general similarity in both groups' patterns of responses. Morals were considered very objective, conventions very relative. Judgments of categories were also similar. Both groups tended to treat basic level animal and object categories as objective. Manipulations of category attributes reduced objectivity scores. Artificial categories were seen as less objective. Children and adults saw superordinate (and/or unfamiliar) categories as less objective than basic level ones. For both groups this effect was more pronounced for object categories than for animals.

General Discussion

One way to view the distinction between natural and artifactual kinds is as a difference in the objectivity of categories. Natural kinds are truly existing groupings we discover. Artifactual kinds are conventional groupings we invent. If category membership is an objective matter of fact, then classification decisions are either correct or erroneous. If criteria for category membership are conventional then it should be acceptable for people with different conventions to classify individuals in different ways. The four studies described in this paper explored children's and adults' intuitions about the acceptability of alternative ways of classifying animals, objects, and artificial stimuli.

Three studies explored children's judgments of objectivity. The major finding was that children distinguished different degrees of acceptability of alternative sorts. In some cases there was only one right way to categorize, in other cases there was more flexibility. This suggests children do not have a general bias to treat all categories as objective matters of fact. Rather, something about children's representation of categories leads them to accept more or less flexibility in categorization.

A second set of questions concerned the sources of objectivity judgments. Conclusions here are more tentative, given the limited number of items studied. Children judged sorts that did not conform to standard basic level categories to be errors. Norms for categorizing animals and objects at this level were held to be objective; as akin to moral rules rather than conventions. Subsequent studies explored superordinate and artificial categories. Children's views of classifying artificial stimuli depended on the attributes of the items; some sorts were treated as objective while others were seen as more conventional. This difference may have been due to salience. When one dimension was the obvious one to use to organize the stimuli (based on inter-subject agreement) it was an error not to sort on this feature. When two features were salient children tended to accept alternative sortings. However, salience (or other structural properties) did not so directly predict responses to superordinate categories of animals and objects. Superordinate categories of objects were treated as conventional. This may reflect the lower coherence of superordinate as opposed to basic level categories. However, superordinate categories of animals were viewed as objective. These items received the same ratings for objectivity as did moral judgments. This was true despite the fact that there was less agreement as to how to classify animals at the super-ordinate level than there was for other categories. This suggests animal categories were less coherent on structural features (e.g., similarity) than other stimuli. Children did not agree which sort was correct, but each individual felt there was only one correct way to categorize animals. While more studies are needed to tease apart all the influences on judgments of objectivity, these data suggest that both structural and knowledge-based factors may play a role.

Adults' intuitions about objectivity generally matched children's. Some judgments were rated as objective; discordant moral and mathematical judgments were errors. Others were treated as relative; alternative food conventions were acceptable. Sorting animals and objects into basic level categories was viewed as objectively correct, similar to moral judgments. However, not all categorization decisions were treated this way. Alternative sortings of artificial stimuli were more acceptable. Some kinds of animal and object categories were also treated as more like conventions. For adults this seemed to depend on familiarity; familiar categories were seen as more objective than unfamiliar. However, the effect of familiarity was more pronounced for object than for animal categories. Thus adults may have a bias to treat animal categories as objective. Finally, there was some evidence that adults may be more relativistic than children. In cases where children and adults judged the same items, adults' responses tended to be less objective. However, this effect was not large and is based on comparisons across different studies.

Familiar, basic level categories of animals and objects were treated as natural kinds by participants in these studies. Decisions about how to sort these items were viewed as objective matters of fact, rather than as conventional opinions. Other categories were seen as less objective. Exactly what determines objectivity judgments is a matter for further research (see below). However, results from the current studies seem to indicate that, other things being equal, categories of animals are considered more objective than are categories of objects.

Sources of Objectivity Judgments

There are several ways to explain why categories might or might not be viewed as objective. One set of accounts stress structural attributes of categories; for example, a Roschian view of naturalness. What makes a category objective is that its members are similar to each other and different from non-members. Factors such as inter-item similarity and category level would be expected to influence judgments. Another factor might be familiarity. Well-known, salient categories will be expected to be objective. The studies included in this paper used inter-subject agreement regarding categorization decisions as a rough measure of these factors. Results suggest that these factors do influence judgments of category objectivity. However, these studies do not address their individual contributions. Future studies, which systematically vary the various structural factors, are needed to address the particular contribution of such attributes as; similarity, level, familiarity, and salience on judgments of objectivity.

While structural factors did influence objectivity judgments results suggests that these factors do not completely account for subjects' responses. The assumption of these studies is that categories with higher inter-subject agreement were better categories on structural principles. Given this assumption the fact that categories with lower agreement (animal superordinates) were rated as more objective than categories with higher agreement (object superordinates) indicates that some other, "non-structural," factors were influencing judgments. Among the possible other influences are theoretical beliefs. For example, people believe that in some domains (e.g., naive biology), categories are tracking the causal structure of world. In other domains (e.g., human artifacts) there is less causal structure and categories are more like conventions (see Keil, 1989).

The rationale for exploring intuitions regarding the objectivity of categories was to provide an additional perspective on the distinction between natural and artifactual kinds. How closely do categories identified as objective match with those traditionally identified as natural? As discussed above there are several ways of characterizing natural kinds. These characterizations tend to agree on a core set of kinds. In particular, categories of animals at the folk-specific/basic level are consensus choices for natural kinds (Atran, 1987; Keil, 1989). These categories were identified as objective in the above studies. However different conceptions of natural kind also differ in their identifications. In the above studies some object and artificial categories were also identified as objective. Artificial categories especially would seem to be poor candidates for natural kinds on theory-based views. It remains to be seen just how category objectivity correlates with other attributes of categories. Will categories identified as natural kinds on other measure (e.g., categories with rich inductive potential, categories named by nouns; see Gelman & Coley, 1991; Markman, 1990) also be treated as objective? Rather than trying to identify a single correct characterization of natural kind it may be more profitable to focus on the particular attributes of categories and their inter-relationships. Natural kindhood may be a matter of degree with categories showing some subset of the attributes of idealized natural kinds (Gelman & Coley, 1991; Markman, 1990).

Objectivity and Constraints on Concepts

The various accounts of natural kind have often been developed in the context of claims about constraints on human concepts. The argument is that human concept learners must have some way of narrowing down the infinite number of possible categorization schemes into a manageable set of plausible hypotheses (Keil, 1981). Natural kinds are those concepts consistent with the (domain general or specific) constraints on concepts. There have been debates concerning just how to characterize these constraints; as limitations, biases, assumptions, preferences, etc. Intuitions about the objectivity or arbitrariness of conformity to constraints may bear on this question. Stronger constraints should produce categories which are seen as objective. For example, categories like GRUE (green before the year 2000, blue thereafter, Goodman, 1955) or categories which violate ontological boundaries (something square and an hour long, Keil, 1979) seem like errors. There is something wrong about these categories. This suggests the constraints they violate are relatively strong. Other kinds of violations evoke a different response. Categories based on size rather than shape, or morphology rather than evolutionary descent (for animals), seem more plausible. These schemes are different but not errors. This suggests the constraints violated in these cases may be closer to assumptions or preferences. One of the things we would like to know about a particular hypothesized constraint is its strength. For example, does the constraint limit what is even conceivable, what is considered as a legitimate possibility, or which possibility is preferred? Children's evaluations of the objectivity of categorization decisions may give us a better appreciation of the kind of constraints guiding concept learning (cf. Keil, 1981).

Categories are formed and evaluated according to rules (constraints, biases, etc.). Just as there are different types of rules governing social behavior (Turiel, 1983), we should expect there to be different kinds of rules governing categorization. In particular, this paper explored judgments of the objectivity of categories; are the ways we categorize seen to be matters of fact (and thus correct or incorrect) or matters of convention (and thus more arbitrary)? Both children and adults made distinctions among categories. Some were seen as objective, some as conventions. While some hypotheses were advanced to account for these variations it remains for future work to carefully sort out the various influences on category objectivity. Nonetheless, these studies have demonstrated that even preschoolers' have robust and consistent intuitions that some categories are more objective than others. In the context of theories of natural kinds and constraints on categories these intuitions seem to be important sources of evidence regarding cognitive development.

References

Atran, S. (1987). Folkbiological universals as common sense. In S. Modgil and C. Modgil (Eds.) Noam Chomsky: Consensus and controversy, (pp. 247-268). London: Falmer Press.

Bigler, R. S., & Liben, L. S. (1992). Cognitive mechanisms in children's gender stereotyping: Theoretical and educational implications of a cognitive-based intervention. Child Development, 63, 1351-1363.

Brown, R. (1965). Social Psychology. New York: Free Press.

Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73-121.

Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press.

Chandler, M. (1988). Doubt and developing theories of mind. In J. W. Astington, P. L. Harris, & D. R. Olson (Eds.), Developing theories of mind (pp. 387-413). New York: Cambridge University Press.

Flavell, J. H. (1988). The development of children's knowledge about the mind: From cognitive connections to mental representations. In J. W. Astington, P. L. Harris, & D. R. Olson (Eds.), Developing theories of mind (pp. 244-267). New York: Cambridge University Press.

Fodor, J. A. (1975) The language of thought, Cambridge, MA: MIT Press.

Forguson, L., & Gopnik, A. (1988). The ontogeny of common sense. In J. W. Astington, P. L. Harris, & D. R. Olson (Eds.), Developing theories of mind (pp. 226-243). New York: Cambridge University Press.

Freund, L. S., Baker, L., & Sonnenschein, S. (1990). Developmental changes in strategic approaches to classification. Journal of Experimental Child Psychology, 49, 343-362.

Gelman, S. A. (1988). The development of inductions within natural kind and artifact categories. Cognitive Psychology, 20, 65-95.

Gelman, S. A., & Coley, J. C. (1991). Language and categorization: The acquisition of natural kind terms. in S. A. Gelman & J. P. Byrnes, (Eds.), Perspectives on Language and Cognition: Interrelations in Development. (pp. 146-196). New York: Cambridge University Press.

Gelman, S. A., & Kalish, C. W. (1993). Categories and causality. in R. Pasnak and M. L. Howe (Eds). Emerging themes in cognitive development. (pp. 3-32). New York: Springer-Verlag.

Gelman, S. A. & Markman, E. M. (1986). Categories and inductions in young children, Cognition, 23, 183-209.

Gelman, S. A., & Wellman, H. M. (1991). Insides and essences: Early understandings of the non-obvious. Cognition, 38, 213-244.

Goodman, N. (1955). Fact, fiction, and forecast, Indianapolis: Bobbs-Merrill.

Helwig, C. C., Tisak, M. S., & Turiel, E. (1990). Children's social reasoning in context: Reply to Gabennesch. Child Development, 61, 2068-2078.

Horton, R. (1970). African traditional thought and western science. in B. Wilson (Ed.) Rationality. (pp. 131-142) New York: Harper & Row.

Kalish, C. W. (1995). Graded membership in animal and artifact categories. Memory & Cognition, 23, 335-353.

Keil, F. C. (1979). Semantic and conceptual development: an ontological approach, Cambridge, MA: Harvard University Press.

Keil, F. C. (1981). Constraints on knowledge and cognitive development. Psychological Review, 88, 197-227.

Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press.

Kendler, H. H., & Kendler, T. S., (1962). Vertical and horizontal processes in problem solving. Psychological Review, 69, 1-16.

Komatsu, L. K., & Galotti, K. M. (1986). Children's reasoning about social, physical and logical regularities: A look at two worlds. Child Development, 57, 413-420.

Kripke, S. (1972). Naming and necessity. In D. Davidson and G. Harman (Eds.), Semantics of natural language (pp. 253-355). Dordrecht, Holland: Reidel.

Lakoff, G. (1986). Women, fire and dangerous things: what categories reveal about the mind. Chicago: University of Chicago press.

Lamberts, K. (1994). Flexible tuning of similarity in exemplar-based categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 1003-1021.

Levy, G. D., Taylor, M. G., & Gelman, S. A. (1995). Traditional and evaluative aspects of flexibility in gender roles, social conventions, moral rules, and physical laws. Child Development, 66, 515-531.

Locke, J. (1961). An essay concerning human understanding, 5th edition, Volume II. New York: E.P. Dutton. (Original work published 1707).

Lockhart, K. L., Abrahams, B., & Osherson, D. N. (1977). Children's understanding of uniformity in the environment. Child Development, 48, 1521-1531.

Malt, B. C. (1990). Features and beliefs in the mental representation of categories. Journal of Memory & Language, 29, 289-315

Markman, E. M. (1990) Categorization and naming in children. Cambridge, MA: MIT Press.

Medin, D. L., & Ortony, A. (1989). Psychological essentialism. In S. Vosniadou & A. Ortony (Eds.) Similarity and analogical reasoning. (pp. 179-195). New York: Cambridge University Press.

Merriman, W. E., & Bowman, L. L. (1989). The mutual exclusivity bias in children's word learning. Monographs of the Society for Research in Child Development, 54 (3-4, Serial No. 220).

Mill, J.S. (1973). System of logic (8th edn.) In J. M. Robson (Ed.), Collected works of John Stuart Mill (Vols VII and VIII). Toronto: University of Toronto Press. (Originally published, 1872).

Nicholls, J. G., & Thorkildsen, T. A. (1988). Children's distinctions among matters of intellectual convention, logic, fact, and personal preference. Child Development, 59, 939-949.

Piaget, J. (1932). The moral development of the child. New York: Harcourt Brace.

Piaget, J. (1965). The child's conception of number. New York: Norton.

Quine, W. V. (1977). Natural Kinds, In S. Schwartz (Ed.), Naming, necessity, and natural kinds. (pp. 155-175). Ithaca, NY: Cornell University Press.

Rosch, E. (1973). On the internal structure of perceptual and semantic categories. In T. E. Moore (Ed.), Cognitive development and the acquisition of language (pp. 111-144). New York: Academic Press.

Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D, & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382-439.

Schommer, M. (1990). Effects of beliefs about the nature of knowledge on comprehension. Journal of Educational Psychology, 82, 498-504.

Siegal, M. (1991). Knowing children: Experiments in conversation and cognition. Hove, England: Erlbaum.

Siegal, M., Waters, L. J., & Dinwiddy, L. S. (1988). Misleading children: Causal attributions for inconsistency under repeated questioning. Journal of Experimental Child Psychology, 45, 438-456.

Smetana, J. G. (1981). Preschool children's conceptions of moral and social rules. Child Development, 52, 1333-1336.

Turiel, E. (1989) Domain-specific social judgments and domain ambiguities, Merrill-Palmer Quarterly, 35, 89-114.

Turiel, E. (1983). The development of social knowledge: Morality and convention. New York: Cambridge University Press.

Turiel, E., & Davidson, P. (1986). Heterogeneity, inconsistency, and asynchrony in the development of cognitive structures. In I. Levin (Ed.), Stage and structure: Reopening the debate. (pp. 106-143). Hillsdale, NJ: Ablex.

Wellman, H. M. (1990). The child's theory of mind. Cambridge, MA: MIT Press.

Whewell, W. (1967). The philosophy of the inductive sciences. London: Cass (original work published 1840).

Appendix

Items used in Studies 1-4

Items Used in Study 1

Item Type	Options	Items
Basic-level Animals	Frog/Fish	Fish1, Fish2, Fish3*
	Deer/Horse	Deer1, Deer2, Deer3*
Basic-level Objects	Bat/Hammer	Hammer1, Hammer2, Hammer3*
	Boat/Truck	Boat1, Boat2, Boat3*
Conventions	Dinner/Breakfast	Corn, Cereal, Salad*
Morals	Nice/Naughty	Hitting, Coloring, Stealing*
Controls	2 Pears/ 1 Apple	Which has two items?*
		Which do you like better?

Items used in Study 2

Item Type	Options	Items
Super-Ordinate Animals	Cat/ Dog	Lion, Hyena, Leopard*
Super-Ordinate Objects	Pot/ Barrel	Bowl, Trashcan, Basket*
Artificial	Small Triangle/Big Square	Small Square, Big Triangle
	3 Purple/ 2 Green Circles	2 Purple, 3 Green
Conventions	Breakfast/ Dinner	Corn, Cereal, Salad*
Morals	Naughty/ Nice	Hitting, Coloring, Stealing*
Controls	2 Pears/ 1 Apple	Which has two items?
		Which do you like better?*

* Puppet's responses were normative for these items

Items used in Study 3

Type	Options	Item	Type	Options	Item
Moral	Mean, OK	Hitting	Conventional	Breakfast, Dinner	Cereal
	Naughty, Nice	Sharing		Breakfast, Dinner	Corn
Animal	Fish, Crab	Frog	Object	Truck, Boat	Train
	Mouse, Turkey	Rabbit		Ruler, Knife	Scissors
	Cat, Dog	Lion		Brush, Basket	Broom

Items used in Study 4

Item Type	Options	Items
Basic-level Animals	Frog/Fish	Fish1, Fish2, Fish3*
	Deer/Horse	Deer1, Deer2, Deer3*
Basic-level Objects	Bat/Hammer	Hammer1, Hammer2, Hammer3*
	Boat/Truck	Boat1, Boat2, Boat3*
Conventions	Dinner/Breakfast	Corn, Cereal, Salad*
Morals	Nice/Naughty	Hitting, Coloring, Stealing*
Super-Ordinate Animals	Cat/ Dog	Lion, Hyena, Leopard*
	Bee/Turtle	Lobster, Hummingbird, Crab*
Super-Ordinate Objects	Pot/ Barrel	Bowl, Trashcan, Basket*
	Guitar/Desk	Piano, Television, Harp*
Artificial	Small Triangle/Big Square	Small Square, Big Triangle
	3 Gray/2 White Circles	2 Gray, 3 White
Mathematical	5+7=12 or 14*	2x3=8 or 6, 4-8=-4 or 0

*Informants' responses matched participants'

Footnotes

[1]In this paper individuals or instances will be indicated by lowercase, categories will be indicated by uppercase words.

[2]It is important to distinguish the idea that categories may be objective or invented from the idea that members of categories may be naturally occurring or constructed (e.g., Gelman, 1988). For example, gold and gems are (usually) naturally occurring. The claim above is that GOLD is also naturally occurring (it is a real kind in nature) while GEM is not. Conversely developments in mathematics and philosophy may suggest that COMPUTER is a natural kind, while any individual computer is a human construction (cf. Keil, 1989).

[3]See also Vygotsky's (1962) distinction between natural and scientific concepts.

[4]This conception of convention seems likely to be culturally specific. One characterization of the difference between traditional and modern cultures hinges on exactly this point. Modern cultures view many (if not all) of their practices as arbitrary conventions while traditional cultures see good reasons and underlying explanations (e.g., see Horton, 1970).

[5]Human-made objects will be referred to as "objects." These items are traditionally labeled "artifacts" (e.g., Gelman, 1988) however this obscures the distinction between the instances being human constructions and the category itself being a human construction.

[6] Normative responses for moral, conventional, and control items were determined based on results from the self condition of Study 2.

[7] Unless otherwise indicated, significance levels reported were based on 1-tailed tests using Holm's control for familywise error. This is a stepwise version of Dunn's test in which the largest test value is compare against a significance level of alpha/N(number of tests), the second largest compared against alpha/(N-1) etc. Results reported as approaching statistical significance were reliable at p<.05 before controlling for familywise error.

[8]Because seriousness judgments were made as ratings on a scale, they were analyzed using parametric measures (e.g., t-tests).