An Apple is More Than Just a Fruit: Cross-Classification ...people.uncw.edu/nguyens/files/nguyen and murphy 2003.pdf · classification in children’s concepts, focusing on taxo-nomic

An Apple is More Than Just a Fruit: Cross-Classification in

Children’s Concepts

Simone P. Nguyen and Gregory L. Murphy

This research explored children’s use of multiple forms of conceptual organization. Experiments 1 and 2examined script (e.g., breakfast foods), taxonomic (e.g., fruits), and evaluative (e.g., junk foods) categories. Theresults showed that 4- and 7-year-olds categorized foods into all 3 categories, and 3-year-olds used bothtaxonomic and script categories. Experiment 3 found that 4- and 7-year-olds can cross-classify items, that is,classify a single food into both taxonomic and script categories. Experiments 4 and 5 showed that 7-year-oldsand to some degree 4-year-olds can selectively use categories to make inductive inferences about foods. Theresults reveal that children do not rely solely on one form of categorization but are flexible in the types ofcategories they form and use.

The history of conceptual development contains anumber of proposals of a developmental shift fromprimitive or simpler conceptual structures to moresophisticated and abstract concepts (e.g., Inhelder &Piaget, 1964; Vygotsky, 1962). The motivation forthese proposals was both theoretical and empirical.Theoretically, the shift was thought to reflectchildren’s inability to identify and represent abstrac-tions. Young children could not identify the proper-ties common to all animals, for example, ignoringtheir superficial differences, because of their relianceon concrete properties. Empirically, many research-ers found that until the age of 7 or 8, children foundit difficult to classify items into categories such asanimals, vehicles, or furniture. Instead, they oftenseemed to group items together by noncategoricalrelations, such as grouping a man and a car, sayingthat the man would drive the car. Thus, perhapschildren’s knowledge of the world is organized

around such kinds of relations rather than adultlikecategories.

We briefly review this literature along with morerecent claims that children do form abstract con-cepts. We argue that the opposition between twodifferent kinds of classification has oversimplifiedour understanding of children’s conceptual abilitiesand that children may be able to use simultaneouslyboth categorical and other kinds of relations. Webegin by describing the different forms of conceptualorganization that past research has identified.

Thematic, Script, and Taxonomic Categories

Researchers in conceptual development haveidentified several category types that children mayuse to classify things. Thematic categories groupobjects that are associated or have a complementaryrelationship; they often are spatially and temporallycontiguous. For example, a dog and its leash form athematic pair because the leash restrains the dog;cereal and a bowl do as well because the bowlcontains the cereal during eating. It is important tonote that the dog and leash are not similar; the cerealand bowl do not share many properties. Becausethematic relations are readily observed and have anassociative basis, they might be easy for children toidentify. Script categories are formed when itemsplay the same role (not complementary roles) in ascript. A script is a schema for a routine event (e.g.,going to the doctor, seeing a movie). For example,eggs and cereal are both in the script category ofbreakfast foods, not because they are spatially ortemporally contiguous (like the leash and dog) but

r 2003 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2003/7406-0014

Simone P. Nguyen, Department of Psychology, University ofNorth Carolina at Wilmington; Gregory L. Murphy, Department ofPsychology, New York University.This research was supported by a National Science Foundation

Graduate Research Fellowship, University of Illinois DissertationFellowship, and Grants MH41704 (National Institute of MentalHealth) and SBR 97-20304 (National Science Foundation). We aregrateful to Brian Ross for detailed advice on this research andconstructive feedback on an earlier draft. We are grateful to KarlRosengren for helpful input and Renee Baillargeon, Kevin Miller,and Cameron Gordon for discussions regarding this research.Special thanks go to Kristy Brooks, Amy Gaither, MychelleDenney, and Kevin Frederick for their invaluable researchassistance. We also thank the children, parents, and teacherswho were involved in this research.Correspondence concerning this article should be addressed to

Simone Nguyen, Department of Psychology, University of NorthCarolina, 601 South College Road, Wilmington, NC, 28403-5612.Electronic mail may be sent to [email protected].

Child Development, November/December 2003, Volume 74, Number 6, Pages 1783 – 1806

because they play the same part in an event schema(what is eaten at breakfast). (As a result, some callthese slot-filler categories.)

Taxonomic categories are so called because theyare usually organized into hierarchies of increasinglyabstract categories, such as terrier-dog-mammal-animal. However, for the present purposes, thecritical aspect of taxonomic categories is that theyare based on common properties or similarity. Dogsare in the same category because they have four legs,bark, breathe, have fur, and so on. (These commonfeatures allow a hierarchical structure in which morespecific categories have all the properties of moregeneral categories plus further, distinguishing prop-erties; see Murphy, 2002, chap. 6). Most categoriespicked out by common nouns are taxonomic: chair,fish, telephone, cloud, party, and vehicle, for exam-ple, are all taxonomic categories whose memberstend to share certain properties. In sum, taxonomiccategories are based on shared properties or, moregenerally, similarity among the category members.

Early work on concepts assumed that taxonomiccategories were the only, or the only ‘‘correct,’’ formof classification possible (as clearly seen in Hull,1920; or Inhelder & Piaget, 1964, p. 7). Thus, the find-ing that preschool children often group objects usingthematic or script categories suggested that theypossess a primitive conceptual system. Althoughthematic and script categories are different, theyshare the property that they are primarily defined byexternal relations rather than by internal properties(Markman, 1989, p. 21). The dog and leash do notboth bark, are not both made of leather, do not botheat meat, and so on, yet they form a coherentgrouping because the leash is spatially and tempo-rally contiguous to the dog during its walk.Similarly, an egg and cereal are both breakfast foodsbecause they have a common relation to an externalevent (eating breakfast), not because they both comefrom a chicken or contain wheat. In contrast, ataxonomic category such as animals is not definedby a situation in which an object occurs, nor itsrelation to other kinds of objects. Instead, somethingis an animal if it has properties such as being alive,reproducing, breathing, and so on. It is possible thatthe external relations involved in thematic and scriptcategories are easier for children to identify than thesimilarities underlying taxonomic categories (seeMarkman’s, 1989, discussion of collections). Lucar-iello, Kyratzis, and Nelson (1992, p. 980) havesuggested that script categories are easier to formbecause they are based on experienced events ratherthan abstractions across items that may not occurtogether.

As we noted, preschool children tend to groupitems together in ways other than the usualtaxonomic category. For example, several researchersfound that young children group items based onthematic relations (Greenfield & Scott, 1986; Kagan,Moss, & Siegel, 1963; Olver & Hornsby, 1967; Smiley& Brown, 1979). Other studies found that youngchildren group items based on their role in scriptsand that script categories may be important in othertasks such as memory and free association (Nelson,1986, 1988; Lucariello et al., 1992). These studiesoften found that children develop taxonomic cate-gories by around age 7 or 8.

Proposals for a shift in the nature of children’sclassification have been criticized in three ways.First, several studies found that young children canform taxonomic categories. Clearly, they have nodifficulty in learning basic-level categories such asdog, telephone, chair, tree, and so on, even at veryearly ages (Horton & Markman, 1980; Rosch, Mervis,Gray, Johnson, & Boyes-Braem, 1976). But childrencan learn and use even superordinate categoriesunder certain conditions (e.g., Markman, 1989;Waxman & Gelman, 1988). Second, several studiesfound that children are not as strongly interested inthematic relations as previously believed, contrary tothe traditional claim (Denney, 1975; Denney &Moulton, 1976; Waxman & Namy, 1997). Third,recent studies found that adults use thematicand script categories when the relations are suf-ficiently strong (Lin & Murphy, 2001; Murphy, 2001;Ross & Murphy, 1999). These results with adultsmake it difficult to argue that such categories areprimitive.

In the adult concept literature, there is increasingrecognition of the fact that items can be cross-classi-fied into more than one category (Barsalou, 1991;Murphy, 1993; Ross & Murphy, 1999). In additionto the usual taxonomic categories that people use,such as dog and animal, they know a variety of otherkinds of categories that could include the sameitems, such as other, more specialized taxonomiccategories (e.g., carnivore, pet), ad hoc or goal-derived categories (e.g., things to carry out of aburning house; Barsalou, 1983, 1991), thematiccategories, and script categories. Although taxo-nomic categories are particularly useful, othermodes of categorization can also be very useful.Markman (1989) pointed out that children mustlearn thematic and script categories such as thingsthat are found at a birthday party, things you bringto school, breakfast foods, and so on, as part oflearning about the activities in their culture. Thechild who brings a bat but no ball to the playground

1784 Nguyen and Murphy

is going to have a hard time playing baseball. Ratherthan outgrowing such categories, adults may con-tinue to use them alongside taxonomic categories(Murphy, 2001).

The present study addressed the issue of cross-classification in children’s concepts, focusing on taxo-nomic and script categories of foods. We used thedomain of food as a test case for two main reasons.First, food is cross-classified into many categories andinto taxonomic and script categories in particular. It isstraightforward to identify taxonomic categories forfoods because they have familiar names such as fruitand meat. Script categories are not as entrenched,although some do have familiar names, such as snackand dessert. Furthermore, several foods are found inmultiple script categories, as muffins might be eatenat breakfast or as a snack, for example. Ross andMurphy (1999) found that adults spontaneouslygrouped foods both taxonomically and accordingto script relations. For example, adults in their studyconsidered a bagel to be both a bread (taxonomic)and breakfast food (script). In fact, food is the onlydomain that we know of that has been shown tohave both strong taxonomic- and script-basedcategories in adults. However, it has yet to be seenwhether young children also cross-classify foods.

The taxonomic categories we tested in the presentstudy were those that arose from Ross and Murphy’s(1999) results, and they were superordinate cate-gories such as meats, fruit, and vegetables. (Thebasic level for our food items, as shown by naming,would be more specific categories such as cookie,chicken, or broccoli.) Thus, this level of taxonomicclassification is similar to that studied in previousstudies of conceptual development in object cate-gories, such as vehicles, tools, insects, and wateranimals (e.g., Mandler, Bauer, & McDonough, 1991;Rosch et al., 1976; Smiley & Brown, 1979; Waxman &Gelman, 1988).

Second, a limitation of past research on categor-ization is the relative lack of focus on domains thatchildren know a lot about and that are connected totheir greater knowledge of the world. Althoughmany studies have examined real-world categoriessuch as animals (e.g., Carey, 1985; Keil, 1989; Mark-man, 1989), these categories are still at the peripheryof most children’s daily lives and everyday thinking,especially those who live in urban settings (Medin &Atran, in press; Coley, 2000). The domain of food,however, is one that is highly central to children’slives and everyday thinking (Birch, Fisher, &Grimm-Thomas, 1999) and is an integral part ofchildren’s broader knowledge about health andillness (Rozin, 1990).

Although there is clearly a limitation in usingcategories from a single domain (addressed furtherin the General Discussion), we focused on foodbecause of the cross-classification found in adultstudies. If we discover that children can also cross-classify in this domain, it will demonstrate that thereis no necessary progression from one form ofcategorization to the other, showing that childrenhave the potential for conceptual flexibility.

The Present Experiments

Our study investigated script and taxonomiccategories using the domain of food. Our goal wasto discover whether children are able to categorizesimultaneously foods using these two very differentbases and, if not, to find out which one developsfirst. Given that adults are clearly able to classifyfoods in two different ways (Ross & Murphy, 1999)and that both script and taxonomic relations areinformative, we expected to find considerable over-lap in children’s use of these categories rather thanone of them replacing the other. We focused on tasksthat directly reflect concept use, namely, categoriza-tion and category-based induction, rather thanindirect measures of categories, such as word asso-ciation or memory (cf. Lucariello & Nelson, 1985).If children are able to use both forms of categor-ization, this will indicate a flexibility in their concep-tual skills that has been ignored in the debate over ashift from one form of categorization to another.

Because generalization from one object to otherobjects is a critical function of categorization, inExperiments 4 and 5, we also investigated concep-tual flexibility in an induction task. For example, ifyou believed that bread is unhealthy, would you alsoconclude that muffins are unhealthy? Althoughtaxonomic categories have been heavily studied incategory-based induction tasks, script categorieshave seldom been studied. Ross and Murphy(1999) found that adults use different food categoriesto make different types of inductions. Adultsdemonstrated inductive selectivity by making moreinferences about the composition of foods fortaxonomic categories and more inferences aboutwhen foods are eaten for script categories (see alsoHeit & Rubinstein, 1994). Few developmental stud-ies have addressed whether children use theircategory knowledge to generalize properties selec-tively. Gelman and Markman (1986) and Gelman(1988) showed that children generalize internalphysiological structure across category membersbut do not generalize superficial properties thatdiffer across examples (e.g., weight or accidental

Cross-Classification 1785

properties such as ‘‘is dirty’’). However, children donot always show a sensitivity to which propertiesshould be induced for different categories (e.g.,Gelman, 1988; Gelman & O’Reilly, 1988). Laterresearch found some sensitivity to category type.For example, Kalish and Gelman (1992) found thatwhen inferring information about the properties offragility and texture, 3- and 4-year-olds consideredmaterial composition as a crucial category. Forexample, when asked if a wooden pillow was hardor soft, children focused on the material as opposedto the object category and answered that it was hard(see also Deak, 2000). The present investigation addsto this body of research by examining children’sselective use of taxonomic and script categories ininductive inferences.

In addition to the more familiar types of cate-gories, we also explored evaluative categories, inwhich foods have the same positive or negativeevaluation. Evaluative categories are not based onsimilarity, like taxonomic categories, nor do theynecessarily occur in similar settings. For example,the evaluative category of healthy foods may includefruits, vegetables, dairy products, and fish, whichare not in the same taxonomic category and are eatenin a wide variety of settings. Although affectivereactions to different stimuli have long been studiedwithin social psychology (e.g., Duckworth, Bargh,Garcia, & Chaiken, 2002; Zajonc, 1980), stimuli thatreceive the same affective response have seldombeen studied as categories and tested in conceptualtasks (Niedenthal, Halberstadt, & Innes-Ker, 1999, isa recent exception). Our studies provide an explora-tory look at evaluative categories in a developmentalcontext.

Evaluative categories seem especially pertinent tothe domain of foods, in which people identify foodsas disgusting (e.g., grasshopper, brussel sprouts),dangerous (e.g., wild mushrooms), or healthy (e.g.,spinach; Rozin, Hammer, Oster, & Horowitz, 1986).We focused on healthy and unhealthy foods, whichcould be drawn from the set of foods used in ourother categories (unlike dangerous foods, for exam-ple) and which are salient categories for adults. Notethat these categories are not based purely onemotion, as foods may receive their evaluation fordifferent reasons (ice cream because it tastes good,broccoli because it is healthy). Thus, evaluative foodcategories are not purely affective but have con-siderable content as well. This is similar to theevaluative category of ‘‘weed trees’’ documented byMedin, Lynch, Coley, and Atran (1997), which areassociated with properties such as being weak-wooded and dropping many twigs. Whether chil-

dren classify foods evaluatively and whether theyuse such categories for induction have apparentlynot been investigated.

Three questions guided the present examination.First, do children have taxonomic, script, andevaluative categories, and do they develop atdifferent rates or simultaneously? Second, do chil-dren cross-classify food into these categories? Thatis, can children classify an individual item into morethan one category? These first two questions are ofparticular interest in the context of research that hasclaimed that children undergo a shift in theircategorization style. Third, do children selectivelyuse taxonomic, script, and evaluative categories tomake inductive inferences about food? This is animportant question because inductive inferences area basic function of categories, and the possibility thatchildren cross-classify food may pose a challenge tothe induction process.

To investigate these questions, we conducted aseries of five experiments. Experiment 1 examinedwhether 4- and 7-year-olds and adults classify foodsinto taxonomic, script, and evaluative categories.The task was extended to 3-year-olds in Experiment2. Experiment 3 examined children’s classification ofthe same foods into more than one category.Experiments 4 and 5 examined 4- and 7-year-olds’and adults’ use of script and taxonomic categories asthe basis for inductive inference.

Experiment 1

Experiment 1 was designed to discover whetherchildren classify foods into taxonomic, script, andevaluative categories and at what age this abilityappears. Categorization studies have tended to useconflict picture triads, in which there is a target anda taxonomic and thematic choice, for example, pittedagainst each other (e.g., dog: leash vs. cow). In thesestudies, children were required to select either thetaxonomic or thematic choice, but not both. Theresults have generally shown that young childrenfavor the thematic or script choice over the taxo-nomic choice, whereas older children and adultsfavor the taxonomic choice (e.g., Lucariello et al.,1992). However, the conflict picture triads do notprovide children the opportunity to select bothchoices even though both are related to the target;therefore, these findings may reflect preferencesrather than knowledge. Indeed, Smiley and Brown(1979) found that adults and children as young as 6knew both the taxonomic and thematic relations intheir stimuli, even if they consistently chose one ofthem in conflict trials. Clearly, selecting a thematic


choice does not show that children do not under-stand the taxonomic relation. Experiment 1 includedconflict triads as a comparison with previousresearch.

In addition to the conflict triads, Experiment 1included nonconflict triads that did not pit twocategories against each other. The nonconflict triadswere composed of a target; an alternative that shareda taxonomic, script, or evaluative relationship withthe target; and an unrelated alternative, for example,corn: carrot (taxonomic choice) versus juice (unre-lated choice). This task permitted independent testsof children’s knowledge of taxonomic categories and(in other trials) script relations. If children havemultiple categories, then they should select thetaxonomic alternatives on the taxonomic trials, scriptalternatives on the script trials, and the evaluativealternatives on the evaluative trials. Category per-formance on one category type does not affectperformance on the others, as in conflict triads. Ifthe traditional view of children’s concepts is correct,script categories should appear before taxonomiccategories.

Evaluative categories have not been tested beforein this paradigm, and one can derive opposingpredictions about them. Perhaps evaluative cate-gories, which do not require a detailed analysis ofthe category’s content, are a more primitive form ofcategorization that will appear before taxonomiccategorization. Alternatively, perhaps children lackthe knowledge of food and physiology that under-lies health judgments about food; therefore, suchcategories will appear after taxonomic categories. Or,perhaps evaluative categories will appear at thesame time as taxonomic categories.

In any test of this sort, there is always somecategory that includes the ‘‘wrong’’ item. Forexample, in the taxonomic versus thematic triad ofitems such as needle: thread versus pin (Smiley &Brown, 1979), even the thematically related items(needle – thread) are also members of general taxo-nomic categories such as artifacts or physical objects.However, children and adults both prefer to use aspecific category when it is available rather thanrelying on abstract categories, such as artifacts orliving things. As all the items in our studies werefoods, participants were not able to choose onealternative over the other based on the fact that itand the target item are both foods. They had to usemore specific relations, either taxonomic or scriptbased. Using the food category as a basis forresponding would result in random choices, and itwill be seen that this never occurred.

Method

Participants. Participants were 32 children: sixteen4-year-olds (M5 4, 5, range5 4,0 –5,0; 7 boys and 9girls) and sixteen 7-year-olds (M5 7;1, range5 6,8 –7,6; 8 boys and 8 girls). Sixteen adults alsoparticipated (M5 20; 4 males and 12 females). Anadditional set of 15 adults also provided ratings ofthe stimuli. The participants were predominatelyEuropean-American and were recruited from amiddle-class community located in the MidwesternUnited States. In particular, child participants wererecruited from schools, and adult participants wererecruited from a university.

Materials. We used several resources to help usgenerate a list of foods that were familiar to children.We referred to children’s picture and word-learningbooks. Research has found that even infants andtoddlers produce and comprehend a large repertoireof food and drink words (Fenson et al., 1994). Wealso relied on child and parent informants. Firstgraders not involved in the primary experimentswere informally interviewed during their lunchperiod at an elementary school. Children were askedto give examples of taxonomic and script categoriesof foods (e.g., ‘‘Can you give some examples of dairyproducts?’’ ‘‘Can you give some examples of break-fast foods?’’) and were asked whether they werefamiliar with particular foods (e.g., ‘‘Do you knowwhat a muffin is?’’). Also, we informally interviewedparents of preschool- and school-aged children whowere not involved in the experiments. Parents wereasked about their child’s experience and knowledgeof particular foods (e.g., ‘‘Does your child knowwhat Twinkies are?’’ ‘‘Does your child eat hot dogsat lunch?’’). Research suggests that children know alot about foods, especially because the typical childeats three meals a day, plus snacks, and is often ex-posed to advertisements for foods (Birch et al., 1999).

Combining these sources with the results of Rossand Murphy (1999) allowed us to identify items thatshare taxonomic, script, or evaluative categorieswithout having a salient relation in one of the othertypes. It was not necessary for our design that eachtarget food be in only one script (or taxonomic)category. (Indeed, as we have pointed out, objectsare seldom in only one category because of cross-classification.) All that was necessary was that thetarget category be salient enough to be identifiedwithin the triad. That is, if muffin is paired withpopcorn, their common membership in the snackcategory will become salient if participants encodeboth of them as snacks. If muffin’s membership in


other possible script categories makes categorizationmore difficult, we should find delayed categoriza-tion of script relative to taxonomic categories. Asdescribed, we took various steps to ensure that ouritems were readily identifiable as members of thetested categories (Experiments 4 and 5 provideverification that we were successful in this goal).

One hundred and two 2.5 in.� 3 in. (6.35 cm�7.62 cm) color photographs of food were collectedfrom Internet sources and food products. Thephotographs could not include more than one food(e.g., tray of fruit), and the target food needed to beeasily recognizable. Photographs were arranged into34 triads consisting of a target and two choices. Anattempt was made to select pictures that did nothave a strong resemblance so that perceptualsimilarity would not dominate the choices.

Each triad was printed on a separate sheet of8.5 in.� 11 in. (21.6 cm� 27.9 cm) white paper withlabels of the foods. There were 12 taxonomic, 12script, 4 evaluative, and 6 conflict triads. Each of thefoods in the triads had a typed label. The taxonomictriads included a target (e.g., corn), a taxonomicchoice (e.g., carrot), and an unrelated choice (e.g.,juice). The taxonomic categories were familiarcategories such as fruit, meat, and breads. The scripttriads included a target (e.g., bacon), a script choice(e.g., pancake), and an unrelated choice (e.g., Jell-O).Script categories were based on setting or time ofeating, such as breakfast foods. The evaluative triadsincluded a target (e.g., Twinkie), an evaluative choice(e.g., Cheetos), and an unrelated choice (apple).Evaluative categories consisted of prototypicalhealthy or unhealthy foods.

The related and unrelated choices were counter-balanced across triads: The script choice for one triadwas the unrelated choice for another script triad.Thus, the results could not be explained by apreference for the related items. The conflict triadsincluded a target (e.g., ice cream), a taxonomicchoice (e.g., cheese), and a script choice (e.g., cookie).See Appendix A for a complete list of the triads.

We asked 10 adults to rate the targets and theircategory matches to ensure that our items weretypical and familiar category members and thatthese factors did not vary systematically across thecategory types. The scales ranged from 1 (low) to 7(high). For the typicality ratings, adults were told torate each food in terms of how typical an instance ofthe given category it is. The ratings were uniformlyhigh and did not differ significantly among thetaxonomic, script, and evaluative categories (M5 5.99,SD5 0.66; M5 5.80, SD5 0.69; M5 6.23, SD5 0.85,respectively). For the familiarity ratings, adults rated

the degree to which they were familiar with afood as belonging to a given category. Again, therewere no significant differences between the taxo-nomic, script, and evaluative categories (M5 6.25,SD5 0.62; M5 5.90, SD5 0.97; M5 5.70, SD5 1.45,respectively).

In addition, 5 adults rated the visual similarity ofeach target and its two choices separately (e.g.,bacon and pancake, bacon and Jell-O). Participantswere told that ‘‘we are interested in whether the foodpairs look like each other in these particularphotographs.’’ The scale ranged from 1 (not at allvisually similar) to 7 (very visually similar). There wasnot a significant difference between the similarity oftargets and their category choices (M5 1.88, SD5 0.40)and the similarity of targets and the unrelated items(M5 1.38, SD5 0.21). Given this nonsignificantdifference and the low absolute level of visualsimilarity in both groups, we were not concernedthat the appearance of the pictures would biaschildren’s answers.

The 34 triads were placed in a three-ring binder.There were three blocks of trials with a fixedpresentation: (a) warm-up; (b) taxonomic, script,and evaluative triads; and (c) conflict triads. Thetrials within each block were presented in one of tworandom orders. The conflict trials were always testedlast because of their potential for biasing other trials.The conflict trials force participants to choosebetween taxonomic and script categories, and thismight cause them to emphasize one and ignore theother in the nonconflict trials. Because all of thecategory types occurred in the nonconflict trials, thisdid not bias any response in the conflict trials.

Procedure. Children were tested individually forapproximately 15min in a quiet area of their school.The first block of the experiment consisted of twoeasy warm-ups to familiarize children with theprocedures. For example, children heard, ‘‘This is acinnamon donut. This is a chocolate donut and thisis broccoli. Now, is the chocolate donut or thebroccoli the same kind of food as the cinnamondonut?’’ If children selected the unrelated picture(i.e., broccoli), the researcher corrected the child andrepeated the warm-up until the child understood theprocedures. The second block consisted of 12taxonomic, 12 script, and 4 evaluative trials. Thethird block consisted of 6 conflict trials. In each ofthese trials, the researcher labeled the pictures andasked which food is the same kind of food as thetarget, as in the warm-up task. If a child did notmake a clear choice, the researcher repeated thequestion. The researcher did not provide childrenwith feedback about their answers.


Adults participated in a paper-and-pencil versionin a large group. Adults did not see the foodphotographs but were asked to read and respondto names of foods.

Results

We scored the nonconflict trials by assigning a 1 tochoices that belonged to the same category as thetarget and a 0 to the unrelated choice. For the conflicttrials, the taxonomic choices were assigned a 1 andthe script choices a 0. We then summed the trials intotheir respective category types and transformed thescores into proportions. Because the scores ofthe nonconflict trials reflect accuracy (sensitivityto the category relationship tested), and the scoresof the conflict trials reflect a preference (both choicesare ‘‘correct’’), we analyze them separately through-out this article.

Nonconflict triads. We conducted a 3 (age: 4-year-olds, 7-year-olds, adults) � 3 (category type: taxo-nomic, script, and evaluative) analysis of variance(ANOVA) for the nonconflict trials. The dependentvariable was accuracy of food choice. Accuracyincreased with age, as 4-year-olds were 70% accu-rate, whereas 7-year-olds and adults were 85%and 91% accurate, respectively, F(2, 45)5 14.12,MSE5 0.56, po.05. There was a significant differ-ence between 4-year-olds and the older groups butnot between 7-year-olds and adults, Tukey’s (HSD)test honestly significant different, po.05. There wasneither a significant main effect of category type nora Category� Age interaction. See Figure 1 for thepercentage of category choices by age and categorytype. All of the age groups performed reliably betterthan chance for each category type, ts(15)43.0,pso.05. Thus, children as young as 4 had all threecategory types.

To determine whether the results were due to oneor two foods, we separated out the different food

groups within each of the category types. Althoughthere are not enough items within each category toallow reliable cross-category comparisons, the ap-parent differences can serve as suggestions forfuture research. Table 1 lists each food categoryand each age group’s accuracy in classifying into it.The table suggests that some taxonomic, script, andevaluative foods are understood better than others.Breads and grains seemed to be particularly difficultfor the children, perhaps because they are perceptuallydiverse. Lunch foods and snacks seemed difficult forthe 4-year-olds, perhaps reflecting differences ineating habits across age groups. Children are morelikely to have hot meals at lunch than our collegestudent participants, and their snacks probably areless dependent on vending machines. Finally, 4-year-olds seemed to have limited awareness of healthyfoods. Perhaps they are more likely to hear com-ments such as, ‘‘You’ve eaten enough junk foodtoday’’ than to hear comments on the healthyproperties of foods.

Conflict triads. Results from the conflict trials weresimilar to past studies that have pitted two forms of

Fig. 1. Percentage of taxonomic, script, and evaluative categorychoices in Experiment 1.

Table 1

Experiment 1: Mean Accuracy (Standard Deviations) of Categorizing

Items Into Food Categories as a Function of Age

4-year-olds 7-year-olds Adults

Taxonomic

Beverages .87 (.028) 1.00 (0) 1.00 (0)

Bread and grains .56 (.025) .59 (.027) .93 (.017)

Dairy .68 (.030) .78 (.025) .78 (.025)

Fruit .84 (.023) 1.00 (0) .96 (.012)

Meat .78 (.031) 1.00 (0) .96 (.012)

Vegetables .71 (.031) .90 (.020) .93 (.017)

Script

Breakfast .68 (.040) .81 (.025) .62 (.034)

Lunch .59 (.037) .93 (.017) .96 (.012)

Dinner .71 (.036) 1.00 (0) .93 (.017)

Birthday .87 (.022) .96 (.012) .96 (.012)

Dessert .75 (.031) .87 (.022) .90 (.020)

Snack .68 (.035) .84 (.030) .81 (.030)

Evaluative

Healthy .56 (.030) .75 (.036) .90 (.020)

Junky .71 (.036) .78 (.036) .96 (.012)

Note. Four-year-olds and 7-year-olds were above chance on all ofthe taxonomic foods except breads and grains, ts(15)42, pso.05.Adults were above chance on all of the taxonomic foods, ts(15)44,pso.05. Four-year-olds were above chance on all of the scriptfoods except breakfast and lunch, ts(15)42, pso.05. Seven-year-olds were above chance on all of the script foods, ts(15)44,pso.05. Adults were above chance on all of the script foods exceptfor breakfast foods, ts(15)43, pso.05. Seven-year-olds and adultswere above chance on both the healthy and junky foods, ts(15)42,pso.05. Four-year-olds were above chance on junky evaluativefoods, but not on healthy evaluative foods, t(15)5 2.40, po.05.


categorization against one another. Seven-year-olds,t(15)5 3.22, po.05, were above chance on theirtaxonomic choices (M5 0.62, SD5 0.15), 4-year-olds(M5 0.44, SD5 0.10) and adults (M5 0.56, SD5 0.20)were not. Within children, then, these results showthe usual finding that children do not consistentlychoose taxonomic relations until around 7 years ofage. Although it is tempting to view the differencebetween 4- and 7-year-olds as reflecting a change incognitive abilities, 4-year-olds’ above-chance perfor-mance on the nonconflict trials argues against suchan explanation. That is, given that 4-year-olds were74% correct on the nonconflict taxonomic trials, wecannot conclude from the conflict trials that they didnot know the taxonomic categories. Similarly, such aconclusion would be absurd for adults. Instead, itseems likely that the 4-year-olds and adults (but notthe 7-year-olds) found the script relations as salientor important as the taxonomic relations.

Discussion

The goal of this experiment was to examinewhether children have multiple category types,including taxonomic, script, and evaluative cate-gories. The answer to this question is yes. By age 4,children do have taxonomic, script, and evaluativecategories. The results of the present study runcounter to past claims that 4-year-olds undergo aqualitative shift in their categorization abilities (e.g.,Greenfield & Scott, 1986; Kagan et al., 1963;Lucariello et al., 1992; Nelson, 1986, 1988; Olver &Hornsby, 1967). An explanation for the difference inthese results is that past research has typically usedconflict trials in which taxonomic and script cate-gories are pitted against each other (e.g., Lucarielloet al., 1992; Sell, 1992). Our conflict triads also founda (nonsignificant) preference for script categories in4-year-olds. However, our nonconflict triads allowedchildren the opportunity to categorize foods inmultiple ways. Children as young as 4 years of ageclassified foods into taxonomic, script, and evalua-tive categories on the nonconflict trials. There was nosign that script categories are acquired beforetaxonomic categories.

Thus, the conclusion of prior research that young-er children think in terms of script or thematiccategories before they understand taxonomic cate-gories is based on a conflation of preference andcategorization. Seven-year-olds show a preferencefor taxonomic over script categories that 4-year-oldsdo not, as shown in the conflict triads (and byLucariello et al., 1992; Smiley & Brown, 1979), butchildren at both ages understand both kinds of

categories and use them in categorization, as shownin the nonconflict triads.

One possible concern about our results is whetherwe were successful in identifying items that were intaxonomic categories but not script categories. Forexample, perhaps butter and yogurt co-occur insome script that we did not successfully identify inour stimulus construction procedure. This possibil-ity is addressed by our induction experiments (seeDiscussion of Experiment 5).

Experiment 2

Experiment 2 examined the early emergence oftaxonomic and script categories by extending amodified version of Experiment 1 to 3-year-olds.Experiment 1 showed that by 4 years of age, childrenhave taxonomic, script, and evaluative categories,but it leaves open the possibility that scriptcategories emerge before taxonomic categories atan earlier age. Examining the development of thesecategories in younger children could help addressclaims of the developmental shift from script totaxonomic categories. Perhaps 3-year-olds will showthe ability to categorize foods only into scriptcategories, as this shift would predict.

Method

Participants. Participants were sixteen 3-year-olds(M5 3,6, range5 3,1 –3,8; 9 boys and 7 girls).Children were drawn from the same demographicpopulation as in Experiment 1.

Materials. The 12 taxonomic and 12 script picturetriads from Experiment 1 were used with 3-year-olds. To accommodate the attention spans of 3-year-olds, we did not include the evaluative or conflicttriads. In addition, a large doll named Jane was usedto make the experiment gamelike for the youngerchildren.

Procedure. Children were tested in a quiet area oftheir day care or in a laboratory setting. Theresearcher told children, ‘‘Jane needs help figuringout which foods are the same kinds of food.’’ Aftercompleting an easy warm-up trial borrowed fromExperiment 1, the children were presented with thetaxonomic and script trials. For each of the trials,the researcher labeled the foods and asked, ‘‘Is Yor Zthe same kind of food as X?’’ These trials werepresented in one of two random orders.

Results and Discussion

We scored the data as in Experiment 1. The 3-year-olds were 60% correct on the taxonomic trials


and 60% correct on the script trials, both of whichdiffered from chance, t(15)5 2.5, po.05; t(15)5 2.05,po.05. We also examined the individual taxonomicfoods: beverages (M5 0.56, SD5 0.30), bread andgrains (M5 0.43, SD5 0.44), dairy products (M5 0.78,SD5 0.31), fruits (M5 0.65, SD5 0.35), meats(M5 0.75, SD5 0.36), and, vegetables (M5 0.43,SD5 0.40). Similarly, we examined the individualscript foods: breakfast (M5 0.53, SD5 0.28), lunch(M5 0.40, SD5 0.37), dinner (M5 0.62, SD5 0.38),birthday (M5 0.78, SD5 0.36), dessert (M5 0.62,SD5 0.28), and snack (M5 0.62, SD5 0.38). Three-year-olds were above chance (pso.05) only on thetaxonomic categories of dairy products and meatsand on the script category of birthday foods, though,as before, there is little power in the tests ofindividual categories.

We also compared the 3-year-olds’ data with thedata from Experiment 1 by combining them in a 4(age) �2 (category type) ANOVA. There was a maineffect of age, F(3, 60)5 19.53, MSE5 0.65, po.05.Three-year-olds were significantly different from4-year-olds, 7-year-olds, and adults, po.05, byTukey’s HSD test, demonstrating that even though3-year-olds were above chance, there is significantcategory learning from 3 to 4 years. There wasno main effect of category nor a Category�Ageinteraction.

These results suggest that by 3 years, childrenhave both taxonomic and script categories, and thatone type of category does not dominate the other. Inconjunction with the results of Experiment 1, theseresults show a steady improvement in children’sacquisition of different categories from 3 to 7 years.The use of nonconflict trials revealed that taxonomicand script categories are emerging simultaneously in3-year-olds. These results do not reveal a develop-mental shift from script to taxonomic categories.Given that 3-year-olds’ level of performance was justabove chance, at 60% on both the taxonomic andscript trials, the possibility that a shift occurs beforethis age seems unlikely.

Experiment 3

Based on the results of Experiments 1 and 2, weconcluded that children have multiple categories inthe domain of food. However, what we do not knowis whether children realize that a single food canbelong to more than one of these categories. In theprevious experiments, we did not systematically testeach food in multiple categories. There are indeedmany ways to categorize a single food, as in the case

of muffins, which can be considered a grain, a snack,a breakfast or lunch food, and a healthy, or even ajunk food, depending on its ingredients. Ross andMurphy (1999) found that adults cross-classifyfoods; apple was categorized as a fruit, snack, andhealthy food. The question of whether childrenwould do so addressed their flexibility as categor-izers. Although Experiments 1 and 2 showed theycan use two different categorization systems withina single domain, the experiments did not show thatchildren can treat the same item as a member of twodifferent categories, especially two categories withvery different bases. True cross-classification re-quires that different systems of categorization beapplied to the same items.

The purpose of Experiment 3, then, was toexamine children’s cross-classification of food. Be-cause children were best at identifying the script andtaxonomic categories (see Table 1), we constructednonconflict triads for these two category types, usinga common set of target items. Specifically, twelvefoods were used as the targets in both the taxonomic(e.g., ice cream: butter vs. sandwich) and script (icecream: cake vs. potato) triads. If children cross-classify, they should select the choice that shares acategorical relationship with the target in bothtaxonomic and script triads. A limitation on thisprediction is that children might be unfamiliar withone of the categories and could therefore fail tocross-classifyFnot because of any problem withcross-classification, per se, but because of ignorance.For example, a child might be perfectly willing toclassify ice cream into different categories but simplyis not aware that it is a dairy food. And, indeed, inExperiment 1, 4-year-olds made about 30% errorsboth in taxonomic and script categories; therefore,one could hardly expect that they would correctlyclassify foods into both categories near to 100% ofthe time.

How, then, should cross-classification be tested? Itis clearest to frame the problem in the negative:What will children do if they cannot cross-classify? Ifcross-classification is difficult, there should be anegative relation between the two tasks, such thatitems that are correctly classified taxonomicallyshould be incorrect on the script trials, and viceversa. Thus, we gave children the same item in twodifferent classification tasks and performed a testof independence to discover whether there is anegative relationship between taxonomic andscript classification. That is, we hoped to discoverwhether children who think of ice cream as a dessertwould then find it difficult to think of it as a dairyproduct.


Method

Participants. Thirty-two children participated: six-teen 4-year-olds (M5 4,0, range5 4,0 –5,0; 6 boysand 10 girls) and sixteen 7-year-olds (M5 7,6,range5 7,0 –8,0; 6 boys and 10 girls). Sixteen adultsalso participated (M5 19,8; 7 males and 9 females).A separate group of 15 adults also completed ratingsof the items. Participants were recruited from thesame demographic population as in the previousexperiments.

Materials. The pictures of foods were selected andprepared the same way as in Experiment 1. Twelvefoods were used as the targets in both the 12taxonomic and 12 script triads. For example, thetaxonomic triads included a target item (e.g., icecream), a taxonomic choice (e.g., butter), and anunrelated choice (e.g., sandwich). The script triadsincluded a target (e.g., ice cream), a script choice(e.g., cake), and an unrelated choice (e.g., potato).The categorical and unrelated choices were counter-balanced across triads (e.g., the categorical choice forone triad became the unrelated choice for anothertriad). See Appendix B for a complete list of thetriads.

As in Experiment 1, 10 adults rated the typicalityand familiarity of the targets and their categorymatches to ensure that these factors did not varyamong category types. The scales again ranged from1 to 7. There was no significant typicality differencebetween the taxonomic and script categories(M5 6.02, SD5 0.47; M5 5.69, SD5 0.72), nor wasthere a familiarity difference (M5 6.11, SD5 0.67;M5 5.66, SD5 0.84).

As in Experiment 1, 5 adults also completed visualsimilarity ratings of the stimuli to ensure that neitherof the choices in each triad appeared significantlymore similar to the target. The level of visualsimilarity was low overall, and there was nosignificant difference between the targets and theircategory choices (M5 2.4, SD5 0.67) and the targetsand their unrelated choices (M5 1.85, SD5 0.26).

The 24 triads were placed in a three-ring binder.There were two blocks. The first block included 12targets placed into six taxonomic triads and sixscript triads. The second block included the same 12targets but in triads from the untested category. Theorder of trials was random within blocks and wasreversed for approximately half of the participants.

Procedure. Children were tested individually in aquiet area of their school. Children initially com-pleted a warm-up task in which they were asked tocategorize a cat into two different categories. Thefirst warm-up was: ‘‘This is a tabby cat. Is a bear or a

Persian cat the same kind of thing as the tabby cat?’’The second warm-up was: ‘‘This is a tabby cat. Is adog or elephant the same kind of thing as the tabbycat?’’ If children selected the incorrect pictures (i.e.,bear or elephant), the researcher corrected thechildren and repeated the warm-up until it wasclear that they understood the procedures. Next,children completed the 24 test trials. For each ofthese trials, the researcher labeled the pictures andasked which choice was the same kind of food as thetarget, as before. If a child did not make a clearchoice, the researcher repeated the question. Nofeedback was given in the test trials.

Adults participated in a paper-and-pencil versionin a group. They did not see the food photographsbut were asked to read and respond to names offoods.

Results

We assigned a 1 to choices that belonged to thesame taxonomy or script category as the target and a0 to unrelated choices, and we converted the scoresinto proportions. We then conducted a 2 (categorytype: taxonomic and script) � 3 (age: 4-year-olds,7-year-olds, and adults) ANOVA. As in the previousstudies, performance on the taxonomic and scripttrials was relatively similar for each age group.There was no effect of category type. Four-year-oldswere 66% accurate on the taxonomic and 60% on thescript trials. Seven-year-olds were 89% accurate onthe taxonomic and 83% on the script trials. Adultswere 89% accurate on both the taxonomic and scripttrials. Clearly, there is no sign of a script-to-tax-onomic shift.

There was a main effect of age, F(2, 45)5 21.24,MSE5 0.65, po.05. Overall, 4-year-olds were lessaccurate than 7-year-olds and adults, who wereabout equally accurate, as shown by Tukey’s HSDtests, pso.05. We also compared participants’ per-formance to chance (50%) and found that all agegroups were above chance on both the taxonomicand script trials, ts(15)42.3, pso.05.

The accuracy of the 7-year-olds and adults was sohigh that participants in these groups must havebeen classifying the same items into both taxonomicand script categories. However, at 66% and 60%accuracy for taxonomic and script triads, the 4-year-olds might have been classifying different itemscorrectly in the two category types much of the time.Thus, it is possible that they were not cross-classifying individual items to any great degree.

We investigated this possibility by asking ifchildren were avoiding cross-classifying items. That


is, if they got ice cream correct on the script triad,this should have decreased their likelihood ofgetting it correct on the taxonomic triad, reflectingtheir inability to think of ice cream as being in twodifferent categories. In short, if children cannotcross-classify, their performance should be nega-tively correlated across triads. We set up as a nullhypothesis the possibility that classification on thetwo category types was independent. Under thishypothesis, we can calculate the number of itemsthat each participant should get correct on bothproblems, to wit, P (correct taxonomic) �P (correctscript) �12 (items). The result of this calculation wasthat children should get 4.87 items correct on bothtests, on average. If the children were unable tocross-classify, the number of items they got correcton both tests should have been significantly less thanthis, as they should have tended to get differentitems correct on the two tests.

In fact, the average number of items that a childcorrectly classified into both categories was 4.75,virtually identical to the number predicted by theindependence assumption, p4.05. Clearly, childrendid not exhibit any tendency to classify items intoonly one of the categories. Instead, it seems likelythat their knowledge of the two kinds of categoriesdevelops independently.

Discussion

These results provide further evidence that 4-year-olds are not limited to script categories and thatchildren do not undergo a shift from script totaxonomic categories with development. The resultsextended the findings of Experiments 1 and 2 bydemonstrating that children as young as 4 canclassify the same object into taxonomic and scriptcategories. These results suggest that young chil-dren’s categorization of objects is flexible. Pastresearch suggesting that children undergo a shiftfrom either thematic to taxonomic (e.g., Inhelder &Piaget, 1964) or from script to taxonomic categories(Nelson, 1988) has typically pitted two categorytypes against each other while ignoring the possibil-ity that children might be able to use more than oneform of categorization. However, a few recentstudies have shown that children are more flexiblein their categorization than once assumed. Theresults of Experiment 3 are consistent with thesestudies. For example, Blaye and Bonthoux (2001)recently showed that by age 5, children can flexiblycategorize an object into a taxonomic and thematiccategory when they are cued to do so. In this study,children were allowed to categorize spontaneously a

target item taxonomically or thematically. One weeklater, children were presented with a drawing of ascene that primed children to use a different form ofcategorization than before (e.g., a thematic scene ifthe child had spontaneously categorized the itemtaxonomically). Our results show that children cancross-classify even without any hint or prime. Otherresearch has also found that even 3-year-olds havedifferent words for a single object at varying levels ofa taxonomy to which the object belongs (Blewitt,1994; Deak & Maratsos, 1998; Waxman & Hatch,1992), although superordinate categorization is notstrictly speaking cross-classification.

In the current study, children categorized itemsinto taxonomic and script categories separately.Thus, the results show that at minimum childrencan represent an item in terms of taxonomic andscript relations at different times. A further question,which we cannot answer, is whether children cansimultaneously think of the same object as belongingto a taxonomic and script category. Clearly, it wouldbe helpful to have access to more than one categoryrepresentation simultaneously. If a lactose-intolerantchild is trying to decide what to eat for dessert,considering the categories of dessert and dairy foodsimultaneously will get the child much further inmaking an appropriate decision than if only the onecategory was activated. We suspect that the ability tocoordinate two different categories in this waydepends on the entrenchment of the categories. Ifchicken automatically activates the concepts ofdinner food and meat, and if these are familiarconcepts, it may be possible for children to entertainthem simultaneously. However, it is not possible todiscover this in the usual paradigms of categorization.

Experiment 4

Experiment 3 demonstrated that children can cross-classify foods, but the fact that food can belong tomore than one category may pose a problem forinductive inferences. How do children know whichcategory to use when making an inductive inferenceabout particular properties of food? In a study withadults, Ross and Murphy (1999, Experiment 6)found an interaction between category and inferencetype, but this depended to some degree on theprocedure used. This interaction was relatively smallwhen adults were asked to judge the probability thatcertain taxonomic and script categories have variousbiochemical and situational properties. For example,adults were asked whether a category of food (e.g.,fruit, breakfast foods) relative to other foods is low,medium, or high in vitamins. Another example was


asking adults whether a category of food (e.g., fruit,breakfast foods) relative to other foods is eatenduring the morning, afternoon, or evening. Aftermaking a response, adults were also asked to ratethe probability that their answer is correct. Adultsprovided slightly higher induction judgments to thetaxonomic categories (e.g., fruit) for biochemicalproperties (e.g., has vitamins) than to situationalproperties (e.g., is eaten during a certain time ofday). But they gave significantly higher ratings tothe script categories (e.g., breakfast) for the situa-tional properties than for the biochemical properties.

The interaction was larger when Ross andMurphy (1999, Experiment 7) directly pitted taxo-nomic and script categories in conflict triads,requiring adults to compare the two categories andselect the one that provided a better basis for aparticular inference. They found that adults tendedto choose the taxonomically related item for bio-chemical inferences and the script-related item forsituational inferences. For example, when told that atarget food had a fictitious enzyme, adults said that afood that shared a taxonomic category with thetarget was more likely to have the enzyme than afood that shared a script category with the target. Incontrast, when told that a target food was eaten on afictitious holiday, adults said that a food that shareda script category with the target was more likely toalso be eaten on the holiday. We used both types oftest (conflict and nonconflict) to see which, if either,would be better at revealing children’s inductionsand to see if the results would parallel those of Rossand Murphy (1999) with adults.

The goal of Experiment 4 was to investigatechildren’s selective use of taxonomic, script, andevaluative categories to make inductive inferences offoods. Children participated in either the biochem-ical or situational condition. Children in the formercondition were asked to make biochemical infer-ences about the ingredients of foods, whereaschildren in the latter condition were asked to makesituational inferences about when foods should beeaten. Children in both conditions received thenonconflict (taxonomic, script, evaluative) and con-flict triads from Experiment 1. The nonconflict triadsincluded a target food and two choices, an alter-native that shared a categorical relationship with thetarget and an unrelated alternative. The conflicttriads included a target food and alternatives thatshared a taxonomic and script relationship with thetarget.

If children distinguish the types of inductions thatare appropriate for different kinds of categoriesFthat is, if they exhibit inductive selectivityFin the

nonconflict trials they should make more biochem-ical inferences for the taxonomic categories than forthe script categories. In contrast, they should makemore situational inferences for the script categories.The current experiment also examined whetherchildren use evaluative categories for biochemicalor situational inferences, or both. To date, no re-search has documented the type of inferences thiscategory type supports. We speculated that evalua-tive categories in the nonconflict triads could beused to make biochemical inferences, because foodsare typically defined as healthy or junky based ontheir shared biochemical properties such as ingre-dients and nutritional value (e.g., sugar and fatcontent). Past research has shown that childrenexhibit some sensitivity to the kinds of propertiesthat they will infer across categories, but theirsophistication in matching properties to categoriestakes some time to develop (Gelman, 1988; Gelman& Markman, 1986; Gelman & O’Reilly, 1988). How-ever, past work has not explicitly contrasted scriptand taxonomic categories in the induction task.

For the conflict triads, we might also expect thatchildren should choose the taxonomic choice whenmaking biochemical inferences but the script choicewhen making situational inferences. Given Ross andMurphy’s (1999) study with adults, it is possible thatthe results will be stronger in the conflict than thenonconflict triads.

Method

Participants. Participants were twenty-eight 4-year-olds (M5 4,4, range5 4,1 –4,9; 12 boys and 16 girls),twenty-eight 7-year-olds (M5 7,6, range5 6,9 –8,0;10 boys and 18 girls), and 28 adults (M5 21;10 males and 18 females). Half of the participantsin each age group were randomly assigned tothe biochemical condition and the other half tothe situational condition. Thirty adults also rated thematerials. The demographic information for theparticipants is similar to that of the previousexperiments.

Materials. The same food triads from Experiment1 were used in Experiment 4. We wanted to ensurethat these materials could be used for inferences thatwere based on categorical relationships between thetargets and their category alternatives. For example,we wanted to ensure that if milk was the target andthe choices were water and beef, that water wouldbe selected because water and milk share a categor-ical relationship, not because those were the onlytwo items that appeared to have any relation at all.Therefore, we pretested the target and related items


to ensure that they would provide a basis forinduction.

We had adults rate a total of 24 pairs of foods: 12taxonomic targets and their taxonomic choices (e.g.,cereal –bread) and 12 script targets and their scriptchoices (e.g., cake– ice cream). The evaluative targetswere not included because we were attempting toreplicate the results of Ross and Murphy (1999), whoused only script and taxonomic categories (but seethe main results). Half of the adults were told thatthe target food had an ingredient, a biochemicalproperty, and then were asked to rate how certainthey felt that the choice also had this ingredient. Theother half of the adults were told that the target foodwas eaten on a holiday, a situational property, andthen were asked to rate how certain they felt that thechoice was also eaten on this holiday. Adults ratedtheir certainty on a probability scale ranging from 0to 100. The food pairs were presented in two randomorders for each group of adults.

The mean probability ratings were subjected to a 2(property: biochemical and situational)�2 (categorytype: taxonomic and script) ANOVA. The taxonomicand script categories had similar inferential power,as indicated by a lack of a main effect of categorytype. See Table 2 for the probabilities by propertyand category type. There was a main effect ofproperty, F(1, 28)5 20.56, MSE5 195.52, po.05,moderated by a Category Type� Property interac-tion, F(1, 28)5 27.37, MSE5 44.44, po.05. The bio-chemical property led to higher probabilities fortaxonomic categories than for script categories,whereas the situational property led to higherprobabilities for script categories than for taxonomiccategories. These probabilities suggest that thetargets and their script or taxonomic alternatives inExperiment 1 have a categorical relation that sup-ports inductions. Thus, we replicated the results ofExperiment 6 of Ross and Murphy (1999) with thepresent items and validated the use of the triadsfrom Experiment 1 in the current experiment.

Procedure. Children were tested individually in aquiet area of their classroom or in a laboratorysetting in one of two conditions: biochemical orsituational. As in Experiment 1, there were threeblocks of trials in each condition. The first blockincluded two easy warm-up trials. Children in thebiochemical condition heard, for example, ‘‘This is arobin. This is a bear and this is a blue jay. The robinhas a body part called an omentum inside. Do youthink the bear or blue jay probably has an omentuminside too?’’ Children in the situational conditionheard that the robin ‘‘makes a sound called a tekkerin the morning.’’ If a child selected the bear, theresearcher corrected the answer and repeated thewarm-up until it was clear the child understood theprocedures.

After the warm-up trials, children in the biochem-ical condition heard the following: ‘‘Now I’m goingto tell you about a place that some people live.yThe foods that people eat in this faraway place havedifferent ingredientsy. This game is about ingre-dients of foods. Some foods have a certain ingre-dient, but not all foods do. Sometimes foods evenhave more than one ingredient, too. So can you helpme figure out which foods have a certain ingredientin them and which foods do not?’’ Children in thesituational condition heard the same descriptionexcept that they were told that there was a foreigncountry where people eat foods on different holi-days.

These descriptions were adapted from Experiment7 of Ross and Murphy (1999). Novel words wereused for the ingredients and holidays because wewere interested in the inferences children make fromthe new information that was provided in theexperiment. We did not want children to base theiranswers on associations that they might havebetween actual ingredients, holidays, and foods.

The second block included taxonomic, script, andevaluative trials, whereas the third block includedconflict trials. For the second block of trials, childrenin the biochemical condition were told that a targetfood had a novel ingredient and were asked to selectanother food that also had that ingredient (e.g.,‘‘Pary is an ingredient in grapes. Do you think astrawberry or bread probably also has Pary in ittoo?’’). Children in the situational condition, incontrast, were told that a food was eaten on a novelholiday and were asked to select another food thatwas also eaten on that holiday (e.g., ‘‘Cake is eatenon a special holiday called Dax. Do you thinkice cream or soup is probably eaten on Dax too?’’). Ifa child did not make a clear choice, the researcherrepeated the question. The trials within each of

Table 2

Adult Probability Ratings of Induction by Property and Category type in

Experiment 4

Category type

Taxonomic Script Mean

Property

Biochemical 51 39 45

Situational 58 65 61

Mean 54 52


the blocks were presented in one of two randomorders.

Adults were not shown the photographs but wereasked to read and respond to the questions withfood names presented in a paper-and-pencil format.

Results and Discussion

To test children’s inductive selectivity, we firstconducted a 2 (property: biochemical and situa-tional) �3 (age; 4-year-olds, 7-year-olds, adults) �3(category type: taxonomic, script, and evaluative)ANOVA on the data from the nonconflict triads.Recall that the nonconflict triads included a targetand a categorically related choice and an unrelatedchoice. If children have inductive selectivity, theyshould select the category choice more when it wasappropriate for the induction. For example, forbiochemical inferences children should select thecategory choice on the taxonomic and (perhaps)evaluative triads but should be less likely to do so onthe script triads. Similarly, on the script triadsparticipants should select the category choice forsituational inferences but should do so less often onthe taxonomic and (perhaps) evaluative triads.

There was a reliable age effect, F(2, 78)5 24.73,MSE5 1.13, po.05: Overall, 7-year-olds (M5 0.77,SD5 0.18) and adults (M5 0.82, SD5 0.17) selectedthe food that shared a categorical relationship withthe target significantly more often than did 4-year-olds (M5 0.60, SD5 0.19), Tukey’s HSD test, po.05.Participants selected the food that shared a categor-ical relationship with the target more often whenmaking biochemical inferences (M5 0.76, SD5 0.19)than situational inferences (M5 0.70, SD5 0.21), F(1,78)5 4.56, MSE5 0.20, po.05. There was no maineffect of category type nor an interaction of CategoryType�Age.

There was a Category Type�Property interaction,F(1, 132)5 4.79, MSE5 0.15, po.05, which wassubsumed by the three-way interaction amongcategory type, property, and age, F(4, 132)5 4.0,MSE5 0.1, po.05. Figure 2 illustrates that 4- and 7-year-olds lacked inductive selectivity, in that chil-dren tended to pick the category choice regardless ofwhether the biochemical or situational inference wasappropriate. Unlike the children, adults made sig-nificantly more biochemical inferences for thetaxonomic than for the script categories, t(13)5 4.7,po.05, and more biochemical inferences for theevaluative than the script categories, t(13)5 –3.88,po.05. Adults made significantly more situationalinferences for the script than for the taxonomiccategories, t(13)5 –4.3, po.05. There was also a

trend toward a difference between the script andevaluative categories, t(13)5 1.9, p5 .08. There wasno significant difference between the taxonomic andevaluative categories.

We examined the conflict triads by comparingparticipants’ taxonomic choices to chance, 50%.Recall that the conflict triads included a taxonomicchoice pitted against a script choice. We expectedthat if children have inductive selectivity, theyshould pick the taxonomic choice when makingbiochemical inferences and the script choice whenmaking situational inferences. Figure 3 shows thatthe results from the conflict trials conform to theresults of the nonconflict trials: Adults had apreference for taxonomic categories when they madebiochemical inferences, t(13)5 4.94, po.05, but thispreference was reversed when they made situationalinferences, t(13)5 –3.60, po.05. Four- and 7-year-olds showed no such preferences.

Overall, the results from the adults are consistentwith Ross and Murphy (1999), revealing that adultshave inductive selectivity. Adults made more bio-chemical inferences for the taxonomic and evalua-tive categories than for script categories. In contrast,adults made more situational inferences for thescript than the taxonomic categories and (margin-

Fig. 2. Percentage of taxonomic, script, and evaluative categorychoices selected for biochemical and situational inferences inExperiment 4.


ally) than the evaluative categories. However, ourresults did not reveal that children have inductiveselectivity. Children made a similar number ofbiochemical and situational inferences for the taxo-nomic, script, and evaluative categories. Thus, theresults show that children can use all three kinds ofcategories from which to draw inferences, which isan important result, as induction is a major functionof categorization (Gelman & Markman, 1986, 1987;Murphy, 2002). However, the results also suggestthat children are not very good at distinguishingwhat kinds of properties are most appropriate forthese different category types.

Experiment 5

One explanation of the results of Experiment 4 isthat children do not have inductive selectivity. Thegoal of Experiment 5 was to examine this possibilitymore fully. Experiment 4 included mainly noncon-flict triads (28), which included an alternative thatshared a categorical relationship with the target andan unrelated alternative. The nonconflict triad taskrequired children always to choose a food even ifthey believed an inference did not make sense forthat triad. For example, situational inferences makesense for script triads, but they make less sense fortaxonomic triads. However, children were not giventhe option of not selecting a food in these cases.Given this circumstance, children might haveattempted to select the best choice possible, whichmay have been to select the food that shared thesame script, taxonomic, or evaluative category as thetarget regardless of the property type. This strategywould tend to reduce any selectivity based onproperty type. A similar explanation was suggestedin a recent study by Diesendrcuk and Bloom (2003,Experiment 2). In this study, although children weremore likely to generalize relevant properties (e.g., ‘‘Itwas made especially to play with cats’’) as opposed

to irrelevant properties (e.g., ‘‘My uncle gave this tome’’) to items that shared the same shape as thetargets, children still made many irrelevant propertygeneralizations. Diesendruck and Bloom (2003,Experiment 2) suggested that when children wereasked to generalize irrelevant properties, they mayhave tried to pick the best answer possible under theforced-choice circumstances, which was the shapechoice.

In contrast, conflict triads directly pit a taxonomicagainst a script choice, and one of them is alwaysappropriate. Although Experiment 4 did containconflict triads, they were small in number (only 6)relative to the nonconflict triads. We increased thenumber of conflict trials to 15 in Experiment 5 to givethe children a greater opportunity to reveal anyknowledge of which properties should be inducedfrom different categories. We did not includeevaluative categories in Experiment 5 to increasethe power of our script – taxonomic comparison. Ifchildren have inductive selectivity, they shouldselect the taxonomic choice in the biochemicalcondition and the script choice in the situationalcondition. Finally, we slightly modified the wordingin the procedure to make it more child friendly.

Method

Participants. The participants were thirty-two 4-year-olds (M5 4,5, range5 4,0 –5,1; 16 boys and 16girls), thirty-two 7-year-olds (M5 7,6, range5 6,7 –8,3; 14 boys and 18 girls), and 32 adults (M5 19; 16males and 16 females). The participants were drawnfrom a similar population as in the previousexperiments.

Materials. The stimuli were selected and preparedin the same way as in the previous experiments. Thephotographs were arranged into 15 conflict triads(shown in Appendix C) in which there was a targetand a taxonomic and a script choice pitted againsteach other. Adults again completed visual similarityratings of the stimuli. There was not a significantdifference between the targets and their taxonomicchoices (M5 3.41, SD5 1.00) and the targets andtheir script choices (M5 3.68, SD5 0.74).

Procedure. A researcher tested children in a quietarea of the classroom. Children participated in eitherthe biochemical or situational condition. The direc-tions children heard in Experiment 5 were asimplified version of those from Experiment 4.Instead of using the word ingredient, we used thewords stuff inside. Also, instead of using the wordholiday, we used the phrase special time. In particular,children in the biochemical condition heard the

Fig. 3. Percentage of taxonomic choices selected for biochemicaland situational inferences in the conflict trials of Experiment 4.


following directions: ‘‘I’m going to tell you about aplace where some people live that’s very far awayfrom here.y This game is about the stuff insidefoods that these people eat. Some foods have thesame stuff inside and some foods have different stuffinside. So can you help me figure out which foodshave the same stuff inside?’’ Children in thesituational condition heard the following directions:‘‘This game is about the foods that these people eatduring special times. Some foods are eaten duringthe same special times and others are eaten duringdifferent special times.’’

Children in both conditions completed an initialwarm-up trial, followed by the 15 conflict triads. Thetriads were presented in one of two random orders.

Results

Experiment 5 investigated whether children se-lectively generalize properties depending on thetype of category. To compare responses as a functionof property, we assigned a 1 to the taxonomic choicesand a 0 to script choices for both the biochemical andsituational conditions. We then converted thesescores into proportions and conducted a 3 (age; 4-year-olds, 7-year-olds, adults) � 2 (property; bio-chemical, situational) ANOVA on these data. Therewere many more taxonomic choices for the bio-chemical (M5 0.69, SD5 0.23) than for the situa-tional (M5 0.20, SD5 0.61) properties, F(1,90)5 275, MSE5 .02, po.05. There were signifi-cantly more taxonomic choices for biochemical thanfor situational properties for each age group exam-ined separately, ts(30)43, pso.05. Overall, adultsand 7-year-olds made more taxonomic choices thandid 4-year-olds, F(2, 89)5 9.0, MSE5 43.4, po.05.Four-year-olds differed significantly from 7-year-olds and adults, but 7-year-olds and adults did notdiffer from each other, Tukey’s HSD, po.05.

Figure 4 shows that these main effects weremoderated by a Property�Age Group interaction,F(2, 90)5 38.6, MSE5 .02, po.05. The adults and7-year-olds gave very different responses based onthe property type, whereas 4-year-olds had a fairlysmall difference. Instead, they tended to choose thescript-related item regardless of property.

We also compared participants’ taxonomic cho-ices to chance (50%). Seven-year-olds and adults, butnot 4-year-olds, were significantly above chance ontheir taxonomic choices for the biochemical proper-ties, ts(15)45, pso.05. The opposite pattern was truefor the situational properties. Four-year-olds, 7-year-olds, and adults chose the taxonomic choices less

often than chance for the situational properties,ts(15)o� 9, pso.05.

Discussion

The purpose of Experiment 5 was to reexaminewhether children can selectively use their categoriesfor induction by testing conflict triads pittingtaxonomic against script categories. The results fromExperiment 5 suggest that 7-year-olds understandthat biochemical inferences are most appropriate fortaxonomic categories, whereas situational inferencesare most appropriate for script categories. Although4-year-olds were not above chance on their taxo-nomic choices for the biochemical properties, therewas still a significant difference in their performanceacross the two property types: They made signifi-cantly more taxonomic choices for the biochemicalthan for the situational properties, t(30)5 3.36,po.05. This suggests that 4-year-olds are beginningto develop their ability for inductive selectivity.These overall results fall in line with researchdemonstrating that children know that certaincategories support certain inductive inferences(Deak, 2000; Kalish & Gelman, 1992).

The fact that 4-year-olds were more accurate onthe situational properties also suggests that childrenmay be able to make inductive inferences from scriptcategories earlier than taxonomic categories. How-ever, this result could also simply reflect a preferencefor the script relation, which would be consistentwith their performance on conflict triads in Experi-ment 1 and previous research. Future research couldexamine how to improve 4-year-olds’ ability to makeappropriate inductive inferences for taxonomiccategories (e.g., priming children with categorylabels for the target foods).

The results may also be related to the findings ofRoss and Murphy (1999), who studied induction

Fig. 4. Percentage of taxonomic choices selected for biochemicaland situational inferences in Experiment 5.


with similar categories in adults. As described earlier,they found some evidence for selective inductionwhen their participants made probability ratings fromindividual categories, but the effects were noticeablystronger in conflict trials. An analogous result can befound in our Experiments 4 and 5. In Experiment 4,we found evidence for selective induction withnonconflict triads in adults only. In Experiment 5,with conflict triads, the effect with adults was muchstronger (see Figure 4), and we now found clearselectivity in 7-year-olds. Thus, the conflict triadswere more powerful in revealing this effect.

That said, it should also be noted that neither kindof task is perfect. In the nonconflict triad, participantsmay be tempted to choose the item that is relatedsomehow to the target even if it is not an appropriatecategory for making the given induction. If one mustchoose between two items that do not seem likely toshare the induced property, participants may simplydecide to choose any item that has a noticeablerelation. In contrast, the conflict task emphasizes thedifference between two kinds of items by presentingthem simultaneously and forcing participants tochoose the better one. This task seems to do betterat revealing inductive selectivity, but whether thesame results would be found when a single item istested (e.g., ‘‘Do I think that this particular muffinhas carbohydrates?’’) has yet to be seen.

As a final note, the results of this experiment,along with the pretest with adults reported in theMethod section of Experiment 4, verify that we weresuccessful in identifying items that were related bytaxonomic or script category in our experiments.That is, the fact that adult participants (and 7-year-olds in Experiment 5) gave stronger biochemicalinductions to our taxonomic items and strongersituational inductions to our script items givesindependent evidence that we successfully identi-fied the categories of the food types. If there was anyconcern that our taxonomic items might also havehad strong script relations, or vice versa, the presentresults contradict this possibility.

General Discussion

The Piagetian view that children do not havetaxonomic concepts has been largely discarded inrecent years. Much research on this topic has beendevoted to finding ways of demonstrating children’searly appreciation of taxonomic concepts (Markman,1989; Waxman & Gelman, 1988). Others have arguedthat Piaget’s results really reflect children’s use ofscript-based concepts, which often include membersof the same taxonomic category. These script-based

categories may eventually develop into completetaxonomic concepts (e.g., Nelson, 1988). Such dis-putes tend to create an exclusive approach tocategorization in which researchers attempt to dis-cover which form of organization children use. Ourapproach has been to ask whether children usemultiple kinds of concepts by measuring themindependently rather than assuming that one formpredominates.

Taxonomic, Script, and Evaluative Categories

Three questions guided our investigation. First,do children have taxonomic, script, and evaluativecategories, and does one of them predate the others?The answer to the first part of this question is yes.Experiment 1 found that 4-year-olds, 7-year-olds,and adults have taxonomic, script, and evaluativecategories of food. Experiment 2 revealed that 3-year-olds have both taxonomic and script categories.The answer to the second part of this question is thatthe categories appear to develop simultaneously.Unlike much research that forced children to chooseeither a taxonomic or script or thematic response,our results did not show any advantage for the scriptcategory. The 3-year-olds were equally accurate withboth types of category.

The results from Experiments 1 and 2 differ fromresearch concluding that young children use scriptcategories but lack taxonomic categories until afterapproximately age 7 (e.g., Lucariello et al., 1992;Lucariello & Nelson, 1985; Nelson & Nelson, 1990;Sell, 1992; Yu & Nelson, 1993). One reason for thisdifference is that past studies have used conflicttriads that pit taxonomic and script categoriesagainst each other, whereas the present studies usednonconflict picture triads in which children wereallowed to categorize items into both categories. Inconflict triads, preference and categorization abilityare conflatedFthere is no way for participants toreveal knowledge of both kinds of category. Thus,this technique cannot reveal cross-classification, asour results did.

Another possible explanation is that our experi-ments used tasks more directly related to categor-ization, as opposed to memory or word-associationtasks. It is possible that script (and thematic)categories have an advantage in such noncategoriza-tion tasks because of factors such as associativestructure affecting memory recall (e.g., Blewitt, 1993;Blewitt & Krackow, 1992; Blewitt & Toppino, 1991).Krackow and Gordon (1998), for example, concludedthat script categories had an advantage over taxo-nomic categories in memory and word-association


tasks because past studies used script categories thatare high on typicality and associativity (e.g., shirt –pants), whereas the taxonomic items were selected soas not to be associated. This advantage disappearedwhen Krackow and Gordon (1998) used scriptcategories that were low on typicality and associa-tivity (e.g., vest –pants) in a cued-recall task. Thus,the apparently earlier appearance of script categoriesin children’s concepts may be due to these aspects ofthe most popular tasks used in prior research.

Note that we are not claiming that script categoriesare not an early and salient form of organization ofchildren’s concepts. In fact, we find evidence for themin children as young as 3. Instead, we are arguing thatboth script and taxonomic forms of organization areprevalent in children’s concepts, and past findings ofa progression from one to the other may reflect achange in preference. This result is consistent withrecent claims that children’s preference for thematicover taxonomic category choices in the triad taskmay reflect a preference rather than a conceptualshift (Lin & Murphy, 2001; Smiley & Brown, 1979;Waxman & Namy, 1997).

Although we did not find the traditionallyclaimed developmental shift, this does not implythat there is no improvement with age. Certainly, theresults from Experiments 1 and 2 show that from 3 to7 years there is a steady improvement in children’sability to categorize foods into taxonomic and scriptcategories. However, this improvement occurs aboutequally across category types. The only qualitativedifference was on conflict trials, when 7-years-oldscategorized foods more often into taxonomic cate-gories than into script categories in Experiment 1(i.e., the usual preference for taxonomic choices inschool-aged children; Smiley & Brown, 1979). Thereason for this change in preference is beyond thescope of the present article, but it may have to dowith formal education (see Markman, 1989; Murphy,2002, chap. 10; Smiley & Brown, 1979, for discussion).

We found that children and adults were bothsensitive to evaluative categories of healthy andunhealthy foods, although there was perhaps agreater developmental difference here than for theother categories. An important question aboutevaluative categories is whether they are constructedsimply on the basis of affective value (healthy isgood; unhealthy is bad) or have additional content.We address this in light of the induction results.

Cross-Classification

Our second question was, do children cross-classify foods into these categories? The answer is

yes. Experiments 1 and 2 examined whether childrenhave multiple systems of categories, finding thatscript and taxonomic categories exist simultaneouslyin the food domain at all age groups tested.Evaluative categories also were found in 4-year-oldsto adults. Experiment 3 showed that 4- and 7-year-olds can categorize a single food into both ataxonomic and script category, such as consideringmilk as a dairy product and a snack. Again, theseresults suggest that children do not rely on a singleform of categorization in which one type of categoryeventually replaces the other. The results from thepresent study are consistent with the findings ofBlaye and Bonthoux (2001), who found that childrencan change their initial form of categorization fromtaxonomic to thematic or vice versa when providedwith a contextual cue. Our results are the first todocument this phenomenon with taxonomic andscript categories, particularly when no prime or hintis provided for children.

Cross-classification is an important conceptualability. As children gain greater knowledge of theworld, they become aware that the same entity canbe perceived in different ways, ranging from specificto general, and differing in the particular perspectivebrought to bear on it. For example, thinking ofsomeone as amother, a woman, an adult, an accountant,a tennis player, an extravert, and so on, may beimportant to understanding the complex behavior ofa single person; thinking of a truck as a moving van,a mode of transportation, a manufactured object, alarge mass, an economic asset, or a source ofpollution may be necessary to dealing with it ondifferent occasions. Thus, it is essential to our fullunderstanding of the world that we be able to cross-classify items and use different categories to deriverelevant information.

Future research should examine the mechanismsthat underlie children’s ability to cross-classify itemsflexibly. It is possible that polynomy, the ability toproduce several words for a single object, is relatedto children’s cross-classification. Research has found,for example, that very young children have multiplewords for a single object (Blewitt, 1994; Deak &Maratsos, 1998; Waxman & Hatch, 1992) and thatdepending on the goals of the conversation, childrenare able to choose labels reflecting different con-ceptual perspectives on a single item (Clark, 1997).

Inductive Selectivity

Our third question was, do children selectivelyuse taxonomic, script, and evaluative categories tomake inductive inferences about food? The answer


to this question is complex. Although we did notfind evidence for children’s inductive selectivity inExperiment 4, we did find evidence in Experiment 5,which included only conflict triads that pittedtaxonomic against script categories. Experiment 5showed that 7-year-olds make more biochemicalinferences for taxonomic categories and more situa-tional inferences for script categories. These resultsare consistent with the small body of researchsuggesting that children have inductive selectivity(e.g., Deak, 2000; Kalish & Gelman, 1992).

Across Experiments 4 and 5, however, we foundlittle evidence for inductive selectivity in 4-year-olds. Because children of this age can classify itemsinto both kinds of categories, that does not seem tobe their problem in induction. Instead, we suspectthat the requirements of matching the property tothe category are what make selectivity difficult. Thatis, the child must sometimes make an induction fromcake to ice cream, say, but sometimes must not makean induction, depending on what property is beingprobed. Recognizing that different properties arerelevant for different categorical inductions andinhibiting the positive response for inappropriateproperties (e.g., saying that two snacks do not sharea biochemical property) may have been too difficultfor the 4-year-olds. The exact reason for 4-year-olds’failing this task needs to be examined in future work.

Murphy and Ross (1999) examined inductions foritems that were in two categories. They found thatadults often paid attention to both categories whenmaking their inductions. For example, participantsread about the level of amino acids in fruits andsnacks (among other properties and other cate-gories). When they were asked to decide the levelof amino acids in apples (which are both a fruit anda snack), they showed evidence of using the levels inboth categories to produce an answer. Their proce-dures differed in several respects from the presentinduction tasks, which were more like the traditionaltask of Gelman and Markman (1986) and thesubsequent cognitive development literature. None-theless, their results point out that multiple cate-gories of an item can in some circumstances besimultaneously taken into account and coordinated.Eventually, children must learn not to decidewhether an apple is a fruit or a snack but to takeboth categories into account in making an accurateinduction.

An important contribution of Experiments 4 and 5is the finding that children can use script categoriesas a basis for induction. So far, the majority ofresearch examining children’s inductive inferenceshas focused on taxonomic categories, and the role of

script categories in induction has not been welldocumented (see Krackow & Gordon, 1998). Ourresults suggest that script categories are useful forinduction, and clearly by age 7, children use scriptcategories preferentially for situational inferences.

Experiment 4 is the first study to investigate thetype of inference evaluative categories support. Wespeculated that evaluative categories would be usedto make biochemical inferences because healthy orjunky foods typically share properties such asingredients and nutrients. Indeed, in Experiment 4,we found that adults strongly made biochemicalinferences for evaluative categories, but they alsomade situational inferences, although less strongly(see Figure 2). These results reveal that evaluativecategories have considerable content beyond simplepositive affect. Less-than-healthy foods such as icecream and cake both share nutritional ingredientsand are often served as desserts. For adults, then,evaluative categories provide a basis for inferences.There is increasing evidence that everyday cate-gories have components of goals and ideals (Barsa-lou, 1985, 1991; Lynch, Coley, & Medin, 2000; Medinet al., 1997), and evaluative food categories seemsimilar to some of these examples, such as Medin etal.’s (1997) weed trees. Unlike purely affectivecategories, these evaluative categories carry consid-erable content and thus can provide the basis forsignificant inductions. Future research needs toexplore when children begin to use evaluativecategories for these types of inferences.

Although Experiment 4 demonstrated that eval-uative categories can be used to make biochemicaland situational inferences (which are associated withtaxonomic and script categories, respectively), thequestion remains whether evaluative categories sup-port a particular type of inductive inferences thatother categories do not. There is reason to believe, forexample, that evaluative categories may be used tomake inductive inferences about the bodily conse-quences of eating these foods, especially becausefood is integrated into children’s knowledge of healthand illness (e.g., Fallon, Rozin, & Pliner, 1984; Rozin,1990; Springer, Ngyuen, & Samaniego, 1996).

Food Categories

We used food as a test case in our experiments. Itis possible that children are particularly precociousin dealing with food categories, given their interestand importance in the lives of children. However,that would not invalidate our results, given that theessential question is whether children have thecompetence to form taxonomic, script, and evalua-


tive categories and to use them appropriately. Therewill always be content differences in the acquisitionof categories because of a lack of knowledge andexperience (Mervis, 1987). Thus, it would not besurprising if 4-year-olds did not know taxonomiccategories of financial instruments or types of searchalgorithms. The finding that 3-year-olds have bothtaxonomic and script categories in the domain offood demonstrates their competence in both forms ofconceptual organization, even at this early age.

It is possible that other domains would not showthe same results as we have obtained. If children donot know of script relations involving animals, forexample, they would of course have to use onlytaxonomic or evaluative categories. Some mighttherefore argue that we should have used a widevariety of stimuli from different domains so that ourresults would be more generalizable. However, ourchoice of foods was directed by the knowledge thatthere are both script and taxonomic organizationspossible within this domain and that adults know ofand use both of them (Ross & Murphy, 1999). Thathas not been extensively documented with animals,clothing, furniture, and so on. Therefore, if we hadnot found cross-classification in a diverse set ofstimuli, it would not be clear whether this was dueto children’s lack of ability or to the lack of one or theother structure in some of these domains. By using adomain in which both kinds of classification canbeFand areFused, our results could be morerevealing of children’s abilities. If future studiesreveal that children do not cross-classify furniture oranimals, say, this would be informative aboutchildren’s conceptual knowledge of those domains.However, to discover such effects, individual do-mains must be studied in some detail rather thanmixing together items from different domains.

We certainly believe that further investigationusing different kinds of categories is called for.However, we would argue that the striking patternof multiple kinds of conceptual organization usedfrom age 3 years through adulthood was onlyobservable by our use of a domain in which differentcategory organizations have been well documented.

Conclusion

In conclusion, the present studies suggest thatchildren can cross-classify items into multiplecategories and use these categories for inductiveinferences. By 3 years, children have multiple cate-gories of food, including taxonomic and scriptcategories. By 4 years, children also have evaluativecategories and can cross-classify foods into more

than one category. Furthermore, 7-year-olds and tosome extent 4-year-olds realize that certain cate-gories of food provide better support for certaininductive inferences. That is, children have inductiveselectivity. These results suggest that young childrenare not restricted to a single form of categorization,as suggested by traditional accounts of children’sconceptual development. Even young children cancategorize aspects of their world flexibly, laying thegroundwork for their ability to cross-classify asadults.

References

Barsalou, L. W. (1983). Ad hoc categories. Memory &Cognition, 11, 211 – 227.

Barsalou, L. W. (1985). Ideals, central tendency, andfrequency of instantiation as determinants of gradedstructure in categories. Journal of Experimental Psychol-ogy: Learning, Memory, and Cognition, 11, 629 – 654.

Barsalou, L. W. (1991). Deriving categories to achievegoals. In G. H. Bower (Ed.), The psychology of learning andmotivation: Advances in research and theory (pp. 1 – 64). SanDiego, CA: Academic Press.

Birch, L., Fisher, J., & Grimm-Thomas, K. (1999). Childrenand food. In M. Siegal & C. C. Petereson (Eds.),Children’s understanding of biology and health (pp. 161 –182). Cambridge, England: Cambridge University Press.

Blaye, A., & Bonthoux, F. (2001). Thematic and taxonomicrelations in preschoolers: The development of flexibilityin categorization choices. British Journal of DevelopmentalPsychology, 19, 395 – 412.

Blewitt, P. (1993). Taxonomic structure in lexical memory.The nature of developmental change. Annals of ChildDevelopment, 9, 103 – 132.

Blewitt, P. (1994). Understanding categorical hierarchies:The earliest levels of skills. Child Development, 6,1279 – 1298.

Blewitt, P., & Krackow, E. (1992). Acquiring taxonomicrelations in lexical memory: The role of superordinatecategory label. Journal of Experimental Child Psychology,54, 37 – 56.

Blewitt, P., & Toppino, T. C. (1991). The development oftaxonomic structure in lexical memory. Journal ofExperimental Child Psychology, 51, 296 – 319.

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

Clark, E. V. (1997). Conceptual perspective and lexicalchoice in acquisition. Cognition, 64, 1 – 37.

Coley, J. D. (2000). On the importance of comparativeresearch: The case of folk biology. Child Development, 71,82 – 90.

Deak, G. O. (2000). The growth of flexible problem solving:Preschool children use changing verbal cues to infermultiple word meanings. Journal of Cognition andDevelopment, 2, 157 – 191.


Deak, G. O., & Maratsos, M. (1998). On having complexrepresentations of things: Preschoolers use multiplewords for objects and people. Development Psychology,34, 224 – 240.

Denney, D. R. (1975). Developmental changes in conceptutilization among normal and retarded children. Devel-opmental Psychology, 12, 359 – 368.

Denney, D. R., & Moulton, P. A. (1976). Conceptualpreferences among preschool children. DevelopmentalPsychology, 12, 509 – 513.

Diesendrcuk, G., & Bloom, P. (2003). How specific is theshape bias? Child Development, 74, 168 – 178.

Duckworth, K., Bargh, J. A., Garcia, M., & Chaiken, S.(2002). The automatic evaluation of novel stimuli.Psychological Science, 13, 513 – 519.

Fallon, A. E., Rozin, P., & Pliner, P. (1984). The child’sconception of food: The development of food rejectionswith special reference to disgust and contaminationsensitivity. Child Development, 55, 566 – 575.

Fenson, L., Dale, P. S., Reznick, S. J., Bates, E., Thal, D. J., &Pethick, S. J. (1994). Variability in early communicativedevelopment. Monographs of the Society for Research inChild Development, 59 (Serial No. 242).

Gelman, S. A. (1988). The development of induction withinnatural kind and artifact categories. Cognitive Psychology,20, 65 – 95.

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

Gelman, S. A., & Markman, E. M. (1987). Young children’sinductions from natural kinds: The role of categoriesand appearances. Child Development, 58, 1532 – 1541.

Gelman, S. A., & O’Reilly, A. W. (1988). Children’sinductive inferences within superordinate categories:The role of language and category structure. ChildDevelopment, 59, 876 – 887.

Greenfield, D. B., & Scott, M. S. (1986). Young children’spreference for complementary pairs: Evidence against ashift to a taxonomic preference. Developmental Psychol-ogy, 22, 19 – 21.

Heit, E., & Rubinstein, J. (1994). Similarity and pro-perty effects in inductive reasoning. Journal of Experi-mental Psychology: Learning, Memory, & Cognition, 20,411 – 422.

Horton, M. S., & Markman, E. M. (1980). Developmentaldifferences in the acquisition of basic and superordinatecategories. Child Development, 51, 708 – 719.

Hull, C. L. (1920). Quantitative aspects of the evolution ofconcepts. Psychological Monographs, 28, 1 – 86.

Inhelder, B., & Piaget, J. (1964). The early growth of logic inthe child. New York: Norton.

Kagan, J., Moss, H. A., & Siegel, I. E. (1963). Psychologicalsignificant of styles of conceptualization. Monographs ofthe Society for Research in Child Development, (2, Serial No.28), 73 – 112.

Kalish, C. W., & Gelman, S. A. (1992). On wooden pillows:Multiple classifications and children’s category-basedinduction. Child Development, 63, 1536 – 1557.

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

Krackow, E., & Gordon, P. (1998). Are lions and tigerssubstitutes or associates? Evidence against slot filleraccounts of children’s early categorization. Child Devel-opment, 69, 347 – 354.

Lin, E. L., & Murphy, G. L. (2001). Thematic relations inadults’ concepts. Journal of Experimental Psychology:General, 130, 3 – 28.

Lucariello, J., & Nelson, K. (1985). Slot-filler categories asmemory organizers for young children. DevelopmentalPsychology, 21, 272 – 282.

Lucariello, J., Kyratzis, A., & Nelson, K. (1992). Taxonomicknowledge: What kind and when? Child Development,63, 978 – 998.

Lynch, E. B., Coley, J. D., & Medin, D. L. (2000). Tall istypical: Central tendency, ideal dimensions and gradedcategory structure among tree experts and novices.Memory & Cognition, 28, 41 – 50.

Mandler, J. M., Bauer, P. J., & McDonough, L. (1991).Separating the sheep from the goats: Differentiatingglobal categories. Cognitive Psychology, 23, 263 – 298.

Markman, E. M. (1989). Categorization and naming inchildren: Problems of induction. Cambridge, MA: MIT Press.

Medin, D. L., & Atran, S.(in press). The native mind:Biological reasoning in development and across cul-tures. Psychological Review.

Medin, D. L., Lynch, E. B., Coley, J. D., & Atran, S. (1997).Categorization and reasoning among tree experts: Do allroads lead to Rome? Cognitive Psychology, 32, 49 – 96.

Mervis, C. B. (1987). Child-basic object categories and earlylexical development. In U. Neisser (Ed.), Concepts andconceptual development: Ecological and intellectual factors incategorization. Emory symposia in cognition (pp. 201 – 233).NY: Cambridge University Press.

Murphy, G. L. (1993). Theories and concept formation. InI. Van Mechelen & J. Hampton (Eds.), Categories andconcepts: Theoretical views and inductive data analysis (pp.173 – 200). San Diego, CA: Academic Press.

Murphy, G. L. (2001). Causes of taxonomic sorting byadults: A test of the thematic-to-taxonomic shift.Psychonomic Bulletin & Review, 8, 834 – 839.

Murphy, G. L. (2002). The big book of concepts. Cambridge,MA: MIT Press.

Murphy, G. L., & Ross, B. H. (1999). Induction with cross-classified categories.Memory & Cognition, 27, 1024 – 1041.

Nelson, K. (1986). Event knowledge: Structure and function indevelopment. Hillsdale, NJ: Erlbaum.

Nelson, K. (1988). Where do taxonomic categories comefrom? Human Development, 31, 3 – 10.

Nelson, K., & Nelson, A. P. (1990). Category production inresponse to script and category cues by kindergartenand second-grade children. Journal of Applied Develop-mental Psychology, 11, 431 – 446.

Niedenthal, P. M., Halberstadt, J. B., & Innes-Ker, A. H.(1999). Emotional response categorization. PsychologicalReview, 106, 337 – 361.


Olver, R. R., & Hornsby, J. R. (1967). On equivalence. In J. S.Bruner, R. R. Olver, & P. M. Greenfield (Eds.), Studies incognitive growth (pp. 68 – 85). New York: Wiley.

Rosch, E., Mervis, C. B., Gray, W., Johnson, D., & Boyes-Braem, P. (1976). Basic objects in natural categories.Cognitive Psychology, 3, 382 – 439.

Ross, B. H., &Murphy, G. L. (1999). Food for thought: Cross-classification and category organization in a complexreal-world domain. Cognitive Psychology, 38, 495 – 553.

Rozin, P. (1990). Development in the food domain.Developmental Psychology, 26, 555 – 562.

Rozin, P., Hammer, L., Oster, H., & Horowitz, T. (1986). Thechild’s conception of food: Differentiation of categoriesof rejected substances in the 16 months to 5 year agerange. Appetite, 7, 141 – 151.

Sell, M. A. (1992). The development of children’s knowl-edge structures: Events, slots, and taxonomies. Journal ofChild Language, 19, 659 – 676.

Smiley, S. S., & Brown, A. L. (1979). Conceptual preferencefor thematic or taxonomic relations: A nonmonotonictrend from preschool to old age. Journal of ExperimentalChild Psychology, 28, 437 – 458.

Springer, K., Ngyuen, T., & Samaniego, R. (1996). Earlyunderstanding of age- and environment-related nox-iousness in biological kinds: Evidence for a naive theory.Cognitive Development, 11, 65 – 82.

Vygotsky, L. S. (1962). Thought and language. Cambridge,MA: MIT Press.

Waxman, S. R., & Gelman, R. (1988). Preschoolers’ use ofsuperordinate relations in classification and language.Cognitive Development, 1, 139 – 156.

Waxman, S. R., & Hatch, T. (1992). Beyond the basics:Preschool children label objects flexibly at multiplehierarchical levels. Journal of Child Language, 19,153 – 166.

Waxman, S. R., & Namy, L. R. (1997). Challenging thenotion of a thematic preference in young children.Developmental Psychology, 33, 555 – 567.

Yu, Y., & Nelson, K. (1993). Slot-filler and conventionalcategory organization in young Korean children. Inter-national Journal of Behavioral Development, 16, 1 – 14.

Zajonc, R. B. (1980). Feeling and thinking: Preferences needno inferences. American Psychologist, 35, 151 – 175.

Appendix A

Food Triads for Experiments 1, 2, and 4

Target Choice 1 (category) Choice 2 (unrelated)Taxonomic

Beverages Soda pop Juice PotatoBeverages Water Milk StrawberryBreads and grains Cereal Bread BeefBreads and grains Pasta Muffin YogurtDairy foods Butter Yogurt CarrotDairy foods Pudding Cheese OrangeFruits Grapes Strawberry BreadFruits Watermelon Orange FishMeats Ham Beef MuffinMeats Turkey Fish MilkVegetables Corn Carrot JuiceVegetables Green beans Potato Cheese

Script

Breakfast foods Waffle Egg CupcakeBreakfast foods Bacon Pancake Jell-OLunch foods Hot dog Chicken nuggets PancakeLunch foods Sandwich Soup Ice creamDinner foods Chicken Pizza CrackerDinner foods Spaghetti Hamburger PeanutsBirthday foods Cake Ice cream SoupBirthday foods Candy Cupcake Chicken nuggetsDesserts Cookie Jell-O EggDesserts Brownie Pie Pizza


Snack foods Pretzel Peanuts PieSnack foods Raisins Cracker Hamburger

EvaluativeHealthy foods Banana Spinach CheetosHealthy foods Celery Apple ChipsJunk foods Twinkie Cheetos AppleJunk foods Chocolate Chips Spinach

Target Taxonomic ScriptIce cream Cheese (dairy) Cookie (dessert)Grapes Apple (fruit) Cracker (snack)Milk Soda pop (beverages) Muffin (breakfast)Potato Celery (vegetable) Turkey (dinner)Chicken nuggets Ham (meats) Sandwich (lunch)Cake Bread (Breads and grains) Candy (birthday)

Appendix B

Food Triads for Experiment 3Target Category choice Unrelated choice

Milk Cheese (dairy) NoodlesMilk Cookies (snack) BolognaIce cream Butter (dairy) SandwichIce cream Cake (birthday/dessert) PotatoTurkey Sausage links (meat) CookiesTurkey Stuffing (dinner) PineappleMeatballs Bologna (meat) PretzelMeatballs Pizza (dinner) JuicePancake Noodles (grains) Potato chipsPancake Fried egg (breakfast) Peanut butterCereal Bread (grains) WatermelonCereal Bacon (breakfast) WaterShake Juice (beverage) SoupShake Cupcake (birthday/dessert) CerealSoda pop Water (beverage) CheeseSoda pop Potato chips (snack) BroccoliApple Watermelon (fruit) StuffingApple Sandwich (lunch) Fried eggGrapes Pineapple (fruit) Sausage linksGrapes Pretzel (snack) BaconSalad Broccoli (vegetables) CupcakeSalad Soup (lunch) ButterCelery Potato (vegetables) PizzaCelery Peanut butter (snack) Cake

Appendix C

Food Triads for Experiment 5Target Taxonomic Script

Cheese Shake (dairy) Cracker (snack)


Milk Butter (dairy) Cookie (snack)Butter Cheesecake (dairy) Toast (breakfast)Sausage links Steak (meat) Pancakes (breakfast)Waffle Pretzel (grains) Sausage links (breakfast)Roll Cookie (grains) Soup (dinner)Ice cream Milk (dairy) Cake (birthday)Cream Cheese Shake (dairy) Bagel (lunch)Bologna Steak (meat) Bread (lunch)Turkey Bacon (meat) Gravy (dinner)Toast Cracker (grains) Fried egg (breakfast)Pancake Pretzel (grains) Syrup (breakfast)Bread Cereal (grains) Peanut butter (lunch)Meatballs Bologna (meat) Spaghetti (dinner)Oatmeal Noodles (grains) Juice (breakfast)


Documents

An Apple is More Than Just a Fruit: Cross-Classification ...people.uncw.edu/nguyens/files/nguyen and murphy 2003.pdf · classification in children’s concepts, focusing on taxo-nomic