How Two- and Four-Year-Old ChildrenInterpret Adjectives and Count Nouns

D. Geoffrey HallM.R.C. Cognitive Development Unit, London

Sandra R. WaxmanNorthwestern University

Wendy M. HurwitzHarvard University

HALL, D . GEOFFREY; WAXMAN, SANDRA R.; and HURWITZ, WENDY M. HOW TWO- and Four-Year-Old Children Interpret Adjectives and Count Nouns. CHILD DEVELOPMENT, 1993, 64,16511664.We examined the role of object kind familiarity (i.e., knowledge of a count noun for an object)on preschoolers' sensitivity to the relation between a novel word's form class (adjective or countnoun) and its reference (to a material kind-property or to an object kind). We used a forced-choicematch-to-target task, in which children learned a word for one object (e.g., a metal cup), andthen chose between 2 other objects. One was from the same object kind but a different materialkind (with different related properties, such as color and texture; e.g., a white plastic cup); theother was from a different object kind but the same material kind (with the same related proper-ties; e.g., a metal spoon). In Experiment 1, children learned either a count noun (e.g., "This isa zav") or an adjective (e.g., "This is a zav one"). Within each form class, we crossed the familiar-ity of the referent object kind (familiar and unfamiliar) with the age of the children (2- and4-year-olds). The principal finding was that in interpreting an adjective, 4-year-olds were morelikely to choose the object sharing material kind with the target if the target was familiar thanif it was unfamiliar. No such familiarity effect was evident among 2-year-olds. In Experiment 2,we employed a more unambiguously adjectival frame (e.g., "This is a very zav-ish one"), andreplicated the results of Experiment 1. We interpret the results in terms of 2 proposed wordlearning biases: one that learners initially expect any word applied to an unfamiliar object torefer to a (basic-level) kind of object, and a second that learners prefer words to contrast inmeaning. We consider several interpretations of the observed age difference.

Young children know that a word may ries are not marked transparently in tbe in-belong to any of several different grammati- put children receive. For example, countcal categories, and that these different gram- nouns do not have a universally constant se-matical categories are associated with differ- rial position, stress or pitch level, or identi-ences in reference. This knowledge is both fying affix (Pinker, 1984).important and impressive. The knowledge isimportant because it gives children a tool for There have been several experimentallimiting hypotheses about word meaning. demonstrations of preschoolers' ability toFor example, if children can identify a novel identify words as coming from differentword as a count noun, then they can infer grammatical categories, and to make rele-that tlie word refers to a kind (or a category) vant inferences about meaning. For exam-of individual (e.g., object). The knowledge pie. Brown (1957) showed 3- and 4-year-oIdis impressive because it is not obvious how children a drawing of a pair of hands per-children come by it; the grammatical catego- forming an action upon a substance in a con-

The research in this article was reported by W.M.H. in a senior honors thesis in the Depart-ment of Psychology at Harvard University. A grant from that department supported this work.We thank the directors, teachers, parents, and children at the following schools for their gracioushelp: Harvard Yard Day Care, Kehillith Israel Nursery School, Serendipity Child DevelopmentCenter, and Soldier's Field Park Day Care. We thank Skyler Vinton for help with data collection.We are grateful to Bill Merriman, Steven Pinker, Nancy Soja, and an anonymous reviewer for veryhelpful comments on an earlier version of the manuscript. Correspondence may be addressed tothe first author at the M.R.C. Cognitive Development Unit, 4 Tavlton Street, London WCIHOBT, England.

[Child Development, 1993,64,1651-1664. 1993 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/93/6406-0007$01.00]

1652 Child Developmenttainer. If children heard the drawing de-scribed with a novel count noun ("a sib"),then they interpreted the word as referringto a kind of object (the container); if childrenheard a mass noun ("some sib"), then theytook the word to refer to a kind of material(the substance); and if they heard a verb("sibbing"), then they interpreted it as refer-ring to an action (the action performed bythe hands). More recently, Katz, Baker, andMacnamara (1974; also Gelman & Taylor,1984) showed that children as young as 2years of age expect that a count noun (e.g.,"This is a zav") applied to an object will re-fer to a kind of object, while a proper name(e.g., "This is Zav") will refer to an indi-vidual.

Recently scholars have attempted to ex-plain how young children learn the distinc-tions in reference associated with wordsfrom different grammatical categories. Theacquisition of the category, count noun, maybe the result of a relation between (1) chil-dren's expectations about the reference ofwords used under certain conditions and (2)caretakers' tendency to provide count nounsunder these same conditions (e.g., Macna-mara, 1982; Pinker, 1984). A growing bodyof evidence suggests that young children ex-pect that a word applied to an unfamiliar ob-ject (i.e., an object for which they know nocount noun for the kind) in an ostensivedefinition will refer to a kind of object. Morespecifically, this expectation appears to bethat the word will refer to a particular kindof object, namely, a basic-level kind. Basic-level kinds have members that, among otherthings, share an intermediate level of shapesimilarity (for discussion, see Hall & Wax-man, 1993). In this paper, our use of the ex-pression "object kind" is always meant toimply "basic-level object kind," unless oth-erwise noted.

Some of tbe evidence that children havean object kind expectation in learning wordsconsists of demonstrations that children willextend a count noun (e.g., "This is a dax")applied to an unfamiliar object to anotberobject of the same object kind, rather thanto an object of a different object kind sharingsome other feature with the target, such asa thematic link, a color, or a texture (e.g.,Baldwin, 1989; Landau, Smith, & Jones,1988; Markman & Hutchinson, 1984; Taylor& Gelman, 1988; Waxman & Kosowski,1990).

More striking evidence of children'spreference to extend a word for an unfamil-

iar object to other objects of the same objectkind consists of findings that children willdo this even if the word is not a count noun,and therefore could not refer to a kind ofobject (in the adult language). For example,Soja (1992) and Soja, Geirey, and Spelke(1991) have shown that 2-year-olds expectthat a word applied to an unfamiliar objectwill refer to a kind of object, whether itsform class implies a count noun interpreta-tion (e.g., "This is a zav"), an unspecifiednoun interpretation (e.g., "This is my zav"),or a mass noun interpretation (e.g., "This issome zav"). Dickinson (1988) has replicatedSoja et al.'s findings with 3-, 4-, and even5-year-old children. A similar demonstrationhas been made by Markman and Wachtel(1988), who found that 3-year-old childrentended to interpret a novel mass noun (e.g.,"This is pewter") applied to an unfamiliarobject as referring to a kind of object. Fur-thermore, Hall (1991) uncovered a similareffect in 2-year-oIds' interpretation of aproper name (e.g., "This is Zav") applied toan unfamiliar object.

Fortunately for children, caretakers'strategies for introducing new words un-der ostension seem to dovetail with chil-dren's interpretative preferences. Caretak-ers strongly prefer to offer count nouns(words that refer to kinds of object), ratherthan mass nouns or proper names, as osten-sive labels for (unfamiliar) objects (see Cal-lanan, 1985; Hall, in press; Ninio, 1980;Shipley, Kuhn, & Madden, 1983). The com-bination of young children's interpretativepreference and caretakers' input suggestshow children may discover the grammaticalcategory, count noun. That is, children ex-pect that words applied ostensively to unfa-miliar objects will refer to kinds of object,and the words caretakers typically provideostensively for novel objects are countnouns. Once children know these words'grammatical correlates (e.g., that they maybe preceded by the indefinite article), thenthey may use this knowledge to assist inidentifying novel count nouns in the inputaddressed to them. In this manner, they maycome to succeed in drawing the relevant in-ferences described earlier in Brown (1957).

If children hold a strong bias to interpreta word applied to an unfamiliar object as re-ferring to a kind of object (i.e., as if it werea count noun), then how do they ever learnthat some words may refer to the individualobject (i.e., proper names), to its materialkind (i.e., mass nouns), to object kinds at dif-ferent hierarchical levels (i.e., subordinate-

Hall, Waxman, and Hurwitz 1653

level or superordinate-level, as opposed tobasic-level count nouns), or to any of itsproperties (i.e., adjectives)? A recurrentfinding from the experimental literature isthat young children appear to be much morelikely to interpret a novel word applied to anobject as referring to something other thana (basic-level) kind of object if the referentobject is familiar (if children already knowa count noun for that kind of object) than ifit is unfamiliar (if they do not know such aword).

For example, Soja et al. (1991; Soja,1992) and Dickinson (1988) had difficulty inteaching 2- through 5-year-olds mass nounsfor material kinds when the referents wereunfamiliar objects. In contrast, Prasada(1993) found that 2V2-3V2-year-olds willreadily interpret a mass noun as referring toan object's material kind when the object isfamiliar.^ Moreover, Markman and Wachtel(1988) found that children interpreted amass noun as referring to a kind of object ifthe referent object was unfamiliar; however,if the object was familiar, children morereadily interpreted the word as referring tomaterial kind. Similarly, Hall (1991) foundthat children were more likely to interpret aproper name as referring to an individual ifthe referent object was familiar than if it wasunfamiliar. And Taylor and Gelman (1988,1989) have presented evidence that childrenare more likely to interpret a count noun asreferring to a subordinate-level kind, ratherthan a basic-level kind, if the referent objectis familiar than if it is unfamiliar.

The preceding effects of familiarityupon interpretation may reflect two things.The first is a strong bias to interpret a wordapplied to a novel solid object as referring tothe (basic-level) object kind (see Markman,1989). The second is a bias to avoid inter-preting two words as having the same mean-ing. Two versions of this second bias havebeen discussed in the literature: mutual ex-clusivity and lexical contrast. According tomutual exclusivity, the more stringent ver-sion, children should expect that any objectwill be a member of only one object kind.

Although some of the observed familiarityeffects are consistent with mutual exclusiv-ity (e.g., Markman & Wachtel [1988]; thefindings of Dickinson [1988], Prasada[1993], Soja [1992], Soja et al. [1991]; andHall [1991]), mutual exclusivity cannot ac-commodate all the observed familiarity ef-fects (e.g., Taylor & Gelman, 1988,1989; seealso Waxman & Senghas, 1992). The failureof mutual exclusivity to account for all thefindings suggests that the bias guiding chil-dren's word learning may be less stringent.Children may instead be adhering to a lexi-cal contrast bias (Clark, 1983, 1987). Ac-cording to lexical contrast, children expectthat no two words will have the same mean-ing, but words may have overlapping ref-erence (see Gathercole, 1987, for criticaldiscussion). Thus, a solid object m,ay beconstrued as a member of more than one ob-ject kind.

The effect of familiarity on word inter-pretation has now been documented forwords from several grammatical categories.But there is one category of word for whichresearch has failed to uncover a clear famil-iarity effect, namely, the adjective. Unlikecount nouns, which refer to kinds of individ-uals (e.g., objects), adjectives refer to proper-ties.^ The pattern of previous results leadsto the prediction that young children shouldbe more likely to interpret a word as refer-ring to a property, rather than to a kind ofobject, if the referent object is familiar thanif it is unfamiliar. Some recent research of-fers indirect support for this prediction.

Consider first a recent paper document-ing considerable difficulty in teaching chil-dren adjectives for unfamiliar objects.Smith, Jones, and Landau (1992) taught onegroup of 3-year-old children an adjective(e.g., "This is a dax one") and another groupa count noun (e.g., "This is a dax") for atarget, an unfamiliar geometric object with asalient property (multicolored paint speck-les or silver-gold glitter). Children then sawa set of other objects that differed from thetarget in shape and/or salient property. Intheir first two experiments. Smith et al.

^ Prasada also mentioned the familiar count noun in teaching the new mass noun. Forexample, to teach the mass noun "sponge" for a ball made of sponge, he said, "This ball is madeof sponge; this is a sponge ball." This procedural detail, mentioning the familiar count noun(e.g., "ball"), may have contributed to his success in getting such young children to make amaterial kind interpretation of the mass noun; see Waxman, Shipley, & Shepperson (1991).

Recent analyses reveal principled distinctions between the semantics of adjectives (e.g.,"brown") and count nouns (e.g., "dog"). For example, count nouns, but not adjectives, provideprinciples of identity and individuation; that is, only count nouns pick out individuals withidentity (see Cupta, 1980; Macnamara, 1986).

1654 Child Developmentfound that children extended a count nounto objects sharing a similar shape, regardlessof salient property; moreover, childrenshowed a similar, though weaker, tendencyto do the same if they heard an adjective.These results are consistent with the claimthat children expect that a word (either acount noun or an adjective) applied to anunfamiliar object will refer to a kind of ob-ject, because shape is correlated stronglywith object kind, at least for kinds of artifact.In a third experiment. Smith et al. made thesalient property extremely salient by pre-senting the target object in a dark toy "cave,"under intense light, so that the surface of theobject glowed. In this experiment, childrencontinued to extend the count noun on thebasis of shape, but now they extended theadjective on the basis of the salient property.Smith et al. thus were able to lead childrento interpret an adjective as referring to a sa-lient property of an unfamiliar object, butthe context had to be made extremely sup-portive of such an interpretation before chil-dren were willing to make it.

Now consider evidence that childrenwho learn adjectives ioi familiar objects areless likely to show a bias to interpret thewords as referring to kinds of object. Wax-man and Kosowski (1990) showed that chil-dren as young as 2 years of age (2- to 4-year-olds) tended not to interpret an adjective asreferring to a kind of object when the refer-ent object was familiar. One group of chil-dren learned a count noun (e.g., "This is afopin") and another learned an adjective(e.g., "This is a fopish one") for a drawingof a familiar object (e.g., a cow). All childrenthen had to choose two other objects towhich the new word applied. Two of thesewere from the same superordinate objectkind (e.g., a fox, a zebra) and two were re-lated thematically to the target (e.g., milk, abam). Children who heard a count noun forthe familiar object tended to select the ob-jects from the same superordinate objectkind; children who learned an adjective didnot. Waxman and Kosowski (1990) did notshow directly that the young childrenmapped an adjective onto a property inter-pretation, but they did show that childrenclearly avoided an object kind interpreta-tion. Waxman (1990) has made a similardemonstration with 3-year-olds using a dif-ferent task.

The preceding studies offer indirect evi-dence of a familiarity effect in the interpreta-tion of novel adjectives, but it would bemore desirable to compare familiar and unfa-miliar objects within a single experiment.

keeping all procedural details aside from fa-miliarity constant. Taylor and Gelman (1988,Experiment 1) reported such a comparison.They tested 2-year-olds in a toy selectiontask. Children in this study saw four toys,members of two object kinds crossed withtwo kinds of material having distinct proper-ties, such as color and texture (pale greenfake fur and yellow and black plaid mate-rial). For half the subjects, both kinds of ob-ject were unfamiliar (two kinds of stuffedcreature); for the other half, the kinds werefamiliar (stuffed dogs and stuffed birds).Within each group, half the children learnedan adjective (e.g., "This is a zav one") forone of the toys; half learned a count noun(e.g., "This is a zav"). After learning theword, children carried out a series of actionsin response to requests that included thenovel word. Taylor and Gelman (1988) in-ferred children's interpretations of the novelword from the toys children selected. An ob-ject kind interpretation was inferred if chil-dren selected toys within the same kind, ir-respective of salient property; a propertyinterpretation was inferred if children se-lected toys that had the same salient prop-erty, regardless of object kind.

Taylor and Gelman (1988) found that 2-year-olds made relatively more property in-terpretations if the word was an adjectiveand more object kind interpretations if theword was a count noun, averaging over ob-ject familiarity. Children thus showed a sen-sitivity to the distinction in reference be-tween the two types of word. In addition,Taylor and Gelman found a familiarity ef-fect, but only in the interpretation of thecount noun. That is, children who learned acount noun were more likely to focus on thenamed object only, and less likely to selectthe other member of the same object kind,if the object was familiar than if it was unfa-miliar. Subsequent research suggested thatthis finding likely reflected the fact that chil-dren were making a subordinate-kind inter-pretation of the count noun applied to thefamiliar object (Taylor & Gelman, 1989).However, Taylor and Gelman (1988) did notfind a familiarity effect in the interpretationof a novel adjective.

Taylor and Gelman's (1988) findingssuggest that 2-year-olds are sensitive tothe relation between adjectival form classand reference to a property, because theyshowed a tendency to make a property inter-pretation of an adjective irrespective of ob-ject familiarity. However, their failure tofind a familiarity effect calls out for further

Hall, Waxman, and Hurwitz 1655investigation for two reasons. First, other in-vestigations have found familiarity effects inthe acquisition of words from other gram-matical categories (as did Taylor & Gelman,1988, themselves for count nouns); those re-sults suggest that children should be morelikely to focus on object kind and less likelyto attend to a property (or to material kind)if the referent object is unfamiliar than if itis familiar. Second, other research on chil-dren's interpretations of adjectives lends in-direct support to the hypothesis that adjec-tives are easier to learn for familiar thanunfamiliar objects (e.g.. Smith et al., 1992;Waxman & Kosowski, 1990).^

In this article, we pursue the study ofchildren's expectations concerning the refer-ence of adjectives. In Experiment 1, we ex-amined the effect of object kind familiarityon children's interpretations of novel adjec-tives. In order to determine whether any ob-served familiarity effect was a general effectof acquiring words rather than adjectives inparticular, we manipulated the form class ofthe words taught; half the subjects learnedadjectives, and the remainder learned countnouns.

The general design of this experimentis reminiscent of Experiment 1 of Taylor andGelman (1988), but it differed in two impor-tant ways. First, in order to see if Taylor andGelman's (1988) findings held up with achange in method, we used a forced-choicematch-to-target task. Use of this task enabledus to employ several sets of objects, ratherthan only one set, for each child. Second,and more important, we were curious aboutany differences in the interpretation of ad-jectives at different times during the pre-school years. Taylor and Gelman (1988)failed to find a familiarity effect with 2-year-olds, and there has been some recent debateabout whether the tendency to assume mu-tual exclusivity (or lexical contrast) is evi-dent in 2-year-olds' word learning. Some ev-idence suggests that 2-year-olds are lesslikely to show a familiarity effect than are4-year-olds (for discussion, see Merriman,1991, and Woodward & Markman, 1991; see

also Merriman & Bowman, 1989; Merriman& Schuster, 1991). To examine the issue ofage-related differences in the impact of ob-ject kind familiarity on interpretation, wetested equal numbers of 2-year-olds and 4-year-olds on the task.

Experiment 1Method

Subjects.Eighty children partici-pated, 40 2-year-olds and 40 4-year-olds. The2-year-olds ranged in age from 2 years, 2months to 2 years, 11 months, with a meanage of 2 years, 7 months. The 4-year-oldsranged in age from 4-1 to 4-11, with a meanage of 4-5. Ten children of each age wereassigned randomly to each of four condi-tions, such that all conditions had approxi-mately equal numbers of boys and girlsand approximately the same mean age. Thefour groups were: unfamiliar targetsnoun(mean ages 2-6 and 4-5), unfamiliar targetsadjective (mean ages 2-6 and 4-6), familiartargetsnoun (mean ages 2-8 and 4-5), fa-miliar targetsadjective (mean ages 2-7 and4-4). Children were tested in their pre-schools during normal school hours. Theywere from primarily white middle- and up-per-middle-class backgrounds.

Materials.Eight triads of objects wereused, four involving unfamiliar objects andfour involving familiar objects. In each triadthere was a target object, for which the chil-dren were taught a novel word. In addition,there was an object that matched the targetin object kind but differed in material kind(and related properties, such as color andtexture). The third object matched the targetin material kind (and related properties) butdiffered in object kind from the target. In theUnfamiliar conditions, the object kinds wereall intended to be unfamiliar (i.e., it was in-tended that children knew no count nounfor the basic-level object kind). The objectkinds in the Familiar conditions were in-tended to be familiar. The material kinds(and related properties) used in each triadof Unfamiliar stimuli corresponded to thoseused in a triad of Familiar stimuli. Table 1describes all triads.

^ There is another study that has examined preschoolers' understanding of adjectives. Cel-man and Markman (1985) report two experiments on children's understanding of adjectives andcount nouns. However, they did not teach children a word for an object in order to examine towhich other objects children would extend it. Instead children in these experiments saw arraysof four drawings. The authors' interest lay in seeing whether children expected an adjective(e.g., "Find the zav one") to pick out a drawing that depicted an object that contrasted on adimension within an object kind, and whether they expected a count noun (e.g., "Find the zav")to pick out a drawing that depicted an object that contrasted with other objects in object kind.They found some evidence that (especially older) preschoolers had this expectation.

1656 Child DevelopmentTABLE 1

DESCRIPTION OF STIMULUS TRIADS

Target Same Object KindUnfamiliar sets:

glass tongs red plastic tongsmetal garlic press white plastic garlic pressstraw cornucopia denim cornucopiablack leather warmer blue wool warmer

Familiar sets:glass plate red plastic platemetal cup white plastic cupstraw hat denim hatblack leather glove blue wool glove

Same Material Kind

glass napkin ringmetal apple corerstraw basket handleblack leather pack

glass cupmetal spoonstraw basketblack leather belt

Familiarity of the object kinds.Weconducted a separate pretest with a differentgroup of children to verify that the intendedunfamiliar objects were unfamiliar (i.e., thatchildren likely knew no count noun for theobject kinds) and that the familiar objectswere familiar. We based this pretest on theone employed by Markman and Wachtel(1988, Experiment 2). Seven 2-year-olds(mean age 2-9; ranging from 2-5 to 2-11) andseven 4-year-olds (mean age 4-5; rangingfrom 4-1 to 4-8) were tested individually.None participated in either of the experi-ments.

The experimenter first provided a prac-tice trial. She showed children a pencil (afamiliar object not included as a target or testitem), and asked them if they knew a namefor the kind of thing. All children answered"a pencil." The experimenter then said"yes," and told them that it was good to saythe names for kinds of things that they knew.She next showed children a plastic object(an unfamiliar object not included as a targetitem) and asked if they knew a name for it.No child provided a label. The experimenteradmitted that she also knew no name for thekind of thing (a true statement), and that itwas good not to say a name for kinds ofthings that one did not know.

Children then saw each of the 24 objectsdescribed in Table 1, one at a time. For eachobject, the experimenter asked the childrenif they knew what it was. If children pro-duced an appropriate count noun, or if theyindicated that they did not know a name, theexperimenter moved to the next item. If chil-

dren attempted some description of the ob-ject, then they were given a forced-choicecomprehension task. In this task, childrensaw the object along with two other unfamil-iar distractors. The distractors were selectedfrom among the following; a white plasticpasta measurer, a pasta measurer made ofwooden dowels, a clear plastic architect'scurve, a piece of a purple plastic clip, a blackmetal three-hole punch, a white plastic potscraper with a thumb hole, and a plastic ac-cordion telescope. Children were asked topoint to the X, where X was the (basic-level)count noun for the object. Thus, an objectwas judged familiar if children generated anappropriate count noun for it, of if theypointed to it in the comprehension task. Anobject was judged unfamiliar if children said"I don't know" or shrugged, or if they failedto point to it in the comprehension task.

We assigned each object a familiarityscore; this score ranged from 0% (neverjudged familiar) to 100% (always judged fa-miliar). The mean familiarity scores for thefamiliar targets (glass plate, metal cup,straw hat, black leather glove) were 100%for the 2-year-olds and 96% for 4-year-olds.The mean familiarity scores for the unfamil-iar targets (glass tongs, metal garlic press,straw cornucopia, black leather warmer)were 7% for the 2-year-olds and 14% for the4-year-olds. Mean familiarity scores alsowere computed for the familiar test objects;this score was 91% for both the 2-year-oldsand the 4-year-olds. Mean familiarity scoresfor the unfamiliar test objects then werecomputed; they were 5% for the 2-year-oldsand 2% for the 4-year-olds.*

* For familiar target items, the 2-year-olds and 4-year-olds received the comprehension teston average on 11% and 7% of trials, respectively. For unfamiliar target items, 2-year-ofds receivedthe comprehension test on 21% of the trials, on average; 4-year-olds did so on 29% of the trials,on average. For familiar test items, the comprehension test was administered on 9% of trials for

Hall, Waxman, and Hurwitz 1657TABLE 2

MEAN NUMBER OF MATERIAL KIND SELECTIONS OUT OF FOUR

Unfamiliar Familiar

Experiment 1:Count noun:

2-year-olds 1.2 (.63)* .7 (.48)*4-year-olds 4 (.70)* 1.0 (1.63)

Adjective:2-year-olds 1.6 (.97) 1.1 (.57)*4-year-olds 8 (1.03)* 3.0 (1.16)**

Experiment 2:Adjective:

2-year-olds 1.3 (1.06) 1.2 (1.03)*4-year-olds 1.4 (1.58) 3.3 (.68)**NOTE.N = 10 per condition. Standard deviations are in parenthe-

ses.* Mean is significantly less than chance, p < .05.** Mean is significantly greater than chance, p < .05.

In summary, the pretest results revealclear differences in object kind familiaritybetween the unfamiliar and familiar stimuli.Whereas the familiarity scores for the famil-iar objects were above 90% on average, thefamiliarity scores for the unfamiliar objectswere on average below 15%.

Procedure.In the experiment proper,children took part in four trials. Childrensaw either four unfamiliar triads or four fa-miliar triads. There were two conditions.

In the Count Noun conditions, the ex-perimenter began each trial by presentingthe ta.rget object from the triad, pointing toit, and labeling it with a count noun. Shesaid, for example, "See this? This is a zav.''The experimenter then asked the child torepeat the count noun. She then repeatedthe count noun up to three times herself.The target was kept in view as the child sawthe tv70 choices, one that matched the targetin object kind but differed in material kind,one tiiat matched in material kind but mis-matched in object kind. Children wereasked, "Can you find another zav?"

Children in the Adjective conditions re-ceived the same treatment, except that theyheard the novel word introduced as an adjec-tive. The experimenter labeled the object bysaying "See this? This is a zav one." Chil-dren were asked, "Can you find another onethat is zav?"

After making these selections, childrenwere asked for explanations. For example,pointing to the chosen object, the experi-menter asked, "Why is this a zav?" or "Whyis this a zav one?"

To counterbalance, we used a Greco-Latin square design. Subjects within eachcondition saw the object triads in a varyingorder and heard the nonsense words in aseparately varying order. The left-right posi-tion of the two choice objects was deter-mined randomly. The same Greco-Latinsquare and the same left-right positionswere used in all conditions. The novelwords used were "blick," "fep," "wug," and"zav."

Results and DiscussionChildren received a score from 0 to 4 to

reflect the number of trials on which theymade a material kind selection. We con-ducted a three-way ANOVA using this de-pendent measure, with age (2-year-olds, 4-year-olds), familiarity (Unfamiliar, Familiar),and form class (Count Noun, Adjective) asbetween-subjects factors. The means fromeach condition appear in Table 2.

There was a significant effect of famil-iarity, F(l, 72) = 4.35, p < .05. As predicted,there were more material kind selections ifthe target was familiar than if it was unfamil-iar. Form class also was a significant factor,F(l, 72) = 13.76, p < .0005, with children

the 2-year-olds and 4% for the 4-year-oIds, on average. For unfamiliar test items, the comprehen-sion test was administered on average on 23% of trials for 2-year-olds and 13% of trials for4-year-olds.

1658 Child DevelopmentTABLE 3

NUMBER OF CHILDREN MAKING THREE OR FOUR MATERIAL KINDSELECTIONS OUT OF FOUR

Unfamiliar Familiar

Experiment 1:Count noun:

2-year-olds 14-year-olds 0*

Adjective:2-year-olds 14-year-olds 1

Experiment 2:Adjective:

2-year-olds 24-year-olds 2

NOTE.N = 10 per condition.* Significantly below chance, p < .05.** Significantly above chance, p < .05.

0*2

0*g**

making more material kind selections if theylesirned an adjective than if they learned acount noun. Finally, the age x familiarityinteraction was significant, F(l, 72) 19.40,p < .0001. Follow-up tests of simple effectsrevealed a significant familiarity effectamong the 4-year-olds (p < .001), but notamong the 2-year-oIds (p > .10).

Three effects approached significance(p = .07). The first two were the age x formclass and the familiarity X form class inter-actions. Because we had a theoretical inter-est in these interactions, we computed sim-ple effects tests. These tests revealed thatthe effect of form class was significant among4-year-olds (p < .001) but not among 2-year-olds (p > .15), and that the effect of familiar-ity was significant in the Adjective condi-tions (p < .01), but not the Count Nounconditions (p > .85). The third near-sig-nificant effect was the three-way interaction,which can be interpreted as showing that theage X familiarity interaction (i.e., the ten-dency for 4-year-olds but not 2-year-olds toshow a familiarity effect) was larger for theAdjective than for the Count Noun condi-tions.

We next considered children's perfor-mance with respect to chance. If childrenhad selected items at random on the four tri-als, then they should have made two mate-rial kind selections out of the four forcedchoices (i.e., they should have made a mate-rial kind selection on 50% of the trials). Theresults showed that 4-year-olds in the Famil-iar Adjective condition were the only groupwho made significantly more material kind

selections than would have been expectedby chance: i(9) = 2.79, p < .05; 4-year-oldsin the Unfamiliar Adjective condition weresignificantly below chance, t(9) = 3.67, p