Lexical Principles May Underlie the Learningof Verbs

Roberta Michnick Golinkoffand Roberta Church JacquetUniversity of Delaware

Kathy Hirsh-PasekTem,ple University

Ratna NandaknmarUniversity of Delaware

GOLINKOFF, ROBERTA MICHNICK; JACQUET, ROBERTA CHURCH; HIRSH-PASEK, KATHY; and NANDAKU-MAR, RATNA, Lexical Principles May Underlie the Learning of Verbs. CHILD DEVELOPMENT, 1996,67, 3101-3119, To examine whether children (mean age 34 months) can fast map and extendnovel action labels to actions for which they do not already have names, the comprehension offamiliar and novel verbs was tested using colored drawings of Sesame Street characters per-forming both familiar and unfamiliar actions. Children were asked to point to the character"verbing," from among sets of 4 drawings. With familiar words and actions, children made correctchoices 97% of the time. With novel action words, children performed at levels mostly signifi-cantly above chance, selecting a previously unlabeled action or another token of a just-names ac-tion. In a second, control experiment children were asked to select an action from among the samesets of 4 drawings, but they were not given a novel action name. Here children mainly demonstratedperformance at levels not significantly different from chance, showing that the results from themain experiment were attributable to the presence of a word in the request. Results of thesestudies are interpreted as support for the availability of principles to ease verb acquisition.

How would we know that children had tional experience with the word in otherlearned a novel verb? First, children should sentential and extralinguistic contexts isbe able to produce or comprehend the newly necessary to refine the originally assignedlearned verb in the original context in which meaning,it was learned. Second, to show true wordlearning, children should be able to extend The Phenomenon of Fast Mappingthe verb to new, nonidentical instances of In Garey and Bartlett's (1978) study ofthe action that the verb labels. Although fast mapping, 3- and 4-year-oId childrenin theory (Quine, 1960) such learning might were introduced to a novel adjective ("chro-be arduous and require many trials, a num- mium") by a teacher who casually askedber of studies show that children can "fast children to "bring me the chromium tray,map" (Garey & Bartlett, 1978) a novel word not the blue one, the chromium one" fromto a meaning in a single trial. That is, upon among a set of trays, Ghildren were able tohearing a novel word in a specific context, respond correctly to the request as well as tochildren implicitly recognize tbat tbey have demonstrate some recollection of the word'sheard a novel word and that it maps to a meaning (e.g., that it was a color word) afterreferent in that context. Of course, from a a delay of 1 week. Heibeck and Markmansingle exposure tbe cbild cannot know tbe (1987) replicated and extended Garey andnuances of a novel word's meaning; addi- Bartlett's results to the domains of texture

This article was presented at the Society for Research in Child Development in April 1991,in Seattle, We thank Kelly Olguin for assistance in running these studies, Carolyn Mervis andMelissa Schweisguth for their comments, and Alice Eyman, Director of the University of Dela-ware Laboratory Preschool for her help in obtaining subjects. The research was supported inpart by grants #l-R01-HD1596'4-01 and #l-R01-HD19568-01 from NICHHD to the first andthird authors. Requests for reprints should be sent to Golinkoff at the College of Education,University of Delaware, Newark, DE 19716, Golinkoff's e-mail address is Roberta @ Udehedu,

[Child Development, 1996,67,3101-3119, 1996 by the Society for Research in Child Development, Inc,All rights reserved, 0009-3920/96/6706-0008801,00]

3102 Child Developmentand shape. To make certain that childrenhad actually learned an abstract meaning forthe novel word and had not just treated it asthe name of only the target item, they askedchildren to extend the newly learned wordto another exemplar of the concept. Theirfindings indicated that fast mapping had in-deed occurred, although more readily in thedomains of shape and color than texture (seealso, e.g., Au & Markman, 1987; Dockrell,1981; Markman & Wachtel, 1988).

Fast mapping (and in some cases, exten-sion of the novel word) also occurs withobject words (Dockrell, 1981; Dockrell &Campbell, 1986; Hutchinson, 1986; Mark-man & Wachtel, 1988; Merriman & Bow-man, 1989; Merriman & Schuster, 1991; Rice& Woodsmall, 1988). In one study by Golin-koff, Hirsh-Pasek, Bailey, and Wenger(1992), whose design was used in the re-search reported here, 28-month-old childrenshowed an impressive degree of object labellearning. On each trial, subjects were shownfour objects, some of whose names theyknew and some of whose names they did notknow. During the course of a single experi-mental session, children were asked to fastmap and extend a total of eight new objectlabels. Children could not have simplylearned to give the experimenter the novelobject on each trial since a familiar objectwas always requested as well. In addition,extension ofthe novel term to another exem-plar was tested in the presence of yet an-other unfamiliar object which had not yetbeen named. This ruled out the possibilitythat children just interpreted the novel termas mapping to any novel object.

While the availability of the fast map-ping process for the learning of adjectivesand nouns has been shown repeatedly,fewer studies have examined the fast map-ping of verbs or action labels. In the studiesthat have examined both object and actionlabels in the same fast mapping task, actionlabels often prove more difficult to fast mapthan object labels. Rice and Woodsmall(1988) showed 3- and 5-year-old childrenvideotapes of events which presented newobject, action, and property labels in a storyformat. Word learning occurred at a signifi-cantly lower rate with action labels com-pared to object and property labels. Merri-man, Marazita, and Jarvis (1993, 1995)similarly found a difference in favor of thefast mapping of object labels over action la-bels in studies with 4-year-olds. Using amodification of the intermodal preferentiallooking paradigm (Golinkoff, Hirsh-Pasek,

Cauley, & Gordon, 1987), children wereasked to point at one of two simultaneouslypresented videotapes. Either a novel and fa-miliar object were paired or a novel and fa-miliar action. These pairs were accompaniedby a novel noun or a novel verb, respec-tively.

Why should the evidence suggest thatobject labels are easier to fast map than ac-tion labels? This phenomenon accords wellwith the eEirly acquisition data that suggestthat object words predominate over actionwords (e.g.. Bates, Bretherton, & Snyder,1988; Dromi, 1987; Gentner, 1983, 1988;Goldin-Meadow, Seligman, & Gelman,1976; but see Bloom, Tinker, & Margulis,1993; Gopnik & Choi, 1995; Tardiff, 1996)even in languages such as Korean whichallow much noun ellipsis and place verbsin sentence final position (Au, Dapretto, &Song, in press). Verbs may appear moreslowly than nouns for at least five reasons:First, verb learning requires uncovering thesemantic components of the verb, for exam-ple, implied causation, direction or locationof action, or manner of action (Behrend,1990; Gentner, 1983; Pinker, 1989). Second,the referents of nounsoften concrete ob-jectsare more stable and perceptuallyavailable than the referents of actions which,unless repetitive, tend to be ephemeral(Gentner, 1988; Givon, 1973); third, it maybe more difficult to detect what is constantin an action compared to what is constantin an object. For example, is "running" stillrunning when you observe both your grand-mother and Frank Shorter do it? Fourth,multiple verbs can be used to describe thesame event. "Running" may also be called"jogging," "dashing," or "sprinting." Andfinally, to use a verb appropriately the childmust have mastered the obligatory argumentstructure it requires (Bloom, Lightbown, &Hood, 1975; Gleitman, 1990; Pinker, 1989)(e.g., does it or does it not require an ob-ject?). Although verb learning may presentthese problems, children do begin to acquireverbs in their productive lexicons by at least21 months (Bloom & Lahey, 1978) and, intheir receptive vocabularies, by at least 16months (Golinkoff et al., 1987).

y Does Fast Mapping Occur?In fast mapping studies it is up to chil-

dren (as it often is in life) to find a likelyreferent for a novel >vord. Although childrenappear to be capable of using social/prag-matic cues such as the speaker's gaze direc-tion and intentionality (Baldwin, 1993; To-masello & Barton, 1994) to assist them, such

Golinkoff et al. 3103cues are not always available and, in anyevent, cannot aid children in deciding on thecorrect construal for the meaning of a novelword. To solve such problems for the learn-ing of novel nouns, several researchers ar-gue that children must possess a principlewhich allows them to readily assign new la-bels to previously unnamed objects. Mark-man (1987; Merriman & Bowman, 1989) of-fered the "mutual exclusivity" principle:children select the unnamed object becausethey are biased against allowing objects tohave more than one name. Clark's (1983)"contrast" principle predicts that childrenchoose the unnamed object on the assump-tion that a new word must differ in meaningfrom the words they already know. Golin-koff, Mervis, and Hirsh-Pasek (1994), find-ing these principles lacking after a carefulreview (see also Gathercole, 1989; Rice,1990), suggested the "novel namenamelesscategory" principle (or N3C): the child willfind an unnamed, as opposed to an alreadynamed, referent for a novel name. Golinkoff",Hirsh-Pasek, Mervis, Frawley, and Peirillo(1995) argued that children must be simi-larly motivated to attach novel action namesto unnamed actions. The rapid word learn-ing observed in the second year of life couldnot occur if children did not search out po-tential referents for novel words in nonos-tensive contexts.

Verb ExtensionThe fast mapping of verb meaning

would be of limited utility if the child couldnot use a newly learned verb to label novelexemplars ofthe action; the economy of lin-guistic representation rests on the fact thatwords label categories (be they categories ofobjects, actions, or properties) and notunique instances. For nouns and adjectives,research indicates that after only one or twononostensive exposures to a word-referentcombination, extension occurs to new exem-plars of the word (e.g., Golinkoff, Hirsh-Pasek, et al., 1992; Heibeck & Markman,1987). Very little is presently known aboutwhether extension of novel verbs occurs asreadily. Behrend (1990; Behrend, Harris, &Cartwright, 1955) and Forbes and Farrar(1993) study verb extension as a way to dis-cover how children construe verb meaning.First, Behrend et al. (1995) taught childrena novel verb for a novel action. For example,children were shown a videotape of a persontwirling a spaghetti server to collect tangledyarn and told that the person is "pinting."Second, children were shown variations onthis event in which either the instrument

was changed (a rachet) or the action changed(scooping) or the result or outcome of theevent was changed (the yarn was left dan-gling). The dependent variable was the fre-quency with which children were willing toextend the novel verb to the different typesof changes. Three-year-old children ex-tended the novel verb most to instrumentchanges and much less to both action andresult changes; 5-year-olds also extendedmost to instrument changes and rejected re-sult changes.

One of the variables in the Forbes andFarrar (1993) study that significantly re-duced extension for both children and adultswas a change in "causative agent" orwhether "the person either moved himselfor was moved by others" (p. 278). A changein who is the agent of an action seems to beone of the most basic kinds of change thatan action can undergo; even 14-month-oldinfants can discriminate changes in theagent-patient relationships in an event frommore superficial changes (such as the left-right direction of the action across a moviescreen) (Golinkoff, 1975; Golinkoff & Kerr,1978). Thus, even before children havelearned many verbs they apparently recog-nize that a change in the agency of a nonlin-guistic event is a significant change in themeaning of the event. And yet, to accuratelyextend a novel verb, children must knowthat the same action can be performed by anumber of agents. Thus, verb extension, likenoun extension, is based on the implicit rec-ognition that a verb labels a category of ac-tions, potentially performed by a variety ofagents, in a variety of locales and orienta-tions.

Once again, researchers who have stud-ied noun acquisition have posited principlesfor noun extension that may work for verbextension. The assumption that object wordslabel categories has been captured in differ-ent ways by three principles: the "taxonomicassumption" (Markman & Hutchinson,1984), the "noun-category bias" (Waxman &Kosowski, 1990), and the "categoricalscope" principle (Golinkoff et al., 1994; Go-linkoff, Shuff-Bailey, Olguin, & Ruan, 1995;Shuff-Bailey & Golinkoff, 1995). For objectlabels, the categorical scope principle statesthat children will extend names to other ex-emplars in the same basic level category.Golinkoff, Hirsh-Pasek, et al. (1995) arguedthat categorical scope applies to actionwords as well. Specifically, an action wordmay be extended to an action of the sametype regardless of (1) the agent performing

3104 Child Developmentthe action and (2) superficial differences inthe manner in which the action is per-formed. That is, each instance of an actionof the same typeeven though performedby a different agent, or in a different place,or in a different mannershould be calledby the same name in order to keep the num-ber of action terms within the realistic limitsof human memory. The present study testsfor the most basic form of action word exten-sion: can a newly learned action word beimmediately extended to a new agent per-forming the same exact action?

To test children's fast mapping and ex-tension of novel action words, 3-year-oldswere shown drawings of popular SesameStreet characters performing some familiaractions (e.g., waving) and some unfamiliaractions (e.g., doing an arabesque). The firstquestion was whether children would as-sume that a novel action word mapped to anunfamiliar action, in the presence of familiaractions. Importantly, no explicit link wasmade between the novel action and thenovel action term; it was up to the child tofigure out to which action the novel termmapped. The second question was whetherchildren would extend newly learned actionwords to other examplars of the same type.Children were shown another character per-forming the just labeled novel action along-side characters performing familiar actionsas well as another unnamed action. If chil-dren select the second exemplar of the justheard action name, and not the novel actionwithout the label, they will be showing thatthey consider the novel action name to be(a) not just linked to the original action (asort of "proper verb" hypothesis) and (b) notjust a label for any unnamed action but for aparticular action, namely, the one just la-beled.

Experiment 1To test for the fast mapping and exten-

sion of novel verbs, children participated ina task where they were asked to select draw-ings of Sesame Street characters performingboth novel and familiar actions. From amonga set of four drawings, children were askedto give the experimenter a picture of both aknown action (e.g., "Which one is waving?")and an unfamiliar action (e.g., "Which one isglorping?") on each trial in counterbalancedorder. Some trials were designed to test forfast mapping and some for extension of thenewly learned verb. Fast mapping would bedemonstrated if children selected the unfa-miliar action in response to a request which

included a novel verb. Verb extension (incomprehension) would be demonstrated ifchildren selected another exemplar of thenovel action, performed by a different char-acter, on a subsequent trial. On one of thetrials that tested for fast mapping, childrenwere given a second novel name and askedto select an action in the presence ofthe pre-viously named novel action and an unnamednovel action. If children are guided by aprinciple to map novel action terms to un-named actions, they should believe that theyalready have a name for the first unnamedaction and select the second unnamed ac-tion. However, given how few exposureschildren will have to the novel verb, thattrial in particular, and this study in general,are a rigorous test of whether children canfast map and extend novel verbs.

SubjectsSix boys and six girls (mean age = 34

months, 21 days; range 32,18-35,29) all na-tive speakers of English producingmultiword speech, were tested individuallyin a university laboratory as their parents si-lently looked on. (Children were selected tobe approximately 6 months older than sub-jects in the parallel noun study [see Golin-koff, Hirsh-Pasek, et al., 1992] in accord withfindings in the literature that knowledge ofverbs generally comes after children havealreadv learned some nouns.)Procedure

Stimuli.Stimuli consisted of 27 col-ored drawings (each 3 inches by 4 inches)depicting actions performed by Big Bird,Cookie Monster, Bert, and Ernie. As Table1 shows, 18 of the 27 were familiar actions(e.g., eating) and five were unfamiliar ac-tions (e.g., doing a split). Four of the unfa-miliar actions were also seen performed bya second character. For the selection of fa-miliar actions, research was consulted forthe names of actions young children werelikely to know (e.g.. Bloom et al., 1975; Go-linkoff et al., 1987). The unfamiliar actionswere selected on the criteria that they couldbe easily discriminated from the familiar ac-tions and were unlikely to be witnessed bychildren or, if witnessed, heard describedwith verbs. Figure 1 shows each ofthe unfa-miliar actions with their phrasal descrip-tions.

To test whether the actions that we la-beled "familiar" and "unfamiliar" had thisstatus for young children, or indeed,whether children interpreted these two-dimensional drawings as depicting actions.

Golinkoff et al. 3105TABLE 1

THE 18 FAMILIAR AND FIVE UNFAMILIAR ACTIONSUSED IN EXPERIMENTS 1 AND 2

Familiar

clappingcoloringcryingdressingdrinkingeatingfallingkickingpointingreadingridingrunningsittingsleepingstandingthrovs'ingwashingwaving

ACTION TYPE

Unfamiliar

doing a split"doing an arabesque"doing a backbend^doing a pushuptilting one's head"

" To test for extension, there were two drawings ofeach of these actions, e.g., one card showed Big Bird do-ing a split and one card showed Ernie doing a split.

we conducted a study with a separate groupof youngsters. Nine slightly older children(six males and three females, mean age =39 months, 9 days; range 35,10-43,28) weretested individually with the action drawings.Children on average 5 months older than tiheexperimental subjects were tested sinceyounger children were reluctant to apply la-bels to the actions. All children were nativespeakers of English and spoke in multiwordsentences.

Each child was told that we were goingto show him or her some pictures ofthe Ses-ame Street characters, some of which wouldbe "silly" and some "okay." In the "silly"ones the child might not know what the per-son was doing and wouldn't know what tocall it. In the "okay" ones, the child wouldknow what the person was doing and whatto call it. Children were told that they shouldput the silly ones in the silly pile (in frontof a picture of a silly face) and the okay onesin the okay pile in front of the smiley face.The child was then handed one card at atime (with the unfamiliar ones interspersedrandomly) and asked to place the card in theappropriate pile. As the child placed eachcard in a pile (whether it was familiar or un-familiar and regardless of the pile to whichit was assigned) the experimenter asked,

"What's he doing here?" or "Can you tellme what character name is doing?" If chil-dren did not interpret these drawings as por-trayals of actions they could have answeredin other ways despite the use of a gerundivesentence frame. Even older children willoverride the form class information presentin experimenters' questions if they are pre-disposed to answer in another way (e.g.,Markman & Wachtel, 1988). Childrengreatly enjoyed this game and were verysuccessful at it.

The results indicated that children didindeed interpret the drawings as depictionsof action. They often answered with a pro-noun and a verb in the progressive form, asin "He's eating." They were also not reluc-tant to indicate that they "didn't know" orthat the action was "silly" when it was unfa-miliar. Of the 18 cards which showed famil-iar actions, children were able to provide thecorrect name for the action 91.4% of thetime. For the five cards which showed unfa-miliar actions, children labeled them only15.6% ofthe time, and when they did, it waswith great hesitation and uncertainty. Thus,children's labeling responses validated ourhunches as to which actions were familiarand which unfamiliar. Additional support forchildren's ability to interpret these cards asdepictions of action comes from their com-prehension responses during the familiariza-tion test.

Familiarization.The 27 stimuluscards were first spread out randomly on thefloor in front ofthe child. Children were firstasked to name the familiar actions in randomorder in response to a question like, "Whatis Big Bird doing here?" Only two of thechildren were willing to label the actions.Therefore, the rest were tested for compre-hension of the action names. Children wereasked to select a card from an assortment ofthree cards arranged side by side in front ofthe child in response to the question. "Canyou find a picture of someone washing?"The position ofthe correct card was counter-balanced across trials. No child persistentlymade choices of a particular character or spa-tial position; indeed, the vast majority ofchildren's choices were correct. In the rarecases (about 10%) when a child selected thewrong card, the experimenter prompted thechild to label the card correctly by re-minding the child when such an actionmight be carried out. If a child asked whata character was doing in an action for whicha label was not requested, attention was di-

3106 Child Development

FIG. 1.The five unfamiliar actions. Going clockwise from the top left, the actions depicted arebending the back, arching one's back (with back facing floor) while on hands and feet, sitting with legsapart with arms to one side, doing a push-up, doing an arabesque.

verted to another card. The cards were thenremoved, and the experiment proper began.

Test.Children were given two blocksof four trials each. Table 2 presents the ex-perimental design for one block of trials. Atrial consisted of two questions, one askingfor a familiar action and one asking for anunfamiliar action. Asking for a familiar ac-tion on each trial eliminated the possibilitythat subjects would simply develop a strat-egy of choosing the unfamiliar action. Sinceeach trial asked two questions of the child,and since there were four trials in a block oftrials, the child responded to a total of eightquestions for each block of trials. The entireexperiment then, consisted of a child's re-sponses to 16 test questionshalf about fa-miliar and half about unfamiliar actions.

The stimulus display for each trial con-sisted of the same drawings used in the fa-miliarization phase now mounted in groupsof four on a piece of 8 inch x 11 inch lami-nated construction paper as shown on Fig-ure 2. To lessen any confusion, each charac-ter was always shown in the same quadranton each sheet (e.g., Ernie always appearedin the upper left cell). The cell in which the"correct" response was found was counter-balanced across trials. In the description ofthe test trials below, familiar actions will belabeled with an "F" and unfamiliar actionswith a "U"; numeral subscripts will be usedto indicate when a different familiar or unfa-miliar action was used. When a second tokenof an unfamiliar action was used, the sub-script will contain an "a" for the original ac-tion and a "b" for the second token.

TABLE 2EXAMPLE OF EXPERIMENTAL DESIGN FOR ONE BLOCK OF TRIALS

Possible PossibleFamiliar Unfamiliar ActionActions' Actions Requested''

Trial 1 eating (Fi) daxing (UjJ Where's daxing?reading (F2) Where's eating?sleeping (F3)

Trial 2 standing (F4) unlabeled (U2) Where's standing?crying (F5) daxing (Ujj) Where's daxing?

Trial 3 sitting (Fg) lorping (U3J Where's falling?falling (F7) daxing (Ui^) Where's lorping?

Trial 4 coloring (Fg) lorping (U3J,) Where's waving?waving (F9) unlabeled (U2) Where's lorping?

" The second block of trials used a diflferent set of familiar and unfamiliar actions (seeTable 1).

* Familiar actions were requested first on half of the trials, according to four random orders.

FIG. 2.A sample stimulus sheet. Going clockwise from the top left, the actions depicted areeating, sleeping, "daxing," and reading.

3108 Child Development

Trial 1 within each set of four trialswas designed to examine whether childrenwould assume that a new term mapped toa previously unnamed action. The stimulusgroup presented to the child consisted ofthree familiar actions (Fj, Fg, and F3) andone unfamiliar action (Uu), making the like-lihood that a child could select the unfamil-iar action by chance 25%. In counterbal-anced order, the experimenter asked for onefamiliar action and for the unfamiliar action.The carrier phrases used to request familiarand unfamiliar actions were the same. Forexample, children were asked, "Can youshow me someone verbing?" or "Do you seesomeone verbing here?" Carrier phraseswere changed when a child hesitated oracted confused by the request. The child'sresponse consisted of pointing. After thechild answered the two questions for thefirst trial, the stimulus group was removedand replaced with a new group for the sec-ond trial.

For trial 2, the stimulus group consistedof two different familiar actions (F4 and F5),a new unfamiliar action (U2), and the unfa-miliar action from Trial 1 now performed bya different character (Ui^). This second char-acter performed the unfamiliar action seenon Trial 1 in exactly the same position andfacing in the same direction as the originalexemplar. Thus, if the two tokens ofthe un-familiar action from Trials 1 and 2 were su-perimposed, they would be highly similar inshape. The purpose of this trial was to seewhether children would extend a newlylearned verb to another token under the cir-cumstances that (1) they had only seen a sin-gle exemplar of it and (2) another unfamiliaraction was present.

The stimulus group for Trial 3 consistedof two different familiar actions (Fg and F7),the original unfamiliar action seen on Trial1 (Via)' 3.nd a new unfamiliar action {U^J.This trial tested whether children preferredto assign a new unfamiliar action name to anas yet unlabeled action {UsJ ^^ opposed tothe previously labeled unfamiliar action(Uja). That is, would subjects, offered an-other new action name, select the new unfa-miliar action rather than the action previ-ously labeled on Trials 1 and 2? This trialrepresented a very stringent test of whetherthe child had learned the name of the origi-nal action since the subject had been ex-posed to the original name/action combina-tion only two times. If children had in factlearned the new verb used in Trials 1 and 2,they should assume that the seeond new

verb referred to the new unnamed actionand not to the action that had already beenlabeled in Trials 1 and 2.

The stimulus group for trial 4 consistedof two different familiar actions (Fg and Fg),the unfamiliar, unlabeled action seen ontrial 2 (U2), and a second token of the unfa-miliar action seen and labeled on Trial 3(Vsh), now performed by another character.The goal of this trial was the same as thegoal of Trial 2: Would subjects extend an-other new action term, heard only once onthe previous trial, to the second token of thatunfamiliar action? Thus, the subject wasasked for the same novel action that had justbeen named on Trial 3.

To summarize, Trial 1 tested principleswhich predict that an unnamed actionshould be selected as the referent for a novelaction term. Children had to figure out towhich action the novel term mapped. Trial2 tested whether children would extend thenew term heard on Trial 1 to another token.Performance here was predicted under prin-ciples that state that children interpret novelterms as applying to categories. Trial 3 in-creased the complexity of the task by pre-senting yet another unnamed action with thepreviously labeled action seen on Trial 1.Trial 4 was also a test ofthe principles whichstate that words are extended categoricallysince children were asked to extend a sec-ond novel action term that they had heardon the previous trial.

ResultsAre there differences by sex of the sub-

jects?We conducted a two-way mixed AN-OVA with the between-subjects factor of sexand the within-subjects factor of familiarity(familiar vs. unfamiliar questions). On eachtrial (there were four trials in each of twoblocks), one question was asked about a fa-miliar action and one about an unfamiliar ac-tion for a total of eight questions about famil-iar actions and eight about unfamiliaractions. Subjects \vere assigned two scoresof 0-8, one for the number of questions an-swered correctly about familiar actions andone for the number of questions answeredaccording to predictions about unfamiliar ac-tions. Because neither the main effect of sexnor the interaction approached significance,subsequent analyses were pooled over sex.

Do children select the correct, familiaractions?The only significant effect foundin the above analysis was question type:Questions asked about familiar actions weresignificantly easier for children to answer

Golinkoff et al. 3109than questions about unfamiliar actions, F(l, type. A superscript of " 1 " means that the " 1 "23) = 27.70, p < .001. The mean number was scored on the first trial and not on thecorrect for familiar actions was 7.75 (SD = second; the absence of a superscript means.45); for unfamiliar actions, the mean num- that the " 1 " was achieved on only the sec-ber answered according to predictions was ond trial. Finally, a score of "2" means that5.58 (SD = 1.31). the child selected the appropriate unfamiliar

Do children select a novel action as the '^^ t^ "" " ^ * ^^^' "^ ^ particular type.referent for a novel action term?Table 3 As can be seen on Table 3, on trial 1shows the proportion of children's seleetions where there were three familiar actions andin response to requests using a novel action one unfamiliar action, children selected theword. Percentages were computed out of 24 novel action 83% of the time. To evaluatepossible responses since on each trial (14), whether, out of a group of 12, children se-the 12 subjects were asked two questions of" lected the novel action at a rate greater thanthat type. Underlined scores represent the that expected by chance, second-order bino-responses that were predicted by the rele- mial probabilities were calculated in the fol-vant lexical principles. When a cell has no lowing way. First, the probability of a singleentry, the response could not occur on that child obtaining a score of 0,1, or 2 was calcu-trial. Table 4 presents the data for each sub- lated. Thus, since the chance probability ofject by trial type. A score of "0" means that a child selecting the unfamiliar item on aon the two occasions when the child re- single trial is .25 (given four choices), theceived a trial of that type, the child did not chance probability of a child selecting theselect the appropriate unfamiliar action on unfamiliar item on both trials of the sameeither trial. Similarly a score of " 1 " means type and scoring 2 out of 2 is (.25) x (.25) =that the child selected the appropriate unfa- .0625. The chance probability of scoring 1miliar action on one ofthe two trials of that out of 2 is (.25) x (.75) X 2 = .3750, and the

TABLE 3PERCENTAGES OF CHILDREN'S ACTION SELECTIONS (Standard Deviations in Parentheses) IN RESPONSE

TO REQUESTS USING A NOVEL VERB (Experiment 1) OR TO REQUESTS WITHOUT A NOVEL VERB(Experiment 2)

ACTION TYPE

First Token of Second Token of DistractorTRIAL AND EXPERIMENT Novel First Novel Novel Second Novel Familiar Novel1: Test of selection of

novel action:"Novel verb 83(25) 17No word 29 (79) 71

2: Test of extension ofjust-learned novelverb:''

Novel verb 71 (40) 17 12No word 25 (52) 42 33

3: Test of whetherjust-learned novelverb will novelverb:*

Novel verb 25 63 (38) 12No word 16 50 (60) 34

4: Test of extensionof second, just-learned novelverb:*"

Novel verb 67 (39) 17 20No word 29 (67) 42 29

NOTE.Underlined numbers were the hypothesized outcomes; cells with no entry could not occur on that particu-lar trial.

" These trials pertain to the principles of mutual exclusivity and novel namenameless category.^ These trials pertain to the taxonomic assumption and to lie categorical scope principle.

3110 Child DevelopmentTABLE 4

INDIVIDUAL CHILDREN'S RESPONSES IN EXPERIMENTS 1 AND 2 BY TRIAL TYPE

AGE SEX

TRIAL TYPE

Experiment 1: Asked fornovel verb by name:

35-24 F 1' 2 2 1'35-7 F 1' 1^ 1' 135-24 F 1' 2 2 234-2 F 2 2 1 232-14 F 1 2 1^ 1'34-2 F 2 2 0 232-24 M 2 0 2 1'33-24 M 2 1' 1' 1'35-29 M 2 1 1 032-24 M 2 0 2 235-12 M 2 2 0 032-18 M 2 2 2 2

Experiment 2: Asked topick one:

33-05 F 0 0 1' 035-18 F 1^ 1 1 035-23 F 0 1 1 234-28 F 0 0 0 035-2 F 1 0 1 1'36-10 F 0 1' 1 1'35-23 M O 1' 2 035-34 M 0 0 1 135-22 M 0 0 1 034-17 M 2 1' 2 1^34-20 M 1' 1 0 134-8 M 2 0 1^ 0

" A score of 0 means that ou neither of the two trials of that type did the child select theunknown action; a score of 2 means that the unknown action was selected on both trials. Ascore of 1 with the superscript 1 means the child selected the unknown action only on the firsttrial of that type; the absence of a superscript means that the child selected the unknown actiononly on the second trial of that type.

chance probability of scoring 0 ont of 2 is(.75) X (.75) = .5625. Second, expected fre-quencies for scores of 2 ,1 , and 0 were calcu-lated by multiplying each of the chanceprobabilities by the number of subjects (12);expected frequencies are .75, 4.5, and 6.75,respectively. Third, a second-order applica-tion of the binomial distribution was used toobtain the probability that as many as eightor more children in a sample of 12 wouldobtain scores of 2 out of 2 by chance. Thisprobability (the sum ofthe probabilities thatexactly 8, exactly 9, exactly 10, exactly 11,and exactly 12 children would score 2) isvery small (.000000091). In the group of 12children there are thus, significantly morechildren responding as predicted to the twotrial 1 questions than would be expected todo so by chance (i.e., given a probability of.25 of choosing in accordance with the pre-

dictions in response to each question aboutan unfamiliar novel action word).

Another way to look at these scores is toexamine the probability that 12 children (theentire sample) would score either 1 or 2 outof 2 by chance. The probability of scoring 1or 2 is the probability of scoring 1 plus theprobability of scoring 2 (.375 -I- .0625 =.4375). By the second-order binomial, thelikelihood that all 12 children in a group of12 would score either 1 or 2 by chance isvery small (.00004917).

children interpret a second new ac-tion word as applying to yet another un-named action?On trial 3, a previously la-beled novel action and a novel, unnamedaction were presented with two familiar ac-tions. The overall rate of selection ofthe un-named novel action was 63% and five of 12

Golinkoff etal. 3111children obtained a score of 2, By the sec-ond-order binomial, the likelihood that fiveor more children in a group of 12 would se-lect the second unfamiliar object by chanceis .000519. Also, five of the 12 children ob-tained a score of 1. We next evaluated thelikelihood that 10 children out of a group ofsize 12 would score either 1 or 2 out of 2 bychance. This probability is also small(.005365). When children erred on trial 3,they selected the just-named action 25% ofthe time and a familiar action 12% of thetime.

Will children extend a just-learnednovel term to another novel action of thesame type?In trial 2, when presented withanother token of the action that they had justselected as the referent for the novel verbin trial 1, children selected the token of thejust-named action 71% ofthe time. Further-more, seven children achieved a score of 2.The likelihood that seven or more childrenwould achieve a score of 2 by chance aloneis .000002227. The remaining five childrenachieved a score of 1 (3) or 0 (2). The proba-bility that 10 children would score either 1or 2 out of 2 by chance is also small(.005365).

Extension was seen also in trial 4 where67% of children's responses were selectionsof another token of the action they had seenon the prior trial. Here five children re-ceived scores of 2, a number significantlygreater than would be expected by chancein a group of size 12 (binomial p = .000519).Of the remaining children, two obtained ascore of 0, and five children obtained a scoreof 1. The probability that 10 or more chil-dren in a group of size 12 would receive ascore of 1 or 2 out of 2 by chance is small(.005365).

Were children consistent in their indi-vidual response patterns?Two types ofanalyses were performed. First, each subjectwas given one point for performance ac-cording to predictions on each of four trials(across two blocks of trials) that tested thesame principle. Thus, trials 1 and 3 testedfor whether a novel name would be mappedto an unnamed action, and trials 2 and 4tested for extension. On trials 1 and 3, eightout of 12 subjects received scores of 3 (fivesubjects), or 4 (three subjects), indicatingthat the majority of subjects mapped the un-known term to the unnamed action on 75%or more of their trials. This frequency (eightor more in a group of 12) can be interpretedin light ofthe binomial: The probabilities of

an individual child's scoring 3 and 4 out of4 by chance are .0469 and .0039, respec-tively. By a second-order application of thebinomial expansion, the chance probabilitythat eight or more children out of a group of12 will score 3 or 4 is .00000002. Thus, thelikelihood that two-thirds of the subjectscould have achieved scores of 3 or 4 bychance is quite slim. Further, this occurreddespite the greater difficulty of trial 3, wherea just-named action was also present. It ap-pears that the majority of the children wereindeed motivated to affix novel names to un-named actions. Each of the four childrenwho did not attain scores of 3 or 4 obtaineda score of 2.

On trials 2 and 4, where another un-named novel action was presented in addi-tion to a new token of the action seen onthe prior trial, six out of 12 children reliablyextended a just-learned novel action term toa new token on at least 75% of their trials,receiving scores of 3 (two subjects) or 4 (foursubjects). As above, the likelihood by the bi-nomial that a single child could answer 3 or4 out of 4 trials according to predictions bychance is (.0469 -I- .0039) = .0508. Thechance probability that six or more childrenout of 12 will score 3 or 4 by chance is.00012. This result suggests that childrenrecognize that wordsin this case actiontermsare applied to categories of action.Ofthe six children who did not attain scoresof 3 or 4, four received scores of 2.

The second intrasubject analysis askedwhether children were consistent from trialto trial in their interpretation of the sameterm. In other words, if a child selected thenovel action on trial 1, did the child thenselect the token of that action on trial 2, andthen a different novel action on trial 3,thereby acting as if the first unknown actionalready had a name? For this analysis, achild needed to answer the first three trialsin a row according to predictions to receivea point, an extremely conservative criterionthat would be expected to be met by chanceonly with the probability of .0156. Subjectscould receive a score of 0, 1, or 2, commen-surate with the number of blocks of trials inwhich three questions in a row were an-swered consistently. Thus, a child who ob-tained a score of 0 failed to answer three ina row consistently on either of the trialblocks. A score of " 1 " means that the childanswered three in a row consistently on onlyone of the two blocks of trials; "2" meansthat the child answered three in a row con-sistently on both blocks.

3112 Child DevelopmentResults again suggest that children are

not operating randomly. The likelihood of asingle individual obtaining a score of either1 or 2 is .031 (the sum of the probability ofgetting 1 and the probability of getting 2).Since seven out of 12 subjects receivedscores of 1 (six subjects) or 2 (one subject),two-thirds ofthe children, seen as individu-als, appear to be treating a just-learned termwith consistency from trial to trial. The databecome more impressive, however, when,using the second-order binomial, the chancelikelihood that seven or more children in agroup of 12 would receive a score of 1 or 2is calculated. This probability is very small(.000000018).

Summary.Results from a task whichgave young children considerable opportu-nity to respond inconsistently suggest thatchildren readily fast map novel verbs tonovel actions. From a single exposure to anovel action word in a nonostensive task,two-thirds ofthe sample appeared to assumethat a new word applies to an action whichdoes not already have a name (trial 1) andthat the novel verb can be immediately ex-tended (trial 2 and trial 4). However, on trial3, where a second novel verb was intro-duced, children seemed somewhat con-fused. This is not surprising in that on thistrial, children were asked to take a secondnovel term to apply to an unnamed action inthe presence of a novel action that had beennamed only twice on the prior trials.

Thus, the data from the trials consideredsingly (that is, trial 1, 2, 3, or 4) present amixed picture of young children's ability tofast map and extend novel verbs. When thedata from the trials that tested the same prin-ciples were taken together, however, the re-sults are more supportive ofthe predictions.Children were consistent in their perfor-mance on the trials that tested for fast map-ping as well as on those that tested for exten-sion of a novel verb. On the fast mappingtrials (1 and 3), there is only a very smalllikelihood that eight children out of a sampleof 12 could obtain scores of 3 or 4 out of4 correct by chance. Similarly, on the trialswhich tested for the extension of a novelverb to another exemplar (2 and 4), there isa very small likelihood that, by chance, sixchildren out of 12 could consistently extenda novel verb to a second exemplar. More im-pressively, eight of the 12 children treatedthe novel verbs as though they mapped tothe same action on the first three consecu-tive trials in either or both blocks of trials.This latter finding means that children fast

mapped (trial 1) and extended the first novelverb (trial 2) and then, when they heard an-other novel verb on trial 3, did not select theaction that had just been named. That is,they correctly treated the new action termas though it applied to the newest unnamedaction.

Before these results can be fully evalu-ated, however, a control experiment isneeded to rule out the possibility that chil-dren would have responded similarly with-out being offered a novel label. If childrenare behaving according to certain lexicalprinciples they should only do so in thepresence of a label. Without a label, theirbehavior should look random.

Experiment 2This control study was designed to de-

termine whether the choice of novel actionsin Experiment 1 was controlled by the useof a novel action label. The results of theprior experiment may be attributable to arti-fact (perhaps a preference for novelty, seeMerriman, 1991) and not to the operation oflexical principles if children behave thesame way when asked to select an actionwithout being offered a name. This experi-ment exactly duplicates the design of themain experiment. Thus, children were stillasked for a familiar action on each trial andstill had to choose from among familiar andunfamiliar actions (see Table 2). However,wherever a novel action was requested byname in Experiment 1, children here wereasked, "Can you show me someone doingsomething?"

SubjectsTwelve children (six boys and six girls)

(mean age = 35 months, 2 days; range =33,05-36,10) were tested. All were Cauca-sian and native speakers of English withmultiword speech.Procedure

The same two-part procedure was usedas in the Experiment 1, with the exceptionthat no novel action was requested by name.Results

Responses to familiar actions.When afamiliar action was requested by name, chil-dren were correct on 100% of die trials.

Did children select the same unknownaction from Experiment 1 when the requestdid not include a novel term?On trial 1,when requests did not include a novel term,children selected the novel action only 29%

of the time (see Table 3). Recall that, as dis-played on Table 4, a child could obtain ascore of 0 (not selecting the novel action oneither ofthe two trials of that type); a scoreof 1 (selecting the novel action on one ofthe two trials of that type); or a score of 2(selecting the novel action on both trials ofthat type). The majority of responses on trial1 were 0 (seven children); the rest werescores of 2 (two children) and 1 (three chil-dren). To evaluate whether the number ofchildren (out of 12 tested) selecting thenovel action was significantly greater thanthe number expected by chance, second-order binomial probabilities were used tocompute change levels at the high tail of thedistribution as in Experiment 1. However,since there were so few scores of 2 in Exper-iment 2, these probabilities were only com-puted by considering whether the numberof children who scored either 1 or 2 out of 2in a group of size 12 exceeded chance. First,the probability of a single child obtaininga score of 0, 1, or 2 was calculated. Theseprobabilities are .5625 for a score of 0, .375for a score of 1, and .0625 for a score of 2.Second, expected frequencies for scores of0, 1, or 2 were calculated by multiplyingeach ofthe chance probabilities by the num-ber of subjects (12); expected frequenciesare 6.75, 4.5, and .75, respectively. Third, asecond-order application of the binomialdistribution was used to obtain the probabil-ity that five or more children in a sample of12 would obtain scores of 1 or 2 out of 2by chance. Since this probability is high(.22619), it appears that children did not se-lect the novel action with a frequency sig-nificantly greater than that expected bychance when it was not requested by nameon trial 1.

Did children select the second novel ac-tion on trial 3, despite the fact that it wasnot requested by name?On trial 3, chil-dren's selection ofthe novel action they hadnot already seen on trial 1 was at a rate of50% (see Table 3). As Table 4 shows, twochildren scored 2, two children scored 0, andthe eight remaining children scored 1. Theprobability that 10 or more children out of12 could obtain a score of 1 or 2 by chanceis small (.005365). This significant resultsuggests that the novelty of the second un-named novel action may have played a rolein children's responses on this trial. None-theless, comparisons of children's perfor-mance across the two experiments (see be-low) suggest that even on trial 3, arguablythe most difficult trial, children in the main

Golinkoff et aL 3113experiment responded systematically whilechildren in this experiment did not.

Did children select the second token ofthe novel action on trials 2 and 4Chil-dren selected the second token of the novelactions at the rates of 25% on trial 2 and 29%on trial 4. On trial 2, six children obtained ascore of 0, and six children obtained a scoreof 1. Trial 4 results were quite similar: sixchildren scored 0, five children scored 1, andone child scored a 2. The likelihood that ina group of 12 children six children wouldscore either 1 or 2 out of 2 by chance (appli-cable to both trials 2 and 4) is .205246.Therefore, when the request was made with-out a label, the numbers of children out of agroup of 12 who selected the second tokenofthe first novel action on trials 2 and 4 werenot significantly greater than would be ex-pected by chance.

Were children consistent in their indi-vidual response patterns?The same twoanalyses that were performed in Experiment1 were performed here. First, each subjectwas given one point for performance ac-cording to predictions on each of four trialsthat tested the same principle. For trials 1and 3 combined (fast mapping trials), thereare two children out of 12 who score either 3or 4. Three children scored a 2, six childrenscored a 1, and one child scored a 0. Byapplying the second-order binomial vfe findthat the probability that two or more chil-dren in a group of 12 would score either 3or 4 is .1215. Thus, since this probability isnot low enough to justify an inference that asignificantly greater number of children re-ceived scores of 3 or 4 than would be ex-pected to do so by chance, we conclude thatchildren do not select the unnamed actionwhen it is not requested by name.

On trials 2 and 4 (extension trials), onlyone child scored 3; no children scored4. Three children scored 2, four childrenscored 1, and four children scored 0. By thesecond-order binomial the probability that 1or more out of 12 children would score ei-ther 3 or 4 is .4649. Thus, when the secondtoken of the unnamed action was not re-quested by name, the children did not selectit more than would be expected by chancein a group of this size (12).

The second intrasubject analysis askedwhether children were consistent from trialto trial in their responses as they had beenwhen a verb was used in the main experi-ment. That is, did children pick the sameaction or a token of it on trials 1 and 2, re-

3114 Child Developmentspectively, and then select the newest novelaction on trial 3? With two blocks of trials,subjects could therefore receive a score of 0,1, or 2. No child received a score of 2, onechild scored 1, and the remaining 11 chil-dren scored 0. By the second-order bino-mial, the probability that 1 or more out of 12children would obtain a score of 1 or 2 is.3147. Thus, children do not appear to beoperating consistently when the unnamedaction was not requested by name.

Comparison of children's responses tothe unfamiliar actions across the two exper-iments.Fisher's exact test was performedto compare across experiments the propor-tions of children with scores on pairs of trialsofthe same type. Thus, the number of chil-dren in each experiment obtaining scores of3 and 4 versus 0, 1 and 2 on trials 1 and 3and on trials 2 and 4 were assembled into 2X 2 tables as shown in Table 5. By theFisher's exact test, the proportion of chil-dren with scores of 3 and 4 for both sets oftrials (1 and 3; 2 and 4) differed significantlybetween the two experiments (p < .05 forboth sets).

Another way in which the Fisher's wasapplied was to compare the number of chil-dren in each experiment who gave a consis-tent interpretation across the first three trialsof a block of trials. Thus, children could haveobtained scores of 2, 1, or 0, commensuratewith the number of trial blocks in whichthey selected the novel action (trial 1), a to-ken of it (trial 2), and then a different actionon trial 3. In Experiment 1, seven childrenreceived scores of 1 or 2, and five childrenreceived scores of 0; in Experiment 2, onechild received a score of 1 and the rest re-ceived scores of 0. The Fisher's test revealedthat a significantly higher proportion of chil-dren received scores of 1 or 2 in the main

experiment than in the control experiment(p < .005).

Summary.The control experimentwas designed to see if children would re-spond similarly to the children in the mainexperiment despite the fact that actions inthe control study were not requested byname. If children had responded similarly,some explanation other than lexical princi-ples, such as a novelty preference, wouldhave been responsible for the results in themain experiment. Clearly, children in thecontrol experiment, considered as a group,for the most part did not obtain a signifi-cantly larger number of high scores thanwould be expected by chance in a group ofsize 12. The only exception to the interpreta-tion that children in the control experimentacted randomly comes from trial 3 when itis analyzed singly. Here, significantly morechildren than would be expected by chancein a group with 12 members selected whatwould have been the newest unnamed ac-tion in the main experiment (i.e., the actionthat should be selected in accordance withpredictions of the lexical principles). How-ever, this result is mitigated when trials 1and 3 are analyzed together.

Another way in which the children inthe control experiment were shown to per-form differently than children in the mainexperiment was in the comparisons acrossexperiments, both by group and by individ-ual subjects. The children in the control ex-periment were much less likely than thosein the main experiment to select the appro-priate unfamiliar action in response to a re-quest without a name. These findings sug-gest that children in Experiment 1 engagedin true (if limited) word learning and re-sponded in a consistent and principled way

TABLE 5OBSEHVED FBEQUENCIES OF HIGH (3-4) AND LOW (0-2) SCORES

ON FAST MAPPING THIALS (Trials 1 and 3) AND EXTENSIONTRIALS (Trials 2 and 4) IN THE MAIN AND

CONTROL EXPEWMENTS

TRIALS

1 and 3

2 and 4

EXPERIMENT

Scores

3-40-2

Fisher exact p < .052-40-2

Fisher exact p < .05

Main

84

66

Control

210

111

Golinkoff et al. 3115when offered a label in the presence of anunfamiliar action.

General DiscussionThe purpose of this research was to see

whether children could fast map and extendnovel motion verbs; prior research alreadyshowed that children could fast map and ex-tend novel nouns (e.g., Golinkoff, Hirsh-Pasek, et al., 1992) and adjectives (e.g.,Heibeck & Markman, 1987). In this experi-ment, children were presented with fournew action names in a single session. Tosucceed, they had to be able to fast map anew term to a novel action and then to ex-tend this term to the same action performedby a different agent. Together with Merri-man et al. (1993, 1995) and Rice and Wood-small (1988), this is one of the few studiesthat have tested for verb learning in a nonos-tensive context. That is, the experimenternever made an explicit link between the ac-tions and the action names; it was up to thechild to infer to which action the novel verbmapped. Nonetheless, we hypothesized thatchildren would have litde difficulty withthis task since by 34 months of age most chil-dren have acquired many verbs. Experiment2 was a control experiment which presentedchildren with the task of selecting a novelaction in the absence of a novel label. If, inthe main experiment, children's responsesare guided by something other than lexicalprinciples (say, a preference for novelty),then performance in the control experimentshould continue to exceed chance levels.First Mapping of Verb Meaning IsPredicted by Lexical Principles

Experiment 1 tested whether childrenwould select a previously unnamed actionin response to requests using a novel termin the presence of both named and unnamedactions. This question was relevant to princi-ples such as mutual exclusivity (Markman,1987), contrast (Clark, 1983), and N3C (Go-linkoff et al., 1994), all of which attempt toaccount for the rapid learning of novelnames. The principle of N3C states thatnovel terms map to unnamed categories. Un-like the principles of contrast (Clark, 1983,1987) or mutual exclusivity (Markman, 1987,1989) which make a number of other as-sumptions, N3C predicts only tha.t childrenwill attach novel labels to unlabeled catego-ries. Results from Experiment 1 indicatedthat N3C applies to novel action names aswell; children selected the novel action tobe the referent for the novel action name onboth trials 1 and 3, although to a lesser de-

gree on trial 3. On that trial, children hadto choose between two known actions (e.g.,eating), one action that had just been named,and a novel action that had not yet beennamed. According to the principles that statethat a novel term should be taken as thename for a referent without a name, childrenshould have reliably chosen the unnamednovel action. This trial, however, representsa highly rigorous test of these principlessince children had only two exposures to thefirst novel verb/action combination. Anotherpossible explanation for the weaker resultson this trial may have to do with the greaterpotential for verbs to overlap in referencemore than do nouns, as discussed above.

Performance on trial 3 looks less ambig-uous in two other analyses. First, when thetwo trials (1 and 3) that tested for fast map-ping are taken together, the results clearlyindicateespecially in comparison to theperformance of the children in the controlstudythat most children did fast map thenovel term to the respective novel actionson these trials. Second, the comparison be-tween the number of children in the mainand control experiment who responded ac-cording to the word learning principles onthree trials in a row within a block alsoshows a clear pattern of adherence to the fastmapping principles. Children who re-sponded consistently selected the first un-named action on trial 1, the second token ofthat action on trial 2, and finally, the secondunnamed action on trial 3. Two-thirds (eight)of the children in the main experiment re-sponded with this degree of consistency; nochild in the control experiment did. No wordlearning experiments of which we are awarehave held children to this high a standard toprovide evidence for the operation of lexicalprinciples.

N3C (or a principle like it) seems to beuseful throughout the life span for the rapidacquisition of object names (Golinkoff, Diz-noff, et al., 1992), property names (e.g.,Heibeck & Markman, 1987), and action la-bels as well (the present study). To use N3C,or for that matter, any principle that statesthat new terms are mapped to unnameditems, children must have the followingknowledge: First, they must be capable ofinterpreting others' communications as ref-erential acts. Golinkoff et al. (1994) calledthis the principle of "reference" and foundevidence of its use at the beginning of thesecond year of life. By the time children are2 years old, the ability to detect a referentialact is quite sophisticated. Tomasello and

3116 Child DevelopmentBarton (1994), inspired by Baldwin (1993),showed that 2-year-old children will notmap an action label to the first action theysee if that action was apparently performedaccidentally. Rather, children will wait tomap a novel verb to the second novel action,apparently performed intentionally. Second,children must recognize that the word theyare hearing is novel. This implies that theymust already have a vocabulary of some sizeto compare the novel word against. In fact,Mervis and Bertrand (1994) showed thatN3C (as it applied to the learning of novelobject labels) was only in place after chil-dren showed evidence of having undergonea vocabulary spurt. Third, children musthave some way to distinguish whether thenovel word is a noun, or a verb, or an adjec-tive. To do this children can rely on one ormore redundant linguistic cues (Goldfield,1993; Golinkoff, Diznoff, Yasik, & Hirsh-Pasek, 1992) such as the presence or absenceof articles, final placement in the sentence,and prosodic cues such as stress, as well asnonlinguistic cues.

Interestingly, our findings on the fastmapping of verbs parallel those of Rice andWoodsmall (1988) and Merriman et al. (1993,1995) in finding verb learning a bit more dif-ficult than noun learning. The presentstudy's design is identical to the design of aprior study on nouns (Golinkoff, Hirsh-Pasek, et al., 1992) except for the use of onlytwo blocks of trials instead of three. Al-though the rate of selection ofthe unfamiliaritem when requested by a novel term wasvirtually identical (72% for nouns, 70% forverbs), the children in the verb study were,on average, 5 months older. Especially withonly two blocks of trials, they might havebeen expected to perform at a higher levelthan the younger children in the noun study.There are two possible explanations for thisdifference, one of which can be ruled out.First, it is not the case that children failedto interpret the stimulus drawings as actionsand were just confused. The stimulus evalu-ation study (see Experiment 1) showed thatchildren interpreted these stimuli as por-traying actions, as they probably do in abook-reading situation.

Second, the fast mapping of verb mean-ing may require more exposures than the fastmapping of noun meaning. This makessense on the rationale that many subeventsmay occur when a new action is performed,and it may take a few exposures to the action/verb combination to tease out vi'hich portion

of the novel action is "core" and which ismore peripheral and a function of the indi-vidual actor. Perhaps until children are fairlycertain about what exactly is being labeled,they are a bit reluctant to use and extend anew verb. Support for this view comes fromForbes and Farrar (1993), who reported that3-year-olds generalized novel verbs signifi-cantly less than 7-year-olds or adults. Fur-thermore, if we assume that children do notencounter much in the way of negative evi-dence (e.g.. Pinker, 1989), it is quite adap-tive for children to be somewhat conserva-tive in their fast mapping and extension ofnovel verbs.Extension of Novel Verbs is Predicted byLexical Principles

Experiment 1 also tested whether chil-dren will readily extend a newly learned ac-tion term after a single exposure to anotherinstance of the action performed by anotheragent. Much ofthe power words have wouldbe lost if childrenlike Victor, the wild boyof Aveyron (Itard, 1932)expected eachword to label a single, unique exemplar.Children fortunately do not hold that expec-tation and extend just-learned nouns (Golin-koff, Hirsh-Pasek, et al., 1992), adjectives(Heibeck & Markman, 1987), and verbs (thepresent study) after only a single exposure tothem. Three principles (the noun-categorybias, the taxonomic assumption, and categor-ical scope) have been proposed to accountfor extension.

Golinkoff et al. (1994) and Golinkoff,Shuff-Bailey, et al. (1995) provided evidencethat the categorical scope principle bettercaptures the basis for object word extensionsince it specifies that novel words are to beextended on the basis of basic level categorymembership. The jury is still out, however,on the issue of whether verb meanings, likenoun meanings, are organized in a subordi-nate-basic-superordinate hierarchy (Rosch,Mervis, Gray, Johnson, & Boyes-Braem,1976). For example, Kegl (1989) made thecase that action terms can be organized hier-archically while the word of Huttenlocherand Lui (1979) and Miller and Johnson-Laird (1976) indicates that verb organizationis flatter and less hierarchical than the orga-nization of nouns. Whether or not verbs turnout to be organized hierarchically the waynouns are, it is still the case that childrenmust construct a principle to guide verb ex-tension to other actions in the same category.Golinkoff, Hirsh-Pasek, et al. (1995) assertedthat "categorical scope for verbs states that

Golinkoff et al. 3117action labels are extended to other actionsthat appear to require the same semanticcomponents as the original action" (p. 208)."Semantic component" refers to elements ofmeaning like "theme," "path," and "man-ner" that the world's languages either con-fiate, within the meaning of their verbs ormark outside the verb with a separate form(Frawley, 1992). Although considerablymore research is needed, it appears that chil-dren can detect these semantic componentsthrough their observation of events (Pinker,1989) as well as from the syntactic frameswhich surround the verb (Gleitman, 1990;Hirsh-Pasek & Golinkoff, in press). When er-rors of verb extension occur, categoricalscope predicts that they will likely be a re-sult of the child failing to note one of thesemantic components conflated in the verb.For example, Choi and Bowerman (1992) re-port that a Korean child used the verb ollita(cause to go up) incorrectly, to mean "put-ting [something] away regardless of direc-tionality" (that is, up or down) because thischild failed to realize that the element ofpath was included in the verb.

This study, which tested for verb exten-sion by showing a new agent performing theaction in exactly the same way, may be aneasier situation than children typically en-counter. In the real world, even when thesame agent performs the same action, the ac-tion is never done in exactly the same way.In addition, the location of the action maychange, as well as the instrument used toperform it, or the manner in which it is per-formed. Further, all the novel actions de-picted were intransitive. Fast mapping re-sults from Merriman et al. (1995) with2-year-olds (but not 4-year-olds, see Merri-man et al., 1993) suggest that intransitive ac-tions without an object may be easier to fastmap (and perhaps, to extend) than transitiveactions. It would be interesting to explorewhether the fast mapping and extension ob-served in this study would occur at the samerate if transitive actions were tested.Conclusions

The present study is important becauseit is among the first to show that fast mappingand rapid extension occur with verbs. Fur-ther, children's word-learning behavior wasin accordance with the prediction of princi-ples that have been demonstrated to hold forobject word learning. The availability ofsuch principles across form classes may helpexplain word acquisition phenomena likethe vocabulary spurt.

ReferencesAu, T. K., Dapretto, M., & Song, Y-K. (in press).

Input vs. constraints: Early word acquisitionin Korean and English. Journal of Memoryand Language.

Au, T. K., & Markman, E. M. (1987). Acquiringword meanings via linguistic contrast. Cogni-tive Development, 2, 217-236.

Baldwin, D. A. (1993). Infants' ability to consultthe speaker for clues to word reference. Jour-nal of Child Language, 20, 377-394.

Bates, E., Bretherton, I., & Snyder, L. (1988).From first words to grammar: Individual dif-ferences and dissociable mechanisms. Cam-bridge: Cambridge University Press.

Behrend, D. A. (1990). The development of verbconcepts: Children's use of verbs to label fa-miliar and novel events. Child Development,61, 681-696.

Behrend, D. A., Harris, L. L., & Cartwright, K. B.(1995). Morphological cues to verb meaning:Verb inflections and the initial mapping ofverb meanings. Journal of Child Language,22, 89-106.

Bloom, L., & Lahey, M. (1978). Language devel-opment and language disorders. New York:Wiley.

Bloom, L., Lightbown, P., & Hood, L. (1975).Structure and variation in child language.Monographs of the Society for Research inChild Development, 40(2, Serial No. 160).

Bloom, L., Tinker, E., & Margulis, C. (1993). Thewords children learn. Cognitive Develop-ment, 8, 431-450.

Carey, S., & Bardett, E. (1978). Acquiring a singlenew word. Papers and Reports on Child Lan-guage Development, 15, 17-29.

Choi, S., & Bowerman, M. (1992). Learning to ex-press motion events in English and Korean:The influence of language specific lexicaliza-tion patterns. In B. Levin & S. Pinker (Eds.),Lexical and conceptual semantics (pp. 83-121). Cambridge, MA: Blackwell.

Clark, E. V. (1983). Convention and contrast inacquiring the lexicon. In T. B. Seiler & W.Wannenmacher (Eds.), Concept developmentand the development of word meaning (pp.67-89). Berlin: Springer-Verlag.

Clark, E. V. (1987). The principle of contrast: Aconstraint on language acquisition. In B. Mac-Whinney (Ed.), Mechanisms of language ac-quisition (pp. 1-34). Hillsdale, NJ: Erlbaum.

Dockrell, J. E. (1981). The child's acquisition ofunfamiliar words: An experimental study.Unpublished doctoral dissertation. Univer-sity of Stirling, Scotland.

Dockrell, J. E., & Campbell, R. (1986). Lexicalacquisition strategies in the preschool child.

3118 Child DevelopmentIn S. A. Kuczaj II & M. D. Barrett (Eds.), Thedevelopment of word meaning (pp. 121-154).New York: Springer-Verlag.

Droml, E. (1987). Early lexical development.Cambridge: Cambridge University Press.

Forbes, J. N., & Farrar, M. J. (1993). Children'sinitial assumptions about the meaning ofnovel motion verbs: Biased and conservative?Cognitive Development, 8, 273-290.

Frawley, W. (1992). Linguistic semantics. Hills-dale, NJ: Erlbaum.

Gathercole, V. C. (1989). Contrast: A semanticconstraint. Journal of Child Language, 16,685-702.

Centner, D. (1983). Why nouns are learned beforeverbs: Linguistic relativity vs. natural parti-tioning. In S. A. Kuczaj (Ed.), Language devel-opment: Language, culture, and cognition(pp. 301-325). Hillsdale, NJ: Erlbaum.

Gentner, D. (1988, October). Cognitive and lin-guistic determinism: Object reference and re-lational reference. Paper presented to BostonChild Language Meetings, Boston.

Civon, T. (1973). The time-axis phenomenon.Language, 49, 890-925.

Gleitman, L. (1990). The structural sources of verbmeanings. Language Acquisition, 1, 3-55.

Goldfield, B. A. (1993). Maternal input and thechild's acquisition of nouns and verbs. Jour-nal of Child Language, 20, 85-100.

Goldin-Meadow, S., Seligman, M., & Gelman, R.(1976). Language in the two-year-old. Cogni-tion, 4, 189-202.

Colinkoff, R. M. (1975). Semantic development ininfants: The concepts of agent and recipient.Merrill-Palmer Quarterly, 21, 181-193.

GolinkofF, R. M., DiznofF, J., Yasik, A., & Hirsh-Pasek, K. (1992, May). How children identifynouns vs. verbs. International Conference onInfant Studies, Miami.

Golinkoff, R. M., Hirsh-Pasek, K., Bailey, L. M.,& Wenger, R. N. (1992). Young children andadults use lexical principles to learn newnouns. Developm,ental Psychology, 28, 99108.

Golinkoff, R. M., Hirsh-Pasek, K., Cauley, K. M.,& Gordon, L. (1987). The eyes have it: Lexicaland syntactic comprehension in a new para-digm. Journal of Child Language, 14, 23-46.

GolinkofT, R. M., Hirsh-Pasek, K., Mervis, G. B.,Frawley, W., & Parillo, M. (1995). Lexicalprinciples can be extended to the acquisitionof verbs. In M. Tomasello & W. Merriman(Eds.), Beyond names for things: Young chil-dren's acquisition of verbs (pp. 185-223).Hillsdale, NJ: Erlbaum.

Golinkoff, R. M., & Kerr, J. L. (1978). Infants' per-ceptions of semantically-defined action rolechanges in filmed events. Merrill-Palm,erQuarterly, 24, 53-61.

Golinkoff, R. M., Mervis, C. B., & Hirsh-Pasek,K. (1994). Early object labels: The case for adevelopmental lexical principles framework.Journal of Child Language, 21, 125-155.

Golinkoff, R. M., Shuff-Bailey, M., Olguin, K., &Ruan, W. (1995). Young childeren extendnovel words at the basic level: Evidence forthe principle of categorical scope. Develop-mental Psychology, 31, 494-507.

Gopnik, A., & Choi, S. (1995). Names, relationalwords, and cognitive development in Englishand Korean speakers: Nouns are not alwayslearned before verbs. In M. Tomasello &W. E. Merriman (Eds.), Beyond names forthings (pp. 63-80). Hillsdale, NJ: Erlbaum.

Heibeck, T., & Markman, E. M. (1987). Wordlearning in children: An examination of fastmapping. Child Development, 58, 1021-1034.

Hirsh-Pasek, K., & Golinkoff, R. M. (in press). Theorigins of grammar: Evidence from early lan-guage comprehension. Cambridge, MA: MITPress.

Hutchinson, J. (1986). Children's sensitivity to thecontrastive use of object category terms. Pa-pers and Reports on Child Development, 25,49-56.

Huttenlocher, J., & Lui, F. (1979). The semanticorganization of some simple nouns and verbs.Journal of Verbal Learning and Verbal Be-havior, 18, 141-162.

Itard, J. (1932). The wild boy of Aveyron. NewYork: Century.

Kegl, J. (1989). Taxonomic structure and objectdeletion in the English verbal system. Paperpresented at Eastern States Conference onLinguistics, Newark, DE.

Markman, E. M. (1987). How children constrainthe possible meanings of words. In U. Neisser(Ed.), Concepts and conceptual development:Ecological and intellectual factors in catego-rization (pp. 255-287). Cambridge: Cam-bridge University Press.

Markman, E. M. (1989). Categorization and nam-ing in children. Cambridge, MA: MIT Press.

Markman, E. M., & Hutchinson, J. E. (1984). Chil-dren's sensitivity to constraints on wordmeaning: Taxonomic vs. thematic relations.Cognitive Psychology, 16, 1-27.

Markman, E. M., & Wachtel, G. A. (1988). Chil-dren's use of mutual exclusivity to constrainthe meanings of words. Cognitive Develop-ment, 20, 121-157.

Merriman, W. E. (1991). The mutual exclusivitybias in children's word learning: A reply toWoodward and Markman. Developmental Re-view, 11, 164-191.

Merriman, W. E., & Bowman, L. (1989). The mu-tual exclusivity bias in children's word learn-ing. Monographs of the Society for Research

GolinkofPetal. 3119in Child Development, 54 (3-4, Serial No.220).

Merriman, W. E., Marazita, J., & Jarvis, L. H.(1993). Four-year-olds' disambiguation of ac-tion and object word reference./ouma/ of Ex-perimental Child Psychology, 56, 412-430.

Merriman, W. E., Marazita, J., & Jarvis, L. H.(1995). Children's disposition to map newwords onto new referents. In M. Tomasello& W. E. Merriman (Eds.), Beyond names forthings (pp. 147-183). Hillsdale, NJ: Erlbaum.

Merriman, W. E., & Schuster, J. M. (1991). Youngchildren's disambiguation of object name ref-erence. Child Development, 62, 1288-1301.

Mervis, C. B., & Bertrand, J. (1994). Acquisitionofthe novel name-nameless category (N3C)principle. Child Development, 65, 1646-1662.

Miller, G. A., & Johnson-Laird, P. N. (1976). Lan-guage and perception. Cambridge, MA: MITPress.

Pinker, S. (1989). Learnability and cognition.Cambridge, MA: MIT Press.

Quine, W. V. O. (1960). Word and object. Cam-bridge: Cambridge University Press.

Rice, M. L. (1990). Preschoolers' QUIL: Quick in-

cidental learning of words. In G. Conti-Ramsden & C. Snow (Eds.), Children's lan-guage (Vol. 7). Hillsdale, NJ: Erlbaum.

Rice, M. L., & Woodsmall, L. (1988). Lessons fromtelevision: Children's word learning whenviewing. Child Development, 59, 420-424.

Rosch, E., Mervis, G. B., Gray, W. D., Johnson,D. M., & Boyes-Braem, P. (1976). Basic ob-jects in natural categories. Cognitive Psychol-ogy, 8, 382-439.

Shuff-Bailey, M., & Golinkoff, R. M. (1995, April).Lexical extension in young children: Percep-tual similarity vs. object kind. Society for Re-search in Child Development, Indianapolis.

TardifF, T. (1996). Nouns are not always leEimedbefore verbs: Evidence from Mandarin speak-ers' early vocabularies. Developmental Psy-chology, 32, 492-504.

Tomasello, M., & Barton, M. (1994). Learningwords in nonostensive contexts. Develop-mental Psychology, 30, 639-650.

Waxman, S. R., & Kosowski, T. D. (1990). Nounsmark category relations: Toddlers' and pre-schoolers' word-learning biases. Child Devel-opment, 61, 1461-1473.