The Role of Social-Referential Context in Verbal and Nonverbal Symbol Learning

The Role of Social-Referential Context in Verbal and

Nonverbal Symbol Learning

Aimee L. Campbell and Laura L. Namy

This study examined the role of social-referential context in 13- and 18-month-olds’ mapping of verbal andnonverbal symbols to object categories. Infants heard either novel words or novel nonverbal sounds in either areferential or nonreferential context. In all conditions, an experimenter engaged in a social-referentialinteraction and the label was produced while the infant’s attention was directed to the referent. In the referentialcondition, labels were produced by the experimenter within the context of a familiar naming routine. In thenonreferential condition, labels were emitted from a baby monitor placed near the infant. The studysubsequently tested infants’ mapping of the symbols to the referent objects using a forced-choice procedure.Although the results for the 18-month-olds were strongest, infants at both ages showed evidence of learningboth words and sounds in the referential condition and failed to learn them in the nonreferential condition.Thus, infants successfully learned both words and sounds under the same circumstances at both ages. Thesefindings suggest that the social-referential context, and not the symbolic form per se, determine infants’ successat symbol learning.

Infants’ emerging symbolic ability is one of the mostamazing and most enigmatic phenomena of theirearly development. Infants as young as 1 year seemto have surprisingly little trouble with the complex-ity of learning that some symbol (often completelyarbitrarily related to an object, action, or event) canbe used to stand for, represent, and convey informa-tion about that entity. Studies investigating infants’ability to learn words indicate that by the end oftheir first year, infants have acquired a keen ability tolearn words that stand for things in their environ-ment (Bates, Benigni, Bretherton, Camaioni, &Volterra, 1979; Bloom, 1993). Woodward, Markman,and Fitzsimmons (1994) demonstrated that infants asyoung as 13 months are able to interpret a novelword as an object name quickly in an experimentalsetting with limited number of exposures, and willapply the word to other category members.

Recent studies of the mechanisms underlyingearly word learning reveal evidence consistent with

a general symbolic ability supporting word learning.Infants exposed to a spoken language are not onlyable to learn words, but are also able to acquirenonverbal symbols such as gestures and tones asnames for things, and appear to acquire them underthe same learning conditions as they do words(Acredolo & Goodwyn, 1985, 1988; Hollich, Hirsh-Pasek, & Golinkoff, 2000; Iverson, Capirci, & Caselli,1994; Namy, 2001; Namy & Waxman, 1998; Wood-ward & Hoyne, 1999).

Specifically, observational evidence shows thatchildren exposed exclusively to a vocal languagewill produce gestural symbols to refer to things inthe environment in the same manner that theyproduce words to refer to things in the environment(Acredolo & Goodwyn, 1985). Acredolo and Good-wyn (1988) found that 87% of the 16- to 18-month-olds they studied had at least one sign in theircommunicative repertoire. These gestures appear tobe acquired at about the same time that thereferential use of words begins to occur (Folven &Bonvillian, 1991; Goodwyn & Acredolo, 1993; Meier& Willerman, 1995). There are strong correlationsbetween children’s early gestural production andlanguage development. For example, gestural com-binations have been noted to occur at the same timein development as two-word combinations, and thenumber of symbolic gestures being produced corre-lates with the rate of vocabulary development(Acredolo & Goodwyn, 1985, 1988). In addition,infants’ use of words and symbolic gestures seems to

r 2003 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/2003/7402-0015

Aimee Campbell, University Honors Program, VirginiaCommonwealth University; Laura L. Namy, Department ofPsychology, Emory University.

This project was supported in part by NIH grant # MH61846-02awarded to the second author. We thank Natalie Roth, RebeccaFiscal, Amy Meyer, and Alysha Jaffer for their assistance withvarious phases of this project. We also thank Alan Cienki, RobynFivush, Anne Fulkerson, Steve Nowicki, Lynne Nygaard, RachelRobertson, and three anonymous reviewers for their helpfulfeedback on early versions of this paper.

Correspondence concerning this article should be addressed toLaura L. Namy, Department of Psychology, Emory University,Atlanta, GA 30322. Electronic mail may be sent to [email protected].

Child Development, March/April 2003, Volume 74, Number 2, Pages 549–563

be complementary rather than redundant; that is,infants tend not to acquire both a gesture and a wordto name the same object (Acredolo & Goodwyn,1985, 1988; Iverson et al., 1994).

This observational evidence is bolstered byexperimental evidence directly comparing infants’acquisition of novel words and novel nonverbalsymbols as object names. Namy and colleagues(Namy, 2001; Namy & Waxman, 1998, 2000, 2002)have found that 17-month-olds interpret novelsymbols including both words and gestures as wellas nonverbal sounds and pictograms as names forobject categories. All gestures, words, sounds, andpictograms were novel and arbitrarily related to theobjects. Woodward and Hoyne (1999) reported asimilar phenomenon for the acquisition of wordsand nonverbal sounds. Infants as young as 13months readily accepted either a word or a soundas an object name. Hollich et al. (2000) have alsodemonstrated that nonverbal mouth noises arereadily mapped to object categories as early as 12months of age.

It is interesting that although the empiricalevidence suggests that initially, infants seem toregard words and nonverbal symbols as part of thesame symbolic system, a shift seems to take placearound 18 months of age (see Volterra & Iverson,1995). For example, Iverson et al. (1994) reported ashift from an equivalent number of words andgestures at 16 months to a strong preference for theverbal modality at 20 months. Experimental evi-dence for a shift from a more general, inclusivesymbolic ability to a priority for acquiring words isreported by Namy and Waxman (1998, 2002) andWoodward and Hoyne (1999). For example, Namyand Waxman (1998) compared 18- and 26-month-olds’ acquisition of a novel word or arbitrary gestureas a name for an object category. Although infants atboth ages learned a novel word as a category name,only the 18-month-old infants accepted the novelgesture as standing for the category. In fact, evenwhen the experimenter strongly emphasized thesymbolic nature of the gestures by embedding themwithin a highly familiar and ritualized namingroutine, 26-month-old infants did not learn thegesture as representing, or naming, the category.Only when the experimenter provided explicittraining and reinforcement for the child’s use ofgestures did the 26-month-olds interpret the ges-tures as standing for the categories. Woodward andHoyne similarly found that in contrast to 13-month-olds, 20-month-olds failed to interpret nonverbalsounds as object names. Namy and Waxman (1998)interpreted these findings as implying that over

time, children begin to discern that words areprivileged with respect to object naming as theyaccrue experience with language. Thus, an initial,general ability to learn symbols gives rise to a morefocused tendency to use words as the predominantsymbolic modality.

This evidence appears to support the notion thatinfants begin word learning with no clear priority forwords over other types of symbols. The expectationthat words are the predominant form of symbolicreference appears to emerge out of infants’ experi-ence with symbols more generally. Although theevidence in support of this account is compelling,there is some evidence that appears to challenge thisdevelopmental account.

Two findings in particular challenge the notionthat infants’ early symbolic abilities are general.These studies suggest that words hold a privilegedstatus in the symbolic repertoire from the onset ofcommunicative development. In fact, these studiessuggest that words have taken on a privileged statuswith respect to symbolic reference even before theonset of productive language in infancy. These twostudies are similar in that they each pit wordsagainst nonverbal sounds in prelinguistic infants.

Balaban and Waxman (1997) looked at the use of anoun phrase (e.g., ‘‘a bird’’) versus a duration-matched sine tone to facilitate perceptual categoriza-tion and category discrimination in 9-month-oldinfants. In their procedure, Balaban and Waxmanhabituated infants to categories of objects presentedon a screen that were accompanied by either thelabel or tone. If words and tones both facilitateattention to objects, infants in both conditions shouldshow evidence of forming object categories bypreferring to look at an object from a novel objectcategory following habituation. Results indicatedthat infants who had heard the novel word duringthe habituation phase looked reliably more at theobject from the new category during the test. Incontrast, infants who had heard the novel tone as aname for the category showed no preference for theobject from the new category. The authors inter-preted this finding as suggesting that infants havecertain expectations that words refer to objectcategories, and on hearing a word, they weremore likely to form a category for the objects. Thisfinding implies that infants do not have the sameexpectation for a novel tone as they do for a novelword.

Xu (1998) has made similar arguments about theprivileged status of words in the formation of objectsortals. She examined whether providing distinctlabels (e.g., ‘‘a ball’’ and ‘‘a duck’’) for two distinct

550 Campbell and Namy

objects facilitated infants’ ability to individuate thetwo objects. In the two-word condition, infants werefamiliarized with two objects emerging from eitherside of an occluder. During the presentation of eachof the objects, they received a label that wasproduced by a loudspeaker. For example, the infantheard ‘‘Look, a duck’’ or ‘‘Look, a ball.’’ In the one-word condition, infants also saw two objects emer-ging from either side; however, both objects receivedthe same label, ‘‘Look, a toy.’’ Next, the occluder waslowered to reveal either the expected outcome of twodifferent objects resting behind the occluder, or theunexpected outcome of a single object. Resultsindicated that infants in the two-word conditionlooked longer at the unexpected outcome, indicatingtheir expectation that there would be two objectsbehind the screen, whereas infants in the one-wordcondition did not show this trend. Xu interpretedthis finding as indicating that the presence of thelabel for each of the objects encouraged infants torecognize them as distinct. In an important manip-ulation, Xu performed the same procedure usingeither two distinct words or two tones as labels forthe objects. If infants interpret the tones, which werean octave apart, in the same manner that theyinterpret the words, as indicating distinct objects, thepresence of tones should facilitate their performanceon the task as well. Results showed that infantslooked longer at the unexpected outcome when theyheard two words labeling the objects but lookedlonger at the expected outcome when they heardtwo tones labeling the object. This finding suggeststhat although hearing two different words for theobjects facilitated their ability to discriminate twodistinct objects, hearing two different tones did notaffect their performance. In this case, then, infantsdid not use words and tones as similar cues to objectindividuation.

These two studies seem to suggest that words arealready a privileged form of reference for infants at 9to 10 months of age. Infants appear to expect wordsto name objects in a way that other arbitrary signalsdo not. This raises the possibility that the similarperformance of slightly older children on verbalversus nonverbal symbol-learning tasks does notimply a general symbolic ability from the onset ofcommunicative development. Perhaps the similarperformance observed for words versus nonverbalsymbols does not reflect similar processing mechan-isms. For example, perhaps the task is elicitingsymbolic mapping in the case of word learning butmerely paired association in the case of nonverbalsymbols, with no real appreciation of the referentialor representational nature of the signal.

There are two reasons to believe this is not thecase. First, the two studies that challenge theequipotentiality of verbal and nonverbal symbolsare visual attention tasks, not symbolic-mappingtasks such as those used by Namy (2001; Namy &Waxman, 1998) and Woodward & Hoyne (1999). Itmay be that the privileged influence of words oninfants’ visual discrimination is by virtue of the factthat one stimulus involves a human voice whereasthe other does not. We know from the literature oninfant attention that from birth, visual orienting isheightened by the human voice (e.g., see Ecklund-Flores & Turkewitz, 1996). This heightened attentionmight, by itself, lead to superior category discrimi-nation or object discrimination, independent of anyexpectations infants might have about the commu-nicative function of language. Second, the verbal andnonverbal stimuli employed in the studies by Balabanand Waxman (1997) and Xu (1998) were not necessa-rily matched for salience and complexity. When asimilar paradigm was employed using instrumentalRaffi music, similar visual attention and categorydiscrimination were found for a verbal versus anonverbal task (Roberts, 1995; Roberts & Jacob, 1991).

Nonetheless, it is a clear possibility that thelearning processes for words versus nonverbalsymbols may differ, despite the similar performancein the range of symbol-learning tasks employed thusfar. If words are interpreted symbolically andnonverbal symbols hold some more associative andless representational status, we should be able toreveal different learning conditions for words andother symbols. Namy and Waxman (2000) havealready revealed one such distinction that mayindicate different learning contexts, namely, the roleof sentence context in word versus gesture learning.Specifically, Namy and Waxman found that 17-month-olds will interpret gestures as object namesregardless of whether they are presented withinfamiliar naming phrases such as ‘‘Look at the y’’ Incontrast, infants more readily interpreted words asobject names when they were embedded in namingphrases than when they were presented in isolation.In the current study, we went one step furtherFweattempted to ascertain the symbolic nature ofinfants’ mapping of verbal and nonverbal signalsto objects by examining the social-referential cues tonaming inherent in a typical naming context.

We know from studies by Baldwin, Tomasello,and colleagues that children as young as 12 monthsand certainly by 18 months appreciate that pointingand eye gaze are important cues to naming, lendingstrong support to the claim that children appreciatethe referential and representational nature of words.

Referential Context and Symbol Learning 551

For example, Baldwin (1991, 1993) found that 18-month-olds are able to determine the appropriatereferent of a novel word by following the attention ofthe experimenter. Tomasello and colleagues (Toma-sello & Barton, 1994; Tomasello, Strosberg, & Akhtar,1996) have found that 18- and 24-month-old childrenare able to use information gained from an experi-menter’s behavior in determining the appropriatereferent for a word under even more complicated orsubtle circumstances.

In perhaps the most direct test of infants’appreciation of social-referential cues versus pairedassociation as a mechanism for word learning,Baldwin et al. (1996) compared infants’ acquisitionof words when they were introduced to novel wordsproduced either with or without cues to referentialintent. In this study, infants were presented with anovel word in either a coupled or decoupledcondition. In the coupled condition, the novel wordwas produced by the experimenter who wasengaged in a joint-attention episode with the child.In the decoupled condition, the novel word wasproduced by a confederate who was out of sight andpretending to talk on a telephone. However, thepresentation of the word by the confederate wascontingent on the child’s attention; the word wasproduced only when the child was engaged with theobject. Mapping of the novel word to the targetobject was tested by a forced-choice paradigmrequiring the child to select the object from a traycontaining both the target and a distractor toy.Results indicated that 18- and 20-month-old infantswere more likely to select the target toy in thecoupled condition. The researchers concluded thatinfants as young as 18 months of age understandthat the speaker identifies the referent of a word, andif cues to referential intent are lacking, no symbolicmapping should be made. In a similar study,Fulkerson, Shull, and Haaf (2002) recently demon-strated that infants as young as 15 months of age areable to use an object label to aid categorization whenit is presented by the experimenter in the context ofan interaction, but not when it is played by a taperecorder placed at a distance from the experimenter,suggesting a sensitivity to context at an even earlierage.

Our primary question of interest in this study waswhether social-referential cues similarly facilitatenonverbal symbol learning. If infants acquire non-verbal object names via associative principles, thesocial-referential context should not influence in-fants’ success at symbol-to-referent mapping. If,however, infants appreciate the symbolic andrepresentational nature of the nonverbal, as well

as the verbal, signals, they should show thesame sensitivity to social-referential cues in asymbol-learning task for both verbal and nonverbalstimuli.

To examine this question, we developed aprocedure similar to that used by Baldwin et al.(1996). Infants were assigned to one of twoconditions, either the referential or nonreferentialcondition. In both conditions, infants engaged in aninteractive play session with an experimenter whowas seated across the table from them, and in bothconditions the production of a novel symbol wasentirely contingent on and associated with infants’visual attention to the target object being labeled. Inthe referential condition, the experimenter, who wasengaged with the infant in a socially dynamicinteraction, produced the symbol as part of thatinteraction. In the nonreferential condition, theexperimenter was also engaged with the infant;however, the symbol was not produced as part ofthis interaction, but rather originated from anexternal source and was not contingent on theexperimenter’s social-referential cues. In each con-dition, half of the infants were introduced to novelwords and the other half were introduced to novelnonverbal sounds as names for object categories.

All infants were subsequently tested on theirmapping of the symbol (word or sound) to the targetobject, as well as their extension of the symbol toadditional members of the target category that hadnot been explicitly labeled. The difference betweenthe referential and nonreferential conditions was asubtle manipulation because, from the infants’perspective, the only difference was the sourceand the timing of the symbol. As a result, thismanipulation provided a strong test of infants’sensitivity to the social-pragmatic factors thataccompany symbols.

To examine the development of infants’ emergingsymbolic ability, both 13- and 18-month-olds’ per-formance on this symbol-learning task was exam-ined. Although a good amount of evidence has beenaccrued supporting the idea that 18-month-olds areproficient and flexible symbol learners, less is knownabout symbol acquisition at 13 months. By examin-ing the performance of infants at these ages, adevelopmental picture examining the emergence ofa symbolic ability can be formed. This also allows usto test the possibility that the equipotentiality ofwords and nonverbal symbols observed at 18months in previous studies has emerged over thefirst several months of word learning, with a priorityfor words evident at the very onset of communica-tive development.


Method

Participants

Participants in this experiment consisted of sixty-four 13-month-old infants (age: M5 13.57 months;range5 12.70–14.74 months) and sixty-four 18-month-old infants (age: M5 17.82 months; range517.04–19.18) who were recruited via direct mailings.Participants were from predominantly White orBlack middle-class families.

A stringent inclusion criterion was imposed, inthat only children who made a clear choice on atleast 13 of the 16 trials presented (discussed later)were included in the analysis. Fifteen additionalinfants were excluded (six 13-month-olds and nine18-month-olds), 6 for failing to make a choice on atleast 13 trials, 7 for failing to complete the task, and 1for exhibiting a side preference on all 16 trials. Thenumber of infants excluded did not differ system-atically by condition or symbol type learned.

Materials

Stimuli consisted of 10 small, manipulable toyreplicas of objects that were used in the context of agame with the experimenter. All of the objects wereselected to be common and familiar to infants atthese ages. Two of the objects, a rabbit and a spoon,were used in a warm-up trial to teach the infant the‘‘finding’’ game. The remaining eight objects wereused in the experimental portion of the procedure.The eight objects consisted of two visually discri-minable examples from each of four categories:apples, cups, planes, and horses.

The experimental session was taped by a videocamera situated to record the ongoing interactionbetween the experimenter and the infant. Thisinteraction was transmitted by live video feed to atelevision monitor outside the testing room, where aconfederate viewed the interaction. Placed near theconfederate was the transmitting end of a babymonitor. The receiving end of the baby monitor wasplaced in the testing room, near the infant. Thevolume of the monitor was set at the same level forall infants in the nonreferential condition, which wascalibrated to be roughly equivalent to the volume ofthe symbols when produced by the experimenterseated across from the infant. Clarity and intellig-ibility of the words and sounds were comparablewhen emitted from the baby monitor to thoseproduced by the experimenter.

The novel symbolic stimuli included two novelwords and two novel nonverbal sounds. Thenovel words employed were foppick and tillen.The novel sound stimuli included a two-tone beep

and a rising glissando. The nonverbal sounds wereproduced by a battery-operated, hand-held devicethat could be hidden out of view of the infant, on theexperimenter’s lap.

Procedure

Infants at each age were randomly assigned toeither the referential or nonreferential condition andto either the word or sound group within eachcondition. Each infant was seated in a booster seatacross a table from the experimenter, with the parentseated next to the infant. Parents were instructed notto speak during the procedure. Where infants wereuncomfortable sitting by themselves in the boosterseat, they were transferred to their parent’s lapduring the procedure. The video camera waspositioned over the experimenter’s left shoulder.The receiving end of the baby monitor was posi-tioned on a window sill to the right of the infant. Theinfants were able to see the monitor if they turnedtheir heads to the right. The procedure wascomposed of three phases: a warm-up phase, anintroduction phase, and a test phase.

Warm-Up phase. The procedure began with awarm-up phase designed to introduce the child tothe finding game that was used to assess symbolinterpretation. This phase was identical across allages and conditions. The infants were shown twofamiliar objects (e.g., a rabbit and a spoon) one at atime and were allowed to play freely with each ofthem. During play, the experimenter called theinfant’s attention to each toy six times. For one ofthe two objects, designated the target object, theexperimenter labeled the toy with its familiar basic-level name (i.e., bunny or spoon) by saying, forexample, ‘‘Look at what you have y bunny! You seethat? y bunny! Do you like that one? y bunny!’’For the remaining toy, designated the distractor, theexperimenter drew similar attention to the toyomitting only the label, saying, for example, ‘‘Lookat what you have! You see that? Do you like thatone?’’ Thus, the two toys received equal attentionfrom the experimenter, but only one was labeled.Assignment of the target and distractor object andorder of presentation of the two objects wasrandomly determined for each infant.

After the experimenter referred to each object sixtimes, she administered a series of four forced-choicetrials during which the infants were presented withboth toys and were asked to choose one. Two of thechoice trials were target trials and two of the choicetrials were control trials. On the target trials, theexperimenter asked for the toy that had been named


in the warm-up. For example, the experimenterplaced both of the toys in front of the child, one oneach side of the table, and said, ‘‘Which one can youget? Can I have the bunny?’’ While requesting the toy,the experimenter placed her hand, with the palm up,at the infant’s midline to indicate her desire for theinfant to choose one of the toys. During the request,her eyes remained focused on the child’s face or herown hand rather than the toys. During the target trialsof the warm-up phase, the infants were reinforced forchoosing the correct toy to convey to the child thegoal of the game. If the child selected the target toy,the experimenter cheered and said, for example,‘‘That’s right! That’s the bunny! You’re doing such agood job!’’ If the child failed to select the target toy,the experimenter highlighted his or her mistake bysaying, for example, ‘‘That’s not the bunny! Look, thisone is the bunny!’’ The control trials proceeded inexactly the same fashion as the target trials; however,the infant was simply asked to choose one of theobjects. Again, the experimenter presented the infantwith both of the toys. However, in the control trials,the experimenter said, ‘‘Which one can you get? Canyou pick one?’’ During the control trials, the infantwas positively reinforced for choosing either toy. Thepurpose of the control trials was to insure that theinfants were attending to the instructions and werenot consistently selecting the target object regardlessof the instructions, simply because labeling height-ened their attention to that object.

Pilot testing suggested that when infants receivedthe target trials first, they were more likely to choosethe target toy during the control trials, whichdefeated the purpose of this training phase. How-ever, when control trials were presented first, infantsappeared to understand that either choice wasacceptable when neither object was explicitly namedduring the questioning. For this reason, we consis-tently presented the two control trials first for allinfants during the warm-up phase. The side pre-sentation of the objects was randomly determinedfor each trial.

Introduction phase. Following the practice trials,the experimenter administered the experiment prop-er. The child was introduced to two sets of objects,each presented in a manner similar to the warm-upphase. Within each set, there were two pairs ofobjects, a mapping pair (e.g., a red apple and a bluecup) and an extension pair (e.g., a green apple and asilver cup). The pairing of the toys and the order ofpresentation of the pairs were randomly assigned.Within each pair, one of the toys was randomlydesignated the target object. As in the warm-upphase, the infants were introduced to each toy one at

a time in a random order. While the target toy waspresented and the child was attending to it, asymbolic form (either word or nonverbal sound)was produced. The key manipulation of socialcontext involved how the symbol was introducedto the infant. In the referential condition, theexperimenter presented the target toy to the infantand labeled it six to eight times in the context of theirinteraction just as she had during the warm-upphase. The labels were specifically produced whenboth the child and the experimenter were looking atthe object. In the word condition, the experimenterverbalized the word. In the tone condition, thenonverbal sound was produced by the experimenterby activating a hand-held device during the interac-tion. The sound-producing device was held in theexperimenter’s lap and was not visible to the child.For example, in both the word and sound condition,the experimenter said, ‘‘Look at what you have![symbol] That’s what we call that one. Do you seewhat you have there? [symbol] That’s what that is!Do you like that one? [symbol]’’ This is similar to thetechnique employed successfully by Namy (2001).

In presenting the distractor toy, the experimenterbehaved identically in both the word and soundconditions. Specifically, the experimenter paid equalattention to the distractor as she did to the target toyduring the familiarization phase, saying, ‘‘Look atwhat this is! See what you have? Do you like thatone?’’ However, the distractor object never receiveda label of any kind.

In the nonreferential condition, when the experi-menter presented the target toy, she interacted withthe infant in a similar manner, saying, for example,‘‘Look at what this is! Do you like that one? Look atwhat you have!’’ However, the symbol (either theword or the nonverbal sound) was not presented bythe experimenter in the context of the interaction, butrather was emitted from the baby monitor. Althoughthe symbols were produced when the infant waslooking at the object, the production of the symbolwas not contingent on the actions of the experimen-ter, as it was during the referential condition. Forexample, the symbol did not always directly followone of the experimenter’s phrases or points, butrather simply occurred when the infant’s attentionwas focused on the object. The nonreferentialmanipulation was accomplished by having a con-federate view the child’s interactions with the toysfrom an adjoining room on a television monitortransmitting a live feed from the video camera. Theconfederate watched the interaction without sound,ensuring that the symbol production was contingenton child gaze but was not explicitly contingent on the


experimenter’s speech or social-referential cues. Thisconfederate then produced the appropriate symbolvia the baby monitor at the appropriate time (i.e.,when the infant was looking at the target object).

When presenting the distractor object, the experi-menter’s behavior was identical to that in thereferential condition. The experimenter drew atten-tion to the object without labeling it.

The experimental set-up minimized the possibilityof the confederate’s producing the symbols inconjunction with the experimenter’s referential acts.First, the confederate could see but not hear theexperimental session on the video monitor. Second,the positioning of the camera provided the confed-erate with a clear view of the infant and the table, butdid not clearly display the experimenter’s move-ments. Although the experimenter’s hands occasion-ally came into view on the video monitor, this did notoccur with sufficient frequency to cue consistentlythe confederate as to the experimenter’s referentialacts. In addition, the confederate was instructed tofocus exclusively on the child’s visual attention to theobject, not on the experimenter. Finally, the confed-erate was blind to the experimental hypotheses.

From the infants’ perspective, the only elementthat differed between the referential and nonrefer-ential conditions was the source and the timing ofthe symbol presented with the target object. Theexperimenter who was interacting with the infantbehaved identically in all other respects, drawing theinfants’ attention to the object by pointing anddirecting her gaze at the objects.

Test phase. As during the warm-up phase, theexperimenter then administered a series of fourforced-choice trials including two target trials andtwo control trials. The infants were first tested ontheir mapping of the symbol to the objects that hadbeen presented during the introduction phase, andthen were tested again with an extension pair on twotarget trials and two control trials.

For both mapping and extension, the presentationof the target and control trials was blocked, in thatthe child either received the two control trials first orthe two target trials first. The presentation order ofthe blocks was randomized. Test trials were identicalacross age, symbol type, and condition. Duringtarget trials, the experimenter placed both toys infront of the child, on each side of the table, andrequested the target object by producing either theword or the nonverbal sound. For example, theexperimenter would say, ‘‘Which one can you get?[symbol] Can you get it? [symbol]’’ During the testphase for either condition, the symbol was alwaysproduced by the experimenter, by verbalizing the

word or by using the hand-held sound-producingdevice. The control trials were identical to thoseduring the warm-up phase. The experimenter pre-sented the child with both of the toys and said,‘‘Which one can you get? Can you pick one?’’

Following the mapping and extension trials for thefirst stimulus set, the introduction phase and the testphase (mapping and extension) were repeated forthe second set of objects. Thus, there were a total offour target and four control trials assessing theinfants’ mapping of the symbol onto the originaltarget object and an equal number of observationsper child for the extension trials.

Assessment of context manipulation. To confirm thatthe referential and nonreferential conditions wereroughly equated with respect to the nature of theexperimenter interaction, a group of six naive codersviewed a randomly selected 12 children’s experi-mental sessions from each condition (equally dis-tributed among the two ages and symbol types) andrated the experimenter’s enthusiasm and engage-ment with the toys and the choosing game. The 24sessions were viewed in a random order. Coderswere told nothing about the nature of the studyexcept that we were interested in the ways childrenlearn names for objects. They were also explicitlyinstructed to ‘‘ignore the exact details of theprocedure and focus only on the experimenter andher overall level of enthusiasm’’ for the objects andthe game. Coders assigned an enthusiasm score toeach session using a 7-point scale with 1 labeled as‘‘completely unenthusiastic,’’ 4 as ‘‘moderately en-thusiastic’’ and 7 as ‘‘extremely enthusiastic.’’ Therange of scores varied from 4 to 7, with an averageenthusiasm rating of 5.97. To ensure that theexperimenter was generating as much enthusiasmfor the objects and the game in the nonreferentialcondition as in the referential condition, we con-ducted a t test comparing the ratings for thereferential (M5 5.88, SD5 .34) and nonreferential(M5 6.07, SD5 .31) conditions. This analysis indi-cated that the experimenter was actually rated asslightly but significantly more enthusiastic in thenonreferential condition, t(5)5 2.62, po.05. Thisheightened enthusiasm toward the objects and thegame in the nonreferential condition, while unex-pected, is valuable in that it biases the outcome, if atall, against our hypothesis in favor of the learning ofsymbols in the nonreferential condition and not inthe referential condition.

In addition, we confirmed that infants in bothcontext conditions received equal numbers of clearpresentations of the label (word or sound) whileattending to the target object. This was a concern


because the confederate could not hear the experi-menter and might have occasionally produced thesymbols while the experimenter was talking, result-ing in a smaller number of clear exposures to thesymbols in the nonreferential condition. For arandomly selected 16 children’s sessions in eachcondition (equally distributed among the two agesand two symbol types), we tabulated the number oftimes the label was presented clearly while the infantwas attending to the object in each session (includingboth Set 1 and Set 2). Labeling instances in which thechild was looking away from the object or when theconfederate’s production of the label was eclipsed bythe experimenter speaking to the child were notincluded in the tally. This analysis indicated that theexperimenter produced symbols an average of 6.4times per set in the referential condition, and theconfederate produced the symbols clearly (withoutvocal competition from the experimenter) an aver-age of 6.8 times per set. A t test comparing thenumber of clear labels produced in the two condi-tions was not significant, suggesting no systematicdifference in the number of presentations of thelabels by condition.

Choice coding. Infants’ choices on each trial werecoded from videotape. On each trial, infants couldselect the target object, the distractor, or make noclear choice between the two objects. A choice wasdefined as the first object a child touched. If the childtouched both toys simultaneously and then pro-ceeded to hand one to the experimenter, this toy wasconsidered the choice. Any child who failed to makea clear choice on more than two target trials, morethan two control trials, or more than three trialsoverall was excluded from the experiment.

One primary coder analyzed all of the videotapesand a secondary coder who viewed only the testphase and coded with the sound off (and wastherefore blind to target object assignment, symboltype, and condition) analyzed a randomly selected25% of the videos of the infants in each condition.Intercoder agreement on individual trials was 95%.

If the infants are mapping the symbols to thetarget object, they should choose the target object atabove-chance rates in the target trials and select thetarget object significantly more in the target trialsthan in the control trials. If the infants fail to map thesymbols to the object, they should respond at chancein both the target and control trials, yielding noreliable differences between trial types. If the infants’attention to the target object is heightened by the actof labeling, but a symbolic interpretation was notmade, they should choose the target object at above-chance rates in both the target and control trials,

again yielding no reliable differences between trialtypes.

Results

The mean proportion of trials on which the infants ineach group selected the target toy in both the targetand control trials is presented in Table 1. Weassessed infants’ performance on the task in threeways: (a) by examining group effects of condition,(b) by comparing infants’ performance to chance,and (c) by examining individual patterns. In allanalyses, we compared the proportion of the trialson which infants chose the target object for the targetversus control trials. Success on this symbol-learningtask was defined as heightened selection of thetarget object on target trials relative to control trialsand relative to chance responding. We first assessedwhether there were a priori differences in infants’success on the task attributable to a clearer under-standing of the finding game. To assess thispossibility, we examined infants’ performance onwarm-up trials. We tested infants’ ability to findobjects corresponding to familiar basic level objectlabels (e.g., bunny).

Warm-Up Trials

To examine any systematic variation in perfor-mance related to the infants’ understanding of thefinding game, we conducted a four-way ANOVA oninfants’ performance on the warm-up trials with age(2: 13-month-olds vs. 18-month-olds), symbol (2:word vs. sound), and context (2: referential vs.

Table 1

Mean (and Standard Deviation) Proportion Target Choices on Target

and Control Trials for Each Context Condition and Symbol Type at Each

Age

Condition

Referential Nonreferential

Target trials Control trials Target trials Control trials

13-month-olds

Word 0.601 0.503 0.548 0.468

(0.258) (0.240) (0.199) (0.199)

Sound 0.559 0.479 0.409 0.473

(0.174) (0.189) (0.214) (0.236)

18-month-olds

Word 0.619 0.454 0.471 0.513

(0.207) (0.203) (0.195) (0.264)

Sound 0.661 0.498 0.519 0.538

(0.195) (0.192) (0.222) (0.217)


nonreferential) as between-subject factors, and trialtype (2: target vs. control) as a within-subject factor.As expected, the analysis revealed a main effect oftrial, F(1, 109)5 9.44, po.01. Infants were more likelyto choose the target object in the target trials(M5 0.59, SD5 0.42) than in the control trials (M5

0.47, SD5 0.42). However, there was also an un-anticipated trial by context interaction, F(1, 109)53.98, po.05. On practice trials, infants in thereferential condition were more likely to choose thetarget object in the target trials (M5 0.65, SD5 0.42)than they were in the control trials (M5 0.45,SD5 0.42). However, performance in the nonrefer-ential condition did not differ by trial type (M5 0.53,SD5 0.42 for the target trials and M5 0.49, SD5

0.42 for the control trials), Tukey’s HSD, po.01. Thatis, infants randomly assigned to the referentialcondition tended to perform better on the practicetrials than those in the nonreferential condition.There were no other main effects or interactions. Weaccounted for these sample differences in the basiccomprehension of the finding game by conductingtwo types of subsequent analyses.

In one set of analyses, we calculated a differencescore on infants’ performance in the warm-upphase (number of target choices on target trials minusnumber of target choices on control trials) andentered it as a covariate in subsequent analyses.This approach is imperfect because the warm-uptrials were actually intended to teach infants thegame rather than to index their a priori under-standing of it. That they received feedback duringthe warm-up trials renders it likely that many of thechildren who did poorly overall on the warm-uptrials had, in fact, learned the game by the time theexperiment proper commenced. However, this co-variate offers the advantage of allowing us toinclude performance for all participants whilemathematically controlling for any variance in theexperiment proper accounted for by initial compre-hension of the task. For 11 infants, a warm-updifference score could not be calculated because theydid not make a clear choice on either of the practicecontrol trials, on either of the practice target trials, orboth. The mean difference score was calculated forthe other 117 infants (M5 0.12), and this value wasentered as the estimated difference score for the 11missing values.

In our second set of analyses, we reconducted thestatistics including only those infants whose differ-ence scores were greater than zero. Although thisapproach eliminated many of the infants from theanalysis (including many who performed well in theexperiment proper, further challenging the validity

of the warm-up measure), it offers the advantage ofeliminating any possible effects of task comprehen-sion on performance. Conducting the analyses bothways provides the most comprehensive character-ization of the phenomena by seeking convergingfindings across these two approaches.

Effects of Condition

We assessed the effects of several factors oninfants’ performance, the form of the symbol (wordvs. sound), the context in which the symbol waspresented (referential vs. nonreferential), the age ofthe infant (13 months vs. 18 months), and mappingvs. extension as well as trial type (target vs. control).We first performed a five-way ANCOVA with age(2), symbol type (2), and context (2) as between-subject factors and trial (target vs. control) and type(mapping vs. extension) as within-subject factors,using warm-up scores as a covariate. This analysisyielded a main effect of trial, F(1, 119)5 7.49, po.01,indicating that overall, infants were more likely tochoose the target toy in the target trials (M5 0.54,SD5 0.21) than in the control trials (M5 0.49,SD5 0.21). This main effect was mediated by aTrial�Context interaction, F(1, 119)5 8.75, po.01.Infants who were presented with the symbols in thereferential context were more likely to choose thetarget toy in the target trials (M5 0.60, SD5 0.22)than they were in the control trials (M5 0.49,SD5 0.22). However, performance in the nonrefer-ential condition did not vary as a function of trialtype (M5 0.48, SD5 0.22 for the target trials andM5 0.49, SD5 0.22 for the control trials), Tukey’sHSD, po.01. This interaction reveals that infantsmore readily mapped symbols to their referents inthe referential than in the nonreferential condition.

Additionally, the analysis revealed a main effectof type (mapping vs. extension), F(1, 119)5 6.61,po.05. We were surprised to find that infants weremore likely to choose the target toy during theextension trials (M5 .54, SD5 0.21) than during themapping trials (M5 .49, SD5 0.21). This effect maybe driven by a relative novelty preference for thedistractor object during the mapping trials afterincreased attention to the target object during theintroduction phase. The effect of the mapping versusextension did not interact with trial (target vs.control), suggesting that the finding reflects moreof an attentional component of the infants’ behaviorin this task rather than their symbolic understand-ing. A similar effect of type (mapping vs. extension)has been reported by Namy and Waxman (2000)using a similar type of paradigm.


There were no main effects or interactionsinvolving symbol type. In fact, the F value for themain effect of symbol type was less than 1, hintingthat the social-referential context had a similarinfluence on both word and sound learning. Thiswas confirmed by effect size analyses. Using the o2

statistic, we calculated that at 18 months, the effect ofcontext on performance was medium to large forboth words and sounds (o25 .048 and .097 forwords and sounds, respectively; see Cohen, 1977, fora discussion of magnitude of effects). At 13 months,there was a medium to large effect for sounds(o25 .049); however, the F value of the context effectwas less than 1 for words, indicating a null effect ofcontext in this condition. Thus, the effect of contexton sound learning was at least equal in magnitude tothe effect on words at both ages.

Next, we conducted a 2 (age)� 2 (symboltype)� 2 (context)� 2 (trial)� 2 (type) ANOVAincluding only those children who succeededon the warm-up trials (difference score40, N5 39).This analysis yielded only a Context�TrialType interaction, F(1, 31)5 6.52, po.05. Posthoc analysis indicates that, as in the ANCOVA,children in the referential condition selected thetarget object reliably more often on the target trials(M5 0.61, SD5 0.22) than on the control trials(M5 0.45, SD5 0.20), but did not do so in thenonreferential condition (M5 0.47 and 0.57 forthe target and control trials, respectively, SD5 0.16and 0.24). Performance on the target trials wasreliably higher for the referential condition than forthe nonreferential condition, but there was nodifference between the two conditions for the controltrials.

Comparisons With Chance

We augmented these group comparisons bycomparing performance with chance in each cell.All comparisons were one-tailed, assessing whetherthe means exceeded chance performance.

As earlier, we conducted the analyses separatelyfor the entire sample and for only those childrenwho succeeded on the warm-up trials. However,the small sample size for the set of childrensucceeding on the warm-up trials offered limitedpower, yielding sample sizes ranging from 2 to 8 percell at each age. As a result, we used a marginalcriterion (po.10) for these reduced samples andinterpreted the outcomes with caution, seeking onlyto bolster or confirm the full-sample analysis ratherthan draw strong conclusions from those analysesalone. Infact the reduced sample analyses closely

mirrored the analyses from the full sample at eachage.

The 18-month-old infants showed clear patternsthat were consistent with the findings from theAN(C)OVAs. Those in the referential conditionselected the target object at a rate significantlyhigher than chance for the target trials in both thesound and word conditions using the full sample ineach group, t(15)5 4.37 and 2.27, respectively, bothpso.01, and were marginally above chance in thereduced sample of those succeeding on the warm-uptrials for both symbol types, t(7)5 1.51, p5 .087 forthe word condition, and t(8)5 1.83, p5 .063 for thesound condition. Performance did not differ fromchance on control trials for either symbol type, usingeither measure. The 18-month-olds in the nonrefer-ential condition failed to choose the target object at arate significantly higher than chance for either trialtype in either the word or sound condition, usingeither measure. These results confirm that at 18months, a symbol-to-object mapping is evident onlyin the target trials of the referential condition, andthat words and sounds follow a similar pattern.

In general, 13-month-olds revealed a similar butless robust pattern. The 13-month-old infants in thereferential condition who were learning wordsselected the target object at a rate significantlyhigher than chance in the target trials using the fullsample, t(15)5 1.92, po.05, but failed to exceedchance using the reduced sample that succeeded onthe warm-up trials. Both samples failed to exceedchance performance on the control trials. The 13-month-old infants learning sounds in the referentialcondition failed to exceed chance responding foreither target and control trials using the full sample,but reached marginal significance for the target trialsusing the reduced sample, t(4)5 1.596, p5 .093. Inthe nonreferential condition, the 13-month-old in-fants failed to choose the target object at a ratesignificantly higher than chance for either trial typein either the word or sound condition, using eithermeasure.

Thus, comparisons with chance largely mirror thefindings of the AN(C)OVAs at 18 months, revealingabove-chance performance in the referential butnot the nonreferential condition for both symboltypes. These analyses reveal a similar, but morevariable pattern for the 13-month-old infants. Inthe referential condition, 13-month-olds exceededchance performance for the word conditionusing one measure and for the sound conditionusing the other. In the nonreferential condition,13-month-olds did not differ from chance for eithersymbol type.


Individual Patterns of Performance

To examine how the performance of the individ-ual infants related to the group trends, we comparedthe number of infants in the referential andnonreferential conditions who chose the target objectmore often on the target trials than they did on thecontrol trials, collapsed across mapping and exten-sion trials. This criterion is liberal in that it includeseven children who selected the target only oncemore in the target than in the control trials.However, it is useful in bolstering our claims to theextent that it captures individual patterns consistentwith group performance. Tables 2 and 3 depict theproportion of children adhering to this criterion forthe full and reduced samples, respectively. Given thesmall sample sizes in each cell of the design for thereduced sample, we do not attempt to analyze these

smaller samples statistically, performing quantita-tive analyses for the full sample only. However, wereport them to demonstrate that the trends found forthe full sample are fully replicated by the reducedsample and, in fact, tend to appear even more robustwithin the selected sample.

For the 18-month-olds overall, significantly moreinfants showed the targeted pattern of success (thatis, more target choices on target trials than on controltrials) in the referential condition than in thenonreferential condition, Fisher’s exact test,p5 .005. Collapsing across conditions, we found nodifference in the number of infants choosing thetarget object more in the target trials in the wordversus the sound condition, Fisher’s exact test,p5 .19, consistent with the similar performance forthe two symbol types in the ANOVAs. Whencomparing the referential condition with the non-referential condition for each symbol type, we foundthat more 18-month-olds learning sounds in thereferential condition demonstrated this pattern ofsuccess than did those learning sounds in thenonreferential condition, Fisher’s exact test,p5 .002. However, the number of infants learningwords who chose the target more in the target trialsthan in the control trials did not differ between thereferential condition and the nonreferential condi-tion, Fisher’s exact test, p5 .22.

At 13 months, we found similar patterns. Agreater number of infants succeeded in the refer-ential condition than in the nonreferential condition,Fisher’s exact test, p5 .044. In addition, the 13-month-olds did not differ in their likelihood ofchoosing the target more in the target trials bysymbol type, Fisher’s exact test, p5 .19. Whenexamining effects of condition separately for eachsymbol type, a marginally significantly greaternumber of infants learning sound showed thispattern of success in the referential condition thanin the nonreferential condition, Fisher’s exact test,p5 .06. However, no difference in responding wasfound for the word condition, Fisher’s exact test,p5 .22. Infants’ performance seems to be morerobustly affected by the context (referential vs.nonreferential) in the sound condition than in theword condition at this age.

Looking Behavior of Infants

As a supplementary analysis, we examined thenumber of times infants looked to the monitor and tothe experimenter during the labeling events. Weinterpreted looks to the monitor as an index ofinterest and attention to the novel source of the

Table 2

Number of Individual Children Selecting the Target Object More Often

on Target Than Control Trials/Total Sample Size in the Full Sample for

Each Symbol Type and Condition, at Each Age

Condition

Referential Nonreferential Total

13-month-olds

Word 9/16 7/16 16/32

Sound 10/16 5/16 15/32

Total 19/32 12/32 31/64

18-month-olds

Word 9/16 7/16 16/32

Sound 13/16 4/16 17/32

Total 22/32 11/32 33/64

Table 3

Number of Individual Children Selecting the Target Object More Often

on Target Than Control Trials/Total Sample Size in the Reduced Sample

(Including Only Those Succeeding on the Warm-Up Trials) for Each

Symbol Type and Condition, at Each Age

Condition

Referential Nonreferential Total

13-month-olds

Word 3/4 1/2 4/6

Sound 4/5 1/7 5/12

Total 7/9 2/9 9/18

18-month-olds

Word 4/6 1/2 5/8

Sound 6/8 1/5 7/13

Total 10/14 2/7 12/21


auditory stimulus. In contrast, we interpreted looksto the experimenter as social-referencing acts, bidsfor information about the meaning of the auditorystimulus. The configuration of the camera relative tothe child did not lend itself to accurate and reliableduration measures; therefore, we employed thenumber of looks as a cruder measure of attentionand social reference.

We examined the number of looks to the monitorfor infants in the nonreferential condition using a 2(age)� 2 (symbol type)� 2 (set) ANOVA. Thisanalysis yielded no effect of age or symbol typebut a main effect of set, F(1, 60)5 30.19, po.001. Thiseffect indicates that infants looked to the monitorsignificant more often on the first set (M5 2.91,SD5 2.26) than on the second set (M5 1.48,SD5 1.93). This pattern reflects the relative noveltyof the sound source in set 1, revealing that infantsbecame more accustomed to the experience by thesecond set. This confirms that infants were indeedattending to the monitor, especially during their firstencounter with sounds emitted from the monitor.This suggests that infants’ lack of mapping of thesymbols in the nonreferential condition was not dueto failure to notice or attend to the sounds emanatingfrom the monitor.

Next, we performed an analysis of the number oflooks to the experimenter’s face during labeling,using a 2 (age)� 2 (symbol type)� 2 (context)� 2(set) ANOVA. This analysis yielded no effect of age,context, or symbol. There was a main effect of set,F(1, 120)5 10.13, po.005, revealing that infantslooked significantly more often to the experimenterduring the first set (M5 4.21, SD5 2.90) than duringthe second set (M5 3.38, SD5 2.48). This effect wasmediated by an interaction between set and symboltype, F(1, 120)5 7.31, po.01. Post hoc analyses(Tukey’s HSD) revealed that infants looked reliablymore often to the experimenter during Set 1 thanduring Set 2 in the sound condition (M5 4.53 and3.00, SD5 2.88 and 2.36 for Sets 1 and 2, respec-tively) but did not demonstrate a decline inexperimenter looks in the word condition (M5 3.89and 3.77, SD5 2.91 and 2.56 for Sets 1 and 2,respectively). Despite this pattern, the number oflooks did not differ reliably between the word andsound condition during either set. This findingsuggests that children were frequently consultingthe experimenter’s face for cues to help interpret thenovel symbolic input, regardless of the contextof symbol type. However, by the second set,children seem to respond differently to wordsand sounds, orienting less toward the experimen-ter’s face as a source of information about the

sound, but maintaining orientation toward theexperimenter as an information source in the wordcondition.

Discussion

These data provide clear evidence that 18-month-oldinfants map both words and nonverbal sounds toobjects when the symbols are presented in areferential context. However, when these symbolsare presented in a nonreferential context, infants failto learn them as names of objects. The data from the13-month-old infants are similar to those for theolder children, but less consistent. There is a generaltrend toward more successful symbol learning in thereferential condition than in the nonreferentialcondition for both symbol types at both ages.However, comparisons with chance and analyses ofindividual patterns indicate more robust effects ofcontext at 18 months than at 13 months for bothsymbol types. Overall, these findings imply that thedriving force behind infants’ interpretation of asymbol as an object name is not the form ofthe symbol per se, but rather the context in whichthe symbol is presented.

These findings are striking for two reasons. First,infants resist interpreting nonreferential symbols asobject names, even when the availability of social-referential cues during the naming episodes is heldconstant. Consistent with work by Baldwin et al.(1996), we find that as early as 13 months, infantsdiscriminate between labeling (those contexts inwhich the symbols originated from the sameanimate source as the social-referential cues) andnonlabeling contexts (those in which the symbolswere produced by an inanimate source that func-tioned independently from the social-referential cuesavailable). Second, we find no systematic differencebetween the conditions under which infants mapwords versus other nonverbal symbols to objectcategories. This suggests that infants are using cuesto intentionality rather than expectations about aparticular modality or symbol type to determinesymbolic mappings.

The finding that infants are able to discriminatebetween labeling and nonlabeling contexts stronglysuggests that infants interpret these words andsounds as symbolic, mapping them to objectcategories only when appropriate cues to namingare present. Infants’ sensitivity to the intentional,communicative nature of symbols argues against amore associative, stimulus–response learning pro-cess as accounting for early symbol learning. Thisfinding does not, of course, imply that infants are


unable to associate a novel symbol with an objectcategory in the absence of referential cues by way ofassociative learning. However, the results suggestthat infants more readily learn the association inreferential (naming) contexts and that it is notsimply a stimulus–response pairing but rather asymbolic and referential mapping for both wordsand nonverbal sounds.

Although 13-month-olds do not show any evi-dence of learning either symbol in the nonreferentialcondition and do show evidence of learning bothsymbols in the referential condition, it is clear thatthe context effect is weaker for infants at this agethan at 18 months. There are several possible reasonswe might expect to find weaker results in such a taskwith these younger infants. First, we might expectany symbol-learning task to be more challenging forchildren at such a young age who are on the cusp ofword learning. Second, this forced-choice paradigm(requiring children to demonstrate their understand-ing of the symbol by choosing the target objectduring the target trials) is a particularly demandingtask requiring significant attention and memoryresources. Although looking-time measures havebeen used successfully with children this age, wefelt that a more stringent criterion of whether asymbolic mapping has been made, even if it resultsin less consistent performances, was an importantcomponent of the study. Although a small number ofstudies have successfully elicited symbolic mappingusing a forced-choice paradigm with this age (inparticular, see Woodward & Hoyne, 1999; Wood-ward et al., 1994), the dynamics of our tasks may nothave been as well suited to this younger populationas to the 18-month-olds.

The results of this study conflict with Balaban andWaxman’s (1997) and Xu’s (1998) findings. Inparticular, the infants in this study failed to mapwords to objects in the nonreferential condition. Amethodological difference between this study andthose of Balaban and Waxman and of Xu mayaccount for the discrepant findings. In particular, thetesting measures were distinct. The aforementionedstudies recorded looking times of infants and usedthis information to infer their understanding ofwords and sounds. The current study required thatthe infants actively map the symbol onto the referentand to demonstrate this knowledge by reliablychoosing the target object in target trials. Althoughthe looking-time measure reflects infants’ decreasedattention to category instances and increasedattention to novel objects as a function of exposureto a label, this measure may not accurately reflectthe extent to which the infants understand the

symbolic relationship between the word and theobject.

The use of a habituation method may also accountfor the different performance in the word versussound condition in these other studies. It may be thatinfants’ attention is more readily directed to objectsby words than by nonverbal sounds, given thegreater history of exposure to words as referentialsymbols. However, the familiar naming context inwhich the sounds were embedded in our referentialcondition was apparently sufficient to override anyimplicit preference for words as symbols. Suchsocial-referential cues to naming were absent inBalaban and Waxman’s (1997) and Xu’s (1998)studies. The degree of discrepancy between thepatterns of findings reported for looking-time versuschoice measures suggests that the two methods maybe tapping different processes.

Implication for Our Model of Word Learning

The finding that infants learn both words andnonverbal sounds in similar contexts is consistentwith the findings of Roberts (1995; Roberts & Jacob,1991), Namy and Waxman (1998; Namy, 2001), andWoodward and Hoyne (1999). These results suggestthat, initially, infants are willing to accept a variety ofsymbol forms as object names when these symbolsare presented in the appropriate social-referentialcontext. It is important to note that not only didinfants succeed at learning both the word and thesound in the referential condition, they showedsimilarly robust effect sizes in the two symbol-learning tasks. In other words, infants appeared tointerpret words and sounds as equipotential sym-bolic forms. This suggests that the mechanismsdriving word learning are not specific to words butrather imply a general symbolic ability early indevelopment.

Of course, there are limits to the degree to whichone can infer common process from commonbehavior. However, evidence from a wide range ofmeasures and methodologies including naturalisticproduction (Acredolo & Goodwyn, 1985, 1988;Goodwyn & Acredolo, 1993; Iverson et al., 1994),experimental comprehension studies (Hollich et al.,2000; Namy, 2001; Namy & Waxman, 1998; Wood-ward and Hoyne, 1999), and experimental produc-tion (Namy & Waxman, 2002) converge on theconclusion that words and nonverbal symbols areequipotential forms of symbolic reference in infants.That infants successfully learn multiple symbolicforms under the same learning conditions and,perhaps even more telling, fail to learn multiple


symbolic forms under the same learning conditions(as the current experiment shows) lends evenstronger support to this conclusion.

Implications for Our Understanding of Infants’ Social-Referential Sensitivity

The findings in this study replicate Baldwinet al.’s finding that infants fail to interpret a labelas an object name when it is produced by a sourcethat does not embed the label in a social-referentialroutine. This implies a fairly subtle appreciation ofhow social-referential cues signal naming events,because in both conditions the social-referential cueswere present during labeling, and the label wasproduced concurrent with infants’ attention to thesame object toward which the referential cues areoriented. Thus, infants understand not only thatsocial-referential cues were reliable indices of nam-ing but also that the social-referential cues mustoriginate from the same source as the label.

In one sense, the manipulation between thereferential and nonreferential conditions was subtle,in that the only difference was the source and thetiming of the symbol. However, in another sense, themanipulation could be seen as obvious. In onecondition, the symbol was presented by the personwith whom the child was interacting. In the othercondition, the symbol originated from a mechanicalobject placed to the side of the infant that seemedunrelated to the ongoing interaction. In future work,it will be important to explore whether it is possibleto elicit symbolic mappings in the absence of social-referential cues, given sufficient alternative indica-tors that the symbols are being produced with theintention of labeling. Put differently, it may be thatsocial-referential cues are sufficient but not neces-sary to elicit symbolic mapping, provided that otherreliable cues to intentional naming are present.Indeed, studies by Balaban and Waxman (1997),Schafer and Plunkett (1998), and Werker, Cohen,Lloyd, Casasola, and Stager (1998) have demon-strated, using looking-time paradigms, that childrenreadily associate novel words with their referentseven in the absence of social-referential cues.However, it is unclear from these passive lookingparadigms whether these associations are symbolicin nature. We propose that the appreciation thatsocial-referential behaviors signal naming is ac-quired because social-referential behaviors emergeas a reliable set of cues to reference over time. Thisimplies that infants should be able to use alternativecues to reference when those cues are established asreliable indicators as well. A test of this hypothesis

will provide important insight into the origins andacquisition process of social-referential understand-ing.

Conclusions

The results from the current study provide severalkey insights into the symbol-acquisition process.First, this study joins a growing body of researchindicating that infants’ early word-learning ability isnot driven by a word-specific mechanism but ratherby a general symbolic ability. Infants, at least untilthe age of 18 months, accept a variety of symbolicforms as object names. Second, this study confirmsthat a driving factor behind whether infants willlearn a symbol as an object name is the context inwhich the symbol is presented. Infants showimpressive sensitivity to subtle referential cues andreveal the ability to use this information to makeappropriate symbolic mappings as well as avoidmaking inappropriate symbolic mappings. Resultsfrom this study suggest that the mechanisms drivingearly symbol learning are those that take intoaccount the social and referential nature of symbolsand are not initially limited to a particular symbolicmedium.

References

Acredolo, L., & Goodwyn, S. (1985). Symbolic gesturing inlanguage development: A case study. Human Develop-ment, 28, 40–49.

Acredolo, L., & Goodwyn, S. (1988). Symbolic gesturing innormal infants. Child Development, 59, 450–466.

Balaban, M., & Waxman, S. R. (1997). Do words facilitateobject categorization in 9-month-old infants? Journal ofExperimental Child Psychology, 64, 3–26.

Baldwin, D. A. (1991). Infants’ contribution to theachievement of joint reference. Child Development, 62,875–890.

Baldwin, D. A. (1993). Infant’s ability to consult thespeaker for clues to word reference. Journal of ChildLanguage, 20, 395–418.

Baldwin, D. A., Markman, E. M., Bill, B., Desjaralins, R. N.,Irwin, J. M., & Tidball, G. (1996). Infants’ reliance on asocial criterion for establishing word–object relations.Child Development, 67, 3135–3153.

Bates, E., Benigni, L., Bretherton, I., Camaioni, L., &Volterra, V. (1979). The emergence of symbols: Cognitionand communication in infancy. New York: AcademicPress.

Bloom, L. (1993). The transition to language: Acquiring thepower of expression. New York: Cambridge UniversityPress.

Cohen, J. (1977). Statistical power analysis for the behavioralsciences (Rev. ed.). New York: Academic Press.


Ecklund-Flores, L., & Turkewitz, G. (1996). Asymmetricheadturning to speech and nonspeech in human new-borns. Developmental Psychobiology, 29, 205–217.

Folven, R. J., & Bonvillian, J. D. (1991). The transition fromnonreferential to referential language in children acquir-ing American Sign Language. Developmental Psychology,27, 806–816.

Fulkerson, A. L., Shull, S. S., & Haaf, R. A. (2002, April).Sensitivity to the source of auditory input mediates thefacilitation of labels on infants’ categorization: A develop-mental trend. Poster presented at the InternationalConference on Infant Studies, Toronto, Canada.

Goodwyn, S., & Acredolo, L. (1993). Symbolic gestureversus word: Is there a modality advantage for the onsetof symbol use? Child Development, 64, 688–701.

Hollich, G. J., Hirsh-Pasek, K., & Golinkoff, R. M. (2000).Breaking the language barrier: An emergentist coalitionmodel for the origins of word learning. Monographs of theSociety for Research in Child Development, 65 (3, Serial No.262).

Iverson, J. M., Capirci, O., & Caselli, M. C. (1994). Fromcommunication to language in two modalities. CognitiveDevelopment, 9, 23–43.

Meier, R. P., & Willerman, R. (1995). Prelinguistic gesture indeaf and hearing infants. In K. Emmorey & J. S. Reilly(Eds.), Language, gesture, and space (pp. 391–409). Hills-dale, NJ: Erlbaum.

Namy, L. L. (2001). What’s in a name when it isn’t a word?17-month-olds’ mapping of non-verbal symbols toobject categories. Infancy, 2, 73–86.

Namy, L. L., & Waxman, S. R. (1998). Words and symbolicgestures: Infants’ interpretations of different forms ofsymbolic reference. Child Development, 69, 295–308.

Namy, L. L., & Waxman, S. R. (2000). Naming andexclaiming: Infants’ sensitivity to naming contexts.Journal of Cognition and Development, 1, 405–428.

Namy, L. L., & Waxman, S. R. (2002). Patterns ofspontaneous production of novel words and gestures

within an experimental setting in children ages 1;6 and2;2. Journal of Child Language, 29, 911–921.

Roberts, K. (1995). Categorical responding in 15-month-olds: Influence of the noun-category bias and thecovariation between visual fixation and auditory input.Cognitive Development, 10, 21–41.

Roberts, K., & Jacob, M. (1991). Linguistic vs. attentionalinfluences on nonlinguistic categorization in 15-month-olds. Cognitive Development, 6, 355–375.

Schafer, G., & Plunkett, K. (1998). Rapid word learning byfifteen-month-olds under tightly controlled conditions.Child Development, 69, 309–320.

Tomasello, M., & Barton, M. (1994). Learning words innonostensive contexts. Developmental Psychology, 30,639–650.

Tomasello, M., Strosberg, R., & Akhtar, N. (1996). Eighteen-month-old children learn words in non-ostensive con-texts. Journal of Child Language, 23, 157–176.

Volterra, V., & Iverson, J. M. (1995). When do modalityfactors affect the course of language acquisition? In K.Emmorey & J. S. Reilly (Eds.), Language, gesture, andspace (pp. 371–390). Hillsdale, NJ: Erlbaum.

Werker, J. F., Cohen, L. B., Lloyd, V. L., Casasola, M., &Stager, C. L. (1998). Acquisition of word–object associa-tions by 14-month-old infants. Developmental Psychology,34, 1289–1309.

Woodward, A. L., & Hoyne, K. L. (1999). Infants’ learningabout words and sounds in relation to objects. ChildDevelopment, 70, 65–77.

Woodward, A. L., Markman, E. M., & Fitzsimmons, C. M.(1994). Rapid word learning in 13- and 18-month-olds.Developmental Psychology, 30, 553–566.

Xu, F. (1998). Distinct labels may provide pointers to distinctsortals in nine-month-old infants. Paper presented at theProceedings of the 22nd Annual Boston UniversityConference on Language Development, Somerville,MA.


Documents

The Role of Social-Referential Context in Verbal and Nonverbal Symbol Learning