Pictures and Words 1 Do Both Pictures and Words Function as Symbols for 18 and 24-month-old Children?

In Experiment 1, 24-month-old toddlers were taught a new word (“whisk”) through the labeling of a picture of a whisk. After repeated pairings of the word and picture, participants were shown the picture and a real whisk and asked to indicate the whisk. They took the word to refer to the real object rather than to the picture. Experiment 2 established that children are not biased to select any novel real object in the test trial. Rather, the results from Experiment 1 reflected the child’s interpretation of the word as referring to the pictured kind. A third study confirmed that a novelty preference within a perceptually specified category could not account for the results of Experiment 1. A final study (Experiment 4) examined whether 18-month old infants also understand pictures and words as symbols, and results were comparable to Experiments 1 and 2. Taken together, these results confirm that the mapping between words and objects for 18 and 24-month-olds is a referential relation, as opposed to an associative one. Furthermore, these results show that children as young as 18-months begin to understand the symbolic nature of pictures. Pictures and Words 3 Do Both Pictures and Words Function as Symbols for 18 and 24-Month-Old Children? In adult human mental life, both pictures and words function as symbols. Many words and pictures stand in a representational relation to entities in the world; these symbols have referential content. The word “dog” refers to dogs in general, or to a particular dog (as in, “my dog, Domino”). Similarly, a picture of a dog may represent dogs in general, or a particular dog. The referential relation is fundamentally intentional; the intentions of a symbol’s maker or user determine its content (see Bloom & Markson, 1998). In discerning the nature of representational understanding, we must consider whether every creature for whom there is a mapping between a word/picture and some set of entities in the world grasps the symbolic nature of the word/picture. This question is trenchant in cases of lexical representations in non-human primates; see the debate between Seidenberg and Petitto (1987), on the one hand, and Savage-Rumbaugh and colleagues (Savage-Rumbaugh, 1982; Savage-Rumbaugh & Brakke, 1986; Savage-Rumbaugh, McDonald, Sevcik, Hopkins, & Rubert, 1986; Savage-Rumbaugh, Shanker & Taylor, 1998) on the other. It also arises when considering the nature of the mapping between pictures and objects, and between words and objects, constructed by low functioning children with autism who have been taught to use pictures of objects as means of requesting objects in the world (Bondy & Frost, 1998; Lancioni, 1984). The question also arises in the case of young infants. An alternative interpretation of the mapping between words/pictures and their referents is that it is merely associational (e.g. Plunkett, 1997; Smith, Jones & Landau, 1996). Associative mappings may be created between any arbitrary stimuli, as when a tone signals shock, or predicts reward following a bar press. Associative mappings reflect frequency and temporal contiguity of pairings. They are created by classical laws of association (see Rescorla & Wagner, 1972) or by Pictures and Words 4 mechanisms that determine causally relevant contingencies (see Gallistel, 1990; Heyes & Dickinson, 1990). When a bee learns that red flowers with a certain odor contain sweet nectar, we do not think that the bee considers the flower to be a symbol of nectar; rather we consider the bee to have learned a predictor of nectar. In terms of word learning, an associative account of the mapping between words and objects in the world holds that it is established through sensitivity to statistical co-variation. On this account, words are non-referential and non-symbolic. Children learn words such as “car” through repeated pairing of the verbal label with experience of cars (e.g. Richards and Goldfarb, 1986). Connectionist treatments of word learning, drawing on statistical relations between words and objects in the environment, are in this tradition (Plunkett, 1997). Indeed, children’s initial word learning can certainly be seen as consistent with the associative view; their first words are often things one can see and information provided by parents tends to be overt (i.e. picking up a cup and repeatedly labeling the item for the child). On the surface, the input conditions for word learning would seem to be consistent with associationist principles. However, deeper analyses question this assumption (Bloom, 2000; Harris, Jones & Grant, 1983) and experimental evidence also undermines this view. There are many empirical studies that suggest infants as young as 18 months understand that words are representations with referential content, for toddlers of this age seek evidence from the word’s introducer as to his or her referential intent. Baldwin (1991, 1993b) conducted a series of experiments in which 18-19-month-old children used the gaze of an adult to map a novel label to an item. One condition explicitly contrasted the symbolic and associationist point of views. The experimenter waited until the infant was attending to an unfamiliar stimuli and unnamed object, and then exclaimed, “Look, it’s a modi.” The experimenter, however, was Pictures and Words 5 looking in a bucket, at an unseen object. The children did not map the term “modi” to the item they themselves were looking at and playing with, but rather looked to see what the experimenter was staring at, and applied the word to the item within the bucket. Baron-Cohen, Baldwin & Crowson (1997) replicated these findings with 24-month-olds, and found that children with autism of the same mental age failed to use the speaker’s gaze in mapping a newly heard word onto an object. These results suggest that normally developing toddlers of this age know that a speaker’s referential intention provides evidence for an object label’s content. Mere associative pairing, at least for normally developing children, does not determine the mapping. Baldwin, Markman, Bill, Desjardins, Irwin, & Tidball (1996) provided additional evidence that the mapping between words and objects is not determined by the laws of association. Infants heard novel labels when they were investigating a single novel object. In one condition, the speaker was seated within the infant's view and displayed concurrent attention to the novel toy when verbalizing the novel label (coupled condition); in the other condition, the label was given by a speaker seated out of the infant's view (decoupled condition). Infants mapped the label to the object in the coupled condition, but not in the decoupled condition, despite the fact that co-variation between label and object was equivalent in both cases. Baldwin’s conclusions from such data have not gone unquestioned. Those who believe that the mapping between words and objects is associative might reply that joint attention is a condition on the child’s making the mapping. This may be so, but, for this reply to have some force, one would need an account of how this condition comes to be learned, through associative mechanisms. The present experiments approach the issue in a different way, seeking convergent evidence for Baldwin’s conclusions from a very different paradigm and a very different reflection of the distinction between associative mapping and referential mappings. Pictures and Words 6 From Baldwin’s work, we have evidence consistent with the claim that words are symbols for children, then, at least by 18 months of age (see also Tomasello, Strosberg & Akhtar, 1996). What about pictures? Pictures are significant stimuli in the conceptual life of young American infants, as they are frequently surrounded by books and photographs. Mature pictorial competence involves appreciating the representational nature of pictures (DeLoache, 1991). In spite of the fact that a realistic picture is perceptible and occupies its own space, its purpose is to depict something else in the real world. As adults, we have the knowledge that pictures refer, but do toddlers share this understanding? When we show a toddler a picture of an aardvark or a dump truck and label it “aardvark” or “dump truck” we assume she takes us to be providing information about the entities in the world that are so named. We assume that toddlers do not assume that these words refer only to the pictures, nor that the facts we tell them about aardvarks or dump trucks are facts about the pictures themselves. But are we right in this assumption? DeLoache & Burns (1993) term the awareness of the symbolic relation between the picture and what it stands for “representational insight”, and suggest that in at least one context in which pictures represent a current situation children do not show representational insight into pictures until 2 _ years of age (DeLoache, 1987; DeLoache, 1989; DeLoache, 1991). The question of when infants understand pictures as representations must be distinguished from two other questions about picture perception. First, when do children recognize the perceptual similarity between pictures and the real items they refer to? Second, when do children recognize the differences between pictures and real objects that are consequences of the 2D nature of pictures? With respect to the first question, habituation studies show that very young infants (5-month-olds) appreciate the perceptual similarity between Pictures and Words 7 pictures and objects. For example, after being habituated to a face (Dirks & Gibson, 1977) or real object (DeLoache, Strauss, & Maynard, 1979), 5-month-old infants generalized habituation to pictures of the same face or object, and dishabituated to pictures of novel faces or objects. Hochberg and Brooks (1962) showed that prior experience with pictures is not necessary for infants to recognize the similarity between 2-D representations and 3-D objects. They meticulously kept their participant away from pictured images until 19 months of age. He had no trouble identifying pictured objects when finally presented with picture books. These studies illustrate that children of 5to 19-months of age perceive pictures and real objects in ways that capture perceptual similarity, but they do not examine whether infants understand pictures as representations. With respect to the second question, it appears that infants fail to fully appreciate the consequences of the 2D nature of pictures until the middle of the second year of life, although they can discriminate between pictures and real objects (DeLoache, Strauss & Maynard, 1979; Slater, Rose & Morison, 1984). In a study of infants looking at picture books consisting of one small photograph per page, DeLoache, Pierroutsakos, Uttal, Rosengren, & Gottlieb (1998) showed that nine-month old infants reach and explore photographs, as if they were trying to pick up the items depicted in the pictures. Even infants from a non-literate West African community, the Beng of Côte d’Ivoire, displayed the same pattern, showing that this is not a culturally specific phenomenon. There have been many informal accounts of similar behavior (Murphy, 1978, Beilin & Pearlman, 1991). Perner (1991) reported that his son, aged 16 months, tried to step into a picture of a shoe, and Ninio and Bruner (1978) described an 8-month-old trying to pick up pictures in a child’s book. In Ninio and Bruner’s report, as in DeLoache et al., this behavior ceased by ages 18-19 months, by which age infants apparently distinguish pictures from Pictures and Words 8 real objects with respect to appearance vs. reality. This distinction is necessary but not sufficient for understanding pictures as representations. The earliest age at which an understanding of pictures as representations has been demonstrated is 30 months by DeLoache & Burns (1994). They presented toddlers with a photograph of a room and indicated on the photo where an object was hidden (e.g. pointing to a couch and stating “it’s under here.”) Although 24-month-old children were unable to utilize the information in the photo to locate the object in the real room, 30-month-olds succeeded at this task. DeLoache and Burns take this as evidence that children aged 30 months understand the relationship between the picture and its referent, specifically that the picture refers to the real room, whereas younger children do not. Could it really be true that children do not understand that pictures are symbols until 2.5 years of age, a full year after they understand words as symbols? Perhaps this failure is restricted to the unusual case in which a picture represents a specific current situation. The hypothesis that the failure of 24-month-olds in the DeLoache and Burns study reflects a general lack of mature pictorial competence has consequences for how we understand young toddlers comprehension of picture books. If they do not take pictures as symbols for real objects in the world, then how do they understand the adult practice of naming pictures, “that’s a giraffe”, and providing facts about giraffes, “giraffes live in Africa?” Do young toddlers think that the labels and facts map onto the pictures, and perhaps other entities perceptually similar to the pictures? Or do they understand that both the word and picture refer to real entities in the world? Experiment 1 explores this question. In Experiment 1 an experimenter taught toddlers a novel word (“whisk”) for a novel object depicted only in a picture. During the teaching phase of the study, the word was Pictures and Words 9 repeatedly paired with the picture. The experimenter then offered the children a choice between the picture and a real whisk (previously unseen) and asked the children to show her a whisk. If the mapping between a word and the stimulus it is paired with is associative, we would expect the children to pick the picture alone, or both the picture and the real object, because of generalization based on perceptual similarity. If, however, the children understand that words and pictures are symbols for real world entities, then they should map the word “whisk” onto a previously unseen object, and should choose the real whisk alone or both the picture and the real object, because pictures are often labeled elliptically. We say, pointing to a picture, “that’s an elephant”, not “that’s a picture of an elephant.” Of course, the selection of both the picture and real object could support either hypothesis. If children choose the real whisk under these circumstances, various controls are needed to rule out the possibility that real objects are so much more salient than pictures that 2-year-olds would choose a real object over a picture no matter what was asked. Therefore, in the Real Item Bias Control Phase of Experiment 1, participants were shown a series of familiar entities, one a real object (e.g., a cup) and one a picture (e.g. a picture of a flower). They were asked to show the experimenter one of the entities; half of the time the pictured entity was requested and half the time the real object was requested. Also, to ensure that children of this age would accept both a real object and a picture of an object as correct choice for “show me an x”, in the Picture Choice Control children were shown pairs of familiar objects, a real object and a picture of that object (e.g. a spoon and a picture of a spoon). They were asked to show the experimenter a spoon. Additional controls are introduced in Experiments 2 and 3, and Experiment 4 examines whether 18-month-old children also understand pictures as representations. Pictures and Words 10 Thus, this experiment had two goals. First, we sought evidence from a new method that would converge with results in the literature which indicate that young children understand words as symbols, and that the mapping between words and objects is a referential, meaningdetermining relation, as opposed to an associative mapping. Second, we sought evidence that children younger than 30-month-olds understand the symbolic nature of pictures. Experiment 1 Method Participants. Twenty normally developing, native English-speaking children (mean age 24.02 months, range 20.28 – 29.02 months) were included in the study. There were 10 males and 10 females. Five additional participants were excluded due to fussiness, and 3 additional participants were excluded because they knew the “novel” item to be taught in the experiment. Participants were recruited through the New York University Infant Cognition Center infant database. Parents were reimbursed up to $10 for travel expenses, and children received a t-shirt or small toy for participation. Stimuli. The stimuli used throughout the experiment were real-world (3-D) items, including some toy models, and 2” X 2” laminated black and white line drawings modeled after Mayer-Johnson stimuli (picture icons utilized within a symbolic system of communication for individuals with autism, www.mayerjohnson.com). Procedure One or both of the child’s parents were present throughout testing. Participants sat next to or in the lap of a parent, separated from the experimenter by a 24-inch wide table. Items were placed within the participants’ reach. Each session was videotaped, with parental consent. Pictures and Words 11 Pre-Training Phase. Participants were taught a new word paired with a picture. We first needed to demonstrate that the word was indeed novel to the participants. Participants were shown one 8 _” x 11” page with seven familiar pictures of foods; they were asked to identify a highly familiar object (apple) from the display, to ensure that they could discriminate from a pool of seven items. Participants were then shown an 8 _” x 11” page with seven unfamiliar pictures of tools/utensils; they were asked to identify an unfamiliar object (whisk) from the display. Participants were expected to fail on this trial, showing they do not know the word whisk refers to the pictured object, thereby qualifying it as novel. Parents were also asked whether their child knew the word before the session began. Training Phase. The participants were then taught the new word ‘whisk’, mapped to the picture of a whisk. The participants were presented with the picture of the whisk in isolation and were told, “This is a whisk. Can you touch the whisk?” All participants complied with this instruction. The whisk picture was then presented to the participants with an apple picture. The participants were instructed, “Can you show me a whisk?” This was repeated until the participants correctly identified the whisk (by touching the picture or handing it to the experimenter) on three consecutive trials, randomized for side of presentation, and with the same instruction to the participants. This teaching procedure is used to teach children with autism novel symbols, in an associative fashion (Bondy & Frost, 1998). If participants failed to correctly discriminate the whisk on consecutive trials, the training procedure was re-implemented from the initial step (presenting the picture in isolation). The participants were then shown one 8 _” x 11” page with five pictures (four unfamiliar pictures and the whisk picture) and were asked, “Can you show me a whisk?” Positive verbal feedback was given for each trial. This trial was performed to ensure that the participants had Pictures and Words 12 indeed learned the new word, and could discriminate between the whisk picture and other pictures not used in the teaching procedure. Test Phase. This was the crucial phase of the experiment, in which participants were presented with the whisk picture and a real whisk. If the participants learned the word through a paired association, without understanding the symbolic role of both words and pictures, they should choose the picture of the whisk when questioned by the experimenter, or they should choose both items due to perceptual similarity between the real whisk and the picture. The word “whisk” had been paired with the picture for a minimum of five mappings (and never paired, or associated, with the real item). If, however, the children understand that both words and pictures refer, they should always include the real whisk in their choice. A choice of both items would also be consistent with referential understanding, since words refer both to pictures and objects. The participants were presented with the whisk picture and a real 3-D whisk (Figure 1) and were instructed, “Can you show me a whisk?” The measure of interest was the response of the child (which item the participant pointed to or gave to the experimenter). Responses were coded as the child’s individual touches of picture and/or real whisk. Only intentional responses were coded (giving a picture or object to experimenter, sliding item to experimenter, pointing to item, or picking up and showing to experimenter with eye contact). If the child played with or explored a picture or object without clearly indicating it a response, this was noted but not included in the final coding. For instance, if a child indicated that an object was correct (by pointing or giving to experimenter) and then merely played with the picture, this would be coded as a “real object alone” response. Conversely, if the child indicated the picture (by pointing or giving it to the experimenter) and then merely played with the object, this would be coded “picture alone” response. The children’s behavior may provide clues about why they respond to Pictures and Words 13 an item, as pointing or handing over an item is clearly different from manipulating without referential intent. Two coders independently coded the videotapes. Agreement on response classification was 92%. Disagreements were settled by discussion. [INSERT FIGURE 1 ABOUT HERE] Real Item Bias Control. This phase of the experiment required participants to choose an item requested by the experimenter, either a picture or real object. This procedure was performed to ensure that participants could choose an item at the request of the experimenter, whether it was a picture or real item, and would not simply be drawn to the real, potentially more salient, item. Participants were presented with one real-world item (for example, book) and one picture (for example, apple) and were instructed to “Show me a/an X” (Figure 2). There were ten trials of this type. Order and side of presentation were randomized. The measure of interest was the intentional response of each child (which items the child pointed to, picked up and showed or gave to the experimenter). If the child did not respond after 5 verbal prompts by the experimenter (or approximately one minute), the next trial was introduced, and ‘no response’ was recorded. [INSERT FIGURE 2 ABOUT HERE] Picture Choice Control. Participants were presented with real items and pictures of those items, e.g. picture of spoon and real spoon and were instructed to show the experimenter a spoon. There were four trials of this type (spoon, dog, button, car). Order and side of presentation were randomized. The measure of interest was the response of child (which items the child pointed to or gave to the experimenter). Responses were coded as the child’s individual touches of picture and/or real item. If the child did not respond after five verbal prompts by the experimenter (or approximately one minute), the next trial was introduced, and ‘no response’ was recorded. The Pictures and Words 14 goal of this control phase was to determine which item (picture, object, or both) the children pair with a known label for familiar objects. Specifically, this phase will see if, like adults, children accept words as referring both to real and depicted objects. There were two types of stimuli used in this phase: real items (button and spoon) and models (stuffed dog and toy car, which are individual objects but still representations of other objects, namely, living dog and real car). These two types of stimuli were selected to examine if there was a difference between real, functional, referents of a familiar word (e.g., spoon) and models that are, after all, representations of other items (e.g., toy car). In other words, we wished to determine whether the children interpreted the real items as truer referents than models. Order of Procedures. The Real Item Bias Control phase was administered first (to give children exposure to both pictures and objects as correct responses), followed by the Picture Choice Control phase and subsequently the Pre-Training, Training and Testing Phases of the Experiment. Results will be reported in the order as presented in the Procedures section. Results The responses we report are intentional acts of indicating the requested items (pointing, showing, giving to the experimenter). Non-referential exploration or playing with an item was not included. Pre-Training Phase. For the pre-trials of the novel word training procedure, all participants correctly discriminated the apple picture from among a seven-icon sample, showing that they could pick a named entity from an array of seven. One parent reported knowledge of the word by their child, and two additional children chose the whisk, and therefore three children were excluded from the study. Twenty participants failed to choose the whisk, which confirmed the word as novel for these participants. Pictures and Words 15 Training Phase. All participants were able to discriminate the whisk picture from the apple picture for three consecutive trials, and to successfully select the whisk from an array of five items, ensuring that they indeed learned the new word. There was an average number of 5.2 pairings between the verbal label and picture. Sixteen participants required the minimum number (5 pairings); three children required 6 pairings and 1 child required 7 pairings. Test Phase. In the last (test) trial, no participants (0%) chose the picture alone, 11 (55%) chose the real item alone, and 9 (45%) chose both items. One child who indicated the real object also explored the picture, but this was not a clear intentional response, hence the trial was coded as “real alone”. All other responses were intentional. Of the 9 children who selected both items, 5 chose the real object first and 4 indicated the picture first. Thus, 16 of 20 children indicated the object first or exclusively. They indicated it by pointing to it or handing/sliding it to the experimenter. There was a significant difference between picture and real item choice (p<.01, 2tailed paired t-test); All children (100%) included the real item in their choice, whereas only 45% included the picture in their choice. These results suggest that although the children were taught the new word in a manner consistent with associative learning (verbal label ‘whisk’ paired repeatedly with the picture), they took the word to refer to a real whisk rather than the picture. The children had no prior experience with the real whisk, and yet all selected the real item when asked to indicate a whisk. None chose the picture alone, in spite of the teaching experience. This suggests that they take pictures to serve as representations of real objects. Some of the verbal responses accompanying the trials were also quite revealing. Seven children made spontaneous verbal responses. Of these, five children labeled the real item “whisk”. One additional child who selected both the picture and real whisk held up the real Pictures and Words 16 whisk and commented, “This is a whisk”, and subsequently presenting the picture added “And this is a picture of a whisk!” Another child picked up the real whisk and made a declarative statement (“ta-da!”) when asked to indicate a whisk. There were also some informal comments at the end of the trial; when questioned, “What is that?” by the experimenter, pointing to the whisk, a child holding the real item pointed to the picture and said, “That’s a picture of it.” These comments provide further suggestive evidence that the children know that a picture refers to an object in the real world. Real Item Bias Control. Here, children were presented with 10 trials consisting of one real, familiar object and one picture of a familiar entity each and were asked to show the experimenter one of the items. The overall accuracy for the real item bias control phase was 96% correct, with an average of 97% accuracy for trials in which the real item was correct, and 95% accuracy for trials in which the picture was correct. There was no significant difference between trial types, and there were no item effects. These results show that children of this age are quite good at the item request task, and have no difficulty with selecting either the picture or the real item. Notably, the children were not more preoccupied with a real, potentially more interesting, familiar object. Children accept a picture of an apple as a referent for “apple”, at least when the only other choice is a real book. Conversely, the participants correctly accepted a real item (such as a cup) as a referent for “cup” when provided with a picture of a flower as an alternative choice. Picture Choice Control. In this control, children were presented with 4 trials consisting of one real, familiar object and one picture denoting the respective object, and were asked, for example, to show the experimenter “a spoon”. There was no response on 1 (1.3%) of the trials. Of the 79 remaining trials, summed across all children, children selected both the picture and the Pictures and Words 17 object on 57% of the trials, the object alone on 40.5% of the trials, and the picture alone on only 2.5% of the trials (Figure 3). There was a significant difference between picture and real item choice (p<.05, 2-tailed paired t-test). There was no significant difference between responses for real items and toy models, showing that participants chose the 3-D item at the same rate whether it was a real object, such as spoon, or a model that represents another entity, such as a toy car. [INSERT FIGURE 3 ABOUT HERE] These results demonstrate that children know that a real (3-D) item is the primary referent of a familiar word, such as “spoon.” The children virtually never choose the picture alone, but in 40.5% of the trials they selected the real item alone. However, these results also demonstrate that children accept that the word “apple” applies to the depicted entity, as evidenced by the high proportion of “both” responses within trials. Adults use language in this manner – when looking through a picture book with a child, we often point to a picture of a monkey and say, “that’s a monkey”, not “that is a picture of a monkey.” Experiment 1 provides convergent evidence for Baldwin’s (1991, 1993b) conclusion that young language learners appreciate the symbolic nature of words. The process of word learning involves meaning assignment; it is not merely a matter of association. If word learning were simply a process of making an associative mapping, then children whose only experience with a word was repeated pairing with a specific picture would be expected to take that picture as the central stimulus mapped to that word. This was not the pattern observed here. The toddlers never chose the picture alone when asked to indicate a whisk. Half the time they picked the real object alone and half the time they picked both the picture and the object. Rather, the pattern observed is that expected if the children understood the picture as a symbol for a real world object, and Pictures and Words 18 understood also that the word referred to that object. The results from the Picture Choice Control confirmed that children consider real objects canonical referents for words, but that, as for adults, it is permissible to directly name a picture with the label of the object pictured. This could explain why they indicated the picture as well as the real whisk when asked to show the experimenter a whisk. The results from the Real Item Bias Control show that children of this age are perfectly capable of ignoring a real, familiar object in favor of a picture under these circumstances. Thus, the findings from the Test trial are not due to overwhelming salience of the real whisk relative to a picture. Additionally, the way in which children interacted with the real whisk, namely pointing or giving it to the experimenter and not merely playing with it, suggests children viewed it as the primary referent of the word. However, perhaps when the label is a newly learned word, the mapping is fragile and a preference for real objects over pictures determines the response. This real item preference explanation for the choice of the whisk predicts that children taught the word “whisk” mapped to a picture of a whisk would choose a real item over the picture on the test trials irrespective of the identity of that item—whether it is a whisk or not. Experiment 2 explores this possibility. Experiment 2 This experiment was performed to ensure that children were not drawn to the real whisk in the test phase of Experiment 1 simply because it was a real object. After all, children were presented with a picture for many consecutive trials in the teaching phase of Experiment 1. Perhaps any novel real object would be extremely salient under these circumstances. Children may be biased towards the real object, no matter what its identity, and irregardless of the experimenter’s question. To address this question, we performed the same experiment, this time Pictures and Words 19 adding a single control trial before the test phase that offered children a choice between the whisk picture and a new novel object, a garbage disposal crusher. Children were asked to show the experimenter a whisk. The control phases, training, and final test trial of the experiment were identical to Experiment 1. Method Participants. Twenty normally developing, native English-speaking children (mean age 24 months, range 22-26 months) participated in the study. There were 10 males and 10 females. Two additional participants were excluded due to fussiness, and two others knew the word “whisk”. Participants were recruited through the Harvard University Laboratory for Developmental Studies database. Parents were reimbursed up to $5 for travel expenses, and children received a t-shirt or small toy for participation. Stimuli. The stimuli for the Real Item Bias Control, Picture Choice Control, PreTraining, Training, and Test Phases were identical to those used in the same phases of Experiment 1. Additionally, a garbage disposal crusher was used as the novel object for the Real Object Preference Probe (Figure 1). Procedure The procedure was identical to Experiment 1 with a single exception: after the Training Phase (word “whisk” paired with novel picture of whisk), children were presented with a Real Item Preference Probe. Children were then given the Test Trial of Experiment 1. Real Item Preference Probe. Participants were presented with the whisk picture and a novel object (garbage disposal crusher) and asked to show the experimenter a whisk. This control trial seeks to establish whether children are simply drawn to new objects in the test phase. If the fact that a novel object is simply more salient or interesting than a previously seen Pictures and Words 20 picture accounts for the choice of the whisk in Experiment 1, then children should select the object in this trial. If, however, children have mapped the word “whisk” onto the kind whisk, they should reject the garbage disposal crusher in favor of the pictured whisk on this trial. Performance on the test trial that follows will then bear, as in Experiment 1, on whether this mapping is associative or referential. Results Pre-Training Phase. All participants correctly discriminated an apple from a seven-object sample, showing that they could pick one picture from an array including six distracters. All participants failed to choose "whisk", ensuring there was no prior knowledge of the word whisk, which was also verified by parental report. Training. All participants were able to discriminate the whisk picture from apple picture for three consecutive trials, and to successfully select the whisk from an array of five items, to ensure they indeed learned the new word. There was an average number of 5.2 pairings between the verbal label and object. Eighteen participants required the minimum number (5 pairings); one child required 6 pairings and one child required 7 pairings. Real Item Preference Probe. Three participants did not respond to this trial. One handled neither of the objects and two played with both without indicating either to the experimenter. Of the remaining 17 children, all selected the picture alone (100%). Twelve of these 17 children clearly indicated the picture, and then additionally explored the novel object. Thus, very few children followed the pattern we might expect if children’s behavior toward a novel object in this experimental setting is merely play or exploration. Most children clearly indicated the picture –pointing at and/or showing it to the experimenter. Later, some children also explored the object. This differential way in which they interacted with the items supports the conclusion that the Pictures and Words 21 children understood the word’s referent as the picture. These results therefore confirm that children are not simply selecting the real whisk in the test trial because it is a salient, real, object. Test Phase (picture of whisk/real whisk trial). In the test trial, one participant (5%) chose the whisk picture alone, 8 (40%) chose the real novel object alone, and 11 (55%) chose both items. One child who indicated the real whisk also explored the picture (non-intentionally), and the single child who indicated the picture also explored the real object (unable to determine if intentional). Of the 11 children who selected both items, 5 indicated the real object first and 6 selected the picture first. There was a significant difference between picture and real item choice (p<.01, 2-tailed paired t-test). See Figure 4 for the pattern of responses on the Test Trial and on the Real Item Preference Probe Trial. [INSERT FIGURE 4 ABOUT HERE] Real Item Bias Control. The results were identical to those of Experiment 1. The overall accuracy when asked to indicate a familiar object from two choices was 96% correct, with an average of 97% accuracy for trials in which the real item was correct, and 95% accuracy for trials in which the picture was correct. Picture Choice Control. These results were also virtually identical to the control phase of Experiment 1 (Figure 3). Shown a picture of a spoon and a real spoon and asked to indicate “a spoon,” toddlers take a 3D object as more canonical referent of a familiar word, and to a lesser extent use the word directly to refer to a pictured object. There was no significant difference between responses for real items and toy models (as in Experiment 1). Conclusions Experiment 2 shows that the salience and/or desirability of a real novel object cannot account for the choice of the whisk in the crucial (whisk/whisk) test trials of Experiments 1 and Pictures and Words 22 2. If our participants indicated the real whisk alone because it was a salient real object they had never seen it before, they should also have chosen the real garbage disposal crusher because it too was a salient real object they had not previously seen. Instead, children clearly indicated the whisk picture was the correct response on the Real Item Preference Probe trial. Another version of a novelty preference may account for the choice of the real item in the test (whisk/whisk) trial. Perhaps the word is indeed mapped associatively onto the picture, and the real object is included in the mapping due to stimulus generalization and not referential understanding. The children could be selecting the real object because it is a novel withincategory perceptual match, not because they understand the symbolic relationship between the two stimuli. If this were true, then if we were to teach the child a novel word for a novel object, repeatedly pairing the word with object, and then offering the same choice as in Experiment 1 (the object and a picture of the object), children should make the picture the primary choice, for now the picture would be relatively salient because of novelty. Experiment 3 explores this possibility. Experiment 3 To ensure that a within perceptual class novelty preference was not responsible for the "whisk" test phases of Experiments 1 and 2, children were taught a novel word applied to a novel object. The control phases of the experiment were identical to the first and second studies; the test phase of the experiment probed whether children, when taught a novel word for a novel object, would apply that word only to a picture of that object (indicating a within perceptual category novelty bias).