Syntactic Cues in the Acquisition of Collective Nouns Method

2. Method 2.1. Subjects The subjects were 16 4- and 5-year-old children from daycares in the Tucson area (10 girls and 6 boys; mean age = 5 years, О months, age range = 4 years, О months to 5 years, 8 months) and 16 University of Arizona undergraduates who received class credit for their participation. All subjects were native English speakers. 2.2. Stimuli For the pretest, there were two pairs of picture cards. The first pair consisted of а line-drawing of а tree and а line-drawing of five trees. The second pair consisted of а line-drawing of а single dog and а line-drawing of five dogs. Eight sets of cards were used in the main experiment. Each set contained three line-drawn pictures: а training picture and two test cards. The training picture showed а collection of five identical novel objects. One of the test cards depicted the same five objects (but in а different configuration) and the other test card showed а single object of the same kind standing alone. For half of the sets, the groups of objects were depicted as unconnected; for the other, they were depicted as physically connected Ьу means of а string looped around а part of each of the objects. Within both the connected and unconnected sets of objects, half were novel animals and the other half were novel machines. An example of а connected set is shown in Fig. 1. All training and testing cards were 12.5 х 12.5 cm. 2.3. Procedure Subjects were introduced to а puppet and told that they were going to look at some pictures with the puppet and that he would ask them some questions about the pictures. They were then told in а stage whisper Ьу the experimenter that the puppet was not very smart and they would have to tel1 him their answers slowly and carefully. (The identical protocol was used for both children and adults. Prior to testing, the adults were told that the study had been designed to Ье appropriate for young children and that parts of the study might seem silly.) Subjects were divided into two groups: half were in the plural count поип group and half were in the singular count поип group. Each subject received four trials in random order: а connected animal trial, an unconnected animal trial, а connected machine trial and an unconnected machine trial. АН of the words used were two-syllaЫe nonsense words: fendle, Ьlicket, kempu, and malto. Each subject heard each of the four labels once. Subjects were presented with one of two assignments of nonsense words to stimuli - the Fig. 1. Exarnple of stirnuli set used in Experirnent 1. nonsense words applied to the connected stimuli in one assignment were applied to the unconnected stimuli in the other. Half the subjects received one assignment, half the other.4 Subjects were tested individually while seated at а tаЫе with an ex­ perimenter. Children were tested in а quite area of their day care and adults were tested in а small testing room. 4 Due to experirnental error, there was an inadvertent relationship between assignrnent of syntactic condition and nonsense word-stirnuli assignrnent, such that rnore than half of the singular count group received one assignrnent and rnore than half of the plural count group received the other. Since there was no effect of the type of stirnuli (connected vs. unconnected) on subjects' responses, it is unlikely that this could have had any effect on the results reported below. Pretest In the pretest, subjects were shown а picture of а tree and а picture of five trees and asked "Сап you point to the forest?" They were then shown а picture of а dog and а picture of five dogs and asked "Сап you point to the dog?" This was included to estaЫish how subjects respond when asked about familiar object names and collective nouns. Main study After completion of the pretest, subjects were told "Now I am going to teach you and Big Bird [the puppet] some new words. I'd like you to help Big Bird learn what the words mean, all right?" А training card showing five objects was then placed in front of the subject. The experimenter waved her hand over the card (taking care not to point to any particular object inside the card) and presented the new word. Subjects in the plural count condition were told (using fendle as an example of the novel word) "These are fendles. There are fendles over here. Сап you tel1 Big Bird what we have over here?" Subjects in the singular count condition were told: "This is а fendle. There is а fendle over here. Сап you tel1 Big Bird what we have over here?" Once the subject repeated the novel word - either "fendles" (for plural count) or "fendle" /"а fendle" (for singular count)- the training card was removed and replaced Ьу the two test cards, one showing the same five objects in another configuration and the other showing а single object. The orientation of the cards (which one was to the right of the subject; which one was to the left) was random. Big Bird then asked "Сап you show me the fendle?" After а response the materials were cleared away. Тhis procedure was the same for all four trials and took approximately 10 minutes to complete. 3. Results and discussion 3.1. Pretest Out of 16 children, 11 children (69%) responded correctly to both parts of the pretest. Five children (31%) failed the pretest. Of the children who failed, 3 (19%) picked the single tree when asked to point to the forest and 2 (12%) picked the five dogs when asked to point to the dog. АН of the adults performed perfectly оп the pretest. It is interesting to note that there appears to Ье по systematic Ьias towards errors with the collective noun: forest did not appear to Ье significantly more difficult than the object name dog. А t-test comparing the ages of children passing the pretest with the ages of those who had failed found по significant age differences. 3.2. Main study For the main analyses, the dependent measure was the number of times that the subjects pointed to the collection of five objects when asked "Сап you point to the fendle?" Preliminary analyses found no effect of animacy for either adults or children. А 2 х 2 х 2 ANOVA was conducted with age (children vs. adults) and syntax (singular count vs. plural count) as between-subject factors and connectedness (connected vs. unconnected) as а within-subject factor. Fig. 2 shows the mean percentage of collective responses in each condition. There was а significant effect of syntax, with collective responses more frequent in the singular count noun condition than in the plural count condition (84% vs. 8%; F(l, 28) = 56.2, р < .005), but there was no effect of age or connectedness. There was also а significant Age х Syntax inter­ action (F(l, 28) = 5.3, р < .05), with adults showing more sensitivity than children to the syntactic manipulation. То explore the interaction, separate 2 х 2 ANOVAs were performed on the children and adults, with syntax as а between-subject factor and connectedness as а within-subject factor. Children's responses showed а % of collective responses 4- and 5-year-olds Adults Ages of subjects Fig. 2. Children and adult interpretations as а function of syntactic cues and connectedness in Experiment 1. □, Singular-connected; О, singular-unconnected, ■, plural-connected; е, plural-unconnected. main effect of syntax (F(l, 14) = 6.8, р < .05); those exposed to singular count noun syntax interpreted the words as collective nouns significantly more than those exposed to plural count noun syntax (68% vs. 15% ). Тhis indicates that syntax is sufficient, most of the time, to force one construal over another in the circumstances studied here. Further analyses found that the proportion of children in the plural count noun condition who pointed to the single object was significantly greater than chance (t(7) = -2.8, р < .05), but the proportion of children in the singular count noun condition who pointed to the collection was not. This could Ье because of the small sample size ( 8 children in the singular count noun group), but it could also Ье the result of developmental differences in sensitivity to singular count syntax and/or the capacity to construe а collection of objects as an individual. We return to these possibilities in the General Discussion. Although the children gave more collective responses with the connected stimuli than with the unconnected stimuli (48% vs. 35% ), this difference was not significant, and there was no interaction between syntax and connectedness. Adult performance was at ceiling with regard to sens1t1v1ty to syntax. Subjects in the single count noun condition understood all novels words as collective nouns and those in the plural count noun condition interpreted all words as referring to single objects. There was therefore no main effect of connectedness and no interaction. In sum, both children and adults are sensitive to syntax when acquiring object names and collective nouns, though adults are more sensitive than children. Contrary to the second prediction, however, the connectedness of the objects had no effect. EXPERIMENT 2 The second experiment was designed to extend the previous one in several ways. First, three different age groups were included - 4-year-olds, 5-year-olds, and adults - so as to better explore any developmental trends. Second, instead of pictures of objects, real objects were used. То the extent that there is а DPO Ьias, it might apply more strongly with actual objects than with depictions of objects. If so, the first experiment, which used pictures, might have made it unrealistically easy for subjects to abandon the object interpretation and choose the alternative collective noun construal. The procedure was also slightly modified, in that instead of pointing to а single group while presenting the new word, there were three groups, each of which was pointed to and named once. This was done to better focus the subjects оп the desired interpretation; if they do possess the capacity to attend to syntactic cues, hearing three distinct groups each called "а fendle" might Ье more effective at eliciting this capacity than hearing the phrase for only а single group. А final change is that half of the stimuli used in this study were familiar objects. А proposal based on mutual exclusivity would suggest that the acquisition of collective nouns would Ье facilitated (and perhaps only possiЫe) when the objects that are parts of the collection already have their own names, since only then would people Ье free to seek out other, non object-kind, interpretations of the word. As а result, one would expect more collective responses when the objects already have names. This hypothesis is explored below. 4. Method 4.1. Subjects The subjects were 24 4-year-old children (9 girls and 15 boys, mean age; 4 years, 6 months; range: 4 years, 1 month to 4 years, 11 months) and 24 5-year-old children (12 girls and 12 boys, mean age: 5 years, 5 months, range: 5 years, О months to 5 years, 11 months) from daycares in the Tucson area. Twenty-four University of Arizona students also participated, receiv­ ing class credit for their participation. АН the subjects were native speakers of English. None of the subjects took part in Experiment 1. 4.2. Stimuli and design Pretest materials were the same as those in Experiment 1. Тhere were four homogeneous sets of test materials. Two sets consisted of familiar objects: plastic spoons and pens. The other two sets consisted of unfamiliar objects: gray metal clips and short Ыuе plastic tubes. АН test materials were contained in separate zip-loc bags. Each subject was presented with four trials which alternated between familiar object trials and unfamiliar object trials. Half of the subjects received an unfamiliar object set in their first trial and half saw а familiar object set. Тhе order of presentation for each of the four object sets and the novel labels ( the same as in Experiment 1) was counterbalanced. Half of the subjects were in the plural count поип group and half were in the singular соипt поип group. Subjects were tested individually while seated at а tаЫе with an ex­ perimenter. Children were tested in а quite area of their day care and adults were tested in а small testing room. 4.3. Procedure The pretest was identical to that of Experiment 1. After the pretest had been completed, the experimenter told the subjects that they were going to Ье taught some new words with the puppet and, since he was not very smart, they should help him learn their meanings. (As in Experiment 1, adults were forewarned that the study had been designed to Ье appropriate for young children and that parts of the study might seem silly.) The experimenter then picked up а zip-loc bag and began the training phase Ьу removing three groups of four objects each and placing them on the tаЫе. Pointing to each group in turn - from the experiment's left to the experimenter's right - the experimenter used the novel word to label each of the groups of objects. In the plural count noun condition, subjects heard "these are fendles ...and these are fendles ... and these are fendles". In the singular count condition subjects heard: "this is а fendle ...and this is а fendle ...and this is а fendle". The groups of objects were then picked up and placed back in the bag.5 Following this, the experimenter placed another group of four objects (of the same kind) in front of the subject and а single object (also of the same kind) next to the group. The experimenter then asked "Сап you point to the fendle?" while the puppet nodded excitedly to encourage the children to respond. After а response, the materials were replaced in the bag and the procedure was repeated for the next three trials. Тhе entire procedure took approximately 10 minutes to complete. 5. Results and discussion 5.1. Pretest Adults performed perfectly on the pretest. Of the 24 4-year-old children, 14 (58%) passed the pretest and 10 ( 42%) failed. Of the 14 children who passed the pretest, 4 children met the criteria of answering both questions correctly but did not adopt an orthodox method in responding to the second question "Сап you point to the dog?" Instead of pointing to the picture depicting а single dog these children carefully pointed to а single dog in the picture depicting five dogs. This was counted as сопесt. (No children pointed to а single tree of the five when asked to point to the forest, and this type of response never occuпed in the main study.) Of the 10 4-year-olds who failed the pretest, seven responded to both questions Ьу pointing to the single object and the other three pointed to the collections of objects on both occasions. А t-test comparing the ages of the 14 children passing the pretest with the 10 children who failed showed no significant age differences. Of the 24 5-year olds, 12 passed the pretest and 12 failed. Of those who failed, six pointed to а single object in response to both questions and the 5 Since multiple groups were used in this study, it tumed out to Ье unfeasiЫe to ask children to repeat the novel word in the same manner as done in Experiment 1. Given that there were three piles in front of the children, if they were given а question such as "Could you tel1 Big Bird what we have over here?", they would Ье correct in answering "fendles" regardless of how the word was presented and how they were interpreting it. other six always pointed to the group of objects. А t-test comparing the ages of the 12 children passing the pretest (mean age: 5 years, 7 months) with the ages of the 12 who failed (mean age: 5 years, 4 months) found а marginally significant age difference (t(22) = 2.0, р < .06). 5.2. Main study The dependent variaЫe in all analyses was the number of times that subjects pointed to the collection of objects. А 3 х 2 х 2 ANOVA was conducted with age ( 4-year-olds vs. 5-year-olds vs. adults) and syntax (singular count noun vs. plural count noun) as between-subject factors and familiarity (familiar vs. unfamiliar) as а within-subject factor. The per­ centage of collective responses made Ьу children and adults in each condition is shown in Fig. 3. There was а main effect of syntax, with collective responses being more frequent in the singular count noun condition than in the plural count noun condition (73% vs. 23%; F(l,66) = 51.4, р < .005). There was also а significant Syntax х Age interaction (F(l, 66) = 4.2, р < .05), but there was по significant effect of age, no effect of familiarity, and no other interaction effects. 100 % of collective responses 80 60 40 20 4-year-olds 5-year-olds Adults Ages of subjects Fig. 3. Children and adult interpretations as а function of syntactic cues and familiarity in Experiment 2. □, Singular-familiar; О, singular-unfamiliar, ■, plural-familiar; 8, plural­ unfamiliar. То explore the results further, separate 2 х 2 ANOVAs were performed оп each of the three age groups, with syntax as the between-subject factor and familiarity as the within-subject factor. The analysis of the 4-year-olds indicated that although responses showed а trend in the right direction, there was по main effect of syntax. For the children exposed to the singular count noun, 60% responded with the collective interpretation; for the children exposed to the plural count noun, 35% responded with the collective interpretation. An analysis of just the 4-year-olds who passed the pretest also failed to find an effect - 64% collective responses for the singular count noun condition versus 39% collective responses for the plural count noun condition. The proportion of the 4-year-olds' collective interpretations in the singular count noun condition was not significantly greater than chance, but there was а marginally significant trend in the plural count noun condition to favor the object interpretation over chance (t(ll) = -1.6, р = .07, one-tailed). There was по main effect of familiarity and по interaction between familiarity and syntax. Analyses of the 5-year-olds found а main effect of syntax. The children in the singular count condition pointed to the collection 73% of the time and children in the plural count condition pointed to the collection 13% of the time (F(l, 22) = 21.2, р < .001). (Children who passed the pretest gave 79% collection responses in the singular count noun condition and 8% in the plural count noun condition; F(l, 10) = 29.5, р < .001). The proportion of 5-year-olds' object responses in the plural count condition was significantly greater than chance (t(ll) = -4.5, р < .005), as was the proportion of collective responses in the singular count condition (t(ll) = 2.3, р < .05). There was no main effect of familiarity and по interaction between familiarity and syntax. Adults also showed а strong effect of syntax (F(l, 22) = 75.7, р < .001), tending to give а collective response with singular count noun syntax (98%) but not with plural count noun syntax (20%). Both the proportion of object responses in the plural count condition (t(ll) = -3.4) and the proportion of collective responses in the singular count condition (t(ll) = 23) were significantly greater than chance (both ps < .005). Interestingly, adults did show а main effect of familiarity; the familiar objects were given а collective interpretation more often than the unfamiliar objects, as predicted Ьу mutual exclusivity (65% vs. 53%; F(l, 22) = 6.4, р < .02). Тhere was по interaction between familiarity and syntax. One interesting phenomenon is that while there is no monotonic trend with age in the nature of responses with plural count noun syntax, there is а gradual increase in the nature of singular count noun syntax responses, with subjects producing increasingly more collection responses as they get older. А contrast analysis on the singular count noun responses for all subjects was carried out, testing the hypothesis that there is а linear trend Ьу age (Rosenthal & Rosnow, 1985). This hypothesis was confirmed (F(l, 33) = 9.04, р < .01). ln addition, а series of pairwise Scheffe F-tests оп singular count noun responses found that adults gave significantly more collective responses than 4-year-olds when exposed to а singular count noun (F(2, 33) = 4.6, р < .05). No other pairwise comparisons were significant. Some explanations for this developmental difference are discussed below. One prediction made above, based on mutual exclusivity, is that collective responses should Ье more prevalent with objects whose names are already known Ьу the subjects. The results were surprising; there was no main effect of familarity and no interaction between age and familiarity. The analysis of the adult data, however, found that they were significantly more likely to give а collective response if the object already had а name, an effect that did not show up with the 4-year-olds or the 5-year-olds. The reason for the lack of an effect of familiarity for the children is unclear, since other investigators have found that 4- and 5-year-olds have no proЫem applying mutual exclusivity, for example, when learning substance names (Markman & Wachtel, 1988). А tentative hypothesis is that the children we tested viewed this study as an exercise in learning an entirely new language. lf so, then mutual exclusivity would not apply, as it does not preclude two words from different languages from having overlapping reference. GENERAL DISCUSSION The main findings from the above two experiments сап Ье summed up as follows. Adults and 5-year-olds (and in Experiment 1, adults and 4- and 5-year-olds) are sensitive to syntactic cues when acquiring object names and collective nouns. Тhеу сап learn а collective noun that describes an array of objects even if the objects that make up the collection are unfamiliar and have no prior name. In fact, for all ages tested, subjects treated the word as an object name over half of the time when it was presented as а plural count noun ("These are fendles"), and they treated the word as а collective noun over half of the time when it was presented as а singular count noun ("This is а fendle"). Sensitivity to syntax appears to improve with age, however, and adults are better at using these cues than children. Finally, there is а gradual increase with age in the tendency to construe а singular noun as referring to а collection, while there is по such increase in the tendency to treat the plural count noun as an object name. Two less central hypotheses were not supported. For children and adults, it made no difference whether or not the groups of objects were depicted as connected, contrary to the hypothesis that connectedness would lead the subjects to better view the groups as single individuals. Also, while adults are more likely to give а collective response when objects already have known names, which is in accord with the predictions of mutual exclusivity, this did not occur with children. These results provide some mixed support for the theory proposed here. Children and adults are сараЫе of learning collective nouns through syntactic support and do not need to know the name for the kind of object for this learning to Ье successful. Оп the other hand, adults are more attentive to syntactic cues than children, significantly so with regard to the difference between the 4- and 5-year-olds versus the adults in Experiment 1 and the 4-year-olds versus the adults in Experiment 2. Further, the 4-year­ olds in Experiment 2 did not show а significant main effect of syntax, and while the children in Experiment 1 were sensitive to syntax, their proportion of collective noun responses in the singular count condition was not significantly greater than chance. How сап we explain these developmental differences? Below we discuss three possibllities. ( 1) Young children have proЫems overriding the whole object Ьias The first type of explanation is most congenial to the special constraint theory. This is that children start off with an overall Ьias to favor the whole object interpretation when learning а novel label, and only later, as а result of language learning and/or conceptual change, do they become сараЫе of attending to syntactic cues and learning labels that refer to individuals other than whole objects. Under this proposal, the children in our experiments performed worse than the adults because they found it difficult to override their Ьias to construe а novel word as referring to а kind of whole object and/or were less sensitive to the semantic implications of the syntactic manipulation. This does not adequately explain the pattern of responses, however. For one thing, it predicts that the object nouns (where the child is shown the group and told "These are fendles") should Ье much easier to learn than the collective nouns (where the child is shown the group and told "This is а fendle"). In fact, the only errors that should occur at all are with the collective nouns; children should systematically misinterpret them as object names. There is some trend in that direction; errors with object nouns are consistently less frequent than errors with collective nouns. But these differences were relatively small and, with the 4-year-olds in Experiment 2, the number of errors in the singular count condition and the plural count condition were virtually identical. In other words, it is not that the failure of 4-year-olds was because they always treated the word as an object name, regardless of how it was presented. On the contrary, their errors were about evenly matched - 40% of the time when the word was presented as а singular count noun, they interpreted it as naming the kind of object, and 35% of the time when the word was presented as а plural count noun, they took it as naming the kind of collection. Note also that in both experiments, the most common response when presented with а singular count noun was the collective interpretation, not the object interpretation. Finally, consider the pretest, in which children were tested on their knowledge of the meanings of the object noun dog and the collective noun forest. While children were not perfect with forest, they were usually correct (Experiment 1, 4- and 5-year-olds: 81%; Experiment 2, 4-year-olds: 71%, 5-year-olds: 75% ), and they did not make significantly more errors with forest than they did with dog. This is not to deny that children might Ье less sensitive to syntax or that they might Ье more prone to favor the object interpretation in certain contexts - both possiЬilities might Ье true (as discussed below). But the considerations above suggest that the behavior of the children in this study is not consistent with the proposal that their errors are due to the existence of special constraints. (2) Young children are less sensitive to syntactic cues than adults А second hypothesis has to do with sensitivity to syntax. There is consideraЫe support for the view that children are less sensitive than adults to syntactic cues when inferring word meanings (e.g., Landau, Smith, & Jones, 1988; Soja, 1992). In the domain here, it is possiЫe that although children know that an NP with а singular count noun denotes а single individual (including, potentially, а collection of objects) and that an NP with а plural count noun denotes more than one individual, they just tend not to рау as much attention to this difference between а fendle and fendles as adults do. In particular, they might know that count nouns сап apply to individuals other than whole objects (even 2-year-olds seems to know this; Soja, 1992), but they do not apply this knowledge as consistently as adults. The existence of such child-adult differences with regard to sensitivity to syntax is fully consistent with the theory proposed here. In fact, any theory of acquisition that assumes that people сап attend to more than one type of information in the course of word learning must admit the possiЬility of this sort of developmental change. Under the theory proposed here, there are multiple cues that conspire to direct а child towards а particular meaning; these include the perceived intention of the language user, perceptual/ cognitive Ьiases in how we construe the world (such as the DPO Ьias), and mappings between syntactic structure and conceptual categories. How these cues are weighed is а subtle matter and there тау well Ье differences between children and adults, as well as individual differences between adults. The difficulty in weighing cues сап Ье illustrated with an example. Imagine someone pointing to а salient part of а novel object and saying "Look at the fendle". Here different cues to word meaning conflict. On the one hand, the person is pointing to the part and so is possiЫy intending to refer to it. Also, the syntax of the word is compatiЫe with this interpreta- tion, since а part Ье construed as an individual and the NP the fendle serves to estaЫish reference to an individual. On the other hand, we are prone to construe the whole object, not the part, as the most salient individual, and the fact we have no prior name for the whole object means that а constraint such as mutual exclusivity will not Ьias against this interpretation. These latter considerations might cause us to assume that the person meant to refer to the whole object but was pointing to the wrong place. Matters get more complicated when we consider what would happen if the person just mumЫed "fendle", or if she was only pointing in а vague manner, or pointing in one direction but looking elsewhere. In these sorts of cases, there is по clear solution, as there might Ье several valid ways of taking into consideration these different cues to word meaning. There is some debate over the extent to which adults intentionally try to help children in the word learning process (for different views, see Nelson, 1988; Gleitman, 1990; Bloom, in press), but nobody argues that adults purposefully use clashing syntactic frames and intentional cues in а malevol­ ent attempt to confuse their children. In this regard, multiple cues to word meaning should Ье viewed in the same light as the multiple cues that interact in the course of certain aspects of visual perception (e.g., those that determine figure-ground segmentation; Peterson & Gibson, 1993) - in the normal environment, they aid successful perception; they do not hinder it. (3) There are developmental differences in the notion of "individual" Children's limited sensitivity to syntactic information might partially explain the child-adult differences, but there is а further possibllity. The stimuli used in Experiment 1 were groups of novel machines or novel animals; in Experiment 2, they were piles of unfamiliar or familiar objects. From the standpoint of language and cognition, these were highly implaus­ iЫe referents for collective nouns. Languages such as English do not have collective nouns that refer to arbltrary groups of machines or animals, or to arbltrary piles of specific objects, and thus such collective nouns might prove particularly difficult to learn. Why would such collections pose difficulties from the standpoint of word learning? One candidate explanation has to do with the nature of the semantic category "individual". Once we determine that dog aod сир are intended to refer to dogs aod cups, we have оо proЫems viewiog these objects as appropriate refereots for couot oouos. Discrete physical objects are highly salieot iodividuals, presumaЫy because our perceptual aod cognitive systems have evolved to coostrue them io this maooer (see Spelke, 1994). But we have оо such iohereot Ьias with regard to arbltrary collectioos of objects. lo order to cooceptualize them as individuals, we need some motivation to see them as distinct entities in their own right. Real collective nouns, such as family, bunch, and army, refer to sets of objects that bear some salient and enduring relationship with one another, either Ьу being spatially and/or physically connected, like the grapes in а bunch, or Ьу having more abstract social connections like the members of а family or the soldiers in an army .6 One proposal is that such individuals are the products of our "naive theories" or "stances". We construe families and bunches and armies as individuals because we construe them as distinct causal entities in our mental models of the physical and social world (for discussion, see Bloom, in press; Dennett, 1991). In contrast, the groups of entities in Experiment 1 and the piles of objects in Experiment 2 have по conceptual status as distinct entities. The only reason to treat them as collections is because - in the singular count condition - this is the only interpretation consistent with the sorts of arrays being pointed to and the syntax used to describe them. It might Ье that the unnaturalness of the stimuli affects children more than adults. Young children may fail to see why the groups of objects constitute individuals and thus sometimes ignore the syntax and treat the labels as object names, since objects are cognitively salient individuals. The gradual increase in collective responses with age might Ье due to an increasing aЬility to construe the groups as collections despite its arЬitrari­ ness, solely because this is the only interpretation consistent with the linguistic and non-linguistic behavior of the experimenter.7 This hypothesis suggests that if young children were exposed to more cognitively natural sets of entities, they would Ье more willing to treat them as individuals and better аЫе to learn collective nouns that name them. It is not impossiЫe to construct such individuals and study how people construe them (В!ооm & Fountain, in prep.). Consider, for example, the display in Fig. 4. In one study, adults were shown such а display оп а computer screen and heard it described with а plural count noun: "These are fendles". There are two conceivaЫe interpretations as to what the word might mean, both 6 Not all nouns have conditions on the kinds of objects they refer to, however. As noted above, group nouns such as group, pile, heap, stack, cluster, multitude, and triad can describe entities that belong to а range of different kinds, as in а pile of books!CDs/sockslsand. Unlike collective nouns like family, these nouns are defined through properties such as the numerosities of the collections they refer to, the collections' configuration and density, and how they are created. The acquisition and representation of these words poses some interesting puzzles not directly addressed Ьу the theory presented here. 7 Тhis capacity to construe а set of objects as а single individual solely on the basis of intentional and syntactic cues shows up in another domain of word leaming: the acquisition of names for modem art. An adult is сараЫе of construing (for instance) several pieces of fiberglass arranged on the floor as an individual (as in Eva Hesse's Tori), and can learn а name for those pieces of fiberglass, even if he or she does not understand why the pieces go together or what the specific intent of the artwork is. ArguaЬ!y, this is because the adult assumes that the group is an individual in the mind of its creator (the artist) and, as а result, it gains psychological coherence; it becomes real because someone else intends it to Ье real (Bloom, 1994а, in press; see also Levinson, 1989). Fig. 4. Ambiguous display: are there 15 individuals (where individual = object) or are there 3 (where individual = collection)? consistent with the syntax; it could Ье an object name, and so there would Ье 15 fendles, or it could Ье а name for the groups of objects, so that there would Ье 3 fendles. When the objects are stationary, adults virtually always interpret fendle as an object name; apparently mere grouping is not enough to drive them towards а collective interpretation. Another group of subjects was shown а similar display, but this time each group of objects was moving across the screen in а unique pattern, distinct from the other groups. This provided the subjects with some motivation to view each group as а distinct causal entity in its own right. As predicted, they then showed а strong Ьias to treat "fendles" as referring to the groups, not the objects. Further experiments are exploring whether this collective interpretation is due to higher-level explanatory considerations (as proposed here) - or whether it is because of factors such as the Gestalt principle of common fate or, alternatively, because these moving groups share some of the same properties that serve to define objects (e.g., Spelke et а!., 1992). It is an interesting question whether а 4-year-old - or а much younger child - shown the same display of moving groups would also converge оп the collective interpretation. More generally, attempts to teach children names for individuals that are not objects, such as collections, sounds, and events, could serve as а useful methodology with which to explore the nature and development of the category of "individual". One prediction is that if the individuals are salient enough, very young children should Ье аЫе to acquire such names, even for collections of objects in which the objects themselves belong to unfamiliar kinds. In support of this speculation, there is some reason to believe that collections do have some psychological reality as individuals for young children. For one thing, words like family and forest show up early in lexical development. Furthernюre, there is evidence that children perform better on Piagetian class-inclusion tasks when tested with collective nouns than with superordinate nouns (Markman, 1973; Markman & Seibert, 1976) and that they even sometimes spontaneously distort superordinate nouns into collective terms (Callanan & Markman, 1982; Markman, Horton, & McLanahan, 1980; but see also Macnamara, 1986 for а critical discussion). The studies reported here found а sensitivity to syntax only in older children and the one semantic cue we manipulated, the connectedness of the objects, had по effect. Because of this, any inferences about younger children must Ье tentative. But the research reviewed above shows that 2and 3-year-olds possess some sensitivity to syntax and some capacity to grasp individuals that are more abstract than whole objects. So the speculation that such children сап acquire novel collective nouns of the sort shown in Fig. 4 is not entirely unwarranted. The goal of the research reported here was to explore а perspective on lexical development in which the bias to interpret novel labels as referring to kinds of whole objects is а by-product of more general aspects of language and cognition. This alternative is motivated Ьу data from children as young as two, such as the fact that they сап acquire pronouns and proper names, and сап use syntax to acquire aspects of word meaning. The positive results reported here with regard to the acquisition of collective nouns Ьу older children suggests that continued research along these lines might g1ve us further insight into the development of language and cognition.