Simplicity and generalization: Short-cutting abstraction in children's object categorizations

Development in any domain is often characterized by increasingly abstract representations. Recent evidence in the domain of shape recognition provides one example; between 18 and 24 months children appear to build increasingly abstract representations of object shape [Smith, L. B. (2003). Learning to recognize objects. Psychological Science, 14, 244-250]. Abstraction is in part simplification because it requires the removal of irrelevant information. At the same time, part of generalization is ignoring irrelevant differences. The resulting prediction is this: simplification may enable generalization. Four experiments asked whether simple training instances could shortcut the process of abstraction and directly promote appropriate generalization. Toddlers were taught novel object categories with either simple or complex training exemplars. We found that children who learned with simple objects were able to generalize according to shape similarity, typically relevant for early object categories, better than those who learned with complex objects. Abstraction is the product of learning; using simplified - already abstracted instances - can short-cut that learning, leading to robust generalization.     

Labelling patterns and object naming

We examined the effects of different labelling patterns on the generalization of object names. Two‐year‐olds, three‐year‐olds and adults were shown two ‘standard’ objects, which were named with the same label, or with two different labels, or with no label at all. Participants were then asked whether objects morphed to be intermediate to the standards belonged to one of the labelled categories or, in the No Label condition, were ‘like’ one of the standards. The Same Label condition showed generalization to all intermediates, whereas the Different Label and No Label conditions showed division of the intermediates into two separate categories, with somewhat sharper division under Different Label. These results suggest two possible mechanisms of lexical learning: ‘boosting’ the equivalence of different exemplars through label identity, and ‘differentiating’ the exemplars through differences in labelling. The studies provided strong evidence for boosting. Learners are sensitive to the distribution of labels across exemplars, and they hold powerful assumptions about the relationship between these distributions and the underlying naming space. These findings have implications for the early emergence of cross‐linguistic differences in lexical learning.     

Beyond fast mapping: young children's extensions of novel words and novel facts

L. Markson and P. Bloom (1997) concluded that there was evidence against a dedicated system for word learning on the basis of their finding that children remembered a novel word and a novel fact equally well. However, a word-learning system involves more than recognition memory; it must also provide a means to guide the extension of words to additional exemplars, and words and facts may differ with regard to extendibility. Two studies are reported in which 2-4-year-old children learned novel words and novel facts for unfamiliar objects and then were asked to extend the words and facts to additional exemplars of the training objects. In both studies, children extended the novel word to significantly more category members than they extended the novel fact. The results show that by 2 years of age, children honor the necessary extendibility of novel count nouns but are uncertain about the extendibility of arbitrary facts.     

The training and transfer of real-world perceptual expertise

A hallmark of perceptual expertise is that experts classify objects at a more specific, subordinate level of abstraction than novices. To what extent does subordinate-level learning contribute to the transfer of perceptual expertise to novel exemplars and novel categories? In this study, participants learned to classify 10 varieties of wading birds and 10 varieties of owls at either the subordinate, species (e.g., "great blue crown heron,"eastern screech owl") or the family ("wading bird,"owl") level of abstraction. During training, the amount of visual exposure was equated such that participants received an equal number of learning trials for wading birds and owls. Pre- and posttraining performance was measured in a same/different discrimination task in which participants judged whether pairs of bird stimuli belonged to the same or different species. Participants trained in species-level discrimination demonstrated greater transfer to novel exemplars and novel species categories than participants trained in family-level discrimination. These findings suggest that perceptual categorization, not perceptual exposure per se, is important for the development and generalization of visual expertise.     

Establishing a generalized repertoire of helping behavior in children with autism

The present study used a multiple baseline across participants design to assess whether 4 children with autism could learn a generalized repertoire of helping adults with different tasks through the use of a multicomponent teaching package. Different helping responses were taught in the presence of multiple exemplars of discriminative stimuli drawn from experimenter-defined categories of helping behavior (e.g., locating objects, putting away items, setting up an activity). During the training condition, video models, prompting, and reinforcement were used. The results showed that all 4 children learned to emit appropriate helping responses in the presence of discriminative stimuli from the helping categories used during training. Generalization of helping responses was observed in the presence of untrained discriminative stimuli during additional probe conditions. Additional pre- and postintervention generalization trials showed that the frequency of helping responses also increased in the presence of novel stimuli, in a novel setting, and with a novel instructor.     

Is this a dax which I see before me? Use of the logical argument disjunctive syllogism supports word-learning in children and adults

Many authors have argued that word-learning constraints help guide a word-learner's hypotheses as to the meaning of a newly heard word. One such class of constraints derives from the observation that word-learners of all ages prefer to map novel labels to novel objects in situations of referential ambiguity. In this paper I use eye-tracking to document the mental computations that support this word-learning strategy. Adults and preschoolers saw images of known and novel objects, and were asked to find the referent of known and novel labels. Experiment 1 shows that adults systematically reject a known distractor (e.g. brush) before mapping a novel label (e.g. "dax") to a novel object. This is consistent with the proposal that participants worked through a Disjunctive Syllogism (i.e. Process-of-Elimination) to motivate the mapping of the novel label to the novel object. Experiment 2 shows that processing is similar for adults performing an implicit Disjunctive Syllogism (e.g. "the winner is the dax") and an explicit Disjunctive Syllogism (e.g. "the winner is not the iron"). Experiment 3 reveals that similar processes govern preschoolers' mapping of novel labels. Taken together, these results suggest that word-learners use Disjunctive Syllogism to motivate the mapping of novel labels to novel objects.     

Study preferences for exemplar variability in self-regulated category learning

Increasing exemplar variability during category learning can enhance classification of novel exemplars from studied categories. Four experiments examined whether participants preferred variability when making study choices with the goal of later classifying novel exemplars. In Experiments 1–3, participants were familiarised with exemplars of birds from multiple categories prior to making category-level assessments of learning and subsequent choices about whether to receive more variability or repetitions of exemplars during study. After study, participants classified novel exemplars from studied categories. The majority of participants showed a consistent preference for variability in their study, but choices were not related to category-level assessments of learning. Experiment 4 provided evidence that study preferences were based primarily on theoretical beliefs in that most participants indicated a preference for variability on questionnaires that did not include prior experience with exemplars. Potential directions for theoretical development and applications to education are discussed.     

Sensitivity to sampling in Bayesian word learning

We report a new study testing our proposal that word learning may be best explained as an approximate form of Bayesian inference (Xu & Tenenbaum, in press). Children are capable of learning word meanings across a wide range of communicative contexts. In different contexts, learners may encounter different sampling processes generating the examples of word-object pairings they observe. An ideal Bayesian word learner could take into account these differences in the sampling process and adjust his/her inferences about word meaning accordingly. We tested how children and adults learned words for novel object kinds in two sampling contexts, in which the objects to be labeled were sampled either by a knowledgeable teacher or by the learners themselves. Both adults and children generalized more conservatively in the former context; that is, they restricted the label to just those objects most similar to the labeled examples when the exemplars were chosen by a knowledgeable teacher, but not when chosen by the learners themselves. We discuss how this result follows naturally from a Bayesian analysis, but not from other statistical approaches such as associative word-learning models.     

Children learn words easier when they are interested in the category to which the word belongs

The overall pattern of vocabulary development is relatively similar across children learning different languages. However, there are considerable differences in the words known to individual children. Historically, this variability has been explained in terms of differences in the input. Here, we examine the alternate possibility that children's individual interest in specific natural categories shapes the words they are likely to learn – a child who is more interested in animals will learn a new animal name easier relative to a new vehicle name. Two‐year‐old German‐learning children (N = 39) were exposed to four novel word–object associations for objects from four different categories. Prior to the word learning task, we measured their interest in the categories that the objects belonged to. Our measure was pupillary change following exposure to familiar objects from these four categories, with increased pupillary change interpreted as increased interest in that category. Children showed more robust learning of word–object associations from categories they were more interested in relative to categories they were less interested in. We further found that interest in the novel objects themselves influenced learning, with distinct influences of both category interest and object interest on learning. These results suggest that children's interest in different natural categories shapes their word learning. This provides evidence for the strikingly intuitive possibility that a child who is more interested in animals will learn novel animal names easier than a child who is more interested in vehicles.     

The reciprocal relation between sleep and memory in infancy: Memory‐dependent adjustment of sleep spindles and spindle‐dependent improvement of memories

Sleep spindle activity in infants supports their formation of generalized memories during sleep, indicating that specific sleep processes affect the consolidation of memories early in life. Characteristics of sleep spindles depend on the infant's developmental state and are known to be associated with trait‐like factors such as intelligence. It is, however, largely unknown which state‐like factors affect sleep spindles in infancy. By varying infants’ wake experience in a within‐subject design, here we provide evidence for a learning‐ and memory‐dependent modulation of infant spindle activity. In a lexical‐semantic learning session before a nap, 14‐ to 16‐month‐old infants were exposed to unknown words as labels for exemplars of unknown object categories. In a memory test on the next day, generalization to novel category exemplars was tested. In a nonlearning control session preceding a nap on another day, the same infants heard known words as labels for exemplars of already known categories. Central–parietal fast sleep spindles increased after the encoding of unknown object–word pairings compared to known pairings, evidencing that an infant's spindle activity varies depending on its prior knowledge for newly encoded information. Correlations suggest that enhanced spindle activity was particularly triggered, when similar unknown pairings were not generalized immediately during encoding. The spindle increase triggered by previously not generalized object–word pairings, moreover, boosted the formation of generalized memories for these pairings. Overall, the results provide first evidence for a fine‐tuned regulation of infant sleep quality according to current consolidation requirements, which improves the infant long‐term memory for new experiences.     

Recombinative generalization of within-syllable units in prereading children

This study demonstrates recombinative generalization of within-syllable units in prereading children. Three kindergarten children learned to select printed consonant-vowel-consonant words upon hearing the corresponding spoken words. The words were taught in sets; there were six sets, presented consecutively. Within sets, the four words that were taught had overlapping letters, for example, sat, mat, sop, and sug. Tests for recombinative generalization determined whether the children selected novel words with the same components as the trained words (e.g., mop and mug). Two children demonstrated recombinative generalization after one training set, and the 3rd demonstrated it after two training sets. In contrast, 2 other children, who received tests but no training, showed low accuracy across six sets. The 3 experimental children then demonstrated highly accurate printed-word-to-picture matching, and named the majority of the printed words. These findings are a promising step in the development of a computerized instructional technology for reading.     

Categorization in infancy: labeling induces a persisting focus on commonalities

Recent studies with infants and adults demonstrate a facilitative role of labels in object categorization. A common interpretation is that labels highlight commonalities between objects. However, direct evidence for such a mechanism is lacking. Using a novel object category with spatially separate features that are either of low or high variability across the stimulus set, we tracked 12‐month‐olds’ attention to object features during learning and at test. Learning occurred in both conditions, but what was learned depended on whether or not labels were heard. A detailed analysis of eye movements revealed that infants in the two conditions employed different object processing strategies. In the silent condition, looking patterns were governed exclusively by the variability of object parts. In the label condition, infants’ categorization performance was linked to their relative attention to commonalities. Moreover, the commonality focus persisted after learning even in the absence of labels. These findings constitute the first experimental evidence that labels induce a persistent focus on commonalities.     

Developmental differences in children's context-dependent word learning

In this study, 2.5-, 3-, and 4-year-olds (N = 108) participated in a novel noun generalization task in which background context was manipulated. During the learning phase of each trial, children were presented with exemplars in one or multiple background contexts. At the test, children were asked to generalize to a novel exemplar in either the same or a different context. The 2.5-year-olds’ performance was supported by matching contexts; otherwise, children in this age group demonstrated context dependent generalization. The 3-year-olds’ performance was also supported by matching contexts; however, children in this age group were aided by training in multiple contexts as well. Finally, the 4-year-olds demonstrated high performance in all conditions. The results are discussed in terms of the relationship between word learning and memory processes; both general memory development and memory developments specific to word learning (e.g., retention of linguistic labels) are likely to support word learning and generalization.     

Infants' reasoning about opaque and transparent occluders in an individuation task

Two experiments document that conceptual knowledge influences 3-year-olds' extension of novel words. In Experiment 1, when objects were described as having conceptual properties typical of artifacts, children extended novel labels for these objects on the basis of shape alone. When the very same objects were described as having conceptual properties typical of animate kinds, children extended novel labels for these objects on the basis of both shape and texture. Moreover, providing a salient perceptual cue (Experiment 2) did not interfere with children's reliance on conceptual information in extending novel words: when an object with eyes was labeled with a novel word in the context of a story describing the object as an artifact, children extended the label on the basis of shape alone (i.e. as though the object were an artifact). These results, which challenge directly the position that 'dumb attentional mechanisms' can account for word learning, stand as evidence for the central role of conceptual information in mapping words to meaning.     

A role for action knowledge in visual object identification

We evaluated the impact of visual similarity and action similarity on visual object identification. We taught participants to associate novel objects with nonword labels and verified that in memory visually similar objects were confused more often than visually dissimilar objects. We then taught participants to associate novel actions with nonword labels and verified that similar actions were confused moreoften th an dissimilaractions. We then paired specific objects with specific actions. Visually similar objects paired with similar actions were confused more often in memory than when these same objects were paired with dissimilar actions. Hence the actions associated with objects served to increase or decrease their separation in memory space, and influenced the ease with which these objects could be identified. These experiments ultimately demonstrated that when identifying stationary objects, the memory of how these object were used dramatically influenced the ability to identify these objects.     

Testing a word is not a test of word learning

Although vocabulary acquisition requires children learn names for multiple things, many investigations of word learning mechanisms teach children the name for only one of the objects presented. This is problematic because it is unclear whether children's performance reflects recall of the correct name–object association or simply selection of the only object that was singled out by being the only object named. Children introduced to one novel name may perform at ceiling as they are not required to discriminate on the basis of the name per se, and appear to rapidly learn words following minimal exposure to a single word. We introduced children to four novel objects. For half the children, only one of the objects was named and for the other children, all four objects were named. Only children introduced to one word reliably selected the target object at test. This demonstration highlights the over-simplicity of one-word learning paradigms and the need for a shift in word learning paradigms where more than one word is taught to ensure children disambiguate objects on the basis of their names rather than their degree of salience.     

Discrimination of artificial categories structured by family resemblances: a comparative study in people (Homo sapiens) and pigeons (Columba livia)

Adult humans (Homo sapiens) and pigeons (Columba livia) were trained to discriminate artificial categories that the authors created by mimicking 2 properties of natural categories. One was a family resemblance relationship: The highly variable exemplars, including those that did not have features in common, were structured by a similarity network with the features correlating to one another in each category. The other was a polymorphous rule: No single feature was essential for distinguishing the categories, and all the features overlapped between the categories. Pigeons learned the categories with ease and then showed a prototype effect in accord with the degrees of family resemblance for novel stimuli. Some evidence was also observed for interactive effects of learning of individual exemplars and feature frequencies. Humans had difficulty in learning the categories. The participants who learned the categories generally responded to novel stimuli in an all-or-none fashion on the basis of their acquired classification decision rules. The processes that underlie the classification performances of the 2 species are discussed.     

Object name learning provides on-the-job training for attention

By the age of 3, children easily learn to name new objects, extending new names for unfamiliar objects by similarity in shape. Two experiments tested the proposal that experience in learning object names tunes children's attention to the properties relevant for naming—in the present case, to the property of shape—and thus facilitates the learning of more object names. In Experiment 1, a 9-week longitudinal study, 17-month-old children who repeatedly played with and heard names for members of unfamiliar object categories well organized by shape formed the generalization that only objects with similar shapes have the same name. Trained children also showed a dramatic increase in acquisition of new object names outside of the laboratory during the course of the study. Experiment 2 replicated these findings and showed that they depended on children's learning both a coherent category structure and object names. Thus, children who learn specific names for specific things in categories with a common organizing property—in this case, shape—also learn to attend to just the right property—in this case, shape—for learning more object names.     

Metamemory and the transfer of an interrogative strategy by EMR children

EMR children were taught to produce an interrogative strategy as an aid to paired-associate learning. Interrogative training was spread over four sessions with experimenter guidance gradually faded; training was followed by tests of long-term retention for the trained items, strategy maintenance, and generalization. Metamemory was assessed prior to training and following generalization. Analysis of recall demonstrated successful maintenance and generalization of the interrogative strategy; the quality of elaborations produced by the trained children predicted recall accuracy. Importantly, metamemory pretest measures for the interrogative trained children were related to strategy use at generalization. Metamemory posttest reliably predicted recall during training, maintenance, and generalization sessions and strategy use at maintenance. Interrogative training did not noticeably improve general aspects of metamemorial knowledge.     

Evidence for a form of mutual exclusivity during label acquisition by grey parrots (Psittacus erithacus)?

The authors present context-dependent evidence for a form of mutual exclusivity during label learning by Grey parrots (Psittacus erithacus). For human children, mutual exclusivity refers to their assumption during early word learning that an object has one and only one label. Along with the whole-object assumption (that a label likely refers to an entire object rather than some partial aspect), mutual exclusivity is thought to guide children in initial label acquisition. It may also help children overcome the whole-object assumption by helping them interpret a novel word as something other than an object label, but for young children, any second label for an object can initially be more difficult to acquire than the first. The authors show that Grey parrots quickly learn object labels for items, then have considerable difficulty learning to use color labels with respect to a previously labeled item unless specifically taught to use a color and object label as a pair.