"Shape bias" goes social: Children categorize people by weight rather than race.

Children tend to categorize novel objects according to their shape rather than their color, texture, or other salient properties-known as "shape bias." We investigated whether this bias also extends to the social domain, where it should lead children to categorize people according to their weight (their body shape) rather than their race (their skin color). In Study 1, participants (n = 50 US 4- and 5-year-olds) were asked to extend a novel label from a target object/person to either an object/person who shared the target's shape/weight, color/race, or neither. Children selected the shape-/weight-matched individual over the color-/race-matched individual (dobjects  = 1.58, dpeople  = 0.99) and their shape biases were correlated across the two domains. In Study 2, participants (n = 20 US 4- and 5-year-olds) were asked to extend a novel internal property from a target person to either a person who shared the target's weight, race, or neither. Again, children selected the weight-matched individual (d = 1.98), suggesting they view an individual's weight as more predictive of their internal properties than their race. Overall, results suggest that children's early shape bias extends into the social domain. Implications for weight bias and early social cognition are discussed. RESEARCH HIGHLIGHTS: Preschoolers extend novel labels based on people's weight rather than their race. Preschoolers infer internal features based on people's weight rather than their race. Shape biases are present, and correlated, across the social and object domains.

Object Shape and Depth of Word Representations in Preschoolers.

This study examined the effect of a shape cue (i.e., co-speech gesture) on word depth. We taught 23 preschoolers (M = 3;5 years, SD = 5.82) novel objects with either shape (SHP) or indicator (IND) gestures. SHP gestures mimicked object form, but IND gestures were not semantically related to the object (e.g., an upward-facing palm, extended toward the object). Each object had a unique IND or SHP gesture. Outcome measures reflected richer semantic and phonological learning in the SHP than in the IND condition. In the SHP condition, preschoolers (a) expressed more semantic knowledge, (b) said more sounds in names, and (c) generalized more names to untaught objects. There were also fewer disruptions to prime picture names in the SHP condition; we discuss the benefit of a co-speech shape gesture to capitalize on well-established statistical word learning patterns.

Late bloomer or language disorder? Differences in toddler vocabulary composition associated with long-term language outcomes.

Children with delays in expressive language (late talkers) have heterogeneous developmental trajectories. Some are late bloomers who eventually "catch-up," but others have persisting delays or are later diagnosed with developmental language disorder (DLD). Early in development it is unclear which children will belong to which group. We compare the toddler vocabulary composition of late talkers with different long-term outcomes. The literature suggests most children with typical development (TD) have vocabularies dominated by names for categories organized by similarity in shape (e.g., cup), which supports a bias to attend to shape when generalizing names of novel nouns-a bias associated with accelerated vocabulary development. Previous work has shown that as a group, late talkers tend to say fewer names for categories organized by shape and are less likely to show a "shape bias" than TD children. Here, in a retrospective analysis of 850 children, we compared the vocabulary composition of groups of toddlers who were late bloomers or persisting late talkers. At Time 1 (13-27 months), the persisting late talkers said a smaller proportion of shape-based nouns than both TD children and late bloomers who "caught up" to typically sized vocabularies months later (18-38-months). Additionally, children who received a DLD diagnosis between 4 and 7 years said a significantly smaller proportion of shape-based nouns in year two than TD children and children with other diagnoses (e.g., dyslexia). These findings bring new insight into sources of heterogeneity amongst late talkers and offer a new metric for assessing risk. RESEARCH HIGHLIGHTS: Toddler vocabulary composition, including the proportion of names for categories organized by shape, like spoon, was used to retrospectively compare outcomes of late talking children Persisting Late Talkers said a smaller proportion of shape-based nouns during toddlerhood relative to Late Bloomers (late talkers who later caught up to have typically-sized vocabularies) Children with later DLD diagnoses said a smaller proportion of shape-based nouns during toddlerhood relative to children without a DLD diagnosis The data illustrate the cascading effects of vocabulary composition on subsequent language development and suggest vocabulary composition may be one important marker of persisting delays.

Beyond the Shape of Things: Infants Can Be Taught to Generalize Nouns by Objects' Functions.

Two-year-olds typically extend labels of novel objects by the objects' shape (shape bias), whereas adults do so by the objects' function. Is this because shape is conceptually easier to comprehend than function? To test whether the conceptual complexity of function prevents infants from developing a function bias, we trained twelve 17-month-olds (function-training group) to focus on objects' functions when labeling the objects over a period of 7 weeks. Our training was similar to previously used methods in which 17-month-olds were successfully taught to focus on the shape of objects, resulting in a precocious shape bias. We exposed another 12 infants (control group) to the same objects over 7 weeks but without labeling the items or demonstrating their functions. Only the infants in the function-training group developed a function bias. Thus, the conceptual complexity of function was not a barrier for developing a function bias, which suggests that the shape bias emerges naturally because shape is perceptually more accessible than function.

What categorical induction variability reveals about typical and atypical development.

Categorical induction abilities are robust in typically developing (TD) preschoolers, while children with Autism Spectrum Disorders (ASD) frequently perform inconsistently on tasks asking for the transference of traits from a known category member to a new example based on shared category membership. Here, TD five-year-olds and six-year-olds with ASD participated in a categorical induction task; the TD children performed significantly better and more consistently than the children with ASD. Concurrent verbal and nonverbal tests were not significant correlates; however, the TD children's shape bias performance at two years of age was significantly positively predictive of categorical induction performance at age five. The shape bias, the tendency to extend a novel label to other objects of the same shape during word learning, appears linked with categorical induction ability in TD children, suggesting a common underlying skill and consistent developmental trajectory. Word learning and categorical induction appear uncoupled in children with ASD.

Effects of past language experience and present language context on the shape bias in Spanish-English bilingual children.

English-monolingual children develop a shape bias early in language acquisition, such that they more often generalize a novel label based on shape than other features. Spanish-monolingual children, however, do not show this bias to the same extent (Hahn & Cantrell, 2012). Studying children who are simultaneously learning both Spanish and English presents a unique opportunity to further investigate how this word-learning bias develops. Thus, we asked how Spanish-English bilingual children (Mage  = 21.31 months) perform in a novel-noun generalization (NNG) task, specifically examining how past language experience (i.e. language exposure and vocabulary size) and present language context (i.e. whether the NNG task was conducted in Spanish or English) influence the strength of the shape bias. Participants completed the NNG task either entirely in English (N = 16) or entirely in Spanish (N = 16), as well as language understanding tasks in both English and Spanish to ensure that they understood what the experimenter was asking them to do. Parents completed a language exposure survey and vocabulary checklists in Spanish and English. There was a significant interaction between condition and choice type: Bilingual children in the English condition showed a shape bias in the NNG task, but bilingual children in the Spanish condition showed no reliable biases. No measures of past language experience were related to NNG task performance. These results suggest that when learning new words, bilingual children are attuned to the regularities of the present language context, and prior language experiences may play a more secondary role.

Atypical shape bias and categorization in autism: Evidence from children and computational simulations.

The shape bias, a preference for mapping new word labels onto the shape rather than the color or texture of referents, has been postulated as a word-learning mechanism. Previous research has shown deficits in the shape bias in children with autism even though they acquire sizeable lexicons. While previous explanations have suggested the atypical use of color for label extension in individuals with autism, we hypothesize an atypical mapping of novel labels to novel objects, regardless of the physical properties of the objects. In Experiment 1, we demonstrate this phenomenon in some individuals with autism, but the novelty of objects only partially explains their lack of shape bias. In a second experiment, we present a computational model that provides a developmental account of the shape bias in typically developing children and in those with autism. This model is based on theories of neurological dysfunctions in autism, and it integrates theoretical and empirical findings in the literature of categorization, word learning, and the shape bias. The model replicates the pattern of results of our first experiment and shows how individuals with autism are more likely to categorize experimental objects together on the basis of their novelty. It also provides insights into possible mechanisms by which children with autism learn new words, and why their word referents may be idiosyncratic. Our model highlights a developmental approach to autism that emphasizes deficient representations of categories underlying an impaired shape bias.

Consider the category: The effect of spacing depends on individual learning histories.

The spacing effect refers to increased retention following learning instances that are spaced out in time compared with massed together in time. By one account, the advantages of spaced learning should be independent of task particulars and previous learning experiences given that spacing effects have been demonstrated in a variety of tasks across the lifespan. However, by another account, spaced learning should be affected by previous learning because past learning affects the memory and attention processes that form the crux of the spacing effect. The current study investigated whether individuals' learning histories affect the role of spacing in category learning. We examined the effect of spacing on 24 2- to 3.5-year-old children's learning of categories organized by properties to which children's previous learning experiences have biased them to attend (i.e., shape) and properties to which children are less biased to attend (i.e., texture and color). Spaced presentations led to significantly better learning of shape categories, but not of texture or color categories, compared with massed presentations. In addition, generalized estimating equations analyses revealed positive relations between the size of children's "shape-side" productive vocabularies and their shape category learning and between the size of children's "against-the-system" productive vocabularies and their texture category learning. These results suggest that children's attention to and memory for novel object categories are strongly related to their individual word-learning histories. Moreover, children's learned attentional biases affected the types of categories for which spacing facilitated learning. These findings highlight the importance of considering how learners' previous experiences may influence future learning.

How we categorize objects is related to how we remember them: The shape bias as a memory bias.

The "shape bias" describes the phenomenon that, after a certain point in development, children and adults generalize object categories based on shape to a greater degree than other perceptual features. The focus of research on the shape bias has been to examine the types of information that learners attend to in one moment in time. The current work takes a different approach by examining whether learners' categorical biases are related to their retention of information across time. In three experiments, children's (N=72) and adults' (N=240) memory performance for features of objects was examined in relation to their categorical biases. The results of these experiments demonstrated that the number of shape matches chosen during the shape bias task significantly predicted shape memory. Moreover, children and adults with a shape bias were more likely to remember the shape of objects than the color and size of objects. Taken together, this work suggests that the development of a shape bias may engender better memory for shape information.

Young children retain fast mapped object labels better than shape, color, and texture words.

We compared short- and long-term retention of fast mapped color, shape, and texture words as well as object labels. In an exposure session, 354 3- and 4-year-old children were shown a set of two familiar and three novel stimuli. One of the novel stimuli was labeled with a new object label, color, shape, or texture word. Retention of the mapping between the new word and the novel object or property was measured either 5 min or 1 week later. After 5 min, retention was significantly above chance in all conditions. However, after 1 week, only the mappings for object labels were retained above chance levels. Our findings suggest that fast mapped object labels are retained long term better than color, shape, and texture words. The results also highlight the importance of comparing short- and long-term retention when studying children's word learning.

Children prefer pattern over shape during complex categorization.

Numerous studies have shown that children tend to view objects with similar shapes as having the same category. However, these studies often adopt simple categorization tasks and ignore the perceptual dimension (e.g., surface pattern of objects) that likely attract children's attention. The purpose of this study was to test how children categorize when pattern competes against shape. In Experiment 1a children were presented with a target and several testing objects that shared the same shape, color, or texture as the target. The results indicated that children preferentially selected the shape-sharing objects. However, when the texture was replaced by pattern (Experiment 1b), there was no significant difference between shape and pattern choices. When shared features were intricately overlapped between different pairs of stimuli (Experiment 2), children preferentially chose objects that shared patterns over those that shared shapes. These findings are the first to reveal children's pattern preference in categorization, supporting the view that children's categorization is flexible.

Hiding a plane with a pixel: examining shape-bias in CNNs and the benefit of building in biological constraints.

When deep convolutional neural networks (CNNs) are trained "end-to-end" on raw data, some of the feature detectors they develop in their early layers resemble the representations found in early visual cortex. This result has been used to draw parallels between deep learning systems and human visual perception. In this study, we show that when CNNs are trained end-to-end they learn to classify images based on whatever feature is predictive of a category within the dataset. This can lead to bizarre results where CNNs learn idiosyncratic features such as high-frequency noise-like masks. In the extreme case, our results demonstrate image categorisation on the basis of a single pixel. Such features are extremely unlikely to play any role in human object recognition, where experiments have repeatedly shown a strong preference for shape. Through a series of empirical studies with standard high-performance CNNs, we show that these networks do not develop a shape-bias merely through regularisation methods or more ecologically plausible training regimes. These results raise doubts over the assumption that simply learning end-to-end in standard CNNs leads to the emergence of similar representations to the human visual system. In the second part of the paper, we show that CNNs are less reliant on these idiosyncratic features when we forgo end-to-end learning and introduce hard-wired Gabor filters designed to mimic early visual processing in V1.

Brain Measures of Toddlers' Shape Recognition Predict Language and Cognitive Skills at 6-7 Years.

While a number of studies have found that an improvement in object shape recognition is associated with language growth in infants and toddlers, no published studies have investigated the longitudinal relation between early shape recognition, and language abilities in later childhood. An electrophysiological measure of semantic processing (the N400) was used to assess shape recognition and general object recognition in a naming context in 20-month-olds. The measures of shape recognition strongly predicted language and cognitive abilities at 6-7 years even after controlling for toddler vocabulary size. The electrophysiological measures of general object recognition were not related to future language or cognitive abilities. These results suggest that early shape recognition abilities may play a role in language acquisition and influence even long-term language outcomes.

Reproducibility and a unifying explanation: Lessons from the shape bias.

The goal of science is to advance our understanding of particular phenomena. However, in the field of development, the phenomena of interest are complex, multifaceted, and change over time. Here, we use three decades of research on the shape bias to argue that while replication is clearly an important part of the scientific process, integration across the findings of many studies that include variations in procedure is also critical to create a coherent understanding of the thoughts and behaviors of young children. The "shape bias," or the tendency to generalize a novel label to novel objects of the same shape, is a reliable and robust behavioral finding and has been shown to predict future vocabulary growth and possible language disorders. Despite the robustness of the phenomenon, the way in which the shape bias is defined and tested has varied across studies and laboratories. The current review argues that differences in performance that come from even seemingly minor changes to the participants or task can offer critical insight to underlying mechanisms, and that working to incorporate data from multiple labs is an important way to reveal how task variation and a child's individual pathway creates behavior-a key issue for understanding developmental phenomena.

Feature Biases in Early Word Learning: Network Distinctiveness Predicts Age of Acquisition.

Do properties of a word's features influence the order of its acquisition in early word learning? Combining the principles of mutual exclusivity and shape bias, the present work takes a network analysis approach to understanding how feature distinctiveness predicts the order of early word learning. Distance networks were built from nouns with edge lengths computed using various distance measures. Feature distinctiveness was computed as a distance measure, showing how far an object in a network is from other objects based on shared and non-shared features. Feature distinctiveness predicted order of acquisition across all measures: Words that were further away from other words in the network space were learned earlier. The best distance measures were based only on non-shared features (object dissimilarity) and did not include shared features (object similarity). This indicates that shared features may play less of a role in early word learning than non-shared features. In addition, the strongest effects were found for visual form and surface features. Cluster analysis further revealed that this effect is a localized effect in the object feature space, where objects' distances from their cluster centroid were inversely correlated with their age of acquisition. Together, these results suggest a role for feature distinctiveness in early word learning.

Are Children with Autism Spectrum Disorder Initially Attuned to Object Function Rather Than Shape for Word Learning?

We investigate the function bias--generalising words to objects with the same function--in typically developing (TD) children, children with autism spectrum disorder (ASD) and children with other developmental disorders. Across four trials, a novel object was named and its function was described and demonstrated. Children then selected the other referent from a shape match (same shape, different function) and function match (same function, different shape) object. TD children and children with ASD were 'function biased', although further investigation established that having a higher VMA facilitated function bias understanding in TD children, but having a lower VMA facilitated function bias understanding in children with ASD. This suggests that children with ASD are initially attuned to object function, not shape.

Attentional Learning Helps Language Acquisition Take Shape for Atypically Developing Children, Not Just Children with Autism Spectrum Disorders.

The shape bias-generalising labels to same shaped objects-has been linked to attentional learning or referential intent. We explore these origins in children with typical development (TD), autism spectrum disorders (ASD) and other developmental disorders (DD). In two conditions, a novel object was presented and either named or described. Children selected another from a shape, colour or texture match. TD children choose the shape match in both conditions, children with DD and 'high-verbal mental age' (VMA) children with ASD (language age > 4.6) did so in the name condition and 'low-VMA' children with ASD never showed the heuristic. Thus, the shape bias arises from attentional learning in atypically developing children and is delayed in ASD.

Development of the shape bias during the second year.

The shape bias is an attentional preference children show for the shape of an object over other aspects of the object in a word-learning context. This bias, which aids in establishing a word-object pairing, was investigated in 12-, 18-, and 24-month-old children (n = 90) across noun, adjective, and no-label conditions. The present research presents evidence of development across this time span; there was a transition from a label reducing the chance of shape extensions to indiscriminate shape extensions to a label increasing the chance of shape extensions. This research supports the notion that children are focusing their extensions more toward shape during the course of development thereby developing a more mature and more specialized shape bias.

Learning to manipulate and categorize in human and artificial agents.

This study investigates the acquisition of integrated object manipulation and categorization abilities through a series of experiments in which human adults and artificial agents were asked to learn to manipulate two-dimensional objects that varied in shape, color, weight, and color intensity. The analysis of the obtained results and the comparison of the behavior displayed by human and artificial agents allowed us to identify the key role played by features affecting the agent/environment interaction, the relation between category and action development, and the role of cognitive biases originating from previous knowledge.

Event-related potentials during word mapping to object shape predict toddlers' vocabulary size.

What role does attention to different object properties play in early vocabulary development? This longitudinal study using event-related potentials in combination with behavioral measures investigated 20- and 24-month-olds' (n = 38; n = 34; overlapping n = 24) ability to use object shape and object part information in word-object mapping. The N400 component was used to measure semantic priming by images containing shape or detail information. At 20 months, the N400 to words primed by object shape varied in topography and amplitude depending on vocabulary size, and these differences predicted productive vocabulary size at 24 months. At 24 months, when most of the children had vocabularies of several hundred words, the relation between vocabulary size and the N400 effect in a shape context was weaker. Detached object parts did not function as word primes regardless of age or vocabulary size, although the part-objects were identified behaviorally. The behavioral measure, however, also showed relatively poor recognition of the part-objects compared to the shape-objects. These three findings provide new support for the link between shape recognition and early vocabulary development.