The present study integrates ideas and approaches from studies of psychophysics and conceptual development. Preschool age children were presented with creatures that either fit into a prototypical category or had features/labels from competing categories. Children were asked to complete classification (e.g., identify a creature's category) and induction (e.g., identify a creatures missing features, given a category label) trials. The feature dimensions tested, shape and color, were selected in order to evaluate the relative salience of different kinds of features. When presented with prototypical category members, children's classifications and inductions were accurate. When presented with creatures with counter-predictive features, neither feature was found to unduly influence children's classifications. In contrast, when children were asked to make inductions about creatures with features that conflicted with their category labels, one feature - color - was found to significantly influence inductions while the other feature - shape - did not. This finding indicates that consideration of psychophysical properties is required in order to accurately interpret studies exploring children's conceptual development.
Relatively early in development, children exhibit an
impressive ability to organize the world around them.
With relatively sparse input, as limited as a verbal label
from a peer or adult, children can form categories that
guide their classifications of objects, individuals, and
events. Moreover, children use these categories to make
projective inductions. They apply their knowledge of
known category members to make educated guesses
about the features of novel category members. So, if a
child knows that their pet cat is safe to approach when
it is purring and dangerous to approach when it wags its
tail, then they can apply that knowledge to draw
inferences about new cats that they encounter
The fact that children adeptly navigate categorization
and induction is uncontroversial. However, the source
of these competencies is a perpetual source of debate
within the field of child development. Although there
are a number of competing theories and models that
might be used to explain how children use and acquire
categorization and induction behaviors, these
explanations can be roughly condensed into two
approaches. Similarity-based approaches focus
primarily on associative, perceptual, and statistical
factors when investigating and characterizing children’s
competencies
These two approaches have remained at odds for
decades because they have complementary advantages
and disadvantages. Similarity-based approaches are
elegant because of their simplicity, deriving
explanatory power from domain-general cognitive
processes related to attention and perception. At some
level, these processes, and their related brain areas,
must be recruited in perceiving and learning concepts
and categories. However, hypotheses and models
grounded in a similarity-based approach have had
difficulty addressing some basic characteristics of
concepts and conceptual development. For example,
children’s inductive inferences are commonly guided
by category membership, even when perceptual
features are available and informative
Theory-based approaches have little difficulty addressing these complications because they attribute deep, and sometimes complex, naïve or intuitive theories to individuals, including young children. Thus, where similarity-based approaches are elegant and efficient, theory-based approaches presuppose substantial innate conceptual acumen. Conversely, theory-based accounts address, and in some cases predict, a wide range of behaviors and intuitions in children and adults that are problematic for similaritybased accounts.
Today, the work produced by researchers employing
each of these approaches continues in parallel,
advancing separate theoretical agendas without directly
addressing the friction between views. However, there
are some notable exceptions. Specifically, Sloutsky and
colleagues
The goal of the current project is to begin evaluating the claims made by Waxman and Gelman using the psychophysical approaches developed by Sloutsky and colleagues. Doing so requires integrating research from perception and psychophysics into approaches used to study children’s conceptual development. This pairing is not new to the field, but the pairing of perceptual methods with developmental topics is particularly challenging because child participants lack the expertise and attention span of adults.
As a first step toward the goal of more tightly
integrating research on perception and conceptual
development, the current project evaluates whether
certain stimulus dimensions influence children’s
classification and projective induction more than others.
There is evidence that some features are more salient
than others
As an initial step, this study focuses on two common
features typically used in studies of classification and
induction: color changes and shape changes. A third
variable, the presence or absence of a third feature, was
also employed in order to add additional variety and
complexity to the test stimuli. The paradigm used in
this study is schematically similar to the design
developed by
Prior studies of classification and induction, particularly using the approaches employed in the present study, largely focus on children between the ages of three and five. Thus, that age group was the focus of the present study. Fourteen preschool aged children (M = 4.64, SD = .25, range: 4.13 to 5.05, 7 female) were recruited from daycares in urban and suburban settings in the greater metro area of Louisville, Kentucky. Children were tested individually by an experimenter at their school, and they received a certificate and sticker as rewards for participation. Data from one additional child was excluded from analysis because the child failed to follow directions.
The materials used in this study consisted of drawings of artificial creatures. Each creature had three features, including color, a “bottom” feature (i.e., a tail), and a “top” feature (i.e., a mouth or horns). The key features in this study were the color and tail feature. As in prior studies, each of these features was set to one of two binary values. Color was either a fully bright and saturated value for red (rgb = 255, 0, 0) or the same color at 50% the brightness. The bottom or tail feature was one of two triangles. Both triangles had the same area, but one was four times the height of the other. The third and final feature was either present or absent on top of each creature. Two artificial prototypes were created by randomly assigning one color and tail to create two prototype creatures for two categories, flurps and jalets, see Figure 1. The third feature appeared on half of the creatures and was category neutral (i.e., each individual flurp or jalet had a 50% chance of having a top feature). In addition to the prototypical creatures, mismatch creatures were generated by mixing features across categories (e.g., a flurp color with a jalet tail, in classification trials) or presenting a creature with a diagnostic feature from one category with a label from the other category (e.g., a creature labeled as a flurp that is the color of a jalet, in induction trials). Other materials included creatures with competing features from both prototypes, color patches depicting both values of color used in the stimuli, black and white line drawings of creatures, disembodied tails, and two audio tracks. The audio tracks consisted of a woman’s voice saying, “this is a flurp,” and “this is a jalet.” Each was approximately four seconds long.
Participants were tested individually in a quiet space at their preschool. Stimuli were displayed using presentation software on a laptop with a 15-inch screen, and the experimenter determined the pace of the session, ensuring that children were looking at the screen before beginning each trial. Three kinds of trials were presented to participants, including training trials, classification trials, and induction trials.
Training Trials The experiment began when the experimenter said, “I’m going to show you two kinds of creatures, flurps and jalets. I want you to look at them and remember what they look like because I will ask you about them later.” Children then saw each prototypical creature six times, for a total of 24 training trails. These trials were split evenly between the two tobe-learned categories, and they depicted prototypical jalets and flurps. Each was accompanied by a four second audio track that both labeled each creature and indicated when the experimenter should progress to the next trial (i.e., exposure to each trial was approximately 4 seconds). Training trials were presented in a random order for each participant.
Classification Trials Classification trials began with the following instructions: “I’m going to show you some more creatures that look a lot like the ones before. This time I’m going to show you a creature and ask you to tell me if it’s a flurp or a jalet.” Classification trials were identical to training trials except that the creatures were not labeled. After each trial, the experimenter asked if the creature was a jalet or a flurp. Participants were presented with eight warm-up trials, divided evenly between categories so that each prototypical creature appeared twice. These trials were presented with feedback. If a creature was incorrectly classified, the experimenter said, “Oops, that’s a _____." Test trials without feedback began immediately after the 8th warm-up trial, and consisted of 16 trials, including two instances of each prototypical creature and two instances of each mismatch creature. Mismatch creatures were constructed using one feature from each category. Classification trials were presented pseudorandomly. The first two trials were always prototypical creatures, but the remaining trials were intermixed and presented randomly.
Induction Trials After the final classification trial, the experimenter said, “Now I’m going to show you some more creatures that look a lot like the ones before. This time I’m going to tell you if it’s a jalet or a flurp, but I’m going to ask you to guess what’s missing.” In each induction trial, participants were presented with labeled creatures that were missing a target feature. Because this experiment focused on color and shape changes, these features were missing from each creature, and participants were prompted to select which of two possible features was missing from the creature (e.g., “Which color/bottom is missing?” see Figure 2). Induction trials featured a warm-up and test phase with the same number, disposition, and pseudorandomization as the previously displayed classification trials
identifications on 80% of trials, which was significantly greater than chance responding, t(13) = 4.67, p < .001. Since the informative features on each mismatch creature were counter-predictive, accuracy for mismatch trials was expected to be random if the tested features were equivalently salient. For the purpose of detecting differences in salience, accuracy on these trials was therefore arbitrarily defined as a classification that was consistent with each mismatch creature’s tail. As can be seen in Figure 3, participants responded rationally and randomly (M = 49%) when presented with conflicting features.
Classification Accuracy 100
Induction For the purpose of evaluating inductions,
label-consistent responses were defined as “accurate.”
In the case of prototypes, this means that participants
chose the missing feature that matched both the label
and the visible informative feature. In the case of
mismatches, responses were labeled as accurate when
they aligned with labels
Participants’ inductive inferences were evaluated with a 2x2 repeated measures ANOVA with Featuretype (color vs. shape) and Creature-type (prototype vs. mismatch) as within-subjects factors. This analysis revealed a significant main effect of Creature-type, F(1,13) = 7.32, p < .05, ηp2 = .36, indicating that accuracy for prototypes was significantly greater than mismatches. Planned comparisons (Bonferronicorrected) further revealed that the difference between prototype and mismatch creatures was only significantly different when color was presented in opposition to labels, p < .05. To put this finding in perspective, the means for prototypes and mismatches across both Feature-types were compared to chance responding using one-sample t-tests (chance = .5). Every set of responses tested in this study significantly exceeded chance responding (p’s < .05) except when labels were placed in competition with visible colors.
Induction Accuracy 100 80 60 40 20 0 Judge Color Judge Shape
Children’s classifications were very accurate when they were presented with prototypical creatures. This result indicates that they learned the target categories and remembered them. When presented with mismatch creatures, children responded randomly but rationally. Because mismatch creatures were constructed of features from competing categories (e.g., a jalet’s tail paired with a flurp’s color), children had only conflicting evidence to guide their classifications. If children had a predisposition to only learn a single feature, or to view one feature as more informative for classification, then children’s responses would have been non-random. Instead, their random responding indicates the quality of their learning and their lack of biases with respect to feature dimensions when making classification judgments.
Children’s inductions revealed a different pattern of behaviors. As in classification trials, children were accurate when making inductions about features missing from prototypical creatures. Thus, when the label and the visible feature were category-consistent, children effectively identified the missing feature, regardless of which feature they were judging.
In contrast, children’s inductions diverged by Feature-type when they were presented with mismatch creatures. When they were asked to judge color, children exhibited a bias to make label-consistent judgments, as indicated by responses that significantly differed from chance but did not differ from judgments of prototypical creatures, which were always labelconsistent if accurate. However, when asked to judge shape, children’s responses were significantly less accurate (i.e., less label-consistent) than their responses to prototypical creatures, and their responses did not differ from chance.
Recall that mismatch creatures were constructed from
a label from one category and a visible feature from a
competing category. Thus, there was no “right” answer,
and either the label or the feature might have guided
children’s intuitions. In prior studies
In the current study, children’s responses revealed
that they found labels to be more salient cues to
category membership than the shape of a part, but they
found labels and colors to be equally salient. This
pattern of results indicates that there were important
differences in salience between the stimulus dimensions
of color and shape. Broadly, these results represent
evidence that stimulus dimensions are not all equally
salient or informative.
The pattern of results recorded here both supports
and undermines similarity-based accounts of
representation and conceptual development. On one
hand, these data reveal that results focusing on salience
may actually be tapping into differences between
stimulus dimensions, and not manipulations of
attention. For example, Deng and Sloutsky’s effect
might be attributable to the part that they selected to
make more salient, and not to the manipulation of
salience that they applied. On the other hand, these
results support claims that feature salience is important
and influences inductions. However, these results also
provide a mechanism by which certain kinds of
comparisons can be combined with specific features in
order to provoke findings that support a false
dichotomy between perceptual and conceptual factors,
as suggested by
Perhaps the most important outcome of this study is
to highlight a fundamental problem with studies that
probe conceptual representations. Specifically, broad
claims about the role of labels or feature salience
cannot be evaluated without a strong understanding of
the between- and within-feature contrasts that they
represent. It is important to acknowledge that decisions
made while designing auditory and visual stimuli may
influence or provoke certain response patterns in young
participants. Historically, these factors have received
little attention, even though they have recently been
identified as powerfully influencing both child and
adult participants
Studies of psychophysics focusing on child participants are relatively rare. Preschoolers especially are not well suited to the number of trials or the kinds of manipulations usually employed by such approaches. However, it is increasingly obvious that the mechanics of perception and attention are playing important and under-represented roles in our modern understanding of children’s conceptual development. The purpose of the current project to address this gap in our understanding, and to begin to contextualize and understand how past design decisions and experimental methodologies have shaped our modern understanding of children’s categorization and induction behaviors.
Future studies should focus on exploring other popular manipulations, such as changes in size and positioning, as well as the presence or absence of features, which was a random factor in the present study but is currently being studied in ongoing projects. Contrasting values of these features are employed in many studies, but they are not well understood. There are also findings and theories in perception that are highly relevant to interpreting findings using different stimulus dimensions (e.g., the work of Wendell Garner), but that are not widely integrated into modern studies of child development. More generally, these studies may hold the key to addressing ongoing debates between theory- and similarity-based approaches to understanding conceptual development.
We thank the staff, children, and parents at Keneseth Israel Preschool, Newburg Kindercare, and St. Paul School in Louisville, KY for their support. Thank you to Emma Groskind, Jacob Messmer, Kristain Pile, Kayla Renner, and Gretchen Santana for their assistance.