Conference PaperPDF Available

Ready to Learn: Predictive Exposure to Category-Relevant Regularities Facilitates Novel Category Learning


Abstract and Figures

Prior evidence suggests that category learning can occur implicitly by detecting regular co-occurrences of features within categories. Less studied is whether regularities wherein category membership predicts other events or actions also foster category learning. Moreover, we know little about whether, and to what degree, exposure to these regularities facilitates subsequent supervised learning. Here, participants were pre-exposed to exemplars from two categories during a cover task, while uninformed of their category membership. Pre-exposure occurred under conditions in which category membership did (Predictive Mapping) or did not (Mere Exposure) predict task events to which participants responded. Baseline participants completed the same task with category-irrelevant stimuli. Subsequently, all participants were taught the categories (using pre-exposure exemplars) under explicit supervision. Whereas neither Predictive Mapping nor Mere Exposure influenced cover task performance (vs. Baseline), Predictive Mapping substantially improved subsequent supervised category learning. These findings point to latent category learning given pre-exposure to Predictive Mapping regularities.
Content may be subject to copyright.
Ready to Learn: Predictive Exposure to Category-Relevant Regularities Facilitates
Novel Category Learning
Layla Unger (
Ohio State University, Department of Psychology, 1835 Neil Avenue
Columbus, OH 43210 USA
Vladimir Sloutsky (
Ohio State University, Department of Psychology, 1835 Neil Avenue
Columbus, OH 43210 USA
Prior evidence suggests that category learning can occur
implicitly by detecting regular co-occurrences of features
within categories. Less studied is whether regularities wherein
category membership predicts other events or actions also
foster category learning. Moreover, we know little about
whether, and to what degree, exposure to these regularities
facilitates subsequent supervised learning. Here, participants
were pre-exposed to exemplars from two categories during a
cover task, while uninformed of their category membership.
Pre-exposure occurred under conditions in which category
membership did (Predictive Mapping) or did not (Mere
Exposure) predict task events to which participants responded.
Baseline participants completed the same task with category-
irrelevant stimuli. Subsequently, all participants were taught
the categories (using pre-exposure exemplars) under explicit
supervision. Whereas neither Predictive Mapping nor Mere
Exposure influenced cover task performance (vs. Baseline),
Predictive Mapping substantially improved subsequent
supervised category learning. These findings point to latent
category learning given pre-exposure to Predictive Mapping
Keywords: category learning, latent learning, implicit
learning, supervision
Learning to group distinct perceptual experiences into
categories is a fundamental cognitive ability that manifests
itself in perception, memory, reasoning, and language. Much
of research on category learning conducted to date has
investigated how we accomplish this feat under explicit
supervision conditions (e.g., Ashby, Maddox, & Bohil, 2002;
Smith et al., 2014). For example, participants are often
informed that the stimuli they will experience belong to
categories whose composition they must try to determine,
often provided with category labels, and given explicit
feedback following their categorization decisions. During
real-world category learning, access to such explicit
information is often very limited. In the absence of explicit
supervision, how does implicit category learning unfold?
Moreover, does implicit learning (or even mere exposure to
stimuli) help learners capitalize on the relatively small
amount of explicit category information they may receive?
Here, we investigate how exposure to to-be-learned stimuli
influence category learning, with an emphasis on whether
such exposure heightens the readiness with which categories
may be learned under explicit supervision.
Category Membership and Perceptual Features
In comparison to research on category learning under
explicit, supervised conditions, the body of research
conducted on implicit or incidental category learning is
relatively small, and has focused primarily on learning that is
driven by sensitivity to regularities in the perceptual
characteristics of category members. Moreover, this research
has treated learning from such regularities as entirely separate
from learning from explicit category-relevant information.
Research in this field is motivated by the observation that,
in many naturalistic categories, members of the same
category often resemble each other more strongly than they
resemble members of other categories (Rosch, 1975). Such
family resemblances may be thought of as resulting from
reliable co-occurrences of perceptual features with each other
within exemplars from the same category, but not with
features that occur in exemplars from other categories. As an
example, consider the category of “birds”: In this category,
perceptual features such as feathers, wings, and beaks
reliably co-occur, and rarely occur with features
characteristic of other categories, such as fur.
Evidence from research in infants (Quinn & Johnson,
2000; Younger & Cohen, 1983), children (Kloos & Sloutsky,
2008) and adults (Clapper & Bower, 1994; Nelson, 1984)
suggests that a sensitivity to these perceptual feature
regularities contributes to category learning from an early
age. Moreover, learners who are given implicit or incidental
exposure to category exemplars are more likely to learn
family resemblance categories, in contrast to learners given
explicit supervision, who are more likely to learn rule-based
categories. (Love, 2002; Nelson, 1984). Accordingly, in
many accounts of implicit category learning, sensitivity to
such regular co-occurrences of perceptual features plays a
key role (Quinn & Eimas, 1997; Sloutsky, 2010; Smith &
Heise, 1992). In prior empirical research, implicit,
unsupervised learning from perceptual feature regularities
has been treated as distinct from learning under explicit
supervision. The possibility that such implicit learning might
help learners take advantage of information about category
membership provided by explicit supervision remains largely
unexplored (though see Folstein, Gauthier, & Palmeri, 2010
for an investigation with equivocal results).
Category Membership and Prediction
Beyond regularities in the features of category exemplars
themselves, the environment may also convey regularities
with which category membership predicts other perceptual
experiences or goal-related actions. With respect to
perceptual experiences, an exemplar’s category membership
may predict the location in which it is seen (e.g., dogs
typically appear on the ground, whereas birds often appear in
the sky), the other entities with which it is seen (e.g., a dog is
often seen with a leash), the type of motion it will exhibit
(e.g., birds often fly, whereas dogs do not), the words that are
heard when it is seen (e.g., the words “dog”, “furry”, etc. are
more likely to be heard when seeing a dog than a bird, even
in the absence of explicit labeling events), and so on.
Similarly, with respect to goal-related actions, an exemplar’s
category membership may predict whether it is an entity to
approach or avoid, the appropriate way in which to interact
with it (e.g., a nail should be hammered, whereas a screw
should be turned with a screwdriver), and so on. Importantly,
these prediction regularities are often implicit: We typically
do not receive explicit feedback when we make a correct or
incorrect prediction. Instead, when predicting a perceptual
experience, we might see what we expect or be surprised, and
when predicting a goal-related action, we might accomplish
or fail at our goal. Both the possibility that such implicit
prediction regularities contribute to category learning, and
how category learning that is fostered by prediction
regularities unfolds, remain relatively unstudied.
To our knowledge, the only research in which implicit,
prediction-driven category learning has been studied to date
consists of a handful of studies conducted in the auditory
domain (e.g., Gabay, Dick, Zevin, & Holt, 2015). In these
studies, participants completed a task in which they made
different responses to visual stimuli that were different from
each other either in appearance, or in the location in which
they appeared. For example, in one version of this task
(Gabay et al., 2015), participants indicated in which of 4
possible locations an X appeared, but were not given
feedback about whether they responded correctly. Critically,
the different visual stimuli were each preceded by task-
irrelevant exemplars from the same number of acoustic or
speech sound categories. The absence of corrective feedback
during the task rendered the predictive mapping between
auditory category and perceptual events/visuomotor
responses implicit. Across studies, when the category
membership of the sounds predicted the location or identity
of the stimulus to which participants respond, participants
showed evidence of having learned both this predictive
mapping, and the relevant auditory categories. Moreover, the
category learning that took place in versions of this paradigm
generalized to novel category exemplars.
The success with which category learning took place in
these studies provides initial evidence that predictive
relationships between category membership and perceptual
events/goal-related actions fosters category learning.
However, this evidence consists of a small number of studies
conducted within the auditory domain, which recruits
different brain systems (Ley et al., 2012; Seger & Miller,
2010) and contains qualitatively different categories (e.g.,
ones that are temporally dynamic) from those in the otherwise
more extensively studied visual domain. Moreover, as in
research on implicit category learning from perceptual
feature regularities, this line of research has not investigated
whether implicit learning from prediction regularities
facilitates subsequent learning from explicit supervision.
Present Experiment
As described above, category learning may be facilitated
implicitly by sensitivities both to the reliable co-occurrence
of perceptual features in category exemplars, and the
reliability with which category membership predicts
perceptual events or goal-directed actions. However, the
effects on category learning of the latter are comparatively
unstudied. Moreover, we know little about whether either
sensitivity improves the readiness with which categories are
learned under explicit supervision. These gaps in our current
understanding of category learning led to our two aims. First,
we aimed to compare category learning under conditions in
which participants were either only exposed to category-
relevant perceptual feature co-occurrences, or additionally
exposed to regularities in which category membership
predicted visual events and goal-related actions. Second, we
aimed to test whether learning under either condition
facilitated subsequent supervised category learning.
To accomplish these aims, we exposed participants to
stimuli that, unbeknownst to them, were members of two
categories within which perceptual features reliably co-
occurred. We accomplished this exposure in the context of a
speeded cover task, which we manipulated such that the
category membership of an exemplar did or did not predict
task events and appropriate responses. Specifically, in the
cover task, participants were given a short period of time
(<500ms) during which to indicate whether a stimulus that
first appeared in a central location had then “jumped” to the
left or right. In the “Predictive Mapping” condition, category
membership perfectly predicted the location to which stimuli
jumped, whereas in the “Mere Exposure” condition, category
membership and jump location were unrelated. As a control,
participants in a Baseline condition completed the same task
with stimuli unrelated to the categories in the former two
conditions. Importantly, participants were not asked to
predict where the stimulus jumped, nor were they given
explicit corrective feedback about whether their responses
were accurate (they were only informed when they had failed
to respond within the time allowance given). Accordingly,
the predictive regularities in the Predictive Mapping
condition were incidental to task performance. After
completing the cover task, all participants then were taught
the two categories under traditional explicit supervised
conditions with corrective feedback.
Using this approach, we investigated: 1) Whether
participants showed evidence of learning the Predictive
Mapping, as indexed by more rapid improvements in RT over
the course of the cover task, and 2) Whether Mere Exposure,
Predictive Mapping, or both facilitated subsequent
supervised category learning.
Participants were 72 adults (Mage=34.21 SDage=11.89)
recruited from Amazon Mechanical Turk who received $1 for
participation for this ~15min study. Participants were
randomly assigned to one of three conditions (N=24):
Predictive Mapping, Mere Exposure, and Baseline.
Category Exemplars The primary stimuli used in this
experiment were exemplars from two categories: Flurps and
Jalets. These exemplars were colorful images of “creatures”
similar to those used in prior category learning research
conducted by Deng and Sloutsky (e.g., Deng & Sloutsky,
2012). Creatures were composed of 7 binary-valued features
including a head, antennae, body, hands, feet, tail, and button.
Category membership was based on a combination of
deterministic and probabilistic features, such that each
category had a family resemblance structure. Specifically,
one feature (antennae) was perfectly associated with category
membership, and therefore deterministic, whereas five were
associated with membership in a given category in 80% of
exemplars, and therefore probabilistic. The remaining feature
occurred equally often in exemplars from both categories,
and was therefore irrelevant to category membership. This
category structure is summarized in Table 1.
Baseline Exemplars An additional set of creatures dissimilar
in appearance from the Category Exemplars were adapted
from stimuli created by Badger and Shapiro (2012) for use in
the Baseline Condition only. Like the Category Exemplars,
these stimuli were composed of binary-valued features.
Unlike the Category Exemplars, the set of Baseline
Exemplars included all possible combinations of feature
values, and was not divided into categories.
Participants followed a link on Amazon Mechanical Turk to
the experiment, which was presented using the Gorilla™
platform. During the experiment, participants proceeded
through three phases: A Practice Phase, an Exposure Phase,
and a Supervised Category Learning Phase.
Practice Phase The purpose of the Practice Phase was to
accustom participants to the task in the subsequent Exposure
Phase (see below). The task in this phase was introduced as
the “Color Jump Game”. A schematic of this task as used in
both the Practice and Exposure Phases is provided in Fig. 1.
In the task, participants watched a star that initially
appeared on the center of the computer screen in between a
red panel on the left, and a blue panel on the right. The star
then disappeared, and reappeared on the left red panel or the
right blue panel. Participants were instructed to hit the “q”
key if the star reappeared on the left, and “p” if it reappeared
on the right. Participants were informed that they would have
only a short amount of time to respond, and that they would
receive feedback indicating whether they were correct,
incorrect, or too slow. (As noted below, corrective feedback
was not provided in the Exposure Phase version of this task.)
Participants then completed the task, which consisted of 20
trials. The star reappeared equally often on the left and right,
in a pseudorandomized order. The amount of time
Fig 1: Color Jump Game schematic. Dotted boxes denote
stimulus locations; “xxx” denotes feedback.
Table 1: Category Structure
Note: The “D” feature is deterministic, “P”s 1-5 are
probabilistic, “I” is irrelevant.
participants had in which to respond after the star reappeared
started at 500ms, then decreased in 25ms increments every 5
trials such that the allowance for the last 5 trials was 425ms.
Exposure Phase In this phase, participants continued to play
the Color Jump Game. However, the star that appeared in the
Practice Phase was replaced by: 1) Category Exemplars
whose category membership perfectly predicted whether they
reappeared on the left red or right blue panel in (Predictive
Mapping Condition), 2) Category Exemplars whose category
membership was unrelated to their reappearance locations
(Mere Exposure Condition), or 3) Baseline Exemplars in the
Baseline Condition. Participation in each of these conditions
was randomly assigned between subjects. To keep any
predictive mapping between category membership and
reappearance location implicit, participants were only given
feedback when they responded too slowly, but were not told
whether their responses were correct or incorrect.
In all conditions, participants completed 80 trials (8 blocks
of 10 trials) of the Color Jump Game. To increase the
potential speed and accuracy benefits of detecting any
predictive mapping between category membership and
reappearance location, the difficulty of the task was increased
every two blocks by reducing the RT allowance. Specifically,
the RT allowance began at 425ms, then decreased by 25ms
every 2 blocks, such that the allowance was 350ms for the
final 2 blocks. Participants were alerted to this reduction in
RT allowance at the beginning of each block in which it
occurred. The outcome measure of interest in this phase is the
rate at which participants’ RTs for accurate trials improved.
Supervised Category Learning Phase The purpose of this
phase was to investigate how well participants in each of the
conditions learned to classify the Category Exemplars into
two categories under traditional supervised learning
conditions. In this phase, participants in all conditions
completed the same task with the same stimuli. First,
participants were informed that they would be learning about
two kinds of creatures: Flurps and Jalets. Participants were
told that for each creature, they should identify whether they
think it is a Flurp or Jalet using onscreen buttons, after which
they would receive corrective feedback.
Participants then proceeded through 30 trials of this task (3
blocks of 10 trials each). On each trial, participants were
presented with a Category Exemplar, and asked whether it
was a Flurp or Jalet. After responding, participants received
a message saying “That’s a [Flurp/Jalet]! preceded by a green
checkmark if they had responded correctly, or a red X if they
had responded incorrectly. The outcome measure of interest
in this phase was participants’ accuracy at categorizing
Category Exemplars over the course of the 3 blocks.
Analyses were conducted in the R environment (R
Development Core Team, 2008) using functions in base R,
Analysis of accuracy also yielded no condition differences.
the lmer function for mixed effects regression from the lme4
package (Bates, Maechler, Bolker, & Walker, 2015), and the
Anova function for deriving F-statistics and p-values for
regression models from the car package (Fox & Weisberg,
Exposure Phase Overall, RT decreased over the Exposure
Phase. Specifically, in a mixed regression model with RT on
accurate trials as the dependent variable, Trial Number as a
fixed effect, and participant as a random effect, Trial Number
had a significant, negative coefficient (β = -0.65 ms, SE =
0.03, p < .0001).
To test whether exposure to Category Exemplars in either
the Predictive Mapping or Mere Exposure conditions
concurrently influenced performance on the Color Jump
Game during the Exposure phase, we fit regression models
for each participant in which Trial Number predicted RT on
accurate trials. We then used the regression coefficient for
Trial Number as a measure of the rate of change in RT over
the course of this phase. Finally, we analyzed whether rate of
change in RT varied across the three conditions. To conduct
this analysis, we fit a regression model in which Condition
predicted rate of RT change. This model revealed no
significant relationship between Condition and rate of RT
change (F(2,69)=0.10, p=.90)
(Fig. 2).
Supervised Category Learning Phase The purpose of this
phase was to test whether the success with which participants
learned to categorize the Category Exemplars given explicit
supervision varied according their preceding experience in
the Exposure Phase. We therefore first tested whether
participants in each condition were able to learn the
categories under explicit supervision. We found that
participants in the Predictive Mapping and Mere Exposure
conditions achieved above-chance performance in Block 1 of
the Supervised Category Learning Phase (both ps < .01); by
Block 2, participants in all three conditions achieved above-
chance performance (all ps < .01).
To investigate whether experience in the Exposure Phase
influenced the degree to which participants successfully
learned the categories, we took as our outcome variable the
accuracy with which participants categorized Category
Exemplars in each of the 30 trials in this phase, and tested
whether accuracy varied with Condition (Predictive
Mapping, Mere Exposure, and Baseline). Because the 30
trials were organized into three blocks, over which
categorization accuracy may improve, we also tested whether
accuracy varied across blocks. Specifically we fit an omnibus
mixed regression model with accuracy as the dependent
variable, Condition, Block, and their interaction as fixed
effects, and participant and trial number within a block as
random effects. We then derived F-statistics and p-values for
each of the fixed effects. This analysis revealed significant
main effects of both Condition (F(2,69)=5.22, p=.008), and
Block (F(1,2076)=20.82, p<.0001), which did not interact
(p=.16) (see Fig. 2).
To further investigate these main effects, we compared
each pair of Blocks or Conditions using three fixed effects
models. These models included the same components as the
omnibus model, but each used data from only one pair of
Blocks or Conditions. The pairwise comparisons of Blocks
revealed that Block 1 accuracy was significantly lower than
in Blocks 2 or 3 (both ps < .01), but that accuracies in Blocks
2 and 3 did not significantly differ (p=.22). More importantly,
the pairwise comparisons of Conditions revealed that the
Predictive Mapping condition exceeded both the Baseline
(F(1,46)=8.51, p=.005) and Mere Exposure (F(1, 46)=6.55,
p=.014) conditions. In contrast, accuracies in the Mere
Exposure and Baseline conditions did not significantly differ
(F(1,46)=.118, p=.733). In sum, participants in the Predictive
Mapping condition learned the categories more successfully
than either participants in the Mere Exposure or those in the
Baseline conditions, whereas the success of category learning
for participants in the latter two conditions did not differ.
Prior evidence has revealed that exposure to regularities in
which perceptual features co-occur in category exemplars can
implicitly facilitate category learning. However, we know
little about whether exposure to regularities in which
category membership predicts other perceptual events or
goal-directed actions can similarly facilitate category
learning. Moreover, we know little about how either form of
exposure may allow us to more readily learn categories under
explicit supervision.
In this experiment, we exposed participants to exemplars
of two categories that possessed distinct perceptual feature
co-occurrence regularities. This exposure took place within a
cover task, under conditions in which category membership
did (Predictive Mapping) or did not (Mere Exposure) predict
task events to which participants responded. Importantly, this
exposure was implicit: Participants were neither informed
that the stimuli they viewed were members of two categories,
nor alerted of any relationship between stimulus appearance
and task events, or given corrective feedback about their
responses to task events. For comparison, Baseline
participants completed this task with other, non-categorized
stimuli. Subsequently, all participants were taught the two
categories under traditional, explicitly supervised conditions.
Although we found no variation across exposure
conditions in performance on the cover task, participants who
were exposed to the category exemplars under Predictive
Mapping subsequently learned the categories under explicit
supervision substantially more successfully than participants
in either the Mere Exposure or Baseline conditions. In
contrast, participants in the Mere Exposure condition were
not measurably more accurate than those in the Baseline
condition. These findings suggest that exposure to category
exemplars under conditions in which category membership
predicts other perceptual events and goal-directed actions
improves the readiness with which categories are learned
given explicit supervision. Exposure to predictive regularities
may therefore promote latent learning (e.g., Kimble &
BreMiller, 1981), in which such exposure promotes the
formation of mental representations that render learners more
receptive to explicit instruction.
Open Questions
The present experiment sets a foundation for further
investigation into both the influence on category learning of
category-relevant prediction regularities, and how exposure
to category-relevant regularities may facilitate subsequent
supervised category learning.
One question to explore in future research is raised by the
superiority of the Predictive Mapping condition during
supervised category learning, despite the lack of condition
effects during the Exposure Phase. Specifically, the
superiority of supervised category learning in the Predictive
Mapping condition suggests that participants were sensitive
to the regularity with which category membership predicted
Exposure Phase task events and appropriate responses. Such
sensitivity could, in principle, have allowed participants to
increasingly anticipate events in the Exposure Phase task, and
therefore improve more rapidly than participants in the other
conditions. However, we observed no such effect, as
evidenced by comparable RT decreases across conditions in
the Exposure Phase (the Color Jump Game). We therefore do
Fig 2: RTs during Exposure Phase (top) and Accuracies
during Supervised Category Learning Phase (bottom)
not have evidence of sensitivity to predictive regularities
while they were being experienced. One possible explanation
for the lack of condition differences during the Exposure
Phase is that the rate at which we shortened the RT allowance
in the Exposure Phase task pushed participants to the limit of
the rate at which they were able to speed up. Future research
might address this issue in multiple ways. For example, a
future version of this experiment might include “guess” trials
interspersed within the Exposure Phase, in which participants
are asked to predict task events. Sensitivity to predictive
regularities could therefore manifest as the gradual
achievement of above-chance performance on guess trials in
the Predictive Mapping, but not the Mere Exposure
Another key direction for future research is to illuminate
the nature of what is learned via implicit sensitivity to
category-relevant regularities. For example, does exposure to
these regularities increase attention to category-relevant
features, and away from those that are irrelevant?
In this experiment, we found that supervised category
learning was facilitated when preceded by implicit exposure
to exemplars from family resemblance categories under
conditions in which category membership predicted other
perceptual events and goal-directed actions. This facilitation
occurred above and beyond the effect of mere exposure to the
exemplars themselves. These findings point to implicit
learning due to exposure to category-relevant predictive
regularities that in turn helps learners capitalize on explicit
information about category membership.
This work was supported by National Institutes of Health
Grants R01HD078545 and P01HD080679 to Vladimir
Ashby, F. G., Maddox, W. T., & Bohil, C. J. (2002).
Observational versus feedback training in rule-based and
information-integration category learning. Memory &
Cognition, 30, 666-677.
Badger, J. R., & Shapiro, L. R. (2012). Evidence of a
transition from perceptual to category induction in 3-to 9-
year-old children. Journal of Experimental Child
Psychology, 113, 131-146.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015).
Fitting Linear Mixed-Effects Models Using lme4. Journal
of Statistical Software, 67, 1-48.
Clapper, J. P., & Bower, G. H. (1994). Category invention in
unsupervised learning. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 20, 443.
Deng, W., & Sloutsky, V. M. (2012). Carrot eaters or moving
heads: Inductive inference is better supported by salient
features than by category labels. Psychological Science,
23, 178-186.
Folstein, J. R., Gauthier, I., & Palmeri, T. J. (2010). Mere
exposure alters category learning of novel objects.
Frontiers in Psychology, 1, 40.
Fox, J., & Weisberg, S. (2011). An R Comparison to Applied
Regression (Second Edition ed.). Thousand Oaks, CA:
Gabay, Y., Dick, F. K., Zevin, J. D., & Holt, L. L. (2015).
Incidental auditory category learning. Journal of
Experimental Psychology: Human Perception and
Performance, 41, 1124.
Kimble, D. P., & BreMiller, R. (1981). Latent learning in
hippocampal-lesioned rats. Physiology & behavior, 26,
Kloos, H., & Sloutsky, V. M. (2008). What's behind different
kinds of kinds: Effects of statistical density on learning and
representation of categories. Journal of Experimental
Psychology: General, 137, 52.
Ley, A., Vroomen, J., Hausfeld, L., Valente, G., De Weerd,
P., & Formisano, E. (2012). Learning of new sound
categories shapes neural response patterns in human
auditory cortex. Journal of Neuroscience, 32, 13273-
Love, B. C. (2002). Comparing supervised and unsupervised
category learning. Psychonomic Bulletin & Review, 9, 829-
Nelson, D. G. K. (1984). The effect of intention on what
concepts are acquired. Journal of Verbal Learning and
Verbal Behavior, 23, 734-759.
Quinn, P. C., & Eimas, P. D. (1997). A reexamination of the
perceptual-to-conceptual shift in mental representations.
Review of General Psychology, 1, 271.
Quinn, P. C., & Johnson, M. H. (2000). GlobalBeforeBasic
Object Categorization in Connectionist Networks and 2
MonthOld Infants. Infancy, 1, 31-46.
Rosch, E. (1975). Basic objects in natural categories:
Language Behavior Research Laboratory, University of
Seger, C. A., & Miller, E. K. (2010). Category learning in the
brain. Annual Review of Neuroscience, 33, 203-219.
Sloutsky, V. M. (2010). From perceptual categories to
concepts: What develops? Cognitive Science, 34, 1244-
Smith, J. D., Boomer, J., Zakrzewski, A. C., Roeder, J. L.,
Church, B. A., & Ashby, F. G. (2014). Deferred feedback
sharply dissociates implicit and explicit category learning.
Psychological Science, 25, 447-457.
Smith, L. B., & Heise, D. (1992). Perceptual similarity and
conceptual structure. In B. Burns (Ed.), Advances in
Psychology: Percepts, Concepts, and Categories (Vol. 93,
pp. 233-272). North-Holland: Elsevier.
R Development Core Team. (2008). R: A language and
environment for statistical computing. Vienna, Austria: R
Foundation for Statistical Computing.
Younger, B. A., & Cohen, L. B. (1983). Infant perception of
correlations among attributes. Child Development, 54, 858-
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Maximum likelihood or restricted maximum likelihood (REML) estimates of the parameters in linear mixed-effects models can be determined using the lmer function in the lme4 package for R. As for most model-fitting functions in R, the model is described in an lmer call by a formula, in this case including both fixed- and random-effects terms. The formula and data together determine a numerical representation of the model from which the profiled deviance or the profiled REML criterion can be evaluated as a function of some of the model parameters. The appropriate criterion is optimized, using one of the constrained optimization functions in R, to provide the parameter estimates. We describe the structure of the model, the steps in evaluating the profiled deviance or REML criterion, and the structure of classes or types that represents such a model. Sufficient detail is included to allow specialization of these structures by users who wish to write functions to fit specialized linear mixed models, such as models incorporating pedigrees or smoothing splines, that are not easily expressible in the formula language used by lmer.
Full-text available
Categorizations which humans make of the concrete world are not arbitrary but highly determined. In taxonomies of concrete objects, there is one level of abstraction at which the most basic category cuts are made. Basic categories are those which carry the most information, possess the highest category cue validity, and are, thus, the most differentiated from one another. The four experiments of Part I define basic objects by demonstrating that in taxonomies of common concrete nouns in English based on class inclusion, basic objects are the most inclusive categories whose members: (a) possess significant numbers of attributes in common, (b) have motor programs which are similar to one another, (c) have similar shapes, and (d) can be identified from averaged shapes of members of the class. The eight experiments of Part II explore implications of the structure of categories. Basic objects are shown to be the most inclusive categories for which a concrete image of the category as a whole can be formed, to be the first categorizations made during perception of the environment, to be the earliest categories sorted and earliest named by children, and to be the categories most codable, most coded, and most necessary in language.
Very little is known about how auditory categories are learned incidentally, without instructions to search for category-diagnostic dimensions, overt category decisions, or experimenter-provided feedback. This is an important gap because learning in the natural environment does not arise from explicit feedback and there is evidence that the learning systems engaged by traditional tasks are distinct from those recruited by incidental category learning. We examined incidental auditory category learning with a novel paradigm, the Systematic Multimodal Associations Reaction Time (SMART) task, in which participants rapidly detect and report the appearance of a visual target in 1 of 4 possible screen locations. Although the overt task is rapid visual detection, a brief sequence of sounds precedes each visual target. These sounds are drawn from 1 of 4 distinct sound categories that predict the location of the upcoming visual target. These many-to-one auditory-to-visuomotor correspondences support incidental auditory category learning. Participants incidentally learn categories of complex acoustic exemplars and generalize this learning to novel exemplars and tasks. Further, learning is facilitated when category exemplar variability is more tightly coupled to the visuomotor associations than when the same stimulus variability is experienced across trials. We relate these findings to phonetic category learning.
A 3-layered backpropagation connectionist network, configured as an autoassociator, learned to form global (e.g., mammal) before basic-level (e.g., cat) category representations from perceptual input. To test the predicted global-to-basic order of category learning of the network, 2-month-olds were administered the familiarization/novelty-preference procedure and examined for representation of global and basic-level categories. Infants formed a global category representation for mammals that excluded furniture but not a basic-level representation for cats that excluded elephants, rabbits, or dogs. The empirical results are consistent with the global-to-basic learning sequence observed in the network simulations.
The controversy over multiple category-learning systems is reminiscent of the controversy over multiple memory systems. Researchers continue to seek paradigms to sharply dissociate explicit category-learning processes (featuring category rules that can be verbalized) from implicit category-learning processes (featuring learned stimulus-response associations that lie outside declarative cognition). We contribute a new dissociative paradigm, adapting the technique of deferred-rearranged reinforcement from comparative psychology. Participants learned matched category tasks that had either a one-dimensional, rule-based solution or a multidimensional, information-integration solution. They received feedback either immediately or after each block of trials, with the feedback organized such that positive outcomes were grouped and negative outcomes were grouped (deferred-rearranged reinforcement). Deferred reinforcement qualitatively eliminated implicit, information-integration category learning. It left intact explicit, rule-based category learning. Moreover, implicit-category learners facing deferred-rearranged reinforcement turned by default and information-processing necessity to rule-based strategies that poorly suited their nominal category task. The results represent one of the strongest explicit-implicit dissociations yet seen in the categorization literature.
This chapter discusses perceptual similarity and conceptual structure. The relevance of perceptual similarity for conceptual structure stems from the role of real world experience on feature weights. One kind of knowledge that pushes feature weights around and stretches the similarity space is implicit knowledge about relations among perceptual features. Perceptual features do not vary orthogonally in the world. They come in causally related clusters; birds with webbed feet tend to have bills and objects with dog-like feet tend to have dog-like heads. Evidence from laboratory experiments indicates that both adults and older infants are sensitive to such correlations. This empirical evidence indicates that experience with correlations causes increased attention to the combinations of features that enter into correlations. Perceptual and conceptual similarity are not the same things. Conceptual similarity does not reduce to perceptual similarity. The chapter discusses how perceptual and conceptual similarities are different and describes the causal dependencies between the two.
The authors discuss the origins of categorical representations in young infants, using recent evidence on the categorization of animals. This evidence suggests that mature conceptual representations for animals derive from the earliest perceptually based representations of animals formed by young infants, those based on the surface features characteristic of each species, including humans. The shift from perceptually to conceptually based representation is a gradual and continuous process marked by initial, relatively simple, perceptually based representations coming to include more and more specific values of common animal properties. Development is thus a process of enrichment by perceptual systems, including that for language, and without the need of specialized processes that alter the nature of human thought and the representation of human knowledge. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
The formation of new sound categories is fundamental to everyday goal-directed behavior. Categorization requires the abstraction of discrete classes from continuous physical features as required by context and task. Electrophysiology in animals has shown that learning to categorize novel sounds alters their spatiotemporal neural representation at the level of early auditory cortex. However, functional magnetic resonance imaging (fMRI) studies so far did not yield insight into the effects of category learning on sound representations in human auditory cortex. This may be due to the use of overlearned speech-like categories and fMRI subtraction paradigms, leading to insufficient sensitivity to distinguish the responses to learning-induced, novel sound categories. Here, we used fMRI pattern analysis to investigate changes in human auditory cortical response patterns induced by category learning. We created complex novel sound categories and analyzed distributed activation patterns during passive listening to a sound continuum before and after category learning. We show that only after training, sound categories could be successfully decoded from early auditory areas and that learning-induced pattern changes were specific to the category-distinctive sound feature (i.e., pitch). Notably, the similarity between fMRI response patterns for the sound continuum mirrored the sigmoid shape of the behavioral category identification function. Our results indicate that perceptual representations of novel sound categories emerge from neural changes at early levels of the human auditory processing hierarchy.
We examined whether inductive reasoning development is better characterized by accounts assuming an early category bias versus an early perceptual bias. We trained 264 children aged 3 to 9 years to categorize novel insects using a rule that directly pitted category membership against appearance. This was followed by an induction task with perceptual distractors at different levels of featural similarity. An additional 52 children were given the same training followed by an induction task with alternative stimuli. Categorization performance was consistently high; however, we found a gradual transition from a perceptual bias in our youngest children to a category bias around 6 or 7 years of age. In addition, children of all ages were equally distracted by higher levels of featural similarity. The transition is unlikely to be due to an increased ability to inhibit perceptual distractors. Instead, we argue that the transition is driven by a fundamental change in children's understanding of category membership.