Evidence for Knowledge-Based Category Discrimination in Infancy

by Sabina Pauen
Citation
Title:
Evidence for Knowledge-Based Category Discrimination in Infancy
Author:
Sabina Pauen
Year: 
2002
Publication: 
Child Development
Volume: 
73
Issue: 
4
Start Page: 
1016
End Page: 
1033
Publisher: 
Language: 
English
URL: 
Select license: 
Select License
DOI: 
PMID: 
ISSN: 
Abstract:

Child Development, July / August 2002, Volume 73, Number 4, Pages 1016-1033
Evidence for Knowledge-Based Category Discrimination in Infancy
Sabina Pauen
Two studies examin~dwhether infan~s' ~at~~ory discrimination in an object-examination task was based solely
on an ad h.oc analysis .of rerceptual similarities among the experimental stimuli. In Experiment lA, Ll-montholds
examined four different ~~emplars of one superordinate category (animals or furniture) twice, followed
by a new exemplar?f the famII~ar cat~gory and an exemplar of the contrasting category. Group A (N = 39) expl~
r~~ natur~l-iooking toy repl.Icas :"Ith low between-category similarity, whereas group B (N = 40) explored
artificial-looking ~oy models wIth. high bet~een-category similarity. Experiment IB (N = 40) tested a group of
IO-m~nth-olds with the .s~~e desI?n. ExperI~ent ic (N = 20) reversed the order of test trials. For Experiment
2 (N - 20), the same artIfiCIa~-lookingtoy animals as in Experiment I (group B) were used for familiarization),
b~t no category change was Introduced at the end of the session. Infants' responses varied systematically only
with the pres~nce of a category change, and not with the degree of between-category similarity. This supports
the hypothesis that performance was knowledge based.
INTRODUCTION
Wh~t kind of information is crucial for category formation
at a preverbal age? Is it information about the
appearance of objects, information about functional
or causal relations, or both? These questions still provoke
heated debates among psychologists interested
in the origins of human cognition (see Goldstone &
Barsalou, 1998; Madole & Oakes, 1999; Mandler,
2000a). One group of researchers suggests that the
first ~ategories ?f preverbal infants are purely perceptual
In nature .In the sense that they are exclusively
based on attributes that can be perceived directly
(e.g., Behl-Chadha, 1996; Quinn & Eimas, 1996; Roberts,
1988; Smith & Heise, 1992). This idea is consistent
w~th tra~itional accounts of cognitive developm~~
t In which conceptual thinking requires the
ability to abstract from the immediate perception, and
to rely on verbally encoded knowledge about defining
characteristics (e.g., Bruner, Olver, & Greenfield,
1966; Inhelder & Piaget, 1964). Another group of rese~
rch~rs su?gests that conceptual thinking may
gUIde Infants categorization, even at a preverbal age
(Mandler, 2000a; Nelson, 1974; Premack, 1990). Following
these authors, the earliest categories are based
on core knowledge about causal, functional, or structural
properties (e.g., Gelman, 1990; Keil, 1991). Such
knowledge could either be innate (Fodor, 1975) or
start to develop during the first year of life (Mandler,
1992):According to the conceptual view, the ability to
perceIve and learn causal patterns is a fundamental
prerequisite for category formation at all ages (Carey
& Spelke, 1994; Keil, Smith, Simons, & Levin, 1998;
Pauen, 1999a).
Recently; the strong dichotomy between perceptionand
conception-based views has been questioned (e.g.,
Jones & Smith, 1993; Madole & Oakes, 1999; Quinn &
Eimas, 1997; Quinn, Johnson, Mareschal, Rakison, &
Younger, 2000). Among others, Madole and Oakes
(1999) have argued that the focus of future investigations
should shift toward a microanalytic consideration
of the diverse sources from which children's
knowledge is derived. Despite such new developments,
the general theoretical dissent regarding the
nature of preverbal categories remains (see Mandler,
2000a; Quinn et al., 2000). One reason may be the diversity
and inconclusiveness of empirical findings.
As discussed below, this problem is closely related to
the methods used for studying categorization in early
childhood.
Categorization research with infants less than 1
year of age is dominated by two paradigms: visualfixation
tasks (either habituation/ dishabituation or
familiarization/ preference for novelty) and the objectexamination
task. Both types of tasks employ the
same general procedure and logic: Infants are first
familiarized with a number of different exemplars
from one category (e.g., dogs). Following familiarization,
two new test stimuli are presented; one is a
new exemplar of the familiar category (e.g., a new
dog) and the other is a new exemplar of a contrasting
~ategory (e.g., a new fish). A clear preference for the
latter is typically interpreted as evidence for category
discrimination.
A general limitation of this procedure is that one
can never be sure whether infants increase attention
to the out-of-category exemplar because they recognize
the category boundary or because of some other
© 2002 by the Society for Research in Child Development, Inc.
All rights reserved. 0009-3920/2002/7304-0002
reason. This ambiguity is reduced to some extent if a
priori preferences for the out-of-category item can be
ruled out as a possible explanation of the observed
findings. Further evidence supporting the categorybased
explanation is provided by data indicating that
infants generalize familiarization to the same-category
test item. Whereas the first condition is typically met
by studies using either visual-fixation tasks or the
object-examination technique, this is not true for the
second condition. Visual-fixation studies typically fail
to report statistical comparisons of mean looking
times for the last familiarization item and the withincategory
test item (e.g., Behl-Chadha, 1996; Eimas &
Quinn, 1994; Quinn & Eimas, 1996). Therefore, these
studies do not test whether infants actually generalize
familiarization to the within-category test item.
The corresponding analysis plays a crucial role in object-
examination studies, however. If no increase in
attention from the last familiarization item to the
same-category test exemplar is observed, but an increase
in attention to the contrasting-category test
item is observed, then infants are assumed to have
categorized the stimuli (Mandler & McDonough,
1993, 1998; Oakes, Coppage, & Dingel, 1997; Oakes,
Madole, & Cohen, 1991). Mandler and McDonough
(1993, 1998) have also found that occasionally, infants
show a linear increase from the last familiarization
trial over the same- to the contrasting-category test
stimulus. They interpret this pattern as "advanced
categorization." This interpretation will be discussed
further in the Results and Discussion section of Experiment
1A. For now, the important point is that in
the examination paradigm too, generalization of familiarization
to the within-category test item suggests
that infants are distinguishing the test items on
the basis of category membership rather than for some
other reason.
Studies that have used the visual-fixation task and
the object-examination task not only differ with respect
to how they treat methodological problems, but
also show considerable variation in experimental
procedure. In a typical visual-fixation task, infants
are presented, often in pairs, with a large set of pictures
of different exemplars of a single category. During
the test phase, infants see one or more pairings of
a novel within-category exemplar and a novel out-ofcategory
exemplar. The dependent measure in visualfixation
tasks is the total amount of time that infants
spend visually fixating each stimulus. In a typical
object-examination task, infants are presented with a sequence
of three-dimensional (3-D) miniature models
of real-world objects that they are allowed to both visually
and manually explore one at a time. Following
repeated exposure to only a small number of familiar-
Pauen 1017
ization exemplars, one new same-category item and
one new different-category item are presented. The
dependent measure in these tasks is the amount of
time that infants spend actively examining each object.
Examination time is a subset of total looking time
(Oakes & Tellinghuisen, 1994; Ruff, 1986). Some researchers
(e.g., Mandler & McDonough, 1993) have
argued that examination reflects conceptual processing
to a greater degree than does looking time, which is
highly influenced by perceptual features of the stimuli.
Not surprising, diverging patterns of results have
been obtained with these tasks, leading to different
conclusions about the processes underlying infants'
performance: Visual-fixation studies have shown that
infants as young as 3 to 4 months can distinguish between
basic-level categories such as cats, dogs, fish,
horses, zebras, and giraffes (Eimas & Quinn, 1994;
Eimas, Quinn, & Cowan, 1994) or chairs, beds, sofas,
and tables (Behl-Chadha, 1996). Apparently, infants
at this age can also form superordinatelike classes,
such as mammals and furniture (Behl-Chadha, 1996).
Quinn and Johnson (2000) reported that 2-montholds
are better at distinguishing superordinate-like
classes than basic-level categories. This suggests the
existence of a global- to basic-level shift in early infancy.
Quinn and Eimas (1996, p. 23) argued that "it is
doubtful and most likely impossible that infants 3 to 4
months of age would have any conceptual knowledge
about animals and instances of furniture at the
basic level or higher." To explain category discrimination,
the authors assumed that categories are formed
"on-line" in the course of the experimental session
(see also Smith, 2000). This on-line process may be facilitated
by the fact that visual-fixation tasks generally
employ a pairwise presentation of multiple exemplars
during familiarization.
More specifically, Quinn and Eimas (1996) assumed
that young infants form a perceptually based
representation during familiarization that generalizes
to novel instances of the familiar category, but not to
an out-of-category exemplar, resulting in a looking
preference for the out-of-category exemplar during
the test phase. This interpretation includes two related
arguments: (1) categorization performance in
infancy can be explained by on-liner processes of visual
abstraction, and (2) these on-line processes are
based on perceptual information about the appearance
of the presented stimuli. (It should be pointed
out, however, that these data do not rule out the possibility
that infants might apply previously acquired
category knowledge.)
Referring to data obtained with object-examination
techniques, Mandler and McDonough (1993,
1998) came to different conclusions. These authors
1018 Child Development
found that children failed to discriminate most basiclevel
categories, such as dogs and rabbits or dogs and
fish, until the end of the first year of life. At the same
time, global-level categories such as animals and vehicles
were differentiated at 6 to 7 months of age (the
earliest age at which the object-examination task can
be employed, because younger infants are not yet
able to handle objects effectively). Response patterns
found for this type of contrast showed a shift from
simple to advanced categorization between 7 and 11
months of age. To explain these findings, Mandler
and McDonough (1993) suggested that the meanings
developed before entering the experimental situation
direct infants' attention during the test phase. This
conclusion rests partly on the argument that it should
be more difficult for young infants to abstract a globallevel
category on-line than to abstract a basic-level category
on-line after having been exposed to just a few
familiarization exemplars. After all, within-category
variability is much higher for global- than for basiclevel
categories (Rosch, Mervis, Gray, Johnson, &
Boyes-Braem, 1976). Mander and McDonough (1993)
further speculated that the presentation of 3-D toy
models may enhance the application of previously
acquired knowledge more than do pictures, because
only the toy models can be explored visually as well
as manually. According to Mandler (1992, 2000b),
infants start with the formation of rather broad conceptual
distinctions, such as the global animateinanimate
distinction (see also Gelman & Spelke, 1981).
This distinction could be based on the causal understanding
of the difference between self-initiated and
externally induced motion (e.g., Gelman, 1990; Mandler,
1992; Premack, 1990). In contrast to authors who
use the visual-fixation task, Mandler and McDonough
thus suggested that (1) pre-existing representations
playa crucial role in explaining categorization
performance in infants less than 1 year of age, and (2)
these representations include conceptual (i.e., causal,
structural, or functional) knowledge.
As this review of the current literature reveals, different
models have been developed to explain categorization
performance. These models typically combine
claims about the role of perceptual and conceptual
attributes in infants' categorization with claims about
the role of pre-existing representations that infants
may (or may not) apply in categorization tasks. However,
to enhance clarity in the theoretical debate, it
may be useful to treat these aspects separately. First, it
must be determined whether categorization performance
in preverbal infants is ever based on pre-existing
representations of the categories under investigation,
or whether infants' performance is always based on
an ad hoc analysis of the perceptual features available
in the experimental stimuli themselves. Only if it can
be shown that pre-existing representations influence
categorization performance at all would it make
sense to speculate further about the role of perceptual
and conceptual knowledge for category formation at
a preverbal age. The series of experiments to be reported
next focused on this issue.
In the present context, categorization was called
on-line when infants relied only on their immediate
perception of the given experimental stimuli to form
a category within the familiarization period. In contrast,
categorization was called knowledge based
when infants referred to some already existing representation
that was activated by the perception of the
given stimuli but included properties other than
those corresponding to the physical attributes of the
experimental material. Such properties could be
either perceptual or conceptual in nature. For example,
a given toy animal without a specific smell (perceptual
attribute) and without the ability to direct its
own behavior (conceptual attribute) may nevertheless
activate the memory of a real-world animal that
has the corresponding perceptual and / or conceptual
attributes. The crucial point is that infants' performance
was only called knowledge based if categorization
responses were at least partly based on perceptual
and / or conceptual properties belonging to a
representation that had been formed prior to the start
of the experimental session. Alternatively, performance
was called on-line if categorization responses
could be explained exclusively by an analysis of perceptual
similarities and differences among the stimuli
presented during the experimental session.
The current research investigated whether infants
less than 1 year of age showed on-line-based or
knowledge-based categorization as defined above,
using the object-examination paradigm. This task
was chosen because strong claims in favor of the
knowledge-based view appeal to data obtained with
this method. The experimental design was based on the
following general logic: If infants showed on-line categorization,
then their responses during the test phase
should vary systematically with the degree of perceptual
between-category similarity of the experimental
material. More specifically, categorization responses
should be stronger in a low between-category similarity
condition than in a high between-category similarity
condition. If infants showed knowledge-based categorization,
however, a manipulation of perceptual
between-category similarity should not greatly influence
infants' performance provided that the contrasted
categories were familiar to the infants.
Following a similar line of reasoning, Oakes et
al. (1997) tested 10- to 13-month-olds in an objectexamination
task contrasting land and sea animals in
a high and a low between-category similarity condition.
They found that performance varied systematically
with the degree of between-category similarity.
As suggested by the on-line view, infants' category discrimination
was better in the low between-category
similarity condition. However, because the authors
used different basic-level exemplars in both similarity
conditions, it is not possible to determine from the reported
findings whether the activation of different
knowledge or the manipulation of between-category
similarity of the experimental stimuli was responsible
for the observed differences in categorization performance.
In an attempt to overcome this problem, the
present experiments used identical basic-level exemplars
in the high- and in the low-similarity condition,
but varied the degree of perceptual between-category
similarity among stimuli. Given the close relation between
perceptual and conceptual attributes (Rosch et
al., 1976), choosing appropriate stimuli for this purpose
turned out to be difficult. To ensure that the
stimuli designed for this purpose were appropriate,
three preliminary experiments were conducted.
Preliminary Studies
Two different sets of stimuli were created, representing
the same basic-level exemplars of animals and
furniture items in both sets: the group of animals included
a giraffe, a zebra, a crocodile, a hippo, and a
turtle, and the group of furniture items consisted of a
cupboard, a chair, a bed, an armchair, and a stool.
These exemplars were represented by toys that differed
systematically in their appearance between sets.
In Set A (see Figure I, top), all items were toy replicas
of real-world objects with realistic natural variation.
Exemplars within each category varied greatly
in shape, color, and surface pattern; systematic perceptual
differences between categories concerned
material aspects (animals in hard plastic, wooden furniture),
characteristic parts (presence or absence of
facelike features), and abstract form attributes (curvy
borders, rounded contours, irregular colored and textured
surfaces for animals; straight borders, sharp
contours, and smooth, regularly colored surfaces for
furniture). Hence, within- and between-category similarity
were both rather low.
In Set B (see Figure I, bottom), all toy models were
made out of wood. Variation in their physical appearance
was highly controlled. They were all marked
with two black-and-white dots that could be interpreted
as eyes in the case of animals, and as knobs or
decoration in the case of furniture. Each item had
curved as well as rectilinear parts, rounded as well as
Pauen 1019
sharp edges, a smooth surface, and leglike appendages.
The furniture items were designed such that
each item had the same global shape as one animal,
the same colors as another animal, and the same surface
pattern as a third animal (see Table 1). Within
each category (animals, furniture), all five exemplars
displayed a different global shape, a different color
combination, and a different surface pattern.
Preliminary Experiment 1. The first preliminary experiment
tested whether the stimuli used in Sets A
and Bdiffered systematically with respect to the degree
of perceptual within- as well as between-category
similarity. A group of undergraduate students (age:
M == 22,1, range == 19~29 years; 10 males, 10 females)
were asked to rate similarity among pairs of objects
on a 10-point rating scale (1 == minimal similarity,
10 == maximal similarity). The experiment consisted
of two parts: In Part I, all possible object combinations
for one given set (either artificial- or naturallooking
stimuli) were presented in random order, and
in Part 2, the same was done with the remaining set.
The order of presentation of the sets was counterbalanced
across the sample. Participants were instructed
to base their similarity estimates on information
about the appearance of the stimuli only, and to disregard
what they thought the objects might represent in
the real world.
Mean scores for within-category similarity in each
set were entered as dependent variables into an analysis
of variance (ANOVA) with order of presentation
for both sets (artificial first, natural first), and
gender (male, female) as independent variables. This
analysis revealed no significant effects, indicating
that the two sets of stimuli were comparable with
regard to perceptual within-category similarity. An
ANOVA with between-category similarity for each
set as the dependent variable revealed that similarity
ratings were significantly higher for the artificiallooking
stimuli than for the natural-looking stimuli,
F(l, 16) == 33.17, P < .001. The intended manipulation
of between-category similarity therefore was
successful. Overall, similarity ratings were rather
low for both sets of stimuli (see Table 2). It is important
to note, however, that between-category similarity
was rated slightly higher than within-category
similarity for the artificial material, t(19) == -1.76,
P == .09, whereas the reverse was true for the naturallooking
stimuli, t(19) == 2.84, P< .05. In both analyses,
neither gender nor order of presentation influenced
the results.
Preliminary Experiment2. This experiment tested
whether between-category similarity among the artificial
stimuli would be rated differently when adults
applied previously acquired knowledge compared with
1020 Child Development
Figure 1 Stimuli used for Experiments lA, IB, and Ie: Natural-looking toy models of animals and furniture items with low
be tween-category similarity presented to Group A (top). Artificial-looking toy models of the same animals and furniture items
with increased between-category similarity presented to Group B (bottom). The artificial-looking animals were also presented
during familiarization in Experiment 2.
when they focused on the appearance of the given
toys alone. The task closely resembled the objectexamination
task that infants received in later experiments.
A group of undergraduate students (age:M =
24 years, range = 19-29 years; 20 males, 20 females)
were given the same series of four exemplars of one
category twice. Each object was presented for 20 s,
during which the adult was allowed to freely explore
the toy. Participants in the perceptual instruction condition
(N = 20) were asked to focus exclu sivel y on
the appearance of the stimuli, whereas tho se in the
knowledge-based instruction condition (N = 20) were
asked to consider all they knew about the nature of
the real-world objects represented by the given toys.
Each object occurred equally often in each position.
Following familiarization, one new exemplar of
the already familiar category and one exemplar of the
contrasting category were presented, with order of
test trial s counterbalanced. Each object in both categories
served as a test item equally often. Participants
rated the similarity of each test item to the previously
seen familiarization items on a 10-point rating
scale (l = minimal similarity, 10 = maximal similarity).
Rating responses were entered into an ANOVA
with instruction (perceptual, knowledge based) as the
independent variable and rating response for the samePauen
1021
Table 1 Form, Color, and Surface Pattern of Artificial-Looking Stimuli
Objects Form Color Surface Pattern
Animals
Giraffe Giraffe Yellow and brown Net shaped
Zebra Zebra Black and white Striped
Crocodile Crocodile Dark and light green Dotted
Hippo Hippo Gray and pink Plain
Turtle Turtle Red and brown Checked
Furniture
Cupboard Similar to giraffe Red and brown (turtle) Plain (hippo)
Chair Similar to zebra Dark and light green (crocodile) Net shaped (giraffe)
Bed Similar to crocodile Gray and pink (hippo) Checked (turtle)
Armchair Similar to hippo Yellow and brown (giraffe) Striped (zebra)
Stool Similar to turtle Black and white (zebra) Dotted (crocodile)
Note: Minimal similarity = 0; maximal similarity = 10.
Table 2 Within-Category Similarity and between-Category
Similarity among Natural-Looking Stimuli, and Artificial-Looking
Stimuli: Adult Ratings
M SD M SD
Natural-looking
stimuli 3.12 .98 2.62 .90
Artificial-looking
stimuli 3.14 1.02 3.41 .96
category and the different-category test exemplar as
the repeated-measurement variable. This ANOVA revealed
a main effect for instruction, F{l, 38) == 4.30,
P< .05, and a main effect for test trial, F{l, 38) == 15.64,
P < .01, as well as a significant interaction between
the two variables, F{l, 38) == 57.00, P < .01. As expected,
the perceptual instruction group rated the
same-category item as less similar to the previously
seen familiarization items (M == 3.80, SO == 2.12) than
the different-category item (M == 5.05, SO == 1.82),
t(19) == -2.80, P < .01, whereas the reverse was true
for the knowledge-based instruction group (samecategory
item: M == 5.45, SO == 1.19; different-category
item: M == 1.45, SO == 1.05), t(19) == 7.50,P< .01. These
results showed that between-category similarity was
perceived as greater than within-category similarity for
the artificial-looking material only when adults focused
on the appearance of the stimuli. Between-category similarity
was perceived as smaller than within-category
similarity when real-world knowledge was applied.
Preliminary Experiment 3. Given that adults judged
Main Experiments
Experiment 1 tested whether two groups of infants
(A, B) showed comparable performance in an objectperceptual
between-category similarity to be higher
than within-category similarity for the artificial stimuli,
one might ask whether young children would still
be able to recognize the category membership of the
artificial stimuli. An experiment with 2-year-olds (10
girls, 10 boys; age: M == 2,5, range == 2,0-2,10) investigated
this issue. Children were given one item at a
time and were asked: "Do you know what this is
called?" In this task, 60% of the animals and 55% of
the furniture items were labeled correctly at the basic
level. When children were later asked to put all animals
into a garden and all furniture into a house, the
percentage of objects sorted correctly was 95% for
the animals and 80% for the furniture items. These
findings suggest that the artificial stimuli were still
recognizable as animals or furniture, respectively.
Based on the results of all three preliminary experiments
it was concluded that perceptual betweencategory
similarity was higher for the artificial stimuli
than for the natural stimuli (Preliminary Experiment
1). Also, perceptual between-category similarity was
higher than within-category similarity for the artificial
material (Preliminary Experiment 2), and this did
not prevent young children from identifying the presented
stimuli as exemplars of their respective superordinate
category (Preliminary Experiment 3). Thus, a
comparison of infants' category performance in a study
using both sets of stimuli should reveal whether preverbal
infants' performance in an object-examination
task is more likely to be on-line based or knowledge
based in the sense defined in the Introduction.
Between-Category
Similarity
Within-Category
Similarity
1022 Child Development
examination task, when replicas of the same realworld
exemplars (i.e., animals and furniture) were
presented that differed systematically only with respect
to their perceptual between-category similarity
(high, low). The same type of task was chosen as described
by Mandler and McDonough (1993, 1998) to
ensure that the findings would be comparable across
both series of studies. In previous studies using this
method, Ll-month-olds showed a linear increase in
attention from the last familiarization trial over the
within-category test item to the out-of-category test
item. Mandler and McDonough (1993, 1998) called
this pattern of responses advanced categorization because
they assumed that infants recognized both test
items to be new in perceptual terms, but treated the
second item as being conceptually new as well.
According to this line of reasoning, the observed
categorization response is based on previously acquired
knowledge. This interpretation can be questioned,
however. Alternatively, infants may have
formed a new categorical representation during the
familiarization period, based on processes of visual
abstraction, or they may not have formed a category
at all, but rather encoded and memorized the appearance
of individual exemplars. In both cases, different
patterns of changes in mean examination time should
be produced by infants who participate in a high and
in a low between-category similarity condition. If infants
do not categorize the given items at all, or if they
form a category on-line, performance during the test
phase should vary systematically with the degree of
perceptual between-category similarity.The knowledgebased
account leads to a different prediction. If infants
show knowledge-based categorization, performance
during the test phase should depend on the presence
or absence of a change in category, even if an analysis
of perceptual between-category similarity suggests a
different pattern of responses. The following experiment
tested this hypothesis.
EXPERIMENT lA
Group A received realistic toy replicas of real-world
objects in which both within- and between-category
perceptual similarity were rather low (see Figure I,
top). Group B received toy models of the same realworld
exemplars, with the only difference being that
perceptual between-category similarity was high (see
Figure I, bottom). If Ll-month-olds did not categorize
the given stimuli at all, or if the on-line view were correct,
Group A (low between-category similarity)
should show an increase in examination time from
a new exemplar of the already-familiar category to a
new out-of-category exemplar. Group B(high betweencategory
similarity), in contrast, should show a decrease
from a new exemplar of the already-familiar
category to a new exemplar from the contrasting category,
because the out-of-category test item has more
perceptual features in common with the familiarization
stimuli than the same-category test item. On the
knowledge-based account, in contrast, no substantial
differences in infants' responding across conditions
should occur. One would expect an increase in mean
examination time from the familiar-category item to
the contrasting-category item.
Method
Participants. A total of 79 infants (age: M = 11
months, 13 days, range = 11 months-Ll months, 30
days) participated. Infants' names were drawn from
birth announcements in a local newspaper. Their families
were contacted via letters and phone calls. Half
of the infants were randomly assigned to Group A
(n = 40, 20 girls, 20 boys), and the other half were assigned
to Group B (n = 39, 21 girls, 18 boys). Four additional
infants were originally tested but started to
cry and did not complete the session. All infants remaining
in the final sample came from a White, middleclass
socioeconomic background.
Procedure. Infants were seated in a high chair. After
a short warm-up period, the experimenter placed the
first object within reach in front of the children. For 20 s,
the infants could do with the toy whatever they
wanted {e.g., play, explore, ignore}. If a given toy fell
off the table, the experimenter quickly picked it up
and placed it back on the table. A small light fixed on
the wall behind the high chair indicated the duration
of each trial. Half of the children in each condition
were familiarized with animals; the others were familiarized
with furniture. The order of presentation
for all stimuli was varied systematically across infants
in each group such that each object occurred
equally often in each position. Items were presented
in the same order to Groups A and B.During familiarization,
a series of four different exemplars from one
category was presented twice {2 X 4 familiarization
trials}, followed by one new exemplar from the familiar
category (first test item = Trial9), and one new exemplar
from the contrasting category {second test
item = Triall0}. The experimenter did not label any
of the objects, and parents were asked to not interact
with the children during the session.
Scoring. For the purpose of later scoring, three different
copies of these videotapes were made. Either
Trial 9 or Trial 10 (random choice) of a given session
was copied onto a first tape. The remaining test trial
of the same session was copied onto a second tape
Pauen 1023
Figure 2 Results of Experiment 1A: Changes in mean examination
times during the test phase (Trial 8, last familiarization
trial; Trial 9, new exemplar of the familiar category; Trial 10, new
exemplar of the contrasting category) split by group (Group A,
low between-category similarity; group B, high between-category
similarity), and familiarization condition (animals, furniture).
Responses to the category change on Trial 10 were
highly similar for both groups and familiarization conditions.
interaction, post hoc tests (Student-Newman-Keuls)
compared mean examination times for both groups
(A, B)in both familiarization conditions (animals, furniture)
on Trials 8, 9, and 10, respectively. When Trial
8 served as the dependent variable, Group A differed
significantly from Group B for those infants who had
been familiarized with furniture, p < .05. Infants who
were familiarized with the natural-looking furniture
items examined the last familiarization stimulus for a
shorter time than did those who had been familiarized
with the artificial-looking furniture items. When
Trials 9 or 10 served as the dependent measures, no
two subgroups differed significantly. Accordingly, the
observed three-way interaction could be attributed
primarily to differences in mean examination times
on the last familiarization trial (see Figure 2). The corresponding
difference on this specific trial was not
replicated in later studies, and was therefore likely a
result of chance variation across groups.
Because examination times for the critical two test
trials (Trials 9 and 10) did not differ substantially
--0- Group A (Animals)
, .• ' . Group A (Furniture)
-0- Group B (Animals)
. , •.. Group B (Furniture)

10
..
9
Trial
8
10
9
8
,.-..
$ 7
Q)
E
E.:: 6
t::
0
'.0
tr:t:l 5 'srtl
X 4 ~
t::
rtl
Q) ~ 3
2
1
0
Mean intercoder reliability for examination was
r = .95,with a range of r = .92 to r = .96. For each trial,
the mean score of both coders provided the raw data
for further analysis.
Familiarization trials. To test whether infants habituated
to the presented material and / or category, two
separate scores were calculated: (1) mean examination
time during the first presentation of all four different
familiarization stimuli, and (2) mean examination
time during the second presentation. These two
scores were entered into a 2 X 2 X 2 (Group X Familiarization
Condition X Presentation) mixed-design
ANOVA. The analysis revealed a significant main effect
for presentation, F(I, 75) = 11.86, P< .01.All other
main effects and interactions were nonsignificant. Infants
examined the material longer during the first
presentation (M = 6.19 s, SD = 2.68 s) than during
the second presentation (M = 5.11 s, SD = 2.69 s).
Neither group (artificial-looking or natural-looking
material) nor familiarization condition (animals or
furniture) affected infants' performance. Hence, no
a priori preferences for any given set of stimuli
were observed.
Test trials. Infants' mean examination times for the
last familiarization trial and both test trials (Trials 8, 9,
and 10) were entered in a 2 X 2 X 3 (Group X Familiarization
Condition X Trial) mixed-design ANOVA,
which revealed a significant main effect for trial, F(2,
150) = 30.58, P < .001, as well as a significant threeway
interaction between group, familiarization condition,
and trial, F(2, 150) = 3.54, P < .05. No other
main effects and interactions were significant. To explore
further the nature of the observed three-way
Results and Discussion
(Trial 10 or Trial 9, respectively). The third tape contained
all familiarization trials (Trials 1-8). Two independent
coders who had not been involved in the
process of data collection and were blind to the hypothesis
first measured examination times on the two
test trials (first and second tape). Observers were unaware
of which of the two test trials they were viewing.
The third tape was scored only after the other two
tapes. This procedure sought to rule out the possibility
that knowledge about the hypothesis could influence
the results. Both observers recorded examining
time, defined as the time during which a given child
was concentrating on and inspecting the object
(Oakes et al., 1991; Oakes & Tellinghuisen, 1994).
This state of focused attention involves active intake
of information about the object as opposed to
activity for its own sake, which may include the object
(Ruff, 1986).
1024 Child Development
Table 3 T-Test Comparisons of Means for the Three Critical
Trials (8, 9, and 10) in Experiments 1 and 2
Trial
Study df 8/9 8/10 9/10
Experiment 1A 78 -5.63*** -6.80*** -2.76***
Experiment 1B 39 -1.35 -3.39** -2.21*
Experiment ic 19 -4.56*** -2.03* 2.74*
Experiment 2 19 -1.52 -.04 1.58
* P< .05; **P< .01; *** P< .001.
across the four tested subgroups but rather showed a
numerical increase from Trial 9 to Trial 10 in all four
cases, the corresponding data were collapsed across
groups. T-test comparisons of mean examination
times on the last three trials (see Table 3) revealed that
infants increased attention from the new exemplar
from the familiar category (Trial 9:M == 7.05 s, SO ==
3.75 s) to the out-of-category item (Trial 10:M == 8.41
s, SO == 4.14 s). To allow for a direct comparison of the
present findings with the categorization patterns
identified by Mandler and McDonough (1993), the
same t test was also used to compare group means for
Trials 8 and 9 (despite the significant difference between
subgroups), and between Trials 8 and 10. The
mean for Trial 9 turned out to be significantly greater
than for Trial 8 (M == 4.80 s, SO == 4.10 s).
The central finding of a significant increase in
mean examination time from the same- to a differentcategory
test item in both conditions (see Figure 2)
was consistent with an analysis of individual datasets.
According to this analysis, 75% of all infants who had
been familiarized with animals dishabituated to a
furniture item. This was true for Group A as well as
for Group B. For those infants who had been familiarized
with furniture items, the corresponding percentages
were somewhat lower, but showed only a small
difference between groups (Group A, 68.4%;Group B,
60%). The total number of infants who showed an increase
in examination time from Trial 9 to Trial 10 did
not differ substantially between Groups A and B,
however, x2(3, N == 79) == 1.44, P > .05.
As in previous studies (Mandler & McDonough,
1998; Pauen, 1997), 11-month-olds showed a linear increase
in mean examination time from the last familiarization
exemplar over the new within-category
item to the out-of-category item, with all means differing
significantly from each other. As suggested by
the knowledge-based view, no substantial differences
in results were found between Groups A and B.
Eleven-month-old infants showed similar responses
to the out-of-category exemplar when that exemplar
had the same material consistency, global shape,
color, and surface pattern as the familiarization
stimuli (Group B), and when its perceptual features
differed greatly from the familiarization stimuli
(Group A).
At this point, it was premature to draw any general
conclusions, however. Results obtained with the
present version of the object-examination task are
associated with some methodological problems that
needed to be addressed. One problem concerns the
explanation for the linear increase in mean examination
time over the last three trials (i.e., for the pattern
of results called "advanced categorization" by Mandler
& McDonough, 1993). If infants treated all animals
or furniture items as equal members of the same
global class, one would expect a generalization of
familiarization to the new exemplar of the familiar
category. Following Mandler and McDonough, 11month-
olds did not show this pattern of results, because
they recognized that the same series of four
items was presented twice during familiarization.
When entering the test phase, 11-month-olds responded
with an increase in mean examination time
to the within-category test item, because they realized
that it was perceptually new. The out-of-category
item was examined for an even longer time than was
the within-category test item because the former
was not only new in perceptual terms, but also in conceptual
terms. Mandler and McDonough assumed
that younger infants were less able to encode and
memorize specific features of individual exemplars.
For that reason, they may have generalized familiarization
to the within-category item. If this line of reasoning
is correct, younger infants participating in the
same task as used in Experiment 1A should show the
pattern of simple categorization. Experiment 1B
tested this developmental hypothesis.
EXPERIMENT 1B
Method
A total of 40 infants (age: M == 10 months, 12 days,
range == 10 months, 1 day-10 months, 3 days) participated.
Infants were recruited in the same way as in
Experiment 1A. Half of the infants were randomly assigned
to Group A (12 girls, 8 boys), and the other half
were assigned to Group B (7 girls, 13 boys). Five additional
infants were originally tested but started to cry
and did not complete the session. All infants remaining
in the final sample came from a White, middleclass
socioeconomic background. The experimental
procedure, stimuli, and scoring technique were identical
to that in Experiment 1A.
Results and Discussion
Mean intercoder reliability on all trials was r == .92,
with a range from r == .81 to r == .99. Mean examination
time decreased significantly from the first part of
the familiarization period (M == 6.73 s, SO == 2.67 s) to
the second part of the familiarization period (M ==
5.28 s, SO == 2.81 s), t(39) == 3.19, P < .01.
To test infants' performance during the test phase,
mean examination times for the last three trials were
entered in a 2 X 2 X 3 (Group x Familiarization Condition
X Trial) mixed-design ANOVA. This analysis
revealed a significant main effect for trial, F(2, 76) ==
6.41, P < .01. No other main effects or interactions
were significant. Mean examination times were M ==
5.17 s (SO == 4.25 s) for the last familiarization trial,
M == 6.10 s (SO == 4.92 s) for the same-category item,
and M == 7.63 s (SO == 4.59 s) for the out-of-category
item. Post hoc comparisons revealed no significant increase
in mean examination time from Trial 8 to Trial
9, but a significant increase from both Trials 8 and 9 to
Trial 10 (see Table 3).
Infants generalized familiarization to the first test
item (new exemplar of familiar category) but not to
the out-of-category item. This was consistent with the
pattern of "simple categorization" observed in previous
studies with infants younger than 11 months of
age (Mandler & McDonough, 1993, 1998). As in Experiment
lA, no differences between the two experimental
conditions were observed during the test
phase, and infants in both experimental groups responded
only with an increase in mean examination
time to the category change. The general finding that
perceptual between-category similarity did not affect
responses during the test phase could thus be replicated
with an independent sample of 10-month-olds.
EXPERIMENT ic
Another methodological problem associated with the
classical version of the object-examination task concerns
the fixed sequence of the test trials. Because the
out-of-category exemplar is always presented after
the within-category exemplar, one could argue that
the significant increase from the first to the second
test trial reflects natural changes in attention throughout
the experimental session. To test whether the order
of presentation of test stimuli accounted for the
observed findings, Experiment 1C was conducted.
Method
A total of 20 infants (age: M == 11 months, 13 days,
range == 11 months-Ll months, 30 days) participated.
Pauen 1025
Children were recruited in the same way as in Experiment
1A. Half of the children were randomly assigned
to Group A (4 girls, 6 boys), and the other ha~f
were assigned to Group B (6 girls, 4 boys). Two additional
children were originally tested but started to
cry and did not complete the session. All children came
from a White, middle-class socioeconomic background.
The experimental procedure and the stimuli we~e
identical to that in Experiment lA, except that all children
examined the out-of-category item first rather
than second during the test phase.
Results and Discussion
Mean intercoder reliability was r == .96 with a
range from r == .91 to r == .99. Mean examination tin:e
decreased significantly from the first half of the familiarization
phase (M == 6.95 s, SO == 2.38 s) to the second
half (M == 5.06 s, SO == 2.89 s), t(19) == 2.93,P< .01.
To test infants' performance during the test phase,
mean examination times for the last three trials were
entered in a 2 X 2 X 3 (Group X Familiarization Condition
X Trial) mixed-design ANOVA. This analysis
revealed a significant main effect for trial, F(2, 32) ==
13.59, P < .001. No other main effect or interaction
reached significance. Mean examination times were
M == 5.19 s (SO == 3.83 s) for the last familiarization
trial, M == 8.58 s (SD == 3.89 s) for the out-of-category
item presented on Trial 9, andM == 6.58 s (SD == 4.04 s)
for the perceptually new same-category item, presented
on Trial 10. Post hoc comparisons revealed a
significant increase in mean examination time from
the last familiarization trial to each of the two perceptually
new test stimuli, as well as a significa~t ~ecline
from the out-of-category item to the new withincategory
item (see Table 3). This replicates the results
obtained in Experiment 1A.
The major findings of Experiment 1 can be summarized
as follows: (1) infants aged 10 to 11 months
showed a decrease in attention during the familiarization
period; (2) following familiarization with four
different 3-D toy models of the same global category,
10- as well as 11-month-olds preferred to examine an
out-of-category item over a new within-category exemplar,
irrespective of order of presentati~n; and ~3)
infants' responses did not vary systematically WIth
the degree of perceptual between-category similarity.
If infants either compared individual exemplars or
formed some type of categorical representation only
on the basis of perceptual information available during
the familiarization phase, a systematic manipu!ation
of between-category similarity should have Influenced
changes in mean examination time during
the test phase. Three independent studies provided
1026 Child Development
evidence against this idea, suggesting that some type
of pre-existing categorical representation may be
guiding infants' attention in the object-examina~on
task presented here. If this pre-existing represent.atIon
refers to the categorical distinction between animals
and furniture, infants should only show an increase
in mean examination time from the first to the second
test trial if the second item belongs to a new realworld
category. However, it is also possible that infants
in Experiment I, Group B were responding to
the fact that the exemplar from the novel category
presented a novel combination of familiar features
rather than to its membership in a different realworld
category. On this account, a same-category
item showing a new combination of already-familiar
features should not elicit an orientation response. Experiment
2 tested this hypothesis.
EXPERIMENT 2
In this experiment, infants received the same objec~examination
task as in Experiment 1A (Group B: arti ficial
material). The only difference was that the second
test item did not represent an exemplar from the
contrasting category, but rather was a true hybrid of
the perceptual features of three of the four familiarization
exemplars. Accordingly, Experiment 2 provided
no change in category at the end of the experimental
session, but provided a similar variation in
perceptual similarity between the test items and the
familiarization stimuli. If infants applied knowledge
about category membership, no increase in attention
from the first to the second test item should be perceived.
Alternatively, infants may be more interested
in toys that provide a new combination of alreadyfamiliar
perceptual features than in toys that look
completely different. In this case, infants who participated
in Experiment 2 should show a pa~t~rn of r:sponses
similar to those infants who participated 10
Experiment 1A.
Method
Participants. A total of 20 infants, 11 months of age
(age:M = 11 months, 18 days, range = 11 months, 2
days-11 months, 29 days) participated. The sample
consisted of 8 girls and 12 boys. Infants were recruited
in the same way as in Experiment 1A. Three
additional infants were originally tested but started to
cry and did not complete the session . All infants came
from a White, middle-class socioeconomic background.
Stimuli and procedure. The stimuli consisted of
artificial-looking animals only. Infants received the
same toys as in Experiment 1A (Group B;see Figure I,
bottom). They were presented in the same order ~nd
manner as in Experiment 1A with the sole exception
of the second test item (Trial 10). Instead of a furniture
item, a hybrid of three familiarization stimuli (animals)
was presented. This hybrid had the same surface
pattern as one of the familiarization exemplars,
the same shape as a second one, and the same colors
as a third one (see Figure 3).
It was expected that whereas infants ~h.o ~art~cipated
in Experiment 1A (Group B/ famlhan.zatIon
with animals) showed a linear increase from Tnal9 to
Trial 10, no such increase would be found in Experiment
2 (due to the lack of a category change). Perceptually,
there were very few differences between the
stimuli that served as second test items in each experiment:
These differences concerned only the specific
outlines and arrangement of parts. Therefore, if infants
only relied on an ad hoc analysis of overall ~imilarities
and differences in the appearance of the gIven
test stimulus to previously seen familiarization exemplars,
responses to the second test. item should not
differ substantially between Expenment 1A (Group
B, familiarization with animals) and Experiment 2.
Results and Discussion
Mean intercoder reliability for Experiment 2 was r=
.95, with a range from r = .91 to r = .98. For each trial,
Figure 3 Artificial-looking animals presented on Trial 10 in Experiment 2.
the mean score of examination time between both
coders provided the raw data for all further analyses.
Infants examined the given animals for a significantly
longer time during their first presentation (M =
5.44 s, SO = 2.38 s) than during their second presentation
(M = 4.23 s, SO = 1.55 s) in the familiarization
phase, t(19) = 2.48, P < .05.
During the test phase, infants who participated in
Experiment 2 showed an increase in mean examination
time from Trial 8 (M = 4.90 s, SO = 2.88 s) to Trial
9 (M = 6.34 s, SO = 3.01 s), and a slight decrease from
Trial 9 to Trial 10 (M = 4.95 s, SO = 3.59 s). None of
these mean differences were significant, however (see
Table 3). To test whether infants' performance differed
systematically between Experiment 1A and Experiment
2 for Trial 10 only, a 2 X 3 (Study X Trial)
mixed-designANOVA was conducted that compared
performance of those infants who participated in Experiment
lA/Group B(familiarization with artificiallooking
animals) with the performance of those infants
who participated in Experiment 2. This analysis
revealed a significant main effectfor trial, F(2, 76)=5.04,
P < .01, as well as a significant interaction between
study and trial, F(2, 76) = 5.49, P< .01 (see Figure 4).
Subsequent analyses revealed no significant differences
between the means in the two studies observed,
either for the last familiarization trial, F(l, 39) = .24,
P> .05, or for the first test trial, F(l, 39) = .41,P> .05.
Up to this point, infants who participated in both experiments
had received identical stimuli. After combining
both datasets (N = 40), infants showed a
significant increase from the last familiarization trial
(M = 5.21 s, SO = 3.94 s) to the first test trial (M =
6.76 s, SO = 4.14 s), t(39) = -2.64, P < .05, suggesting
that they increased their attention to a perceptually
different-looking exemplar of the already-familiar
animal category.
As hypothesized, means in the two studies differed
significantly only for the last test trial, F(l' 39) =
11.67, P < .01. In Experiment 1, an increase in mean
examination time was observed between Trial 9 and
Trial 10, t(19) = 2.32,P < .05. In Experiment 2, examination
times tended to decrease (see Figure 4), but
this decrease failed to reach the level of significance
(see Table 3). An analysis of individual response patterns
was consistent with this finding: Whereas 75%
of all infants who participated in Experiment 1A
(Group B, familiarization with animals) showed an
increase in examination time from Trial 9 to Trial 10,
this was true for only 30% of the infants who participated
in Experiment 2. The remaining 25% (Experiment
1A) and 70% (Experiment 2) showed a decrease
in examination time. This difference between the
studies was significant, X2(1, N = 40) = 8.12, P < .01.
Pauen 1027
10
9
8
$'""'
OJ 7
S
E=: l::: 6
0
'.0
r:l l:: 5 ·sr:l
~X 4
S
r:l
OJ 3
~
2
-0-Experiment lA
1 ~Experiment 2
0
8 9 10
Trial
Figure 4 Changes in mean examination times during the test
phase for infants who participated in Experiment lA (familiarization
with artificial-looking animals; category change on
Trial 10) and Experiment 2 (familiarization with artificiallooking
animals; no category change on Trial 10). The presence
or absence of a category change predicted whether infants increased
examination time from Trial 9 (new exemplar of familiar
category) to Trial 10 (new exemplar of the contrasting category).
The corresponding pattern of findings thus suggests
that infants' responses to the test stimuli varied systematically
with the presence or absence of a category
change. A new combination of familiar attributes
did not elicit the same increase in mean
examination time as did a perceptually similar object
from a novel category.
GENERAL DISCUSSION
Four related object-examination studies tested whether
infants' categorization performance could be attributed
to on-line processes of category formation during
familiarization, or to the application of previously
acquired knowledge. Experiment 1 contrasted animals
with furniture, using natural-looking toy models
(Group A) as well as toy replicas with artificially increased
between-category similarity (Group B). Both
Il-month-olds (Experiment 1A) and 10-month-olds
(Experiment 18) noticed the category change equally
well for both sets of stimuli. This pattern of findings
1028 Child Development
did not depend on the order of test trials (Experiment
1C). Experiment 2 used only artificial-looking stimuli
and compared infants' responses during the test phase
with and without introducing a category change at
the end of the experimental session. In both cases, the
critical test item had the same global shape, coloring,
and surface pattern as did the exemplars from the familiarization
category. An increase in mean examination
time occurred only after a change of category.
As is argued below, results of both experiments
seem to support the knowledge-based account. This
conclusion rests on the assumption that performance
in the object-examination task reflected infants' ability
to discriminate animals and furniture in categorical
terms. Whether this was actually the case must be
addressed before further implications of the reported
findings can be examined.
Does the Object-Examination Task Assess
Categorization Skills?
In classification tasks that use either visual-fixation
tasks or the object-examination technique, category
discrimination is typically inferred if infants generalize
familiarization to a new member of the alreadyfamiliar
category, and increase attention to a new
member of a different category (see Introduction). In
Experiment lA, 11-month-olds showed a different
pattern of responses, however. Mean examination
times increased from the last familiarization item to the
new within-category exemplar, and they increased
still more from the within-category exemplar to the
out-of-category exemplar. This raises the question of
whether infants categorized the given stimuli at all.
It is possible that infants simply compared the perceptual
similarity between individual familiarization
stimuli on the one hand and both test stimuli on the
other hand. In this case, the out-of-category exemplar
may be examined for a longer time than the
same-category test item because it shares fewer perceptual
attributes with each of the previously seen familiarization
items. Several arguments can be raised
against this interpretation. First, if changes in mean
examination time were based on a comparison of perceptual
similarity between individual stimuli, infants
should respond differently in the high and in the low
between-category similarity condition. More specifically,
infants in the low between-category similarity
condition should examine the out-of-category item
for a comparably longer time than the same-category
test item. Infants in the high between-category condition
should show the reverse pattern of preference;
that is, they should prefer the within-category test
stimulus, because it differs in shape, texture, and
color from the familiarization stimuli, whereas the
out-of-category stimulus is similar to the familiarization
stimuli on these dimensions. Three independent
studies (Experiments lA, 1B, and 1C) provided evidence
against this hypothesis. Second, in Experiment
1B, 10-month-olds produced a pattern of responding
consistent with the idea that they categorized the
given stimuli (i.e., they gerleralized familiarization to
the first test item despite the fact that this item was
perceptually new as compared with the familiarization
stimuli). It seems rather unlikely that infants who
are only 4 weeks older no longer categorize the giverl
items when tested with the same material and the
same type of task. Finally, Pauen (1997) tested 11month-
olds' categorical distinction between animals
and furniture with a slightly modified version of the
target task. In this study, 10 exemplars from the same
global domain were presented during the familiarization
phase, followed by one out-of-category exemplar.
Despite the fact that no item was shown twice
and highly different-looking exemplars from each
category were presented, 11-month-olds decreased
examination time throughout the familiarization period
(Trial 1 to Trial 10). They only increased attention
when a new exemplar of the contrasting category
was presented on Trial 11 (see also Pauen, 2000a). The
general finding that infants treat different-looking
members of the same category equivalently while
treating an out-of-category exemplar differently could
thus be demonstrated with 11-month-olds who participated
in an object-examination task with the same
categories employed in the current research. In sum,
these findings suggest that the object-examination
task provides a suitable tool to measure categorization
performance in infants of the tested age range.
This raises the question of whether infants form a category
on-line or whether they apply previously acquired
knowledge.
Is Categorization Performance On-Line Based or
Knowledge Based?
The on-line-based account assumes that infants
base their responses in the object-examination task on
similarity comparisons of the appearance of the given
experimental stimuli. These comparisons could either
refer to (1) overall similarity, (2) a single critical feature,
or (3) a specific configuration of attributes. Each
of these alternatives will be discussed in turn.
Preliminary Experiments 1 and 2 revealed that,
overall, perceptual similarity between categories is
low for stimulus set A, but high for stimulus set B. If
infants had formed a category on-line and responded
to the test exemplars in terms of their degree of mismatch
with that categorical representation, systematic
differences between the two conditions should
have occurred. In the high between-category similarity
condition, the increase in attention should have been
stronger when the same-category item was presented
than when the out-of-category item was presented. In
the low between-category similarity condition, the reverse
pattern should have been found. The fact that
infants showed similar responses in both experimental
groups thus argues against the idea that comparisons
related to overall similarity can account for the
observed pattern of responding.
Alternatively, infants may have focused attention
on a single perceptual attribute to form a category online.
Potentially relevant attributes that would allow
for category discrimination can easily be identified
for the natural-looking toys, but not for the artificiallooking
toys: The artificial-looking items from both
categories had the same material consistency, were
painted with similar colors, and had the same type
of surface pattern, but the natural-looking stimuli
varied substantially across categories with respect to
these features (see Figure 1). Therefore, none of these
features can explain the reported findings. On this account,
then, categorization performance in the low
between-category similarity condition should have
been better than in the high between-categorysimilarity
condition. However, performance in the
two conditions was comparable.
At a more abstract level, certain perceptual features
varied systematically between animals and furniture
items in both experimental conditions. The
proportion of rectilinear outlines provides an example
(Van de Walle, Spelke, & Carey, 1997). This proportion
appeared to be somewhat higher for furniture
items than for animals even in the high betweencategory
similarity condition. To test whether infants'
discriminated artificial-looking animals and furniture
items based on this type of information, Pauen
(2000b) recently replicated Experiment 1 with the
same objects, except that the objects were painted in
one homogeneous color, thereby removing all information
about surface patterns and eyelike markings
that allowed for an identification of individual
exemplars as animal or furniture items, respectively.
Only information about the specific outlines of each
given object remained the same. This set of stimuli
was presented to a group of twenty 11-month-olds.
No significant changes in mean examination time
were observed across the three critical test trials, suggesting
that the proportion of rectilinear outlines
alone cannot account for the positive categorization
response observed in Experiment 1.
Other single features that might vary systemati-
Pauen 1029
cally between animals and furniture items could be
the presence or absence of specific parts. A look into
the developmental literature reveals that parts are
considered to playa special role in young children's
perceptual categorization (Tversky, 1989). Using a sequential
touching task, Rakison and Butterworth
(1998a, 1998b) found that 14- and 18-month-olds behaved
systematically toward categories with different
parts (i.e., animals with legs versus vehicles with
wheels), but did not do so toward objects with similar
parts (animals with legs versus furniture items with
legs). In the present series of studies, 10- and 11month-
olds clearly differentiated between animals
and furniture, despite the fact that items from both
categories had legs (see also Mandler & McDonough,
1998; Pauen, 1997). The same was true for eyes in the
high between-category similarity condition. In the low
between-category condition, however, both animal
and furniture items had eyelike markings. Because infants
performed equally well in the two conditions,
the presence or absence of eyes cannot account for the
observed findings either.
Finally, rather than focusing on the presence or absence
of specific features, infants may respond to a
different configuration of attributes. Part configuration
is known to be highly relevant for object identification
and categorization in adults as well as in children
(Biederman, 1993; Hummel & Biederman, 1992;
Madole & Cohen, 1995; Saiki & Hummel, 1996; Tversky
& Hemenway, 1984; Younger, 1990; Younger &
Cohen, 1986). Experiments conducted by Rakison
and Butterworth (1998a) revealed that a confounded
spatial configuration of body parts could impair categorization
performance in 14- to 18-month-old children.
The authors displaced legs of animals and furniture
items by either changing their orientation but
maintaining the configuration (e.g., placing four legs
on the back of an animal), or by destroying the natural
configuration (e.g., placing each leg at a different
side and angle of the main body). Infants' performance
was only impaired in the latter case. In
Experiments 1 and 2 of the present research, however,
none of the animals or furniture items had misplaced
legs. All items were presented as standing on
their legs.
Considering other body parts and their spatial arrangement,
one could argue that the location of a
headlike part containing facial features might allow
infants to discriminate animals from furniture items.
Quinn and Eimas (1996) reported that information
about the head region seems crucial for 3- to 4-montholds'
ability to categorize animals. Based on the same
kind of information, 7- to 11-month-olds were able to
discriminate between animals and humans in an
1030 Child Development
object-examination task conducted by Pauen (1999b).
Furthermore, it is known that infants can discriminate
faces with two eyes located above a mouth from
other stimuli that contain the same elements but are
presented in a nonbiological configuration (Johnson
& Morton, 1991). In the light of such findings, one
might speculate that infants in the current research
recognized animals by the presence of two eyes above
a mouth painted on some kind of headlike part. Facial
features were barely visible in several animals, however
(natural- as well as artificial-looking stimuli).
Furthermore, infants who were familiarized with animals
and who therefore were provided with more
facial information did not perform any better than
infants who were familiarized with furniture items.
These observations provide evidence against the idea
that infants learned to distinguish the two categories
on-line by focusing exclusively on head and face
information.
In sum, although many perceptual attributes
known to be relevant for object categorization varied
substantially between categories in the natural-looking
set of stimuli, these attributes were shared by exemplars
from both categories in the artificial-looking set
of stimuli and did not, therefore, provide a basis for
distinguishing these stimuli. Despite substantial differences
in overall similarity between familiarization
and test stimuli, infants performed equally well in
both conditions (see Experiments lA, IB, and lC).
Furthermore, neither a single perceptual feature, nor
a specific configuration of attributes could be identified
that varied systematically between categories in
both conditions that would have allowed infants to
form a category on-line during familiarization, and
to discriminate animals from furniture items even in
the high between-category similarity condition.
Taken together, these analyses point to the conclusion
that the on-line view cannot explain the reported
findings.
According to the knowledge-based account, the
toy models used in each condition may activate the
same type of categorical representation-a representation
that infants formed prior to the start of the experimental
session and that contained more information
than was displayed in the given stimuli. In this
case, the appearance of individual items still played
an important role, but rather than providing the sole
basis for on-line processes of category formation during
the familiarization period, it may have helped to
determine membership of each individual exemplar
in the same category. Perceptual cues that allow infants
to identify category membership of toy models
may vary substantially between stimuli, thus leading
to a flexible and adaptive strategy for determining
class membership of each item presented (Jones &
Smith, 1993; Thelen & Smith, 1994).
According to Mandler and McDonough (1993,1998),
infants may identify each given object as a toy replica
of some real-world category (e.g., the global category
"animals"). When different exemplars of the same
category are presented in series during the familiarization
period, the same type of pre-existing knowledge
may be activated on each trial. If a given out-ofcategory
exemplar activates some other type of
pre-existing representation than all exemplars examined
before, infants should show a significant increase
in mean examination time from the first to the second
test item. The reported findings of both experiments
in the present research are consistent with this
interpretation. Infants' responses varied systematically
with the presence or absence of a category change
(see Experiment 2), but not with the degree of perceptual
between-category similarity (see Experiment 1).
Similar findings have been obtained with preschoolers
in studies using a quite different paradigm.
For example, Gelman and Coley (1990) reported that
4- to 5-year-olds preferred to project a hidden property
of a particular animal (e.g., an internal organ) to
a different-looking animal that belonged to the same
category than to an animal that was highly perceptually
similar to the target but belonged to a different
category. Inductive judgments did not cross ontological
boundaries; that is, structural and causal attributes
of living entities were not projected onto nonliving
entities and vice versa (see also Carey, 1985;
Keil, 1989). This general pattern of results also has
been documented for 14-month-olds (Mandler & MeDonough,
1996). Studies on inductive reasoning thus
suggest that knowledge about category membership
has a comparatively stronger impact on children's responses
than does perceived similarities in the appearance
of the presented experimental stimuli. The
studies reported in this article reveal similar responses
in an object-examination task conducted
with 10- and Ll-month-olds.
When thinking about the type of attributes that
might be relevant for infants' category formation, it
seems important to keep in mind that perceptual and
conceptual attributes are closely linked in many cases.
For example, Tversky and Hemenway (1984) showed
that different parts suggest distinct functions,
whereas similar parts tend to have comparable functions
(see also Madole & Cohen, 1995). If one assumes
that infants are able to perceive and learn causal patterns
early in life, they should also learn quickly to
pay attention to the part configuration of a given exemplar
to predict its causal and functional properties.
Mandler and McDonough (1993, 1998) proposed that
infants may be specifically interested in whether an
object is able to perform self-initiated movement (see
also Pauen, 2001; Premack, 1990; Spelke, Phillips, &
Woodward, 1995). This may motivate them to pay
special attention to the form and location of leglike
appendages. Furthermore, infants seem highly interested
in whether an object can interact socially with
others (Carey & Spelke, 1994; Johnson, Booth, &
O'Hearn, 1998; Meltzoff, 1995). This may lead them to
pay special attention to the head region and to facial
features (Pauen, 1999b; Quinn & Eimas, 1996). Causal
knowledge about the distinction between self-starters
and non-self-starters, as well as causal knowledge
about social versus nonsocial beings could provide
the first basis for a conceptual distinction between animates
and inanimates, which may in turn guide infants'
search for specific perceptual attributes in the
given set of experimental stimuli.
As already pointed out, results obtained with an
object-examination task do not allow for conclusions
with regard to the nature of those properties underlying
knowledge-based category discrimination. Preexisting
representations activated in the given task
context could include perceptual as well as conceptual
information about objects other than those presented
during the experimental session. The reported
findings suggest that infants went beyond the perceptual
information provided by the given stimuli, but
leave open the question of the nature of the activated
knowledge.
Categorization tasks that test infants' performance
must address the fact that both pre-existing representations
and an ad hoc analysis of perceptual attributes
related to any given stimuli may simultaneously influence
infants' responses (e.g., Smith, 2000). For that
reason, the specific relation between both aspects
needs to be explored in more detail. Studies that focus
on this question may not only contribute to a better
understanding of category formation in preverbal infants,
but also could be useful for the development of
theoretical models that describe general principles
of human category formation and object identification.
ACKNOWLEDGMENT
The reported studies were conducted at the University
of Tiibingen and at the University of Magdeburg
as part of a larger project (Object Categorization in
Preverbal Children), funded by the German Research
Foundation. Part of this work was presented at the
1997 biennial meeting of the Society for Research in
Child Development in Washington, DC, as well as at
the 1998 meeting of experimental psychologists in
Marburg, Germany. The author would like to thank
Pauen 1031
Birgit Trauble, Nicola Zauner, Andrea Schreier, Michael
Schmidt (Tiibingen), Diana Sodtke, Annika
Falkner, and Denis Krappe (Magdeburg) for their assistance
with data collection and video coding. Special
thanks also go to Gretchen Van de Walle for helpful
comments on earlier drafts of this manuscript.
ADDRESS AND AFFILIATION
Corresponding author: Sabina Pauen, Department of
Psychology, University of Heidelberg, Hauptstrasse
47-51, 69117 Heidelberg, Germany; e-mail: sabina.
pauen@psychologie.uni-heidelberg.de.
REFERENCES
Behl-Chadha, G. (1996). Basic-level and superordinate-like
categorical representations in early infancy. Cognition,
60, 105-141.
Biederman, I. (1993). Visual object recognition. InA. I. Goldman
(Ed.), Readings in philosophy and cognitivescience (pp.
9-22). Cambridge, MA: Bradford Press.
Bruner, J. S., Olver, R. R., & Greenfield, P. M. (1966). Studies
in cognitivegrowth.New York: Wiley.
Carey, S. (1985). Conceptual change in childhood. Cambridge,
MA: Cambridge University Press.
Carey, S., & Spelke, E. (1994). Domain-specific knowledge
and conceptual change. In L. A. Hirschfield & S. A. Gelman
(Eds.), Mapping the mind: Domainspecificity in cognition
and culture (pp. 169-200). Cambridge, MA: Cambridge
University Press.
Eimas, P. D., & Quinn, P. C. (1994). Studies on the formation
of perceptually based basic-level categories in young infants.
Child Development, 65, 903-917.
Eimas, P. D., Quinn, P. C., & Cowan, P. (1994). Development
of exclusivity in perceptually based categories of
young infants. Journal of Experimental Child Psychology,
58, 418-431.
Fodor, J. A. (1975). The language of thought. New York:
Thomas Y. Corwell.
Gelman, R. (1990). First principles organize attention to and
learning about relevant data: Number and the animateinanimate
distinction as examples. Cognitive Science, 14,
79-106.
Gelman, R., & Spelke, E. (1981). The development of
thoughts about animate and inanimate objects: Implications
for research on social cognition. In J. Flavell & L.
Ross (Eds.), Social cognitivedevelopment (pp. 43-66). New
York: Cambridge University Press.
Gelman, S. A., & Coley, J. D. (1990). The importance of
knowing a dodo is a bird: Categories and inferences in 2year-
old children. Developmental Psychology, 26, 796-804.
Goldstone, R. L., & Barsalou, L. W. (1998). Reuniting perception
and conception. Cognition, 65,231-262.
Hummel, J. E., & Biederman, I. (1992). Dynamic binding in
a neural network for shape recognition. Psychological Review,
99, 480-517.
1032 Child Development
Inhelder, B., & Piaget, J. (1964). Theearlygrowthof logic in the
child. New York: Norton.
Johnson, M. H., & Morton, J. (1991). Biology and cognitive development.
Cambridge, Ml\.: Blackwell.
Johnson, S. C., Booth, A., & O'Hearn, K. (1998, April). Inferring
the unseen goals of a non-human agent. Poster presented
at the biennial meeting of the International Society
on Infant Studies, Atlanta, GA.
Jones, S. S., & Smith, L. B. (1993). The place of perception in
children's concepts. CognitiveDevelopment, 8, 113-119.
Keil, F. C. (1989). Concepts, kinds, and cognitive development.
Cambridge, MA: MIT Press.
Keil, F. C. (1991). The emergence of theoretical beliefs as
constraints on concepts. In S. Carey & R. Gelman (Eds.),
Epigenesis of mind: Essayon biology and cognition (pp. 237321).
Hillsdale, NJ: Erlbaum.
Keil, F. C., Smith, W. C., Simons, D. J., & Levin, D. T. (1998).
Two dogmas of conceptual empiricism: Implications for
hybrid models of the structure knowledge. Cognition, 65,
103-135.
Madole, K. L., & Cohen, L. B. (1995). The role of object parts
in infants' attention to form-function correlations. Developmental
Psychology, 31, 637-648.
Madole, K. K., & Oakes, L. M. (1999).Making sense of infant
categorization: Stable processes and changing representations.
Developmental Review, 99, 587-604.
Mandler, J. M. (1992). How to build a baby: II. Conceptual
primitives. Psychological Review, 99, 587-604.
Mandler, J. M. (2000a). Perceptual and conceptual processes
in infancy. Journal of Cognition and Development, 1, 3-36.
Mandler, J. M. (2000b). What global-before-basic trend?
Commentary on perceptually based approaches to early
categorization. Infancy, 1, 99-110.
Mandler, J. M., & McDonough, L. (1993).Concept formation
in infancy. Cognitive Development, 8, 237-264.
Mandler, J. M., & McDonough, L. (1996).Drinking and driving
don't mix: Inductive generalization in infancy. Cognition,
59, 307-335.
Mandler, J. M., & McDonough, L. (1998). On developing a
knowledge base in infancy. Developmental Psychology, 34,
1274-1288.
Meltzoff, A. N. (1995). Understanding the intention of
others: Re-enactment of intended acts by 18-month-old
children. Developmental Psychology, 31, 838-850.
Nelson, K. (1974). Concept, word, and sentence: Interrelationships
in acquisition and development. Psychological
Review, 81, 267-285.
Oakes, L. M., Coppage, D. J., & Dingel, A. (1997).By land or
by sea: The role of perceptual similarity in infants' categorization
of animals. Developmental Psychology, 33,396-407.
Oakes, L. M., Madole, K. L., & Cohen, L. B. (1991). Infants'
object examining: Habituation and categorization. Cognitive
Development, 6, 377-392.
Oakes, L. M., & Tellinghuisen, D. J. (1994). Examining in infancy:
Does it reflect active processing? Developmental
Psychology, 30, 748-756.
Pauen, S. (1997). Kategorisierung im Sauglingsalter: Die
Unterscheidung globaler Objektklassen. Zeitschrift fur
Experimentelle Psychologie, 43, 600-624.
Pauen, S. (1999a). Developing ontological categories of matter:
Stable dimensions and changing concepts. In W.
Schnotz, S. Vosniadou, & M. Carretero (Eds.), New perspectives
on conceptual change (pp. 15-31). Amsterdam:
Elsevier.
Pauen, S. (1999b, April). Are human beings somethingspecial?
Early conceptual differentiation within the animate domain.
Paper presented at the biennial meeting of the Society
for Research in Child Development, Albuquerque, NM.
Pauen, S. (2000a). Early differentiation within the animate
domain: Are humans something special? Journal of Experimental
Child Psychology, 75, 134-151.
Pauen, S. (2000b, April). The relevance of form-bias for preverbal
categorization. Paper presented at the meeting of
experimental psychologists, Braunschweig, Germany.
Pauen, S. (2001, June). Evidence for knowledge-based categorization
in infancy: The animate-inanimate distinction. Paper
presented at the international conference on Foundations
of Human Knowledge Acquisition: New evidence
from infant research and from neuroscience, Delmenhorst,
Germany.
Premack, D. (1990). The infant's theory of self-propelled objects.
Cognition, 36, 1-6.
Quinn, P. C., & Eimas, P.D. (1996). Perceptual cues that permit
categorical differentiation of animal species by infants.
Journal of Experimental Child Psychology, 63, 189211.
Quinn, P. C., & Eimas, P. D. (1997). Reexamination of the
perceptual-to-conceptual shift in mental representation.
Reviewof General Psychology, 1,271-287.
Quinn, P. C., & Johnson, M. H. (2000). Global-before-basic
object categorization in connectionist networks in 2month-
old infants. Infancy, 1, 31-46.
Quinn, P. C., Johnson, M. H., Mareschal, D., Rakison, D. H.,
& Younger, B.A. (2000). Understanding early categorization:
One process or two? Infancy, 1, 111-122.
Rakison, D. H., & Butterworth, G. E. (1998a). Infants' attention
to object structure in early categorization. Developmental
Psychology, 34, 1310-1325.
Rakison, D. H., & Butterworth, G. E. (1998b). Infants' use of
object parts in early categorization. Developmental Psychology,
34, 49-62.
Roberts, K. (1988). Retrieval of a basic-level category in prelinguistic
infants. Developmental Psychology, 24, 21-27.
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., &
Boyes-Braem, P. (1976). Basic objects in natural categories.
Cognitive Psychology, 8, 382-439.
Ruff, H. A. (1986). Components of attention during infants'
manipulative exploration. Child Development, 57, 105-114.
Saiki, J., & Hummel, J. E. (1996). Attribute conjunctions and
the part configuration advantage in object category
learning. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 22, 1002-1019.
Smith, L. B. (2000). From knowledge to knowing: Real
progress in the study of infant categorization. Infancy, 1,
91-98.
Smith, L. B., & Heise, D. (1992). Perceptual similarity and
conceptual structure. In B. Burns (Ed.), Percepts, concepts,
and categories (pp. 233-272). New York: North-Holland.
Spelke, E. 5., Phillips, A., & Woodward, A. (1995). Infants'
knowledge of object motion and human action. In D.
Sperber, D. Premack, & A. Premack (Eds.), Causal cognition:
A multidisciplinarydebate (pp. 44-78). Oxford, U.K.:
Oxford University Press.
Thelen, E., & Smith, L. B. (1994). A dynamic system approach
to thedevelopment of cognition and action. Cambridge, MA:
MIT Press.
Tversky, B. (1989). Parts, partonomies, and taxonomies. Developmental
Psychology, 25, 983-995.
Tversky, B., & Hemenway, K. (1984). Objects, parts and cat-
Pauen 1033
egories. Journal of Experimental Psychology: General, 113,
1169-1913.
Van de Walle, G., Spelke, E. 5., & Carey, S. (1997, April). Conceptsand
categorization in infancy. Paper presented at the
biennial meeting of the Society for Research in Child Development,
Washington, DC.
Younger, B. A. (1990). Infants' detection of correlations
among feature categories. ChildDevelopment, 61, 614-620.
Younger, B. A., & Cohen, L. B. (1986). Developmental
change in infants' perception of correlations among attributes.
Child Development, 57, 803-815.

Comments
  • Recommend Us