Baby Wordsmith
From Associationist to Social Sophisticate
Roberta Michnick Golinkoff1 and Kathy Hirsh-Pasek2
1
University of Delaware and
2
Temple University
ABSTRACT—How do infants acquire their first words? Word
reference, or how words map onto objects and events, lies
at the core of this question. The emergentist coalition model
(ECM) represents a new wave of hybrid developmental
theories suggesting that the process of vocabulary devel-
opment changes from one based in perceptual salience and
association to one embedded in social understanding. Be-
ginning at 10 months, babies learn words associatively,
ignoring the speaker’s social cues and using perceptual
salience to guide them. By 12 months, babies attend to so-
cial cues, but fail to recruit them for word learning. By 18
and 24 months, babies recruit speakers’ social cues to learn
the names of particular objects speakers label, regardless
of those objects’ perceptual attraction. Controversies about
how to account for the changing character of word ac-
quisition, along with the roots of children’s increasing re-
liance on speakers’ social intent, are discussed.
KEYWORDS—word learning; language development
There is power in language. It can start wars or ruin marriages.
Readers of these words barely remember a time when they did
not have language. But every word you know had to be learned.
Imagine bending over your car engine with your mechanic and
being told, ‘‘Your zorch is shot.’’ You follow your mechanic’s eyes
and body orientation to the part he is examining. That rusty metal
protrusion must be the zorch. How do we learn the mapping
between words and the objects and events they represent?
THE WORD-LEARNING PROBLEM
Establishing a Word’s Referent: Perceptual, Social, and
Linguistic Cues
Infants are motivated to learn names for the same reason that
adults are: Knowing what to call something allows one to share
the contents of one’s mind with another person (Bloom & Tinker,
2001), even when the object is not present. Indeed, a great
deal is known about the course of word learning. At 10 months,
babies have an average comprehension vocabulary of 50 words,
saying virtually nothing. By 30 months, average production
vocabulary soars to 550 words (see Table 1; Fenson, Dale,
Reznick, Bates, Thal, & Pethick, 1994), and children speak in
full sentences.
Describing vocabulary growth, however, is only a first step
toward unpacking the mechanisms behind word learning. How
do words get ‘‘hooked’’ to objects and events? How do we (or
children) learn that zorch refers to that whole rusty protrusion
rather than to the object’s color or size? Any object presents an
array of possible referents, a problem Quine (1960) called the
indeterminacy of reference. A number of diverse theories have
arisen to explain how children solve this problem.
One theory is that children approach the word-learning
problem with a set of constraints or principles biasing them ...
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
Baby WordsmithFrom Associationist to Social SophisticateRo
1. Baby Wordsmith
From Associationist to Social Sophisticate
Roberta Michnick Golinkoff1 and Kathy Hirsh-Pasek2
1
University of Delaware and
2
Temple University
ABSTRACT—How do infants acquire their first words? Word
reference, or how words map onto objects and events, lies
at the core of this question. The emergentist coalition model
(ECM) represents a new wave of hybrid developmental
theories suggesting that the process of vocabulary devel -
opment changes from one based in perceptual salience and
association to one embedded in social understanding. Be-
ginning at 10 months, babies learn words associatively,
ignoring the speaker’s social cues and using perceptual
salience to guide them. By 12 months, babies attend to so-
cial cues, but fail to recruit them for word learning. By 18
2. and 24 months, babies recruit speakers’ social cues to learn
the names of particular objects speakers label, regardless
of those objects’ perceptual attraction. Controversies about
how to account for the changing character of word ac-
quisition, along with the roots of children’s increasing re-
liance on speakers’ social intent, are discussed.
KEYWORDS—word learning; language development
There is power in language. It can start wars or ruin marriages.
Readers of these words barely remember a time when they did
not have language. But every word you know had to be learned.
Imagine bending over your car engine with your mechanic and
being told, ‘‘Your zorch is shot.’’ You follow your mechanic’s
eyes
and body orientation to the part he is examining. That rusty
metal
protrusion must be the zorch. How do we learn the mapping
between words and the objects and events they represent?
THE WORD-LEARNING PROBLEM
Establishing a Word’s Referent: Perceptual, Social, and
3. Linguistic Cues
Infants are motivated to learn names for the same reason that
adults are: Knowing what to call something allows one to share
the contents of one’s mind with another person (Bloom &
Tinker,
2001), even when the object is not present. Indeed, a great
deal is known about the course of word learning. At 10 months,
babies have an average comprehension vocabulary of 50 words,
saying virtually nothing. By 30 months, average production
vocabulary soars to 550 words (see Table 1; Fenson, Dale,
Reznick, Bates, Thal, & Pethick, 1994), and children speak in
full sentences.
Describing vocabulary growth, however, is only a first step
toward unpacking the mechanisms behind word learning. How
do words get ‘‘hooked’’ to objects and events? How do we (or
children) learn that zorch refers to that whole rusty protrusion
rather than to the object’s color or size? Any object presents an
array of possible referents, a problem Quine (1960) called the
indeterminacy of reference. A number of diverse theories have
4. arisen to explain how children solve this problem.
One theory is that children approach the word-learning
problem with a set of constraints or principles biasing them to
entertain certain hypotheses about word reference over others.
For example, children seem to attach names to whole objects
rather than to parts (Markman, 1989; Golinkoff, Mervis, &
Hirsh-Pasek, 1994).
A second theory dismisses Quine’s conundrum, claiming that
children map words onto the most salient objects or actions in
the
environment. Early word learning is but word–object associa-
tions (learned links) between noticeable (moving, brightly-
colored) objects and concurrent sound sequences (words).
Finally, a third solution suggested by the family of social -
pragmatic theories proposes that infants are attuned to the social
cues speakers offer when labeling objects. Eighteen-month-
olds, for example, only learn novel words when it is clear that
objects are being labeled for their benefit (Baldwin, Markman,
5. Bill, Desjardins, Irwin, & Tidball, 1996). If a speaker is on the
telephone, toddlers resist learning a novel name for an object in
front of them, even if the name is uttered with great excitement.
This ‘‘failure’’ is adaptive: Fully 50% of parents’ talk is not
about
the child’s focus of attention (Baldwin et al., 1996). To learn
words, children must note more than just the temporal
contiguity
between a verbal label and an object they are attending to.
Although the description of these theories is a bit of a cari-
cature, each family of theories emphasizes only a part of the
Address correspondence to Roberta M. Golinkoff, School of
Educa-
tion, University of Delaware, Newark, DE 19716; e-mail:
[email protected]
udel.edu.
C U R R E N T D I R E C T I O N S I N P S Y C H O L O G I C
A L S C I E N C E
30 Volume 15—Number 1Copyright r 2006 Association for
Psychological Science
word-learning process and appeals to one causal mechanism
6. as paramount. There have been a number of calls for hybrid
theories that recognize the complexity of word acquisition and
integrate diverse inputs (e.g., Waxman & Lidz, 2006). The
emergentist coalition model (ECM; Hollich et al., 2000) offers
one example that has yielded fruit.
Tracing the Changing Process of Word Learning
The ECM recasts the issue of word learning by asking which
components of which theories govern word learning at different
phases of development; rather than providing an overall snap-
shot, it tracks changing strategies over time. Progress has been
made in testing this more complex account (e.g., Hollich et al.,
2000); in fact, the ECM is currently the only hybrid model that
has been empirically evaluated. Here we present evidence il -
lustrating how complex models of word learning can be put to
the
test. By examining infants’ shifting use of associative and
social
strategies across time, we offer a glimpse of evidence for a
piece
of the ECM.
7. Three Fundamental Tenets of the ECM
The model has three basic tenets. First, children are sensitive to
multiple cues in word learning: perceptual, social, and linguistic
(see Fig. 1). Second, word-learning cues change their relative
importance over time. Although a range of cues in the coalition
is
always available, not all cues are equally utilized in the service
of word learning. Children beginning to learn words rely on a
perceptual subset of the available cues in the coalition. Only
later do they recruit social cues like other people’s eye gaze and
handling of objects to learn words.
Third, the principles of word learning are emergent, changing
over time. Infants may start with an immature principle of ref-
erence, such that a word will be mapped to the most salient
object from the infant’s point of view. Later, children sensitive
to
speaker intent map a word onto an object from the speaker’s
point of view.
CHANGING PROCESSES IN WORD LEARNING:
8. THE EVIDENCE
To investigate transformations in the word-learning process, a
method was needed that could measure comprehension and that
could succeed with children between 10 and 24 months. Com-
prehension provides a sensitive index of word-learning compe-
tence not restricted by factors that may limit language
production, such as articulatory control or motivation to talk.
Further, the method must permit the putative cues involved in
word learning to be placed in competition so that infants’
relative
reliance on these cues can be gauged. Hollich et al. (2000)
created just such a method using infant visual fixation on target
objects as the dependent variable. Babies saw two objects, one
interesting (more salient) and one boring (colorless and mo-
tionless). Standing between the objects, which were placed on a
table out of the infant’s reach, a speaker verbally labeled either
the interesting or boring object and used social cues like eye
gaze and sometimes handling to indicate which object was being
9. labeled. The method ensured that children were learning words
and not just examining interesting objects. After word training
and a test to see if children learned the name of the target
object,
another new, deliberately ambiguous label was introduced that
the child had not previously heard. This ‘‘new label’’ trial
tested
whether children would continue to stare at the named, target
object even in the presence of another name or would look away
or look at the alternative object upon hearing a new name. In a
final ‘‘recovery trial,’’ infants were asked to look again at the
original object.
These different kinds of test trials in combination constitute
a powerful test of word learning. If children are operating at
the associative level, failing to use social cues, they should
simply attach the label to the object they find most interesting.
TABLE 1
Median Number of Words (and Ranges) in the Comprehension
and Production Vocabularies of Children at Different Ages,
10. According to Parental Report From the MacArthur
Communicative Development Inventory (CDI)
Age (months)
Comprehensionn Production
Median Range Median Range
10 42 11–154 2 0–10
12 74 31–205 6 2–30
18 – – 75 14–220
24 – – 308 56–520
30 – – 555 360–630
Note. From ages 18 to 30 months the CDI does not include
comprehension
vocabulary. This table is adapted from figures in Fenson, Dale,
Reznick, Bates,
Thal, & Pethick (1994).
Fig. 1. The coalition of cues available for establishing word
reference
and utilized differently across developmental time. Children
shift from
Phase I, a reliance on attentional cues such as how compelling
an object
is (perceptual salience) and the coincident appearance of an
object
and a label (temporal contiguity), to Phase II, a greater
dependency on
11. social and linguistic cues like eye gaze and grammar. By 12
months,
dependence on Phase I cues has begun to wane and shift to the
social cues
in Phase II.
Volume 15—Number 1 31
Roberta Michnick Golinkoff and Kathy Hirsh-Pasek
Alternatively, children sensitive to social cues should learn the
name for the object that the speaker labels, even if it is boring.
Hollich et al. (2000) found that, by 24 months, children
convincingly used social information, learning the names for
both the interesting and boring objects. Nineteen-month-olds
were still attracted to perceptual cues even though they could
use social information to learn the label for the boring object.
Twelve-month-olds showed an entirely different pattern. So-
cial information was necessary, but not sufficient, to ensure
word
learning. They only learned the novel word when social and
perceptual cues were ‘‘in alignment,’’ or when children heard
the
12. speaker label the interesting object. They failed to learn a word
when the speaker labeled the boring object. Had 12-month-olds
been pure associationists, they should have mismapped the
word, thinking that the novel word labeled the interesting object
regardless of speaker cues. The fact that they did not do this
suggests they detected the speaker’s social cues.
Is there ever a time in word acquisition when children mismap
labels, relying totally on the use of perceptual salience? Pruden,
Hirsh-Pasek, Golinkoff, and Hennon (in press) found that, un-
like their older peers, 10-month-olds were pure associationists,
mapping a novel word onto the object that they found the most
interesting, regardless of which object the speaker labeled. Ten-
month-olds acted as if social cues to reference did not exist.
These data suggest that the processes infants use for word
learning change over time. Beginning as associationists, chil -
dren move to attending to social cues, and then to recruiting the
speaker’s social cues to decide which object is being named.
WHAT CAUSES DEVELOPMENTAL CHANGE?
13. The discovery that word-learning processes change across the
first 2 years of life raises additional questions. How does the
perceptually driven 10-month-old become the socially aware
19-month-old? One interesting possibility is that around the end
of the first year, infants come to recognize people as intentional
beings who have goals, act autonomously, and act rationally
(e.g.,
Gergely & Csibra, 2003). Once infants understand other beings
as having minds and intentions distinct from their own, they can
recognize the relevance of those intentions for word learning.
Noting speaker intent allows infants to tap into the lexicons of
accomplished word learners so that they might add to their store
of vocabulary items.
There is, however, another account of the shift to the use of
social cues. As Perner and Ruffman (2005) argued, what
appears
to be sensitivity to social intent may be the ability to form an
association between speaker gaze and speaker talk. With word-
learning experience, children may note that people generally
14. look at things they talk about. This may lead children to begin
to
use social cues like eye gaze for word mapping even before they
understand the social intent behind the use of those cues. Even
this more restrictive social sensitivity would confer advantages
to learners. Once children restrict word-to-world mappings to
those objects that adults look at, they have narrowed the range
of
word referents. Of course, this restricted range still leaves many
alternative referents for a novel label (e.g., the shape, color, or
size
of the object). Here, principles or constraints such as ‘‘pay
atten-
tion to shape’’ (Smith, 2000) or ‘‘label the whole object’’
(Mark-
man, 1989) may also help children narrow their referent
choices.
A less restrictive social account that assumes children have
access to speaker intent allows for more rapid word learning.
Once children can make inferences about what the speaker in-
tends to label, they can learn words incidentally, from conver -
15. sation. Thus, the use of social information under either account
enhances word learning. Whether the use of social information
is
seen as accessing a speaker’s intent or as a more restricted as-
sociation of words with social cues, one can begin to explain
why
early word learning (at least production) is so slow (1 to 2
words
per week) relative to the fast-paced learning that occurs around
19 months of age, when children use social information.
Is it ever possible to distinguish between the use of associative
cues versus social intent? Finding definitive evidence to dis -
confirm reductionist views and affirm accounts that impute
more
sophisticated capabilities to children is not a simple problem.
One preliminary way is to pit associative and social cues
against
each other, as is done in the ECM framework (Hollich et al.,
2000). Another is to examine corollary findings suggesting
children’s sensitivity to social intent. That research is abundant.
By 18 months, for example, children complete a task on which
16. an
adult has feigned failure. Successful completion depends on
inferring adult intention (Meltzoff, 1995). Finally, studies
involving autistic children allow us to separate the effect of
attention to social cues from that of interpreting the intention
behind them. Since autistic children appear not to have the
ability to detect social intent, they learn words associatively
(Hennon, 2002; Priessler & Carey, 2005). Perhaps autistic
children’s vocabularies fail to grow at a rapid rate for this
reason.
Tests of hybrid models of word learning speak to much larger
issues within developmental psychology. Word learning is de-
ceptively simple, calling upon a range of processes that seem to
take on different values for the learner over time. The ECM
provides a window onto cognitive complexity and onto the var -
ious routes by which word learning can occur. In this way, the
ECM parallels current trends in the exploration of social de-
velopment that take multiple levels of influence into account. In
the study of cognitive development, this is rare. Understanding
17. the various pathways to word learning has implications for
crafting interventions that are targeted to the processes children
actually use at different ages to learn words.
Considering word learning from this vantage point might also
inform the controversy about how and when children develop an
understanding of other minds (e.g., Perner & Ruffman, 2005).
That is, children must come to appreciate that others have
thoughts, feelings, and perceptions different from their own. In
communication, this translates into the very motivation for
learning language: Wanting to share what one knows with
32 Volume 15—Number 1
Baby Wordsmith
someone who does not yet know it (Bloom & Tinker, 2001).
Concurrently, children should pay special attention to social
cues used in the context of language, for these cues provide a
conduit to what is on the mind of another person. In this sense,
the study of word learning offers yet another inroad into the
18. study
of how children conceptualize other minds.
Since Plato’s time, scholars have discussed how words map
onto the world. We have witnessed unparalleled progress in
understanding both the course of word learning and the mech-
anisms fueling that development. The birth of words is a psy-
chological watershed in language acquisition. The research
reviewed here suggests that competing theories are best united
under a hybrid view that incorporates changing mechanisms of
development. Our next task is to understand how children who
begin as associationists become social sophisticates.
Recommended Reading
Golinkoff, R.M., Hirsh-Pasek, K., Bloom, L., Smith, L.,
Woodward, A.,
Akhtar, N. Tomasello, M., & Hollich, G. (Eds.) (2000).
Becoming a
word learner: A debate on lexical acquisition. New York:
Oxford
University Press.
Hollich, G.J., Hirsh-Pasek, K., Golinkoff, R.M., Brand, R.J.,
Brown, E.,
Chung, H.L., Hennon, E., & Rocroi, C. (2000). (See References)
19. Waxman, S.R. & Lidz, J. (2006). (See References)
Acknowledgments—This research was supported by Grants
SBR9601306 and SBR9615391 to both authors from the Na-
tional Science Foundation.
REFERENCES
Baldwin, D.A., Markman, E.M., Bill, B., Desjardins, N., Irwin,
J.M., &
Tidball, G. (1996). Infants’ reliance on a social criterion for
establishing word–object relations. Child Development, 67,
3135–3153.
Bloom, L., & Tinker, E. (2001). The intentionality model and
language
acquisition: Engagement, effort, and the essential tension. Mon-
ographs of the Society for Research in Child Development,
66(Serial
No. 267).
Fenson, L., Dale, P., Reznick, S., Bates, E., Thal, D., &
Pethick, S.
(1994). Variability in early communicative development. Mono-
graphs of the Society for Research in Child Development,
59(Serial
No. 242).
20. Gergely, G., & Csibra, G. (2003). Teleological reasoning in
infancy: The
naı̈ ve theory of rational action. Trends in Cognitive Sciences, 7,
287–292.
Golinkoff, R.M., Mervis, C.V., & Hirsh-Pasek, K. (1994). Early
object
labels: The case for a developmental lexical principles
framework.
Journal of Child Language, 21, 125–155.
Hennon, E.A. (2002). How children with autistic disorder use
attentional
and intentional social information for word learning.
Unpublished
doctoral dissertation, Temple University.
Hollich, G.J., Hirsh-Pasek, K., Golinkoff, R.M., Brand, R.J.,
Brown, E.,
Chung, H.L., Hennon, E., & Rocroi, C. (2000). Breaking the
lan-
guage barrier: An emergentist coalition model for the origins of
word learning. Monographs of the Society for Research in Child
Development, 65(3, Serial No. 262).
Markman, E.M. (1989). Categorization and naming in children:
Prob-
lems of induction. Cambridge, MA: The MIT Press.
Meltzoff, A.N. (1995). Understanding the intentions of others:
21. Re-
enactment of intended acts by 18-month-old children. Develop-
mental Psychology, 31, 838–850.
Perner, J., & Ruffman, T. (2005). Infants’ insight into the mind:
How
deep? Science, 308, 214–216.
Preissler, M.A., & Carey, S. (2005). The role of inferences
about ref-
erential intent in word learning: Evidence from autism.
Cognition,
97, 813–823.
Pruden, S.M., Hirsh-Pasek, K., Golinkoff, R.M., & Hennon,
E.A. (in
press). The birth of words: Ten-month-olds learn words through
perceptual salience. Child Development.
Quine, W.V.Q. (1960). Word and object. Cambridge, England:
Cam-
bridge University Press.
Smith, L. (2000). Learning how to learn words: An associative
crane. In.
R.M. Golinkoff, K. Hirsh-Pasek, L. Bloom, L.B. Smith, A.L.
Woodard, N. Akhtar, M. Tomasello, & G. Hollich (Eds.),
Becoming
a word learner: A debate on lexical acquisition (pp. 51–80).
22. New
York: Oxford University Press.
Waxman, S.R. & Lidz, J. (2006). Early word learning. In D.
Kuhn &
R. Siegler (Eds.), Handbook of child psychology (6th ed.,Vol.
2).
New York: Wiley.
Volume 15—Number 1 33
Roberta Michnick Golinkoff and Kathy Hirsh-Pasek
Research Article
When Development and
Learning Decrease Memory
Evidence Against Category-Based Induction in Children
Vladimir M. Sloutsky and Anna V. Fisher
The Ohio State University
ABSTRACT—Inductive inference is crucial for learning: If one
learns that a cat has a particular biological property, one could
expand this knowledge to other cats. We argue that young
children perform induction on the basis of similarity of com-
23. pared entities, whereas adults may induce on the basis of cat-
egory information. If different processes underlie induction at
different points in development, young children and adults
would form different memory traces during induction, and
would subsequently have different memory accuracy. Experi -
ment 1 demonstrates that after performing an induction task,
5-year-olds exhibit more accurate memory than adults. Experi-
ment 2 indicates that after 5-year-olds are trained to perform
induction in an adultlike manner, their memory accuracy drops
to the level of adults. These results, indicating that sometimes
5-
year-olds exhibit better memory than adults, support the claim
that, unlike adults, young children perform similarity-based
rather than category-based induction.
The ability to make inductive generalizations is crucial for
learning: If
one learns that a cat has a particular unobserved biological
property,
one could extend this knowledge to other cats, and possibly to
other
24. mammals. Furthermore, by some accounts, ‘‘inductive inference
is the
only process . . . by which new knowledge comes into the
world’’
(Fisher, 1935/1951, p. 7).
There is much evidence that even infants and young children
can
perform simple inductions (Baldwin, Markman, & Melartin,
1993;
Gelman & Markman, 1986; Sloutsky, Lo, & Fisher, 2001;
Welder &
Graham, 2001). However, the representations and processes
under-
lying this ability remain unclear.
According to one view, people, including young children, hold
several conceptual assumptions that drive their induction (see
Keil,
Smith, Simons, & Levin, 1998, and Murphy, 2002, for reviews
of these
assumptions). In particular, people hold a category
assumption—they
assume that each individual entity is a member of a class or
category,
25. that count nouns refer to categories, and that members of the
same
category share many unobserved properties. Conceptual
assumptions
are a priori—they are not learned, but are rather a precondition
of
learning, and are present early in development (Gelman &
Hirschfeld,
1999; Keil et al., 1998). In the course of induction, people first
identify presented entities as members of categories and then
perform
inductive inferences on the basis of categorization (Gelman,
1988;
Gelman & Markman, 1986). Therefore, when presented with a
rabbit
and told that it has hollow bones inside its body, a child is more
likely
to generalize this property to another rabbit than to a dog
because the
child (presumably) understands that both rabbits belong to the
same
category, and members of the same category share many
properties. It
has been argued that this tendency to perform induction on the
26. basis
of categorization, or category-based induction, is especially
pro-
nounced when entities are members of familiar categories
(Davidson
& Gelman, 1990). In short, according to this view, induction is
a
function of categorization, whereas categorization is a function
of a
priori conceptual assumptions.
According to another view, young children perform induction
(as
well as categorization) by detecting multiple correspondences,
or
similarities, among presented entities (e.g., see Jones & Smith,
2002;
McClelland & Rogers, 2003; Sloutsky, 2003; Sloutsky et al.,
2001).
Because members of a category often happen to be more similar
to
each other than they are to nonmembers, young children are
more
likely to induce unobserved properties to members of the
category than
27. to nonmembers. According to this view, induction and
categorization in
young children are variants of the same process, which is driven
by the
detection of multiple correspondences rather than by a priori
con-
ceptual assumptions. Furthermore, conceptual knowledge often
found
in adults (e.g., knowledge that entities are members of
categories) is
not a priori, but is a product of learning and cognitive
development.
Learning accounts of conceptual knowledge support this
position,
while weakening the claims that conceptual knowledge is a
priori. For
example, it has been claimed that young children’s tendency to
use
similar shape as a reliable categorization cue is a product of a
priori
conceptual knowledge (Diesendruck & Bloom, 2003; Soja,
Carey, &
Spelke, 1991), whereas a convincing learning account of this
shape
28. bias (Smith, Jones, Landau, Gershkoff-Stowe, & Samuelson,
2002) has
weakened the a priori claims by rendering them unnecessary.
Address correspondence to Vladimir M. Sloutsky, Center for
Cogni-
tive Science, 208C Ohio Stadium East, 1961 Tuttle Park Place,
The
Ohio State University, Columbus, OH 43210; e-mail:
[email protected]
osu.edu.
PSYCHOLOGICAL SCIENCE
Volume 15—Number 8 553Copyright r 2004 American
Psychological Society
Overall, the two positions have several fundamental differences.
According to the former position, when entities are members of
fa-
miliar categories, induction is a function of categorization, and
cate-
gorization is a function of conceptual knowledge. Therefore,
induction
is a function of conceptual knowledge. In addition, conceptual
knowledge is a priori rather than learned. According to the
latter
29. position, early in development induction and categorization are
a
function of perceptual similarity among entities, whereas
conceptual
knowledge is a product of learning and development. Thus, the
two
positions assume different kinds of processing underlying
induction
and different developmental courses of induction and
categorization.
One way of contrasting these theoretical positions is to compare
pre-
dictions derived from them. For example, there is evidence in
memory
research that spontaneous categorization of items may lead to
memory
distortions, such as false recognition of critical lures, or
nonpresented
items that belong to the same category as previously presented
items
(Koutstaal & Schacter, 1997). These distortions may occur
because
participants form category-level or gist representations, whereas
de-
30. tails of each individual item are not encoded or are encoded
poorly
(Brainerd, Reyna, & Forrest, 2002; Koutstaal & Schacter,
1997). When
participants are required to focus on perceptual properties of
items,
they amply encode individual items, thus exhibiting accurate
memory
(Marks, 1991; McDaniel, Friedman, & Bourne, 1978).
Thus, similarity-based induction and category-based induction
may
result in differential remembering of information presented
during an
induction task: Whereas similarity-based induction may lead to
accu-
rate memories for perceptually distinct individual items,
category-based
induction may result in memory distortions, such as poor
discrimination
of presented items and critical lures. Therefore, if an induction
task
precedes a memory test, the memory test would reveal
processing un-
31. derlying induction. If people perform category-based induction
and
form category-level memory traces, their ability to discriminate
items
seen during the induction task from critical lures should be poor
(compared with their performance on a baseline no-induction
task).
However, if they perform similarity-based induction, they
should amply
encode perceptual information, forming item-specific memory
traces,
and their discrimination should be as high as the baseline level.
It has been argued that when entities are members of familiar
categories, adults may perform induction in a category-based
manner
(Osherson, Smith, Wilkie, Lopez, & Shafir, 1990), in which
case an
induction task should attenuate their recognition memory
compared
with the baseline. In contrast, if young children perform
induction in a
similarity-based manner, they should exhibit high accuracy in
both
32. baseline and induction conditions. Thus, following an induction
task,
young children may exhibit greater memory accuracy than
adults. The
prediction is nontrivial because typically adults’ memory is
markedly
more accurate than that of young children (see Schneider &
Bjork-
lund, 1998, for a review).
If adults’ induction with familiar categories is indeed category
based, whereas young children’s induction is similarity based,
how
does this category-based induction develop? The category-based
po-
sition argues that conceptual knowledge (e.g., the category
assump-
tion) is a priori rather than learned (Gelman & Hirschfeld, 1999;
Keil
et al., 1998). However, providing a learning account of
category-based
induction weakens this position by rendering the a priori nature
of
conceptual knowledge unnecessary.
33. To test the target prediction, we conducted Experiment 1, in
which
we compared the effects of an induction task on recognition
memory of
5-year-olds and adults. In Experiment 2, we trained 5-year-olds
to
perform category-based induction and examined the effects of
training
on their recognition memory.
EXPERIMENT 1
Method
Participants
Participants were 77 young children (M age 5 5.43 years, SD 5
0.28
years) and 71 introductory psychology students (M age 5 19.3
years,
SD 5 1.33).
Materials, Design, and Procedure
Materials were 44 color photographs of animals presented
against a
white background (see Fig. 1 for examples of the stimuli).
During the
34. study phase, participants were presented with 30 pictures, 1
picture at
a time, from three categories (10 cats, 10 bears, and 10 birds).
During
the recognition phase, they were presented with 28 pictures, 1
picture
at a time, and were asked whether they had seen each exact
picture
during the study phase. Half of the recognition pictures had
been
presented during the study phase, and the other half were new
pic-
tures. These recognition pictures also represented three
categories:
cats (7 old and 7 new), bears (all 7 old), and squirrels (all 7
new). To
ascertain that all of these animals were well familiar to
children, we
pretested them with 5-year-olds in an earlier naming study.
Only those
pictures that were consistently named by a basic-level name
(i.e.,
‘‘cat,’’ ‘‘bear,’’ ‘‘bird,’’ or ‘‘squirrel’’) by more than 85% of
the children
35. were selected for the present study.
The experiment included three between-subjects conditions:
baseline, induction, and blocked categorization, with each
condition
consisting of a study phase and a recognition phase. The
recognition
phase was identical in all three conditions, whereas the study
phase
differed across the conditions. Participants were randomly
assigned to
Fig. 1. Examples of stimuli used in the study and recognition
phases
(copied from the Corel Draw database).
554 Volume 15—Number 8
Induction and Memory
one of the three conditions. There were 27 children and 29
adults in the
induction condition, 24 children and 23 adults in the blocked-
categori-
zation condition, and 26 children and 19 adults in the baseline
condition.
In the study phase of the baseline condition, participants were
36. presented with 30 pictures of animals, and their task was to
remember
these pictures for a subsequent recognition test.
In the study phase of the induction condition, participants were
first
presented with a picture of a cat and informed that it had ‘‘beta
cells
inside its body.’’ Participants were then presented with 30
pictures of
animals (identical to those presented in the baseline condition)
and
were asked whether each of the presented animals also had beta
cells.
After responding, the participants were provided with yes/no
feedback
indicating that only cats, but not bears or birds, had beta cells.
The
fact that 5-year-olds do not know what beta cells are was of no
concern
because young children easily induce unfamiliar properties or
blank
predicates, and as we show in the Results and Discussion
section, they
37. had no difficulty inducing this property. The recognition test
was not
mentioned in the study phase of this condition.
In the study phase of the blocked-categorization condition, par-
ticipants were first presented with a picture of a cat and
informed that
it was young. Participants then were presented with 30 pictures
of
animals (identical to those presented in the baseline and
induction
conditions) and asked whether each of the presented animals
was
young or mature. Participants were provided with random
yes/no
feedback. The purpose of this random feedback was to block
infer-
ences based on the animal-kind information and to force
participants
to focus on perceptual features of individual items. As in the
induction
condition, the recognition test was not mentioned in the study
phase.
The recognition phase was presented immediately after the
study
38. phase. At recognition, participants were presented with 28
pictures
and were asked to determine whether each was ‘‘old’’ (i.e.,
exactly the
one presented during the study phase) or ‘‘new.’’ No feedback
was
provided during the recognition phase. The young children were
tested individually in their day-care centers by female
experimenters
blind to the hypotheses. The undergraduate students were tested
in-
dividually in a laboratory on campus. For all participants,
stimuli were
presented on a computer screen in a self-paced manner, and
stimulus
presentation was controlled by SuperLab Pro 2 (1999) software.
Results and Discussion
Recall that young children were expected to perform induction
by
comparing each animal with the target animal and thus to
remember
study-phase animals well, accurately accepting old animals and
re-
39. jecting new ones. At the same time, it was expected that adults
would
spontaneously categorize animals as cats, bears, and birds when
performing induction, and thus they would form gist or
category-level
memory traces. As a result, it was expected that they would
remember
category information, but not item-specific information, and
thus fail
to discriminate between old items and critical lures (i.e., new
mem-
bers of a studied category). Also, recall that the blocked-
categoriza-
tion condition was identical to the induction condition, except
that the
categorization of animals as cats, bears, and birds was blocked.
Be-
cause categorization was blocked, it was expected that memory
ac-
curacy for both children and adults in this condition would be
comparable to accuracy in the baseline condition.
After several trials, the majority of 5-year-olds and adults
realized
40. that the property of having beta cells should be induced to cats,
but
not to bears or birds, and they accurately performed this
induction:
The average rate of correct induction was over 75% of trials for
both
children and adults. Also, in the recognition phase, both
children and
adults exhibited above 92% accuracy across conditions in
rejecting
distractors from an unstudied category (i.e., squirrels).
Therefore,
participants took the task seriously and paid attention to stimuli
during the study phase.
Recall that we were interested in participants’ discrimination of
old
items and critical lures across conditions. Percentages of hits
(i.e.,
correct recognitions) and false alarms on critical lures by age
group and
condition are presented in Table 1. Data in the table indicate
that
whereas children exhibited equivalent accuracy (i.e., hits �
41. false
alarms) across the baseline, induction, and blocked-
categorization con-
ditions, F(2, 76) < 1, adults’ accuracy was dramatically lower in
the
induction condition than in the other two conditions, F(2, 68) 5
11.5,
p < .0001, Zp
2
5 .252, both ps < .01 in post hoc Tukey tests.
To further examine the ability to discriminate old items from
critical
lures, we computed memory sensitivity A0 scores. A0 is a
nonpara-
metric analogue of the signal detection statistic d0 (Snodgrass
&
Corwin, 1988; Wickens, 2002). If participants do not
discriminate old
items from critical lures, A0 is at or below .5. The greater the
dis-
crimination accuracy, the closer A0 is to 1. A0 scores are
presented in
Figure 2. As predicted, 5-year-olds discriminated old items
from
critical lures well across the three conditions (in all conditions,
42. A0s > .5, one-sample ts > 3, ps < .005). Adults were accurate in
the
baseline and blocked-categorization conditions (both A0s > .5,
one
sample ts > 7, ps < .001), whereas they were not accurate in the
induction condition: Unlike the A0s of the young children,
adults’ A0s
in this condition were not different from .5, t < 1, indicating the
adults did not discriminate between old items and critical lures.
A0 values were submitted to a two-way (Age � Experimental
Con-
dition) analysis of variance. The analysis confirmed a
significant age-
by-condition interaction, F(2, 142) 5 4.64, p 5 .001, Zp
2
5 .06.
Whereas 5-year-olds exhibited no differences in accuracy across
the
conditions, all ps > .87, A0s in adults were markedly lower in
the
TABLE 1
Mean Proportions of Hits and False Alarms (FA) and Mean
Accuracy in Experiment 1
43. Condition
Children Adults
Hits FA
Accuracy
(hits � FA) Hits FA
Accuracy
(hits � FA)
Baseline .77 (.19) .50 (.32) .27 .89 (.10) .47 (.31) .42
Induction .72 (.24) .41 (.34) .31 .83 (.20) .76 (.25) .07
Blocked categorization .78 (.13) .51 (.20) .27 .80 (.18) .50 (.24)
.30
Note. Standard deviations are in parentheses.
Volume 15—Number 8 555
Vladimir M. Sloutsky and Anna V. Fisher
induction condition than in the other two conditions, F(2, 68) 5
13.5,
p < .001, Zp
2
5 .29, post hoc Tukey test ps < .001. Furthermore, as
predicted, in the induction condition, 5-year-olds exhibited
greater
44. accuracy than adults, one-tailed t(54) 5 2.36, p 5 .011.
In short, the induction task markedly attenuated adults’
recognition
accuracy, whereas young children remained accurate. These
results
suggest that whereas adults performed category-based
induction, young
children performed similarity-based induction.
It could be argued, however, that the children were accurate be-
cause of extraneous factors. For example, the children could
have
been more interested in the pictures than the adults, or the
children
could have forgotten gist information faster than item-specific
infor-
mation, whereas the adults could have forgotten item-specific
infor-
mation faster than gist information. The goal of Experiment 2
was to
eliminate these explanations by training children to perform
category-
based induction. If our hypothesis is correct, such training
should
45. differentially affect young children’s memory across conditions:
Although their accuracy should drop in the induction condition
(analogous to the drop for adults in Experiment 1), it should not
drop
in the baseline condition. In addition to providing controls,
Experi-
ment 2 (if successful) would provide a learning account of
category-
based induction found in adults, thus weakening the claim that
cat-
egory-based induction is based on a priori conceptual
knowledge.
EXPERIMENT 2
Method
Participants
Participants were 42 young children (M age 5 5.25 years, SD 5
0.21
years), with 26 participating in the baseline condition and 16
par-
ticipating in the induction condition.
Materials, Design, and Procedure
46. Materials in both conditions were identical to those in
Experiment 1.
There were two between-subjects conditions, induction and
baseline,
and participants were randomly assigned to one of the tw o
conditions.
The procedure of Experiment 2 differed from that of Experiment
1
in that prior to the recognition phase, the 5-year-olds were
presented
with training in which they were taught to perform category-
based
induction. They were first taught that animals that have the
same name
belong to the same category—‘‘they are the same kind of
animal.’’
They were then given three boxes, with each box identified by a
black
outline of a lion, a rabbit, or a dog, and were presented with
pictures of
lions, rabbits, and dogs (none of these categories was presented
during
the main experiment). They were told that animals that have the
same
47. name are the same kind of animal and could be placed in the
same
box. The children were asked to place the pictures in the boxes
face
down. All presented pictures had been pretested in a prior
naming
study that revealed that each of the depicted animals could be
reliably
named by 5-year-olds. The children were presented with six
catego-
rization trials, and yes/no feedback was given after each trial.
Both
types of feedback were accompanied by an explanation that
animals
that have the same name belong to the same kind and should be
placed in the same box.
The categorization training was followed by induction training.
Participants were first reminded that animals that have the same
name
are the same kind of animal. They were then told that animals of
the
same kind have ‘‘the same stuff inside.’’ Then participants were
given
48. six induction trials, each accompanied with yes/no feedback. On
each
trial, they were shown a picture and told that the animal had a
par-
ticular biological property (e.g., ‘‘this dog has thick blood
inside its
body’’), and asked to place the picture in an appropriate box.
Feed-
back was followed by an explanation that animals of the same
kind
have the same name and same stuff inside. All children
completed
training successfully, giving either five correct answers out of
six or
four correct answers in a row in the induction training task. At
the
conclusion of the training session, they were reminded that
‘‘animals
that have the same name are the same kind of animal, and these
animals have the same stuff inside.’’ They were then presented
with
the main experiment, which was identical to Experiment 1.
Results and Discussion
49. Hits, false alarms, and recognition accuracy (hits � false
alarms)
across the conditions are presented in Table 2. Data in the table
point
to a marked difference between 5-year-olds’ high accuracy in
the
baseline condition and low accuracy in the induction condition,
F(1,
40) 5 12.18, p < .005, Zp
2
5 .24. Although accuracy in the baseline
condition remained as high as in Experiment 1, accuracy in the
in-
duction condition dropped dramatically to the level of adults in
the
induction condition of Experiment 1.
Children’s A0 scores across the two experiments are presented
in
Figure 3. As shown in the figure, children’s accuracy in the
induction
condition of Experiment 2 dropped compared with their
accuracy in
the baseline condition of Experiment 2, one-tailed independent-
samples t(40) 5 3.4, p < .005, as well as compared with their
accu-
50. racy in the induction condition of Experiment 1, one-tailed
inde-
pendent-samples t(41) 5 1.7, p < .05. Furthermore, after
training,
their accuracy in the induction condition did not differ
significantly
from .5, or from that of adults in Experiment 1, both ps > .28,
thus
Fig. 2. Memory sensitivity scores (A0) of children and adults
across the
three experimental conditions in Experiment 1. The dashed line
repre-
sents the point of no sensitivity. Error bars represent the
standard errors
of the mean.
556 Volume 15—Number 8
Induction and Memory
indicating that like adults in Experiment 1, 5-year-olds who had
re-
ceived training in category-based induction failed to
discriminate
between old items and critical lures. At the same time, their
accuracy
51. remained high in the baseline condition (A0 > .5), one-sample
t(25) 5 7.7, p < .0001.
In short, training to perform category-based induction
attenuated
memory accuracy of 5-year-olds in the induction, but not the
baseline,
condition. These findings suggest that the high recognition
accuracy
exhibited by 5-year-olds in Experiment 1 did not stem from
extra-
neous factors, but rather stemmed from similarity-based
induction
leading to accurate representation of item-specific information.
These
findings also provide a learning account of category-based
induction.
GENERAL DISCUSSION
In the present study, (a) the induction task decreased
recognition
accuracy of adults, whereas young children exhibited high
recognition
accuracy, and (b) training to perform category-based induction
de-
52. creased recognition accuracy of young children in the induction,
but
not in the baseline, condition.
Recall that similarity-based induction is expected to result in
item-
specific representations, whereas category-based induction is
ex-
pected to result in category-level representations. Therefore,
category-
based, but not similarity-based, induction may lead to errors in
rec-
ognition memory. Hence, finding that young children (unlike
adults)
exhibit high recognition accuracy after an induction task
suggests that
young children do not spontaneously perform category-based
induc-
tion, but rather perform similarity-based induction. This finding
is new
evidence challenging the idea that when categories are familiar,
in-
duction in young children is driven by the category assumption.
There is mounting evidence challenging the claim that young
53. children hold a priori conceptual assumptions; however, much
of this
evidence challenges the centrality assumption—the idea that
young
children assume rather than learn differential importance of
different
properties. It was shown that participation in a learning task in
which
allegedly central features (i.e., matching labels) were poor
predictors
of biological properties, whereas allegedly peripheral features
(i.e.,
similar appearances) were good predictors, resulted in young
chil-
dren’s ignoring labels in favor of appearances in an induction
task
presented 3 months later in a different context by a different
exper-
imenter (Sloutsky & Spino, in press). In addition, children’s
reliance
on linguistic labels (which are allegedly more central than
perceptual
similarities) in an induction task is more pronounced for line-
drawing
54. pictures than for real three-dimensional objects (Deak & Bauer,
1996). Furthermore, perceptual similarity could be more
important for
induction than are matching labels: Young children are more
likely to
rely on similarity of motion than on a matching linguistic label
(Mak &
Vera, 1999). This evidence seriously undermines the idea of an
a
priori centrality assumption because properties of ‘‘peripheral’’
in-
formation should not affect the centrality of ‘‘essential’’
information.
Unlike previous research, the current research challenges the
cate-
gory assumption: There is little evidence that young children
spon-
taneously perform induction in a category-based manner.
Finding that training to perform induction in a category-based
manner reduced memory accuracy of young children to the level
of
adults supports a learning account of category-based induction,
sug-
55. gesting that it is unnecessary to posit that conceptual knowledge
is a
priori. Recall that in Experiment 2, participants were taught that
(a)
animals that have the same name belong to the same kind, (b)
animals
that belong to the same kind have the same stuff inside, and (c)
animals that have the same name have the same stuff inside. It
is
possible that the first two points are taught in school, whereas
the third
is a direct consequence of the first two. Therefore, the results of
Experiment 2 may explain the transition from the similarity-
based
induction exhibited by children to the category-based induction
ex-
hibited by adults, suggesting that category-based induction and
req-
uisite conceptual knowledge could be a product of feedback-
based
learning. These findings are consistent with previous research
indi-
cating that smart behaviors do not have to be a priori —they can
de-
56. velop from simpler representations and processes (e.g., Jones &
Smith,
2002; Smith, Jones, & Landau, 1996; Smith et al., 2002).
Could the reported results stem from extraneous factors, such as
differential forgetting of gist and item-specific information or
differ-
ential interest in pictures in children and adults? Although these
factors could account for the results of Experiment 1, it is
unclear how
they can account for the results of both experiments. Recall that
training to perform category-based induction attenuated
recognition
memory of 5-year-olds in the induction but not the baseline
condition.
TABLE 2
Mean Proportions of Hits and False Alarms (FA) and Mean
Accuracy in Experiment 2
Condition Hits FA
Accuracy
(hits � FA)
Baseline .74 (.16) .42 (.28) .32
57. Induction .85 (.21) .77 (.29) .08
Note. Standard deviations are in parentheses.
Fig. 3. Young children’s memory sensitivity scores (A0) in the
induction
and baseline conditions in Experiments 1 and 2. The dashed line
rep-
resents the point of no sensitivity. Error bars represent the
standard
errors of the mean.
Volume 15—Number 8 557
Vladimir M. Sloutsky and Anna V. Fisher
It seems very difficult to come up …