Young Children Know That Trying Is Not Pretending: A Test of the "Behaving-As-If" Construal of Children's Early Concept of Pretense

Developmental Psychology
Copyright 2004 by the American Psychological Association
2004, Vol. 40, No. 3, 388 –399
0012-1649/04/$12.00
DOI: 10.1037/0012-1649.40.3.388
Young Children Know That Trying Is Not Pretending: A Test of the
“Behaving-As-If” Construal of Children’s Early Concept of Pretense
Hannes Rakoczy, Michael Tomasello, and Tricia Striano
Max Planck Institute for Evolutionary Anthropology
In 3 studies, young children were tested for their understanding of pretend actions. In Studies 1 and 2,
pairs of superficially similar behaviors were presented to 26- and 36-month-old children in an imitation
game. In one case the behavior was marked as trying (signs of effort), and in the other case as pretending
(signs of playfulness). Three-year-olds, and to some degree 2-year-olds, performed the real action
themselves (or tried to really perform it) after the trying model, whereas after the pretense model, they
only pretended. Study 3 ruled out a simple mimicking explanation by showing that children not only
imitated differentially but responded differentially with appropriate productive pretending to pretense
models and with appropriate productive tool use to trying models. The findings of the 3 studies
demonstrate that by 2 to 3 years of age, children have a concept of pretense as a specific type of
intentional activity.
Young children engage with regularity in pretend activities, but
counterfactual proposition that one believes to be false. Rather,
it is not entirely clear how they understand what they (or others
young children have a concept of “pretending-that-p” as “behaving
when they are pretending) are doing. There are currently two main
in a way that would be appropriate if p (the counterfactual situa-
theories. On the one hand, Leslie (1987, 1988, 1994, 2002) and
tion) were the case” (Nichols & Stich, 2000, p. 139). That is,
Fodor (1992) have argued that children as young as 2 years of age
young children’s concept of pretense is much more coarse-grained
apply the same concept of pretense as do adults. In pretending that
than the mature one and has a much bigger extension than the class
a telephone is a banana (holding the telephone to the mouth,
of pretense actions. Accordingly, it does not allow for distinguish-
making chewing movements, saying “Mmm,” etc.), for example,
ing pretending from other kinds of as-if behaviors, for example,
and in observing someone else pretend in this way, children do not
mistakes such as biting into the telephone because one thinks it is
represent the counterfactual situation “this is a banana” as literally
really a banana (or biting into it by accident).
true (or else they would sink their teeth into receivers). Rather, to
The behaving-as-if theory thus predicts that young children
avoid this, the child makes use of a specialized innate cognitive
should make overextension mistakes, applying their concept of
architecture involving an adult concept of pretense, meta-
pretending both to behaving-as-if unknowingly and to behaving-
representing his or her own and others’ pretense in the form
as-if unintentionally. Evidence for the first kind of mistake in
“person pretends (this is a banana).”
young children comes from studies by Perner et al. (1994) and
The competing account is what could be called the “behaving-
Lillard (1993, 1996). In Perner et al.’s (1994) study, 3-year-olds
as-if” construal of children’s early pretense performance and un-
tended to say that a person who behaved as if there was a rabbit in
derstanding (e.g., Harris, 1994; Jarrold, Carruthers, Smith, &
a cage because he or she mistakenly believed there was one there
Boucher, 1994; Lillard, 1994; Nichols & Stich, 2000; Perner,
was “pretending” that there was a rabbit in the cage. In a series of
Baker, & Hutton, 1994). The basic contention of this theory is that
experiments by Lillard (1993), 4- and 5-year-old children were
young children do not yet have the mature adult concept of
told that a troll named Moe (a) hopped around like a rabbit and (b)
pretending as acting intentionally and knowingly according to a
lacked any knowledge about rabbits. The test question was
whether Moe was pretending to be a rabbit as he hopped around.
The vast majority of 4-year-olds and even many 5-year-olds
wrongly answered “yes.” That is, children up to 5 years of age
Hannes Rakoczy and Michael Tomasello, Department of Developmen-
seemed not to understand that pretense is an activity performed on
tal and Comparative Psychology, Max Planck Institute for Evolutionary
the basis of a mentally represented counterfactual situation. In
Anthropology, Leipzig, Germany; Tricia Striano, Junior Research Group
another set of studies, 4-year-olds tended to say that one could
for Cultural Ontogeny, Max Planck Institute for Evolutionary Anthropol-
ogy, Leipzig, Germany.
pretend without using one’s brain or needing a mind (Lillard,
We thank Josef Perner, Paul Harris, Angeline Lillard, Tim German, and
1996). Most surprisingly, when presented with inanimate objects
Maxim Abelev for helpful discussion and Dorit Jaschke, Kristin Liebal,
that showed as-if behavior, 4-year-olds answered the question
Jana Jurkat, Anja Lange, and Andrea Kobiella for help with data collection
“Can it pretend?” affirmatively (Lillard, Zeljo, Curenton, & Kau-
and coding. Many thanks also to Eva Leerman, Jana Jurkat, and Angela
gars, 2000). It thus seems that, in their verbal discourse at least,
Loose for help with recruiting children and to all day-care centers and
young children do not understand the cognitive prerequisites of
parents for their friendly cooperation.
pretense. This seems to make sense in light of the fact that children
Correspondence concerning this article should be addressed to Hannes
before the age of 4 years do not master standard false belief tasks
Rakoczy, Department of Developmental and Comparative Psychology,
Max Planck Institute for Evolutionary Anthropology, Deutscher Platz 6,
(indexing, on many accounts, that they do not yet have a proper
D-04103 Leipzig, Germany. E-mail: rakoczy@eva.mpg.de
concept of belief). There is thus solid experimental evidence, using
388

---

UNDERSTANDING TRYING AND PRETENDING
389
verbal paradigms at least, for the claim of the behaving-as-if theory
tively and creatively. Our prediction was that if children under-
that young children do not understand the cognitive dimensions of
stand pretending and trying as two different intentional activities,
pretense.
they should show the following response pattern: After pretense
But it is still not clear whether young children might understand
demonstrations, they should perform the pretense action them-
the intentional dimensions of pretense, that is, that the person
selves (Studies 1 and 2) or a productive pretense action that
pretending has a special goal that is different from the goal in other
“follows” from the experimenter’s pretense (Study 3). After trying
types of intentional activity. In a recent study with 4-year-olds
demonstrations, in contrast, they should try to perform the real
(Lillard, 1998), children were again told that Moe hopped like a
action themselves, or really perform the real action, or indicate in
rabbit, but then they were told that Moe did not want to, or was not
some other way that their goal is to perform the action properly
trying to, hop like a rabbit. When asked “Is Moe pretending to be
(Studies 1 and 2) or to try to perform the real action productively
a rabbit?” most 4-year-olds wrongly answered affirmatively. Lil-
with different means (Study 3).
lard concluded from these findings that 4-year-olds’ concept of
pretense is not one of intentionally acting-as-if but is a superficial
Study 1
one only of behaving-as-if. However, this study used a very
stringent task to tap children’s understanding of pretense as an
Method
intentional activity, involving them in very complicated discourse
about a rather artificial scenario. How does the child understand a
Participants
story about a character hopping like a rabbit without wanting to
Twenty-four young 2-year olds (25–29 months, mean age
27 months;
when there is no further explanation given for the character’s
12 boys and 12 girls) and 24 young 3-year-olds (34 –38 months, mean
behavior? Indeed, in a recent study, Richert and Lillard (2002)
age
36 months; 14 boys and 10 girls) were included in the final sample.
found that when a reason was given for Moe’s behavior— he was
Children were recruited in urban day-care centers (n
41) or by telephone
walking on hot pavement and did not want to burn his feet—the
from a list of parents and children who had volunteered for studies of child
children performed better. These results suggest that the children
development (n
7). Children came from mixed socioeconomic back-
in the original study might have simply ignored the premise that
grounds and were all native German speakers. Testing was done by one
experimenter in a separate quiet room of the children’s day-care center or
Moe did not want to hop like a rabbit in coming up with their
in a child psychology laboratory, and sessions were videotaped for subse-
answers.1
quent analysis. An additional 4 two-year-olds and 4 three-year-olds were
In the current study, we pursued the hypothesis that young
tested but had to be excluded from the study because they were
children do not have only one undifferentiated category of as-if
uncooperative.
behaviors but have a concept of pretending as intentionally acting-
as-if, which is different from other types of as-if-behaviors, and
Materials and Design
that they employ this concept before they become proficient with
the lexical semantics of words for pretending in such difficult tasks
Figure 1 shows the objects that were used in the test phase. Object Sets
as the Moe test. This hypothesis is plausible given recent findings
A and B were used to try/pretend to write. In both cases, the experimenter
that from their 2nd year, children show some understanding of the
would use the object to make writing or drawing movements on paper. He
made these movements, with some pauses in which he looked at the object,
intentional structure of different human behaviors, as indexed, for
for about 15 s. In the pretending case, his overall expression was playful,
example, in their differential imitation of others’ intentional, ac-
he looked at the object during the pauses in an amused way, and he marked
cidental, and failed acts (e.g., Carpenter, Akhtar, & Tomasello,
his writing movements with sounds (“Hmm. Ahh” as if looking at his
1998; Gergely, Bekkering, & Kira´ly, 2002; Meltzoff, 1995) and in
graphic production). In the trying case, he looked at the object with a
the light of findings that young children from at least 2 years of age
surprised, frustrated expression and made a corresponding sound effect
imitate other people’s pretense actions (e.g., Rakoczy, Tomasello,
(“Hmm?” as if saying “What is wrong here?”). The object sets could be
& Striano, 2002; Watson & Fischer, 1977).
made to work by taking off the caps, behind which there were pencil leads.
We tested this hypothesis by showing children a person either
Object Sets C and D were used to try/pretend to pour. Both contained
pretending or trying to do something (in both cases the person did
water, visible to the child. From the container in Object Set C the exper-
not actually perform the act to a final result). Thus, children saw
imenter tried/pretended to pour water into a cup by making repeated
the same act presented in one of two different forms. In the first
form, the demonstrator was trying (unsuccessfully) to do some-
1 There is also one recent study that questions the validity of the Lillard
thing, for example, to write with a pen. In the second form, the
(1998) findings by showing some awareness of the relations between
demonstrator was pretending to do something, for example, to
intending and pretending in 3-year-olds (Joseph, 1998). But this study, too,
write with a pen. Both models were superficially alike: The dem-
involved children in confusing discourse and had some methodological
onstrator made writing movements with the pen on a sheet of
shortcomings. Children in this study were presented with stories about two
paper, but no marks were made on the paper. The first model was
persons. A showed an involuntary behavior, for example, sneezing; B
marked by signs of surprise and frustration as trying to write; the
pretended to show the same behavior, for example, pretending to sneeze.
second model was marked by signs of fun and playfulness as
The test question was “Who is trying to sneeze?” and the correct answer
pretending to write. The child was then given the object. Impor-
was supposed to be B (because the behavior happened to A, whereas B
performed it intentionally). Sixty percent of the 3-year-olds gave this
tantly, the object could be made to work on closer inspection; for
supposedly correct answer. However, this answer in fact is not correct for
example, the pen could be made to really write. In Studies 1 and
at least two reasons (actually the correct answer would be “neither A nor
2, children were instructed to play an imitation game, and differ-
B”). First, part of the point of pretending is that one does not try to perform
ential imitation in response to the experimenter’s model action was
the real action. Second, “try to X” can only be used when X is an action
the dependent measure. In Study 3, children were not instructed to
verb (see, e.g., Goldman, 1970), but “sneeze” is not an action verb.
strictly imitate but were given the chance to react more produc-
Therefore, it remains unclear how children’s answers are to be interpreted.

---

390
RAKOCZY, TOMASELLO, AND STRIANO
topic to the pretending versus trying conditions were all systematically
varied across children. The order of the topics within both blocks was the
same (e.g., when a child had “writing” first in the pretending block, he or
she also had “writing” first in the trying block).
Procedure
In the beginning, the experimenter and the child played freely with
different toys until the child felt comfortable. Then there was an introduc-
tion phase in which the imitation game was set up. The experimenter
explained to the child that he was now going to show him or her some
interesting things he had brought along, that he was going do something
with these things, and that the child himself or herself could then perform
the same action with the objects that the experimenter had performed. He
then performed an action (e.g., pressing a wooden nail into a pegboard),
gave the objects to the child while saying “It’s your turn now!” and
reinforced the child for imitation. If the child did not imitate, the experi-
menter repeated that the child should do what he had done until the child
imitated. When the child participated well in the imitation game, the
Figure 1.
Objects used in the two test model blocks of Studies 1 and 2.
experimenter then started the actual test phase, which consisted of the two
model blocks. For half of the children, the first model block was pretense;
for the other half, it was trying. Before each model block there was a short,
pouring movements over the cup for about 15 s, again with some pauses in
specific warm-up. The experimenter showed the child three simple pre-
which he looked at the container. The pretending case was marked by a
tense actions that could not really be performed (e.g., pretending to brush
general playful expression with corresponding sound effects (“shhh”—like
his teeth with an object) before the pretense block and three simple trying
the sound of water flowing into the container). The trying case, in contrast,
actions that could not be pretended before the trying block (e.g., trying to
was marked again with a puzzled, disappointed expression while looking at
open an object). He then gave the object to the child, saying “It’s your
the object (“Hmm?” in the sense of “Why is there no water?”). The cases
turn!” In this phase, the experimenter still differentially reinforced imita-
with Object Set D were the same except that the pouring movements were
tions. In the two model blocks, after the eight actual test trials, the
done above the flower. In both cases, real pouring could be achieved by
experimenter then no longer differentially reinforced any responses by the
removing a lid from the container.
child but reacted in an equally positive way to all actions shown by the
Object Sets E and F were used to try/pretend to drink. Both the semi-
child. A session lasted approximately 15 min.
novel bottle (Object Set E) and the container from Object Set F contained
water, and the child could see and hear the water. In pretending/trying to
drink from the bottle, the experimenter held the bottle above his mouth and
Observational and Coding Procedure
shook it several times (as if making water come out into his mouth), with
some pauses in which he looked at the bottle, for about 20 s. The
All sessions were videotaped and coded from the videotape by a single
pretending case was marked by playful expression and drinking sound
observer. Each response of a child to the eight test models was coded.
effects. The trying case was marked by surprised and frustrated looks at the
There were three categories. A given response was coded as a pretense
bottle in the pauses and corresponding sound effects (“Hmm?” as if
response when the child himself or herself clearly pretended to perform the
expecting to be able to really drink). Real drinking could be achieved by
action the experimenter had pretended to perform. The criteria for pretense
opening the bottle at the top. In pretending/trying to drink with Object Set
were that the child showed nonseriousness, playfulness, exaggerated or
F, the experimenter put the straw above the smaller of the two holes that
truncated movements typical of pretense, and appropriate sound effects or
the container had at the top, making sucking movements with some pauses
language and did not care about any real effects of his or her actions (e.g.,
to look at the straw for about 20 s. Pretending and trying were marked in
did not look surprised when the pen left no marks on the paper, did not
a manner analogous to the way they were marked with the bottle. The child
investigate the pen, and did not say that the pen was not working or
could really drink by putting the straw into the bigger of the two holes on
anything similar). An action was coded as a trying response when the child
the container.
either performed the real action himself or herself (e.g., took off the cap of
Objects G and H were used to try/pretend to eat. The experimenter put
the pen and really wrote) or clearly tried to really perform the action. The
both the orange and the nut to his mouth and bit on them, with short pauses
criteria for clear trying were extended examination of the object, obvious
to look at them, for about 15 s. Trying was marked by surprised and
execution of effort, expressions of surprise and frustration when the object
disappointed looks at the object and corresponding sound effects; pretend-
did not work, comments on the object’s malfunctioning (e.g., “does not
ing was marked by a playful expression and eating sound effects
work”), “advises” to the experimenter during or after his performance of
(“Nyum!”). Note that there is one difference between this topic and the
the model (e.g., “You cannot do it this way!” or “Look! I’ll show you how
other three topics: Here the child could not perform the real action by
to do it.”), and pleas for help directed at the experimenter (e.g., “I cannot
himself or herself. Therefore the experimenter closely observed the child’s
do it. Can you help me?”). In terms of our theoretical framework, we
response and interfered if he considered the response to be a clear instance
considered “correct” responses to be trying responses after trying models
of trying (e.g., the child bit on the object with obvious effort or tried to peel
and pretense responses after pretense models. Trying responses after pre-
it or crack it by hand). He then asked the child “Well, what shall we do?”
tense models and pretense responses after trying models were considered
and helped the child to peel the orange or crack the nut if the child
“incorrect.” Responses that fulfilled the criteria neither for a trying re-
requested help.
sponse nor for a pretending response were given the code of the remainder
Each child saw eight action models, a block of four pretense models and
category unclear. In this category there were mainly responses in which the
another block of four thematically matched trying models. For example, a
child copied only some surface behavior of the experimenter or performed
given child would see the experimenter try to write with Object Set A and
some completely different action with the object. A second independent
pretend to write with Object Set B. The order of the two blocks, the
observer coded a random sample of 25% of the sessions for reliability. To
within-block order, and the assignment of the two object sets with the same
test whether the first observer’s codes were influenced by having seen the

---

UNDERSTANDING TRYING AND PRETENDING
391
model action before the child’s response, reliability coding was done from
Arguably, however, this way of analyzing the data poses a very
edited tapes: From the original tapes, only the response period was cut out
strict criterion for successful perception of pretending and trying as
and transferred to a new tape, so that the experimenter’s model action was
such: It requires children to respond significantly more often with
not visible. Interrater reliability was 89%, and Cohen’s kappa was .82.
the correct than with the incorrect response. Against this it can be
argued that such an analysis does not take into account perfor-
Results
mance factors, such as the prepotency of one reaction type—in this
Figure 2 shows for the 2- and 3-year-olds, respectively, the
case, trying responses. In an alternative approach, then, we spe-
numbers of pretense and trying responses as a function of model
cifically looked at children’s pretense responses as a function of
type. In a first statistical analysis, two difference scores were
model type to see whether the 2-year-olds, if not fulfilling the strict
computed for each child: For both model type conditions (pretense
criterion of the first analysis, showed differential pretense re-
and trying), the number of incorrect responses (pretense responses
sponses such that they more often pretended after the pretense
after the trying model and vice versa) was subtracted from the
model than after the trying model. A 2(age)
2(order)
2(model
number of correct responses (pretense responses after the pretense
type) ANOVA on the number of pretense responses yielded sig-
model, trying responses after the trying model); these difference
nificant main effects of model, F(1, 44)
94.10, p
.01 (there
scores could range from
4 to 4. A 2(age)
2(order of model
were more pretense responses after the pretense models), of age,
blocks)
2(model type: pretense vs. trying) mixed-factors anal-
F(1, 44)
20.68, p
.01 (such that the 3-year-olds pretended
ysis of variance (ANOVA) on these difference scores yielded a
more overall), and of order, F(1, 44)
6.76, p
.02 (such that
significant main effect of model, F(1, 44)
4.14, p
.02, such
there were more pretense responses when the pretense models
that the difference score was bigger in the trying model condition
were first). There was also a significant Age
Model interaction
than in the pretense model condition. There were two significant
effect, F(1, 44)
18.14, p
.01, such that the difference in the
interaction effects: a Model
Order effect, F(1, 44)
5.47, p
number of pretense responses between the model conditions was
.03, such that the difference between the scores in the two model
bigger for the 3-year-olds than for the 2-year-olds. Most impor-
conditions was bigger when the trying models were first, and a
tantly, post hoc t tests revealed that both age groups performed
Model
Age effect, F(1, 44)
6.56, p
.02. This Model
Age
significantly more pretense responses after pretense models than
interaction was due to the fact that the 3-year-olds showed signif-
after trying models in both order conditions (all ps
.05,
icantly more correct than incorrect responses after both pretense
one-tailed).
models, t(23)
6.63, p
.01, and trying models, t(23)
7.85,
p
.01. The 2-year-olds, in contrast, showed significantly more
Discussion
correct than incorrect responses only after trying models, t(23)
7.02, p
.01, but showed as many incorrect as correct responses
The 3-year-olds in the present study showed the exact pattern of
after pretense models, t(23)
0, ns. More detailed tests taking
differential imitation that we had expected on the basis of our
order of model blocks into account showed that these results held
theoretical framework: When they saw an adult pretending to
across order conditions: The 3-year-olds performed more correct
perform an action, they then only pretended to perform this action
than incorrect responses after both models, irrespective of order,
themselves and did not care about the real physical effect of their
and the 2-year-olds performed significantly more correct than
performing, whereas when they watched an adult trying to perform
incorrect responses only after trying models, but not after pretense
the same actions, they then really performed the action or tried to
models, irrespective of order (all ps
.01).
really perform it. That is, according to our interpretation, they
perceived the two superficially analogous as-if behaviors in radi-
cally different ways. They perceived pretending to X as such,
under the description of the goal to only act-as-if X. They per-
ceived trying to X as such, under the description of the goal to
really do X.
The 2-year-olds, in contrast, showed this clear pattern only after
trying models. After pretense models they gave, overall, as many
pretense responses as trying responses. That is, under this strict
criterion (more correct than incorrect responses to a given model),
they showed clear signs only of perceiving trying as such, whereas
it remains unclear how they perceived the pretense actions in this
study. One possibility is that they did not really have a clear
conceptual grip of the fact that in pretense, one does not want to
perform the real action. This seems somewhat in conflict with
other findings showing that children at this age competently imi-
tate pretense actions (e.g., Rakoczy et. al., 2002; Watson & Fi-
scher, 1977) and reason about pretense sequences (e.g., Harris &
Kavanaugh, 1993). The other possibility—supported by the find-
ing that the 2-year-olds performed more pretense responses after
pretense models than after trying models—is that the 2-year-olds
did differentially perceive pretending and trying as such but that
Figure 2.
Mean numbers of children’s pretense and trying responses as a
the real actions might have simply been more interesting and
function of age and model type in Study 1 (n
48).
somehow irresistible to them. That is, broadly executive problems

---

392
RAKOCZY, TOMASELLO, AND STRIANO
may explain the 2-year-olds’ bad performance after pretense mod-
dren were differentially reinforced for imitations, there is a serious
els: Although they perceive the model action as pretending, they
possibility that operant conditioning and simple response priming
see that the object can be used to perform the action really, they
could account for the findings. In a second study, we therefore
perceive its function, and they enter into a state of “functional
tested whether these two simpler learning processes—mimicking
fixedness.” They cannot overcome the prepotent action tendency
and priming— could explain the positive findings with the
to then use the object in a functional way themselves. Informal
3-year-olds.
pilot-study observations lend prima facie support to this line of
reasoning. In these we showed 3-year-olds some of the model pairs
Study 2
in the same sort of imitation game, but we used known objects, for
example, a normal pen in the writing topic. Even 3-year-olds in
The same pairs of as-if model actions as in Study 1— one trying
this task hardly showed any pretense responses after pretense
to X, the other one pretending to X—were presented to children.
models. This suggests that when the object is too familiar and its
Two modifications were added to test simpler mimicking and
function too obvious, children neglect the model and the instruc-
priming explanations. First, the warm-up in this study was not
tion to imitate, unable to overcome the tendency to use the object
administered in blocks—as it was in Study 1— but pretense and
in the conventional functional way.2 Relatedly, the 2-year-olds
trying actions were presented in the warm-up in alternating order.
might have simply ignored the experimenter’s instructions more
This was done to rule out simple priming of one response type.
often than the 3-year-olds. Future studies will have to clarify
Second, before performing the model actions, the experimenter
whether the results of the 2-year-olds can be accounted for by
now verbally announced “I am going to X now” (where X was the
these kinds of performance factors or whether a real conceptual
action he then pretended or tried to perform). This was done to
immaturity is responsible.
give children more cues as to what the experimenter was doing and
It might also be objected that our interpretation of the results in
thus to make an explanation in terms of blind mimicking less
terms of differential imitation has created false positives. It might
plausible.
be argued, for example, that simpler kinds of social information
transmission can explain the results, such as mimicking or emu-
Method
lation learning, rather than imitation proper (see Tomasello,
Participants
Kruger, & Ratner, 1993, and Want & Harris, 2002, for overviews
of types of social learning). An emulation explanation, for exam-
Twenty-four young 3-year-olds (34 –38 months, mean age
36 months;
ple, would doubt that children perceived the adult’s models as
16 boys and 8 girls) were included in the final sample. All children were
intentional activities and imitated these under the corresponding
recruited in urban day-care centers. Children came from mixed socioeco-
descriptions; such an explanation would claim instead that the
nomic backgrounds and were all native German speakers. Testing was
children only learned something about the objects and how they
done by one experimenter in a separate quiet room of the children’s
causally work. Two points speak against such an explanation.
day-care center. Three further children were tested but had to be excluded
First, in neither case did the children see that the object really
from the study because they were uncooperative.
worked causally. They did not learn that the pen could be used to
write by seeing how it made marks on a sheet of paper. Second, it
Materials and Design
remains unclear, anyway, what an emulation explanation for chil-
The same materials as in Study 1 were used, and the same experimenter
dren’s pretense responses would look like because in pretense
presented the same eight actions with the materials to the children in two
actions there is no concrete causal effect that the object brings
blocks (four pretense actions and four corresponding trying actions). The
about that could have been learned by observation.
model actions, the systematic variation of the order of the test blocks, and
The more serious concern with the present study is that the
the order of actions within blocks and of the assignment of object pairs to
findings might be accounted for by a simple mimicking explana-
conditions were all exactly as in Study 1. The two theoretically motivated
tion. Such an explanation would claim that children just blindly
differences from Study 1 were as follows: First, the specific warm-up
copied the adult’s surface behavior. Although such an account
blocks before the two test blocks consisted of both pretense and trying
actions in alternating order (see Appendix A for the warm-up actions used).
does not seem very plausible for the trying cases (children often
Second, before producing the model actions in the test blocks, the exper-
made speech acts referring to the malfunctioning of the objects,
imenter verbally announced what he was going to do. For example, in the
indicating that they understood what the goal of the action in
two eating scenarios he said, “I am going to eat something now.” He made
question was), it is a serious possibility for the pretense cases:
the announcements in a neutral voice before trying actions and in a playful
Children rarely performed any speech acts that made clear that
voice before pretense actions.
they were pretending or what they were pretending; rather, this
was mainly coded from their nonverbal behavior. As the experi-
Procedure
menter had not announced what he was going to do, children might
simply not have understood what the action was supposed to be
In the beginning, the experimenter and the child freely played, and the
and so may have mimicked only superficial behavior without
child was introduced to the imitation game as in Study 1. When the child
having a deeper understanding that the experimenter had pretended
felt comfortable and participated in the imitation game, the experimenter
began the first warm-up block, which consisted of three pretense actions
or of what he had pretended. To the eye of the beholder, this
mimicry might then mistakenly have looked like real pretense.
It can also be argued that simple conditioning and priming could
2 However, for an interesting opposing view on the development of
explain parts of the present findings: As there were pretense
functional fixedness, see German and Defeyter (2000), who found that—
warm-up blocks before pretense test trials, and trying warm-up
counterintuitively—functional fixedness arises only later in development,
blocks before trying test trials, and because in the warm-up chil-
at the end of the preschool period.

---

UNDERSTANDING TRYING AND PRETENDING
393
that could not really be performed and three trying actions that could not
different from zero (post hoc analyses of the difference scores as
(or not very naturally) be pretended (see Appendix A). In the warm-up
a function of model type and order revealed that children produced
blocks, the experimenter did not explicitly reinforce the child but laughed
significantly more correct than incorrect responses after trying
and reacted amusedly when the child pretended and reacted with surprise
models irrespective of the order of the model blocks, and signifi-
and appreciation when the child really performed an action the experi-
cantly more correct than incorrect responses after pretense models
menter had unsuccessfully tried to perform (“Wow! You did it.”). Then
when pretense models were first; all ps
.05). Regarding chil-
came the first test block of four model actions (pretense for half the
dren’s responses to pretense models, when trying models were
children, trying for the other half), followed by another warm-up block of
three pretense and three trying actions. Finally, the second test block of
first, there was only a trend for the difference score ( p
.16,
four model actions (trying, if pretending had been first and vice versa) was
one-tailed). That is, after both pretense and trying models, children
administered. Each session lasted approximately 15 min.
produced significantly more correct than incorrect responses.
Overall, the results from the 3-year-olds in Study 1 were
Observational and Coding Procedure
replicated.
All sessions were videotaped and edited afterward: Only the child’s
reaction periods were selected and transferred to a new tape. A single
Discussion
observer then coded the child’s reactions, unaware of what the experi-
menter had demonstrated. Each response of a child to the eight test models
In this study, we presented 3-year-olds with the same model
was coded. The same coding scheme as in Study 1 was used. A second
actions as in Study 1 but with a better-controlled warm-up proce-
independent observer coded a random sample of 25% of the sessions for
dure. The findings from Study 1—that 3-year-olds show differen-
reliability. Interrater reliability was 91.7%, and Cohen’s kappa was .86.
tial and systematic imitation of pretending and analogous trying
actions—were replicated, and simple explanations in terms of
Results
priming or conditioning can be ruled out. Another concern with
Figure 3 shows the pretense and trying responses as a function
Study 1 was that children might have just mimicked the experi-
of model type. For the statistical analysis, the same two difference
menter’s surface behavior, especially his pretending, without un-
scores calculated in the main analysis of Study 1 were computed:
derstanding its meaning. Therefore, in Study 2 we had the exper-
For each child, for both model type conditions (pretense and
imenter verbally announce “I am going to X now” (where X was
trying), the number of incorrect responses (pretense responses after
the action he then pretended or tried to perform) in order to convey
the trying model and vice versa) was subtracted from the number
to children independent information about which action the exper-
of correct responses (pretense responses after the pretense model,
imenter pretended or tried to perform. Three-year-olds surely do
trying responses after the trying model), which yielded a score
understand what “I am going to eat/drink/pour/write” means, and
ranging from – 4 to 4. A 2(model type: pretense vs. trying)
the fact that they showed the same systematic pattern of differen-
2(order of model blocks) mixed-factors ANOVA on the difference
tial imitation as in Study 1 makes it seem less plausible that the
scores yielded no significant main effects but a significant Model
findings in the two studies can be accounted for by a simple
Type
Order interaction effect, F(1, 22)
10.14, p
.01, such
mimicking explanation. However, against this plausibility argu-
that the difference between the two model conditions was bigger
ment remains the concern that children might have understood the
when trying was first.
announcement in the first place but then ignored it in interpreting
Dependent sample t tests revealed that the difference scores both
the behavior or just not have understood the relation between the
in the pretense model condition, t(23)
3.89, p
.01, and in the
announcement and the following behavior. That is, with regard to
trying model condition, t(23)
5.55, p
.01, were significantly
the mimicking concern, the present results remain less than
conclusive.
To rule out simple mimicking explanations more stringently, in
a third study we therefore extended the methodology from Studies
1 and 2 and tested for more productive reactions than imitation as
indicators of children’s understanding of pretending and trying as
different forms of behaving-as-if.
Study 3
Understanding pretense, as all understanding, is inferentially
integrated and structured: When I see you pretending that the
telephone is a banana, I understand that it “follows” within the
pretense scenario that it has to be peeled first (see Harris &
Kavanaugh, 1993; Leslie, 1987, 1988). In a similar way, under-
standing trying essentially involves the appreciation of some in-
ferences: When I see you trying to turn on the light by flipping
Switch A, which does not work, and I know that Switch B works,
then when I want to do what you tried to do, I understand that it
follows that I should use Switch B. However, the inferential
Figure 3.
Mean numbers of children’s pretense and trying responses as a
understanding required for pretense comprehension is even more
function of model type in Study 2 (n
24).
complex than that required in interpreting trying: Pretense infer-

---

394
RAKOCZY, TOMASELLO, AND STRIANO
ences have to be kept apart from inferences about reality (the
In the warm-up, several actions and props were introduced to give
telephone cannot be peeled).
children the opportunity to show creative pretense and trying responses in
Several studies have shown that 2-year-old children show some
the test phase. Three of these props were placed on a piece of cardboard
proficiency at drawing counterfactual pretense inferences (Harris
and stayed on the table throughout the session, reachable by the child: a
teddy bear (which one could pretend to feed or give a drink to), a bowl with
& Kavanaugh, 1993; Walker-Andrews & Harris, 1993; see
a toy fork (which could be used to pretend to eat with or to pretend to feed
Kavanaugh & Harris, 1999, for a review). For example, 2-year-
the teddy bear with and, alternatively, to open or cut objects), and a wrench
olds presented with an adult who pretended to pour into two cups
(which could be used to open containers and cut things; see Figure 4). As
and then pretended to drink from one cup could correctly answer
in Study 2, the pretending and trying warm-up actions were not done in
the question of which cup was “empty” and which was “full”
blocks in order to rule out simple priming of one action type.
(Walker-Andrews & Harris, 1993).
Each child saw four action models, a block of two pretense models and
On the basis of these studies, the rationale for the present study
another block of two thematically matched trying models. For example, a
was as follows: We presented children with model action pairs
given child would see the experimenter try to eat with Object G and
from the previous studies but did not instruct them to strictly
pretend to eat with Object H. The order of the two blocks, the within-block
imitate; in addition, we supplied them with a richer warm-up and
order, and the assignment of the two object sets with the same topic to the
pretending versus trying conditions were all systematically varied across
more props to act on than the objects involved in the model
children. The order of the topics within both blocks was the same (e.g.,
actions. The purpose of these modifications was to supply children
when a child had “eating” first in his or her pretense block, he or she also
with the opportunity to react more productively and unambigu-
had “eating” first in his or her trying block).
ously. Our prediction was that if children perceive pretending to X
as such, that is, as intentionally acting only as if X, and if they have
Procedure
the opportunity to show more productive reactions, then they
should often perform a pretense action in response that “follows”
In the beginning, the experimenter and the child played freely, and the
from the model action. When they see the experimenter pretend to
experimenter performed some simple actions (e.g., building a tower with
pour from a container (which actually contains water) into a cup,
building blocks). He then performed two pretense actions (pretending to
for example, they often should—when handed the container and
make a phone call and pretending to dig a hole) with a novel object (a
doorstopper) and one trying action (trying to make music with a small toy
the cup—pretend to drink from the cup themselves. In contrast, if
piano by pushing a wrong button) and asked the child to do the same after
they perceive trying to X as such, that is, as intentionally trying to
each action. The experimenter reacted with laughter and amusement to
X really, and are given the opportunity to react more productively,
children’s pretense and with surprise and appreciation when children
they should often creatively try to X, going beyond the experi-
performed successfully with the piano.
menter’s behavior. When they see the experimenter try to pour
Then came the warm-up period in which the three additional props were
from a container with water in it that does not come out, for
introduced. First, the experimenter put the teddy bear on a piece of
example, and when there is a tool available they know can be used
cardboard on the table, saying “Look! This is Teddy. I am going to show
to open containers, they often should—when handed the container
you some things I can do, and then it’s your turn and we can do something
and the cup—make use of the tool to open the container first and
with Teddy as well.” The experimenter demonstrated three pretense actions
then pour.
(pretending to brush his teeth, pretending to take a shower, and pretending
to drink) with novel objects and handed the objects to the child. Pretense
imitations by the child and the corresponding pretense actions with Teddy
Method
(brushing Teddy’s teeth, giving Teddy a shower, giving Teddy a drink)
were reinforced by the experimenter. If the child did not spontaneously
Participants
perform the pretense actions with Teddy, the experimenter pushed the
cardboard with Teddy toward the child, saying “And now?” If the child did
Eighteen young 3-year-olds (34 –38 months, mean age
36 months; 8
not react, the experimenter finally explicitly asked the child to perform the
boys and 10 girls) were included in the final sample. Children were
pretense action with Teddy.
recruited in urban day-care centers (n
7) or by telephone from a list of
Second, the experimenter brought out a novel container, announced “I
parents and children who had volunteered for studies of child development
am going to open it,” and tried unsuccessfully to open it. He then said
(n
11). Children came from mixed socioeconomic backgrounds and were
“Hmm. It does not work. Ah, then we can take the wrench” and brought out
all native German speakers. Testing was done by one experimenter in a
a toy wrench, saying “With the wrench we can open things” and opened the
separate quiet room of the children’s day-care center or in a child psychol-
container. The child was then given the wrench (“Now you may do it as
ogy laboratory, and sessions were videotaped for subsequent analysis. Two
well”), and the experimenter helped the child to open the container with the
additional children were tested but had to be excluded from the study, 1
wrench if necessary. The wrench was placed on the cardboard, beside
because of experimental error and 1 because he was uncooperative.
Teddy, and the experimenter brought out another novel container and
unsuccessfully tried to open it. The child was then given the container (“It’s
your turn”). If the child spontaneously used the wrench to try to open the
Materials and Design
container, the experimenter reacted with appreciation (“Ah! That’s how we
can do it!”) and helped the child if necessary. If the child tried to open the
Children were presented with two model action pairs from the previous
container without the wrench, the experimenter pushed the cardboard
studies: pretending/trying to eat and pretending/trying to pour. The same
toward the child, saying “What do we do then?” If the child still did not
objects used in Studies 1 and 2 were used with one exception: The
take the wrench, the experimenter said “Hmm. Or we could use the
container from Object Set D was used, but the experimenter now pre-
wrench!” To introduce the wrench as a multifunctional tool, the experi-
tended/tried to pour into a glass instead of onto a flower.3 We used only
two model action pairs in this study in order to avoid memory overload and
a too lengthy procedure (each action pair needed the introduction of
3 The reason for this modification was that pretending to pour into a cup
additional props for possible corresponding inferential reactions in the
lends itself better to inferential reactions—above all, pretending to drink—
warm-up).
than does pretending to pour onto a flower.

---

UNDERSTANDING TRYING AND PRETENDING
395
than in the previous studies was used. There were five categories into
which a given reaction could be classified (for each test model, a given
child could show several responses in succession, and these responses
could be from different categories): A reaction was coded as inferential
pretense when the child performed a thematically appropriate pretense
action that went beyond what the experimenter did in his pretense model
actions. The criterion for this code was that the child revealed by an action
or by a nonserious speech act an appreciation of the pretense scenario in
question. For the eating topic, these were speech acts such as “I have eaten
it up” or “Mmm, delicious” (after the child had obviously not really eaten
but just put the orange or the nut in front of his or her mouth and made
chewing movements) and actions such as “feeding” Teddy (putting the
orange or the nut to Teddy’s mouth and making chewing movements and
corresponding sound effects) or putting the orange or nut into the bowl,
moving the fork to the bowl and then in front of the mouth, and making
chewing movements and appropriate sound effects. For the drinking topic,
these were, above all, speech acts such as “Hmm! Some tea in there!” and
actions such as pretending to drink and pretending to give Teddy a drink.
Figure 4.
Additional props used in Study 3.
A reaction was coded as simple pretense when the child performed what
looked like a clear pretense action (with playfulness, exaggerated move-
ments, appropriate sound effects, etc.) but without any elements that went
menter in addition showed the child that the wrench could be used to cut
beyond what the experimenter did in his corresponding model action (i.e.,
things, cutting apart a piece of play dough, and the child was given the
actions that were coded in this study as simple pretense would have been
chance to do the same. (Actually, the three actions with Teddy and the
in the category pretense response in Studies 1 and 2). When the child first
three actions with the wrench were not done in blocks, but—to avoid
performed a merely simple pretense action and then immediately went on
simple priming of one action type—were temporally interspersed: two
to perform pretense that went beyond the experimenter’s model (e.g., first
actions with Teddy, then two actions with the wrench, then one action with
pretended to pour and then pretended to drink; first pretended to eat and
Teddy, then one action with the wrench. See Appendix B for details.)
then pretended to feed Teddy) in a thematically appropriate way, this whole
Finally, the bowl with the fork was introduced. The experimenter
episode was given the code inferential pretense.
brought out a cardboard box and a replica carrot, said “Let’s cook some-
Analogously, responses were coded as inferential trying when the child
thing,” and pretended to cook the carrot in the cardboard box. He then
revealed by an action or by a speech act an appreciation that his or her goal
pretended “Now it’s cooked,” brought out the bowl with the fork, put the
carrot in the bowl, and gave it to the child. If the child did not spontane-
was to perform the action in question really and successfully. This code
ously pretend to feed Teddy, the experimenter pushed the cardboard with
was given when the child really performed the action successfully, or tried
Teddy and the wrench on it toward the child and asked “And now?” If the
to perform it with means other than that used by the experimenter in his
child did not pretend to feed Teddy, the experimenter explicitly asked,
corresponding model action (above all, by using the wrench to open
“What can you do with Teddy?” If that did not help, the experimenter
containers, crack the nut, or peel the orange), or said something that made
finally said, “Can you give Teddy something to eat?” The experimenter
reference to the goal (e.g., “We have to open it first”) or the malfunctioning
reinforced appropriate pretense actions (pretending to eat, pretending to
of the objects (e.g., “It does not work,” “It’s broken”). Responses were
feed Teddy). (The same procedure was repeated with a replica sausage that
coded as simple trying when the child performed what looked like a clear
the experimenter pretended to cook.) The bowl with the fork was then
instance of trying behavior (extended examination of the object, obvious
placed on the cardboard, and the cardboard stayed on the table throughout
execution of effort, expression of frustration, etc.) but without any elements
the rest of the session, reachable by the child (approximately 40 cm in front
that went beyond what the experimenter had done in his corresponding
of the child).
model action. When a child first performed a merely simple trying behavior
Then came the first test block, which consisted of two model actions
and then immediately went on to try with different means (e.g., first trying
(pretense/trying for half of the children each). The experimenter presented
to eat by biting on the nut and then taking the wrench to crack it; trying to
the actions in exactly the same way as in Study 2, with a verbal announce-
pour by shaking the container and then making use of the wrench to open
ment “I am going to . . . [action] now” before the action.4
it), the whole episode was coded as inferential trying. Finally, there was a
After performing the model action, the experimenter gave the object(s)
remainder category, unclear, for responses that fulfilled none of the criteria
to the child. He did not reinforce specific responses but reacted equally
for the four categories mentioned above. In this category were mainly
positively to all responses. If the child did not spontaneously make use of
responses in which the child performed some different action with the
one of the props, the experimenter pushed the cardboard with the three
object, or did nothing at all, or performed an ambiguous action (e.g.,
props on it a bit closer to the child.
The first test block was followed by another warm-up block consisting
of two pretense and two trying actions in alternating order (see Appendix
4 There were only two small modifications: First, before pretending/
B). Finally came the second test block, which consisted of two model
trying to pour with the novel container from Object Set D (now into a glass
actions (trying for the children that had seen pretense first and vice versa).
instead of onto a flower), the experimenter said “I am going to pour
The whole session lasted approximately 15 min.
something now” instead of “I am going to water the flower now.” Second,
after the experimenter had performed the pouring actions, whereas in Study
Observational and Coding Procedure
2 he had put the container and the cup or flower equidistant in front of the
child, in this study, he put the glass or cup close in front of the child with
All sessions were videotaped and edited afterward: Only the child’s
one hand and the container at a slightly further distance in front of the child
reaction periods were selected and transferred to a new tape. A single
with the other hand. The reason for this modification was that it made the
observer then coded the child’s reactions, unaware of what the experi-
sequence more naturally interpretable as part of an extensible scenario—“I
menter had demonstrated. Each response of a child to the four test models
pour, you drink” in the pretense case and “Look! Nothing in there, it did
was coded. Importantly, a different and more differentiated coding scheme
not work!” in the trying case.

---

396
RAKOCZY, TOMASELLO, AND STRIANO
bringing the orange to the mouth but without signs of either pretending or
differential responses to thematically matched pretending and try-
trying to eat).
ing models that they perceived these two kinds of as-if behaviors
A second independent observer coded a random sample of 25% of the
under different intentional descriptions. The present study, how-
trials for reliability. Interrater reliability was 95%, and Cohen’s kappa was
ever, extended the previous findings in important ways: Not only
.91.
did young children differentially imitate these different kinds of
model actions, they also showed systematic productive and infer-
Results
ential responses that respected the logical structure of the actions
Only the children’s first responses to a given model action were
they had seen. After pretense model actions, they performed pre-
entered into the analysis, because only the first reaction was
tense actions that fit the stipulated pretense scenario (e.g., they
considered an indicator of how children understood the action they
pretended to drink from a cup into which the experimenter had
had seen the experimenter perform. Figure 5 shows the mean
pretended to pour). After trying model actions, they tried to per-
number of simple and inferential trying and pretending responses
form the action the experimenter had tried to perform but made use
(that children gave as a first response) as a function of model
of different means and verbally commented about their goal and
action. As the main purpose in this study was to test more strin-
the obstacles to its achievement (e.g., after the experimenter had
gently for children’s understanding of pretending and trying and to
tried to pour, they made use of the wrench to open the container
rule out simple mimicking as an alternative explanation, children’s
first and then to pour, saying something like “We have to open it
inferential pretending and trying responses were the focus of the
first”). These findings also rule out a simpler explanation in terms
analysis. After pretense models, only inferential pretending re-
of superficial mimicking that had been a concern in Studies 1 and
sponses were considered correct, and only inferential trying re-
2. Whereas in the previous studies, it could have been objected that
sponses were counted as incorrect; after trying models, the anal-
children’s responses, especially to pretense model actions, only
ogous situation held. As in Studies 1 and 2, for each child, for the
looked like insightful imitation but were in fact dumb mimicking,
two model action conditions, a difference score (correct minus
this objection does not hold for the present study: In their re-
incorrect responses) was computed, which ranged from
2 to 2. A
sponses, children went beyond what the experimenter had done in
2(model action: pretense vs. trying)
2(order of model blocks)
a systematic way, indicating that they drew thematically appropri-
ANOVA on these difference scores yielded no significant effects.
ate inferences. Generally, parsimonious explanations of the present
Next, these difference scores were tested against zero. For both
data in terms of mimicking, simple conditioning, or some such
model type conditions, the difference scores were significantly
superficial mechanism seem implausible given the systematic and
bigger than zero: pretense models, t(17)
1.88, p
.04; trying
productive inferential responses the children gave.
models, t(17)
6.23, p
.01 (one-tailed). That is, after both
pretending and trying models, 3-year-olds produced significantly
General Discussion
more correct inferential responses than incorrect inferential
responses.
In the current studies, we tested young children for their under-
standing of different forms of as-if behaviors, specifically, pre-
tending to perform an action and trying to perform the same action.
Discussion
This presented an interesting case with which to test standard
The results of Study 3 replicated the findings from the previous
behaving-as-if theories against our revised acting-as-if construal.
two studies: 3-year-old children show by their systematic and
Standard behaving-as-if theories claim that young children’s con-
cept of “pretending-that-p” is one of “behaving in a way that
would be appropriate if p were the case” (Nichols & Stich, 2000,
p. 139).