Content uploaded by Luca Bonatti
Author content
All content in this area was uploaded by Luca Bonatti
Content may be subject to copyright.
Psychological Science
22(5) 674 –681
© The Author(s) 2011
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0956797611404086
http://pss.sagepub.com
Performing actions on moving objects typically requires a high
level of accuracy. The fact that people can perform tasks such as
hitting or catching a ball shows that they can accurately repre-
sent the timing of a visible moving object and anticipate its
future positions (Regan, 1992). Low-level visual mechanisms
are available to track single and even multiple object displace-
ments (Pylyshyn, 2001, 2003; Scholl & Pylyshyn, 1999; Scholl,
Pylyshyn, & Feldman, 2001). However, in natural scenes, mov-
ing objects are not always visible. Thus, observers must often
rely on mental representations of nonperceived dynamic events.
In studies of mental imagery, it is generally assumed that the
mind builds analog representations that are isomorphic with
external events and allow the observer to “run” simulations of
unseen situations (e.g., Johnson-Laird, 1983; Kosslyn, 1994);
it is also assumed that these simulations even recruit the
same neural tissue involved in the direct perception of visible
events (Borst & Kosslyn, 2008; Klein et al., 2004; Kosslyn,
Thompson, Kim, & Alpert, 1995). The ability to simulate events
has been considered a hallmark of thinking; as the philosopher
Colin McGinn (1989) wrote, “A thinking system, we might say,
is a simulation engine—a device that mimics, copies, replicates,
duplicates, imitates, parallels reality” ( p. 176). This approach
naturally invites the supposition that, in continuity with mecha-
nisms for the perception of visible displacements, analog simu-
lations provide the means to represent invisible dynamic events
(Schwartz, 1999; Shepard & Cooper, 1986).
However, knowledge about dynamic events goes beyond
the simple representation of invisible positions. Adults and
infants can perceive high-level properties of dynamic scenes,
such as causal relations (Leslie, 1982; Michotte, 1963) or
agency status (Gergely, Nádasdy, Csibra, & Bíró, 1995). Thus,
a plausible (although not necessary) view of the relation
between perception and dynamic imagination is that informa-
tion about the dynamic properties of a scene (e.g., their causal
relations or the physical forces acting on objects) contribute to
create a unique mental representation that simulates future
states of invisible events, allowing the observer to predict
them. Indeed, it has been claimed that the internalization of
Corresponding Author:
Luca L. Bonatti, ICREA and Universitat Pompeu Fabra, C. Roc Boronat, 138,
Edifici Tanger, 55.110, 08018 Barcelona, Spain
E-mail: lucabonatti@mac.com
A Dissociation Between Judged
Causality and Imagined Locations in
Simple Dynamic Scenes
Florent Levillain1,2 and Luca L. Bonatti3
1Department of Psychological and Brain Sciences, Johns Hopkins University; 2Laboratoire “Cognitions Humaine et Artificielle,”
Université Paris 8; and 3ICREA and Universitat Pompeu Fabra
Abstract
To mentally extrapolate the trajectory of a moving object that disappears from sight, different sources of information can be
exploited: memory of its last visible position, its inferred movement through time, and general understanding of the causal
structure of the scene. It is often assumed that these cues are integrated into unified analog mental representations. In our
experiment, participants predicted the position of an object that disappeared behind an occluder and estimated the degree
to which the movement was caused by another object. They made considerable errors in predicting imagined displacements.
Moreover, their predictions were misaligned with their judgments of causality. They predicted the positions of the invisible
moving objects better in events that they judged less causally correct than in events that they judged more causally correct.
These results suggest that physical and cognitive parameters of imagined dynamic events do not merge into unitary mental
models simulating actual states of the world.
Keywords
mental imagery, dynamic imagination, motion prediction, perception of causality
Received 7/9/10; Revision accepted 12/13/10
Research Article
Causality and Prediction 675
invariant properties of the environment is evolutionarily adap-
tive (Hubbard, 1995; Shepard, 2002).
In the experiment reported here, we tested both observers’
accuracy in estimating displacement of invisible objects and
the degree to which high-level properties of a dynamic scene,
such as its degree of causal connectedness, are integrated into
a unified mental representation. Participants estimated the
timing of successive invisible displacements within dynamic
scenes and judged the causal correctness of the scenes. We
reasoned that if a unified mental representation grounds
motion prediction by mental simulation, participants who
viewed a scene that afforded reliable information about the
dynamic properties of occluded objects would (a) estimate the
objects’ positions reasonably accurately and (b) make more
accurate predictions for causally correct than for causally
anomalous scenes. Alternatively, errors in estimating imag-
ined positions, and possibly a dissociation between accuracy
of on-line predictions and judgments of causal correctness,
would indicate that there was no common representation of
dynamic events.
Method
Participants
Nineteen participants completed the experiment (mean age =
24.4 years, range = 20–31 years). Ten (naive group) had
learned physics only in middle or high school, 4 (intermediate
group) had taken physics in college, and 5 (professional group)
were physicists.
Stimuli and apparatus
We created realistic movies in which one ball (the “launcher”)
moved toward a second ball at the center of a scene (the “tar-
get”). One second after the beginning of each movie, the
launcher appeared from one side of the scene and traveled
horizontally toward the target at a constant speed of 25.8°/s,
19.3°/s, or 12.9°/s (see Videos S1–S5 in the Supplemental
Material available online). The target began moving at the
same speed and in the same direction at the moment the
launcher stopped moving or after a delay (see the descriptions
of the causality conditions later in this section).
Targets’ trajectories were either occluded or visible. In the
target-occluded movies, after initiating its movement, the tar-
get continued its trajectory behind an occluder, so that its
actual position could only be imagined. Three vertical lines
were drawn on the occluder, in one of two different configura-
tions (Table 1). Movement direction (rightward/leftward) was
counterbalanced across trials. The target-visible movies were
identical to the target-occluded movies except that the position
of the occluder was changed so that the target was always vis-
ible. The bars remained in the same positions as in the target-
occluded movies (i.e., they were not moved with the occluder),
so that participants would estimate the same movement dis-
tances in the target-visible and target-occluded conditions and
accuracy in these estimates would therefore be directly com-
parable between the conditions. Figure 1 shows examples of
the two conditions, illustrating each bar configuration.
The causality of the scenes varied. In the contact condition,
the motion of the launcher ceased immediately after the
launcher hit the target, which started to move immediately
after contact. In the delay condition, the launcher contacted the
target and stopped moving, but the target began moving after
an interval of either 480 or 640 ms. In the space condition, the
launcher stopped exactly when the target started moving, but it
stopped before any contact, at a distance of either 100 or 130
pixels from the target.1
Finally, for both the target-visible and the target-occluded
conditions, we created 20 distractor movies in which the
launcher fell from above, landing in the same positions as in
the experimental sequences. Because the movements of the
targets and launchers were orthogonal in the distractor movies,
they broke up the sequence of horizontal movements created
by the experimental movies. Thus, there were 160 animations
in total: 120 experimental animations, crossing five causality
conditions (one contact condition, two delay conditions,
two space conditions) with two target-visibility conditions
Table 1. Locations of the Vertical Bars and Hypothetical Times at Which the Targets Would
Arrive at the Bars as a Function of Launcher Speed
Hypothetical arrival time (s)
Bar
Distance
from origin
Launcher speed:
12.9°/s
Launcher speed:
19.3°/s
Launcher speed:
25.8°/s
Configuration 1
1 44.2° 0.44 0.28 0.24
4 60.8° 1.72 1.16 0.88
6 71.8° 2.60 1.72 1.32
Configuration 2
2 49.7° 0.88 0.56 0.44
3 55.3° 1.28 0.84 0.68
5 66.3° 2.16 1.40 1.12
676 Levillain, Bonatti
(visible, occluded), two bar configurations, three speeds, and
two direction of movement, plus 40 distractor trials. At the end
of each movie, a graded scale appeared so that participants
could make causal judgments.
The experiment ran on a PowerMac G4, programmed with
PsyScope X (for more information on this open-source soft-
ware, see http://psy.cns.sissa.it). The animations were projected
on a 200-cm × 135-cm screen by an Epson EMP 8100 projector
placed behind the participants. The scenes filled the entire visual
space, covering a visual angle of 70°. Response times (i.e., esti-
mates of when the target reached the vertical bars) were col-
lected with a button box that was placed together with a
numerical keypad on a table in front of the participants.
Procedure
Participants sat in a dimmed room, 2.5 m from the screen.
Each session began with an example of the contact condition
as a practice trial. Participants were instructed to visually track
the launcher and press the button on the button box when they
felt that the target reached the position corresponding to each
bar. They were told to press the button only three times (once
for each bar), and to do so both if the target remained visible
and if it disappeared behind the occluder. They were also
informed that the balls moved at identical and constant veloci-
ties, so that the speed of the launcher was entirely predictive of
the speed of the target.
Participants were also instructed to indicate the perceived
strength of the causal relation between the launcher and the
target by entering a number on the keypad when the scale
appeared on the screen (0 = not at all causal, 9 = completely
causal). No explicit connection was drawn between predicting
the target’s position and making the causality judgment. No
feedback was given during the experiment.
Participants initiated each trial by pressing a button. The
movies were presented in two blocks of 80, arranged in a pseu-
dorandom order, with the constraint that the same causality
condition could not occur more than three times in a row. The
target-occluded movies were presented in the first block, and
the target-visible movies in the second block. The experiment
lasted approximately 1 hr.
Results
A trial was excluded from analysis (0.23% of the data) if the
participant did not indicate the target’s position exactly three
times, if the participant pressed the button before the disap-
pearance of the target, or if a response time was more than 2.5
standard deviations from the mean response time for that
trial’s condition. The two variables of interest were mean tim-
ing error (MTE) and causality judgments. MTE was calculated
as the difference between the response time indicating when
the target was estimated to reach a bar and the actual time
when the target crossed that bar. A positive value indicates that
the response was entered after the target crossed the bar, and a
negative value indicates that the response was given before the
actual arrival time. The frame in which an occluded target
reached each bar was determined off-line as the first frame in
which the target made contact with the bar in the 3–D model
generating the movie. All analyses we report are two-way
repeated measures ANOVAs, and all post hoc tests are alpha-
adjusted with Bonferroni correction.
As movement direction had no main effect and did not
interact with any other independent variable, we collapsed
across this factor. We also collapsed across the two temporal
and spatial intervals of the delay and space conditions, respec-
tively, as restricted analyses showed that they had no influence
on MTE. Figure 2 shows the average MTE at each bar posi-
tion, for each combination of causality and target-visibility
conditions.
We analyzed the difference in individual MTE means
between the target-visible and target-occluded conditions by
means of an ANOVA with target occlusion (visible, occluded),
Fig. 1. Overall structure of the experimental movies. In the target-occluded
condition, the movement of the target (the green ball) could only be imagined;
the example shown here (a) represents the launcher (the white ball) in its
initial position. In the target-visible condition, the movement of the target was
always visible; the example shown here (b) represents the stopping position
of the launcher in the space condition with the larger spatial interval. These
examples illustrate the two different configurations of the vertical bars.
Causality and Prediction 677
speed (25.8°/s, 19.3°/s, 12.9°), bar (1, 2, 3, 4, 5, 6), and scene
causality (contact, delay, space) as independent variables.
Occlusion had a strong effect, F(1, 18) = 100, p < .0001, prep =
.99, ηp
2 = .77. When the target was visible, participants cor-
rectly determined the exact moment of arrival at each position.
This result not only shows that they could accurately predict
contact points when direct visual feedback was available, but
also shows that the task of tracking successive positions with
the spatiotemporal parameters used was totally feasible. How-
ever, in the target-occluded condition, when participants had to
imagine the target’s position, they made important systematic
errors.
Post hoc tests revealed that in the target-visible condition,
participants were as precise in monitoring contact with the last
bar as in monitoring contact with the first bar; error did not
increase with traveled distance. However, the pattern of results
in the target-occluded condition was markedly different. First,
participants overestimated the time to contact at each tested
position, which indicates that overestimation error began right
when visual feedback for the target position was terminated.
Second, although overestimation increased during occlusion,
the increase was not continuous, as a simulation model would
predict. Indeed, in the target-occluded condition (but not in the
target-visible condition), participants could not differentiate
between the close positions tested in the different bar configu-
rations (MTEBar2 – MTEBar1 = 0.85 ms, p = 1; MTEBar4 –
MTEBar3 = 72.11 ms, p = .4; MTEBar6 – MTEBar5 = 61 ms, p =
0.8). That is, error increased between responses, but not
between equally spaced positions. This result suggests that
participants had at best a rather limited ability to predict the
position of an invisible target.
Speed had no effect on error, nor did speed and target
visibility have an interactive effect on error. However, the
interaction of speed and causality condition was significant,
F(4, 18) = 5.1, p ≤ .0004. Also, the significant three-way inter-
action of speed, causality condition, and target visibility,
F(4, 18) = 4.5, p ≤ .0012, showed that speed had an effect only
when causality was violated in imagination.
To further clarify how the experimental manipulations
affected participants’ imagination of dynamic scenes, we ran
restricted analyses on the target-occluded condition, collapsing
across the two bar configurations, which did not differentially
affect MTE. An ANOVA with response order (Response 1, or
R1; Response 2, or R2; Response 3, or R3), speed, and causality
condition as independent variables showed a strong effect of
response order, F(2, 18) = 99.13, p ≤ .0001, prep = .99, ηp
2 = .85.
The overestimation increased at each successive response
(MTER2 – MTER1 = 621 ms, p ≤ .0001; MTER3 – MTER2 = 698
ms, p ≤ .0001). The causal nature of the movies was also rele-
vant, F(2, 18) = 91.9, p ≤ .0001, prep = .99, ηp
2 = .99. Post hoc
analyses showed that participants were less accurate in the delay
condition than in contact condition (MTEdelay – MTEcontact =
325 ms, p ≤ .0001), and less accurate in the delay condition than
in the space condition (MTEdelay – MTEspace = 284 ms, p ≤
.0001). However, they were equally (in)accurate in the contact
and space conditions (MTEspace – MTEcontact = 31 ms, p = .3).
1
0
400
800
1,200
Error (ms)
1,600
2,000
Contact
Space
Delay
Occluded Target
Visible Target
3
Bar Position
Causality Judgment
4 5 6
0
1
2
3
4
5
6
7
8
9
2
Fig. 2. Mean timing errors (left panel) and causal-strength judgments (right panel). Timing error is plotted as a function of bar
position, separately for the contact, space, and delay conditions, both when the target was visible and when it was occluded.
Causality judgments were made on a scale from 0 (not at all causal) to 9 (completely causal). Error bars represent 95% within-
participants confidence intervals (Loftus & Masson, 1994) in the respective conditions.
678 Levillain, Bonatti
Speed had no main effect, but it interacted with causality condi-
tion, F(4, 18) = 9.4, p ≤ .0001. Post hoc analyses showed that
the interaction was due to the fact that MTEs in the space and
contact conditions were not distinguishable when the target and
launcher moved at speeds of 12.9°/s and 19.3°/s, but at speeds
of 25.8°/s, scenes with spatial violations generated higher errors
than causally correct scenes, p ≤ .0001. In contrast, scenes with
temporal violations generated a much higher error than scenes
with spatial violations or causally correct scenes at all three
speeds (all contrasts, p ≤ .00001) Thus, in all tested conditions,
a violation of the temporal relation of causality resulted in an
error that was greater than the error observed when there was a
spatial violation, or no violation at all.
The causal nature of the scenes also influenced participants’
explicit judgments of causality (see Fig. 2), as revealed by an
ANOVA with causal judgment as the dependent variable and
causal condition as the independent variable, F(2, 18) = 69.5,
p ≤ .0001, prep = .99, ηp
2 = .79. As expected, the contact condi-
tion was judged more causally correct (M = 8.25, SD = 0.71)
than the delay condition (M = 3.99, SD = 2.86) and the space
condition (M = 1.83, SD = 2.35). Causality judgments differed
significantly between all pairs of conditions (contact vs. space:
p ≤ .0001; contact vs. delay: p ≤ .0001; delay vs. space: p ≤
.002), but the difference was greatest between the contact and
space conditions: Whereas causality was considered strongest
in the contact condition, it was considered null in the space con-
dition. It is notable that participants rated events involving tem-
poral violations as more causal than those involving spatial
violations. Thus, paradoxically, in the implicit task of predicting
the position of an invisible target, participants performed better
in the space condition than in the delay condition, even though
they explicitly judged causality to be weaker in the space condi-
tion than in the delay condition. That is, prediction abilities and
assessment of causality did not align.
Participants had to base their predictions about the position
of the invisible target on the velocity of the visible launcher.
Thus, it could be argued that although participants simulated the
movement of the target, the representational momentum of the
launcher, which has been argued to affect responses to Michotte-
like scenes (Hubbard, Blessum, & Ruppel, 2001), influenced
these simulations. Specifically, because the target in the space
condition began to move exactly when the launcher stopped, the
target could have “inherited” the representational momentum of
the launcher. In contrast, in the delay condition, the launcher
stopped before the target started moving, and thus there was
much less opportunity for the target’s speed to be simulated via
representational momentum.2 We were able to test this alterna-
tive explanation by exploiting the distractor trials. Because the
target moved orthogonally to the launcher in these trials, there
could be no transfer of representational momentum. Hence,
according to this alternative explanation, causality condition
should have had no effect on MTE in the distractor trials. In
other words, the data should show an interaction between trajec-
tory (orthogonal in distractor trials, identical in experimental
trials) and causality condition.
To test this alternative explanation, we first checked that dis-
tractors engaged participants in processing the causal nature of
the scenes. An ANOVA with causality judgment as the depen-
dent variable and causality condition as the independent vari-
able showed that they did, F(2, 18) = 38.7, p < .001. As we
found for the experimental scenes, participants judged the
events in the space and delay conditions as less causal than the
events in the contact condition (Mspace = 2.18, SD = 2.8; Mdelay =
3.51, SD = 2.4; Mcontact = 7.33, SD = 1.2). We then ran an
ANOVA with MTE in the target-occluded condition as the
dependent variable and trajectory (orthogonal, identical), cau-
sality condition, and response order as independent variables.
Again, causality condition had a very strong effect, F(2, 18) =
57.5, p ≤ .0001, and errors were higher in the delay condition
than in both the space and the causality conditions (ps ≤
.000001), but errors in the space and causality conditions did not
differ (p = .2). Yet there was no effect of trajectory, F(1, 18) =
0.05, p = .8, and trajectory and causality condition did not have
an interactive effect, F(2, 18) = 0.35, p = .7. That is, errors did
not differ between the distractor and experimental movies. This
result excludes an explanation in terms of transfer of representa-
tional momentum.
Finally, because many of our participants were highly
skilled physicists, we could also control for expertise effects.
An ANOVA with level of expertise as the independent variable
(naive, intermediate, professional) revealed that expertise had
no effect on prediction accuracy, F(2, 18) = 0.35, p = .71, or on
causality judgments, F(2, 18) = 1.35, p = .29.
Discussion
How does the cognitive system compensate when information
about the trajectory of a moving object is incomplete? The lit-
erature on mental imagery and on the prediction of motion
frequently appeals to mental analog representations as a poten-
tial substrate for spatial computations and understanding of
dynamic events. Here, we have provided evidence that this
approach may not offer an adequate account of how observers
represent dynamic stimuli.
We devised a motion-prediction task that directly probed
participants’ on-line ability to predict the future position of a
moving object, rather than testing memory for past positions,
as is done in other paradigms (e.g., Hubbard, 1995; Hubbard
et al., 2001). We found that participants were highly accurate
at determining the exact moment of arrival of a moving tar-
get only when the target was continuously visible. When it
was occluded and they had to rely on imagination, their esti-
mates were highly inaccurate, even though the scenes were
very simple and brief; the amount of error was as high as
70% of the scenes’ durations (see Table 1 and Fig. 2). These
results are in accordance with and expand on previous results
obtained with different paradigms (e.g., Gilden, Blake, &
Hurst, 1995). Neither participants’ knowledge that the move-
ment of the visible launcher was completely predictive of the
invisible movement nor their thorough understanding of the
Causality and Prediction 679
underlying physical laws could reduce or prevent the error.
Furthermore, the error did not appear to increase continu-
ously as traveled distance increased: Not only were partici-
pants unable to simulate a simple uniform movement, but
they could not even distinguish close intervals when these
intervals were not estimated by immediate sequential
responses; this was true regardless of the speed of the objects
and the causal correctness of the scenes. Taken together, this
quantum increase in error is difficult to reconcile with a sim-
ulation theory of imagined movement.
Finally, by coupling the movement-prediction task with
explicit judgments of causal strength, we showed that intuitive
judgments of causality do not accord with predictions of imag-
ined object movements. Participants’ predictions of the dis-
placements of occluded objects were better for scenes that
participants judged as highly causally incorrect (space condi-
tion) than for scenes they judged as more causally correct
(delay condition). Moreover, we showed that this dissociation
cannot be explained by an influence of representational
momentum on simulations of the targets’ movement. This dis-
sociation casts strong doubt on the possibility that a common
substrate integrates simulations of physical variables and
high-level knowledge about the simulated scenes into analog
representations that replicate and parallel reality.
How lower- and higher-level properties of visual representa-
tions are integrated is a widely discussed topic. Some authors
argue that visual perception is cognitively impenetrable: Low-
level visual mechanisms tie the mind to the world in a manner
that is unaffected by what an organism thinks or knows about
the world (Pylyshyn, 1986, 1999). Others argue that the high-
level interpretation of a scene influences low-level perceptual
properties. Of particular relevance to our topic is Buehner and
Humphreys (2010) argument that “the higher-level concept of
causality has a profound influence on humans’ perception of
both space and time” (p. 44), influencing low-level properties,
such as the perception of the size of an object. Our results add to
this debate, suggesting instead that causality does not influence
even imagined object movements. Indeed, even Buehner and
Humphreys’s experiments contain traces of the same dissocia-
tion we found. They suggested that space-time perception is
warped along the causality dimension because, in a memory
task, participants underestimated the actual size of an object
involved in causal scenes. However, scenes judged as highly
noncausal induced object-size recollections that were closer to
those induced by fully causal scenes than to those induced by
scenes that were similar to the scenes of our delay condition and
were similarly judged as violating causality only mildly. That is,
a causal violation judged to be more severe exerted less “spatial
warping” than a less severe violation.
These results, as well as ours, suggest that there is no sim-
ple way to understand the influence of causality on other cog-
nitive processes, nor any simple way to categorize causality in
humans’ cognitive landscape. Impressions of causality lie at
the “intersection between perception and cognition” (Scholl &
Nakayama, 2002, p. 493). Clearly, some low-level properties
that induce impressions of the presence or absence of causality,
such as contact between objects or temporal asynchronies
between the movements of two objects, have a profound effect
on other lower-level cognitive processes, such as memory for
an object’s size or reconstruction of a dynamic sequence. But
it remains a wide-open question whether it is causality per se,
as opposed to such low-level properties, that influences mem-
ory, perception, or imagination.
How can one account for participants’ poor prediction of
imagined movements, as well as for the dissociation between
assessment of causality and prediction of movement? Work
by Pylyshyn and his collaborators suggests one radical pos-
sibility: The visual indexes assigned to objects during track-
ing encode no object properties at all (Pylyshyn, 2007), or at
most encode objects’ last visible locations (Keane &
Pylyshyn, 2006), and if location information is encoded, it
cannot be updated by computations extrapolating trajecto-
ries. Indeed, in a multiple-object tracking paradigm, Keane
and Pylyshyn (2006) showed that participants track moving
objects that disappear better when the objects reappear at the
location of their disappearance than when they reappear at
the location where they should be, according to an extrapola-
tion updating their locations. Our data are compatible with
Pylyshyn’s hypothesis. This framework suggests that the
position of an object will be encoded correctly when continu-
ous visual feedback updates location information, but not
when such feedback is lacking. Therefore, our participants’
inaccurate yet systematic errors in estimating the location of
an invisible object may indicate that the imagined estima-
tions come from postperceptual processes, possibly involv-
ing neither spatial representations nor analog simulation of
trajectories. Indeed, although the visual system tends to
anticipate the position of moving objects (Nijhawan, 2008),
imagined position is characterized by severe time overesti-
mation, which suggests that a mechanism unrelated to visual
tracking per se may be involved.
Some studies suggest that subjective durations of events
involving moving objects tend to be longer than subjective
durations of events involving stationary stimuli (Brown, 1995;
Kanai, Paffen, Hogendoorn, & Verstraten, 2006). We suggest
that time dilation, although not sufficient to explain all our
results, may be partially responsible for the direction of the tim-
ing errors we observed. If participants tried to reproduce the
time necessary for a target to cover a certain distance and expe-
rienced time dilation induced by movement, they would overes-
timate how long it would take for the target to cover the distance.
In other words, rather than extrapolating an invisible object’s
position by means of an analog mental simulation of real physi-
cal forces, participants may have used an internal clock to esti-
mate the position of the object. Such an estimate would be so
coarse and detached from sources of knowledge about dynamic
events that it would not integrate an event’s basic causal struc-
ture. Further experiments are needed to explore this possibility.
Note that because time estimates recruit brain networks other
than those involved in spatial representation (e.g., Meck, 2005),
680 Levillain, Bonatti
our explanation suggests that the mechanisms involved in the
representation of physical scenes may be functionally and neu-
rally distinct from the mechanisms involved in the representa-
tion of imagined dynamic movement.
In conclusion, our results point toward the existence of dif-
ferent independent systems that may be involved in imagining
dynamic movement. One of these systems may translate the
passing of time into rough estimates of object positions, and
another may compute causal relations in the world. These
results show that dynamic imagination, if it exists at all, is
severely limited. They call into question the idea that the
essence of a thinking system is the ability to generate richly
detailed analog representations.
Acknowledgments
We thank J. Mehler and the members of the Language, Cognition and
Development Lab in Trieste for discussions and their help in setting
up our experimental paradigm; Z. Pylyshyn, M. Buehner, and M.
Nathan for their insightful comments; and L. Filippin and A. Isaja for
their technical support.
Declaration of Conflicting Interests
The authors declared that they had no conflicts of interest with
respect to their authorship or the publication of this article.
Funding
This research was supported by the Spanish Ministry of Science and
Innovation (Ref. PSI2009-08232) and by McDonnell Foundation
Grant 220020096.
Supplemental Material
Additional supporting information may be found at http://pss.sagepub
.com/content/by/supplemental-data
Notes
1. The supplemental videos illustrate the following combinations
of variables—Video S1: contact condition, target occluded, 19.3°/s
launcher speed, Bar Configuration 1, rightward movement; Video
S2: space condition, target occluded, 12.9°/s launcher speed, Bar
Configuration 1, rightward movement, 100-pixel spatial interval;
Video S3: space condition, target visible, 25.8°/s launcher speed,
Bar Configuration 2, leftward movement, 130-pixel spatial interval;
Video S4: delay condition, target occluded, 12.9°/s launcher speed,
Bar Configuration 1, rightward movement, 480-ms temporal interval;
Video S5: delay condition, target visible, 25.8°/s launcher speed, Bar
Configuration 2, leftward movement, 640-ms temporal interval.
2. We thank Marc Buehner for raising this possibility.
References
Borst, G., & Kosslyn, S.M. (2008). Visual mental imagery and visual
perception: Structural equivalence revealed by scanning pro-
cesses. Memory & Cognition, 36, 849–862.
Brown, S.W. (1995). Time, change, and motion: The effects of stim-
ulus movement on temporal perception. Perception & Psycho-
physics, 57, 105–116.
Buehner, M.J., & Humphreys, G.R. (2010). Causal contraction: Spa-
tial binding in the perception of collision events. Psychological
Science, 21, 44–48.
Gergely, G., Nádasdy, Z., Csibra, G., & Bíró, S. (1995). Taking the
intentional stance at 12 months of age. Cognition, 56,165–193.
Gilden, D., Blake, R., & Hurst, G. (1995). Neural adaptation of imag-
inary visual motion. Cognitive Psychology, 28, 1–16.
Hubbard, T.L. (1995). Environmental invariants in the representation
of motion: Implied dynamics and representational momentum,
gravity, friction, and centripetal force. Psychonomic Bulletin &
Review, 2, 322–338.
Hubbard, T.L., Blessum, J.A., & Ruppel, S.E. (2001). Representa-
tional momentum and Michotte’s ‘launching effect’ paradigm
(1946/1963). Journal of Experimental Psychology: Learning,
Memory, and Cognition, 27, 294–301.
Johnson-Laird, P.N. (1983). Mental models. Cambridge, MA:
Harvard University Press.
Kanai, R., Paffen, C.L., Hogendoorn, H., & Verstraten, F.A. (2006).
Time dilation in dynamic visual display. Journal of Vision,
6(12), Article 8. Retrieved from http://www.journalofvision.org/
content/6/12/8.full
Keane, B.P., & Pylyshyn, Z.W. (2006). Is motion extrapolation
employed in multiple object tracking? Tracking as a low-level,
non-predictive function. Cognitive Psychology, 52, 346–368.
Klein, I., Dubois, J., Mangin, J.-F., Kherif, F., Flandin, G., Poline,
J.-B., et al. (2004). Retinotopic organization of visual mental
images as revealed by functional magnetic resonance imaging.
Cognitive Brain Research, 22, 26–31.
Kosslyn, S.M. (1994). Image and brain: The resolution of the imag-
ery debate. Cambridge, MA: MIT Press.
Kosslyn, S.M., Thompson, W.L., Kim, I.J., & Alpert, N.M. (1995).
Topographical representations of mental images in primary visual
cortex. Nature, 378, 496–498.
Leslie, A.M. (1982). The perception of causality in infants. Percep-
tion, 11, 173–186.
Loftus, G.R., & Masson, M.E.J. (1994). Using confidence intervals
in within-subject designs. Psychonomic Bulletin & Review, 1,
476–490.
McGinn, C. (1989). Mental content. Oxford, England: Basil Blackwell.
Meck, W.H. (2005). Neuropsychology of timing and time perception.
Brain and Cognition, 58, 1–8.
Michotte, A. (1963). The perception of causality. Oxford, England:
Basic Books.
Nijhawan, R. (2008). Visual prediction: Psychophysics and neuro-
physiology of compensation for time delays. Behavioral & Brain
Sciences, 31, 179–198.
Pylyshyn, Z. (1999). Is vision continuous with cognition? The case
of impenetrability of visual perception. Behavioral & Brain Sci-
ences, 22, 341–365.
Pylyshyn, Z. (2003). Seeing and visualizing: It’s not what you think.
Cambridge, MA: MIT Press.
Pylyshyn, Z.W. (1986). Computation and cognition: Toward a foun-
dation for cognitive science. Cambridge, MA: MIT Press.
Pylyshyn, Z.W. (2001). Visual indexes, preconceptual objects, and
situated vision. Cognition, 80, 127–158.
Causality and Prediction 681
Pylyshyn, Z.W. (2007). Things and places: How the mind connects
with the world. Cambridge, MA: MIT Press.
Regan, D. (1992). Visual judgements and misjudgements in cricket
and the art of flight. Perception, 21, 91–115.
Scholl, B.J., & Nakayama, K. (2002). Causal capture: Contextual effects on
the perception of collision events. Psychological Science, 13, 493–498.
Scholl, B.J., & Pylyshyn, Z.W. (1999). Tracking multiple items
through occlusion: Clues to visual objecthood. Cognitive Psy-
chology, 38, 259–290.
Scholl, B.J., Pylyshyn, Z.W., & Feldman, J. (2001). What is a visual
object? Evidence from target merging in multiple object tracking.
Cognition, 80, 159–177.
Schwartz, D.L. (1999). Physical imagery: Kinematic versus dynamic
models. Cognitive Psychology, 38, 433–464.
Shepard, R.N. (2002). Perceptual-cognitive universals as reflections
of the world. Behavioral & Brain Sciences, 24, 581–601.
Shepard, R.N., & Cooper, L.A. (1986). Mental images and their
transformations. Cambridge, MA: MIT Press.
