Time’s arrow and pupillary response
ANTJE NUTHMANN and ELKE VAN DER MEER
Department of Psychology, Humboldt University at Berlin, Berlin, Germany
Abstract
The psychological arrow of time refers to our experience of the forward temporal progression of all natural processes.
To investigate whether and how time’s arrow is mentally coded in individual everyday events, a relatedness judgment
task was used. The items each consisted of a verb (probe) and an adjective or participle (target). The temporal
orientation between probe and target was varied either corresponding to the chronological orientation (e.g., shrink-
ingFsmall) or corresponding to the reverse orientation (e.g., shrinkingFlarge). Reaction times, error rates, and
pupillary responses were recorded. Chronological items were processed faster than reverse items. These findings
suggest that time’s arrow is mentally coded in single everyday events. Pupil dilation and results of principal component
analyses suggest top-down influences in the processing of temporally related items.
Descriptors: Temporal orientation, Events, Pupillary response, Cognitive load
The phrase ‘‘time’s arrow’’ was first introduced by Sir Arthur
Eddington (1928) in ‘‘Gifford Lectures’’ to describe the irre-
versible increase of entropy in isolated systems. ‘‘An arrow of
time is a physical process or phenomenon that has (or, at least
seems to have) a definite direction in time’’ (Savitt, 1995, p. 1).
Penrose (1979) was concerned with seven possible ‘‘arrows,’’ in-
cluding the process of measurement in quantum mechanics,
along with its attendant ‘‘collapse of the wave function.’’ the
expansion of the universe, and the direction of psychological time.
The latter alludes to our experience of the relentless forward
temporal progression of all natural processes. Surprisingly, in the
microscopic world of atomic particles laws of nature seem to
make no difference between forward and backward direction.
That is, time’s arrow is not found in the basic equations of
physics, but only in boundary and initial conditions that are open
to explanation (Vollmer, 1985). Therefore, complex questions
regarding the nature of time’s arrows must be addressed. Sklar
(1995) argues in favor of time symmetry at the microlevel, time
asymmetry at themacrolevel, and no fully compelling connection
between the two.
The present study investigates the macrolevel, namely the
psychological arrow of time. In everyday experience, most event
sequences are organized unidirectionally. For example, we can
witness the aging of a friend and his death, but we cannot ex-
perience this in the reverse order. Friedman (2002) provided ev-
idence that even 8-month-old children are highly sensitive to
temporal directionality in gravity-related events. These examples
point to the existence of a psychological arrow of time, that is, a
sensitivity to temporal directionality in real-life events.
It is characteristic for event sequences and events, about
which we have background knowledge, that they typically have
causal relations of some sort (cf. van der Meer, 2003). Trabasso,
van den Broek, and Suh (1989) differentiated, for example, mo-
tivational, physical, psychological, and enabling relations. Riedl
(1992) assumed that evolution structured our cognitive system to
reflect all environmental events as causally related. Classical
conditioning, for example, is based on animals’ and humans’
disposition to interpret events as causally related, if there is a
temporal relationship between them. Most physicists and phi-
losophers agree that there is a hierarchy of causality conditions.
‘‘The basic presupposition of the causality hierarchy is that of
temporal orientability’’ (Earman, 1995, p. 274). That is, causality
acts toward the future only. This widely accepted approach ex-
plains causality by means of time’s arrow. Alternatively, one
could explain time’s arrow by means of causality as proposed by
Reichenbach (1956) and Gru¨nbaum (1973). They proposed that
time’s arrows trace back to a causal arrow. In doing so, the
asymmetrical causal relation would be required as an undefined
basic concept. However, it remains completely open how events
might be identified as either causes or consequences independ-
ently from time’s arrows (cf. Vollmer, 1985).
The present article will consider the property of temporal ori-
entability or directionality as a basic presupposition of causality.
According to Friedman (2002), there are at present very limited
insights into the psychological processes underlying the sensitiv-
ity of humans to temporal directionality in real-life events. A
question that is fundamental to ask is: Is the psychological arrow
of time mentally coded? Freyd’s (1987, 1992) theory of dynamic
mental representations provides a general theoretical framework.
She assumes the temporal dimension to be inextricably embed-
ded in the mental representation of the external world and to be
directional. Similarly, Barsalou (1999) argues that our mental
representations of events are not arbitrary, but do preserve as-
Antje Nuthmann is now at the University of Potsdam.
Address reprint requests to: Elke van der Meer, Department of Psy-
chology, Humboldt University at Berlin, Rudower Chaussee 18, 12489
Berlin, Germany. E-mail: vdMeer@rz.hu-berlin.de.
Psychophysiology, 42 (2005), 306–317. Blackwell Publishing Inc. Printed in the USA.
Copyrightr 2005 Society for Psychophysiological Research
DOI: 10.1111/j.1469-8986.2005.00291.x
306
pects of the initial perceptual and experiential input. For routine
events, there is empirical evidence for this assumption. Routines
are descriptions of stereotypical, frequently encountered se-
quences of events (Galambos & Rips, 1982). Several studies
demonstrated the preference for the chronological order of rou-
tine events compared either with the reverse order or with a ran-
dom order in using a variety of different paradigms (cf., Mandler
& McDonough, 1995; Nelson & Gruendel, 1986; van der Meer,
Beyer, Heinze, & Badel, 2002).
On the other hand, there is very limited evidence on the rep-
resentation of time’s arrow within individual events (Zwaan,
Madden, & Stanfield, 2001). Adopting the framework proposed
by Freyd (1987, 1992) and Barsalou (1999), time’s arrow should
not only be coded in mental representations as a connection be-
tween events, but also in the mental representation of individual
events. The event shrinking shall serve as an example. Shrinking is
a temporally unidirectional event. An object is related to the
event shrinking. Among others, the object is characterized by the
opposing features largeFsmall. That is, the event shrinking re-
fers to an object changing from large to small. This transforma-
tion might imply temporal order information. This was the
starting point for the present study. According to Freyd (1987,
1992), mental representations of real-life events have an inherent
time component, making them dynamic representations. This
internal temporal dimension is directional, like external time.
Thus, items with a temporal orientation toward future time (e.g.,
shrinkingFsmall) are expected to be processed faster and with
higher accuracy than items with a temporal orientation toward
past time (e.g., shrinkingFlarge). The first aim of the present
study was to test this hypothesis. A relatedness judgment task
was used. Participants had to decide whether probe–target pairs
were related. The probe was a verb naming an event (e.g.,
shrinking), whereas the target named a feature of an object re-
lated to the event (e.g., small). Relatedness of probe and target
was assumed when the target was a feature that correctly char-
acterized the event. For related items, the temporal orientation
between probe and target was varied: It could either correspond
to the chronological orientation (chronological items, e.g.,
shrinkingFsmall) or to the reverse temporal orientation (reverse
items, e.g., shrinkingFlarge).
In addition, the time interval between the presentation of the
probe and the presentation of the target (stimulus onset as-
ynchrony, SOA) was varied: 250 ms vs. 1000 ms. Characteristic
time constants for automatic spreading activationmechanisms are
a mere 200–250 ms (Fischler & Goodman, 1987; Neely, 1977). If
the SOA is considerably longer, strategic processes can modify
results of automatic activation (Neely, 1991). A frequently used
SOA that enables strategic processing is 1000 ms. In probe–target
paradigms, SOA effects do not strictly argue for either automatic
activation or controlled access to mental representations (cf. van
der Meer et al., 2002). However, compared with priming tasks,
recognition procedures provoke elaborate, semantic processing of
information and are considered to be a more direct method of
measuring howmemorablemental representations are (Gernsbac-
her & Jescheniak, 1995). For that reason, the recognition proce-
dure was used in the current experiment.
Pupillometrics
A second aim of the study was to support behavioral data, that is,
reaction times (RTs) and error rates, with psychophysiological
data. The pupillary response has proved to be a sensitive, reliable,
and consistent measure of the processing load induced by a task,
orFmore broadly definedFresources allocated to a task (cf.
Beatty & Kahneman, 1966; Beatty & Lucero-Wagoner, 2000;
Goldwater, 1972; Hess & Polt, 1964; Loewenfeld, 1993). The
following rule applies: The more difficult a task is or the more
complex a cognitive process is, the more the pupil dilates. Like
eye movements (see Rayner, 1998), pupillary movements are a
good index of moment-to-moment on-line processing activities.
Different aspects of cognitive activity have been successfully in-
vestigated using the pupillary response during the last decade:
language processing (Hyo¨na¨, Tommola, & Alaja, 1995; Just &
Carpenter, 1993), perception (Verney, Granholm, & Dionisio,
2001), memory performance (Granholm, Asarnow, Sarkin, &
Dykes, 1996; van der Meer, Friedrich, Nuthmann, Stelzel, &
Kuchinke, 2003), and attention (Kim, Barrett, & Heilman,
1998).
For the current study, the following global hypothesis holds:
Processing of reverse items consumes more resources than
processing of chronological items. To test this hypothesis, peak
dilation and latency to peak were determined as parameters of
the pupillary response. For reverse items, these parameters were
expected to have higher values than for chronological items.
Principal Component Analysis (PCA) of Pupillary Responses
In addition, the current study had a third, methodological aim
motivated by an apparent paradox in pupillometric research (cf.
Schluroff et al., 1986): On the one hand, pupillary movements
are considered to be a reliable physiological index of resource
consumption. On the other hand, typical measures of the pupil-
lary response are comparatively unidimensional. Thus, the ques-
tion arises how to compress and analyze all the information
represented by a pupillary response. In event-related brain po-
tentials (ERP) research, PCA in combination with analysis of
variance (ANOVA) has proven to be meaningful and successful
(Donchin & Heffley, 1978). The advantage of PCA for the eval-
uation of pupillary responses lies in the fact that all information
of the pupil data is taken into consideration rather than that of
single data points. To further investigate the usefulness of PCA in
pupillometric research, we subjected averaged pupillary respons-
es to PCAs (cf. Granholm & Verney, 2004; Schluroff et al., 1986;
Siegle, Granholm, Ingram, & Matt, 2001; Siegle, Steinhauer, &
Thase, 2004; Verney,Granholm,&Marshall, 2004).We expected
to identify a component reflecting the distinct processing de-
mands associated with chronological and reverse items. As for
the time course of the pupillary response waveform, the differ-
ence in processing chronological and reverse items was expected
to appear in a rather late processing stage associated with de-
cision processes.
Method
Participants
Ninety-six psychology students of Humboldt University in Ber-
lin participated in the experiment. They received either course
credit or DM 10 payment for their participation. All of them had
German as their mother tongue. Twenty students (17 women and
3 men; mean age: 26.1 years) participated in a first pretest to
generate the experimental materials and to examine their ade-
quacy. Twenty students (13 women and 7 men; mean age: 24.3
years) participated in a second pretest to examine the temporal
relatedness of items. Twenty students (12 women and 8 men;
mean age: 26.3 years) participated in a post hoc free association
Time’s arrow and pupil response 307
study to explore the association strength between probe and tar-
get, which is assumed to indicate the general semantic relatedness
of the experimental materials (Strube, 1984). Thirty-six students
participated in the main experiment. Six participants had to be
excluded from all analyses because of technical difficulties. For
themain experiment, the final sample consisted of 30 students (21
women and 9 men; mean age: 24.7 years). Students could only
participate in one of these studies.
Stimuli and Materials
In a first pretest, participants had to generate verbs that described
individual events. Additionally, they were asked to produce pairs
of adjectives that are highly familiar past- and future-oriented
characterizations of the previously generated events (e.g., shrink-
ing: largeFsmall). In total, participants generated 136 different
triplets. These triplets were examined in a second pretest. Par-
ticipants were presented with a verb (e.g., shrinking) describing a
change in time. The verb was accompanied by a pair of adjectives
or participles (e.g., largeFsmall). Participants had to rate on a 5-
point scale (from 15 very bad to 55 very good) how well the
word
word文档格式规范word作业纸小票打印word模板word简历模板免费word简历
pair reflected the change in time. The ratingwas assumed to
show how well the word pair was able to depict changes in per-
sons or objects, associated with a specific event. Those triplets
(individual event and feature pair) that reached a median of at
least 4 on the rating scalewere selected. Next, highly emotional as
well as especially short or long triplets were excluded. The re-
maining triplets were believed to best represent the temporal di-
rectionality of real-life events. The chronological and reverse
items (i.e., related items) were constructed in the following way:
For chronological items, an individual event was combined with
its future-oriented feature (e.g., steamingFtender). For reverse
items, an individual event was combined with its past-oriented
feature (e.g., shrinkingFlarge).
Because the temporal relationship is a special case of semantic
relationship, we intended to control the experimental materials
for global semantic relatedness, too. In a post hoc free associ-
ation study, the participants were presented with the probes (e.g.,
shrinking) and were asked to utter the first words that came to
mind. All free associations that were generated within 10 s were
recorded. For every participant and every related item, four bi-
nary scores (yes vs. no) were determined, scoring 1 as ‘‘yes’’ and 0
as ‘‘no’’: (1) Was the first associative response to the presented
probe the targetword? (2)Was the first response aword similar to
the meaning of the target (e.g., a synonym)? (3) Was the target
word within the top five responses to the probe? (4) Was a word
similar to the target within the first five responses? Next, for each
of the four association strength measures, the proportion of par-
ticipants for whom a positive response was found was calculated.
Of course, the first association measure exhibits the lowest mean
probe–target association frequencies (chronological items: 0.09;
reverse items: 0.07) whereas the fourth shows the highest values
(chronological items: 0.24; reverse items: 0.19). These free asso-
ciation findings correspond with results reported in the literature
(see Strube, 1984, for a complex review). For verbs, adjectives are
associated with low frequency and rather late in the association
sequence. For statistical analysis, we used the mean of the four
association strengthmeasures as a combinedmeasure. A 2 (SOA:
250 vs. 1000 ms) � 2 (temporal orientation: chronological vs.
reverse items) item ANOVA yielded no significant effects,
SOA: F(1,36)5 0.180, MSE5 0.021, p5 .674, Z25 .005;
temporal orientation: F(1,36)5 0.744, p5 .394, Z25 .020;
SOA � temporal orientation: F(1,36)5 1.419, p5 .241,
Z25 .038.1 Thus, probe–target association frequency is equal
for the experimental item groups.
The main experiment consisted of two item blocks, each con-
taining 12 practice and 40 test items. Each item was composed of
the probe (e.g., shrinking) and the target (e.g., large). Fifty per-
cent of the items were related (e.g., shrinkingFlarge); the re-
maining 50% of items were unrelated (e.g., shavingFfar). For
related (i.e., experimental) items, the temporal orientation be-
tween probe and target could either correspond to the chrono-
logical order (e.g., steamingFtender), in which case the items
were referred to as chronological items, or it could run against the
chronological order, in which case the items were referred to as
reverse items (e.g., shrinkingFlarge). The chronological and re-
verse item groups were also controlled for the number of letters
(for probes, mean5 8.1 letters; for targets, mean5 5.2 letters)
and word frequency (for probes, mean5 16.5 occurrences/mil-
lion; for targets, mean5 248.1 occurrences/million; CELEX da-
tabase; Baayen, Piepenbrock, & Gulikers, 1995).
The unrelated probe–target pairs (filler items) were con-
structed by using the same 40 individual events as for the related
items. They were combinedwith features that had occurred in the
unused triplets. Thus, in the main experiment every individual
event (probe) appeared twice: In one item block it was part of a
related item whereas in the other item block it was part of a filler
item. Because the block order was switched between participants,
the word repetition was not supposed to have a confounding
effect.
The experiment was run in German. All examples have been
translated into English. The original materials, both in German
and English, may be obtained from Antje Nuthmann.
Design
The following independent variables were considered in the ex-
periment (within subjects): SOA (250 ms and 1000 ms) and tem-
poral orientation (chronological and reverse). The participants
were presented half of the items with an SOA of 250 ms (Block 1)
and the other half with an SOA of 1000 ms (Block 2). The block
order was switched between participants, who were randomly
assigned to one of the two versions. Probe and target were either
related (50%) or unrelated (50%). For related items, the tem-
poral orientation between probe and target was varied: either
corresponding to chronological order (50%; e.g., steamingFten-
der) or reverse order (50%; e.g., shrinkingFlarge). Unrelated
items (i.e., filler items) had no meaningful relation (neither tem-
poral order nor global semantic relation) between probe and
target (e.g., shavingFfar). These filler items were included in the
experiment so that participants would not only be exposed to
related items. No hypotheses weremade regarding the processing
of filler items. Still, they were included in some exploratory
analyses. Within an SOA condition, items were presented ran-
domly.
The following dependent variables were recorded: RTs, error
rates, and pupillary responses.
Procedure
The experiment took place in a quiet, medium-illuminated room
(background luminance5 500 lux). The participants received
written instructions. They were seated comfortably i