Hoopen 1993 A New Illusion of Time Perception-II

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

A New Illusion of Time Perception—II

Author(s): Gert Ten Hoopen, Gaston Hilkhuysen, Gert Vis, Yoshitaka Nakajima, Fumihiko
Yamauchi and Takayuki Sasaki
Source: Music Perception: An Interdisciplinary Journal, Vol. 11, No. 1 (Fall, 1993), pp. 15-38
Published by: University of California Press
Stable URL: http://www.jstor.org/stable/40285597
Accessed: 14-11-2015 11:52 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

University of California Press is collaborating with JSTOR to digitize, preserve and extend access to Music Perception: An
Interdisciplinary Journal.

http://www.jstor.org

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Music Perception © 1993 by the regents of the
Fall 1993, Vol. 11, No. 1, 15-38 university of California

A New Illusion of Time Perception- II

GERT TEN HOOPEN, GASTON HILKHUYSEN, &


GERT VIS
Leiden University,The Netherlands

YOSHITAKA NAKAJIMA & FUMIHIKO YAMAUCHI


KyushuInstituteof Design,Japan

TAKAYUKI SASAKI
MiyagiGakuinWomen'sCollege,Japan
When one very short empty time interval follows right after another, the
second one can be underestimated considerably, but only if it is longer
than the first one. We coined the term "time-shrinking" for this illusory
phenomenon in our previous studies. Although we could relate our
finding to some studies of rhythm perception, we were not able to
explain the illusion. The present article presents our attempt to under-
stand the mechanism that causes the time-shrinking. Four experiments
are reported. The first one ruled out the possibility that the illusion
results from a difficulty in resolving the temporal structure. The second
experiment showed that the listener was not inadvertently judging the
duration of the first interval instead of that of the second one. In addi-
tion, this experiment yielded more information about the time window
within which the illusion occurs. The third experiment showed that
forward masking of the sound markers, delimiting the empty durations,
could not explain the illusion either. Furthermore, this experiment re-
vealed a clue to the mechanism of time-shrinking: competition between
expected and observed temporal positions. The fourth experiment fur-
ther examined the temporal conditions that give rise to the illusion and
showed that categorical perception plays a crucial role in the formation
of the illusion. In the general discussion, we argue that the illusion is due
to an asymmetric process of temporal assimilation.

Introduction
In a previous paper, we reported a new illusion of time perception
that we recently found. Although this illusion could be connected to
-
Requests for reprints may be sent to either Gert ten Hoopen, Leiden University, FSW
Exper. Psychology, P.O. Box 9555, 2300 RB Leiden, The Netherlands, or to Yoshitaka
Nakajima, Kyushu Institute of Design, Dept. of Acoustic Design, 4-9-1 Shiobaru, Minami-
ku, 815 Fukuoka, Japan.

15

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
16 ten Hoopen et al.

some phenomenain the field of time and rhythmperceptionand rhythm


production,the mechanismof the illusion itself could not be explained.
The present paper describesour attempt to find such an explanation.
We brieflysummarizethe characteristicsof the illusion in the following
paragraphs (see Nakajima, ten Hoopen, &cvan der Wilk, 1991, for
details).
Whenlistenerswere confrontedwith a shortstandardemptytime inter-
val (S) of 120 ms, precededby a shorterneighboringemptytime interval
(N), the durationof S was underestimatedto a considerableextent com-
pared with the control condition in which S was presentedalone (see
Figure 1 for a diagram of the stimulus patterns).When the preceding
neighbors(N) were 45, 70, and 95 ms, the points of subjectiveequality
(PSEs)of the S of 120 ms were about 70, 85, and 100 ms, respectively.A
slight underestimationof S also occurredwhen N equalledS (120 ms).
When the precedingN was longer (145, 170, 195 ms) than S, no clear
underestimationor overestimationoccurred(seeFigure2). Nakajimaet al.
(1991) also found that the conspicuousunderestimationsof S were not
paralleledby overestimationsof N of the samedegree,as a simpleassimila-
tion hypothesiswould predict.The simple assimilationhypothesisholds
that the subjectivedurationsof N and S tend to get closer to each other
than is expected from their physicaldurations.In addition,no underesti-
mations of S could be found when the shorterN succeededS. Further-
more, this study showed that the shrinkingeffect of a shorterpreceding
neighborwas verysmallwhen S was 240 ms (in whichcaseN was 90, 140,
or 190 ms).
In summary,the shrinkingof a time intervaloccurs when it is a rela-
tively short interval(^240 ms), and it is precededimmediatelyby an even
shortertime interval.In the presentstudy,we will first examinewhether
erroneousjudgmentsby listenerscan explain this phenomenon.Possible
erroneousjudgmentscan be dividedinto two maincategories.

Fig. 1. Stimulus patterns used in previous studies that established the time-shrinking illu-
sion. In the control condition (top), a standard time interval (S) had to be matched by a
variable comparison time interval (C). In the experimental condition (bottom), S was
immediately preceded by a neighboring time interval (N). The vertical bars represent very
short sound markers delimiting the empty time intervals. The broken vertical bars indicate
markers whose temporal position could be changed by the subjects.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 17

Fig. 2. The first and prototypical pattern of time-shrinking we found (Nakajima & ten
Hoopen, 1988). Preceding time intervals that were shorter than or equal to a standard time
interval (S) of 120 ms caused underestimations of S. The amount of overestimation is plotted
on the ordinate, thus negative values mean that the point of subjective equality (PSE)of S is
smaller than its point of objective equality (POE). The inserted vertical bars represent the SD
of the PSEs. Triangles are the PSEs of subject GtH, circles those of subject YN.

One type of erroneousjudgmentwould be when the listenerhas a hard


time resolvingthe temporalstructureof the /N/S/ pattern,because both
intervals are very short, and thereforeresorts to a composite duration
judgment.We call this the temporalresolution(TR)explanation.Another
type of erroneousjudgmentis when the listenerquite often matchesthe
duration of N instead of S, which we call the probabilisticmismatch
(PMM)explanation.
One cannot decide between these explanationsof our time-shrinking
notion on the basis of the averagednumericaloutcome of the duration
judgments,because the compositionrule (in the TR explanation)or the
probabilities(in the PMM explanation)can alwaysbe chosenso that they
yield the averagePSEpatternthat we found. Hence, our firsttwo experi-
ments approachthis problemby rearrangingthe temporalpatternsto be
used for the durationjudgments.In our third experiment,we will deter-
mine whether a peripheralprocess, namely forwardmasking,can cause
the underestimationof S.
In our previousexperiments,we kept S constantand studiedthe influ-
ence of a varying N on the PSE of S (Nakajima&cten Hoopen, 1988;
Nakajimaet al., 1991). Recentlywe were able to elucidatethe mechanism
of the illusion in a betterway by keepingN fixed and varyingS instead
(Nakajima, ten Hoopen, Hilkhuysen, &cSasaki, 1992; Nakajima, ten
Hoopen, &cSasaki,1992). Thesestudies,and unpublisheddata, suggested

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
18 ten Hoopenet al.

stronglythat the shrinkingof S by N disappearedwhen the time difference


betweenthese contiguousintervalsexceededabout 100 ms.
Furthermore,the patternof shrinkingand the releaseof shrinkingof-
fered strong evidence for the existence of categoricalperceptionon the
time dimension. However, Nakajima, ten Hoopen, Hilkhuysen, and
Sasaki(1992) studiedonly the effect of a precedingneighborof 50 ms on
an S rangingfrom 40 through280 ms. The purposeof the fourthand last
experimentto be reportedherewas to enlargethis scope:we plannedthree
fixed values of N, namely40, 80, and 160 ms with appropriaterangesof
S, in orderto establishhow time-shrinkingproceedswith increasingvalues
of N and S.

Experiment 1

This experimentwas designedto investigatewhetherthe TR hypothesis


could explain the underestimationof S. Underthis hypothesis,the listener
has difficultyresolvingthe temporalstructureof the N/S patternbecause
both intervalsare very short. Preparedto accomplishthe task, the subject
may thereforehave made an adjustmentof S, basedon an averageimpres-
sion of N and S. In that case, the N/S patterns45/120, 70/120, and 95/120
will yield PSEsof S of (45+ 120)/2 = 82.5 ms, (70 + 120)/2 = 95 ms, and
(95 + 120)/2 = 107.5 ms, respectively.Suchhypotheticaloutcomesmimic
the patternof underestimation,as found in previousexperiments,rather
well (cf. Figure2).
However, if listenershave difficultyresolvingthe temporalstructure,
simplybecauseN and S are too short,they shouldhave the sameproblem
when the N/S patternsare reversedto 120/45, 120/70, and 120/95 and the
task is still to match the second intervalsby a comparison.Underthe TR
hypothesis, overestimations of the second interval should now be ex-
pected. [PSEsof 82.5, 95, and 107.5 ms, respectively,on the basis of an
averageduration(N + S)/2].

METHOD

Six subjectswith normalhearingparticipated(leftearswerescreenedat 1000 Hz by an


audiometer).Theywere studentsof psychologyat LeidenUniversity,20-23 yearsold, two
women and four men.Theywere not paid, they agreedto participatein the experimentto
earncoursecredits.
The experimentalpatterns(firstinterval/secondinterval)were45/120, 70/120, 95/120,
120/45, 120/70, 120/95, 90/240, 140/240, 190/240, 240/90, 240/140, and 240/190 (ms).
The last six patternswere includedfor two reasons.First, as rewardingor motivating
conditions, because the matchingtask was easier with them than with the very short
patterns.Second,becausein our previousstudiesthe subjects,to a smallextent,underesti-
matedeven an S of 240 ms, when it was precededby a slightlyshorterinterval.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 19

The controlconditionscomprisedonly the S interval(45, 70, 90, 95, 120, 140, 190, and
240 ms) to be matchedby C. The soundmarkers,delimitingthe emptytimeintervals,were
approximationsto a squarewave of 1000 Hz, lasting10 ms and startingand stoppingat
zero-crossingpoints. The sound level of the markersignal was about 90 dBA, when
presentedcontinuously,as measuredby a precisionsound-levelmeter (Brüel& Kjaer
2203), mountedwith an artificialear (BrüelÔCKjaer4152) and a microphone(Brüel&
Kjaer4144).
The 20 differentpatterns(12 experimentaland 8 control conditions)were presented
eight timesto each subject,four timesin an ascendingfashion,wherethe initialCompari-
son (C) was subjectivelyshorterthan the Standard(S), and four times in a descending
pattern(initialC > S, subjectively).The subjectheardthe stimulion line, generatedby a
CommodoreAmiga1000 computer,via an amplifier(JVCAX 11) and headphones(AKG
K140) in the left ear. Five secondsafter S started,C was presented,and the task for the
subjectwas to equalizeC to S subjectively.This adjustmentwas done by pressinga mouse
buttonon a "shorten"or "lengthen"icon on the monitorscreen:the longerthe buttonwas
held down, the morethe finalsound markerof C was displacedtowardor away fromthe
initial sound marker.Clickingon a "presentation"icon initiatedthe presentationof the
changedpattern.Therewas no limit to the numberof presentations.Whensatisfiedwith
the matcheddurationbetweenS and C, the subjectclickedon a "finish"icon, at which
point the computerregisteredthe finalvalueof C as PSE.The next trialcould be startedby
clickingon the presentationicon again.
One blockconsistedof 40 trials(20 differentpatterns,once ascendingandonce descend-
ing).These40 trialswere randomizedand precededby 8 warmups.Aftera trainingblock,
the subjectsdid four blocksto be analyzed.The timingof the stimuluspatternsand PSEs
was carefullycalibratedand checked.Adjustmentcould be done in stepsof 1 ms, but the
subjectscould not monitorthe values.

RESULTSAND DISCUSSION

For each condition, the medianvalue of the 48 PSEs[6 (subjects)x 2


(ascending/descending) x 4 (replications)]was established(see Table 1).
Inspectionof the median PSEsin the control conditions shows that the
performanceof the subjectswas precise:the constant error was almost
negligible,andthe semi-interquartile rangeswererathersmall.Suchbehav-
ior formeda base rateagainstwhich the experimentalPSEscould be safely
contrasted.When we comparedthe 45/120, 70/120, and 95/120 patterns
with their control (120 ms), we found that the systematicunderestima-
tions of S, which we had foundpreviously,werereplicated(as was verified
by Wilcoxon tests). Inspectionof the 120/45, 120/70, and the 120/95
patternsclearlyshows that the subjectswere able to makenearlyveridical
matcheswith the second intervalS. Apparentlythe perceptionof the sec-
ond intervalwas not confoundedby the presenceof the firstone, that is,
the temporalstructurecould be resolved.Thus, these data do not support
the TR hypothesis.One may wonderwhetheror not the perceptionof the
first time intervalwas affectedby the presenceof the second intervalin
thesepatterns.Data fromNakajimaet al (1991) showedthat therewas no
such effect.This also is evidenceagainstthe TR hypothesis.
It is also worthwhileto examinethe otherconditions:an S of 240 ms was

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
20 ten Hoopen et al.

TABLE 1
Median Point of Subjective Equality (PSE) of Standard Time Interval (S)
and Semi-interquartile Range as a Function of Values of Preceding
Neighbor (N) and S
Semi-interquartile
N/S Pattern PSE Range Constant Error

0/45 44.0 4.25 -1.0


0/70 71.0 5.25 +1.0
0/95 96.5 8.00 +1.5
0/120 120.5 7.00 +0.5
45/120 83.0 21.50 -37.0
70/120 82.0 9.25 -38.0
95/120 92.5 10.50 -27.5
120/45 41.5 3.75 -3.5
120/70 68.0 5.50 -2.0
120/95 91.0 9.25 -4.0
0/90 93.0 8.50 +3.0
0/140 145.0 7.00 +5.0
0/190 191.0 9.25 +1.0
0/240 242.5 11.75 +2.5
90/240 241.0 28.25 +1.0
140/240 240.0 27.00 0.0
190/240 223.5 13.75 -16.5
240/90 94.0 9.25 +4.0
240/140 148.5 10.50 +8.5
240/190 202.0 11.25 +12.0

note. All values in milliseconds.

not underestimatedwhen the N was 90 or 140 ms. However,when it was


190 ms, a small but significantunderestimationof S occurred(5% as veri-
fied by a Wilcoxon test). This tendencywas also found in our previous
studies(Nakajimaet al., 1991), but therewe did not pay so muchattention
to it. Becausethis phenomenonseemsto be recurrent,we will discussit in
the generaldiscussion.In the 240/190 condition,the secondtime interval
(S)seemedto be overestimatedslightly(themedianPSEwas 202 ms). This
does not support the TR hypothesis,however,becausethe same kind of
overestimationdid not appearin the 120/95 condition,wherethe temporal
resolutionshould have been worse accordingto the TR hypothesis.

Experiment2

Althoughwe ruledout theTR explanationas the causeof time-shrinking,


anotherdeficitwith graspingthe fast temporalpatterncould in principle
also explainthe underestimationof S. In the preceding-neighbor
condition,
all subjectsclaimedthat they indeedattemptedto adjustC to S, but some

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 21

subjectsstatedto havedifficultiesrelatingthe secondandthirdsoundmark-


ers (delimitingS) when the firstand secondones (delimitingN) were closer
to each other. Accordingto their reports,the first two markersgave the
impressionof a group, and the thirdtrailingmarkerseemedisolated.
A reasonablehypothesisthen, is that the listenermay adjustC eitherto
N or to S probabilistically.1
Becausethe valuesof the PSEswere quiteclose
to the physicaldurationsof N, as can be seenfromFigure2, the probabilis-
tic balance may have been toward matchingN. So we consideredthe
possibilitythat subjects,despitethe instructions,often matchedthe shorter
N instead of S (PMM hypothesis).The shrinkingnotion, on the other
hand,holds that subjectsadjustC to S, as required,the subjectiveduration
of S being shrunkby a shorterprecedingN. In order to decide between
these two alternatives,we devised a stimuluspatternin which the posi-
tions of S and C were reversed(see Figure3).
Underthe shrinkinghypothesis,listenersshould overestimateS appar-
ently,that is, they should make C longerbecausenow C is subjectto the
shrinkingpower of the immediatelyprecedingN. Underthe PMM hy-
pothesis, the subjectshould exhibit greatdifficultiesin accomplishingthe
adjustmenttask. Let us take an examplewhere N = 45 ms, S = 120 ms,
and the initial C (which is positioned directly after N) is 60 ms. The
subject,who hearsthat this C is much shorterthan S, makesC longer by
successive presentations.Because C is longer than N, the relationship
betweenthe second and third (variable)sound markersas delimitersof an
intervalis less clear becausethe firstand the second markerare grouped,
therebyisolatingthe third one. Sincethe intervaldifficultto graspis also
the one to be manipulated(namelyC), the subjectshould displayhesitat-
ing behavior or confused adjustments.In the previous stimuluspattern
that we used (S immediatelyafterN), the subjectcould escapethe problem

Fig. 3. Stimuluspatternsused in Experiment2. The positionsof the standardtime interval


(S)and the comparisontime interval(C)were reversedas comparedto the arrangementin
Figure 1. In the control condition (top), C was presentedin isolation, whereas in the
experimentalcondition (bottom),C was immediatelyprecededby the neighboringtime
interval(N).

1. When two of us (GtH & YN) presentedthe time-shrinkingillusion to the 3rd


Workshopon RhythmPerceptionand Production(June1990, Horssen,The Netherlands),
some colleagues pointed out the possibility that the listener might erroneouslyhave
matchedN insteadof S.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
22 ten Hoopenet al.

of relatingthe second and thirdsound markersby unconsciouslyresorting


to a judgmentof the intervalbetween the first and second markers.The
present stimulus pattern prevents such behavior because this time the
intervalto be adjustedis the one that may be difficultto grasp.Perceiving
the firstdurationinsteadof the secondone would just increasethe confu-
sion of the subjectin the presentconfiguration.If C weremadelongerthan
S systematically,this would mean that time-shrinkingtook place without
any possibilityof mismatching.
A secondpurposeof the presentexperimentwas to includemorevalues
of S. In previousstudies (e.g., Nakajimaet al., 1991), we found that the
illusion did sometimesappearwhen S was 240 ms, but neverwhen it was
longer. However, no values of S between 120 and 240 ms were tested
systematically.We wanted to establishthe vanishingpoint of the illusion
between 120 and 240 ms more precisely.This goal can of courseonly be
reachedif the shrinking-hypothesis holds.

METHOD

Forthe presentstimuluspattern,in whichthe comparison(C)immediatelyfollowedthe


neighbor(N), 10 valuesfor the standard(S)as well as for N werechosen: 45, 70, 90, 120,
140, 170, 195, 240, 340, and 390 ms. Of the 10 x 10 possibleexperimentalconditions,60
informativeones were selected(seeTable2). Therewerealso 10 controlconditions:the 10
valuesof S. Notice that, in the controlcondition,C camebeforeS too. Becausein the initial
presentationof the pattern,C shouldbe eitherlongeror shorterthanS subjectively,a block
comprised2 (ascending/descending) x 70 (10 control + 60 experimental)= 140 trials.
Four male subjectsparticipated(the three Dutch authorsand a coauthor [Dutch]of
our 1991 article).Eachsubjectdid four blocksdividedover 16 sessions.Therewas ample
training,and beforeeach sessiontherewas a warm-up.For each condition,eight PSEs[2
(ascending/descending) x 4 (replications)]were gatheredfrom each subject.The rest of
the method was the same as in Experiment1, except that the computerregisterednot
only the PSEs but in addition the histories of the adjustments,that is, all series of
approachesto the PSEs.

RESULTSAND DISCUSSION
As can be seen from the top entriesof the columnsin Table2, the PSEs
in the control conditions(N = 0) almostequalledthe POEs.The constant
errorswere small:between+8.9% and -7.1%. Moreinterestingwas that
the resultsof the experimentalconditionsshowedthat the PMM hypothe-
sis had to be rejected.
Table 2 shows that the shrinkinghypothesiswas supported:subjects
were able to adjust C to S in a systematicway. When S was 170 ms or
shorter,N exerted the effect as expectedfrom the shrinkingnotion: if N
was shorterthan or equal to S, C could be eitherequalto or longerthan S
to get the same subjectivedurationas S had. It is impossibleto explainthe
existenceof PSEssystematicallylongerthanS withinthe frameworkof the
PMM hypothesis.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 23

TABLE 2
Mean Point of Subjective Equality (PSE) and Pooled Standard Deviation
(SD) as a Function of Duration of Standard Empty Time Interval (S) and
Duration of Empty Time Interval (N) Neighboring the Comparison
S = 45 S = 70 S = 90
N PSE(SD) N PSE(SD) N PSE(SD)

0 49(9.7) 0 73(15.2) 0 91(15.2)


45 72(16.5) 45 104(16.7) 45 128(32.6)
70 52(15.1) 70 91(20.9) 70 122(23.1)
90 46(7.8) 90 90(27.5) 90 108(21.7)
120 47(8.7) 120 69(11.5) 140 83(17.2)
170 43(9.2) 170 67(11.5) 240 87(15.1)
195 44(9.1) 195 66(14.0) 340 83(16.6)
390 81(15.2)

S = 120 S = 140 S = 170


N PSE(SD) N PSE(SD) N PSE(SD)

0 116(18.8) 0 130(20.8) 0 168(26.8)


45 149(14.1) 45 154(35.9) 45 177(26.4)
70 145(24.6) 70 149(27.3) 70 177(22.4)
120 134(23.8) 90 166(24.7) 120 190(20.8)
170 107(13.4) 140 146(29.5) 170 186(35.8)
195 109(16.8) 240 133(28.3) 195 163(22.6)
340 138(22.7)
390 123(20.0)

S = 195 S = 240 S = 340 S = 390


~N PSE(SD) N PSE(SD) N PSE(SD) N PSE(SD)

0 193(20.6) 0 233(27.5) 0 335(32.0) 0 375(26.1)


45 195(26.5) 45 233(44.3) 45 322(46.5) 90 363(47.5)
70 192(23.6) 70 223(36.1) 70 311(39.7) 140 363(36.9)
120 209(24.4) 90 245(27.6) 90 320(37.6) 240 417(63.7)
170 198(29.4) 140 228(28.8) 140 311(42.0) 340 374(44.3)
195 190(35.1) 240 226(36.1) 240 362(33.3) 390 373(51.2)
340 223(30.9) 340 318(51.9)
390 239(22.0) 390 307(31.6)

note. All values in milliseconds. The N entries of 0 ms indicate the control conditions
in which there was no neighbor. PSEs printed in boldface differed significantly (5%)
from the PSEs in the control condition as established by Duncan's multiple range tests.

Persuasiveevidencesupportingthe idea that subjectswere able to make


systematic instead of confused adjustments(as the PMM hypothesis
would predict) comes from the matchinghistories. Figure4 depicts one
typicalexampleof the 1120 matchinghistorieswe gathered(70 patternsx
4 subjects x 4 replications),and it can be seen that the ascendingand
descendingapproachesto the PSE are ratherregular.The fact that the

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
24 ten Hoopen et al.

Fig. 4. Example of the matching histories of an ascending (open circles) and a descending
(filled circles) adjustment trial by subject G.V. with a /70/C/-/120/ temporal pattern. The
comparison time interval (C), immediately preceded by a neighboring interval of 70 ms,
had to be adjusted to the subjective duration of a standard of 120 ms. PSEa(=135 ms) is
the point of subjective equality resulting from the ascending series of presentations,
whereas PSEd(=158 ms) resulted from the descending trial. The PSEof S in this replication
of the condition /70/C/-/120/ was the average: 147 ms.

subjects could perform systematic adjustments means that they could per-
ceive the duration of the second time interval, C, constantly, and the PSEs
longer than the POEs indicate that C was shrunk subjectively.
The data also provided an answer to the question of the vanishing point
of the illusion. At each value of S, we did an analysis of variance, and it
turned out that the significant main effect of N on the PSE of S ceased
beyond an S of 170 ms. Additional analysis was done by means of Dun-
can's multiple range tests (5%). The experimental PSEs that were signifi-
cantly longer than the control PSEs are printed in boldface in Table 2.2
2. In a study (Nakajima, ten Hoopen, Hilkhuysen, ÔCSasaki, 1992) that we carried out
after the present experiment was done, we found that the PSE distribution was bimodal
when S was preceded by an N that was about 70 ms shorter than S. It is evident that
analysis of variance is not warranted then.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 25

When S was 170 ms, and N was 120 ms, C was adjustedto a signifi-
cantlylongervalue of 190 ms (comparedwith the controlPSEof 168 ms).
When S was 195 ms and N was 120 ms, however,C was adjustedto a
value of 209 ms, which did not differfrom the correspondingcontrolPSE
of 193 ms. That suggeststhat a time intervalof 190 ms is shrunk,whereas
an intervalof 209 ms is not shrunkremarkablyanymoreby its preceding
N of 120 ms. (Recallfor a correctinterpretationof this analysisthat in the
presentpatternsit is not S, but C that is liableto shrinking.)Thus, 200 ms
is a roughestimateof the vanishingpoint of the shrinkingillusionwhen N
is 120 ms. The data in generalsupportedthe idea that thereis a vanishing
point at about 200 ms. In most cases in which a differencebetweenS and
PSEappeared,the PSEwas shorterthan 200 ms.
An interestingtendencycan be deducedfrom the data in Table2: the
maximumamount of shrinkingappearsto be dependenton the relation-
ship between S and N. The longer S becomes, the more N should be
increasedin orderto get the largestamountof shrinking.Weshallexamine
this relationshipmore systematicallyin Experiment4.

Experiment 3

This experimentwas performedto test a possible peripheralexplana-


tion of the time-shrinkingphenomenon:forward masking. Studentsof
psychoacoustics,reacting to our finding, often proposed that forward
maskingmust have causedour illusion.Indeed,the firstand secondsound
markers in those /N/S/- /C/ patterns where the illusion is most pro-
nounced,are temporallyvery close (between45 and 95 ms), and forward
masking,operatingover a rangeof about 100 ms (Elliot, 1971; Zwicker,
1982), may have takenplace. Althoughthis primarilyaffectsthe loudness
of the second marker,as a consequenceit could be that its temporal
position is less cleartoo. Becausethe maskedsecondsound markeris also
the initial markerof S, listenersmay have had difficultyestimatingthe
durationof S.
It was not quite clear from the psychacousticians'proposalswhy S in
particularshould be underestimated.But one could conceiveof a percep-
tual strategyin which the subject,hearingthreemarkersin successionof
which the secondis temporallyvague,locatesit subjectivelyhalfway.Such
a perceptualsolutionwould indeedgive riseto underestimationsof S. It is,
however,not clearhow the amountof maskingwould affectsucha percep-
tual bias.
A more preciseprediction,in which the differenceintensitylevel takes
effect, could be that more maskingintroducesmore temporalvagueness.
On that account,greatervarianceof the PSEsof S has to be expected,the

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
26 ten Hoopen et al.

higher the intensitylevel of the first markeris with regardto the second
marker.
Althoughwe couldnot deriveveryprecisepredictionsfromthe forward-
maskingproposal,we neverthelessconsideredit worthwhileto examineit.
We systematicallyvariedthe differencein intensitylevel betweenthe first
sound marker(= masker)and the consecutivesound markers.

METHOD

Eight subjects participated. The four Dutch subjects of Experiment 2 performed this
experiment in the Netherlands. Another four Japanese male subjects (authors Y.N. and F.Y.
and two students) executed the experiment in Japan. The apparatus and procedure were
comparable to those in the second experiment, except that the computer in Fukuoka was a
TEAC PS-9000-216, and the headphones in Leiden were a Beyer DT483 and in Fukuoka
were a Rion AD-02. Because a large dynamic range had to be covered in this experiment,
the headphones were connected to the speaker outlet via a resistor.
In the experimental trials, S was either 120 ms, preceded by an N of 45 or 95 ms, or S
was 240 ms, in which case N was 90 or 190 ms. Control conditions were also included,
where an S of 60, 120, 240, or 480 ms was presented without any preceding N.
Within each trial, the sound markers defining S and C had an identical intensity. In the
experimental trials, the relative intensity level of the first marker against the succeeding
markers was -36, -27, -18, -9, 0, 9, 18, 27, or 36 dB. The higher level among the two
levels was always fixed at 82 dBA. In the control conditions the intensities of all the sound
markers in a pattern were the same, but the level varied over trials between 46, 55, 64, 73,
and 82 dBA. Intensities were defined as estimated peak values in the fast mode on the
sound-level meter.
In total, there were 112 trials [56 patterns x 2 (ascending/descending)], which were
randomly divided among four equally long sessions. Each subject did three replications, so
that there were 12 sessions. The first four sessions were considered as training. Data from
the remaining eight sessions (two replications) were used for the analysis. Each session
started with seven warm-up trials. Other procedural measures were the same as before.

RESULTS AND DISCUSSION

Foreach of the 56 stimuluspatterns,the medianPSEwas obtainedfrom


the 32 individualPSEs[8(subjects)x 2 (ascending/descending) x 2(replica-
tions)]. Table 3 gives the median PSEsin the control conditions,and it is
clear that the intensity of the sound markersexerted no effect. Table 4
presentsthe PSEsand theirsemi-interquartile rangein the 45/120 and 95/
120 conditions,where the intensitydifferencebetweenthe firstand other
markersvariedbetween -36 and +36 dB. For reference,the appropriate
control PSEs(drawnfrom Table3) are included.
When the relativelevel of the firstmarkerrangedfrom -18 through36
dB, the amount of time-shrinking(in the N/S patternsof 45/120 ms) was
roughly of the same degree: about 33 ms on average. Note that it is
difficult to defend any substantialeffect of forward maskingwhen the
relative first markerlevel is -18 dB. Nevertheless,time-shrinkingtook
place, posing a seriousproblemto the forward-maskingaccount.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
TimeIllusionII 27

TABLE3
Median Point of Subjective Equality (PSE) and Semi-interquartileRange
(in parentheses) in Milliseconds for the Control Conditions (Standard
Time Interval without Neighbor) as a Function of the Standard (S) and
the Intensity (dBA) of the Sound Markers
Standard Time Interval (ms)

dBA 60 120 240 480

46 61(7) 120(11) 246(13) 486(10)


55 58(8) 120(10) 249(9) 492(13)
64 60(5) 122(9) 244(9) 480(19)
73 61(6) 120(9) 246(9) 481(17)
82 61(6) 118(7) 244(11) 483(15)

TABLE4
Median Point of Subjective Equality (PSE) and Semi-interquartileRange
(in italics) for the Control and the Experimental Conditions as a
Function of N, S, and the Relative First Marker Level
Relative First Marker Level (dB)
N/S
Pattern -36 -27 -18-9 0 9 18 27 36

0/120 118 118 118 118 118 120 122 120 120
7 7 77799 10 11
45/120 116 111 93 87 91 91 75 92 79
10 12 22 28 23 23 23 23 22
95/120 110 105 108 101 100 107 109 106 101
9 8 12 13 11 10 9 12 11
0/240 244 244 244 244 244 246 244 249 246
11 11 11 11 11 9 9 9 13
90/240 250 247 252 249 247 251 250 248 250
10 11 12 13 11 17 14 30 30
190/240 238 227 225 220 223 226 230 223 222
11 12 12 11 10 14 12 10 10

note. All values in milliseconds.Controlconditionsare 0/120 and 0/240 N/S pat-


terns.Experimentalconditionsare 45/120, 95/120, 90/240, and 190/240 N/S patterns.

Also of interest was the pattern of experimentalsemi-interquartile


rangesthroughoutthis range of first markerlevels: If more maskingin-
deed causesmore temporaluncertaintyof the second marker,and thus of
the durationof S, then the semi-interquartile rangeshouldhave increased
as the relativelevel of the first marker This was not the case at
increased.
all, and the data do not support the predictionswe derived from the
forward-maskingproposal.
Of particularinterestwere the PSEvaluesin the 45/120 conditionat the
relativefirst-markerlevels of -36 and -27 dB. Therewas no (2 ms) or

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
28 ten Hoopen et al.

nearly no (7 ms) shrinkingwhen the initial markerwas far less intense


than the following ones (also the semi-interquartilerangesat those levels
were far smaller).This unexpectedresultmay be a clue to the mechanism
underlyingthe time-shrinkingphenomenon.
We hypothesizeas follows: If both markersdelimitingN are not of too
differentloudnesses,a beat patternis induced(both markersget a beat),
and the next sound markeris expectedto arriveafteranotherN interval.
However, if the first markeris too soft, the first beat is perceivedto be
absorbed by the second marker (the initial one of the S interval),and
consequentlythere is no expectationabout the arrivalof the finalmarker
of S. Becausethere is no conflictthen betweenexpectationand observa-
tion, the durationjudgmentof S is not influencedby N.

Experiment4

In Experiment2, we usedpatternsof the formVN/C/-/S/,andthe task of


the listenerwas to adjustC to S. The predictionunderthe shrinkingnotion
was that S shouldbe apparentlyoverestimatedbecauseN shrinksits imme-
diate successor(C). This was indeedthe case with severalcombinationsof
N and S, as shown in Table 2. However, the adjustmenttask was not
always easy. With particularcombinationsof N and S, some subjects
reporteda curiousphenomenon.WhenC, which followedN immediately,
was increasedor decreasedobjectivelyby means of pressingthe mouse
button, a slight changeresultedin a disproportionalchangeof the subjec-
tive durationof C.
Since this effect seemed most pronounced when N was short, we
(Nakajima,ten Hoopen, Hilkhuysen,& Sasaki, 1992) investigatedthe
mattermoresystematicallyin an experiment.WekeptN constantat 50 ms
andvariedS, whichimmediatelyfollowedN, in smallstepsbetween40 and
280 ms. From the method of adjustment,it turnedout that the PSEof S
increasedmuch less than its POE did, when S increasedfrom 50 ms to
about 100 ms. Thatis, S was shrunkconsiderablythroughouta largerange,
while N remainedconstantat 50 ms. Beyond100 ms, the PSEof S suddenly
startedto increaserapidlyand reachedits veridicalat about 160 ms.
Whenwe did an informallisteningtest in which subjectswerepresented
with the patterns50/50, 50/60, 50/70, . . . 50/160 ms (the secondinterval
increasedin steps of 10 ms), we had a perceptualimpressionthat all
temporalpatternsup to about 50/100 ms soundedas though they had a
duration ratio of about 1:1. Despite that, the objective ratio changed
graduallyfrom 1:1 to 1:2. The differencebetweenthe neighboringdura-
tions suddenlybecameclear after that. Such a patternof judgmentssug-

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 29

geststhe existenceof categoricalperceptionof the ratiobetweenneighbor-


ing durations.
The aim of the fourthexperimentwas to determinewhetherthis pattern
of shrinkingcan also be found if we enlarge the time scope. There is
evidencefrom Experiment1 (see Table1) and Experiment3 (see Table4)
that shrinkingoccurseven with a 190/240 ms temporalpattern.The data
from Nakajima et al. (1991) also showed this, although they did not
describethe fact in the text. And Experiment2 (inspect the boldfaced
entriesin Table2) suggestedthat it seems to be the relationshipbetween
the magnitudesof the first and the second interval (Tl and T2) that
determinesthe occurrenceof shrinking.For significantshrinkingto occur,
Tl seemsto have to increasewhen T2 increases.On examiningour previ-
ous experimentaldata, includingthose of unpublishedexperiments,we
find that shrinkingseems to take place only when the differencebetween
T2 and Tl is less than about 100 ms.

METHOD

We employed a method of constant stimuli in a paradigm of three-alternative forced


choice (3-AFC). A pilot with a two-alternative forced choice task (2-AFC) and a 3-AFC
task showed that the latter was preferred by the subjects, but a more important reason for
choosing the latter was that the shape of the PSE distribution stemming from the 2-AFC
task was complicated. One Dutch and one Japanese author (GtH and TS) participated in
the experiment: TS in Leiden, and GtH in Leiden and Fukuoka (the experimental ambience
was almost the same in both places).3 The subjects' task was to compare the subjective
duration of the standard interval (S = T2) with the subjective duration of the comparison
(C) and to make a response by clicking on one of the three response icons on the screen
indicating that C was (1) shorter than, (2) equal to, or (3) longer than T2. In the experimen-
tal conditions, T2 was contiguously preceded by another short empty interval (N or Tl). In
the control condition, T2 was presented alone, that is, without Tl.
The duration of Tl in the experimental condition was 40, 80, or 160 ms. An empty T2
duration of 40, 50, 60, 70, 80, 100, 120, 140, or 160 ms followed the Tl of 40 ms. An
empty T2 duration of 50, 60, 70, 80, 100, 120, 140, 160, or 200 ms followed the Tl of 80
ms. The Tl of 160 ms was followed by a T2 duration of 80, 100, 120, 140, 160, 200, 240,
280, or 320 ms.
The 13 control conditions covered all the durations used as T2 in the experimental
conditions. Consequently the number of the standard patterns was 40 in total (9x3
experimental +13 control conditions). Each standard pattern was paired with a range of
comparisons and made up one experimental block.
The duration ranges and the step size of C were determined on the basis of the results of
a pilot experiment. A step size of 5 ms was chosen and the number of Cs embracing T2
varied from 9 to 25. The presentation order of Cs was randomized in the block and so was
the presentation order of the blocks. Both orders differed between subjects. The whole set
of judgments was repeated 25 times in 25 experimental sessions of 3 hr each.
A standard pattern was presented 2000 ± 200 ms after clicking on the presentation

3. The only serious difference in experimental setting between Leiden and Fukuoka
was a severely shaking audiobooth due to an earthquake (about 6 on the Richter scale)
during the 13th block of the 23rd session of subject GtH. He escaped the booth and
finished the session the next day. The data showed no anomalies.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
30 ten Hoopen et al.

icon, and C was presented after a period of 1700 ± 200 ms. The markers were 3000-Hz
sinusoidal tone bursts, whose duration was about 7 ms including a rise and a fall time of
about 1 ms. They were presented monaurally from headphones (JVC type HA-D515). The
intensity of the markers was set at 92 dBA (continuous).

RESULTS AND DISCUSSION

Because the number of response categorieswas three, we obtained


two cumulativefunctionsfor each standardpattern:One functionrelat-
ing the percentagejudgments"longer"and the other one relating the
percentagejudgments"shorter"to the value of C. By meansof the least
square solution using the Müller-Urbanweights, we calculated two
psychometricfunctions. The C values that correspondedto the 50%
points of the functions were taken as lower and upper limens, and we
obtainedthe PSEson the basis of the principleof equal likelihood(Guil-
ford, 1954, p. 138).
In Figure5, the resultsare depictedin the following way: The differ-
ences betweenthe experimentalPSEs(whereT2 was precededby Tl) and
the correspondingcontrol PSEs (T2 presentedalone) are plotted as a
function of the POE of T2 and of the value of Tl (40, 80, and 160 ms).
Such a portrayalgives a clearview of the interactionsbetweenthe neigh-
boring durationsTl and T2. If the PSEdifferenceon the ordinateis zero
(the dashedhorizontallines in the figure),then the durationjudgmentof
T2 was veridical.If the PSEdifferenceis negative,then T2 was underesti-
mated,or shrunkin our terminology,whereasT2 was overestimatedif the
differenceis positive.
The resultsof subjectGtH at a Tl of 40 ms (circlesin Figure5) show
that if T2 increasesfrom 40 to 120 ms, T2 is underestimatedmore and
more. Beyonda T2 of 120 ms, thereis a suddenreleaseof shrinking.The
pattern of subject TS at a Tl of 40 ms shows almost the same trend,
maximumshrinkingnow takingplacewhen T2 is 100 ms. Thesepatterns,
obtained by the constant method, are in close correspondencewith the
shrinkingpattern we establishedpreviouslyby means of the method of
adjustment(Nakajima,ten Hoopen, Hilkhuysen,&cSasaki,1992). In that
study,however,only a very shortTl (50 ms) was used. The presentstudy
offers an opportunityto see whetherthe typicalshrinkingpatternremains
the same when the time scope is enlarged.
The patternsof GtH and TS at a T2 of 80 ms (trianglesin Figure5)
appear to follow the same trend as those at 40 ms. When T2 increases
from 50 to 140 ms, the amount of shrinkingincreasesto about 35 ms.
Beyond 140 ms, thereis again a releasefrom shrinking.It is interestingto
note that both in the 40-ms and the 80-ms case, the maximumamountof
underestimationtakes place when the differencebetween T2 and Tl
amountsto 60-80 ms.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 31

Fig. 5. Points of subjective equality (PSEs) of T2 preceded by Tl minus the corresponding


control PSEs of T2, as a function of the point of objective equality (POE) of T2 and the
value of the preceding neighbor Tl (40 ms [circles], 80 ms [triangles], 160 ms [squares] )
for subjects GtH (top) and TS (bottom).

Let us finallyinspectthe patternsat a Tl of 160 ms (squaresin Figure


5). Again the maximumamountof shrinkingof T2 appearsto take place
when the the differencebetweenT2 andTl is 60-80 ms. In contrastto the
40-ms and the 80-ms cases there was, however,one differencebetween
GtH and TS: subjectGtH overestimatedT2 when it was shorterthan Tl.
(Thiseffectmay elucidateaspectsof assimilation,but as it is not pertinent
to our presentquestions,we will not considerit here.)
The fact that time-shrinkingis observedeven when Tl is 160 ms sup-
ports our discussionof Experiment3 that it is not a result of auditory
temporal masking. When the intervalbetween the first and the second
markeris no longer than 100 ms, we can hardlythink of any substantial
effectof forwardmasking.

General Discussion

In Experiments1 and 2, we attemptedto show that the time-shrinking


illusion is a real perceptualphenomenoninsteadof an artifactof erroné-

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
32 ten Hoopen et al.

ous matchingbehavior.Devastatingany explanationof the phenomenon


in termsof erroneousjudgmentsis an unpublishedexperiment,carriedout
by a groupof psychologystudentsfromLeidenUniversity.Theypresented
temporalpatternsin which a precedingneighborN was immediatelyfol-
lowed by a comparisonC, and by a standardS, 5 sec later.Sucha pattern
was used in Experiment2, wherethe valueof C had to be adjustedto S. In
the students'experiment,however, the observerwas requiredto change
the physicaldurationof N insteadof that of C, C beingheld constant.The
rationaleof the studentswas that, even thoughC was physicallyconstant,
its perceiveddurationcould be changedby manipulatingN. So the task
they set out for the subjectsconsistedof adjustingN until the subjective
durationof C (thatwas physicallyfixed)equalledthat of S. In orderto be
able to make predictions about the final value of N that satisfied the
subjectiveequalityof C and S, the studentsfirstreplicateda subsetof the
conditionsused in Experiment2: They adjustedC, precededby a fixedN,
until it equalledS subjectively.This finalvalue of C (thatis, the PSEof S)
was then taken as fixed in their experimentproper.Table5 gives a few
examplesdemonstratingthat theiridea worked,even thoughthe task was
very difficult, since now three instead of two time intervalshad to be
monitoredin an adjustmenttrial.If the subjectshad accomplishedthe task
by mismatching,that is, by matchingN and S, N should have had about
the same value as S, which was clearlynot the case. So all in all, Experi-
ments 1 and 2, and the students'experiment,make erroneousmatching
behavioruntenableas an explanationof the time-shrinkingphenomenon.
The possibilitythat maskingmight have causedour illusionwas com-
pletely refuted by Experiment3. This experimenthad an interestingby-
product:When the intensitylevel of the firstsound markerwas very low
with respect to the intensityof the following markers,delimitingS, the

table 5
Fixed Neighbor/Variable Comparison vs Variable
Neighbor/Fixed Comparison
/Nfix/Cvar/- /S/ /Nvar/Cfix/- /S/

45 58 45 47 58 45
70 116 90 73 116 90
90 159 140 93 159 140
120 188 170 102 188 170

Left Panel: Averaged values of the variable comparison (Cvar), pre-


ceded by a fixed neighbor (Nfix) that subjectively equalled the duration
of a standard time interval (S). Right Panel: Averaged values of the
variable neighbor (Nvar) that yielded subjective equality between the
durations of a fixed comparison (Cfix) and S. All values in milliseconds.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 33

shrinkingillusion disappeared(cf. Table4). This suggestedto us that the


illusionmaybe causedby a temporalset or expectation,builtup by N. If the
initialsound markerof N is too soft comparedwith the finalone, no clear
timepatternis perceived,althoughthe firstmarkeris abovethe threshold.If
the initialmarkerof N is intenseenough,the timepatternof N is clearand
can be a templateto perceivethe next time intervalthat is unfolding.
In general,one is not inclinedto speakof anticipationor expectationfor
a time window as short as this one. Anticipatoryor expectantbehavior
often seems to be based on, or accompaniedby, overt or covert forms of
motor behaviorlike tappingor implicitcounting,for which a time win-
dow of about 200 ms seems necessary.Nevertheless,for simplicity'ssake,
we used the term expectation.If one does not grantthe termexpectation,
one may conceptualizethe processas follows: by scanningthe contentsof
auditorystorage,the subjectjudgeswhetherthe momentof arrivalof the
thirdclick was consistentwith the simpleisochronicrule.
Havingheardtwo clicksthat spanTl (e.g., 40 ms), one mightapplythe
rule of isochrony:one judgeswhetherthe next click came after another
subjectiveduration correspondingto Tl. If the click arrivedat the iso-
chronicmoment,the rule is obeyed. If the click came too late, the rule is
violated,but the rule mightbe so strongthat it resistsa certainamountof
violation and lures away the actual, too late moment of arrival, back
toward the rule-basedmoment in auditorystorage. In other words, the
timing mechanismmakes a compromisein auditorystoragebetween ex-
pected and actualmoments.This meansthat scanningauditorystorageis
not a passiveregistration,but can be a processof activereconstructionof
the perceptualtrace. This active process may be called "regularization."
Elementsof this possible explanationare relatedto the ideas proposed
by Jones and Boltz (1989) in their temporallybased theory of attending.
Whatwe calledthe violationof the isochronicruleseemsto be an instance
of what they called "failed attunements."They define this term as
"asynchronybetweensome attendingrhythm(dT'n)and the relevantfocal
level of an event (dTn)"(Jones& Boltz, 1989, p. 473).
And we quote further:"One expects a certainfuturetime course . . . ,
which does not occur.The differencebetweenobservedand expectedfocal
periodsat some level, n, representsan asynchronythat is experiencedas a
-
temporal contrast: dTn dT'n." (p. 473). The authors suggested that
temporalcontrastmay influencetime discriminationand estimation.
We will now quantifythe regularizationprocess.A simplecompromise
betweenthe rule-governedand actual momentsof arrivalcould be some
linearlyweightedcombinationthereof.Our data allow us to estimatethe
weightingcoefficientfrom the equation:
PSEexp(of T2) = w x Tl + (1-w) x PSEcon(of T2) (1)

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
34 ten Hoopenet al.

where w is the weightingcoefficient.But since the PSEcon of T2 is nearly


equal to the POEof T2, we substituteit:
PSEexp(of T2) = w x Tl + (1 - w) x T2 = w(Tl - T2) + T2 (2)
The amountof shrinking(the negativeoverestimation)is
- = - =
-[PSEexp(ofT2) T2] -[MT1-T2) + T2] T2] m/(T2-T1)(3)
In Figure 5 (Expt. 4) this amount of underestimationwas plotted as a
functionof T2. BecauseTl is constant,the coefficientw can be estimated
as the slope of the linear regressionline through the part of the curve
between T2 = Tl and the T2 value where the underestimationwas the
largest.For a Tl of 40 ms, this rangeis froma T2 of 40 to a T2 of 120 ms
for subjectGtH. The slope turnedout to be - 0.45. This meansthat this
subjectweighted the expectedarrivaltime (40 ms) and the actual arrival
time of the last click almost equally.
We also estimatedthe coefficientw from previousdata (Nakajima,ten
Hoopen, Hilkhuysen,& Sasaki, 1992) becausein that study,where a Tl
of 50 ms was used, the PSEswere basedon eightsubjects.Forthe T2 range
from 50 to 100 ms (wherethe underestimationwas the biggest),the slope
of the linearregressionline turnedout to be - 0.71.
Superficially,our quantificationlooks like Jones and Boltz'sformaliza-
tion of their temporal contrast model (cf. p. 475). However, what is
weighted in their model is temporalcontrastwith mnemonicactivity.In
addition,theirweightingcoefficientswtand (1-w) standfor the (hypothe-
sized) psychologicalprocessesof future-orientedand analyticattending,
respectively.Our quantificationis ad hoc and less elaboratethan theirs.
We only weighted the rule-basedmomentand actual momentwithin the
temporalcontrastterm (in the terminologyof Jones& Boltz,we weighted
dT'nwith dTn)andestimatedw fromour data,withoutgivinga psychologi-
cal underpinningof the coefficient w. The main conceptual difference
betweenJones and Boltz's and our modelingis that one can hardlycon-
ceive of future-orientedand analyticattendingwith very fast patternsas
we studied.
A seriouscriticismcan be raisedto our combinationmodel:Why don't
listenersmakesuch a weightingbetweenTl and T2 if the latteris shorter?
Experiment 1 (see Table 1) and other studies from our laboratories,
showed that T2s that are shorterthan Tls are not, or are only slightly,
overestimated.Such a criticismcan be answeredonly by a posteriorirea-
soninglike: It seemsas if the simplepatternrule (1:1) is not violatedin this
case, but changed.Whenthe soundcomestoo soon, the timingmechanism
may be able to neglect the 1:1 hypothesis, and processes the T2 in a
veridicalway, becausethe expectationworks stronglyonly when the event
is too late.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 35

We admitthat our regularizationnotion, althoughnearlythe last step in


unravelingthe time-shrinkingmechanism,is not completely satisfying.
The notion cannot easily account for the asymmetryof the illusion. At
present,we are tryingto explain this final intricacyof the time-shrinking
mechanismin view of the model of emptydurationperception,proposed
by Nakajima (1987). Our attemptto solve the problemof asymmetryis
underway successfully(Nakajima,ten Hoopen, & Sasaki,1992).
We must emphasizethat temporalprocessingis characterizedby pro-
foundasymmetry.Numerouspsychoacousticstudies,for instance,showed
that forwardmaskingcan operateover time spans up to 100 ms or even
more, whereas backwardmasking,if presentat all, ceases within 50 ms
betweentargetand masker(cf. Moore, 1989).
At a higherperceptualor cognitivelevel, temporalasymmetryhas been
lucidly described in an essay by the famous Ernst Mach: "Wenndie
Zeitempfindungan die . . . stetig wachsendeArbeit der Aufmerksamkeit
gebunden ist, so wird es verständlich,warum die physiologischeZeit
ebenso wie die physikalischeZeit nicht umkehrbarist, sondern nur in
einem Sinneablauft.. . . Die beidennebenstehendenTakte[see Figure6],
welche für das Auge und den Verstandeine Symmetriedarbieten,zeigen
nichts Derartiges in Bezug auf die Zeitempfindung.Im Gebiete des
Rhythmusund der Zeit überhauptgibt es keine Symmetrie."("If time
perceptionis relatedto ... the continuouslyincreasingoperationof atten-
tion, then it becomesclearwhy physiologicaltime is irreversiblelike physi-
cal time, and only proceedsin one direction.. . . Both measuresdepicted
here {seeFigure6) which appearsymmetricto the eye and to the mind, do
not appearso with regardto time perception.In the field of rhythmand
time in generalthereis no symmetry."[(Mach, 1922, p. 209), translation
ours].
As is well known amongstudentsof time and rhythm,it was Fraisse(cf.
1946, 1956, 1982), who made a strongargumentfor assimilation(andits
opposite: distinctionor contrast)of temporalintervalsthat constitute a
pattern.By analyzingthe resultsof productionandreproductionof tempo-
ral patterns,he demonstratedthat listenersbasicallyuse two categoriesof
durations:"shorttimes" (200-300 ms) and "long times" (450-900 ms).
Moreover,Fraisse'sresultsshowed that therewas a tendencyto assimilate
durationsbelongingto the same categoryand to contrastdurationsfrom

Fig. 6. The two measures that Ernst Mach used to illustrate his argument for the asymme-
try of time and rhythm perception. The symmetric structure is grasped at once visually, but
when the notes are played, our ear has difficulty inferring symmetry.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
36 ten Hoopenet al.

differentcategories.Fraissearguedthat these tendenciesreflecta common


principle of economy in perceptualorganization.The results from our
laboratoriesare in accordancewith the fruitfulstipulationFraissemade,
but allow considerablerefinementfor the rangeof timeswe studied.
First, assimilationoccurs also for durationsshorter than the "short
times."Second,the assimilationappearsto be asymmetric:a durationin a
temporalpattern is much more assimilatedtoward precedingdurations
than toward succeedingdurations.Third, the assimilationof an interval
occurs only when the precedingdurationis shorter.And finally:Fraisse
suggestedthat a ratio (of about 1.5) seems to form the divisionbetween
short times and long times, whereas our studies show that assimilation
dependson the absolutedifferencebetweenconsecutivetime intervals.
Assimilationis strongestwhen the differenceis about 80 ms, is effective
for smallerdifferences,but disappearsrapidlywith largerdifferences.In
Experiment4, we found by meansof the constantmethodthat this assimi-
lation patternholds over a rangeof T1/T2 combinationsof 40/80 to 160/
240 ms. This was recentlyconfirmedby an experimentin which we used
the method of adjustment(Nakajima,ten Hoopen, &cSasaki,1992). Re-
cently,we showed that time-shrinkingoperateseven when Tl and T2 are
not contiguous,but are separatedin time. Shrinkingcan be observedif the
separatingintervalis smallerthan 400 ms, but is strongerthe closer T2
follows Tl in time (Sasaki, ten Hoopen, &cNakajima, 1992). Further
research(ten Hoopen, 1992) suggests that time-shrinkingbreaks down
when Tl exceeds 200 ms and that the contextualinfluencein slower Tl/
T2 patternsis governedby a differentmechanism.
In Nakajimaet al. (1991), we arguedthat our illusionmighthave been
responsiblefor some so-calledanomaliesof rhythmperceptionandproduc-
tion such as those reportedby Gabrielsson(1974). A problemthat we had
at that time was that the rhythmicpatternsstudiedby Gabrielssonwere
somewhat slower than the patternsfor which we demonstratedour illu-
sion. The suspicion we had, that time-shrinkingmay have caused the
anomaliesthe authorreported,seems warrantedin view of the resultsof
the presentstudy:shrinkingoperatesin a time window that is longerthan
we supposedbefore.
But one very important difficulty remains when trying to interpret
the anomaliesin terms of time-shrinking:In all cases in which the illu-
sion occurred, we used very short sound markers (between 5 and 10
ms) to delimit Tl and T2. Can we generalizeour finding to temporal
patterns comprisingmusical tones? Although we did no formal experi-
ments, we constructedtemporal patterns like T1/T2 = 40/40, 40/50,
40/60, . . . 40/140 ms, onset to onset, where the delimitingsounds were
quasi-musicaltones lasting 40 ms with very short attack times (see Fig-
ure 7). Informallisteningrevealedthat with this materialalso, a sudden

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
Time Illusion II 37

Fig. 7. Example of a temporal pattern in which a 40-ms and an 80-ms interval (onset to
onset) neighbor each other, marked by quasi-musical sounds. Despite an objective 1:2
ratio, the pattern is heard as 1:1.

change of duration ratio could be heard when T2 passed beyond about


90 ms. This supports our conviction that the time-shrinking illusion
alos works with patterns comprising musical sounds.4"6

References
Elliot, L.L. Backward and forward masking. Audiology, 1971, 10, 65-76.
Fraisse, P. Contribution a l'étude du rythme en tant que forme temporelle. Journal de
Psychologie Normale et Pathologique, 1946, 39, 283-304.
Fraisse, P. Les structures rythmiques. Louvain: Publications Universitaires de Louvain,
1956.
Fraisse, P. Rhythm and tempo. In: D. Deutsch (Ed.), The psychology of music. New York:
Academic Press, 1982.
Gabrielsson, A. Performance of rhythm patterns. Scandinavian Journal of Psychology,
1974, 15, 63-72.
Guilford, J.P. Psychometric methods (2nd ed.). New York: McGraw-Hill, 1954.
Jones, M.R., & Boltz, M. Dynamic attending and responses to time. Psychological Review,
1989, 96, 459-491.
Mach, E. Die Analyse der Empfindungen (Neunte Auflage). Jena: Gustav Fischer Verlag,
1922.
Moore, B.C.J. An introduction to the psychology of hearing (3rd ed.). London: Academic
Press, 1989.

4. The second experiment was presented as a paper at the first International Confer-
ence on Music Perception and Cognition, Kyoto, Japan, October 1989.
5. Leon van Noorden, Dirk-Jan Povel, Jaan Ross, Takashi Tsumura, Dirk Vorberg, and
Piet Vos gave us valuable opinions. Joke de Vaal, Jan Vereecke, René Vink, Michel
Warmenhoven, Inge Willemsen, and Marcel Zeelenberg, students of psychology, are
thanked for carrying out Experiment 2. Erik Wagemans did a lot of pilot work for Experi-
ment 4.
6. Our collaboration was supported by The Netherlands Organization for the Advance-
ment of Pure Research (Grant number 09-60 in the year 1987), the Kyushu Institute of
Design (for the trips of YN and GtH in the fiscal years 1987, 1989, and 1992), the Sound
Technology Promotion Foundation (in the fiscal year 1989) and the Ministry of Education,
Japan (for YN's trips in the fiscal years 1989 and 1990), the Miyagi Gakuin Women's
College, Sendai, Japan (for the study leave of TS to The Netherlands from April 1991 to
April 1992), and the Stichting Canon Foundation in Europe (for selecting GtH and YN as
1991 Canon Visiting Research Fellows).

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions
38 ten Hoopen et al.

Nakajima, Y. A model of empty duration perception. Perception, 1987, 16, 485-520.


Nakajima, Y., & ten Hoopen, G. The effect of preceding time intervals on duration
perception (in Japanese). Proceedings of the Autumn Meeting of the Acoustical Society
of Japan. 1988,381-382.
Nakajima, Y., ten Hoopen, G., ÔCvan der Wilk, R.G.H. A new illusion of time perception.
MusicPerception,1991, 8, 431-448.
Nakajima, Y., ten Hoopen, G., Hilkhuysen, G., & Sasaki, T. Time-shrinking: A discontinu-
ity in the perception of auditory temporal patterns. Perception & Psychophysics, 1992,
51, 504-507.
Nakajima, Y., ten Hoopen, G., & Sasaki, T. A discontinuity of time perception. In:
C. Auxiette, C. Drake, & C. Gérard (Eds.), Proceedings of the "Fourth Rhythm Work-
shop: Rhythm Perception and Production." Bourges, France, 1992, 7-12.
Sasaki, T., ten Hoopen, G., & Nakajima, Y. Time-shrinking in temporally separated condi-
tions. In: C. Auxiette, C. Drake, & C. Gérard (Eds.), Proceedings of the "FourthRhythm
Workshop: Rhythm Perception and Production," Bourges, France, 1992, 13-16.
ten Hoopen, G. Temporal processing of fast auditory patterns. Paper presented at the 2nd
International Conference on Music Perception and Cognition, Los Angeles, February
1992.
Zwicker, E. Psychoakustik. Berlin: Springer Verlag, 1982.

This content downloaded from 128.184.220.23 on Sat, 14 Nov 2015 11:52:15 UTC
All use subject to JSTOR Terms and Conditions

You might also like