Beruflich Dokumente
Kultur Dokumente
1.31, 95% CI [0.15, 2.40]; see gray bars in Figure 2, and the
experienced delay intervals on F-cued probe trials and regular
R-cuedtrialsdidnot differ, t(5) 0.40, p .71. AsinExperiment
1A, accuracyonthefirst 20andlast 20F-cuedprobetrialsdidnot
differ significantly, t(5) 0.74, p .49, indicatingthat monkeys
did not learn to respond indiscriminately on probetrials. In con-
trast to thedifferenceinaccuracy betweenR andF-cuedmemory
trials, accuracy onR-cuedprobetrials remainedas highas that of
regular F-cued trials (regular F-cued trials: M .97, SD .06;
R-cuedprobetrials: M 1.00, SD .01). As inExperiment 1A,
therewas no significant differencein experienced delay intervals
betweenregular F-cuedtrialsandR-cuedprobetrials, t(5) 1.88,
p .12, d
z
0.77, 95% CI [0.18, 1.66], and monkeys were
significantly faster in responding on R-cued discrimination trials
than on R-cued memory trials, t(5) 2.61, p .05, d
z
1.06,
95% CI [0.01, 2.06], consistent with a lack of hesitation in re-
spondingonR-cueddiscriminationtrials. Theseresults, collected
under conditions that prevent prospectivememory for discrimina-
tiontargets, reproducethefindingsfromExperiment 1A indicating
activecontrol of memory on R-cued trials.
Experiment 1C
Thedisruptioninmemory accuracy followingF cues observed
inExperiments1A and1B providesstrongevidencefor theability
of monkeys to actively regulatethecontents of workingmemory.
However, monkeys performed almost perfectly on each discrimi-
nationproblem, earningarewardnearly every trial, whereas their
memoryaccuracyaveragedonly.64. Thisdiscrepancymayleadto
differences in expectation of food rewards after R and F cues. If
monkeys expect a food reward after seeing an F cue but then
receiveamemorytest that hasamuchlower probabilityof reward,
they may befrustratedor less motivated, providinganalternative
nonmnemonicexplanationfor thelower accuracyonF-cuedprobe
trials. ThisconcernwasaddressedinExperiment 1Cbyyokingthe
reinforcement onregular F-cuedtrialswiththat onregular R-cued
trials to equatetheexpectation of food reward.
Method
Thesamesix monkeys andtitrateddelays wereused. No addi-
tional trainingwasneeded. TrialsproceededasinExperiment 1B,
but theprobability of reinforcement on regular F-cued trials was
yoked with that on regular R-cued trials. Becausemonkeys were
nearly 100% correct on F-cued discrimination trials, this was
accomplished by calculating the proportion of R-cued trials re-
wardedintheprevioussessionandrewardingthesameproportion
of correct discrimination responses in the current session. Mon-
keysranonesessionof 400R-cuedmatchingtrialsand400F-cued
discriminationtrialsbeforewestartedyoking. Theperformanceon
theseR-cuedtrials was calculatedandusedfor yokingtheF-cued
trials in the first session of the current experiment. All settings,
including number of discrimination tasks used and trial numbers,
remainedthesameas inExperiment 1B. Probetrials wereimple-
mentedinthefirst yokedsession, andtwosessions wereobtained
fromeach monkey. The accuracy and response latencies of the
resulting80probetrials of each typewerecomparedin thesame
way as in Experiments 1A and 1B.
Results and Discussion
Monkeys were again more accurate on R-cue trials than on
F-cueprobetrials, evenafter theprobabilityof reinforcement onR
andF trialswasequated, t(5) 2.77, p .04, d
z
1.13, 95%CI
[0.05, 2.15]; see open bars in Figure 2. Further analysis showed
T
h
i
s
d
o
c
u
m
e
n
t
i
s
c
o
p
y
r
i
g
h
t
e
d
b
y
t
h
e
A
m
e
r
i
c
a
n
P
s
y
c
h
o
l
o
g
i
c
a
l
A
s
s
o
c
i
a
t
i
o
n
o
r
o
n
e
o
f
i
t
s
a
l
l
i
e
d
p
u
b
l
i
s
h
e
r
s
.
T
h
i
s
a
r
t
i
c
l
e
i
s
i
n
t
e
n
d
e
d
s
o
l
e
l
y
f
o
r
t
h
e
p
e
r
s
o
n
a
l
u
s
e
o
f
t
h
e
i
n
d
i
v
i
d
u
a
l
u
s
e
r
a
n
d
i
s
n
o
t
t
o
b
e
d
i
s
s
e
m
i
n
a
t
e
d
b
r
o
a
d
l
y
.
5
MEMORY CONTROL IN MONKEYS
that this differencewas not causedby adifferenceinexperienced
delayinterval, t(5) 0.06, p .96. AsinExperiments1A and1B,
performanceonF-cuedprobetrials didnot decreasesignificantly
betweenthefirst andlast 20trials, t(5) 1.36, p .23, d
z
0.56,
95%CI [0.33, 1.40], indicating that themonkeys did not learn
that probe trials were differently rewarded than regular trials.
Performanceinthediscriminationtests was high, andequivalent,
on F-cued and R-cued trials (regular F-cued trials: M 1.00,
SD .00; R-cued probe trials: M 1.00, SD .01). Again,
experienceddelayintervalsonregular F-cuedtrialsandonR-cued
probe trials did not differ, t(5) 0.57, p .59, suggesting that
there was no disruption in responding to R-cued probe trials. In
addition, experienceddelay interval onR-cueddiscriminationtri-
als was significantly faster than those on R-cued memory trials,
t(5) 2.86, p .04, d
z
1.17, 95%CI [0.07, 2.20], consistent
with a lack of hesitation in responding on R-cued discrimination
trials.
It might takesomeexperiencewiththeyokingprocedurefor the
monkeys to learn about the new reinforcement rate for the dis-
criminationtests. If thiswerethecase, wemight not expect tosee
aneffect of yokinguntil thesecondsessionof testing. Toevaluate
whether a yoking effect might emerge only after experience,
accuracies in the first and second session of testing were com-
pared. Theperformanceinthefirst andthesecondyokedsessions
didnot differ significantlyonregular R-cuedtrials(M.64inthe
1st session, M .66 in the 2nd session), t(5) 0.83, p .44,
regular F-cuedtrials (M .99, SD .01inthe1st session, M
1.00, SD .00inthe2ndsession), R-cuedprobetrials(M1.00,
SD .00inthe1stsession, M.99, SD .01inthe2ndsession),
or F-cuedprobetrials(M .46inthe1st session, M .50inthe
2nd session), t(5) 1.13, p .31, d
z
0.46, 95% CI [0.41,
1.29]. Theseresults indicatethat thepoorer memory performance
on F-cued probetrials was not dueto lower motivation or lower
expectation of reward.
Takentogether, Experiments1A, 1B, and1Cdemonstrateactive
control of memorybymonkeysandeliminateconfoundingfactors
that commonly contaminate studies of active memory control in
nonhuman animals. Specifically, our data suggest that monkeys
areabletofollowtheR andF cuesandactively maintainmemory
when it is necessary to keep themost current and relevant infor-
mation availablein working memory.
Experiment 2
Theability to actively maintain working memory was demon-
strated in Experiment 1. Although this establishes aparallel with
the item method used in studies of human working memory,
human subjects areoften presented with lists of stimuli followed
by an instruction to remember or forget in a list method. To
determinewhether themonkeyswouldgeneralizetheRandF cues
to amodifiedtest situation andto provideacloser parallel to the
variety of methodsusedinhumans, wepresentedtwoconsecutive
sample images followed by a single cue and tested memory for
bothof thestudiedimages. If monkeysgeneralizeuseof theR and
F cuesfromthepreviousexperimentstothecurrenttestswithshort
lists of two images, weshouldcontinueto observesuperior accu-
racy on R-cued trials compared with F-cued probetrials.
Method
Subjects and apparatus. Thesamesix monkeys and testing
equipment wereused.
Cued matching to sample. Thefour images used on match-
ing trials in Experiment 1 were always grouped together as a
quad at test, withoneof themrandomlychosenasthesampleas
before. Four new color clip-art images, 160 pixels high 200
pixels wide, were added and grouped together to make a second
quad. Thus, one image from each quad could be randomly
selectedoneachtrial tocreateatwo-imagelist of samplesfor that
trial. Trials progressed as described in Experiment 1, except that
after touching (FR2) the first sample image, it disappeared and
another sampleimageappearedinthesamelocationafter 200ms.
Monkeys had to touch (FR2) the second sample image before
receivingtheR or F cueandthedelayinterval. At test, if anR cue
had followed the two sample images, two matching tests were
presented one-by-one in the same order as the corresponding
sampleimages hadbeenpresentedinthestudy phase. If anF cue
had followed the two sample images, two of the 10 pre-trained
discrimination tests were randomly selected and presented one
after the other (Figure 1B). Correct responses were immediately
rewardedwithafoodpellet andanexcellent! sound. Reinforce-
ment onF-cuedtrials was not yokedinthis experiment. Incorrect
responses resultedinadoh! sound, but no timeout period. The
timeout periodwasomittedtokeepthedelay interval betweenthe
offsetof eachsampleimageandtheonsetof thecorrespondingtest
constant, evenwhentherewereincorrect responses. Eachsession
contained192trials, half of whichwereR-cuedandtheother half
F-cued. The selection of samples for the two matching tests, the
order in which these two samples appeared in a trial, and the
discrimination problems used were counterbalanced in each ses-
sion. Thesamesymbols wereusedas R cues andF cues for each
individual.
Accuracy titration. Becausetherewerenowtwo, rather than
just one, to-be-remembered images, performance on the R-cued
trials was retitrated by the same procedure described in Experi-
ment 1. All trials initially hada3-s delay interval fromtheoffset
of thesampletotheonset of thecorrespondingtest. Monkeyswere
movedtothenext stagewhentheaverageproportioncorrect of the
twomatchingtrialsafter R cuesfell between.50and.70for three
consecutivesessions.
Probe testing. Eachsessionconsistedof 256R-cuedtrialsand
256 F-cued trials, among which 16 R-cued and 16 F-cued trials
wereselectedto beprobes. OnF-cuedprobetrials, bothdiscrim-
inations werereplacedby matchingtests intheorder correspond-
ingto samplepresentations, whereas onR-cuedprobetrials, both
matchingtestswerereplacedbydiscriminationtests. Regular trials
were rewarded when the correct image was selected, whereas
probe trials were rewarded no matter which image was selected.
Each monkey was tested for 8 sessions, yielding 128 probetrials
of each type.
Data analysis. Theeffects of cues and presentation order on
accuracy and experienced delay interval were both analyzed by
repeated-measuresANOVAs. All accuracy dataweretransformed
by takingthearsineof thesquareroot of correct proportionsprior
toanalysis(Kirk, 1982). Experienceddelayintervalsweredefined
astheinterval fromtheoffset of eachsampletotheregistrationof
aresponse(FR2) tothecorrespondingtest. Mediandelayintervals
T
h
i
s
d
o
c
u
m
e
n
t
i
s
c
o
p
y
r
i
g
h
t
e
d
b
y
t
h
e
A
m
e
r
i
c
a
n
P
s
y
c
h
o
l
o
g
i
c
a
l
A
s
s
o
c
i
a
t
i
o
n
o
r
o
n
e
o
f
i
t
s
a
l
l
i
e
d
p
u
b
l
i
s
h
e
r
s
.
T
h
i
s
a
r
t
i
c
l
e
i
s
i
n
t
e
n
d
e
d
s
o
l
e
l
y
f
o
r
t
h
e
p
e
r
s
o
n
a
l
u
s
e
o
f
t
h
e
i
n
d
i
v
i
d
u
a
l
u
s
e
r
a
n
d
i
s
n
o
t
t
o
b
e
d
i
s
s
e
m
i
n
a
t
e
d
b
r
o
a
d
l
y
.
6
TU AND HAMPTON
wereanalyzed. Significant main effects werefurther analyzed by
plannedpairedt tests. For eacht valuelarger than1, wereported
the standardized Cohens d
z
for within-subjects designs as sug-
gestedbyLakens(2013) aswell asthe95%confidenceinterval for
this measure of effect size calculated using a bootstrapping pro-
cedurein R (Cumming, 2012).
Results and Discussion
Accuracy titration. Titrateddelaysrangedfrom612s(M
8.50 s, SD 2.26 s). Thesedelays aresignificantly shorter than
those used in Experiments 1B and 1C, indicating that it is more
difficult for themonkeystoremember twoimagesthanone, t(5)
3.76, p .01, d
z
1.54, 95%CI [0.28, 2.73]. Accuracyinthelast
three sessions of titration averaged .62 .04 on the first test of
regular R-cued trials and .55 .04 on thesecond test of regular
R-cued trials. Monkeys were more accurate in first tests than in
secondtests, t(5) 3.96, p .01, d
z
1.62, 95%CI [0.33, 2.85],
averaged across thelast threesessions.
Probe testing. Whenthecuewasfollowedby matchingtests,
therewasasignificant maineffect of cue, F(1, 5) 15.30, MSE
0.01, p .01,
p
2
0.75. Further analysisshowedthatbothsample
imageswererememberedbetter followingRcuesthanfollowingF
cues: first tests: t(5) 3.99, p .01, d
z
1.63, 95%CI [0.34,
2.86]; secondtests: t(5) 3.43, p .02, d
z
1.40, 95%CI [0.21,
2.53]; see Figure 3. This discrepancy was not caused by differ-
ences in experienced delay intervals F(1, 5) 1.14, MSE
52,667.09, p .34,
p
2
0.19. The performance on the first 20
F-cuedprobetrialsdidnot differ significantlyfromthat onthelast
20 F-cued probe trials, t(5) 0.79, p .47, indicating the
monkeysdidnot learntorespondindiscriminately onprobetrials.
Performanceondiscriminationtrials was almost always at ceiling
without variation for conducting statistical analysis regardless of
whichcueprecededthetest(after Rcues: M.98, SD .02; after
F cues: M 1.00, SD .00), and no significant differencewas
foundinexperienceddelay intervals after F or R cues, F(1, 5)
0.12, p .74. These results reinforce the findings from the
preceding experiments and extend themto memory for multiple
sample images. As in Experiment 1, we compared the response
latencies on R-cued discrimination trials. Latencies for regular
F-cuedtrials andR-cuedprobetrials didnot differ either for first
tests, t(5) 0.61, p .57, or second test, t(5) 0.22, p .84,
indicatingthat performancewas not disruptedonR-cueddiscrim-
ination trials. In addition, response latencies for R-cued probe
trials were significantly shorter than those for regular R-cured
trialsonthefirst tests, t(5) 9.33, p .0001, d
z
3.81, 95%CI
[1.39, 6.22], and not significantly different than thosefor regular
R-cuedtrialsonthesecondtests, t(5) 1.91, p .11, d
z
0.78,
95% CI [0.17, 1.68], indicating that the monkeys were not
hesitant on R-cued probetrials.
When the cue was followed by matching tests, there was a
significant main effect of test order on performance, F(1, 5)
11.45, MSE 0.01, p .02,
p
2
0.70, andonexperienceddelay
interval, F(1, 5) 80.14, MSE 361,793.29, p .001,
p
2
0.94,
but theinteractionbetweencueandtest order wasonlysignificant
in performance, F(1, 5) 10.16, MSE 0.001, p .02,
p
2