Brainerd,
C.J., Reyna, V.F., & Wright, R. (2003). Recollection Rejection:
False-Memory
Editing in Children. Psychological Review, 110, 762-784.
There
has been an increased interest in identifying cognitive methods for humans to
use as a means to avoid false reports of events that support their gist
memory. In this article we will look at
an editing operation that satisfies two main criteria: 1. we will review the
process of recollection rejection, and 2. we will look at four classes of
phenomena supporting recollection rejection.
These four classes are broken down into qualitative (compare false
memory rates across different experimental conditions) and quantitative
measures (estimate actual amounts of recollection rejection using mathematical
models).
Evidence
that adults edit false-but-gist consistent information is seen in
self-monitoring studies, yet there are limitations when using this method. That is why a specific editing operation for
false memories is imperative. This
process theory comes from the dual-retrieval models of memory.
Dual-Retrieval
Models. There is a classic account of recognition,
which attributes recognition judgments to familiarity. Familiarity is a process based on a
continuous strength scale. However,
Mandler (1980) proposed that recognition involves both familiarity and
recognition, which can be explained in the following equation:
P(H) = R + (1-R)F
Since Mandler’s proposition many other
elaborations have stemmed from this perspective: 1. Tulving’s (1985) remember-know
judgments, 2. disassociations between these judgments based on manipulations
(e.g. divided vs. full attention), 3. the development of Jacoby’s (1991)
process-disassociation model, and later Brainerd et al.’s (1998) conjoint
recognition model, and 4. fuzzy trace theory.
Recollection
Rejection. Recollection rejection occurs when a related
distractor is rejected due to recollection of the target (e.g. “I know I didn’t
see Phoneix because I saw Tuscon.”).
This occurs because mismatches between surface forms of distractors and
recollected targets are generated (Brainerd et al., 1995). The fact that distractors cannot be
recollected, yet targets can be increases the probability for rejecting the
distractors. In short, recollection of
true events provides substantial support against the false events (Reyna &
Lloyd, 1997).
There are three important components of
recollection rejection that must be present for the process to occur: 1.
availability of verbatim traces of targets, 2. tendency for semantically
related distractors to enhance retrieval of verbatim trace for the targets, 3.
perception that recollected targets and semantically related distractors differ
in surface details.
Recollective
Suppression of Semantic False Memory. In this
section we will look at various manipulations in an attempt to make verbatim
traces more accessible, increasing recollection rejection.
Augmenting
the Accessibility of Verbatim Traces at Test.
Among the articles mentioned (p. 767), test phase cuing was the focused
manipulation. Two further manipulations
were made in the test phase cuing that should affect the ability of
meaning-preserving distractors to access verbatim traces of the targets. These manipulations include 1. target
priming – presenting the target immediately prior to the distractor, and 2.
presentation delay – varying the time between study and test. It was assumed that target priming (first
manipulation) would make verbatim traces more accessible, and that by
shortening the time between study and test (second manipulation) verbatim
traces should be more accessible because memory for surface details fade
quicker than gist. Both were found to
reduce semantic false-recognition effects.
Strengthening
Verbatim Traces During Study.
Accessibility of verbatim traces may also be strengthened by varying the
rate of presentation of target materials and by varing the content of such
materials. Research suggests that when
targets are familiar words, memory involves two stages: recognition (meaning)
and postrecognition (surface form).
From this it can be inferred that gist traces are stored before verbatim
traces. Therefore, fast presentation
rates should impair recollection rejection, and increase
false-recognition. Toglia &
Neuschatz (1994) produced these results for false recall with DRM lists
presented at 1 second and 4 seconds.
Seamon et al. (1998) found that false-alarms initially increased when
DRM lists were presented from 20 ms to 250 ms, but later decreased at 2000
ms. This inverted U-relation suggests
that 20 ms is too short for semantic processing to occur, so false-alarm rates
increase until semantic processing is completed. Then, false-alarm rates begin to decrease as verbatim processing
is completed.
In addition, it is argued from previous
research that narratives presented in a linear order between objects (e.g. The
coffee is hotter than the tea. The tea
is hotter than the cocoa. The cocoa is
sweet.) give rise to stronger verbatim
traces than those specifying spatial relations (e.g. The flowers are on the
table. The table is under the
light. The flowers are in a green
pot.). Reyna & Kiernan found this
to be true, finding that linear were recognized more accurately than
spatial.
Israel and Schacter (1997) argue that verbatim
traces are stronger for pictures than for words by producing lower false-alarm
rates for critical unpresented words in the picture + word vs. the letter +
word. Schacter et al. (1999) explained
their findings as supporting the distinctiveness heuristic (p. 769) rather than
recollection rejection. Yet this
explanation is not consistent with other paradigms.
Inverted
U-Relations Between Retrieval Time and Semantic False Recogntion. This qualitative measure follows the assumption that recollection
is slower than familiarity. Therefore,
as previously discussed, false-alarms should initially increase as familiarity
accumulates and then decrease as recollection accumulates. More time is needed for related distractors
to access verbatim traces than gist, which is why false-alarms should initially
increase. These predicted results have
been proven thus far for both pair and item recognition. Rotello & Heit (1999) determined that
criterion shifts could not account for inverted U-relations for pair recognition.
Measuring
Recollection Rejection With Conjoint-Recognition Methodology.
Separating the Contributions of
Target Recollection and Familiarity to False
Reports. Estimates
of recollection rejection and familiarity for semantically related distractors
are measured according to different sets of instructions at test: verbatim
(accept only targets), gist (accept only semantically related distractors), and
verbatim + gist (accept both targets and semantically related distractors). These can be shown in the following
formulas:
Prd v = (1-Trd)Frd
+ (1-Trd)(1-Frd)Bv
Prd g = Trd
+ (1-Trd)Frd + (1-Trd)(1-Frd)Bg
Prd v+g = Trd
+ (1-Trd)Frd + (1-Trd)(1-Frd)B
v+g
These
equations have the following assumptions: 1. responses supported by
recollection take precedence over responses supported by familiarity, and 2.
recollection rejection and familiarity are independent processes. Also, if recollection rejection is occurring
for the types of related distractors that produce false memories, Trd
should not equal 0.
The
manipulation used when measuring recollection rejection is repetition. In Table 1 (p. 774) it can be seen that all
estimates of recollection rejection are greater than 0, and that repetition
produced greater recollection rejection.
Target priming also produced increased recollection rejection 4 x the
level of unprimed category names.
See Table 2 (p. 775) for further results of
recognition experiments. Overall, in
the presentation of new experiments, it is clear that by measuring recollection
rejection, related distractors were rejected through target recollection. Also, recollection rejection increased as
the repetition of targets increased and as the interval between study and test
decreased. Recollection rejection
decreased in the misinformation conditions.
Measuring
Recollection Rejection with ROCs. ROCs plot
hits (targets) against false alarms (unrelated distractors) to corresponding
confidence levels that participants make on a scale from sure-old to
sure-new. Each point on an ROC curve
measures the cumulative probability of giving targets a specific confidence
value (and all of the values to the left of the value) and the cumulative
probability of giving unrelated distractors the same confidence value. If the number of hits and false alarms are
equal then a symmetrical ROC along the diagonal will result. The better recollection is the more the ROC
curve should bow to the point (1,0).
Although the skewed ROC violates signal detection theory’s process
assumptions, it is predicted by a dual-retrieval model. Recollection should cause an increased
confidence for targets; therefore, the ROC curve should intercept the y-axis
where no false alarms are accepted. The
distance from point (0,0) to the y-intercept can be measured as recollection.
Rotello showed recollection rejection as
the point at which the curve intercepts the upper x-axis. See Figure 4 (p. 778). The measure can be
calculated as the distance between the right y-axis and the point on the upper
x-axis.
Rotello’s findings fit the conjoint
recognition model and ROC model, in particular showing that 1. estimates of
recollection rejection proved that related distractors are being rejected by
recollection of the corresponding targets, and 2. the manipulations were consistent
with the predictions for producing recollection rejection.
A
Complementary Operation: Erroneous Recollection Rejection. Although erroneous recollection rejection should not frequently
occur, it sometimes does. Erroneous
recollection rejection occurs when a target is erroneously rejected because
another object is remembered as being studied as the “target.” Brainerd et al. (2003) states that joint
processing of targets and verbatim traces of semantically related targets could
generate mismatches that yield rejections, meaning that participants might
treat related targets as related distractors.
Erroneous recollection rejection can be added to the conjoint
recognition equations as the following:
P(Hv) = Tt
+ (1-Tt)(1-Et)Ft
P(Hm) = Tt
+ (1-Tt)Et + (1-Tt)(1-Et)Ft
Table
3 (p. 779) shows the measures of erroneous recollection rejection in
Experiments 10-12 reported by Brainerd et al. (2001). Experiment 11 showed no signs of erroneous recollection
rejection; however Experiments 10 and 12 did.
Summary. As seen in numerous studies, the false-memory editing process of
recollection rejection follows dual-process theory and has produced predicted
measurements that it is occurring under variations of manipulations. The findings prove recollection rejection to
be an advantageous strategy to help suppress false-memories by allowing
accessible verbatim traces to counter the familiar false-but-gist consistent
events. One last note is that Rotello
et al. (2000) found that adults still used recollection rejection without
awareness, yet increased levels were reported when they had such an
awareness.