The Basic-level convergence effect in memory distortions, Pansky & Koriat, 2004
Introduction
Research has frequently
replicated results showing how memory is lost more rapidly for specific
details. More general or abstract
information is usually preserved after long delays whereas verbatim details are
lost. Subjects have a harder time
discriminating between categorically related items than different category
items, producing false recognition of items.
Oddly, there have also been findings showing that more specific
information is gained over time. It is
argued that memory preferentially stores information at the basic level (BL),
an intermediate level of abstractness and specificity. This is the level at which objects are
spontaneously categorized, and that maximizes within-category similarity while
minimizing between-category similarity.
In this study, the researchers attempted to assess whether there was a
bidirectional shift towards this intermediate level of abstractness of brief
and long delays by use of recognition and recall testing for narrative stimuli.
Experiment 1
Methods
60 undergraduates from the
Results
Two judges determined if cued
recall fit target taxonomy, removing all responses that did not. Immediate testing produced 88% valid
responses while the delayed condition only produced 56%. There was a significant interaction such that
more items were recalled at the BL than at either superordinate
or subordinate level. The latter two
cases did not significantly differ. This
held true for both immediate and delayed recall. There was a trend for subjects to recall the
targets at the same level as at which they were presented, but this trend was
only present at immediate test. There
was a bidirectional and symmetrical effect where the majority of items were
recalled at the BL than any other level.
At immediate testing, the recognition of BL items was not significantly different
than for either superordinate or subordinate
items. At delay, however, recognition of
BL items was significantly greater than for other hierarchical levels. S’s chose BL items over the correct level
items even when all choices were available to them. Using a joint analysis ANOVA, a significant
BLC was found that was more pronounced for recall than recognition and greater
for delay than immediate testing. Incorrect
BL recognition of items presented at the superordinate
level produced higher confidence ratings than those correctly recognized at
this level. The opposite trend was
present for items presented at the subordinate level.
Experiment 2
In an effort to rule out
guessing as a potential cause of BL response trends in Experiment 1, the
researchers composed new sentences that were relatively context free that could
be completed with a range of different taxonomies. Finding a BLC in this condition was believed
to show it as a memory phenomenon.
Method
The same stories as in
Experiment 1 were used with 6 new target items added, and indications of
context removed from all sentences. 30
subjects were in each of three conditions: immediate testing, delayed testing
(24 hr), and control. In the control condition,
the subjects did not read any of the original stories and instead just filled
in the target blanks with any words they wished to make the story coherent. Testing involved a single cued-recall test
where the S’s completed the blanks where targets were omitted. In the first phase, S’s were discouraged from
guessing by a point system punishing incorrect responses. The second phase had S’s fill in all skipped
questions.
Results
As in Experiment 1, judges
omitted responses that did not belong to the target taxonomy. 68% of immediate testing responses were valid
and 46% of delayed. The control group
had a 0.07 proportion of responses in the target taxonomy. The previously observed BLC is extremely
unlikely to have been caused by guessing based on this finding. The overall BLC effect was present in both
free-choice and forced-choice testing phases at a significant level in both
immediate and delayed testing. Verbatim
accuracy was significantly higher for free report than forced report.
Discussion
Fuzzy-trace theory argues
that multiple levels of abstraction are processed simultaneously. The researchers add to this idea that BL
abstractions may tend to dominate and my decay more slowly over longer
intervals, resulting in the BLC effect experimentally observed. They also argue that the BLC effect could be
due to a reconstructive process at retrieval where fragments of memories are
reassembled and used to select a possible target candidate with BL aspects
being prominent. BLC could be due to nonmemorial factors as well, such as BL terms being shorter
and more frequent, people’s tendencies to communicate only as much specific
information as is necessary, measurement limitations in the current study, and
the possibility that the upward and downward shifts that converge at BL may be
two independent processes. They
encourage the exploration of these issues in future research.