Cognitive Psychology and Its Implications, Ch. 7
7
Human Memory:
Retention and Retrieval
Popular fiction frequently has some protagonist who is unable to recall some critical
memory—either because of some head injury, repression of some traumatic
experience, or just because the passage of time has seemed to erase the memory.
The critical turning event in the story occurs when the protagonist is able to recover
the memory—perhaps because of hypnosis, clinical treatment, returning to an old
context, or (particularly improbable) being hit on the head again. Although our everyday
struggles with our memory are seldom so dramatic, we all have had experiences
with memories that are just at the edge of availability. For instance, try remembering
the name of someone who sat beside you in class in grade school or a teacher of a
class. Many of us can picture the person but will experience a real struggle with retrieving
that person’s name—a struggle at which we may or may not succeed. This chapter
will answer the following questions: • How does memory for information fade with the passage of time? • How do other memories interfere with the retrieval of a desired memory? • How can other memories support the retrieval of a desired memory? • How does a person’s internal and external context influence the recall
of a memory? • How can our past experiences influence our behavior without our being able
to recall these experiences?
•Are Memories Really Forgotten?
Figure 7.1 repeats Figure 6.1, identifying the prefrontal and temporal structures
that have proved important in studies of memory. This chapter will focus more
on the temporal (and particularly the hippocampal) contributions to memory,
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 176
Are Memories Really Forgotten? | 177
which play a major role in retention of memory. One of the earliest studies of
the role of the temporal cortex in memory seemed to provide evidence that
forgotten memories are still there even though we cannot retrieve them. As part
of a neurosurgical procedure, Penfield (1959) electrically stimulated portions of
patients’ brains and asked them to report what they experienced (patients were
conscious during the surgery, but the stimulation technique was painless). In
this way, Penfield determined the functions of various portions of the brain.
Stimulation of the temporal lobes led to reports of memories that patients
were unable to report in normal recall, such as events from childhood. This
seemed to provide evidence that much of what seems forgotten is still stored
in memory. Unfortunately, it is hard to know whether the patients’ memory
reports were accurate because there is no way to verify whether the reported
events actually occurred. Therefore, although suggestive, the Penfield experiments
are generally discounted by memory researchers.
A better experiment, conducted by Nelson (1971), also indicates that forgotten
memories still exist. He had participants learn a list of 20 paired associates,
each consisting of a number for which the participant had to recall a
noun (e.g. 43-dog). The subjects studied the list and were tested on it until
they could recall all the items without error. Participants returned for a retest
2 weeks later and were able to recall 75% percent of the associated nouns when
cued with the numbers. However, interest focused on the 25% that they could
no longer recall—were these items really forgotten? Participants were given
new learning trials on the 20 paired associates. The paired associates they had
missed were either kept the same or changed. For example, if a participant had
learned 43-dog but failed to recall the response dog to 43, he or she might now
be trained on either 43-dog (unchanged) or 43-house (changed). Participants
were tested after studying the new list once. If the participants had lost all
memory for the forgotten pairs, there should have been no difference between
recall of changed and unchanged pairs. However, participants correctly recalled
78% of the unchanged items formerly missed, but only 43% of the changed items.
FIGURE 7.1 The brain structures
involved in the creation and
storage of memories. Prefrontal
regions are responsible for
the creation of memories. The
hippocampus and surrounding
structures in the temporal
cortex are responsible for the
permanent storage of these
memories.
Brain Structures
Prefrontal regions
active when information
is retrieved
Hippocampal regions
(internal) active during
retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 177
This large advantage for unchanged items indicates that participants had retained
some memory of the original paired associates, even though they had been unable
to recall them initially.
Sometimes we can recognize things we cannot recall. So, Nelson (1978) also
conducted a similar test involving recognition rather than recall. Four weeks
after the initial learning phase, participants failed to recognize 31% of the
paired associates they had learned. As in the previous experiment, Nelson asked
participants to relearn the missed items. For half the stimuli, the responses were
changed; for the other half, they were left unchanged. After one relearning trial,
participants recognized 34% of the unchanged items but only 19% of the
changed items. Thus, even when participants failed this sensitive recognition
test, however, it appears that a record of the item was still in memory—the
evidence again being that relearning was better for the unchanged pairs than
for the changed ones.
These experiments do not prove that everything is remembered. They show
only that appropriately sensitive tests can find evidence for remnants of some
memories that appear to have been forgotten. In this chapter, we will discuss
first how memories become less available with time, then some of the factors
that determine our success in retrieving these memories.
Even when people appear to have forgotten memories, there is evidence that
they still have some of these memories stored.
•The Retention Function
The processes by which memories become less available are extremely regular,
and psychologists have studied their mathematical form. Wickelgren did some
of the most systematic research on memory retention functions, and his data
are still used today. In one recognition experiment (Wickelgren, 1975), he presented
participants with a sequence of words to study and then examined the
probability of their recognizing the words after delays ranging from 1 min to
14 days. Figure 7.2 shows performance as a function of delay. The performance
measure Wickelgren used is called d (pronounced d-prime), which is derived
from the probability of recognition. Wickelgren interpreted it as a measure of
memory strength.
Figure 7.2 shows that this measure of memory systematically deteriorates
with delay. However, the memory loss is negatively accelerated—that is, the rate
of change gets smaller and smaller as the delay increases. Figure 7.2b replots
the data as the logarithm of the performance measure versus the logarithm
of delay. Marvelously, the function becomes linear. The log of performance is
a linear function of the log of the delay T; that is,
where A is the value of the function at 1 min [log(1) = 0] and b is the slope
of the function in Figure 7.2b, which happens to be 0.30 in this case.
log d¿ = A - b log T
¿
178 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 178
This equation can be transformed to
where c _ 10A and is 3.62 in this case. That is, these performance measures are
power functions of delay. In a review of research on forgetting, Wixted and
Ebbesen (1991) concluded that retention functions are generally power functions.
This relationship is called the power law of forgetting. Recall from Chapter
6 that there is also a power law of learning: Practice curves are described by
power functions. Both functions are negatively accelerated, but with an important
difference. Whereas practice functions show diminishing improvement
with practice, retention functions show diminishing loss with delay.
A very dramatic example of the negative acceleration in retention function
was produced by Bahrick (1984), who looked at participants’ retention of English-
Spanish vocabulary items anywhere from immediately to 50 years after they
had completed courses in high school and college. Figure 7.3 plots the number
of items correctly recalled out of a total of 15 items as a function of the logarithm
of the time since course completion. Separate functions are plotted for
students who had one, three, or five courses. The data show a slow decay of
knowledge combined with a substantial practice effect. In Bahrick’s data, the
retention functions are nearly flat between 3 and 25 years (as would be predicted
by a power function), with some further drop-off from 25 to 49 years (which is
more rapid than would be predicted by a power function). Bahrick (personal
communication, circa 1993) suspects that this final drop-off is probably related
to physiological deterioration in old age.
There is some evidence that the explanation for these decay functions may
be found in the associated neural processes. Recall from Chapter 6 that longterm
potentiation (LTP) is an increase in neural responsiveness that occurs as a
reaction to prior electrical stimulation.We saw that LTP mirrors the power law
d¿ = cT-b
The Retention Function | 179
(a) (b)
1.0
2.0
Delay, T (days)
3.62T −0.321
Measure of retention (d ʹ)
10 15 20 1
.1
.2
.5
1.0
2.0
3.0
4 7 15 30 50 1 2 4 71014
Minutes
log d ʹ
Days
log T
FIGURE 7.2 Results from Wickelgren’s experiment to discover a memory retention function.
(a) Success at word recognition, as measured by d’, as a function of delay T. (b) The data
in (a) replotted on a log-log scale. (After Wickelgren, 1975. Adapted by permission of the publisher. © 1975
by Memory & Cognition.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 179
of learning. Figure 7.4 illustrates some data from Raymond & Redman (2006)
that shows a decrease in LTP in the rat hippocampus with delay. Plotted there
are three conditions—a control condition that received no LTP, a condition
that received just a single stimulation to induce LTP, and another condition that
received 8 such stimulations.While the level LTP is greater in the condition with
8 stimulations than 1 (a learning effect), both conditions show a drop off with
delay. The smooth lines in the figure represent the best-fitting power functions
180 | Human Memory: Retention and Retrieval
Completion
12
10
14 38
log (time+1) (months)
Five courses
Test score
69 114 175 301 415 596
Three courses
One course
FIGURE 7.3 Results from Bahrick’s experiment that measured participants’ retention over
various time periods of English-Spanish vocabulary items. The number of items correctly
recalled out of a total of 15 items is plotted as a function of the logarithm of the time
since course completion. (From Bahrick, 1984. Reprinted by permission of the publisher. © 1984 by the American
Psychological Association.)
180
1 TBS
8 TBS
No TBS
150
120
90
60
30
−30
Minutes
EPSC (% change)
0 30 60 90 120
FIGURE 7.4 From Raymond and Redman (2006), 1 or 8 theta-burst stimulations (TBS) are
presented at 10 minutes in the experiment. The changes in ESPC (excitatory postsynaptic
current—a measure of LTP) are plotted as a function of time. Also, a control condition is
presented that received no TBS. The two lines represent best-fitting power functions.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 180
and show that maintenance of LTP has the form of a power function. Thus, the
time course of this neural forgetting mirrors the time course of behavioral forgetting,
just as the neural learning function mirrors the behavioral learning
function. In terms of the strength concept introduced in Chapter 6, the assumption
is that the strength of the memory trace decays with time. The data
on LTP suggest that this strength decay involves changes in synaptic strength.
Thus, there may be a direct relationship between the concept of strength
defined at the behavioral level and strength defined at the neural level.
The idea that memory traces simply decay in strength with time is one of
the common explanations of forgetting; it is called the decay theory of forgetting.
We will review one of the major competitors of this theory next: the
interference theory of forgetting.
The strength of a memory trace decays as a power function of the retention
interval.
•How Interference Affects Memory
The discussion to this point might lead one to infer that the only factor affecting
loss of memories is the passage of time. However, it turns out that retention is
strongly impacted by another factor: interfering material. Much of the original
research on interference involved the learning of paired associates. The interest
focused on how the learning of one list of paired associates would affect the
memory for another list. In the typical interference experiment, two critical
groups are defined (Table 7.1). The A–D experimental group learns two lists of
paired associates, the first list designated A–B and the second designated A–D.
These lists are so designated because they share common stimuli (the A terms).
For example, among the pairs that the participants study in the A–B list might
be cat-43 and house-61, and in the A–D list cat-82 and house-37. The C–D control
group also first studies the A–B list but then studies a completely different
second list, designated C–D, which does not contain the same stimuli as the first
list. For example, in the second C–D list, participants might study bone-82 and
cup-37. After learning their respective second lists, both groups are retested for
memory of their first list, in both cases the A–B list. Often, this retention test is
administered after a considerable delay, such as 24 hours or a week. In general,
the A–D group does not do as well as the C–D group with respect
either to rate of learning of the second list or to retention of the
original A–B list (see Keppel, 1968, for a review). Such experiments
provide evidence that learning the A–D list interferes with retention
of the A–B list and causes it to be forgotten more rapidly.
More generally, research has shown that it is difficult to maintain
multiple associations to the same items. It is harder both to learn
new associations to these items and to retain the old ones if new
associations are learned. These results might seem to have rather
dismal implications for our ability to remember information. They
How Interference Affects Memory | 181
TABLE 7.1
Experimental and Control Groups Used
in a Typical Interference Experiment
A–D Experimental C–D Control
Learn A–B Learn A–B
Learn A–D Learn C–D
Test A–B Test A–B
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 181
would appear to imply that it would become increasingly difficult to learn new
information about a concept. Every time we learned a new fact about a friend,
we would be in danger of forgetting an old fact about that person. Fortunately,
there are important additional factors that counteract such interference. Before
discussing these factors, however, we need to examine in more detail the basis for
such interference effects. It turns out that a rather different experimental paradigm
has been helpful in identifying the cause of the interference effects.
Learning additional associations to an item can cause old ones to be forgotten.
The Fan Effect: Networks of Facts
Referring back to the activation equation in Chapter 6, the interference effects
we have studied can be understood in terms of the amount of activation that
spreads to stimulate a memory structure. The basic idea is that when participants
are presented with a stimulus such as cat, activation will spread from it to
all of its associates. There is a limit, however, to the amount of activation that
can spread from a source; the more things associated with that source, the less
the activation that will spread to any particular memory structure. In one
experiment illustrating these ideas (Anderson, 1974a), I asked participants to
memorize 26 facts, of the form A person is in a location. Some persons were
paired with only one location, and some locations with only one person. Other
persons were paired with two locations, and other locations with two persons.
For instance, suppose that participants studied these sentences:
1. The doctor is in the bank. (1-1)
2. The fireman is in the park. (1-2)
3. The lawyer is in the church. (2-1)
4. The lawyer is in the park. (2-2)
Each statement is followed by two numbers, reflecting the number of facts associated
with the person and the location. For instance, sentence 3 is labeled 2-1
because its person occurs in two sentences (sentences 3 and 4) and its location
in one (sentence 3). Participants were drilled on this material until they knew it
well. Before beginning the reaction-time phase, participants were able to recall
all the locations associated with a particular person (e.g., doctor) and all the
persons associated with a particular location (e.g., park). Thus, unlike the interference
experiments reviewed earlier, all the material was memorized, and interest
focused on the speed with which it could be retrieved. After memorizing
the material, participants were presented with sentences and had to judge
whether they recognized the sentences from the study set. These studied sentences
were mixed in with new sentences created by re-pairing people and locations
from the study set. The reaction times involving sentences such as those
listed above are displayed in Table 7.2, which classifies the data as a function of
the number of facts associated with a person and a location. As can be seen,
recognition time increases as a function of both the number of facts studied
about the person and the number of facts studied about the location. The slowdown
in recall is not much more than a hundred milliseconds, but such effects
182 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 182
can add up in situations like taking a time-pressure test. Taking a little more
time to answer each question can mean not finishing the test.
These interference effects can be explained by applying the concept of spreading
activation to propositional networks (described in Chapter 5). Figure 7.5
shows the network representation for sentences 1 through 4. By applying the
activation concept to this representation, we can account for the increase in
reaction time. Consider how the participant might recognize a test sentence
such as A lawyer is in the park. According to the spreading-activation theory,
recognizing this proposition involves the following discrete steps:
1. Presentation of the test sentence activates the representation of the terms
lawyer, in, and park in this network. These are the sources of activation.
2. Activation spreads from these sources along the various paths leading
from the nodes. A critical assumption is that these sources have a fixed
capacity for emitting activation; thus, the more paths there are leading
from a source, the less activation will go down any one path.
3. Activation coming down various paths converges at proposition nodes.
These activations sum to produce an overall level of activation of the
proposition node.
4. The proposition is recognized in an amount of time that is inversely
related to its level of activation.
So, given a structure like that shown in Figure 7.5,
participants should be slower to recognize a fact
involving lawyer and park than one involving doctor
and bank because more paths emanate from
the first set of concepts. That is, in the lawyer and
park case, two paths point from each of the concepts
to the two propositions in which each was
studied, whereas only one path leads from each
of the doctor and bank concepts. The increase in
reaction time related to an increase in the number
of facts associated with a concept is called the fan
effect. It is so named because the increase in reaction
time is related to an increase in the fan
How Interference Affects Memory | 183
TABLE 7.2
Results of an Experiment to Demonstrate the Fan Effect
Mean Recognition Time for Sentences (s)
Number of Sentences about a Specific Person
Number of Sentences
Using a Specific Location 1 2
1 1.11 1.17
2 1.17 1.22
From Anderson, 1974a. Reprinted by permission of the publisher. © 1974 by Cognitive Psychology.
Lawyer Subject Location
Doctor Bank
Park
Relation
In
Relation
Subject
Church Fireman
Subject Relation
Relation
Location
Location Subject
Location
FIGURE 7.5 A network
representation of four sentences
used in the experiment of
Anderson (1974a) demonstrating
how spreading activation works.
The sentences are The doctor is
in the bank; The fireman is in
the park; The lawyer is in the
church; and The lawyer is in
the park.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 183
of facts emanating from the network representation
of the concept.
In an fMRI brain-imaging study, Sohn et al.
(2003) looked at the response in the prefrontal
cortex during the verification of such facts. They
contrasted high-fan sentences (in which the items
appeared in many other sentences) with low-fan
sentences (in which the items appeared in few
sentences). Figure 7.6 compares the hemodynamic
response in the two conditions and shows that
there is greater response in the high-fan condition.
Note that there is a higher hemodynamic response
in the condition where there is lower activation of
the concepts. One might have expected lower concept
activation to map onto weakened hemodynamic
response. However, the prefrontal structures
must work harder to retrieve the memory in conditions of lower activation.
As we will see throughout the later chapters of this text, in which we look
at higher mental processes like problem solving, more difficult conditions are
associated with higher metabolic expenditures, reflecting the greater mental
work required in these conditions.
The more facts associated with a concept, the slower is retrieval of any one
of the facts.
The Interfering Effect of Preexisting Memories
Do such interference effects occur with material learned outside of the laboratory?
As one way to address this question, Lewis and Anderson (1976)
investigated whether the fan effect could be obtained with material the participant
knew before the experiment.We had participants learn fantasy facts about
public figures; for example, Napoleon Bonaparte was from India. Participants
studied from zero to four such fantasy facts about each public figure. After
learning these “facts,” they proceeded to a recognition test phase, in which they
saw three types of sentences: (1) statements they had studied in the experiment;
(2) true facts about the public figures (such as Napoleon Bonaparte was an
emperor); and (3) statements about the public figures that were false both in the
experimental fantasy world and in the real world. Participants had to respond
to the first two types of facts as true and to the last type as false.
Figure 7.7 presents participants’ reaction times in making these judgments as
a function of the number (or fan) of the fantasy facts studied about the person.
Note that reaction time increased with fan for all types of facts. Also note that
participants responded much faster to actual facts than to experimental facts.
The advantage of actual facts can be explained by the observation that these true
facts would be much more strongly encoded in memory than the fantasy facts.
The most important result to note in Figure 7.7 is that the more fantasy facts
participants learned about an individual such as Napoleon Bonaparte, the longer
184 | Human Memory: Retention and Retrieval
0 1
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
2 3 4
Time from stimulus onset (s)
Increase in BOLD response (%)
5 6 7 8
High fan
Low fan
FIGURE 7.6 Differential
hemodynamic response in the
prefrontal cortex during the
retrieval of low-fan and high-fan
sentences. The increase in BOLD
response is plotted against the
time from stimulus onset. (After
Sohn et al., 2003. Adapted by permission
of the publisher. © 2003 by the National
Academy of Sciences.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 184
they took to recognize a fact that they already knew about the individual; for example,
Napoleon Bonaparte was an emperor. Thus, we can produce interference
with preexperimental material. For further research on this topic, see Peterson
and Potts (1982).
Material learned in the laboratory can interfere with material learned
outside of the laboratory.
The Controversy over Interference and Decay
We have seen two mechanisms that can produce forgetting: decay of trace
strength and interference from other memories. There has been some speculation
in psychology that what appears to be decay may really reflect interference. That
is, the reason memories appear to decay over a retention interval is that they are
interfered with by additional memories that the participants have learned. This
speculation led to research that studied whether material was better retained over
an interval during which participants slept or one during which they were awake.
The reasoning was that there would be fewer interfering memories learned during
sleep. Ekstrand (1972) reviewed a great deal of research consistent with the
conclusion that less is forgotten during the period of sleep.However, it seems that
the critical variable is not sleep but rather the time of day during which material
is learned. Hockey, Davies, and Gray (1972) found that participants better remembered
material that they learned at night, even if they were kept up during
the night and slept during the day. It seems that early evening is the period of
How Interference Affects Memory | 185
Number of fantasy facts learned (fan)
Reaction time (ms)
1300
1400
1500
1600
1700
1800
1900
2000
2100
2200
2300
2400
1 2 3 4
False
Experimental true
Actual true
FIGURE 7.7 Results from Lewis and Anderson’s study to
investigate whether the fan effect could be obtained using
material participants knew before the experiment. The task
was to recognize true and fantasy facts about a public
figure and to reject statements that contained neither true
nor fantasy facts. Participants’ reaction times in making
these judgments are plotted as a function of the number
(or fan) of the fantasy facts studied. The time participants
took to make all three judgments increased as they learned
more fantasy facts. (After Lewis and Anderson, 1976. Adapted by
permission of the publisher. © 1976 by Cognitive Psychology.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 185
highest arousal (at least for typical undergraduate participants) and that retention
is best for material learned in a high arousal state. See Anderson (2000) for
a review of the literature on effects of time of day.
There has been a long-standing controversy in psychology about whether
retention functions, such as those illustrated in Figures 7.2 and 7.3, reflect decay
in the absence of any interference or reflect interference from unidentified
sources. Objections have been raised to decay theories because they do not
identify the psychological factors that produce the forgetting but rather assert
that forgetting occurs spontaneously with time. It may be possible, however,
that there is no explanation of decay at the purely psychological level. The
explanation may be physiological, as we saw with respect to the LTP data (see
Figure 7.4). Thus, it seems that the best conclusion, given the available data, is
that both interference and decay effects contribute to forgetting.
Forgetting results both from decay in trace strength and from interference
from other memories.
Redundancy Protects Against Interference
There is a major restriction on the situations in which interference effects are
seen: Interference occurs only when one is learning multiple pieces of information
that have no intrinsic relationship to one another. In contrast, interference
does not occur when the information is somewhat redundant. An experiment by
Bradshaw and Anderson (1982) illustrates the contrasting effects of redundant
versus irrelevant information. These researchers looked at participants’ ability
to learn some little-known information about famous people. In the single condition,
they had participants study just one fact:
Newton became emotionally unstable and insecure as a child.
In the irrelevant condition, they had participants learn a target fact plus two
unrelated facts about the individual:
Locke was unhappy as a student at Westminster.
plus
Locke felt fruits were unwholesome for children.
Locke had a long history of back trouble.
In the relevant condition, participants learned two additional facts that were
causally related to the target fact:
Mozart made a long journey from Munich to Paris.
plus
Mozart wanted to leave Munich to avoid a romantic entanglement.
Mozart was intrigued by musical developments coming out of Paris.
Participants were tested for their ability to recall the target facts immediately
after studying them and after a week’s delay. They were presented with names
such as Newton, Mozart, and Locke and asked to recall what they had studied.
Table 7.3 displays the results. Comparing the irrelevant condition with the single
condition, we see the standard interference effect: Recall was worse when there
186 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 186
were more facts to be learned about an item.
However, the conclusion is quite different when
we compare the relevant condition to the single
condition. Here, particularly at a week’s delay,
recall was better when the participant had to
learn additional facts causally related to the
target facts.
To understand why the effects of interference
are eliminated or even reversed when
there is redundancy among the materials to be
learned requires that we move on to discussing
the retrieval process and, in particular, the role
of inferential processes in retrieval.
Learning redundant material does not interfere with a target memory
and may even facilitate the target memory.
•Retrieval and Inference
Often, when people cannot remember a particular fact, they are able to retrieve
related facts and so infer the target fact on the basis of the related facts. For example,
in the case of the Mozart facts just discussed, even if the participants
could not recall that Mozart made a long journey from Munich to Paris, if they
could retrieve the other two facts, they would be able to infer this target fact.
There is considerable evidence that people make such inferences at the time of
recall. They seem unaware that they are making inferences but rather think that
they are recalling what was actually studied.
Bransford, Barclay, and Franks (1972) reported an experiment that demonstrates
how inference can lead to incorrect recall. They had participants study
one of the following sentences:
1. Three turtles rested beside a floating log, and a fish swam beneath them.
2. Three turtles rested on a floating log, and a fish swam beneath them.
Participants who had studied sentence 1 were later asked whether they had
studied this sentence:
3. Three turtles rested beside a floating log, and a fish swam beneath it.
Not many participants thought they had studied this sentence. Participants
who had studied sentence 2 were tested with
4. Three turtles rested on a floating log, and a fish swam beneath it.
The participants in this group judged that they had studied sentence 4 much
more often than participants in the other group judged that they had studied
sentence 3. Of course, sentence 4 is implied by sentence 2, whereas sentence 3 is
not implied by sentence 1. Thus, participants thought that they had actually
studied what was implied by the studied material.
Retrieval and Inference | 187
TABLE 7.3
The Contrasting Effects of Relevant and Irrelevant Information
Recall (%)
Condition Immediate Recall Recall at 1 Week
Single fact 92 62
Irrelevant facts 80 45
Relevant facts 94 73
From Bradshaw & Anderson, 1982. Reprinted by permission of the publisher. © 1982
by the Journal of Verbal Learning and Verbal Behavior.
Anderson7e_Chapter_07.qxd 8/21/09 7:52 PM Page 187
A study by Sulin and Dooling (1974) illustrates how inference can bias participants’
memory for a text. They asked participants to read the following passage:
Carol Harris’s Need for Professional Help
Carol Harris was a problem child from birth. She was wild, stubborn, and violent.
By the time Carol turned eight, she was still unmanageable. Her parents
were very concerned about her mental health. There was no good institution
for her problem in her state. Her parents finally decided to take some action.
They hired a private teacher for Carol.
A second group of participants read the same passage, except that the name
Helen Keller was substituted for Carol Harris.1 A week after reading the passage,
participants were given a recognition test in which they were presented with a
sentence and asked to judge whether it had occurred in the passage they read
originally. One of the critical test sentences was She was deaf, dumb, and blind.
Only 5% of participants who read the Carol Harris passage accepted this
sentence, but a full 50% of the participants who read the Helen Keller version
thought they had read the sentence. This is just what we would expect. The second
group of participants had elaborated the story with facts they knew about
Helen Keller. Thus, it seemed reasonable to them at test that this sentence had
appeared in the studied material, but in this case their inference was wrong.
We might wonder whether an inference such as She was deaf, dumb, and blind
was made while the participant was studying the passage or only at the time of
the test. This is a subtle issue, and participants certainly do not have reliable intuitions
about it. However, a couple of techniques seem to yield evidence that the
inferences are being made at test. One method is to determine whether the inferences
increase in frequency with delay.With delay, participants’ memory for the
studied passage should deteriorate, and if they are making inferences at test, they
will have to do more reconstruction, which in turn will lead to more inferential
errors. Both Dooling and Christiaansen (1977) and Spiro (1977) found evidence
for increased inferential intrusions with increased delay of testing.
Dooling and Christiaansen used another technique with the Carol Harris
passage to show that inferences were being made at test. They had the participants
study the passage and then told them a week later, just before test, that
Carol Harris really was Helen Keller. In this situation, participants also made
many inferential errors, accepting such sentences as She was deaf, dumb, and
blind. Because they did not know that Carol Harris was Helen Keller until test,
they must have made the inferences at test. Thus, it seems that participants do
make such reconstructive inferences at time of test.
In trying to remember material, people will use what they can remember
to infer what else they might have studied.
Plausible Retrieval
In the foregoing analysis, we spoke of participants as making errors when they
recalled or recognized facts that were not explicitly presented. In real life,
188 | Human Memory: Retention and Retrieval
1 Helen Keller was well known to participants of the time, famous for overcoming both deafness and blindness
as a child.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 188
however, such acts of recall often would not be regarded as errors but as intelligent
inferences. Reder (1982) has argued that much of recall in real life involves
plausible inference rather than exact recall. For instance, in deciding that Darth
Vader was evil in Star Wars, a person does not search memory for the specific
proposition that Darth Vader was evil, although it may have been directly
asserted in the movie. The person infers that Darth Vader was evil from memories
about the Stars Wars movies.
Reder has demonstrated that people will display very different behavior,
depending on whether they are asked to engage in exact retrieval or plausible
retrieval. She had participants study passages such as the following:
The heir to a large hamburger chain was in trouble. He had married a lovely
young woman who had seemed to love him. Now he worried that she had
been after his money after all. He sensed that she was not attracted to him.
Perhaps he consumed too much beer and French fries. No, he couldn’t give up
the fries. Not only were they delicious, he got them for free.
Then she had participants judge sentences such as
1. The heir married a lovely young woman who had seemed to love him.
2. The heir got his French fries from his family’s hamburger chain.
3. The heir was very careful to eat only healthy food.
The first sentence was studied; the second was not studied, but is plausible; and
the third was neither studied nor plausible. Participants in the exact condition
were asked to make exact recognition judgments, in which case they were to
accept the first sentence and reject the second two. Participants in the plausible
condition were to judge whether the sentence
was plausible given the story, in which case they
were to accept the first two and reject the last.
Reder tested participants immediately after
studying the story, 20 min later, or 2 days later.
Reder was interested in judgment time for
participants in the two conditions, exact versus
plausible. Figure 7.8 shows the results from her
experiment, plotted as the average judgment
times for the exact condition and the plausible
condition as a function of delay. As might be
expected, participants’ response times increased
with delay in the exact condition. However, the
response times actually decreased in the plausible
condition. They started out slower in the
plausible condition than in the exact condition,
but this trend was reversed after 2 days. Reder
argues that participants respond more slowly in
the exact condition because the exact traces are
getting weaker. A plausibility judgment, however,
does not depend on any particular trace
and so is not similarly vulnerable to forgetting.
Retrieval and Inference | 189
none 20 min
Reaction time (s)
Delay
2.40
2.60
3.00
2.80
3.20
2 days
Exact recall
Plausible retrieval
FIGURE 7.8 Results from Reder’s experiment showing that people
display different behavior depending on whether they are asked to
engage in exact retrieval or plausible retrieval of information. The
time required to make exact versus plausible recognition judgments
of sentences is plotted as a function of delay since study of a story.
(From Reder, 1982. Reprinted by permission of the publisher. © 1982 by Psychological Review.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 189
Participants respond faster in the plausible condition with delay because they
no longer try to use inefficient exact retrieval. Instead they use plausibility,
which is faster.
Reder and Ross (1983) compared exact versus plausible judgments in another
study. They had participants study sentences such as
Alan bought a ticket for the 10:00 a.m. train.
Alan heard the conductor call, “All aboard.”
Alan read a newspaper on the train.
Alan arrived at Grand Central Station.
They manipulated the number of sentences that participants had to study
about a particular person such as Alan. Then they looked at the times participants
took to recognize sentences such as
1. Alan heard the conductor call, “All aboard.”
2. Alan watched the approaching train from the platform.
3. Alan sorted his clothes into colors and whites.
In the exact condition, participants had to judge whether the sentence had
been studied. So, given the foregoing material, participants would accept test
sentence 1 and reject test sentences 2 and 3. In the plausible condition, participants
had to judge whether it was plausible that Alan was involved in the activity,
given what they had studied. Thus, participants would accept sentences 1 and 2
and reject sentence 3.
In the exact condition, Reder and Ross found that participants’ response
times increased when they had studied more facts about Alan. This is basically
a replication of the fan effect discussed earlier in the chapter. In the plausible
condition, however, participants’ response times decreased when they had
learned more facts about Alan. The more facts they knew about Alan, the more
ways there were to judge a particular fact to be plausible. Thus, plausibility
judgment did not have to depend on retrieval of a particular fact.
People will often judge what plausibly might be true rather than try
to retrieve exact facts.
The Interaction of Elaboration and Inferential Reconstruction
In Chapter 6, we discussed how people tend to display better memories if they
elaborate the material being studied.We also discussed how semantic elaborations
are particularly beneficial. Such semantic elaborations should facilitate the process
of inference by providing more material from which to infer. Thus, we expect
elaborative processing to lead to both an increased recall of what was studied and
an increase in the number of inferences recalled. An experiment by Owens,
Bower, and Black (1979) confirms this prediction. Participants studied a story that
followed the principal character, a college student, through a day in her life: making
a cup of coffee in the morning, visiting a doctor, attending a lecture, shopping
for groceries, and attending a party. The following is a passage from the story:
Nancy went to see the doctor. She arrived at the office and checked in with the
receptionist. She went to see the nurse, who went through the usual procedures.
190 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 190
Then Nancy stepped on the scale and the nurse recorded her weight. The doctor
entered the room and examined the results. He smiled at Nancy and said,
“Well, it seems my expectations have been confirmed.”When the examination
was finished,Nancy left the office.
Two groups of participants studied the story. The only difference between the
groups was that the theme group had read the following additional information
at the beginning:
Nancy woke up feeling sick again and she wondered if she really were pregnant.
How would she tell the professor she had been seeing? And the money
was another problem.
College students who read this additional passage characterized Nancy as an unmarried
student who is afraid she is pregnant as a result of an affair with a college
professor. Participants in the neutral condition, who had not read this opening
passage, had no reason to suspect that there was anything special about Nancy.
We would expect participants in the theme condition to make many more themerelated
elaborations of the story than participants in the neutral condition.
Participants were asked to recall the story 24 hours after studying it. Those
in the theme condition introduced a great many more inferences that had not
actually been studied. For instance, many participants reported that the doctor
told Nancy she was pregnant. Intrusions of this variety are expected if participants
reconstruct a story on the basis of their elaborations. Table 7.4 reports
some of the results from the study. As can be seen, many more inferences were
added in recall for the theme condition than for the neutral condition. A second
important observation, however, is that participants in the theme condition
also recalled more of the propositions they had actually studied. Thus,
because of the additional elaborations these participants made, they were able
to recall more of the story.
We might question whether participants really benefited from their elaborations,
because they also misrecalled many things that did not occur in the story.
However, it is wrong to characterize the intruded inferences as errors. Given the
theme information, participants were perfectly right to make inferences and
to recall them. In a nonexperimental setting, such as recalling information for
an exam, we would expect these participants to recall such inferences as easily
as material they had actually read.
Retrieval and Inference | 191
TABLE 7.4
The Interactive Effects of Elaboration and Inference
Number of Propositions Recalled
Theme Condition Neutral Condition
Studied propositions 29.2 20.3
Inferred propositions 15.2 3.7
After Owens et al. (1979). Adapted by permission of the publisher. © 1979 by Memory &
Cognition.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 191
When participants elaborate on material while studying it, they tend to
recall more of what they studied and also tend to recall the inferences that
they did not study but made themselves.
Eyewitness Testimony and the False-Memory Controversy
The ability to elaborate on and make inferences from information, both while it
is being studied and when our recall is being tested, is essential to using our
memory successfully in everyday life. Inferences made while studying material
allow us to extrapolate from what we actually heard and saw to what is probably
true.When we hear that someone found out in a doctor’s visit that she was
pregnant, it is a reasonable inference that she was told so by the doctor. So such
inferences usually lead to a much more coherent and accurate understanding of
the world. There are circumstances, however, in which we need to be able to
separate what we actually saw and heard from our inferences. The difficulty of
doing so can lead to harmful false memories; the Gargoil example above is only
the tip of the iceberg.
One situation in which it is critical to separate inference from actual experience
is in eyewitness testimony. It has been shown that eyewitnesses are often
inaccurate in the testimony they give, even though jurors accord it high weight.
One reason for the low accuracy is that people confuse what they actually
observed about an incident with what they heard from other sources. Loftus
192 | Human Memory: Retention and Retrieval
this commercial, all 15 of his participants recalled that
“gargling with Gargoil Antiseptic helps prevent colds,”
although this assertion was clearly not
made in the commercial. The Federal Trade
Commission explicitly forbids advertisers
from making false claims, but does the
Listerine ad make a false claim? In a
landmark case, the courts ruled against
Warner-Lambert, makers of Listerine, for
implying false claims in this commercial.
As a corrective action the court ordered
Warner-Lambert to include in future advertisements
the disclaimer “contrary to
prior advertising, Listerine will not help
prevent colds or sore throats or lessen
their severity.” They were required to
continue this disclaimer until they had
expended an amount of money equivalent to their prior
10 years of advertisement.
Implications
How have advertisers used knowledge of cognitive psychology?
Advertisers often capitalize on our tendency to embellish
what we hear with plausible inferences. Consider the
following portion of an old Listerine
commercial:
“Wouldn’t it be great,” asks the mother, “if
you could make him cold proof? Well, you
can’t. Nothing can do that.” [Boy sneezes.]
“But there is something that you can do that
may help. Have him gargle with Listerine
Antiseptic. Listerine can’t promise to keep
him cold free, but it may help him fight off
colds. During the cold-catching season, have
him gargle twice a day with full-strength
Listerine. Watch his diet, see he gets plenty
of sleep, and there’s a good chance he’ll have
fewer colds, milder colds this year.”
A verbatim text of this commercial,
with the product name changed to “Gargoil,” was used in
an experiment conducted by Harris (1977). After hearing
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 192
Retrieval and Inference | 193
(1975, 1996; Loftus, Miller, & Burns, 1978) showed that subsequent information
can change a person’s memory of an observed event. In one study, for
instance, Loftus asked participants who had witnessed a traffic accident about
the car’s speed when it passed a Yield sign. Although there was no Yield sign,
many participants subsequently remembered having seen one, confusing the
question they were asked with what they had actually seen. Another interesting
example involves the testimony given by John Dean about events in the Nixon
White House during the Watergate cover-up (Neisser, 1981). After Dean testified
about conversations in the Oval Office, it was discovered that Nixon had
recorded these conversations. Although Dean was substantially accurate in gist,
he confused many details, including the order in which these conversations
took place.
Another case of memory confusion that has produced a great deal of recent
notoriety concerns the controversy about the so-called false-memory
syndrome. This controversy involves cases where individuals claim to recover
memories of childhood sexual abuse that they had suppressed (Schacter,
2001).Many of these recovered memories occur in the process of therapy, and
some memory researchers have questioned whether these recovered memories
ever happened and hypothesized that they might have been created by
the strong suggestions of the therapists. For instance, one therapist said to
patients, “You know, in my experience, a lot of people who are struggling with
many of the same problems you are, have often had some kind of really
painful things happen to them as kids—maybe they were beaten or molested.
And I wonder if anything like that ever happened to you?” (Forward & Buck,
1988, p. 161). Given the evidence we have reviewed about how people will put
information together to make inferences about what they should remember,
one could wonder if the patients who heard this might remember what did
not happen.
A number of researchers have shown that it is possible to create false memories
by use of suggestive interview techniques. For instance, Loftus and Pickerall
(1995) had adult participants read four stories from their childhood written by
an older relative—three were true, but one was a false story about being lost in
the mall at age 5. After reading the story, about 25% of participants claimed to
remember the event of being lost in a mall. Ceci, Loftus, Leichtman, and Bruck
(1995) were similarly able to create false memories in 3- to 6-year-old children.
The process by which we distinguish between memory and imagination is quite
fragile, and it is easy to become confused about the source of information. Of
course, it would not be ethical to try to plant false memories about something
so traumatic as sexual abuse, and there are questions (e.g., Pope, 1996) about
whether it is possible to create false memories as awful as those involving childhood
sexual abuse.
There is an intense debate about how much credibility should be given to
recovered memories of childhood abuse. Although there is a temptation to want
to come to either the simple conclusion that all reports of recovered memories
of abuse should be believed or the conclusion that all should be discounted, it
does not appear to be so simple. There are cases of recovered memories of abuse
that seem to have strong documentation (Sivers, Schooler, and Freyd, 2002) and
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 193
there are cases where the alleged victims of such abuse have subsequently
retracted and said they were misled in their memories (Schacter, 2001).
Serious errors of memory can occur because people fail to separate what they
actually experienced from what they inferred, imagined, or were told.
False Memories and the Brain
Recently, researchers have developed the ability to explore the neural basis of
false memories. They have used a less exotic example than getting people to
believe they had been lost in the mall. In a Deese-Roediger-McDermott
paradigm originally invented by Deese (1959) and elaborated by Roediger and
McDermott (1995), participants study lists of words. One list might contain
thread, pin, eye, sewing, sharp, point, prick, thimble, haystack, thorn, hurt, injection,
syringe, cloth, knitting; a second list might contain bed, rest, awake, tired,
dream, wake, snooze, blanket, doze, slumber, snore, nap, peace, yawn, drowsy. In a
later test, participants are shown a series of words and must decide whether
they have studied those words. There are three types of words:
True (e.g., sewing, awake)
False (e.g., needle, sleep)
New (e.g., door, candy)
The true items were in the lists; the false ones are strongly associated with items
in the list but were not on the list; and the new ones are unrelated to items on
the list. Participants accept most of the true items and reject most of the new
ones, but they have difficulty in rejecting the false items. In one study, Cabeza,
Rao,Wagner, Mayer, and Schacter (2001) found that 88% of the true items and
only 12% of the new items were accepted, but 80% of the false items were also
accepted—almost as many as the true items.
Cabeza et al. examined the activation patterns that these different types of
words produced in the cortex. Figure 7.9 illustrates such activation profiles in
the two hippocampal structures. In the hippocampus proper, true words and
false words produced almost identical fMRI responses, which were stronger
than the responses produced by the new words. Thus, these hemodynamic responses
appear to match up pretty well with the behavioral data where participants
cannot discriminate between true items and false items. However, in the
parahippocampal gyrus, an area just adjacent to the hippocampus, both false
and new items produced identical weak responses, whereas the true items produced
a larger hemodynamic response. The parahippocampus is more closely
connected to sensory regions of the brain, and Cabeza et al. suggested that it
retains the original sensory experience of seeing the word, whereas the hippocampus
maintains a more abstract representation and this is why true items
produce a larger hemodynamic response. Schacter (e.g., Dodson & Schacter,
2002a, 2000b) has suggested that people can be trained to pay more attention
to these distinctive sensory features and so improve their resistance to false
memories. As one application, distinctiveness training can be used to help
194 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 194
elderly patients who have particular difficulty with false memories. For instance,
older adults sometimes find it hard to remember whether they have seen
something or just imagined it (Henkel, Johnson, & DeLeonardis, 1998).
The hippocampus responds to false memories with as high activation as it
responds to true memories and so fails to discriminate between what was
experienced and what was imagined.
Retrieval and Inference | 195
Hippocampus
Image number
True
False
New
(a)
Normalised MR signal
−1.5
−1.0
−0.5
0.5
1.0
2 3 4 5
True
False
New
(b)
Normalised MR signal
−1.5
−1.0
−0.5
0.5
1.0
Image number
Parahippocampal gyrus
1 2 3 4 5
FIGURE 7.9 Results from the fMRI study by Cabeza et al. of activation patterns produced
by participants’ judgments of true, false, and new items on a previously learned word list.
(a) Bilateral hippocampal regions were more activated for true and false items than for
new items, with no difference between the activations for true and false items. (b) A left
posterior parahippocampal region (the parahippocampal gyrus) was more activated for
true items than for false and new items, with no difference between the activations for
false and new items. (After Cabeza et al., 2001. Adapted by permission of the publisher. © 2001 by the National
Academy of Sciences.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 195
•Associative Structure and Retrieval
The spreading activation theory described in Chapter 6 implies that we can
improve our memory by providing prompts that are closely associated with a
particular memory. You may find yourself practicing this technique when you
try to remember the name of an old classmate. You may prompt your memory
with names of other classmates or memories of things you did with that
classmate. Often, the name does seem to come to mind as a result of such efforts.
An experiment by Tulving and Pearlstone (1966) provides one demonstration
of this technique. They had participants learn lists of 48 words that contained
categories such as dog, cat, horse, and cow, which form a domestic mammal
category. Participants were asked to try to recall all the words in the list. They
displayed better memory for the word lists when they were given prompts such
as mammal, which served to cue memory for members of the categories.
The Effects of Encoding Context
Among the cues that can become associated with a memory are cues from the
context in which the memory was formed. If such contextual cues could be
revived when the memory was being tested, a person would have additional
ways to reactivate the target memory. There is ample evidence that context can
greatly influence memory. This section will review some of the ways in which
this influence can occur. Context effects are often referred to as encoding effects
because the context is affecting what is encoded into the memory trace that
records the event.
Smith, Glenberg, and Bjork (1978) performed an experiment that showed
the importance of physical context. In their experiment, participants learned
two lists of paired associates on different days and in different physical settings.
On day 1, participants learned the paired associates in a windowless room in a
building near the University of Michigan campus. The experimenter was neatly
groomed, dressed in a coat and a tie, and the paired associates were shown on
slides. On day 2, participants learned the paired associates in a tiny room with
windows on the main campus. The experimenter was dressed sloppily in a flannel
shirt and jeans (it was the same experimenter, but some participants did not
recognize him) and presented the paired associates via a tape recorder. A day
later, participants were tested for their recall of half the paired associates in one
setting and half in the other setting. They could recall 59% of the list learned in
the same setting as they were tested, but only 46% of the list learned in the
other setting. Thus, it seems that recall is better if the context during test is the
same as the context during study.
Perhaps the most dramatic manipulation of context was performed by
Godden and Baddeley (1975). They had divers learn a list of 40 unrelated words
either on the shore or 20 feet under the sea. The divers were then asked to recall
the list either in the same environment or in the other environment. Figure 7.10
displays the results of this study. Participants clearly showed superior memory
when they were asked to recall the list in the same environment in which
they studied it. So, it seems that contextual elements do get associated with
196 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 196
memories and that memory is improved when
participants are provided with these contextual
elements when being tested. This result actually has
serious implications for diver instruction, because
most of the instructions are given on dry land but
must be recalled under water.
The degree to which such contextual effects are
obtained has proved to be quite variable from experiment
to experiment (Roediger & Guynn, 1996).
Fernandez and Glenberg (1985) reported a number
of failures to find any context dependence; and
Saufley, Otaka, and Bavaresco (1985) reported a
failure to find such effects in a classroom situation.
Eich (1985) argued that the magnitude of such contextual
effects depends on the degree to which the
participant integrates the context with the memories.
In his experiment, he read lists to two groups
of participants. In one condition, participants were
instructed to imagine the referents of the nouns
alone (e.g., imagine a kite in the sky); in the other,
they were asked to imagine the referents integrated
with the context (e.g., imagine a kite on the table in the corner of the room).
Eich found a much larger effect of a change of context when participants were
instructed to imagine the referent integrated with the context.
Bower, Monteiro, and Gilligan (1978) showed that emotional context can
have the same effect as physical context. They instructed participants to learn
two lists. For one list, they hypnotically induced a positive state by having participants
review a pleasant episode in their lives; for the other, they hypnotically
induced a negative state by having participants review a traumatic event. A later
recall test was given under either a positive or a negative emotional state (again
hypnotically induced). Better memory was obtained when the emotional state
at test matched the emotional state at study.2
Not all research shows such mood-dependent effects. For instance, Bower
and Mayer (1985) failed to replicate the Bower et al. (1978) result. Eich and
Metcalfe (1989) found that mood-dependent effects tend to be obtained only
when participants integrate their memories at study with the mood information.
Thus, like the effects of physical context, mood-dependent effects occur
only in special study situations.
Perhaps a more robust effect is mood congruence, the fact that it is easier to
remember happy memories when one is in a happy state and sad memories when
Associative Structure and Retrieval | 197
Dry
10
11
12
13
Learning environment
Mean number of words recalled
Wet
Wet recall environment
Dry recall environment
FIGURE 7.10 Results of a study by Godden and Baddeley to
investigate the effects of context on participants’ recall of words.
The mean number of words recalled is plotted as a function of
the environment in which learning took place. Participants recalled
word lists better in the same environment in which they were
learned. (After Godden & Baddeley, 1975. Adapted by permission of the publisher.
© 1975 by the British Journal of Psychology.)
2 As an aside, it is worth commenting that, despite popular reports, the best evidence is that hypnosis per se
does nothing to improve memory (see Hilgard, 1968; M. Smith, 1982; Lynn, Lock, Myers, & Payne, 1997),
although it can help memory to the extent that it can be used to re-create the contextual factors at the time
of test.However,much of a learning context can also be re-created by nonhypnotic means, such as through
free association about the circumstances of the event to be remembered (e.g., Geiselman, Fisher,Mackinnon,
& Holland, 1985).
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 197
one is in a sad state.Mood congruence is an effect of
the content of the memories rather than the emotional
state of the participant during study. For instance,
Teasdale and Russell (1983) had participants
learn a list of positive, negative, and neutral words in
a normal state. Then, at test, they induced either
positive or negative states. Their results, illustrated
in Figure 7.11, show that participants recalled more
of the words that matched their mood at test.When
a particular mood is created at test, elements of that
mood will prime memories that share these elements.
Thus, mood elements can prime both memories
whose content matches the mood, as in the
Teasdale and Russell experiment, and memories that
have such mood elements integrated as part of the
study procedure (as in Eich & Metcalfe, 1989).
A related phenomenon is state-dependent
learning. People find it easier to recall information if
they can return to the same emotional and physical
state they were in when they learned the information.
For instance, it is often casually claimed that
when heavy drinkers are sober, they are unable to
remember where they hid their alcohol when drunk, and when drunk they are
unable to remember where they hid their money when sober. In fact, some experimental
evidence does exist for this state dependency of memory with respect to
alcohol, but the more important factor seems to be that alcohol has a general debilitating
effect on the acquisition of information (Parker, Birnbaum, & Noble,
1976). Marijuana has been shown to have similar state-dependent effects. In one
experiment (Eich, Weingartner, Stillman, & Gillin, 1975), participants learned a
free-recall list after smoking either a marijuana cigarette or an ordinary cigarette.
Participants were tested 4 hours later—again after smoking either a marijuana
cigarette or a regular cigarette. Table 7.5 shows the results from this study. Two effects
were seen, both of which are typical of research on the effects of psychoactive
drugs on memory. First, there is a state-dependent effect reflected by better recall
198 | Human Memory: Retention and Retrieval
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
Mood state at test
Words recalled
Elation Depression
Positive words
Neutral words
Negative words
FIGURE 7.11 Results from
Teasdale and Russell’s study of
mood congruence. The number
of words recalled from a
previously studied list is plotted
against the mood state at test.
Participants recalled more of
the words that matched their
mood at test. (From Teasdale &
Russell, 1983. Reprinted by permission
of the publisher. © 1983 by the British
Psychological Society.)
TABLE 7.5
State-Dependent Learning: The Effects of Drugged State at Study and at Test
At Test (% correct)
Ordinary Marijuana
At Study Cigarette Cigarette Average
Ordinary cigarette 25 20 23
Marijuana cigarette 12 23 18
From Eich et al. (1975). Reprinted by permission of the publisher. © 1975 by the Journal of Verbal Learning
and Verbal Behavior.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 198
when the state at test matched the state at study. Second, there is an overall higher
level of recall when the material was studied in a nonintoxicated state.
People show better memory if their external context and their internal states
are the same at the time of study and the time of the test.
The Encoding-Specificity Principle
Memory for material to be learned can also depend heavily on the context of
other material to be learned in which it is embedded. Consider a recognitionmemory
experiment by D. M. Thompson (1972). Thompson’s participants
studied pairs of words such as sky blue. Participants were told that they were
responsible only for the second item of the pair—in this case blue; the first word
represented context. Later, they were tested by being presented with either blue or
sky blue. In both cases, they were asked whether they had originally seen blue. In
the single-word case, they recognized blue 76% of the time, whereas in the pair
condition their recognition rate was 85%. Thus, even though they were tested
only for blue, their memories were better in the context of the other word.
A series of experiments (e.g., Tulving & Thompson, 1973;Watkins & Tulving,
1975) has dramatically illustrated how memory for a word can depend on how
well the test context matches the original study context. There were three
phases to the experiment:
1. Original study:Watkins and Tulving had participants learn pairs of
words such as train-black and told them that they were responsible only
for the second word, referred to as the to-be-remembered word.
2. Generate and recognize: Participants were given words such as white and
asked to generate four free associates to the word. So, a participant might
generate snow, black, wool, and pure. The stimuli for the task were chosen
to have a high probability of eliciting the to-be-remembered word. For
instance, white has a high probability of eliciting black. Participants were
then told to indicate which of the four associates they generated was the
to-be-remembered word they had studied in the first phase. In cases
where the to-be-remembered word was generated, participants correctly
chose it only 54% of the time. Because participants were always forced to
indicate a choice, some of these correct choices must have been lucky
guesses. Thus, true recognition was even lower than 54%.
3. Cued recall: Participants were presented with the original context words
(e.g., train) and asked to recall the to-be-remembered words (i.e., black).
Participants recalled 61% of the words—higher than their recognition
rate without any correction for guessing.Moreover,Watkins and Tulving
found that 42% of the words recalled had not been recognized earlier
when the participants gave them as free associates.3
Recognition is usually superior to recall. Thus, we would expect that if participants
could not recognize a word, they would be unable to recall it. Usually,
Associative Structure and Retrieval | 199
3 A great deal of research has been done on this phenomenon. For a review, read Nilsson and Gardiner (1993).
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 199
we expect to do better on a multiple-choice test than on a recall-the-answer
test. Experiments such as the one just described provided very dramatic reversals
of such standard expectations. The results can be understood in terms of
the similarity of the test context to the study context. The test context with the
word white and its associates was quite different from the context in which
black had originally been studied. In the cued-recall test context, by contrast,
participants were given the original context (train) with which they had studied
the word. Thus, if the contextual factors are sufficiently weighted in favor of
recall, as they were in these experiments, recall can be superior to recognition.
Tulving interprets these results as illustrating what he calls the encodingspecificity
principle: The probability of recalling an item at test depends on the
similarity of its encoding at test to its original encoding at study.
People show better word memory if the words are tested in the context of the
same words with which they were studied.
•The Hippocampal Formation and Amnesia
In Chapter 6, we discussed the fictional character Leonard, who suffered amnesia
resulting from hippocampal damage. A large amount of evidence points to
the great importance of the hippocampal formation, a structure embedded
within the temporal cortex, for the establishment of permanent memories.
In animal studies (typically rats or primates; for a review, see Eichenbaum &
Bunsey, 1995; Squire, 1992), it has been shown that lesions in the hippocampal
formation produce severe impairments in the ability to learn new associations,
particularly those that require remembering combinations or configurations of
elements. Damage to the hippocampal area also produces severe amnesia
(memory loss) in humans. One of the most studied amnesic patients is known
as HM.4 In 1953 when he was 27 years old, large parts of his temporal lobes
were surgically removed to cure epilepsy. He had one of the most profound
amnesias ever recorded and was studied for decades. He had normal memories
of his life up to the age of 16 but forgot most of 11 years before the surgery.
Moreover, he was almost totally unable to remember new events. He appeared
in many ways as a normal person with a clear self-identity, but his identity was
largely as the person he was when he was 16 where his memories stopped
(although he realized he is older and had learned some general facts about the
world). His surgical operation involved complete removal of the hippocampus
and surrounding structures, and this is considered the reason for his profound
memory deficits (Squire, 1992).
Only rarely is there a reason for surgically removing the hippocampal formation
from humans. However, for various reasons, humans can suffer severe
damage to this structure and the surrounding temporal lobe. One common
200 | Human Memory: Retention and Retrieval
4 Henry Gustav Molaison recently died at the age of 82. His obituary is available at http://www.nytimes
.com/2008/12/05/us/05hm.html.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 200
cause is a severe blow to the head, but other frequent causes include brain
infections (such as encephalitis) and chronic alcoholism, which can result in a
condition called Korsakoff syndrome. Such damage can result in two types of
amnesia: retrograde amnesia, which refers to the loss of memory for events
that occurred before the injury, and anterograde amnesia, which refers to an
inability to learn new things.
In the case of a blow to the head, the amnesia often is not permanent but
displays a particular pattern of recovery. Figure 7.12 displays the pattern of
recovery for a patient who was in a coma for 7 weeks following a closed head
injury. Tested 5 months after the injury, the patient showed total anterograde
amnesia—he could not remember what had happened since the injury. He
also displayed total retrograde amnesia for the 2 years preceding the injury and
substantial disturbance of memory beyond that. When tested 8 months after
The Hippocampal Formation and Amnesia | 201
Trauma
Coma
7 weeks
Examination
Gross disturbance
of memory back
to infancy
RA, total: 2 years AA, total:
not fixed
Trauma
Trauma
Coma
7 weeks
Coma
7 weeks
Examination
RA, partial:
4 years patchy
memory
RA, total: 1 year
RA, total:
2 weeks
Memory normal Memory
precise
A few
memories
recalled
AA, total:
3 months
AA, total:
3.5 months
Examination
23 weeks
Residual permanent memory loss
(a)
(b)
(c)
FIGURE 7.12 The pattern of a patient’s recovery from amnesia caused by a closed head injury:
(a) after 5 months; (b) after 8 months; (c) after 16 months. RA _ retrograde amnesia;
AA _ anterograde amnesia. (From Barbizet, 1970. Reprinted by permission of the publisher. © 1970 by
W. H. Freeman.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 201
the injury, the patient showed some ability to remember new experiences, and
the period of total retrograde amnesia had shrunk to 1 year. When tested
16 months after injury, the patient had full ability to remember new events
and had only a permanent 2-week period before the injury about which he
could remember nothing. It is characteristic that retrograde amnesia is for
events close in time to the injury and that events just before the injury are
never recovered. In general, anterograde and retrograde amnesia show this
pattern of occurring and recovering together, although in different patients
either the retrograde or the anterograde symptoms can be more severe.
A number of striking features characterize cases of amnesia. The first is that
anterograde amnesia can occur along with some preservation of long-term
memories. This was particularly the case for HM, who remembered many things
from his youth but was unable to learn new things. The existence of such cases
indicates that the neural structures involved in forming new memories are
distinct from those involved in maintaining old ones. It is thought that the hippocampal
formation is particularly important in creating new memories and
that old memories are maintained in the cerebral cortex. It is also thought that
events just prior to the injury are particularly susceptible to retrograde amnesia
because they still require the hippocampus for support. A second striking feature
of these amnesia cases is that the memory deficit is not complete and there
are certain kinds of memories the patient can still acquire. This feature will be
discussed in the next section on implicit and explicit memory. A third striking
feature of amnesia is that patients can remember things for short periods but
then forget them. Thus, HM would be introduced to someone and told the
person’s name, would use that name for a short time, and would then forget it
after a half minute. Thus, the problem in anterograde amnesia is retaining the
memories for more than 5 or 10 seconds.
Patients with damage to the hippocampal formation show both retrograde
amnesia and anterograde amnesia.
•Implicit versus Explicit Memory
So far, this chapter has mainly focused on memories to which people have conscious
access. However, some of the most interesting research in the field of
memory focuses on memories that we are not aware that we have. Occasionally,
we will become aware that we know things that we cannot describe. One example
that some people can relate to is memory for the keyboard of a keyboard.
Many accomplished typists cannot recall the arrangement of the keys except by
imagining themselves typing. Clearly, their fingers know where the keys are, but
they just have no conscious access to this knowledge. Such implicit memory
demonstrations highlight the significance of retrieval conditions in assessing
memory. If we asked the typists to tell us where the keys are, we would conclude
they had no knowledge of the keyboard. If we tested their typing, we would conclude
that they have perfect knowledge. This section discusses such contrasts, or
dissociations, between explicit and implicit memory. In the keyboard example
202 | Human Memory: Retention and Retrieval
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 202
above, explicit memory shows no knowledge, while
implicit memory shows total knowledge. Explicit
memory is the term used to describe knowledge
that we can consciously recall. Implicit memory is
the term used to describe knowledge that we cannot
consciously recall but that nonetheless manifests
itself in our improved performance on some task.
Cases of total dissociation of implicit and explicit
knowledge, such as the typing example, are
rare with normal individuals. Such cases are more
common in patients who suffer from certain amnesias.
Amnesic patients display implicit memories of
many experiences that they cannot consciously recall.
For instance, Graf, Squire, and Mandler (1984)
compared amnesic versus normal participants with
respect to their memories for a list of words such as
banana. After studying these words, participants
were asked to recall them. The results are shown in
Figure 7.13. Amnesic participants did much worse
than normal participants. Then participants were
given a word-completion task. They were shown the
first three letters of a word they had studied and
were asked to make an English word out of it.
For instance, they might be asked to complete
ban______. There is less than a 10% probability that
participants can generate the word (banana) they
studied by chance alone, but the results show that
participants in both groups were coming up with the studied word more than
50% of the time. Moreover, there was no difference between the amnesic and
normal participants in the word-completion task. So, the amnesic participants
clearly did have memory for the word list, although they could not gain conscious
access to that memory in a free-recall task. Rather, they displayed implicit
memory in the word-completion task. The patient HM had also been shown to
be capable of implicit learning. For example, he was able to improve on various
perceptual-motor tasks from one day to the next, although each day he had no
memory of the task from the previous day (Milner, 1962).
Amnesic patients often cannot consciously recall a particular event but will
show in implicit ways that they have some memory for the event.
Implicit versus Explicit Memory in Normal Participants
A great deal of research (for reviews, read Schacter, 1987; Richardson-Klavehn &
Bjork, 1988) has looked at dissociations between implicit and explicit memory in
normal individuals. It is often not possible with this population to obtain the
dramatic dissociations we see in amnesic individuals, who show no conscious
memory but normal implicit memory. It has been possible, however, to demonstrate
that certain variables have different effects on tests of explicit memory than
Implicit versus Explicit Memory | 203
Amnesic
10
20
30
40
50
60
Normal
Participants
Word completion
Words recalled (%)
Word recall
FIGURE 7.13 Results from an experiment by Graf, Squire, and
Mandler comparing the ability of amnesic patients and normal
participants to recall words studied versus the ability to complete
fragments of words studied. Amnesic participants did much worse
than normal participants on the word-recall task, but there was
no difference between the amnesic and normal participants in the
word-completion task. (From Graf, Squire, & Mandler, 1984. Reprinted by
permission of the publisher. © 1984 by the Journal of Experimental Psychology:
Learning, Memory, and Cognition.)
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 203
on tests of implicit memory. For instance, Jacoby
(1983) had participants just study a word such
as woman alone (the no-context condition),
study it in the presence of an antonym manwoman
(the context condition), or generate the
word as an antonym (the generate condition). In
this last condition, participants would see manand
have to say woman. Jacoby then tested the
participants in two ways, which were designed to
tap either explicit memory or implicit memory.
In the explicit memory test, participants were
presented with a list of words, some studied and
some not, and asked to recognize the old words.
In the implicit memory test, participants were
presented with one word from the list for a brief
period (40 ms) and asked to identify the word.
Figure 7.14 shows the results from these two
tests as a function of study condition.
Performance on the explicit memory test was
best in the condition that involved more semantic
and generative processing—consistent with earlier research we reviewed on elaborative
processing. In contrast, performance on the implicit perceptual identification
test got worse. All three conditions showed better perceptual identification
than would have been expected if the participants had not studied the word at all
(only 60% correct perceptual identification). This enhancement of perceptual
recognition is referred to as priming. Jacoby argues that participants show greatest
priming in the no-context condition because this is the study condition in which
they had to rely most on a perceptual encoding to identify the word. In the generate
condition, participants did not even have a word to read.5 Similar contrasts
have been shown in memory for pictures: Elaborative processing of a picture will
improve explicit memory for the picture but not affect perceptual processes in its
identification (e.g., Schacter, Cooper, Delaney, Peterson, and Tharan, 1991).
In another experiment, Jacoby and Witherspoon (1982) wondered whether
participants would display more priming for words they could recognize than
for words they could not. Participants first studied a set of words. Then, in one
phase of the experiment, they had to try to recognize explicitly whether or not
they had studied the words. In another phase, participants had to simply say
what word they had seen after a very brief presentation. Participants showed
better ability to identify the briefly presented words that they had studied than
words they had not studied. However, their identification success was no different
for words they had studied and could recognize than for words they had
studied but could not recognize. Thus, exposure to a word improves normal
participants’ ability to perceive that word (success of implicit memory), even
when they cannot recall having studied the word (failure of explicit memory).
204 | Human Memory: Retention and Retrieval
No context
Recognition judgment
(explicit memory)
50
60
70
80
90
Context
Experimental condition
Correct judgment (%)
Generate
Perceptual identification
(implicit memory)
FIGURE 7.14 Results from
Jacoby’s experiment
demonstrating that certain
variables have different effects
on tests of explicit memory
than on tests of implicit
memory. The ability to recognize
a word in a memory test versus
the ability to identify it in a
perceptual test is plotted as a
function of how the word was
originally studied. (From Jacoby,
1983. Reprinted by permission of the
publisher. © 1983 by the Journal of Verbal
Learning and Verbal Behavior.)
5 Not all research has found better implicit memory in the no-context condition.However, all research finds
an interaction between study condition and type of memory test. See Masson and MacLeod (1992) for
further discussion.
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 204
Research comparing implicit and explicit memory suggests that the two types
of memory are realized rather differently in the brain.We have already noted that
amnesiacs with hippocampal damage show rather normal effects in studies of
priming, whereas they can show dramatic deficits in explicit memory. Research
with the drug midazolam has produced similar results in normal patients.
Midazolam is used for sedation in patients undergoing surgery. It has been noted
(Polster,McCarthy, O’Sullivan, Gray, & Park, 1993) that it produces severe anterograde
amnesia for the period of time it is in a patient’s system, although the patient
functions normally during that period. Participants given the drug just before
studying a list of words showed greatly impaired explicit memory for the words
they studied but intact priming for these words (Hirshman, Passannante, & Arndt,
2001).Midazolam has its effect on the neurotransmitters that are found throughout
the brain but which are particularly abundant in the hippocampus and prefrontal
cortex. The explicit memory deficits it produces are consistent with the association
of the hippocampus and the prefrontal cortex with explicit memory. Its lack of
implicit memory effects suggests that implicit memories are stored elsewhere.
Neuroimaging studies suggest that implicit memories are stored in the cortex.
As we have discussed, there is increased hippocampal activity when memories
are explicitly retrieved (Schacter & Badgaiyan, 2001). During priming, in
contrast, there is often decreased activity in cortical regions. For instance, in one
fMRI study (Koutstaal et al., 2001), priming produced decreased activation in visual
areas responsible for the recognition of pictures. The decreased activation
that we see with priming reflects the fact that it is easier to recognize the primed
items. It is as if their representation has been strengthened in the cortex.
A general interpretation of these results would seem to be that new explicit
memories are formed in the hippocampus; but with experience, this information
is transferred to the cortex. That is why hippocampal damage does not
eliminate old memories formed before the damage. The permanent knowledge
deposited in the cortex includes such information as word spelling and what
things look like. These cortical memories are strengthened when they are
primed and become more available in a later retest.
New explicit memories are built in hippocampal regions, but old knowledge
can be implicitly primed in cortical structures.
Procedural Memory
Implicit memory is defined as memory without conscious awareness. By this
definition, rather different things can be considered implicit memories. Sometimes,
implicit memories involve perceptual information relevant to recognizing
the words. These memories result in the priming effects we saw in experiments
such as Jacoby’s. In other cases, implicit memories involve knowledge about
how to perform tasks. A classic example of such an implicit memory involves
procedural knowledge, such as riding a bike. Most of us have learned to ride a
bike but have no conscious ability to say what it is we have learned.Memory for
such procedural knowledge is spared in amnesic individuals.
An experiment by Berry and Broadbent (1984) involved a procedural learning
task with a more cognitive character than riding a bike. They asked participants
Implicit versus Explicit Memory | 205
Anderson7e_Chapter_07.qxd 8/20/09 9:47 AM Page 205
to try to control the output of a hypothetical sugar factory
(which was simulated by a computer program) by manipulating
the size of the workforce. Participants would see the month’s
sugar output of the factory in thousands of tons (e.g., 6,000 tons)
and then have to choose the next month’s workforce in hundreds
of workers (e.g., 700). They would then see the next month’s
output of sugar (e.g., 8,000 tons) and have to pick the workforce
for the following month. Table 7.6 shows a series of interactions
with the hypothetical sugar factory. The goal was to keep sugar
production within the range of 8,000 to 10,000 tons.
One can try to infer the rule relating sugar output to labor
force in Table 7.6; it is not particularly obvious. The sugar output
in thousands of tons (S) was related to the workforce input in
hundreds (W), and the previous month’s sugar output in thousands
of tons (S1), by the following formula: S _ 2 _W_ S1.
(In addition, a random fluctuation of 1,000 tons of sugar is
sometimes added, and S and W stay within the bounds of 1
to 12.) Oxford undergraduates were given 60 trials at trying to
control the factory. Over those 60 trials, they got quite good at
controlling the output of the sugar factory. However, they were
unable to state what the rule was and claimed they made their
responses on the basis of “some sort of intuition” or because it “felt right.”
Thus, participants were able to acquire implicit knowledge of how to operate
such a factory without acquiring corresponding explicit knowledge. Amnesic
participants have also been shown to be capable of learning this information
(Phelps, 1989).
In recent years, sequence learning (Curran, 1995) has become a major
model for study of the nature of procedural memory, including its realization
in the brain. There are a number of sequence-learning models, but in the basic
procedure, a participant observes a sequence of lights flash and must press corresponding
buttons. For instance, there may be four lights with a button under
each, and the task is to press the buttons in the same order as the lights flash.
The typical manipulation is to introduce a repeating sequence of lights and
contrast how much faster participants can press the keys in this sequence than
when the lights are random. For instance, in the original Nissen and Bullemer
(1987) study, the repeating sequence might be 4-2-3-1-3-2-4-3-2-1. People are
faster at the repeated sequence. There has been much interest in whether participants
are aware that there is a repeating sequence. In some experiments, they
are aware of the repetition; but in many others, they are not. They tend not to
notice the repeating sequence when the experimental pace is fast or when they
are performing some other secondary task. Participants are faster at the repeated
sequence whether they are aware of it or not.
It does not appear that the hippocampus is critical to developing proficiency
in the repeated sequence, because amnesiacs show an advantage for the repeated
sequence, as do normal patients with pharmacologically induced amnesia. On
the other hand, a set of subcortical structures, collectively called the basal ganglia
(see Figure 1.8), does appear to be critical for sequence learning. It had long
206 | Human Memory: Retention and Retrieval
TABLE 7.6
Procedural Memory: An Illustrative Series of
Inputs and Outputs for a Hypothetical Sugar
Factory
Workforce Input Sugar Output (tons)
(W) (S)
700 8,000
900 10,000
800 7,000
1,000 12,000
900 6,000
1,000 13,000
1,000 8,000
Anderson7e_Chapter_07.qxd 8/20/09 9:48 AM Page 206
been known that the basal ganglia are critical to motor control, because it is
damage to these structures that produces the deficits associated with Huntington’s
and Parkinson’s diseases, which are characterized by uncontrolled movements.
However, there are rich connections between the basal ganglia and the
prefrontal cortex, and it is now known that the basal ganglia are important in
cognitive functions. They have been shown to be active during the learning of a
number of skills, including sequence learning (Middleton & Strick, 1994). One
advantage of sequence learning is that it is a cognitive skill one can teach to nonhuman
primates and so perform detailed studies of its neural basis. Such primate
studies have shown that the basal ganglia are critical to early learning of a
sequence. For instance, Miyachi, Hikosaka, Miyashita, Karadi, and Rand (1997)
were able to retard early sequential learning in monkeys by injecting their basal
ganglia with a chemical that temporally inactivated it. Other neural structures
appear to be involved in sequence learning as well. For instance, similar chemical
inactivation of structures in the cerebellum retards later learning of a sequence.
All in all, the evidence is pretty compelling that procedural learning involves
structures different from those involved in explicit learning.
Procedural learning is another type of implicit learning and is supported
by the basal ganglia.
•Conclusions: The Many Varieties of Memory
in the Brain
Squire (1987) proposed that there are many different varieties of memory.
Figure 7.15 reproduces his classification. The major distinction is between
explicit and implicit memory, which he calls declarative memory and nondeclarative
memory. Declarative memory basically refers to factual memories we
can explicitly recall. It appears that the hippocampus is particularly important
Conclusions: The Many Varieties of Memory in the Brain | 207
Memory
Semantic
(facts )
Episodic
(events)
Procedural skills
(e.g., motor,
perceptual,
cognitive )
Priming
( perceptual,
semantic )
Conditioning Nonassociative
(habituation,
sensitization )
Declarative Nondeclarative
FIGURE 7.15 The varieties of memory proposed by Squire. (From Squire, 1987. Reprinted by permission
of the publisher. © 1987 by Oxford University Press.)
Anderson7e_Chapter_07.qxd 8/20/09 9:48 AM Page 207
for the establishment of declarative memories. Within the declarative memory
system, there is a distinction between episodic and semantic memory. Episodic
memories include information about where and when they were learned. For
example, a memory of a particular newscast can be considered an episodic
memory. This chapter and the previous Chapter 6 have discussed these kinds of
memories. Semantic memories, discussed in Chapter 5, reflect general knowledge
of the world, such as what a dog is or what a restaurant is.
Figure 7.15 makes it clear that there are many kinds of nondeclarative, or
implicit, memory.We have just completed a discussion of procedural memories
and the critical role of the basal ganglia and cerebellum in their formation.We
also talked about priming and the fact that priming seems to entail changes to
cortical regions directly responsible for processing the information involved.
There are other kinds of learning that we have not discussed but that are particularly
important in studies of animal learning. These include conditioning,
habituation, and sensitization, all of which have been demonstrated in species
ranging from sea slugs to humans. Evidence suggests that such conditioning in
mammals involves many different brain structures (Anderson, 2000). Many
different brain structures are involved in learning, and these different brain
structures support different kinds of learning.
208 | Human Memory: Retention and Retrieval
1. One of the exceptions to the decay of memories with time
is the “reminiscence bump” (Berntsen & Rubin, 2002)—
people show better memory for events that occurred in
their late teens and early 20s than for memories earlier
or later.What might be the explanation of this effect?
2. The story is told about David Starr Jordan, an ichthyologist
(someone who studies fish), who was the first
president of Stanford University. He tried to remember
the names of all the students but found that whenever
he learned the name of a student, he forgot the name
of a fish. Does this seem a plausible example of interference
in memory?
3. Do the false memories created in the Deese-Roediger-
McDermott paradigm reflect the same sort of underlying
processes as false memories of childhood
events?
4. It is sometimes recommended that students study for
an exam in the same room that they will be tested in.
According to the study of Eich (1985; see discussion on
p. 197), how would one have to study to make this an
effective procedure? Would this be a reasonable way
to study for an exam?
5. Squire’s classification in Figure 7.15 would seem to
imply that implicit and explicit memories involve different
memory systems and brain structures—one called
declarative and the other, nondeclarative. However,
Reder, Park, and Keiffaber (2009) argue that the same
memory system and brain structures sometimes display
memories that we are consciously aware of and others of
which we are not.How could one determine whether
implicit and explicit memory correspond to different
memory systems?
Questions for Thought
Key Terms
amnesia
anterograde amnesia
decay theory
declarative memory
Deese-Roediger-
McDermott paradigm
dissociations
encoding-specificity
principle
explicit memory
fan effect
false-memory syndrome
implicit memory
interference theory
Korsakoff syndrome
mood congruence
power law of forgetting
priming
procedural knowledge
retrograde amnesia
state-dependent learning