INTRODUCTION
For several decades, we have known that the medial temporal lobes
(MTLs) and diencephalon are critical for episodic memory. For example,
bilateral damage to the MTLs usually produces a severe amnesia, as in the
case of Scoville and Milner's patient HM (1957). From subsequent lesion and neuroimaging
studies, we have learned that the hippocampus and adjacent medial temporal
structures support the fundamental processes of encoding, storage, and
possibly also retrieval in episodic memory (see Nadel
& Moscovitch, 1997; Squire et al.,
2004).
In recent years, frontal lobe (FL) contributions to memory have also
become better understood. Frontal lesions tend to have more subtle effects
on memory than MTL lesions, but convergent evidence suggests that
interaction between FL and MTL systems is essential for normal memory
function. Most theories propose that the FLs modify, control, or
“work with” the operations of the MTL in memory. These
putative control processes are usually described as domain-general, and
include organization, search, selection, and monitoring or decision-making
(e.g., Moscovitch & Winocur, 2002; Shallice, 2002; Shimamura,
2002). The general idea is that MTL memory processes are powerful
but inflexible, so that frontal control is needed to
“supervise” or “confer intelligence on” them, to
make them more useful in an ever-changing environment.
Many excellent recent reviews of the FLs and memory are available
(e.g., Baldo & Shimamura, 2002; Petrides, 2000), but most emphasize basic research. In
this article, we aim to highlight recent laboratory findings on the
frontal lobes and memory that may inform the evaluation and management of
memory disorders in the real world. This should be useful in at least two
ways. First, a clearer picture of what is impaired, and what is spared,
may emerge if clinical evaluation examines separate components of memory,
including those that are dependent on the FLs (e.g., search and
monitoring). Unfortunately, frontal lobe processes are not emphasized in
most standardized tests of memory. Second, clues to rehabilitation might
also emerge from such a review. For example, in management of amnesia
following MTL damage, most efforts are focused on bypassing or replacing
the normally-used processes, instead of trying to restore them (which, for
all intents and purposes, is impossible; Glisky &
Glisky, 2002; Wilson, 2002). A similar
approach may be fruitful with FL patients.
In this review, we will first outline some current models on the FLs
and memory. Second, we will highlight a few processes that seem to rely on
frontal cortex. For each of these processes, we will pay particular
attention to studies in which experimenters have been able to attenuate FL
patients' deficits, and discuss related neuroimaging studies where
applicable. In each section, we briefly will discuss consequences for
clinical evaluation and rehabilitation. Finally, we will cover some more
general issues that must be taken into account when considering the role
of the FLs in memory. Although we will focus primarily on focal frontal
lesion patients for the sake of brevity and clarity, our review may also
apply to other kinds of patients who show frontal or strategic memory
problems. Impairments in “working with memory” have been
reported in a broad range of disorders that appear to involve dysfunction
of frontal lobe or connected structures, including Parkinson's
disease, Huntington's disease, closed head injury, schizophrenia,
attention deficit disorder, fronto-temporal dementia, and Alzheimer's
disease.
MODELS OF FRONTAL FUNCTION IN MEMORY
Most theorists agree that the FLs support strategic modification and
control of memory processes implicating posterior neocortex and MTL, but
they disagree concerning the details. For example, most now concur that
strategic, executive, and control are umbrella terms,
which can be fractionated into multiple processes. There are competing
models of what these processes are, and of the relations among them. In
the first of three example models, Shallice (2002) has proposed a Supervisory Attentional
System, in which frontal processes are required for solving
nonroutine memory problems via goal setting for the task,
development and implementation of a strategy to achieve those
goals, and evaluation of the results. Next, Shimamura (2000; 2002) has outlined
four major components in his Dynamic Filtering Theory: selection
or focused attention on perceptual or mnemonic information,
maintenance of it in mind, updating or manipulation of
it, and rerouting or switching from one behavioral or mental set
to another. Finally, Moscovitch and Winocur (2002) cite four major elements in their
Working-with-Memory model: response inhibition to allow
for some operations to be selected over others, working memory
for monitoring, manipulation, and evaluation, cue specification and
maintenance to guide encoding and retrieval, and a sense of
“felt rightness” when endorsing or rejecting
information. Note that all of these theories of FL memory processes have
much in common with general models of executive function and working
memory, but are applied specifically to long-term memory (see Moscovitch, 1992; Wagner,
2002).
As there is disagreement regarding these putative processes, it is
also unclear whether and how to localize them to specific frontal lobe
areas. Some models, however, make proposals with a fair degree of
specificity. For example, Shallice (2002) posits
that goal setting is supported by anterior prefrontal cortex,
development and implementation of a memory search strategy is
carried out by left dorsolateral prefrontal regions, searching
through memory requires right ventrolateral regions, and
evaluation of the results depends on right dorsolateral cortex
(see Petrides, 2002, for a similar view on
memory search and evaluation processes). Moscovitch and Winocur (2002) are also specific as to localization. In their
view, response inhibition depends on premotor regions, the
manipulation of information in working memory (including setting
goals, implementing search strategies and evaluating outcomes under
uncertainty) on mid-dorsolateral prefrontal cortex, cue specification
and maintenance on ventrolateral regions, and a sense of
“felt rightness” when rejecting or endorsing
information on ventromedial and anterior frontal cortex, respectively. A
similar model has been proposed by Fletcher and Henson (2001), who argue that the FLs' three main tasks
in memory are (1) goal setting and strategy selection (supported by
anterior regions; see also Buckner, 2003; Simons & Spiers, 2003), (2) maintenance of
information (ventrolateral regions), and (3) the selection, manipulation,
and evaluation of information (dorsolateral regions).
These models give an idea of current thinking on the frontal lobes and
memory. Each emphasizes different aspects of frontal function in memory,
although it is too early to tell whether one model is more appropriate
than another. Broadly speaking, however, they agree that the FLs are
important when selecting and encoding information, searching through
memory for information that is not readily available, and monitoring or
evaluating the products of that search. We will discuss each of these
features in turn, focusing on when FL patients do, and do not, show
impairments.
STAGES/PROCESSES
Encoding
Semantic organization
In general, FL patients use fewer strategies than healthy controls to
work with memory (e.g., Gershberg & Shimamura,
1995). One strategy that is normally quite beneficial involves
organization of study materials with respect to semantics or meaning
(Bousfield, 1953; Mandler,
1967; Tulving, 1962). The frontal lobes
seem to be especially important for such organization during encoding.
Hirst and Volpe (1988) examined this phenomenon
in a small group of frontal patients. Participants were asked to memorize
two sets of words for later recall: one set contained unrelated words, and
the other contained categories of related words, randomly intermixed with
one another (plant names, occupations, etc.). Healthy controls showed much
better memory for the categorizeable words than the unrelated ones,
whereas frontal patients showed no benefit. In another version of the
task, Hirst and Volpe gave participants a set of categorizeable words to
study, along with detailed instructions on how to organize them. These
instructions did not confer any additional benefit on the controls, but
helped the frontal patients improve substantially. Incisa della Rocchetta
and Milner (1993) and Gershberg and Shimamura
(1995) reported similar findings, although both
emphasized that FL damage impairs organization and strategy implementation
during encoding and retrieval (see also Baldo
et al., 2002; but see Alexander et al.,
2003; Stuss et al., 1994). Taken
together, these findings suggest that FL patients are capable of
organizing materials and benefiting from this in memory, although they
fail to do so spontaneously. This may be especially true when the lesion
encroaches on dorsolateral frontal cortex (Baldo et
al., 2002; but see Alexander et al.,
2003), consistent with the function assigned to this region in the
various models.
Clinical evaluation and real world performance. The main way
to assess the influence of semantic organization on episodic memory in the
clinic is by using a verbal memory test, such as the California Verbal
Learning Test (CVLT-II; Delis et al., 2000), and
examining indices of semantic clustering during the initial learning
trials. Patients with focal FL lesions obtained lower scores on the
CVLT-II semantic clustering measure in comparison to normal controls in
Baldo et al.'s (2002) report, although
neither Stuss et al. (1994) nor Alexander et al.
(2003) found any regional differences for this
measure. Related information can be obtained when testing nonverbal
memory, such as memory for the Rey-Osterreith Complex Figure (1941). Detailed scoring systems (e.g., Boston
Qualitative Scoring System; Stern et al., 1999)
can provide information about how well organized copying is (e.g., shows
good planning and little fragmentation).
A deficit in organizing materials along semantic lines may have some
impact on FL lesion patients in the real world, although there appears to
be little research on this question. For instance, it is conceivable that
patients may have greater difficulty recalling a shopping list (which is
essentially what the CVLT-II is) or any other material where
organizational principles are helpful, such as in studying any subject or
complex set of instructions or procedures. Autobiographical memory also
may be impaired if patients are less able to organize and create
connections among life experiences, or even in encoding and recounting an
extended, complex episode, such as a long conversation, a busy day, a
vacation, or a movie or play one had just seen. On the whole, however,
this may not be the most debilitating memory problem faced by FL patients.
Furthermore, the Hirst and Volpe (1988) results
suggest that if such memory problems are apparent in real life, organizing
stimuli for patients, or providing explicit instructions for them to do so
on their own, will aid them greatly. A similar conclusion was reached by
Incisa della Rocchetta and Milner (1993), who
reported that even though FL patients had trouble recalling the words on
their experimental task, they showed relatively good recall of prose
passages from the Wechsler Memory Scale Logical Memory subtest (Wechsler, 1997). The authors argued that the prose
passages required less self-initiated organization on the part of
subjects. These studies indicate that supplying external organization can
help, at least if the material is relatively simple and the organizing
principles are obvious, such as they are in a list of semantically related
items, which is rarely the case for real-world episodes. As yet, there are
few tests of real world memory to examine the effects of frontal damage or
dysfunction (discussed later).
Contextual information
The frontal lobes may be especially important for remembering the
context within which information was acquired. For example, frontal
patients are usually impaired in memory for the sequence or temporal order
in which items occur, even when memory for the items themselves is intact
(Butters et al., 1994; Johnson et al., 1997; Kesner et
al., 1994; Mangels, 1997; Parkin et al., 1988; Shimamura et
al., 1990; Swain et al., 1998).
Dorsolateral FL regions may be particularly important for these operations
(Kopelman et al., 1997; Milner et al., 1991), as many of the models outlined
in the introduction suggest. Functional neuroimaging data fit with this
idea, although it is unclear whether one hemisphere is more critical than
the other, or whether left and right hemispheres make material-specific
contributions (e.g., Cabeza et al., 1997b, 2000; Dobbins et al., 2002;
Fan et al., 2003; Nolde et
al., 1998; Nyberg et al., 1996; Slotnick et al., 2003).
FL impairments in temporal order memory are not inevitable. They can
be overcome if subject-performed tasks (SPTs) are employed during
encoding. For instance, McAndrews and Milner (1991) had FL patients and controls either say the name
of objects presented during study or perform a predetermined action for
each (e.g., squeeze a sponge). On a subsequent test of memory for the
order in which the objects had been presented, FL patients were reliably
impaired in the naming condition, but performed the same as controls in
the action condition (see Butters et al., 1994,
for similar findings). McAndrews and Milner suggested that
subject-performed tasks create more distinctive encoding of events, by
providing extra cues (e.g., motor cues) with which to retrieve
object-order information.
Another example of impaired memory for context concerns
source, which can consist of various types (location, perceptual
attributes, surrounding information, etc.). When FL patients are exposed
to information from more than one source within a session (e.g., seeing
some words and hearing others, or hearing different sentences from
different experimenters), they are usually able to remember the content
itself, but are impaired when asked about the source (e.g., Janowsky et al., 1989a; Johnson et
al., 1997; Schacter et al., 1984;
Shimamura & Squire, 1987, 1991; cf. Thaiss & Petrides,
2003). Glisky and colleagues (1995)
reported analogous findings in older adults who were characterized
preexperimentally on the basis of their frontal lobe function using
neuropsychological tests. The experimenters presented sentences spoken by
one of two voices (sources) to the older adults, and found that
better memory for source was related to higher FL function.
Recently, however, Glisky and colleagues (2001) showed that the source memory deficit associated
with poor FL function can be reduced or eliminated. By instructing
participants to think about the relation between content and source during
study (e.g., by asking “How well do this voice and sentence go
together?”), the experimenters eliminated the usual source memory
impairment in older people with mild FL dysfunction. These older people
may be less likely to attend to contextual information spontaneously,
and/or link context with content, but the deficit apparently can be
overcome by changing task instructions. Such a study deserves to be
extended to focal FL lesion patients.
Clinical evaluation and real world performance. Taken
together, the temporal order and source memory studies indicate that FL
patients' memory problems are attributable, at least in part, to poor
attentional control or selection and organization of information at
encoding. Unfortunately, there are few standardized clinical measures with
appropriate norms for the evaluation of context memory. There are,
however, tests that measure memory for two similar sets of information
(e.g., both the CVLT-II and Wechsler Memory Scale–III
[WMS-III] Word Lists contain two lists of words, WMS-III Logical
Memory contains two stories). Intrusions of words or details from one list
or story when subjects are trying to remember those from the other may be
a good indicator of impaired memory for temporal order. Indeed, patients
with focal FL lesions are more likely than control participants to intrude
words from a distractor list during yes/no recognition (Baldo et al., 2002).
One limiting factor for clinical use is that these tests do not have
normative data for these specific types of intrusions. Moreover, they
examine memory for just one kind of context, and use only verbal
materials, and so it might be an idea to adapt some of the experimental
tasks outlined above for clinical testing. After all, memory for when an
event occurred or from whom one learned a fact is important for
functioning in the real world, allowing patients to answer questions like
“Did I take my pill before or after lunch?” or “Did I
read that fact in the New York Times or the National
Enquirer?” (Senkfor & Van Petten,
1998).
The few studies that exist suggest that problems remembering context
are not restricted to the laboratory. For example, Davidson et al. (2005) asked FL lesion patients about what they knew
about the terrorist attacks in the United States on September 11th, 2001,
and the source of this information. The patients appeared to have normal
knowledge of what happened on September 11th, but were impaired when asked
about the source. Similarly, Shimamura and colleagues (1990) had FL patients organize historical events into
chronological order, and found that the patients were impaired, despite
the fact that their knowledge about the events appeared normal. Some
researchers have argued that poor temporal context memory may underlie
other problems, including confabulation (see Schnider,
2001), though Moscovitch and Gilboa (Gilboa,
2005; Gilboa & Moscovitch, 2002;
Moscovitch, 1994; Moscovitch
& Melo, 1997) consider the deficits in temporal order and
confabulation to be symptoms of impairment to more fundamental processes,
such as search, evaluation, and “felt rightness” associated
with strategic retrieval.
It thus might be fruitful to develop real-world tests of memory for
context to probe for these deficits, and if they exist, to try to overcome
them. For example, the Shimamura et al. (1990)
historical events task could easily be used to examine whether a patient
has a retrograde temporal ordering impairment. An analogous task involving
recent events could be developed to test for an anterograde deficit,
although it would have to be updated every few months. Memory for source
could be probed by initially presenting stimuli in different modalities
(e.g., seeing some words and hearing others), to determine whether a
patient has disproportionately poor source memory.
The experimental data outlined earlier suggest that FL patients'
context memory deficits can be minimized if information is more richly and
distinctively encoded (in temporal order memory), or if patients are
encouraged to link content and context information (in source memory). A
subject-performed task or attentional manipulation could easily be
included in a clinical assessment of context memory. Furthermore, as far
as we are aware, no one has tried to adapt either of these paradigms
outside of the laboratory, to see if FL patients' memory problems can
be attenuated in the real world. It seems that this would be a worthwhile
endeavor. Of course, there may be limitations on how far the beneficial
effects of such manipulations can generalize. For example, in Butters and
colleagues' (1994) replication of the
McAndrews and Milner (1991) study, they included
a condition in which subjects merely imagined performing a task
with each studied object, but this did not ameliorate FL patients
performance. As well, the Shimamura et al. (1990) historical events data suggest that the temporal
ordering deficit is not based only on encoding, and cannot be ameliorated
entirely by encoding strategies (because the patients encoded these events
when they were healthy). Providing appropriate cues or organizational
strategies at retrieval may thus prove useful too.
Search at Retrieval
Memory impairment following FL damage is not limited to encoding.
Retrieval of previously learned information requires FL-mediated search
processes that are organized and strategic, if the memory test itself does
not provide sufficient cues to specify the target event. This can be
studied by comparing memory tests that vary in their requirements of
self-initiated search processes at retrieval.
Effects of cuing
Retrieval of previously learned information can be tested by using
free recall, cued recall, or recognition. Free recall arguably makes the
greatest demands on self-initiated strategic processing, whereas the cuing
inherent in cued recall and recognition decreases the need for strategic
processing. Surprisingly, there have been few studies in which recall and
recognition have been compared in the same FL patients, and when they
have, the findings have been mixed (e.g., Alexander et
al., 2003; Baldo et al., 2002; Kopelman & Stanhope, 1998; Stuss et al., 1994). A recent meta-analysis (Wheeler et al., 1995) suggested that free recall was
most likely to produce memory deficits in FL patients (i.e., in 80% of
reviewed studies), and recognition was least likely (i.e., in only 8% of
reviewed studies). Cued recall was intermediate in producing memory
deficits in these patients (i.e., in 50% of studies). Note, however, that
for many of the studies in that review, recall was for organized
materials, whereas recognition was for unorganized materials, so whether
there is truly a difference between recall and recognition is still not
clear. Neuroimaging studies with healthy participants have shown
increased activation in anterior regions (among others) during cued recall
in comparison to recognition of word pairs (e.g., Cabeza et al., 1997a, 1997c), although as
Shallice (2003) notes, it is challenging to
compare lesion with function imaging data because of the different
assumptions inherent to each method. Thus, although the FLs appear to be
important for episodic memory tasks regardless of the procedure used for
retrieval, they may be especially important for free recall. Based on the
theories of frontal function, one would expect that the regions likely to
be activated more during recall than recognition include right
dorsolateral and left ventrolateral cortex, although there are too few
studies and too much variability to determine whether this prediction is
consistent with the results of the neuroimaging findings (for a brief
discussion, see Fletcher & Henson,
2001).
This recall/recognition dissociation may provide additional
insight into how different memory processes are dependent on the FLs. For
example, dual-process theories of memory distinguish between
recollection (controlled retrieval of item-specific or contextual
information) and familiarity (the mere sense that an item is old;
e.g., Jacoby, 1991; Mandler,
1980; Tulving, 1983; for a review, see
Gardiner & Richardson-Klavehn, 2000).
Although the definitions of recollection and familiarity are contentious,
some theorists have suggested that recall relies more heavily on
recollection, whereas recognition depends on both but can often (but not
always) be performed on the basis of familiarity alone (see Mandler, 1980). There is some neuropsychological and
neuroimaging evidence that the FLs may be more involved in recollection
than familiarity (e.g., Davidson & Glisky,
2002; Henson et al., 1999; Wheeler & Stuss, 2003). Although some recent
neuroimaging studies have additionally found activity in frontal regions
that is correlated with familiarity (e.g., Ranganath et al., 2000, 2004; Yonelinas et al., 2005), the implications are unclear.
Neuroimaging data on whether the frontal lobes are differentially involved
in familiarity versus recollection may be confounded with the
degree of confidence participants have in their decisions. When
recollection is associated with highly confident decisions, and
familiarity with relatively less confident decisions, then regions of
frontal cortex associated with evaluation, such as the dorsolateral
aspects, will show greater activation for the latter (Henson et al., 1999; 2000;
but see Yonelinas et al., 2005).
Clinical evaluation and real world performance. Many
standardized clinical memory tests use both recall and recognition,
including the CVLT-II, WMS-III subtests (Logical Memory, Paired
Associates, Word Lists, Visual Reproduction), Kaplan-Baycrest
Neurocognitive Assessment memory tasks (Word Lists and Complex Figure;
Leach et al., 2000), and the Doors and People
test (Baddeley et al., 1994). Although most of
these tests do not provide normative data specifically on the difference
between recall and recognition, separate norms are provided for each
measure, allowing the clinician to judge whether there are meaningful
differences between recall and recognition in a patient.
This relative difficulty with free recall as opposed to recognition
has obvious implications for compensating for everyday memory problems of
FL patients. These patients may be better able to answer questions that
rely on recognition (e.g., “Did you talk to your sister on the
telephone yesterday?”) rather than on recall (e.g., “Who did
you talk to on the telephone yesterday?”). Providing organization at
encoding (Gershberg & Shimamura, 1995; Incisa della Rocchetta & Milner, 1993) or giving
cues in the form of notes or reminders should also decrease the demands on
strategic search at retrieval.
Given the difficulty that FL patients have when presented with
insufficient cues, it would be useful to develop a self-cuing retrieval
strategy for everyday life. Most mnemonics (such as the method of loci)
essentially provide a framework of cuing, but the problem is that one has
to use the framework at encoding for it to be effective at retrieval. Can
there be some procedure designed to help people cue themselves at
retrieval without also engaging in some stilted methods of encoding? At
present, there seem to be few studies concerning this question. Yet even
if one could devise a good retrieval strategy, it might be difficult to
train patients to implement it. In order to use a retrieval strategy in a
sophisticated, flexible way, a patient would need (at a minimum) to (1)
recognize there is a problem, (2) retrieve the appropriate strategy, (3)
initiate and maintain it, (4) evaluate the result, and (5) switch to a new
strategy if he or she recognizes that the current one is ineffective.
Performing this series of operations requires good monitoring, metamemory,
and inhibition—the very processes with which FL patients have
difficulty (see Burgess et al., 2000).
Nevertheless, research in the rehabilitation of executive function and
problem-solving is currently in its infancy (see Turner
& Levine, 2004), and so, if in the future, FL patients can be
taught to implement other sorts of problem-solving strategies, perhaps
they can be taught to do so for memory as well.
Monitoring at Retrieval
There are at least two cases where FL damage seems to impair
monitoring of memories at retrieval (i.e., decision-making about whether
the products of a memory search are accurate or not). We discuss each, in
turn, next.
Liberal response bias
Although FL patients have fewer difficulties with recognition than
recall, there are aspects of recognition that appear to be abnormal. For
example, on a yes-no recognition memory test, people are shown a mix of
previously seen items and new ones, and have to discriminate the old from
the new. A person endorsing more of the new items (false alarms)
relative to the old ones (hits) is said to have a more liberal
response bias. Persons with MTL damage usually have a reduced hit rate,
but show normal (or conservative) response bias. In contrast, FL damage
may have little effect on hit rate, but can lead to an elevated false
alarm rate, suggesting more liberal bias (e.g., Swick
& Knight, 1999; but see Verfaellie et al.,
2004). Frontal patients are especially susceptible to producing
false alarms when the foils are perceptually or conceptually similar to
targets (Baldo et al., 2002; Melo et al., 1999; Schacter et al.,
1996). There may be some regional, and hemispheric, specialization
within FL with regard to false alarms. For example, when Rapcsak and
colleagues (2001) divided FL patients into those
with unilateral left versus right hemisphere damage, the right
frontal group was more liberal on a face memory task than the left frontal
and control groups (compare to Alexander et al.,
2003, who reported abnormal bias for verbal materials in left
frontal patients).
The studies outlined earlier concern large groups of FL patients, who
usually show a small but significant shift in bias, but there are single
cases that are more extreme. Rapcsak and colleagues have reported several
patients with right FL damage who show pathologically high false
recognition rates, across a variety of stimuli. Most of these patients
also show false recognition of information in the retrograde domain,
suggesting a retrieval deficit. In one study, Rapcsak et al. (1999) examined whether changing retrieval instructions
could attenuate false recognition. They administered a face recognition
task, in which famous and nonfamous faces were presented, and subjects
were asked to endorse only the famous ones. The FL patients showed normal
memory for the famous faces, but pathologically high false recognition of
the nonfamous ones. The investigators then administered a similar task,
but this time they changed the instructions during the test, asking
participants only to endorse a face if they could state the name,
occupation, or other identifying information for it. Giving these extra
instructions had little effect on controls but led to a dramatic
improvement in the patients, presumably by encouraging them to be more
cautious, and/or make an endorsement based on more detailed
information (i.e., using “recollection” rather than mere
“familiarity” or “gist;” Jacoby, 1991; Mandler,
1980).
Schacter and colleagues (1996) made a
similar argument in describing B.G., a right FL lesion patient who also
showed pathological false alarming. The authors found that when foils were
conceptually similar to targets (e.g., members of the same category), B.G.
was very likely to false alarm to the foils, but when they were not
similar, B.G.'s false alarm rate fell greatly. Note, however, that
Parkin and colleagues (1999) reported a similar
patient, but argued that his deficit involved faulty encoding. It could be
that some patients are liberal due to poor encoding (e.g., impoverished,
general coding of items) and others due to poor retrieval (e.g., abnormal
bias and/or confidence), and this may depend on the hemisphere or
frontal region damaged. Although many of the models mentioned at the
outset describe criterion setting as a function of prefrontal cortex
(e.g., Henson et al., 2000), Moscovitch and
Winocur's (2002) is the only one which
considers these biases explicitly by assigning criterion-setting and
automatic monitoring functions to the ventromedial and frontopolar
regions, respectively, with damage to them leading to changes in negative
and positive biases, respectively. Thus, damage to ventromedial prefrontal
cortex can lead to acceptance of false information typical of
confabulation, and damage to frontopolar regions can lead to impairment in
endorsing true memories or imbuing them with a sense of self-relatedness
and recollection (for further discussion see Moscovitch
& Winocur, 2002).
Clinical evaluation and real world performance. Some clinical
memory measures include yes-no recognition, but the most popular of these
uses only verbal stimuli (the CVLT-II), and liberal bias could be a
material-specific problem (cf. Alexander et al.,
2003; Rapcsak et al., 2001). The WMS-III
has a yes-no face recognition subtest, but does not include normative data
for false alarms or a criterion measure. The ideal clinical measure would
compare yes-no to forced-choice recognition across verbal and nonverbal
stimuli, because the latter may be easier for FL patients.
In the real world, it is currently unclear what proportion of FL
patients have major problems resulting from a liberal bias in recognition,
but there are at least a few cases that do. A good example is described by
Ward and colleagues (1999). This patient would
approach strangers on the street, asking if they were personal
acquaintances or television stars, because they seemed so familiar to him.
This spurious sense of familiarity might cause FL patients to be more
vulnerable to scams and exploitation in the real world, where they might
be more likely to believe what is told to them because it seems familiar
(Jacoby, 1999). Harassment or other
inappropriate social interactions may also be a concern. Fortunately, even
in more extreme cases, encouraging patients to use a more conservative
response criterion, or to search their memory carefully before making a
decision, may help reduce false alarm rates greatly (as in Rapcsak et al., 1999; Schacter et
al., 1996; but see Parkin et al.,
1999).
Confabulation
Confabulation (or “honest lying;” Moscovitch, 1989; for recent reviews, see Gilboa & Moscovitch, 2002; Johnson et al., 2000), describes the phenomenon in
which patients retrieve information that is untrue, accompanied by a
strong feeling that it is true, and are unaware of any
discrepancy, and usually of any memory impairment at all. Confabulation is
most common in autobiographical memory, although it can be evoked in other
domains, including semantics or general knowledge (Moscovitch & Melo, 1997). Many theorists have
dichotomized confabulations (albeit, using different nomenclatures and
criteria) into those that are severe (e.g., bizarre, spontaneous,
persistent, frequent, etc.) versus those that are mild (e.g.,
realistic, provoked, momentary, rare, etc.), although a comprehensive
taxonomy of confabulation has yet to be established.
The most parsimonious explanation of confabulation attributes it to a
retrieval deficit, because patients often confabulate regarding memories
acquired long before brain damage. In separate reviews of the literature,
Gilboa and Moscovitch (2002) and Johnson et al.
(2000) found that ventromedial FL regions (in
either hemisphere) were most often implicated in confabulation, consistent
with the function ascribed to this region in various models. Although
basal forebrain damage (as in cases of anterior communicating artery
aneurysm; AcoA) is common in confabulating patients, it does not always
lead to confabulation, and may not be necessary to produce it. When
confabulation resolves over time, it does so in a gradual and seemingly
spontaneous manner, and does not necessarily improve at the same rate as
other memory abilities (see Gilboa & Moscovitch,
2002).
Clinical evaluation and real world performance. Confabulation
may be difficult to detect in the clinic, because we often take patients
at their word, and do not seek corroboration of their reports unless we
suspect a need. Although confabulation is associated with poor memory and
executive function, the literature suggests these correlations are not
very strong. Spontaneous or fantastic confabulations may be easier to spot
than mild ones (simply because they are implausible, inconsistent, or
bizarre), but mild ones may also cause problems in daily functioning. Two
published methods may aid in documenting confabulation in the clinic.
Dalla Barba (1993) has designed a series of
questions to which most normal people would answer “I don't
know” (e.g., “What did you have for lunch on March 12,
1998?”). As well, Moscovitch and Melo (1997) used a cue-word procedure and found that
confabulations could be elicited for both personal and semantic
information. Another good way to detect confabulation is by asking both
the patient and a relative or friend about a few of the patient's
life experiences, and seeing how the two reports square with one
another.
Despite its rarity, confabulation can have severe real-life
consequences. As stated earlier, it is usually accompanied by unawareness
of deficit. If there is no reason for a patient to think he is
confabulating, there is no reason for him not to act according to his
memory. Few reports of rehabilitation or management of confabulation exist
in the literature, possibly because it is so difficult. Asking the patient
to use a stricter criterion at retrieval, or to reason through a series of
logical steps to see that his confabulation cannot be true, may do little
to convince him, especially in the long run (see Moscovitch, 1989). Burgess and McNeil (1999), however, did report a successful case of
rehabilitation of an AcoA patient who would try to leave his house each
morning for a nonexistent job. The authors had the patient keep a diary of
his daily experiences, and trained him to check it at the beginning of
each day to convince himself that he had not been working recently, and
had nothing scheduled for the day. Of course, this method may not be
successful with every patient. Confabulation may have multiple possible
causes, for example, faulty memory search, decision-making, or temporal
context memory (Burgess & Shallice, 1996;
Moscovitch & Melo, 1997; Schnider, 2001; for reviews of models, see Gilboa & Moscovitch, 2002; Johnson et al., 2000). If so, different rehabilitation
strategies may work with different patients.
ADAPTING CURRENT MEMORY REHABILITATION
METHODS FOR FL DYSFUNCTION
Given the memory impairments resulting from FL damage or dysfunction,
a number of rehabilitation or compensatory techniques may be particularly
useful in this group. Although few studies exist on the efficacy of
traditional memory training techniques in FL patients, many of these
techniques address general problems such as decreased self-initiation and
interference, and thus may be relevant to this population. We outline a
few briefly.
Memory book training is frequently used in patients with
impaired memory resulting from MTL lesions (Kapur,
1995). The use of a memory book bypasses impaired memory processes
by using the external aid as a memory storage and retrieval device.
Procedural/implicit memory can be used to learn the new procedures
inherent in the use of a memory book. This may be also fruitful in FL
patients, because procedural/implicit memory is by and large intact in
them (Robinson, 2001). This approach is being
evaluated by our colleagues at Baycrest (Richards et
al., 1990; Wu et al., 2004), as well as
elsewhere.
Prospective memory involves remembering to do something in
the future. The ability to perform future intended tasks is impaired in FL
patients (e.g., Brunfaut et al., 2000), likely
because of poor initiation of the action or distraction before the action
is completed. The formation of implementation intentions is one way to
circumvent these difficulties (Gollwitzer,
1999). This technique simply involves forming and mentally
rehearsing plans of action to occur in response to specific situations
(e.g., “Each night when I go to bed, I will take my white
pill.”) Forming these implementation intentions is thought to create
situational cues that, when they are encountered later, automatically
activate the desired behavior and effectively bypass the need for
effortful self-initiation. Although to our knowledge this technique has
not been applied to patients with focal FL damage, it can improve
prospective memory in healthy older adults (Chasteen et
al., 2001), a population with subtle but reliable decline in
frontal and executive function (e.g., Raz et al.,
1997; West, 1996).
Another potentially useful procedure involves training
recollection, using a repetition lag procedure (Jennings & Jacoby, 2003). As previously described,
FL patients may rely more on automatic influences on memory (i.e.,
familiarity) than consciously controlled memory processes (i.e.,
recollection). The repetition lag procedure trains recollection over
multiple sessions using a continuous recognition task that requires
recollection (as opposed to familiarity) of repeated items over gradually
increasing intervals. Training with older adults has shown enhanced
recollection of information within the training task (Jennings & Jacoby, 2003) and generalization of
training effects to other memory tasks (Jennings et
al., 2005).
Finally, errorless learning may be particularly important for
patients with FL dysfunction (Baddeley & Wilson,
1994). In general, the ability to learn new information is enhanced
when errors are prevented (reviewed in Kessels & de
Haan, 2003). Frontal patients may be likely to incorporate errors
into memory when learning, and, having done so, be unlikely to inhibit or
unlearn them (Wilson, 2002). Errorless learning
is more effective than an errorful technique in the acquisition of new
information by patients with amnesia if they have FL dysfunction (Komatsu et al., 2000), presumably because errorless
learning decreases interference from incorrect responses.
NEUROTRANSMITTERS AND PHARMACOTHERAPY
The previous sections discussed cognitive manipulations that may
attenuate FL memory deficits, but others kinds of manipulations may also
be successful. For example, Kramer and colleagues have suggested that
aerobic exercise can improve cerebral function and thereby attenuate
age-related declines in memory and executive function (for a review, see
Colcombe et al., 2004). It is possible that such
an approach might also benefit patients with brain injuries.
As well, there is a small but growing literature on the
neuropharmacology of memory. Many neurotransmitters are implicated in
memory (for a review, see Arnsten & Robbins,
2002), but we will focus on one, dopamine (DA), to illustrate
current knowledge and future potential. One clue regarding dopamine comes
from studies of Parkinson's disease (PD), which is associated with
decreased levels of DA in the basal ganglia and cerebral
cortex.1
Dopamine has also been implicated in
animal studies of memory. For example, the prefrontal cells that fire
during delayed objects and location memory tasks are usually dopaminergic
(Goldman-Rakic, 1998), although there is some
controversy over whether the D1 or D2 receptor is most important, and other
neurotransmitters undoubtedly play a role, as well.
Although motor
problems are the most obvious aspect of the disease, Parkinson's
patients have subtle memory problems akin to those of frontal patients
(especially those with dorsolateral FL lesions; for reviews, see
Bondi & Troster, 1997;
Saint-Cyr,
2003). These deficits likely occur because low levels of DA cause
dysfunction of FL cortex itself, which is rich in dopamine receptors, or
interfere with the interaction between FL cortex and the basal ganglia.
When PD patients are given DA agonist drugs to improve their motor signs
(e.g., l-dopa), they show a modest improvement on many of the memory and
executive tasks on which they are normally impaired (e.g.,
Gotham et al., 1988;
Lange et al.,
1992). The FLs are probably implicated in this effect, although to
date there are few functional neuroimaging data linking drug-related
improvements in patients' memory with changes in FL activation (but see
Cools et al., 2002 for an initial step).
Could these drugs ameliorate memory problems in people with other
kinds of FL damage or dysfunction? McDowell et al. (1998) recently conducted one of the few human studies
along these lines. They gave either a DA agonist (bromocriptine) or
placebo to TBI patients (who showed impaired executive function, and
probably had some FL damage). The drug improved performance on many of the
executive measures, such as reducing perseverative errors on the Wisconsin
Card Sorting Test, but did not improve performance on nonexecutive tasks,
such as letter cancellation. Perhaps results of this kind could generalize
to other patients with FL damage or dysfunction. There are, however, at
least three reasons to be cautious. First, there may be vast individual
differences in who benefits from a drug; for example, some PD patients
show little response to DA agonists. Second, drugs may improve some
processes but have a minimal, or detrimental, effect on others. For
example, Cools et al. (2003) reported that
giving DA agonists to PD patients improved their task-switching
performance, but made the same patients abnormally liberal in a
decision-making task (see also Gotham et al.,
1988). They argued that that by increasing levels of DA they
restored normal function to the system supporting task-switching, but
“overdosed” the system responsible for decision-making.
Finally, in cases of focal lesion there may not be enough tissue left to
function properly, even if the patient is given the right drugs (Arnsten & Robbins, 2002).
EVALUATING MODELS OF FRONTAL FUNCTION IN
MEMORY
A few decades ago, it was widely held that the FLs had little to do
with episodic memory. It was only in the 1980s that a first generation of
models began to differentiate clearly frontal from medial temporal
contributions to memory (e.g., Moscovitch and Winocur,
1992; Petrides & Milner, 1982). Since
then, a second generation of models has emerged, some of which we outlined
at the beginning of this article. Each posits different cognitive
processes, and (with a few exceptions) maps these onto different frontal
lobe regions. Because many of the models are arguably heuristic and still
in their early stages, at present it may be more useful to focus on
similarities among them than differences.
The first, and possibly most important, point of agreement among
current models is that there are multiple frontal processes supporting
memory. In the past, this would have been a more controversial statement,
but an abundance of dissociations in lesion and neuroimaging studies
supports the fractionation of frontal function into multiple subcomponents
(even if there is disagreement as to how many there are, and what they
each do). Second, there is some convergence among current models
already. For example, with respect to retrieval, there appears to be a
consensus that at least two major stages are required: Cue specification
and memory search have been attributed to ventrolateral FL by many authors
(e.g., Moscovitch & Winocur, 2002; Petrides, 2002; Shallice,
2002; Simons & Spiers, 2003), whereas
post-retrieval evaluation of the output of the memory search, especially
when the decision is uncertain, has been localized to dorsolateral FL
(Fletcher & Henson, 2001; Moscovitch & Winocur, 2002; Petrides, 2002; Shallice,
2002; Simons & Spiers, 2003). There
also appears to be a growing consensus regarding the role of the
ventromedial FL in confabulation.
A third generation of model development and testing is clearly needed,
one which will allow us to identify more precisely the memory operations
supported by the frontal lobes. Many of the present models are still
somewhat underspecified, so that any given finding can be explained
according to multiple theories, and there seem to be few cases in which
one can unambiguously pit one theory of FL function and memory against
another (although for initial attempts see Burgess
& Shallice, 1996; Shimamura, 2000).
Further complicating matters, of course, is the high degree of within- and
between-subject variability following frontal damage. Nevertheless, as our
understanding of frontal lobe function and memory improves, this will
obviously help to guide evaluation and rehabilitation.
EVALUATION OF FRONTAL MEMORY PROBLEMS
As we have discussed, the FLs likely support multiple processes in
episodic memory. Consequently, no single test will provide a comprehensive
measure of the ability to work with memory. Undoubtedly, new tasks will
become available as researchers develop more precise and sophisticated
theories of FL memory functions. However, of all the currently available
clinical measures of memory, perhaps the CVLT and CVLT-II cover the most
ground when it comes to frontal memory impairments. This test allows one
to estimate organization (both semantic and subjective), context memory
(albeit, only for temporal order memory), memory search (by contrasting
recall and recognition), and response bias (in yes/no recognition). It
would be useful if there were a similar nonverbal test, because frontal
memory deficits can be material-specific; however, clinicians may be able
to use some of the nonverbal tasks we have reviewed in a piecemeal
fashion.
Two other tests that have made their way from the experimental
literature to the clinical domain are Petrides and coworkers'
self-ordered pointing (Petrides & Milner,
1982) and conditional-associative learning tasks (Petrides, 1985). In the self-ordered pointing task,
participants are shown a set of items (words or pictures) on a sheet, and
are required to point to one of them. Then they are shown another sheet
with the same items in new locations and they are required to point to a
new item, and so on until all the possible items have been sampled.
Although patients with MTL damage fail on this test because they cannot
remember the items they have seen, patients with FL damage fail because
they cannot monitor their responses effectively. The deficit is associated
with mid-dorsolateral lesions (Brodmann area 9/46) on the left or
right depending on whether the material is verbal or spatial. The
conditional-associative learning test is associated with response
selection deficits caused by lesions in premotor cortex (Brodmann area
6/8). In this task, participants are required to map different
responses to different stimuli, which patients with lesions to those areas
have difficulty accomplishing, as do patients with Parkinson's
disease (Vriezen & Moscovitch, 1990).
Collecting subjective information (from the patient and a relative or
caregiver) may also be useful for evaluation. First, memory problems may
be evident in the real world, but not in the clinic. Second, comparing a
patient's own ratings to those of a relative or caregiver can give
one the sense of whether insight and meta-memory are normal.
Unfortunately, although many rating scales of memory, executive ability,
and/or behavior exist (e.g., Frontal Systems Behavior Scale, Grace & Malloy, 2001; Memory Assessment Clinics
Self-Rating Scale, Crook & Larrabee, 1990;
Metamemory in Aging, Dixon et al., 1988), none
are specific to the memory problems commonly seen in FL disorders. We
would suggest having the patient and a family member answer questions
about frontal memory processes by using a Likert-scale rating to describe
the frequency and/or severity of FL-related memory problems, or
forcing them to choose between “worse than before” and
“same as before” to answer. Table 1
shows examples of the kinds of questions we would include in such a
questionnaire.
Examples of possible items for a frontal memory
questionnaire
Finally, more general problems may underlie what appear to be
“frontal” memory impairments, and must be addressed in
diagnosis and rehabilitation. First, apathy, depression, and other mood
disorders are common in brain injury and many brain diseases. Left
hemisphere lesions may be more likely to produce depression, whereas right
hemisphere damage may yield indifference or euphoria (for a review, see
Mayberg, 2002). Patients with mood disorders may
be less likely to want to work with memory, or work with anything else for
that matter (Wilson, 2002). Second, FL damage
may greatly reduce metamemory and metacognition (e.g., Janowsky et al., 1989b). In more extreme cases, such
as confabulation, FL patients are typically unaware of their deficit. If a
patient is not aware that he or she has a memory problem, then he or she
is less likely to engage the self-initiated strategies that are the
hallmarks of FL memory function. The former problem (mood) may be more
amenable to therapy, whereas the latter (metacognition and awareness of
deficit) may be less so, but both should be considered when evaluating and
planning a rehabilitation program for FL memory impairment.
SUMMARY AND CONCLUSIONS
In this review, we have outlined the effects of FL lesions on memory,
and highlighted cases where FL impairment can be attenuated or eliminated
in the laboratory. Evaluating FL memory processes in the clinic may give
the clinician a better idea of what processes are impaired, what kinds of
problems patients may face in the real world, and what to do to minimize
such problems. Administering a brief memory test may not be enough to
detect a memory problem. It seems critical to measure performance on some
of the FL memory tasks outlined earlier, or at least to examine the more
“frontal” scores on clinical tasks like the CVLT, because
patients have to work-with-memory very often in everyday life.
Two kinds of questions, theoretical and practical, must be answered in
future research. On the theoretical side, we must learn more about the
basic functioning of the FLs in memory, and the connections and
interactions among frontal, medial temporal, and other brain regions.
Recently it has become easier to do so, using functional neuroimaging.
Regional specificity within the FLs means that damage to different
subregions or subsystems will produce different kinds of memory problems,
and as we learn more about these we will be better able to know exactly
what to expect in individuals with different kinds of FL lesions. In
rehabilitation theory, unanswered questions include why some functions are
more likely to recover than others, and how best to encourage recovery (by
teaching alternative cognitive strategies, and/or promoting neural
plasticity).
On the practical side, we must develop applied neuropsychological
studies of the FLs and memory. We are a long way from applying the
detailed experimental knowledge and theoretical models to the clinic, but
the broad conclusions derived from the laboratory can guide clinical
practice. Indeed, the time is ripe for such an undertaking. Few of the
behavioral interventions outlined here seem to have been attempted in the
real world. Although it remains unclear whether what we have reviewed can
translate from the lab to a patient's home or work, we believe that
some procedures are highly likely to succeed. Fortunately, research on the
frontal lobes and memory has grown rapidly in recent years, and should
continue to do so. When it comes to linking basic research with real world
management and rehabilitation of frontal memory impairment, hopefully we
will make as much progress in next few decades as we have in the last
few.
ACKNOWLEDGMENTS
This research was supported by a fellowship to P. D. and a grant to M.
M. from the Canadian Institutes of Health Research. We thank Tiffany Chow,
Asaf Gilboa, and Brian Levine for suggesting papers for review, and Larry
Leach for suggesting questions for the frontal memory questionnaire.
