INTRODUCTION
Patients with organic amnesia due to cerebral damage affecting medial
temporal lobe structures and related neural circuitry show impaired
performance on direct tests of memory (e.g., free recall, recognition).
Memory for recent information can also be investigated by means of
indirect measures of memory such as repetition priming. A large body of
literature documents that patients with organic amnesia show intact
repetition priming when the experimental paradigm involves facilitated
reprocessing of single items (e.g., words or pictures of real objects) or
between-item associations (e.g., related word pairs) with pre-existing
representation in memory (see for reviews Gooding et
al., 2000; Schacter & Buckner, 1998).
Instead, mixed results have been obtained regarding the capacity of
amnesic patients to be primed by individual items (e.g., nonwords or
abstract pictures) or between-item associations (e.g., unrelated word
pairs) that lack pre-existing memory representations. In particular, some
studies reported comparable priming for novel between-word associations in
amnesics and normal controls (Gabrieli et al.,
1997; Goshen-Gottstein et al., 2000;
Graf and Schacter, 1985) and others revealed
reduced priming in patients with amnesia (Cermak et
al., 1988; Mayes & Gooding, 1989;
Paller and Mayes, 1994; Schacter & Graf, 1986; Shimamura & Squire, 1989).
In their meta-analysis of available experimental evidence, Gooding et
al. (2000) noted a number of variables that can
affect amnesics' performance on repetition priming procedures for
novel material. The most relevant one is the tendency of healthy controls
to contaminate performance on the priming test for novel information by
using explicit stimulus retrieval strategies (e.g., Gooding et al., 1999; McKone &
Slee, 1997). Depending on how susceptible the performance index in
the particular experimental paradigm is to explicit memory contamination
(e.g., because it is based on accuracy rather than on speed of stimulus
processing or because experimental subjects engage in strategic expectancy
to respond faster), amnesic patients are likely to show a reduced priming
effect. A second experimental variable that may affect amnesic
patients' performances is the number of stimulus presentations during
the study phase. It has been suggested that novel information may be
acquired via neural circuits that do not include medial temporal
lobes; however, in this case, the learning process is slower and requires
multiple exposures to material (Holdstock et al.,
2002a). Accordingly, amnesic patients with medial temporal lobe
damage should be particularly disadvantaged by experimental procedures
with one stimulus presentation during the study phase and should perform
better if there are more study presentations. Finally, amnesic
patients' performances on priming tests for novel material should be
particularly poor if the experimental task is based on the generation of
the response cued by a fragment (as, for example, in word-stem
completion). Indeed, if these patients have some degree of frontal lobe
damage, then their ability to generate responses should be particularly
impaired (Brokate et al., 2003).
Here we report the results of a study further investigating priming
for newly formed associations in patients with organic amnesia. The
priming procedure we used consisted of adapting the widely used semantic
priming paradigm so that newly formed between-word associations replaced
the traditionally used semantic associates as prime–target pairs
(Dagenbach et al., 1990; Pecher & Raaijmakers, 1999; Schrijnemakers & Raaijmakers, 1997). To our
knowledge, the semantic priming paradigm has never been utilized to assess
whether patients with organic memory disorders can be primed by recently
established novel associations. To prevent the possible reliance of
healthy subjects on an episodic retrieval strategy to perform the task, we
established experimental parameters in the priming experiment (both SOA
and proportion of prime–target related pairs) needed to obtain
automatic priming (see Neely, 1977, 1991). Moreover, to maximize our chances of obtaining
a significant priming effect and to circumvent amnesic patients'
supposed difficulty in establishing new memory representations following
single stimulus exposure, we administered the priming experiment following
six study trials with the word pairs. Finally, the priming effect was
revealed by speeded lexical decision for previously processed words, thus
circumventing the amnesic patients' possible disadvantage with
priming procedures requiring item generation.
METHODS
Research Participants
Twenty subjects participated in this study. The first group consisted
of 10 patients with organic amnesia. All patients were referred to our
laboratory because of an impairment in recalling day-to-day events. The
neuropsychological evaluation (see Table 1)
showed a severe long-term episodic memory deficit in the context of
substantially preserved general cognition. In particular, these patients
had high range WAIS verbal IQs (M = 112.0; SD = 9.2) but
their performance scores on tests of verbal and visuo-spatial episodic
memory fell on average from 1.9 to 2.7 standard deviations below the
normal controls' mean.
Personal and clinical data, neuroradiological findings and performance
scores on some neuropsychological tests of the amnesic patients who
participated in the present study
The patients in the amnesic group could be classified according to the
locus of the neuropathological damage likely responsible for their memory
disorder. Five patients had bilateral atrophy of the hippocampus. Three of
them developed amnesia following cerebral hypoxia. Histopathological and
neuroimaging data have been reported indicating that temporary loss of
oxygen leads to severe hippocampal atrophy (Gadian et
al., 2000; Hopkins et al., 1995; Rempel-Clower et al., 1996). The 4th patient, an
insulin-dependent diabetic, became amnesic after a period of coma due to
iatrogenically caused hypoglycemia. Animal models demonstrate that
prolonged reduction of blood glucose concentration leads to relatively
selective damage of hippocampal cell sub-populations (Tekkok et al., 2002). The amnesic syndrome in the 5th
patient arose during chemotherapy for the treatment of leukemia. However,
no direct relationship could be established between the basic disease or
the chemotherapy and the hippocampal atrophy. Visual inspection and
morpho–volumetric analysis of high quality cerebral MRI images,
according to Pruessner et al. (2000), revealed
bilateral atrophy of the hippocampus in these 5 patients. In particular,
left and right hippocampal volumes ranged from −5.27 to −1.97
SDs below the mean of a normative group of healthy controls (Pruessner et al., 2000). There was no indication of
significant volume reduction in other brain regions, including the medial
temporal lobe cortices (perirhinal, parahippocampal and entorhinal
cortices). However, as we do not have a morphovolumetric analysis of these
brain regions at our disposal, the claim of completely spared
parahippocampal cortices should be regarded with caution.
One patient had memory problems as a result of surgical section of the
right column of the fornix (see Bonanni et al.,
2004). One patient became amnesic following the removal of a
hypophysary tumor. In this patient, the cerebral MRI showed the presence
of an ischemic area in the anterior region of the left thalamus. Two
patients had amnesia as a chronic sequela of subarachnoidal hemorrhage for
rupture of an anterior communicating artery aneurysm. In these patients,
the amnesic syndrome was thought to be the result of damage to septal
nuclei (Gade, 1982; Gade
& Mortensen, 1990). Morphovolumetric analysis of medial
temporal lobe structures in these 4 patients revealed hippocampal volumes
ranging from −0.3 to +1.0 standard deviations from normal
controls' mean, according to Pruessner et al. (2000). In fact, there was no overlap between
hippocampal volumes in the hippocampal and non-hippocampal patients and
the difference between the average volumes of the two groups was
significant for both the right hippocampus (t = 3.32, p
= .01) and the left hippocampus (t = 3.15, p = .02). The
last patient, a chronic survivor of severe closed-head trauma, wears a
vertebral metallic support and is ineligible for MRI investigation. In
such patients, the memory impairment is generally due to contusional and
diffuse axonal changes at many levels of cortical and subcortical cerebral
structures.
Biographical, clinical and neuroradiological characteristics of the
amnesic patients as well as performance scores on some tests of the
neuropsychological screening battery are summarized in Table 1.
The normal controls (NCs) for the patients in the amnesia group were
10 healthy volunteers matched for age, sex, and literacy with the
patients. Exclusion criteria for this group were the following: a history
of brain trauma, brain or psychiatric disease, diagnosed internal diseases
or taking drugs or medications potentially able to interfere with
cognitive efficiency. Average ages of amnesics and normal controls were
44.7 (SD = 16.1) and 39.7 (SD = 15.2) respectively
[F(1,18) = 0.5, p > .10]. Average years of
schooling were 12.9 (SD = 2.1) and 14.1 (SD = 3.3)
[F(1,18) = 0.9, p > .10].
Experimental Task
Material
The overall experimental material consisted of 180 word–word
pairs and 180 word–non-word pairs. Thirty critical
word–word pairs were chosen so as not to present any
pre-experimental association or semantic relation. In each pair, the first
word represented the prime and the second was the target in the priming
experiment. The words in the pairs were four to eight letters in length
and had an average frequency of occurrence in the Italian lexicon of
30.7:1,000,000 (range 4–233) according to Bortolini et al. (1971). The 30 word–word pairs were divided into
three subsets, each formed by 10 pairs. Thirty recombined
word–word pairs were obtained by re-pairing the primes and the
targets in each 10-pair subset. Thirty neutral word–word
pairs were obtained by combining the word NULLA (BLANK in Italian) as
prime and the same targets as the critical word–word pairs. Thirty,
4-to-8-letter legal nonwords were obtained by substituting one or two
consonants in real words; these nonwords were paired with the primes of
the word pair list to form 30 word–nonword pairs. Thirty
neutral word/nonword pairs were obtained by combining the
word NULLA as prime and the same nonwords as the previous word/nonword
pairs as targets. In order to surreptitiously increase the number of
trials in the priming experiment, 90 word–word and 120
word–non-word pairs were also selected to be used as fillers. Words
and nonwords for the filler and practice pairs were selected with the same
criteria as those used for the critical pairs.
Procedure
The overall experiment consisted of three sessions, each given on 1 of
3 consecutive days. The first two sessions were aimed at learning the 30
word–word pairs. For this purpose, in each session the subject
underwent three consecutive study-test trials. In order to make the
associative link between the words in the pair as rich and varied as
possible, in each session the encoding instructions in the study phase
changed from one trial to the next. Instead, the memory test always
consisted of an episodic cued-recall task. On the 3rd day (about 24 hr
after the last study–test session), the subject underwent the final
cued recall task immediately followed by the priming experiment.
Study phase
In the first study phase of the first-day session, the examiner
presented the two words in the pair auditorily (e.g., envelope–hair)
and the subject was requested to say a sentence aloud that included both
words (e.g., a hair was stuck to the envelope). In the second study phase,
the subject was requested to generate a similarity between the two words
in the pair which, also in this case, were presented auditorily by the
examiner (e.g., both of them are very thin). In the third study phase, the
subject read 30 sentences, each containing the first word and a fragment
(i.e., two or three letters) of the second word of a pair (e.g., In
the envelope there was a brown h_i_). The subject was asked to
complete the word fragment by writing down the missing letters. The same
three study phases were also given in the experimental session on the 2nd
day. In the first two phases, the subject was allowed to repeat the same
sentences or to generate the same similarities. Instead, in the third
study phase the sentences were varied from the first to the 2nd day. The
pair presentation order was randomized across trials.
Cued recall task
For each item of the cued recall task, the examiner said the first
word of a pair aloud and the subject was requested to recall the second
word. With the exception of the delayed trial on the third day, the
examiner provided the correct answer if the subject made an error.
Priming for new associations
The overall priming experiment consisted of three blocks of 120
trials, each containing an equal number (10) of intact, recombined and
neutral word–word pairs, 10 word–nonword pairs, 10 neutral
word–nonword pairs, 30 filler word–word pairs, and 40 filler
word–nonword pairs. The probability of encountering a word rather
than a non-word in the target position was 50% and the intact (i.e.,
studied) pairs represented 16.6% of the total number of word–word
pairs. To avoid presenting the same prime or target words after too short
an interstimulus lag, stimuli pertaining to each of the three 10-pair
subsets contributed differently to each experimental block. In particular,
for each block the 10 critical word–word pairs came from the first
subset, the 10 recombined word–word pairs from the second subset
and, finally, the 10 neutral word–word pairs from the third subset.
In this way, each prime or target word appeared only once in each block.
The presentation order of each block varied among subjects in a
pseudo-random fashion.
The experiments were run individually and controlled by a 386 PC. The
stimuli were presented in white capital letters in the center of a black
computer screen; they occupied about 5.7° of visual angle. Each trial
started with the appearance of a small central cross for 2 s, indicating
the fixation point. The cross was followed by the prime word, which
remained for 300 ms. This was followed immediately by the target, which
remained until the subject responded or, if there was no response, for 5
s, after which it was replaced by a blank screen. The experimenter
controlled the beginning of each trial by pressing a key on the computer
keyboard. The subjects were placed approximately 50 cm from the computer
screen and were instructed to read the prime silently and to perform a
lexical decision task on the target. They had to press a green button for
words and a red button for nonwords using their dominant hand. If the
subjects pressed the wrong button, the mistake was indicated by an
auditory signal. The PC recorded response reaction times and accuracy.
RESULTS
Cued Recall Task
Results of the word pair recall (Figure 1)
were analyzed by means of a two-way mixed ANOVA with group (amnesics
vs. normal controls) as the between-subject factor and trial
(from 1 to 6) as the within-subjects factor. Overall, average accuracy of
recall was much higher in the normal (M = 24.8) than in the
amnesic (M = 7.1) group [F(1,18) = 75.2, p
< .001]. The trial effect was also highly significant
[F(5,90) = 44.1, p < .001], reflecting the
progressive increase of recall accuracy across successive study–test
trials. LSD post-hoc tests revealed significant differences among
adjacent trials in each case, p < .05, except for the
difference between the fourth and fifth trial, p > .10. The
Group × Trial interaction was also significant
[F(5,90) = 2.6, p < .05]. The interaction
did not imply a different learning rate in the two groups. Indeed, the
average performance improvement passing from the first to the sixth trial
was comparable in the two groups [M = +9.2 and +10.6 in the
amnesic and control groups, respectively; F(1,18) = 0.4,
p = n.s.]. Rather, the significant interaction was the
expression of differently shaped learning curves in the two groups. As a
matter of fact, the polynomial contrasts revealed that while the amnesic
patients' learning curve was only fitted by a linear slope
[F(1,18) = 31.0, p < .001], in the normal
controls both the linear and the quadratic components of the slope
contributed to significant prediction of the learning curve
[F(1,18) = 40.2 and 29.5, respectively, p < .001
in both cases]. The significance of this latter component likely
reflects the precocious reaching of a ceiling effect in the normal control
group's performance accuracy across successive study–test
trials (as also revealed by artificially small variances in this
group's performance; see Figure 1).
Performance accuracy of amnesic patients and normal controls on the
six immediate and the delayed trials of the cued recall
task.
A further two-way mixed ANOVA comparing performance on the sixth and
on the seventh cued recall trial revealed significant main effects of
group [F(1,18) = 95.8, p < .001] and trial
[F(1,18) = 9.7, p < .05]. The significant
Group × Trial interaction [F(1,18) = 9.8, p =
.02], documented discrepant performance changes in the two groups as
a function of test trial. Indeed, while the performance decrement in the
normal control group was very limited and actually not significant
[M = −0.6; F(1,18) = .02], it was
substantial and highly significant in the amnesic patients' group
[M = −5.8; F(1,18) = 16.0, p <
.001].
Priming for New Associations
The average number of lexical decision errors made by amnesic patients
and normal controls on the target words of the intact, recombined and
neutral word–word pairs of the priming experiment was very low
(M = 1.6 and 0.7, respectively). For this reason, accuracy data
on this task will not be further analyzed.
As for reaction times (RTs), only those on correct lexical decisions
were considered for the statistical analysis. In addition, trials with RTs
outside a range represented by the individual's mean RT of more or
less than 2 standard deviations were considered outliers and excluded from
all the statistics. The average RTs obtained by amnesics and normal
subjects as a function of priming condition are reported in Table 2. These data were analyzed by means of a
two-way mixed ANOVA with group as the between-subjects factor and priming
condition (intact vs. recombined vs. neutral
word–word pairs) as the within-subjects factor. The group effect was
significant [F(1,18) = 6.1, p < .02], due
to average longer RTs in the amnesic group (M = 796 ms) than in
the normal control group (M = 668 ms). The priming condition
effect was also significant [F(2,36) = 4.4, p <
.02]. Planned comparisons between the three conditions revealed that
average RTs on the intact word–word pairs (M = 718 ms) were
significantly faster than on the recombined pairs (M = 741 ms)
[F(1,18) = 7.9, p = .01], while the difference
with respect to the neutral condition (M = 736 ms) approached
significance [F(1,18) = 3.7, p = .07]. RTs on
the recombined and neutral word–word pairs did not differ
[F(1,18) = 0.6]. The Group × Priming condition
interaction did not approach significance [F(2,36) =
0.3], thus suggesting comparable priming effects in the two
groups.
Lexical decision RTs across the different experimental conditions of
the priming paradigm in the amnesic and normal control group
Relationship Between Performance on Cued Recall Task
and Priming for New Associations
In order to investigate the relationship between episodic and implicit
learning in the amnesic patients, we split the group according to the
performance mean on the delayed cued recall task (M = 6.2). The
low performer group included 6 patients (performance range in the
delayed cued recall test from zero to 5), while the remaining 4 patients
constituted the high performer group (performing from 10–16
in the delayed recall test). Interestingly, 5 of the 6 patients in the low
performer group (designated with asterisk in Table
1) were affected by bilateral hippocampal atrophy. The 6th patient
had undergone surgical section of the right column of the fornix.
As shown in Table 3, the performance
patterns of these two groups of amnesics in the semantic priming paradigm
differed. In fact, while the high performer group revealed a 47 ms
facilitation in the intact compared to the recombined word-word pairs, no
evidence of facilitation was detectable in the low performer group. A
two-way mixed ANOVA provided partial statistical support for this claim.
Indeed, while the group effect was not significant [average RTs in
the high and low performer groups were 851.5 and 757.8, respectively;
F(1,8) = 0.8], both the priming condition effect
[F(1,8) = 4.7, p = .06], and the Group ×
Priming condition interaction [F(1,8) = 4.8, p =
.06] were close to significance. As a matter of fact, separate
post-hoc analysis of the priming effect in the two subgroups by
means of the Newman-Keuls test showed no sign of priming in the low
performer group, p = .99, while the effect in the high performer
group reached significance, p = .01, despite the extremely small
sample size.
Lexical decision RTs in the intact and recombined pair conditions of
the priming paradigm in the amnesic patients' subgroups that scored
below (low performers) or above (high performers) the group mean on the
delayed cued recall task
DISCUSSION
As expected, the healthy subjects' episodic memory for the word
pair list (as measured by the cued recall test) was much better than that
of the amnesic patients. Moreover, a qualitative analysis of performance
showed that the two groups had comparable learning rates passing from the
first to the sixth study–test trial but that the amnesic patients
exhibited much greater forgetting than the normal control group passing
from the sixth to the delayed recall trial. However, due to a ceiling
effect in the healthy subjects' cued recall performance, the
interpretation of both of these findings is somewhat uncertain. In fact,
the precocious reaching of a quasi-perfect performance level could have
led to an underestimation of both the learning and the forgetting rates in
this group of subjects, thus giving rise to spurious results in the
between groups comparisons.
In the lexical decision task, the healthy subjects revealed a
significant priming effect. Indeed, their lexical decision was 28 ms
faster following intact than recombined pairs. In the amnesic group,
making a lexical decision was also faster in the intact than in the
recombined pair condition, but the facilitation effect was smaller (19 ms)
than in the normal control group. Moreover, in the amnesic group the
emergence of a relational priming effect was related to performance
accuracy on the cued recall task. Indeed, only the amnesic patients who
correctly recalled 10 or more word pairs in the delayed cued recall test
displayed a significant facilitation in the priming procedure. No sign of
a priming effect was shown by amnesics who scored 5 or less on the recall
task. It should be noted that the strength of our conclusions about a
direct relationship between performance on direct and indirect associative
memory tests in patients with amnesia is somewhat undermined by the low
patient sample size. As a matter of fact, despite a robust priming effect
in the high performer amnesic group (47 ms) and the lack of any priming in
the low performer group, the critical Group × Priming condition
interaction only approached significance. It is a fact, however, that
difficulty in recruiting adequately sized samples of patients with
selective memory disorders underlain by well defined cerebral lesions is a
constant in the neuropsychological literature on organic amnesia, so that
firm conclusions can only be reached by converging evidence from different
group studies.
To summarize, in a group of patients with organic amnesia we found
evidence of a relationship between deficient ability to episodically
remember novel between-word associations and impaired ability to rely on
these previously studied associations to facilitate lexical decision. How
can these results be reconciled with the contradictory data emerging from
the existing literature on priming for novel between-word associations in
amnesic patients? First of all, it seems very unlikely that poor priming
in patients with severe amnesia is just an epi-phenomenon of the reliance
of healthy controls on explicit recollection of verbal associations during
the lexical decision task. As previously noted, experimental parameters of
the priming experiment (both SOA and proportion of intact pairs) were
established to obtain an automatic form of priming. Experimental results
supported this prediction. Indeed, the amnesics and healthy controls took
about the same time to make the lexical decision in the recombined and
neutral pair conditions. This suggests that following presentation of the
first word of a pair, the subjects did not generate the conscious
expectancy of the paired word (Neely, 1977).
Since preparatory strategies did not contribute to performance, the
subjects could not rely on memory for the study episode, and the priming
phenomenon reflected the unconscious influence exerted by the recently
stored verbal associations in facilitating the lexical decision.
The hypothesis that poor priming in amnesic patients results from
insufficient exposure to the study material also seems untenable (Gooding et al., 2000). In the present study, the
verbal associations were given six times before the priming experiment. In
most previous investigations (e.g., Goshen-Gottstein et
al., 2000; Paller & Mayes, 1994), who
reported variable results, subjects studied novel verbal associations only
once or twice before performing the priming experiment.
Finally, it is unlikely that the observation of reduced relational
priming in the severe amnesics' subgroup was biased by the kind of
experimental paradigm used. Subtle frontal damage could have been
responsible for the amnesics' reduced ability to generate a word as a
completion of a stem, or to guess a word following degraded stimulus
presentation (e.g., in conditions of masked and tachistoscopic exposure).
In the present study, patients were only requested to make a lexical
decision about a letter string. This is likely the least demanding
condition in a word processing task and, as such, the least likely to
present a problem for the possibly reduced generation abilities of brain
lesioned patients.
Based on previously available evidence and on the presently reported
data, we hypothesise that two factors are crucial for determining whether
normal or deficient priming for novel associations will be observed in
amnesics. The first factor concerns the characteristics of the amnesic
sample investigated. In keeping with previous investigations (Schacter & Graf, 1986; Shimamura & Squire, 1989), our data show that the
severity of the anterograde memory impairment predicts the probability
that the individual amnesic will reveal a significant priming effect for
recently established verbal associations. Moreover, in our sample of
patients the emergence of a relational priming was also predicted by the
putative locus of cerebral damage responsible for amnesia. In fact, all
but 1 patient in the low performer subgroup suffered from hippocampal
atrophy. Visual inspection of the parahippocampal gyrus in these patients
did not show any sign of atrophy in the cortical areas surrounding the
hippocampus. Unfortunately, since we have no disposable morphovolumetric
analysis, we could not quantitatively support the argument of a complete
sparing of these areas. The 6th patient had undergone sectioning of the
right fornix column. Since the fornix is the main pathway connecting the
hippocampus with other subcortical structures critical for memory, this
patient's amnesic syndrome can also be considered as resulting from
functional deafferentation of the hippocampus proper (Aggleton et al., 1992). Instead, patients in the high
performer group had cerebral lesions outside the medial temporal lobes and
their hippocampal volumes were in the range of normal controls.
The second factor likely able to influence the quality of relational
priming in patients with organic amnesia is the nature of the stored
representation underlying the priming effect. In particular, it seems that
when the procedure stresses the perceptual aspects of the stored
representation, then the amnesics' priming is normal. Instead, when
the reprocessing facilitation is based on more conceptual features of the
memory representation, then it is more likely that the priming effect will
be deficient. In particular, in studies of word identification, patients
with amnesia revealed normal priming in all experiments in which the words
forming the pairs (which appeared side by side during the priming
paradigm) actually reproduced the same perceptual format as in the study
phase (Gabrieli et al., 1997; Goshen-Gottstein et al., 2000). Instead, in Paller and
Mayes's (1994) study and in the present
research in which the words were presented sequentially (thus not
reproducing the perceptual condition of the study phase), relational
priming in severely amnesic patients was poor. This is not to say that
perceptual factors had no role in the reprocessing facilitation of word
pairs in our study (and possibly in Paller & Mayes's). However,
the emphasis in the study phase on establishing multiple and meaningful
relationships between words in the pairs, together with the change in the
perceptual format from study to test, likely stressed the role of more
conceptual levels of memory representation at the origin of the priming
effect. If this hypothesis is true, then it can be supposed that patients
with organic amnesia, especially those with lesions affecting the
hippocampal formation, are impaired in storing novel conceptual
associations between previously unrelated word pairs.
In this perspective, the presently reported data might also be
relevant for the debate over the possible role of the hippocampus in
episodic and semantic learning. Indeed, while its emergence from repeated
study–cued recall trials suggests that relational priming originates
in the episodic memory system, the nature of the priming paradigm (a
lexical decision task facilitated by the creation of meaningful
relationships between words in the pair) might suggest the alternative
view that the priming effect originates in the knowledge system; that is,
in semantic memory. Espousing one or the other of the two above-mentioned
hypotheses regarding the origin of the priming effect clearly constrains
our interpretation of the presently reported data. Indeed, if we advocate
the first hypothesis (relational priming as rising from the episodic
memory), then poor cued recall and lack of a priming effect would be two
expressions of the same difficulty of hippocampal amnesics in storing new
associative memories in the episodic memory system (Henke et al., 1997; Holdstock et
al., 2002b; Turriziani et al., 2004).
However, if one takes performance on the cued recall task as indicative of
the ability to store new episodic memories and, in contrast, level of
relational priming as indicative of the ability to store new
representations in the semantic memory system, then patients with
hippocampal damage would be equally impaired in both kinds of learning.
Following this assumption, the presently reported data are incompatible
with the neurobiological implementation of Tulving's dualist model
(Tulving & Markowitsch, 1998), which assumes
that the parahippocampal regions are responsible for establishing new
memory representations in the semantic system and that the hippocampus
(which receives its major cortical input from the parahippocampus) is the
critical structure for episodic memory. Rather, the results of the present
study are entirely consistent with a theory that posits a single system
for the storage of new episodic and semantic memory traces (Anderson & Ross, 1980; Cohen et
al., 1997; Squire & Knowlton, 1995;
Squire & Zola, 1998). In this unitary view
of human memory, the episodic–semantic distinction describes either
two different portions of the stored memory representation (the semantic
one possibly deriving from the primitive episodic trace after being set
free from the record of the temporo–spatial context in which the
information was initially acquired) or two different retrieval modes for
the same memory representation (inserted in an autobiographical context in
one case, free from any reference to personal events in the second case)
and the amnesic syndrome consists of an impairment in storing new memory
representations in this unitary system.
