The cognitive neuroscience of memory: perspectives ?· The cognitive neuroscience of memory: perspectives…
Post on 11-Jul-2018
The cognitive neuroscience of memory: perspectivesfrom neuroimaging research
DANIEL L. SCHACTER
Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, MA 02138, USA
Cognitive neuroscience approaches to memory attempt to elucidate the brain processes and systems thatare involved in dierent forms of memory and learning. This paper examines recent research from brain-damaged patients and neuroimaging studies that bears on the distinction between explicit and implicitforms of memory. Explicit memory refers to conscious recollection of previous experiences, whereasimplicit memory refers to the non-conscious eects of past experiences on subsequent performance andbehaviour. Converging evidence suggests that an implicit form of memory known as priming is associatedwith changes in posterior cortical regions that are involved in perceptual processing; some of the sameregions may contribute to explicit memory. The hippocampal formation and prefrontal cortex also playimportant roles in explicit memory. Evidence is presented from recent PET scanning studies that suggeststhat frontal regions are associated with intentional strategic eorts to retrieve recent experiences, whereasthe hippocampal formation is associated with some aspect of the actual recollection of an event.
Cognitive neuroscience is an interdisciplinary enterprisethat draws on data from studies of brain-damagedpatients, functional neuroimaging techniques, computa-tional models and research with non-human animals(for reviews, see Gazzaniga 1995; Kosslyn & Koenig1992). Cognitive neuroscience approaches to memoryhave grown explosively in recent years. During the pasttwo decades impressive progress has been made in elu-cidating the functional properties of memory and inrelating them to underlying neuroanatomical substrates(for reviews and discussion, see Damasio 1989; Schacter1996; Moscovitch 1994; Squire et al. 1993; Thompson &Krupa 1994)One of the most important new developments in the
cognitive neuroscience of human memory involves theadvent of functional neuroimaging techniques such aspositron emission tomography (PET) and functionalmagnetic resonance imaging (functional MRI). These`windows on the brain' (Posner & Raichle 1994)provide new tools for probing dierent aspects ofmemory; they have led to an explosion of recentstudies and a rapid accumulation of novel information(for reviews, see Buckner & Tulving 1995; Ungerleider1995). In this paper I relate ndings and ideas fromneuroimaging studies to an issue that has assumedparamount importance in recent research: the notionthat memory is not a unitary or monolithic entity, butinstead can be fractionated into a number of distinct butinteracting component processes and systems (forreview and discussion, see Schacter & Tulving (1994)).This conclusion has been based largely on converging
evidence from studies of human amnesic patients andlesion studies of experimental animals. I considerrecent neuroimaging research that has begun to illu-minate the nature of the relations among variousmemory processes and systems.
2 . FORMS OF MEMORY ANDFUNCTIONAL NEUROIMAGING
The past two decades worth of research in cognitiveneuroscience have pointed towards a fundamentaldistinction between an explicit or declarative form ofmemory that is critically dependent on the integrity ofmedial temporal lobe and diencephalic brain struc-tures, and an implicit or non-declarative form ofmemory that involves a variety of systems outside themedial temporal/diencephalic region (cf. Eichenbaum1994; Schacter 1987, 1994; Squire 1992; Squire & Zola-Morgan 1991). Explicit memory is reected byconscious recall and recognition of recent informationand events. Implicit memory is reected by non-conscious eects of past experience on subsequentbehaviour, as expressed by such phenomena aspriming (Tulving & Schacter 1990), skill learning(Cohen & Squire 1980; Salmon & Butters 1995) andhabit formation (Knowlton & Squire 1993; Mishkin1984).While conclusions about brain systems and forms of
memory have relied heavily on research concerninghuman patients and experimental animals with brainlesions, the advent of functional neuroimaging techni-ques provides a new opportunity to map the
Phil.Trans. R. Soc. Lond. B (1997) 352, 1689^1695 1689 & 1997 The Royal SocietyPrinted in Great Britain
component processes and systems underlying implicitand explicit forms of memory. In studies using PETand functional MRI, estimates of regional cerebralblood ow are obtained in dierent experimentalconditions. Ideally, experimental conditions areconstructed so that they are similar in all respectsexcept for a critical feature that diers between condi-tions. Comparisons are then made between conditionsof interest by subtracting estimates of blood ow in onecondition from estimates of blood ow in another. Thelogic of this approach holds that any resulting dier-ences in regional cerebral blood ow are associatedwith the critical feature that distinguishes one condi-tion from another. Because the inferences that onedraws about memory or any other cognitive processesin such a study depends crucially on the specicsubtraction that is performed, it is important in PETstudies to hold constant between conditions all vari-ables except for the key parameter of interest (forfurther discussion of the logic of PET studies ofmemory, see Buckner & Tulving (1995)).
3. IMPLICIT AND EXPLICIT MEMORYFOR WORDS : STEM COMPLETION ANDCUED RECALL
An important early PET study of explicit andimplicit memory was reported by Squire et al. (1992).Subjects in Squire et al.'s experiment initially studied alist of familiar words prior to PET scanning (e.g.garden). They were then scanned during a stemcompletion task in which subjects provided the rstword that came to mind in response to three-letterword stems; during one scan stems could be completedwith study list words (priming) and during anotherscan stems could only be completed with new words(baseline). In a separate scan, subjects were given acued recall task in which they were provided withthree-letter stems of study-list words and were asked tothink back to the study list (explicit memory). Thepriming versus baseline comparison revealed decreasedblood ow in extrastriate occipital cortex. This ndingis important theoretically, because a number of investi-gators have argued that visual word priming dependson posterior cortical regions (Gabrieli et al. 1995;Keane et al. 1995; Schacter 1990, 1994; Squire 1992;Tulving & Schacter 1990). However, Squire et al.(1992) also reported signicant blood ow increases inthe right hippocampal formation in the priming condi-tion compared to the baseline condition. This ndingwas surprising because previous research with amnesicpatients characterized by medial temporal lobe damagehas shown clearly that such patients typically exhibitnormal priming (for reviews, see Bowers & Schacter1993; Schacter et al. 1993a,c; Squire 1992; Shimamura1986). Based on these ndings, various investigatorshave argued that priming occurs independently of thehippocampal formation (cf. Gabrieli et al. 1995; Keaneet al. 1995; Moscovitch 1994; Schacter 1990, 1994;Squire 1992, 1994). In addition, Squire et al. (1992)observed right parahippocampal gyrus blood owincreases in the cued recall versus baseline comparison,
implicating the hippocampal region in some aspect ofexplicit recall.Why was the hippocampal region active during
priming in Squire et al.'s (1992) experiment, givenprevious results from amnesic patients indicating thatnormal priming can occur even when the hippocampalformation is damaged? Consideration of Squire et al.'sbehavioural data revealed unusually high levels ofpriming. In fact, the percentage of stems completedwith study list targets in the priming condition (72%)was nearly identical to the percentage of stemcompleted with study list targets in the explicit recallcondition (76%). These ndings raise the possibilitythat priming was contaminated' by explicit memoryfor study list words. It is known that subjects sometimesexplicitly remember target items on nominally implicitmemory tasks, particularly when the targets are easilyaccessible (cf. Jacoby 1991; Schacter et al. 1989). Consis-tent with this idea, during the study phase of theexperiment, subjects saw all target words twice andperformed a `deep' encoding task (pleasantness rating)that promotes high levels of explicit memory. Becauseshort study lists and brief study-test delays were used,all of these factors operating together may have led tothe evident contamination from explicit memory.Given the theoretical importance of the idea that
priming can occur independently of hippocampal acti-vation, we have recently conducted two PETexperiments that examine issues raised by the Squire etal. ndings (Schacter et al. 1996). Our rst experimentaddressed the possible role of contamination' fromexplicit memory in the hippocampal activation thatSquire et al. (1992) observed in their priming condition.To evaluate the issue, we conducted an experiment inwhich we used a `shallow' or non-semantic studytaskcounting the number of t-junctions in a wordin an attempt to eliminate explicit contamination.Previous studies of stem completion priming haveshown that the t-junction counting task can supportrobust priming even when subjects have little or noexplicit memory for the target items (Bowers &Schacter 1990; Graf & Mandler 1984). Accordingly, ifthe priming-related hippocampal activation observedby Squire et al. (1992) reects contamination fromexplicit memory, using the t-junction encoding taskshould eliminate both the explicit contamination andthe associated hippocampal blood ow increases.After carrying out the t-junction task (no scanning
was performed during the encoding task), two separatescans were conducted during which subjects completedthree-letter stems with the rst word that came tomind. One scan consisted of stems that could becompleted with study list words (priming); the otherconsisted of stems that could not be completed withstudy list words (baseline). To maximize the likelihoodof detecting relevant blood ow changes, the sameseries of non-scanned study list, baseline stem comple-tion and primed stem completionwhat we call a`study-test unit'was carried out three times. Orderof conditions and assignment of items to conditionswas completely counterbalanced.The experiment yielded three key results. First,
analysis of behavioural data indicated that we
1690 D. L. Schacter The cognitive neuroscience of memory
Phil.Trans. R. Soc. Lond. B (1997)
successfully eliminated explicit contamination: theabsolute magnitude of the priming eect was compar-able to priming in previous experiments in whichexplicit contamination could be ruled out (Bowers &Schacter 1990; Graf & Mandler 1984). Also, there wasno indication that priming increased across the threestudy-test units, which would have been expected hadsubjects caught on' to the nature of the experimentand begun to engage in explicit retrieval morefrequently as the experiment progressed. Second, therewas no evidence of blood ow increases in the vicinityof the hippocampal formation associated with priming.Third, we replicated Squire et al.'s (1992) nding ofdecreased blood ow in extrastriate occipital cortex inassociation with priming. In addition, we observed avariety of other priming-related blood ow increasesand decreases (see Schacter et al. (1996), for additionalndings, details on PETscanning procedures and othermethodological information).These ndings have potentially important theore-
tical and methodological implications. On thetheoretical side, the data support the position thatperceptual priming occurs independently of the hippo-campal formation. On the methodological side, theSquire et al. (1992) and Schacter et al. (1996) resultsindicate that PET scanning experiments on primingthat use stem completion or similar tasks that areeasily contaminated by explicit memory must also useprocedures that minimize the possibility of explicitcontamination, such as non-semantic study tasks.However, both of these conclusions are compromisedsomewhat by the fact that the failure to observe hippo-campal activity during priming must be viewedcautiously. Negative ndings in PET studies do notcarry the force of positive ndings (e.g. Buckner &Tulving 1995). More importantly, as noted earlier,various studies of explicit retrieval have failed toobserve hippocampal activation (cf. Andreason et al.1995; Shallice et al. 1994; Tulving et al. 1994b). Moreover,in studies that followed-up the initial Squire et al. (1992)nding of hippocampal activation during stem cuedrecall (explicit memory), Buckner et al. (1995) failed toobserve hippocampal blood ow increases whenmodality changed between study and test (i.e. auditorystudy-visual cued recall test) or when the typecase oftarget stimuli changed between study and test (i.e.study words in lower case-test in upper case). If activityin the hippocampal formation is not reliably increasedduring explicit retrieval, then it may not be surprisingthat we failed to detect hippocampal activity duringprimed stem completion performance.In view of these considerations, we performed a
second experiment that examined explicit retrieval onthe stem cued recall test to determine whether wecould observe hippocampal activation. Subjectsstudied a list composed of two dierent types of words.Words in the high recall condition appeared four timesand subjects made semantic encoding judgements aboutthem (judging the number of meanings associated witheach word); words in the low recall condition appearedonce and subjects made a non-semantic judgementabout them (judging the number of t-junctions in aword). After seeing study lists in which both types of
words were presented for 5 s each in a random order(no PET scanning was performed during the encodingtask), stem cued recall was tested during separate scansfor high recall words and low recall words. Three suchstudy-test units were carried out to maximize the sensi-tivity of our procedure. The study-test units werepreceded and followed by two separate scans duringwhich subjects were given word stems that could notbe completed with study-list targets and were asked torespond with the rst word that came to mind. Thisbaseline condition provided a control for simpleperceptual, motor and lexical activities.Behavioural data showed, as expected, that subjects
remembered many more words in the high recallcondition (79%) than in the low recall condition(35%). Results revealed clear evidence of blood owincreases in the hippocampal formation in the highrecall condition, but not in the low recall condition. Inthe high recall minus baseline comparison, we observedbilateral hippocampal activation; in the high recallminus low recall condition, we observed right-sidedhippocampal blood ow increases. These resultsconrm earlier ndings of hippocampal activationduring stem cued recall (Sq...