(1985).

Elects on Health of Exposure toAsbestos (Her Majesty's Stationery Office, London,

84. V. Henderson and P. Enterline, J. Nati. Cancer Inst. 79, 31 (1987). 1985).
85. J. R. Hughes and H. Weill, Am. Rev. Respir. Dis. 133, 5 (1986). 88. M. Corn, K. E. Crump, D. McFee, R. Lee, in preparation.
86. S. Cordier et al., Arch. Envirotn. Health 42, 303 (1987). 89. Battelle Columbus Division, Price Associates, Alliance Technologies Corporation,
87. National Research Council, Committee on Nonoccupational Health Risks of Energy Technology Consultants and Midwest Research Institute, Assessitng Asbestos
Asbestiform Fibers, Asbestiform Fibers: Notnoccupational Health Risks (National Acad- in Public Builditngs (EPA Contr. No. 68-02-4294, draft report for the Exposure
emy Press, Washington, DC, 1984); Report to the U.S. Consumer Product Safety Evaluation Division, 1989).
Commission by the Chronic Hazard Advisory Panel on Asbestos (Consumer 90. H. Weill and J. M. Hughes, Am. Rev. Public Health 7, 171 (1986).
Product Safety Commission, Washington, DC, 1983); Airborne Asbestos Health 91. We thank L. Sabens, V. Kelleher, and M. Bergeron. Research in the laboratory of
Assessmnetnt Update (Environmental Protection Agency, Washington, DC, 1986); E. B.T.M. has been supported by grants from the National Cancer Institute (ROI
D. Acheson and M. J. Gardner, Asbestos: The Control Limit for Asbestos (Her CA33501), National Institute of Environmental Health Sciences (ROl ES03878),
Majesty's Stationery Office, London, 1983); Report on matters of health and safety National Heart, Lung, and Blood Institute (ROI HL39469 and SCOR grant
arising from the use of asbestos in Ontario (Ontario Royal Commission, Ontario 14212), and American Cancer Society (BC 415). We also thank H. C. W. Skinner
Ministry of the Attorney General, Toronto, 1984); R. Doll and J. Peto, Asbestos: for her helpfiil comments.

Priming and Human Memory Systems

Downloaded from www.sciencemag.org on December 3, 2014

ENDEL TULVING AND DANIEL L. SCHACTER

and memory, one that is not procedural, semantic, or episodic. It has

Priming is a nonconscious form of human memory, come to be known as priming (2). Its function is to improve
which is concerned with perceptual identification of identification of perceptual objects. Priming is a type of implicit
words and objects and which has only recently been memory; it does not involve explicit or conscious recollection of any
recognized as separate from other forms of memory or previous experiences. It has affinities to both procedural and seman-
memory systems. It is currently under intense experimen- tic memory. Priming resembles procedural memory in that it
tal scrutiny. Evidence is converging for the proposition enhances perceptual skills. It also resembles semantic memory in that
that priming is an expression of a perceptual representa- it involves cognitive representations of the world and expresses itself
tion system that operates at a pre-semantic level; it in cognition rather than behavior.
emerges early in development, and access to it lacks the The prototypical priming experiment consists of two stages. In
kind of flexibility characteristic of other cognitive memo- the first (study) stage, the subject is presented with a stimulus object
ry systems. Conceptual priming, however, seems to be (target). Target stimuli may comprise words, line drawings of
based on the operations of semantic memory. objects, drawings of faces, and the like. In the second (test) stage,
which may follow the first after an interval that can vary from
seconds to months, the subject is given reduced perceptual informa-
tion about the object and asked to name or categorize it. Reduced
ME EMORY WAS TRADITIONALLY THOUGHT TO BE A UNITARY cues may consist of initial letters or graphemic fragments of words,
faculty of the mind. Recently, however, many researchers partially obliterated words or figures, originally presented faces in a
have adopted the hypothesis that memory consists of a more highly schematized form, or tachistoscopic presentation of
number of systems and subsystems with different operating charac- stimuli. Priming is said to have been demonstrated if the probability
teristics. The problem of what these systems and their properties are, of the identification of the previously encountered targets is in-
and how they are related to one another, now occupies the center creased, or the latency of the identification response is reduced, in
stage in research on memory. comparison with similar measures for nonstudied control items. The
One broad, as yet tentative, organizational scheme distinguishes difference between performance on the target items and the nonstu-
procedural, semantic, and episodic memory (1). Procedural memory died items provides a measure of the magnitude of the priming
underlies changes in skillful performance and appropriate respond- effect.
ing to stimuli; semantic memory has to do with acquisition and use Although priming and other kinds of implicit memory have been
of factual knowledge in the broadest sense; and episodic memory reported from time to time, systematic attempts to explore it began
enables people to remember personally experienced events. The about 10 years ago (3). One of the triggers for the study of priming
domain of procedural memory is behavior, whereas that of semantic turned out to be experiments by Warrington and Weiskrantz (4)
and episodic memory is cognition or thought. Cognitive memory showing that densely amnesic patients, who were severely impaired
systems have the capability of modeling the external world-that is, in their ability to remember recently seen information, exhibited
of storing representations of objects, events, and relations among near-normal learning when they were tested by methods that tapped
them-whereas procedural memory does not have this capability. what we now know is priming. A second stimulus for the study of
Evidence is accumulating about yet another category of learning priming lay in research concerned with the nature of and access to
lexical representations (5). A third source of influence was the
growing interest in the classification of memory into distinctive
E. Tulving is a university professor at the University of Toronto, Toronto, Canada,
M5S lAl. D. L. Schacter is a professor of psychology at the University of Arizona, categories such as episodic and semantic memory (6) and procedural
Tucson, AZ, 85721. and declarative memory (7).
19 JANUARY I990 ARTICLES 301
 We still know relatively little about priming at this early stage of tion of the study material enhances explicit memory, but has little
research. Nevertheless it seems clear that it plays a more important effect on priming (17). On the other hand, the relation between the
role in human affairs than its late discovery would suggest. Although physical format of the studied material and that of the retrieval cues
priming is typically observed only under carefully controlled experi- has relatively little effect on explicit memory, but may greatly affect
mental conditions, similar conditions frequently occur naturally, the magnitude ofpriming (18). Similarly, loss of retention over time
outside the laboratory. It is reasonable to assume, therefore, that often seems to proceed differently in priming and in explicit
priming represents a ubiquitous occurrence in everyday life. memory, possibly because priming is little affected by the kinds of
One remarkable feature of priming is that, unlike other forms of interference manipulations that reduce retention in explicit tasks
cognitive memory, it is nonconscious (8). A person perceiving a (19).
familiar object is not aware that what is perceived is as much an 5) Stochastic independence between successive tests on the same
expression of memory as it is of perception. The fact that people are items. Priming effects are as large for the words that the subjects
not conscious of priming probably accounts for its late discovery. It recognize as having been presented in the study phase as for the
is difficult to study phenomena whose existence one does not unrecognized words (9, 20).
suspect. The conclusions drawn from all of these empirical facts point in
The juxtaposition of its surmised ubiquity in human cognition the same direction. At the psychological level of analysis it looks as if
and the lateness of its discovery, together with its nonconscious normal people faced with ambiguous stimuli are capable of adopting
nature, have inspired an intense experimental and theoretical interest either a perceptual or a memory mode of cognitive operation. In the
in priming. A central issue concerns its nature. What kind of perceptual mode, the operations involve relating the present stimu-
memory is it? lus to the information stored in PRS. This operation reflects
In an early article on priming, we conjectured that the effects we priming; perception is facilitated independently of any recollection
had observed might "reflect the operation of some ... as yet little ofthe learning episode. In the memory mode, the operations consist
understood memory system" other than semantic or episodic memo- of matching the cue information to the information stored in
ry [p. 341 in (9)]. We now present a more seasoned version of this episodic memory. If successful, the product of this match, or
hypothesis. There is additional evidence to support the early idea, synergistic ecphory (21), is recollection of the event of the target
and we can say a bit more about the "little understood" system. item's occurrence in the study list. Our hypothesis is that cognitive
This evidence comes in two major categories. First, the many operations in the perceptual mode involve PRS without any obliga-
different kinds of dissociations between priming and explicit memo- tory engagement of other memory systems, whereas operations in
ry can best be explained by the idea of a pre-semantic perceptual the memory mode depend on the resources of semantic and episodic
system that can operate independently of episodic memory. Second, memory.
neuropsychological studies have revealed dissociations between the At the physiological level of analysis, the evidence points to
reading of words and perceptual identification of other kinds of distinct brain mechanisms subserving priming. At least some of the
objects on the one hand, and semantic knowledge of words and computations involved in and necessary for retrieval of episodic
objects on the other. These dissociations, too, point to a pre- information are disabled when the brain has been damaged, when it
semantic perceptual system that can operate independently of has not yet developed fully or has deteriorated in old age, or when
semantic memory. We believe that priming and perceptual identifi- the influence of certain drugs results in impairment of explicit
cation are expressions of a single perceptual representation system memory. These same computations, however, are not necessary for
(PRS), which exists separately from but interacts closely with other priming. Such a state of affairs signals the distinction between brain
memory systems (10, 11). We present the evidence and reasoning for systems concerned with explicit recollection of past events and
this PRS hypothesis in what follows. primed identification of previously encountered objects-that is,
between episodic memory and PRS.
We next elaborate on the evidence pertaining to two properties of
Phenomena of Prining PRS: (i) access to information in PRS is hyperspecific, probably
because, unlike other cognitive memory systems, it contains no
The evidence in the first category that supports the PRS hypothe- abstract focal traces, and (ii) its domain extends to nonverbal
sis comprises five different kinds of dissociations, involving different objects.
tasks, different tests, different types of retrieval cues, different kinds
of materials, and different subject populations (12), as follows:
1) Intact priming in densely amnesic patients. Amnesic patients
cannot remember the study episode in the priming experiment even Hyperspecificity of Access
after a short interval, yet they show priming effects that are A number of experiments have shown that priming is stochastical-
frequently as large as those in normal subjects (13). ly independent of explicit memory (11, 14, 22). In these experi-
2) Developmental dissociations between priming and explicit ments, joint performance on two successive tests is analyzed item by
memory. Recognition memory in children increases with age, but item for each subject, and the outcome of the experiment is
priming effects can be as large in 3-year-olds as in college students. summarized in terms of relevant data pooled over all subject-items
Similarly, elderly subjects have difficulty with recalling and recogniz- (23).
ing items presented earlier, but their priming effects are indistin- The original discovery of such independence (9, 20) was surpris-
guishable from those of young adults (14). ing, because the typical result of similar explicit memory experi-
3) Drug-induced dissociations. Drugs such as alcohol and sco- ments is one of dependence. For a while it was not known why
polamine reduce performance in explicit recall and explicit recogni- priming measures yielded different results. We now have reason to
tion, but have little or no effect on priming (15). believe that the finding reflects a basic property of PRS: the system
4) Functional independence of priming and explicit memory in seems to work without the kinds of stored focal traces that support
normal subjects. A large number of experiments showing such the operations of semantic and episodic memory.
independence have been reported, involving different kinds of In experiments that support this proposal (24), subjects saw a
priming tasks and tests (16). Thus, for example, semantic elabora- long list of words, including words such as PYRAMID and
302 SCIENCE, VOL. 247
MOSQUITO, and then took two successive fragment completion 1.0- Fig. 1. Stochastic inde-
tests directed at these words. The cues used on the tests were either o_ / pendence between the
the same (for example, "- Y - A - ID" on both tests) or they had outcomes of successive
minimal overlap (for example, "- 0 - Q - - TO" and "- - S - UI - 0" 00
0 0.8
~
* memory tests inan-
i a num-
s ber of experiments. One
on the first and second test, respectively). The task was to complete 2 0.6- * e 0.6- / of the two tests was al-
the fragments, regardless of whether the subject remembered the Co0 ° ways a prnimng test, the
words' earlier presentation. The dependent variable was the degree oc other was either an ex-
of correlation between the two successive tests, measured by Yule's Q 0.4 plicit recognition test (9,
11) or the same priming
Q (a measure of correlation in the fourfold contingency table that . c *
test but with cues differ-
varies from -1.0 to + 1.0). The results showed that with identical 0.2 * ent from those used on
cues on both tests, the correlation between the tests was reliable (Q U c - / the first test (10, 24, 25).
values around 0.90). With different cues on the two tests, however, 00 1 In the graph, the condi-
the correlation showed a drop to zero. 0.oo- 0.2
0 0.4 0.6 0.8 1.0 tional probability of suc-
In further experiments (25), subjects were given successive frag- Probability correct on test 1 cessful item
production of a
on the sec-
target
ment completion tests with nested cues. Again subjects saw a long ond test, given success on the first, is plotted as a function of the simple
list of words, such as AARDVARK and UMBRELLA, and then probability of successful recognition or production of the same target on the
first test. Stochastic independence (absence of item-by-item correlation)
took two successive tests. In the first, they saw three-letter fragments between successive tests is indicated by the diagonal. The stochastic indepen-
(for example "- A - D - - R -" and "U - - R - L -"); in the second dence exhibited by these data differs from the outcomes of explicit memory
test they saw five-letter fragments that included the three letters tests under otherwise identical conditions [for example, figure 1 in (27)].
previously seen ("- A R D - A R -" and "U - B R - L A"). The re-
sults showed that the two successive tests again were independent.
The data from successive tests of implicit memory (10, 24, 25) are ceming the nature of the memory system that subserves priming.
summarized in Fig. 1, along with the data depicting the relation Priming did occur for structurally possible objects, but only if, at the
betwveen implicit measures of memory and explicit recognition (9, time of study, the attention of the subject was directed to the global,
11). Each bivariate data point in Fig. 1 represents the outcome of an three-dimensional structure of the stimuli. Under these conditions,
experiment or a condition in an experiment. The simple probability priming was found to be dissociated from recognition both func-
of success on one test is plotted against the conditional probability tionally and stochastically. Mere exposure to the structurally possible
of success on the other, given success on the first. Stochastic objects, whether at study or in a recognition test preceding the
independence between the tests, indicated by the main diagonal of object decision test, did not produce priming. Nor was any priming
the graph, holds equally for all the experiments. of the structurally possible objects found as a result of semantically
These facts suggest that access to the information that supports rich elaborative coding of the kind that greatly enhances explicit
priming is very inflexible, or hyperspecific. Success or failure of memory. Finally, no evidence for priming of structurally impossible
gaining access to a representation through one cue has no implica- objects was observed.
tions for success or failure of access to the same representation These findings suggest that priming ofobject perception critically
through a different cue, although the tests are otherwise highly depends on perception of objects as structured wholes, implying
reliable (26). Such a state of affairs suggests that priming of words is that PRS is dedicated to the processing of such structural descrip-
not supported by abstract focal traces representing these words, tions (29). The fact that the elaborative encoding task, in which the
because PRS does not contain such traces. If it did, two sets of subjects were required to think of real-world objects ofwhich target
different cues directed at the same set of targets should exhibit at drawings reminded them, did not produce priming in these experi-
least moderate dependency, as they do in explicit memory (27). Thus ments, whereas encoding tasks involving information about the
it looks as if PRS, instead of containing focal traces of words, global three-dimensional structural relations that define each object
contains a multitude of distributed representations of particular did, suggests that PRS operates at a pre-semantic level. The fact that
words, each accessible through specific cues. priming did not occur for impossible objects in these experiments
implies that PRS has evolved to perform only ecologically valid
computations.
We now tum to the second general category of evidence for PRS:
Primimng of Nonverbal Information neuropsychological dissociations between perceptual identification
Most experiments on priming have been conducted with verbal
materials. But priming, dissociated from explicit memory, also
occurs with nonverbal stimuli such as pictures, shapes, and faces Fig. 2. Examples of pos-
(28). sible and impossible ob-
In research on the priming of novel visual objects (11), subjects jects used in experiments
were shown two-dimensional line drawings depicting three-dimen- on object priming (11).
The two upper drawings
sional objects such as those in Fig. 2. All of the objects were novel in depict possible objects
the sense that subjects had never seen them before. Half of the line that could exist in three-
drawings depicted structurally possible objects that can exist in three dimensional form; the
dimensions, and half depicted structurally impossible objects that two lower drawings de-
contained surface and edge violations that ruled out their three- pict impossible objects
that contain structural
dimensional existence. violations that would
In the priming test the subjects made "object decisions." They had prohibit them from ex-
to decide whether a briefly flashed object was possible or impossible. isting in three-dimen-
Subjects were also tested for explicit recognition ofthe target stimuli sional form. Priming
was found under certain
after the object decision test. encoding conditions
The results of these experiments provided additional clues con- with possible objects; it was not found with impossible objects.
I9 JANUARY 1990 ARTICLE 303
and semantic memory. These dissociations, originally observed a that play an important role in explicit remembering. It has been
hundred years ago (30), were seen as related to phenomena of suggested that priming depends on changes in cortical modules that
priming and explicit memory only recently (11). are involved in processing specific attributes ofstimulus information
(42). Inherent in the concept of PRS are suggestions about which
cortical modules may be involved. For example, studies performed
Neuropsychology of Perceptual Representation with the neuroimaging technique of positron emission tomography
have shown that passive reading of familiar words produces selective
Systems bilateral activation in the extrastriate cortex, thus suggesting that
Neuropsychological studies have delineated different types of visual identification of words has an anterior occipital locus (43).
alexias or reading disorders that occur as a consequence of specific This conclusion is consistent with the neuropsychological findings
brain lesions (31). The critical data for our purposes are provided by from patients with selective preservation of the word form system
studies of patients who exhibit the phenomenon of "reading with- (33). Neuropsychological findings also indicate that object identifi-
out semantics" (32). Such patients are able to read aloud printed cation depends on the integrity of posterior cortical areas, especially
words, but they have little or no comprehension of these words in the right hemisphere (39).
when tested with various semantic or associative probes (33). Most Although this evidence bears directly on the neural bases of
important is that their ability to read irregular words, such as perceptual identification of words and objects, it can support only
"cough" or "blood," is almost entirely preserved. Irregular words- indirect inferences concerning the anatomical or physiological un-
unlike regular words-cannot be read on the basis of grapheme-to- derpinnings of priming. If we assume that PRS subserves priming,
phoneme conversion rules; their pronunciations cannot be "sound- then these results provide preliminary hints concerning the likely
ed out." Preserved reading of irregular words thus indicates that brain loci of priming phenomena. However, studies that examine
patients are able to gain access to stored representations of the the matter directly are currently lacking and badly needed.
words' visual forms. This dissociation between intact access to word
form and impaired access to semantic information again supports
the hypothesis that PRS operates at a pre-semantic level (11, 34). If Perceptual Versus Conceptual Priming
the hypothesis is true, these patients, despite their semantic impair-
ments, should show robust priming of words on appropriate tests. We have been concerned in this article with priming as expressed
Although evidence that bears directly on this prediction is not yet on perceptual tasks, in which processing is determined largely by
available, some relevant data are provided by experiments with an physical properties of test cues. However, priming effects have also
alexic patient (P.T.) who exhibits the phenomenon of letter-by- been observed on conceptual tests, in which semantic processing is
letter reading (35). Letter-by-letter readers are unable to recognize required. For example, priming effects on both amnesic patients and
or read printed words unless they identify each letter sequentially normal subjects have been demonstrated for a task in which subjects
(36). Neuropsychological assessment of P.T. suggested that her are given the name of a category, for example "bird," and are asked
reading deficit stems from an impaired ability to transmit letter to produce the first instance that comes to mind, for example,
information in parallel to PRS, which is itself preserved. This "eagle" (44). Similarly, priming effects involving the acquisition of
analysis suggests that P.T., despite her alexia, should nevertheless new associations between unrelated words have been observed on
show priming with words under conditions in which access to PRS cued word stem completion and free association tests in normal
occurs. And it was indeed found that, after letter-by-letter study of a subjects (45), but only after semantic study. A theory that priming is
list of words, P.T. showed priming in a word identification test. The an operation of PRS does not account for such effects.
priming was modality-specific, and it occurred despite P.T.'s great We acknowledge that PRS plays little if any role in the semantic
difficulty in identifying nonstudied words. effects that have been observed on conceptual tests. Indeed, we
Studies of patients with pronounced impairments in recognizing believe that such effects have a different basis than the phenomena
everyday objects provide additional independent evidence for the with which we have been concemed (11). In our view, what has
existence of a pre-semantic system (37). The cognitive deficit of been termed conceptually driven priming reflects a process of
agnosic patients who are severely impaired in visual recognition of semantic learning: the modification of, or adding of new informa-
objects (38) seems to stem specifically from their inability to gain tion to, semantic memory.
access to semantic or associative information about objects from Three pieces of relevant evidence exist. First, conceptual priming
visual input. For example, they have great difficulties with a task in is enhanced by semantic encoding (45). Second, dissociations
which pictures of three objects are shown and the two that perform between performance on perceptual and conceptual tests of priming
the same function must be selected (39). In addition, these agnosic have been observed in studies of college students (46), thereby
patients are impaired when probed with questions concerning the suggesting that different processes support priming on the two types
functions or associative properties of visually displayed objects; yet of tests. Third, amnesic patients can learn some new facts in the
the same patients show relatively intact performance on visual tests absence of episodic memory, although such learning is substantially
of the structural features of objects, such as tests of copying and impaired relative to the performance of normal subjects (47).
judgments that objects seen from different perspectives are identical Additional evidence that semantic learning can be dissociated
(40). It is this contrast between impaired access to semantic knowl- from both episodic memory and perceptual priming (25) was
edge and relatively normal access to structural knowledge that has obtained in research conducted with K.C., an amnesic patient whose
led to the proposal of a system that is separate from, but interacts episodic memory is totally dysfunctional. He does not remember a
with, the semantic system (or systems). Our analysis leads to the single event from his life (48). Any new learning that he exhibits,
prediction that priming should be observed in such patients on tests therefore, must be based on a system or systems other than episodic
that selectively engage PRS. memory. In a recent experiment, K.C. was presented with a long
Relatively little is known about the neural substrates of priming series of complex pictures and three-word phrases (for example, a
(41). Observations of preserved priming in amnesic patients imply picture of a group of fierce-looking native warriors was paired with
that priming is mediated by neural systems outside the medial the phrase STRONGMAN STARTED DYNASTY). The last word
temporal and diencephalic regions that are damaged in amnesia and of the phrase was the target word to be learned by K.C. The
304- SCIENCE, VOL. 247
materials had been constructed in such a manner that the target was understanding of memory processes and mechanisms. The systems
not predictable either from the picture or from the first two words approach combined with appropriate processing theories seems to
of the phrase. After multiple distributed exposures to 64 picture- provide the most direct route to the future (10, 11).
phrase pairs, K.C. was given both perceptual tests that involved
REFERENCES AND NOTES
completing a graphemic fragment of a given target (for example, D-
N-S--) and conceptual tests in which all or a part of the original 1. E. Tulving, Elements of Episodic Memory, (Oxford Univ. Press, New York, 1983);
Human Neurobiol. 6, 67 (1987). Other classificatory schemes of human memory
picture-sentence context, but no fragment, was presented, and K.C. systems include M. K. Johnson, in The Psychology of Leaming and Motivation, G.
had to produce a word that fit the context. Substantial learning, Bower, Ed. (Academic Press, New York, 1983), vol. 17, pp. 81-123; N. J. Cohen,
retained for a number of weeks, was observed on both tasks. The in Neuropsychology ofMemory, L. Squire and N. Butters, Eds. (Guilford, New York,
1984), pp. 83-103; D. L. Schacter and M. Moscovitch, in Infant Memory, M.
critical result, however, was stochastic independence between the Moscovitch, Ed. (Plenum, New York, 1984), pp. 173-216; L. R. Squire, Science
two tasks; the item-by-item correlation between the fragment tests 232, 1612 (1986); L. Weiskrantz, Human Neurobiol. 6, 93 (1987); A. Wilkinson
and C. X. Poulos, in Recent Developments in Alcoholism, M. Galantes, Ed. (Plenurn,
and the conceptual tests was zero. This outcome is consistent with New York, 1987), vol. 5, p. 5; J. D. E. Gabrieli, in The Handbook ofNeuropsycho-
the view that perceptual priming is mediated by PRS, whereas logy, F. Boiler and J. Grafman, Eds. (Elsevier, Amsterdam, in press). Systems views
conceptual priming involves the modification of semantic memory. in the study of memory in animals are represented by R. Hirsh, Behav. Biol. 12,
421 (1974); J. O'Keefe and L. Nadel, The Hippocampus as a Cognitive Map (Oxford
Univ. Press, London, 1978); D. S. Olton, J. T. Becker, G. Handelmann, Behav.
Brain Sci. 2, 213 (1979); M. Mishkin, B. Malamut, J. Bachevalier, in The
Neurobiology ofLeaming and Memory, G. Lynch, J. L. McGaugh, N. M. Weinberger,
Conclusion Eds. (Guilford, New York, 1984), pp. 65-77; M. Mishkin and H. L. Petri, in
Neuropsychology ofMemory, L. R. Squire and N. Butters, Eds. (Guilford, New York,
The evidence we have reviewed converges on the PRS hypothe- 1984), pp. 287-296; R. F. Thompson, Science 233, 941 (1986).
2. The type of priming with which we are concerned in this article is more precisely
sis-that is, priming reflects the enhancement of the neural compu- referred to as direct priming (or repetition priming). We will not discuss indirect
tations and correlated cognitive operations of the perceptual repre- (or semantic or associative priming) for which a separate literature exists; for
sentation system, PRS. This evidence also delineates some proper- example, D. E. Meyer and R. W. Schvaneveldt, J. Exp. Psychol. 90, 227 (1971);
M. J. Farah, J. Exp. Psychol. Hum. Percept. Perform. 15, 188 (1989). The two kinds
ties of PRS: (i) it is concerned with identification of perceptual of priming do not seem to have much in common.
objects, including words; (ii) its neural computations are not 3. For early examples, see U. Neisser, J. Exp. Psychol. 47, 399 (1954); C. Clifton, Jr.,
critically dependent on the brain regions necessary for episodic and J. Verb. Leam. Verb. Behav. 5, 167 (1967); C. N. Cofer, Psychol. Bull. 68, 1
(1967); P. Cramer, Word Association (Academic Press, New York, 1968); W. A.
semantic memory operations; (iii) it develops early and is differen- Winnick and S. A. Daniel, J. Exp. Psychol. 84, 74 (1970). The beginning of the
tially preserved late in life; (iv) its operations are disconnected from current wave of interest is exemplified by D. L. Scarborough, C. Cortese, H. S.
Scarborough, J. Exp. Psychol. Hum. Percept. Perform. 3, 1 (1977); L. L. Jacoby and
consciousness, and its products do not provide a basis for awareness M. Dallas, J. Exp. Psychol. Gen. 110,306 (1981); P. Graf, G. Mandier, P. Haden,
of previous experience; (v) it is relatively immune to the effects of Science 218, 1243 (1982).
drugs that affect other memory systems; (vi) information in it is 4. E. K. Warrington and L. Weiskrantz, Nature 217, 972 (1968); ibid. 228, 628
(1970).
distributed in multiple representations of particular words and 5. K. Kirsner and M. C. Smith, Mem. Cognit. 2, 637 (1974); J. Morton, in The
objects; and (vii) access to representations is hyperspecific. Processing of Visible Language, P. A. Kolers, M. E. Wrolstad, H. Bouma, Eds.
In our discussion of priming we have focused on differences (Plenum, New York, 1979), pp. 259-268.
6. E. Tuiving, in Organization of Memory, E. Tulving and W. Donaldson, Eds.
between priming and other forms of memory. But similarities (Academic Press, New York, 1972), pp. 381-403; M. Kinsbourne and F. Wood,
between them also exist; like other forms of memory, priming in Short-term Memory, D. Deutsch and J. A. Deutsch, Eds. (Academic Press, New
benefits from repetitions, exhibits forgetting over time, and varies York, 1975), pp. 258-291.
7. N. J. Cohen and L. R. Squire, Science 210, 207 (1980); L. R. Squire and N. J.
with the relation between the conditions of encoding and retrieval. Cohen, in Neurobiology of Leaming and Memory, G. Lynch, J. L. McGaugh, N. M.
We think that these and other such "parallel effects" are theoretically Weinberger, Eds. (Guilford, New York, 1984), pp. 3-64.
8. E. Tulving, Can. Psychol. 26,1 (1985); J. F. Kihlstrom, Science 237, 1445 (1987);
uninteresting, since some similarities would be expected of all forms D. L. Schacter, in Varieties of Memory and Consciousness: Essays in Honor of Endel
of memory-otherwise it would be difficult to justify their general Tulving, H. L. Roediger III and F. I. M. Craik, Eds. (Erlbaum, Hillsdale, NJ,
label. 1989), pp. 355-389.
9. E. Tulving et al., J. Exp. Psychol. Leam. Mem. Cognit. 8, 336 (1982).
Nevertheless, some theorists, concentrating on similarities be- 10. PRS replaces the term of quasi-memory (QM) system suggested previously as the
tween priming and other forms of memory, and keen on upholding name of the system mediating priming [C. A. G. Hayman and E. Tulving, J. Exp.
the parsimonious conceptualization of memory as a unitary cogni- Psychol. Learn. Mem. Cognit. 15, 941 (1989)]. We view PRS as a complex system
that comprises several subsystems, including word form, structural description, and
tive system, have argued that priming and explicit forms of memory other subsystems (11).
reflect task-dependent differences in utilization of various aspects of 11. D. L. Schacter, L. A. Cooper, S. M. Delany, J. Exp. Psychol. Gen., in press; D. L.
Schacter, in Development and Neural Bases of Higher Cognitive Function, A. Diamond,
the information stored during a learning episode. These processing Ed. (New York Academy of Science, New York, in press).
theories are usually based on a limited domain of data, such as 12. Different kinds of dissociation can be distinguished, depending on the method of
cognitive psychology experiments with normal subjects. The pro- measuring correlation between two tests. In some experiments, the effects of an
independent variable are compared for a single population of subjects on two tests.
cessing theorists usually argue against the systems approach, claim- If the variable affects performance in one of the two tests, but not in the other, or in
ing that they can explain the results of the experiments they consider the opposite direction in the other, a functional dissociation is said to have been
without postulating different memory systems (49). observed. In another type of experiment, two (or more) populations of subjects-
such as normal people and amnesic patients, or different age groups-may be
We agree that the understanding of processes and mechanisms is compared. If they show different effects on two tests, a neuropsychological,
as vital an objective in the study of priming as it is in the study of developmental, or an alcohol-induced dissociation is said to have been demonstrat-
ed. In yet a third type of comparison, a single population of subjects may be tested
other forms of memory. But we also wish to underscore the twice in succession for the same set of target items. If the results of the two tests
importance of the systems point of view, for two reasons. First, the show no correlation, a contingent dissociation, or stochastic independence,
systems view allows organization and integration of phenomena of between the tests is said to have occurred. The logic of the dissociation experiment
is straightforward. If the two tests that are being compared represent the workings
priming in a manner that has not been realized within monolithic of a single set, or a highly correlated set, of processes, then a comparison of those
processing theories. Second, and more important, it is becoming tests should reveal no substantial dissociation. If it does, the processes involved
must be different.
increasingly clear that there are no universal principles of memory 13. E. K. Warrington and L. Weiskrantz, Neuropsychologia 12, 419 (1974); M.
and that facts discovered about one form of memory need not hold Moscovitch, in Human Memory and Amnesia, L. S. Cermak, Ed. (Erlbaum,
for other forms. This is why systematic classification of memory Hillsdale, NJ, 1982); P. Graf, L. R. Squire, G. Mandier, J. Exp. Psychol. Learn.
Mem. Cognit. 10, 164 (1984); L. S. Cermak et al., Neuropsychologia 23, 615 (1985);
systems, both psychological and physiological, is an essential prereq- D. L. Schacter, Ann. N. Y. Acad. Sci. 444, 41 (1985); for review see A.
uisite for the successful pursuit of the empirical and theoretical Shimamura, Q. J. Exp. Psychol. 38A, 619 (1986).

19 JANUARY 1990 ARTICLE 305

14. A. J. Parkin and S. Streete, Br. J. Psychol. 79, 361 (1988); L. L. Light, A. Singh, J. 32. T. Shailice and E. Warrington, Q. J. Exp. Psychol. 35A, 111 (1983).
L. Capps, J. Clin. Exp. Neuropsychol. 8, 62 (1986). 33. E. Funnell, Br. J. Psychol. 74, 159 (1983); G. Sartori and R. Job, in The Cognitive
15. S. Hashtroudi, E. S. Parker, L. E. DeLisi, R. J. Wyatt, S. A. Mutter, J. Exp. Neuropsychology of Language, M. Coltheart, G. Sartori, R. Job, Eds. (Eribaum,
Psychol. Learn. Mem. Cognit. 10, 156 (1984); E. S. Parker, R. Schoenberg, B. S. Hilisdale, NJ, 1987), pp. 59-78; M. F. Schwartz, E. M. Saffran, 0. S. M. Marin, in
Schwartz, E. Tulving, Bull. Psychon. Soc. 21, 363 (1983); M. J. Nissen, D. S. (31), pp. 259-269.
Knopman, D. L. Schacter, Neurology 37, 789 (1987); M. D. Kopelman and T. H. 34. Some evidence suggests that word form system is a subsystem of PRS; for exampk,
Corn, Brain 111, 1079 (1988). E. K. Warrington and T. Shallice, Brain 103, 99 (1980), and it has been so treated
16. D. L. Schacter, J. Exp. Psychol. Learn. Mem. Cognit. 13, 501 (1987); A. elsewhere (11).
Richardson-Klavehn and R. A. Bjork, Annu. Rev. Psychol. 39, 475 (1988). 35. D. L. Schacter, S. Rapcsak, A. B. Rubens, M. Tharan, J. Laguna, in preparation.
17. P. Graf and G. Mandler, J. Verb. Leam. Verb. Behav. 23, 553 (1984); D. L. 36. K. Patterson and J. Kay, Q. J. Exp. Psychol. 34A, 411 (1982); see also (34).
Schacter and P. Graf, J. Exp. Psychol. Leam. Mem. Cognit. 12, 432 (1986). 37. E. K. Warrington, Philos. Trans. R. Soc. London Ser. B 298, 15 (1982); M. J.
18. H. L. Roediger III and T. A. Blaxton, in Memory and Cognitive Processes: The Riddoch, G. W. Humphreys, M. Coltheart, E. Funnell, Cognit. Neuropsychol. 5, 3
Ebbinghaus Centennial Conference, D. S. Gorfein and R. R. Hoffman, Eds. (Erlbaum, (1988). The pre-semantic system for objects has been referred to as the structural
Hillsdale, NJ, 1987), pp. 349-379; Mem. Cognit. 15, 379 (1987). description system in light of evidence that suggests that it may be a subsystem of
19. L. L. Jacoby, J. Exp. Psychol. Learn. Mem. Cognit. 9,21 (1983); S.-I. Komatsu and PRS (11, 40).
N. Ohta, Jpn. Psychol. Res. 26, 194 (1984); P. Graf and D. L. Schacter, J. Exp. 38. R. Bauer and A. Rubens, in Clinical Neuropsychology, K. M. Heilman and E.
Psychol. Leam. Mem. Cognit. 13, 45 (1987); S. A. Sloman, C. A. G. Hayman, N. Valenstein, Eds. (Oxford Univ. Press, New York, 1985).
Ohta, J. Law, E. Tulving, ibid. 14, 223 (1988). 39. E. K. Warrington and A. M. Taylor, Perception 7, 695 (1978).
20. L. L. Jacoby and D. Witherspoon, Can. J. Psychol. 36, 300 (1982). 40. E. K. Warrington, Q. J. Exp. Psychol. 27, 635 (1975); M. J. Riddoch and G. W.
21. L. S. Cernak, in Varieties of Memory and Consciousness: Essays in Honor of Endel Humphreys, Cognit. Neuropsychol. 4, 131 (1987).
Tulving, H. L. Roediger III and F. I. M. Craik, Eds. (Erlbaum, Hillsdale, NJ, 41. L. Weiskrantz, in Memory Systems of the Brain, N. Weinberger, J. McGaugh, G.
1989), pp. 121-131. Lynch, Eds. (Guilford, New York, 1985), pp. 380-418.
22. E. Tulving, Am. Psychol. 40,385 (1985); D. B. Mitchell and A. S. Brown, J. Exp. 42. L. R. Squire, Memory and Brain (Oxford Univ. Press, New York, 1987).
Psychol. Learn. Mem. Cognit. 14, 213 (1988); C. A. G. Hayman and E. Tulving, 43. S. E. Petersen, P. T. Fox, M. I. Posner, M. Mintum, M. E. Raichle, Nature 331,
ibid. 15, 228 (1989). 585 (1988).
23. A methodological problem, priming of target items on the first test by virtue of 44. P. Graf, A. Shimamura, L. R. Squire, J. Exp. Psychol. Learn. Mem. Cognit. 11, 385
their successful retrieval, was circumvented through the use of the so-called (1985); J. Kihlstrom, Cognit. Psychol. 12, 227 (1980).
reduction method. The fourfold contingency tables, from which the relevant 45. P. Graf and D. L. Schacter, J. Exp. Psychol. Learn. Mem. Cognit. 11, 501 (1985);
measures were calculated, were constructed without relying on any "contaminated" D. L. Schacter and S. M. McGlynn, Am. J. Psychol. 102, 151 (1989); S. B.
data; E. Tulving and M. J. Watkins, Psychol. Rev. 82, 261 (1975); M. J. Watkins Hamann, thesis, University of Toronto (1989).
and A. K. Todres, J. Verb. Learn. Verb. Behav. 17, 621 (1978). [See also (10)]. 46. T. Blaxton, J. Exp. Psychol. Learn. Mem. Cognit. 15, 657 (1989).
24. C. A. G. Hayman, C. A. Macdonald, E. Tulving, unpublished data. 47. D. L. Schacter, J. L. Harbluk, D. R. McLachlan, J. Verb. Leam. Verb. Behav. 23,
25. E. Tulving, C. A. G. Hayman, C. A. Macdonald, in preparation. 593 (1984); E. L. Glisky, D. L. Schacter, E. Tulving, Neuropsychologia 24, 313
26. Hyperspecificity of access was originally discussed in relation to studies of implicit (1986); A. P. Shimamura and L. R. Squire, J. Exp. Psychol. Leam. Mem. Cognit.
learning in amnesic patients; for example, D. L. Schacter, in Memory Systems of the 13, 464 (1987); E. L. Glisky and D. L. Schacter, Neuropsychologia 27, 107 (1989).
Brain, N. Weinberger, J. L. McGaugh, G. Lynch, Eds. (Guilford, New York, 48. E. Tulving, D. L. Schacter, D. R. McLachlan, M. Moscovitch, Brain Cognit. 8, 3
1984), pp. 351-379. It has now been observed by others; for example, J. M. (1988); E. Tulving, Am. Sci. 77, 361 (1989).
Gardiner, A. J. Dawson, E. A. Sutton, Am. J. Psychol. 102, 295 (1989). 49. Representative processing accounts of priming can be found in L. L. Jacoby, J.
27. A. J. Flexser and E. Tulving, Psychol. Rev. 85, 237 (1978); see also (10). Verb. Learn. Verb. Behav. 22, 485 (1983); D. L. Nelson, J. J. Canas, M. T. Bajo, P.
28. W. R. Kunst-Wilson and R. B. Zajonc, Science 207, 557 (1980); V. Bruce and T. D. Keelean, J. Exp. Psychol. Leam. Mem. Cognit. 13, 542 (1987); L. L. Jacoby, in
Valentine, Br. J. Psychol. 76, 373 (1985); S. Bentin and M. Moscovitch, J. Exp. Remembering Reconsidered: Ecological and Traditional Approaches to the Study of Memory,
Psychol. Gen. 117, 148 (1988); G. Musen and A. Treisman, J. Exp. Psychol. Leam. U. Neisser and E. Winograd, Eds. (Cambridge Univ. Press Cambridge, 1988), pp.
Mem. Cognit., in press. 145-177; R. Ratcliff and G. McKoon, Psychol. Rev. 95, 385 (1988); H. L.
29. D. Marr and H. K. Nishihara, Proc. R. Soc. London Ser. B 200, 269 (1978). Roediger III, M. S. Weldon, B. H. Challis, in Varieties ofMemory and Consciousness:
30. The important distinction between apperceptive and associative agnosia was first Essays in Honor of Endel Tulving, H. L. Roediger III and F. I. M. Craik, Eds.
proposed by H. Lissauer, Arch. Psychiatrie Nervenkrankh. 21, 222 (1890) [English (Erlbaum, Hillsdale, NJ, 1989), pp. 3-41; M. S. Humphreys, J. D. Bain, R. Pike,
translation by T. Shallice and M. Jackson, Cognit. Neuropsychol. 5, 153 (1988)]; for Psychol. Rev. 96, 208 (1989); D. Witherspoon and M. Moscovitch, J. Exp.
subsequent developments see A. Ellis and A. Young, Human Cognitive Neuropsycho- Psychol. Learn. Mem. Cognit. 15, 22 (1989).
logy (Erlbaum, Hillsdale, NJ, 1988); T. Shallice, From Neuropsychology to Mental 50. Supported by Natural Sciences and Engineering Research Council of Canada
Strncture (Cambridge Univ. Press, Cambridge, 1988). (A8632), National Institutes of Aging (1 RO1 AGO8441-01), and a Special
31. M. Coltheart, K. Patterson, J. C. Marshall, Eds., Deep Dyslexia (Routledge and Research Program Grant from the Connaught Fund, University of Toronto. We
Kegan Paul, London, 1980). thank Larry R. Squire for constructive criticisms.

306 SCIENCE, VOL. 247