Encoding Specificity In Memory Research Paper

Nowadays, it is easy to dismiss this notion as obvious, but at the time of its introduction around 1970 it was more commonly dismissed as obviously wrong. Key to this change in fortunes has been an increasing appreciation of the role of encoding, and in particular idiosyncratic encoding, in the memory process.

1. Idiosyncratic Encoding

Tulving and his associates were quick to credit Richard Semon with a biological rendition of the encoding specificity principle, and indeed of their entire conception of the episodic memory system (Schacter et al. 1978). But prior to this acknowledgment Semon was virtually unknown to psychology. By the time his views were translated into English (Semon 1921), mainstream American experimental psychology was taking an objectivist turn, and before long encoding was thought to add nothing to what could be objectively discerned from the stimulus conditions.

There were, of course, other approaches to psychology, some of which underscored the subjective nature of encoding. For example, gestalt psychologists used ambiguous figures to demonstrate that one and the same stimulus could result in distinctly different perceptions, and moreover they conceptualized memory for such perceptions as embodied in memory traces. Nevertheless, it was not until the adoption of the information-processing metaphor as the dominant model of cognitive psychology in the second half of the twentieth century that the concept of encoding assumed its current theoretical prominence.

Among the many demonstrations of the role of encoding in memory that came with the dawning of the information-processing era was Conrad’s (1964) finding that an error in the written reproduction of a short visual sequence of letters tended to be acoustically similar to the correct letter. Thus, the letter S was 65 times more likely to be recalled as an F than as a B, despite not looking more like an F than a B. The implication is that performance in this task is mediated by inner speech, even though the letters are presented and reported silently. Such encoding is perhaps virtually universal under these conditions, but under other conditions encoding can be willful and idiosyncratic. The cocktail party phenomenon, for example, illustrates an ability to attend selectively to just one of multiple speakers. On the other hand, that such willful control is imperfect is shown by a special sensitivity to our own name spoken in what would appear to have been an unattended conversation. In this case, encoding is unwilled, yet clearly idiosyncratic.

Throughout the 1960s Tulving, among others, emphasized the hitherto neglected recall phase of remembering. Recall of an episode at any given time, he argued, depended on what was being encoded at that time, as well as on how the episode had been encoded into its memory trace. More particularly, he came to promulgate the encoding specificity principle—the dictum that recall of the episode depends on the relation between current encoding and the information available in the episode trace. Thus, the encoding specificity principle is best illustrated by experiments in which control is brought to bear on the encoding conditions prevailing at both the study and test stages of a memory procedure. A few examples warrant review.

2. Item-Specific Context Effects

In a seminal demonstration of encoding specificity, Tulving and Osler (1968) presented a list of ‘target’ words (e.g., STOMACH, HEALTH, EAGLE), each being shown alone for some subjects (STOMACH), in the presence of a weakly associated ‘cue’ word for others (empty – STOMACH), and in the presence of two weakly associated cue words for yet others (empty, hurts – STOMACH). They then tested for recall of the target words. Of the subjects in each of the three study conditions, some were given free recall instructions (i.e., no specific cues), others were given one of the two cue words of each target word (e.g., empty – ?), and yet others were given both of the cue words (e.g., empty, hurts – ?). Subjects who saw cues at neither study nor test recalled, on average, 44 percent of the target words. The various groups of subjects for whom the cue, or at least one of the two cues, presented at study was also presented at test recalled an average of 60–62 percent of the target words, depending on the group. Subjects for whom no cue word presented during study appeared at test, or for whom the target words had been presented alone at study and tested with one or both of their cues, recalled an average of 29–36 percent of the target words, again depending on group. The authors drew the then-provocative conclusion that, ‘specific retrieval cues facilitate recall if and only if the information about them and about their relation to the (to-be-remembered) words is stored at the same time as the information about the membership of the (to-be-remembered) words in a given list’ (Tulving and Osler 1968, p. 599).

With the distinction between episodic memory and general knowledge yet to be drawn, it would have been natural to attribute the ineffectiveness of the extralist cues (i.e., words not presented at study but presented at test) to the weakness of their associations to the target words, and the effectiveness of the intralist cues to the temporary boosting of these associations in the study phase of the procedure. Evidence contrary to this account was provided by Thomson and Tulving (1970). In one of their experiments, target words were studied either alone (e.g., FLOWER) or in the context of a weakly associated cue ( fruit – FLOWER), and tested under conditions of free recall (i.e., in the absence of explicit cues), or in the presence of the weakly associated cue ( fruit – ?) or of a strongly associated extralist cue (bloom – ?), with a different group of subjects serving under each of the six combinations of study and test conditions. When no context words had been present at study, the proportion of targets recalled was, relative to conditions of free recall (.49), increased by the strong cues (.68) but not by the weak cues (.43). By contrast, when weakly associated context words had been present at study, the proportion of targets recalled was, relative to conditions of free recall (.30), increased by the weak cues (.82) but not by the strong cues (.23). This pattern of results is not readily reconciled with what Thomson and Tulving called the hypothesis of associative continuity, whereby no distinction is made between episodic memory and general knowledge. It is, on the other hand, in keeping with a principle of encoding specificity operating within the realm of a distinct episodic memory system.

The critical finding in each of these two studies is the study–test interaction, which indicates that the relative effectiveness of different test conditions depends on the conditions under which the items were studied. Such interactions constitute the evidence for encoding specificity, and more generally for the premise of a distinct episodic memory system. In one of the many subsequent illustrations of study–test interactions (Fisher and Craik 1977, see also Morris et al. 1977), target words were presented in the context of either semantically or phonemically related words (sleet – HAIL or pail – HAIL). Recall was then measured in response to extralist cues that also were either semantically related or phonemically related to the target words (snow or bale). The semantically related cues were more effective than the phonemically related cues for words studied in the context of semantically associated words (.36 versus .22), but not for words studied in the context of phonemically associated words (.16 versus .18). This interaction is at variance with one particularly influential idea of the time, namely Craik and Lockhart’s (1972) notion that memory for an item is determined by the depth to which it was processed, with semantic encoding being assumed to reflect greater depth than phonemic encoding. Fisher and Craik suggested supplementing this idea with that of encoding specificity.

That specificity of encoding can occur entirely within the semantic realm is implied by the finding that a recall cue for an unambiguous target word may be more or less effective depending on its consonance with the context of the target word at study. Thus, something hea y is a more effective cue than something with a nice sound for the target word PIANO presented in the study sentence, ‘The man lifted the piano,’ but not for the same target word presented in the study sentence, ‘The man tuned the piano’ (Barclay et al. 1974).

Perhaps the most influential support for the encoding specificity principle has been the phenomenon of recognition failure of recallable words. In his quest for a truly general theory of episodic memory, Tulving came to relinquish the widely held assumption of a qualitative difference between the ‘copy’ cue in a recognition test (i.e., the target word itself presented as a test item) and other kinds of memory cue, and suggested instead that the copy cue was, like any other cue, fallible, and so could prove ineffective even though another cue would be effective (Tulving 1983). To evaluate this suggestion, Tulving and Thomson (1973) presented three consecutive lists of 24 weakly associated word pairs (e.g., ground COLD, head LIGHT, bath NEED), and following each of the first two lists they re-presented in a different order the left-hand words as recall cues for the target right-hand words. The subjects expected to be tested the same way for the third list, and indeed they were, but not until they had engaged in two other tasks. First, they free associated to each of a sequence of stimulus words (e.g., want, dark, hot). Unknown to the subjects, each of these stimulus words had been selected as likely to yield one of the target words from the just-studied list (namely, NEED, LIGHT, and COLD, respectively). The subjects then looked over their free associates and circled any they recognized as a target from the study list. Finally, they were given the recall test they had expected, with the left-hand words from the third study list presented as cues for their respective paired words. A substantial proportion of the words produced in the free association task went unrecognized and yet were produced in the ensuing recall test. Across three experiments, the probability of a target free associate being recognized as a target was only about half that of its being recalled in response to the weak associate with which it had been paired. Subsequent research has shown this phenomenon of recognition failure of recallable words to be remarkably robust. Tulving and his associates have interpreted it in terms of the encoding specificity principle: the targets were encoded into memory traces in such a way that the traces tended to be more responsive to the context words with which the target words had been studied than to copies of the targets themselves.

This interpretation implies that level of performance on a recognition test may well depend on the congruence between the cognitive environments at study and test. Just such a relation was demonstrated at the very time the encoding specificity principle was being developed (Light and Carter-Sobell 1970). Subjects studied sentences that included a capitalized two-word phrase (e.g., ‘The STRAWBERRY JAM tasted great’) in anticipation of a memory test for these phrases. In fact, they were given a recognition test in which the test items were word pairs and the task was to decide whether the second word of each pair (JAM) had been included in the study list. Recognition of a target word was much more probable if its context was the same as at study (STRAWBERRY JAM) than if it was different (TRAFFIC JAM). Across various subconditions and experiments, probability of recognition, corrected for guessing, ranged between two and three times greater in the same context condition than in the different context condition.

Of course, jam is a homograph, as were the other target words in this study. According to the encoding specificity principle, however, context may affect recognition of any word, and indeed context effects in the recognition of words that are not transparently homographic have been demonstrated. Thus, Tulving and Thomson (1971) found that a target word (e.g., WATER) studied alone was more likely to be picked out in a recognition test when presented alone (0.70) than when presented in the context of either a weakly (WHISKEY) or strongly (LAKE) associated word (0.49 and 0.59), and that a target word studied in the context of a weakly or strongly related context word (WHISKEY–WATER or LAKE–WATER) was more likely to be picked out at test when presented in the same context (0.85 and 0.88) than when presented alone (0.77 and 0.70) or in a different context (0.58 and 0.66).

Context effects in recognition memory have been shown to extend to nonverbal stimuli, such as faces. In one study (Watkins et al. 1976), subjects were shown photographs of unfamiliar faces, each paired with a descriptive phrase (e.g., keeps tropical fish, drives an Italian sports car, works in cancer research). The phrases and faces were then shown in a recognition test, with half paired as before and half re-paired. The subjects’ task was simply to identify the faces regardless of context. Probability of identification was greater for same-context faces than for re-paired faces (0.84 versus 0.71).

In each of these examples, as with most other studies cited as support of the encoding specificity principle, encoding was controlled by manipulation of the physical environment. According to the thinking behind the encoding specificity principle, however, the effects of the physical environment are mediated by the internal, or cognitive, environment. Consistent with this premise, it should be possible to show, for example, context effects in recognition memory by having the subjects impose contexts in a purely mental way, and such effects have indeed been demonstrated. For example, if instructions in the study phase of a recognition memory experiment call for the mental filling in of irregular shapes with specified colors, then subsequent identification of the shapes, regardless of color, tends to be more accurate if they are shown filled in with the imagined color than with another color (Watkins and Schiano 1982).

3. General Context Effects

Memory may depend on the reinstatement of, not only item-specific contexts, but also general contexts shared by many or all of the target items (Davies and Thomson 1988). For example, there have long been claims of a beneficial effect of using the same room for study and test (Abernathy 1940, Smith et al. 1978; but see Saufley et al. (1985) for an example of a failure to replicate this effect). Similarly, an item encountered in the presence of a particular odor is subsequently more likely to be remembered in the presence of the same odor than in the presence of a different odor (Cann and Ross 1989, Angleton and Waskett 1999). In the same vein, recall has been found to be more successful when study and test are both on land or both underwater than following a study-to-test switch in these environments (Godden and Baddeley 1975). Also, memory tends to be more effective when drug or mood state at test matches that at study. And, like item-specific context effects, general context effects can be psychogenic. For example, adopting the perspective of a burglar or of a home buyer while recalling a story about an old house has been shown to differentially affect recall of such items as ‘rare coin collection’ and ‘leaky roof’ (Anderson and Pichert 1978).

4. Criticisms

Not all aspects of Tulving’s account of encoding specificity are widely accepted. For example, its logic notwithstanding, there is little appeal to Tulving’s (1983) argument that the interaction of study and test conditions precludes characterizing one episode as more memorable than another. We do not deny that Mother Teresa was a better person than Adolf Hitler just because in some ways (e.g., as an orator, artist, or military strategist) she probably was not, so why should we deny that some memories (e.g., my memory for the birth of my first child) are stronger than others (e.g., my memory for putting on my socks that same day)? More importantly, the very idea of episodes being encoded into discrete traces within a distinct episodic memory system is problematic. For one thing, it is an instantiation of Tulving’s notion of availability. This notion has never been adequately operationalized (Watkins 1978), and consequently the distinction between encoding specificity and such process oriented notions as transfer-appropriate processing (Morris et al. 1977) is shaky at best. Also, the relation among traces has not been worked out. If, for example, memory for an item from a study list is represented in a trace, then how is memory for the list represented?

It is important to be clear that such problems refer to a particular conception of encoding specificity, and not to the idea itself. Indeed, in one guise or another, encoding specificity has become essential to contemporary interpretations of episodic memory.

Bibliography:

1. Abernathy E M 1940 The effect of changed environmental conditions upon the results of college examinations. Journal of Psychology 10: 293–301
2. Anderson R C, Pichert J W 1978 Recall of previously unrecallable information following a shift in perspective. Journal of Verbal Learning and Verbal Behavior 17: 1–12
3. Angleton J P, Waskett L 1999 The ability of odours to serve as state-dependent cues for real world mnemonics: can Viking smells aid the recall of Viking experiences? British Journal of Psychology 90: 1–7
4. Barclay J R, Bransford J D, Franks J J, McCarrell N S, Nitsch K 1974 Comprehension and semantic flexibility. Journal of Verbal Learning and Verbal Behavior 13: 471–81
5. Cann A, Ross D A 1989 Olfactory stimuli as context cues in human memory. American Journal of Psychology 102: 91–102
6. Conrad R 1964 Acoustic confusions in immediate memory. British Journal of Psychology 55: 75–84
7. Craik F I M, Lockhart R S 1972 Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior 11: 671–84
8. Davies G M, Thomson D M (eds.) 1988 Memory in Context: Context in Memory. Wiley, Chichester, UK
9. Fisher R P, Craik F I M 1977 Interaction between encoding and retrieval operations in cued recall. Journal of Experimental Psychology: Human Learning and Memory 3: 701–11
10. Godden D, Baddeley A D 1975 Context-dependent memory in two natural experiments: On land and under water. British Journal of Psychology 66: 325–31
11. Light L L, Carter-Sobell L 1970 Effects of changed semantic contexts on recognition memory. Journal of Verbal Learning and Verbal Behavior 9: 1–11
12. Morris C D, Bransford J D, Franks J J 1977 Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior 16: 519–33
13. Saufley W H, Otaka S R, Bavaresco J L 1985 Context effects: Classroom tests and context independence. Memory and Cognition 13: 522–8
14. Schacter D L, Eich J E, Tulving E 1978 Richard Semon’s theory of memory. Journal of Verbal Learning and Verbal Behavior 17: 721–43
15. Semon R 1921 The Mneme. Allen and Unwin, London
16. Smith S M, Glenberg A M, Bjork R A 1978 Environmental context and human memory. Memory and Cognition 6: 342–53
17. Thomson D M, Tulving E 1970 Associative encoding and retrieval: Weak and strong cues. Journal of Experimental Psychology 86: 255–62
18. Tulving E 1972 Episodic and semantic memory. In: Tulving E, Donaldson W (eds.) Organization of Memory. Academic Press, New York
19. Tulving E 1983 Elements of Episodic Memory. Oxford University Press, New York
20. Tulving E, Osler S 1968 Effectiveness of retrieval cues in memory for words. Journal of Experimental Psychology 77: 593–601
21. Tulving E, Thomson D M 1971 Retrieval processes in recognition memory: Effects of associative context. Journal of Experimental Psychology 87: 116–24
22. Tulving E, Thomson D M 1973 Encoding specificity and retrieval processes in episodic memory. Psychological Review 80: 352–73
23. Watkins M J 1978 Theoretical issues. In: Gruneberg M M, Morris P E (eds.) Aspects of Memory. Methuen, London
24. Watkins M J, Ho E, Tulving E 1976 Context effects in recognition memory for faces. Journal of Verbal Learning and Verbal Behavior 15: 505–17
25. Watkins M J, Schiano D J 1982 Chromatic imaging: An effect of mental colouring on recognition memory. Canadian Journal of Psychology 36: 291–99