Proust nose best: Odors are better cues of autobiographical memory

Simon Chu Proust nose best: Odours are better cues of autobiographical memory Simon Chu & John J. Downes Department of Psychology, University of Liverpool Running Head : PROUST NOSE BEST Address for correspondence : John J. Downes, Department of Psychology, University of Liverpool, Eleanor Rathbone Building, Bedford Street South, Liverpool L69 7ZA, England Telephone : +44 (0)151 794 1477 Fax : +44 (0)151 794 2945 E-mail : [email protected] Acknowledgements Thanks are extended to Richard Cotler, Daniel Farrelly, Alison Gall, Rosie Hales, Emily Highfield and Kevin Haigh for assistance in data collection. We also thank David C. Rubin and two anonymous reviewers for helpful critiques on an earlier draft of this paper. This research was supported by Unilever Research and by an Economic and Social Research Council studentship award to SC. 1 Simon Chu Abstract The Proust phenomenon is an enduring piece of folk wisdom which asserts that odours are particularly powerful autobiographical memory cues. We provide a more formal exposition of this phenomenon and test this in two experiments using a novel double-cueing methodology designed to negate less interesting explanations for the phenomenon.. In both studies, recall of an autobiographical event was initially cued by a verbal label (an odour name) for a fixed period, following which a second, extended recall attempt was cued by the same verbal label, the relevant odour, an irrelevant odour or a visual cue. The focus of Experiment 1 was participants’ ratings of the emotional quality of their autobiographical memories, whereas in Experiment 2, content analysis was employed to determine the quantity of information in participants’ recollections. Results revealed that odour-cued autobiographical memories were reliably different in terms of qualitative ratings and reliably superior in the amount of detail yielded. Moreover, visual cues and incongruent olfactory cues appeared to have a detrimental effect on the amount of detail recalled. These results support the proposal that odours are especially effective as reminders of past experience. In explaining these effects, we draw upon Conway’s (1992) theoretical model of autobiographical memory and expand on Conway’s conjecture that sensory cues have direct access to event-specific knowledge in the autobiographical knowledge base. 2 Simon Chu The belief that odours are especially evocative reminders of past experiences can be traced to the literary anecdote described by Marcel Proust, in which he was vividly reminded of his childhood when he dipped a madeleine biscuit into his tea (Proust, 1922/1960). The socalled Proust phenomenon has been defined more formally in a number of different ways (cf. Chu & Downes, 2000). These vary from the belief that odours themselves are not forgotten as rapidly as other perceptual events (Engen & Ross, 1973), to the belief that odours evoke older memories than other stimuli (Rubin, Groth & Goldsmith, 1984), to the belief that odours evoke more emotionally loaded recollections (Herz & Cupchik, 1995). Engen and Ross (1973), for example, observed that while memory for odours relative to stimuli from other perceptual modalities is initially poor, subsequent testing reveals very little forgetting. This led them to conjecture that memory for odours differs in important ways from that for stimuli from other modalities, a view which has since been corroborated by a sizeable body of empirical evidence (for reviews, see Schab, 1991; Herz & Engen, 1996; White, 1998), and which has led to the stronger claim that a separable memory system for olfaction exists, and even that odour memory itself may dissociate into different component subsystems (Lehrner, Walla, Laska & Deecke, 1999). Findings concerning the relative persistence of odours in memory are clearly important to proponents of Proustian retrieval because, for a stimulus to be an effective retrieval cue, it must first be present as part of the memory trace. However, the development of testable hypotheses about the Proustian phenomenon will necessarily involve more detailed consideration of the unique ways in which odours interact with other event information to produce mnemonic representations which are more detailed, more stable and more emotionally loaded. There have been surprisingly few attempts at investigating the Proust phenomenon. Some of these have examined experimentally induced memories, and reported them to be significantly more emotionally-toned when cued by an odour compared to the odour’s verbal label (Herz & Cupchik, 1995), or the sight or touch of the ‘source object’ (Herz, 3 Simon Chu 1996). Other studies have examined naturally occurring autobiographical memories and are therefore of more direct relevance to the Proust phenomenon. Herz and Cupchik (1992) concluded that odour-evoked memories tended to be highly emotional, vivid, specific, rare and relatively old. However, the main purpose of that study was to determine if sex differences were evident in these qualitative aspects of autobiographical memory retrieval, and odours were not compared with other types of sensory cue. This comparison was made by Rubin et al. (1984) who gave either odours, verbal labels or photographs to participants and asked them to briefly describe and rate the retrieved autobiographical memories. Compared with the other cue types, odour–cued memories were rated as more pleasant and had been thought of and spoken of less often. Rubin et al’s (1984) study highlights the obvious methodological approach to evaluating hypotheses which claim that autobiographical memories cued by odours differ in important ways from those cued by other sensory cues, an approach we refer to as the single cue comparison method. The inclusion of a verbal label condition is important because it allows one to evaluate an alternative verbal mediation hypothesis. According to this, it is the generated odour name, rather than the odour itself, which operates as the retrieval cue. Finding that odour cues are significantly different on some critical dimension(s) from labels and from other types of sensory cue would therefore both negate this less interesting hypothesis, and provide support for the Proust phenomenon. Although often not made explicit, the phenomenon has been commonly interpreted as a retrieval effect. A more formal version of this, which we refer to as the differential cue affordance hypothesis, is as follows: Cues of many different types become associated with autobiographical memories, but may differ in terms of their associated affordance value, a hypothetical measure of how efficiently event details can be accessed. Proustian phenomena are explained by the higher cue affordance values of olfactory compared to other cue types. 4 Simon Chu It must be realised, however, that other interpretations of the finding that odours are better cues than stimuli from other sensory modalities are possible. In contrast to the latter retrieval based hypothesis, factors operating at encoding may be important in producing the effect. One possibility, which we refer to as the differential encoding bias hypothesis, follows: According to this hypothesis, autobiographical memories differ in terms of the complexity of the underlying representations; that is, the number of event details which have been encoded and consolidated. However, the more complex a particular memory representations is, the greater the likelihood that ‘peripheral’ details were initially encoded and consolidated. Olfactory details, then, which are peripheral in the sense that they do not usually alter the meaning or interpretation of an event (cf. Baddeley, 1982), may tend to be encoded only with more complex autobiographical memories. It follows that, other things being equal, olfactory cues will lead to the recovery of more detailed autobiographical memories than other cue-types. Thus, although the accepted interpretation of the Proust phenomenon is that smells are better cues, a plausible alternative exists. Obviously, using the single cue comparison method, it is not possible to discriminate between these two alternatives, and the main aim of the present investigation is to provide unambiguous evidence by which the differential cue affordance hypothesis can be judged. To achieve this aim, it becomes necessary to compare different cues anchored to the same autobiographical event, and below, we describe a methodology which we believe accomplishes this. In the present study, rather than comparing different single cues, a double-cueing methodology is adopted (see Tulving & Bower, 1974, for a similar approach). Thus, all autobiographical episodes are initially retrieved in response to a verbal label (an odour name), but this is followed by an extended retrieval prompted by a second cue, which can be either the odour (corresponding to the verbal label), a literal repetition (the verbal label), or, in Experiment 2, a conceptual repetition (a picture). In this way, participants’ retrieval 5 Simon Chu attempts are first anchored to specific autobiographical episodes using a label, and the relative effectiveness of cues from different sensory modalities can be evaluated in the second retrieval phase. The value of this approach is that any advantage observed for odours cannot be due to a differential encoding bias, because for this to operate would require that the autobiographical episodes be initially cued by the odours rather than their corresponding labels. In the experiments which follow, we do not sample all possible cue combinations. Consideration of these other conditions may help in evaluating more complex hypotheses about the relative contributions of encoding and retrieval factors in odour-cued autobiographical memory. However, our remit for the present work was more limited, and we included only those cue combinations which are minimally required to evaluate the differential cue affordance hypothesis. The hypotheses described above do not exhaust the mechanisms which potentially operate in influencing the retrieval and/or interpretation of odour-cued memories. A brief discussion of some of these, and how they can be controlled for, is particularly important for Experiment 1 which focuses on the emotional aspects of retrieved memories. One problem relates to the fact that the hedonic valence of an odour is known to influence emotion (Baron & Bronfen, 1994; Ehrlichman & Bastone, 1992; Ehrlichman & Halpern, 1988; Lorig & Schwartz, 1988; Roberts & Williams, 1992), a factor which may, in turn, determine either the actual or perceived emotional valence of retrieved episodes. Thus, mood-congruent retrieval effects (e.g. Clark & Teasdale, 1982), whereby the hedonic tone of retrieved memories matches that induced by the odour, may operate. Under the double cueing methodology, however, this particular type of biased retrieval cannot operate because the same cue-type (label) is used for the initial retrieval of an episode across all conditions. A second type of bias might occur if the interpretation of emotion (believed to be) experienced during a past episode was influenced by the emotional state induced by the odour at the time of retrieval (cf. Levine, 1997). It is this type of explanation which Herz and Cupchik (1995) 6 Simon Chu seem to invoke when discussing their finding that odour cues led to memory retrievals which were more emotionally toned. The contribution of this potential bias can be evaluated by the use of an additional control condition, in which a verbal label is paired with an incongruent odour . If this bias operates, then the effect should emerge irrespective of the congruence between label and odour, thereby allowing one to discriminate between it and the differential cue affordance hypothesis In what follows, two experiments which employ the double cueing methodology are described. The first of these is concerned with memory for the emotional aspects of retrieved autobiographical episodes whereas the focus of the second is the total amount of event detail recovered. Experiment 1 Method Participants Forty-two volunteer participants were drawn equally from the undergraduate population at Liverpool University and people in full-time employment known to the experimenters. Materials Ten odours were chosen on the basis of familiarity and ease of procurement. These were coffee, vinegar, ginger, chocolate, whisky, onion, parmesan cheese, peanut, cinnamon, and lemon. A small amount of each substance was placed in a wrapped glass jar, with a punctured lid to allow participants to smell the contents without it being visible. Contents of the jars were replaced regularly to maintain odour quality and freshness. Seven rating scales were constructed to measure participants’ evaluations of recalled memories on a selection of dimensions. Each scale used a 7-point Likert scale and the 7 Simon Chu selection of targeted dimensions was determined primarily by previously published work. The dimensions were pleasant, painful, embarrassing, anxious, vivid, unique and personally significant. For each dimension, participants were asked, “On a scale of 1 - 7, how [pleasant / painful / etc] are the events in this memory, where 1 is not [pleasant / painful / etc.] at all and 7 is extremely [pleasant / painful / etc.].” Design A mixed design was used. All participants were cued twice, with the first cue always being the verbal label, and following each retrieval attempt, participants completed the seven rating scales in the same fixed order. The second cue was either the (repeated) verbal label or an odour, and the odour was either of the same nominal identity as the first (verbal label) cue or different. The three levels of cue combination - label, congruent odour, and incongruent odour - comprised the between-subjects factor of the design. In other words, participants always received the same cue-combinations. The ten odours were divided into two stimulus sets of five each, and participants were presented with labels from only one of the sets, with the other set forming the alternate odours for the incongruent odour condition. Stimulus sets were used equally often across participants and groups. Procedure Participants were tested in quiet distraction-free surroundings, in which there were no strong odours. All participants were provided with the following instructions: “This is a study investigating memory for life events of the past. I am now going to give you a word. I would like you to relate an event in your past history which you associate with that word. When you do, please give as much detail about the event as you can remember. Please restrict yourself to a single event and not a series of events which occurred around the same time.” 8 Simon Chu Following this, an odour label was presented auditorily, and the participant was allowed three minutes to think of and relate verbally an autobiographical memory cued by the label, following which the rating scales were completed. No record was taken of participants’ accounts of their memories. The second cue (label or odour) was then introduced with the following instructions: Label group: “It has been suggested that additional effort may be a factor in improving memory performance, and so, given the word _______, I would like you to try as hard as you can for a further few minutes to remember anything else about that event that you have just described.” Congruent and incongruent odour groups “It has been suggested that additional effort may be a factor in improving memory performance and that an odour might stimulate memory and help remember details about past events. Here is an odour... sniff at it and try as hard as you can for a further few minutes to remember anything else about the event you have just described.”1 Participants were given a further three minutes on this task, following which they completed the ratings scales again. This procedure was repeated for 5 trials, using different verbal cues on each trial. At the end of the study, all participants were fully debriefed. Results and Discussion The data from the rating scales were transformed by calculating the change in ratings between the first and second cues. For each participant, change scores were averaged across the 5 trials to give a single mean change score for each scale. As it was not possible to specify a priori the direction of change for five of the scales (pleasant, personal, painful, embarrassing and anxious), the absolute change on these scales served as the unit of analysis. If congruent odour cues afford access to a more veridical record of the emotional 9 Simon Chu experiences associated with a retrieved episode, then it is equally likely that, for example, pleasantness will be re-evaluated more or less pleasantly than beforehand. The actual direction will depend on the event which has been retrieved, something over which we have no control. For the remaining two scales (vivid and unique), direction of change can be specified a priori because access to a more detailed record of the emotion experienced during an event should render the episode more vivid and more unique. Thus, for these two scales, the prediction is that significant changes will occur in the positive direction. Table 1 around here please Mean initial and subsequent ratings for each scale in each condition are shown in Table 1 from which it is apparent that memories are rated as being low on the painful, anxiety and embarrassing scales with a large degree of consistency across all three conditions, while the vividness of the memories is reported be quite high. Mean changes in ratings for each scale in each condition are also shown in Table 1 where it is clear that the changes in ratings in the congruent cue condition are greater than those in the other two conditions. Significantly, on the scales where a direction of change could be specified a priori, the changes are both in the correct direction. Owing to a strong positive skew, the data for each scale were analysed using nonparametric Kruskal-Wallis one-way ANOVAs, with group as the between-subjects factor. As non-parametric analyses are used, median change in ratings for each of the scales are also shown in Table 1. These analyses revealed significant main effects for six of the seven scales: (for all reported statistics, df=2) anxious, χ2 =13.11, p=.0014; embarrassing, χ2=7.91, p=.0192; painful, χ2 = 19.32, p<.001; personal, χ2 = 8.78, p=.0124; pleasant, χ2 = 17.52, p<.001; vivid, χ2 = 13.08, p=.0014. While we believe non-parametric analyses to be more appropriate in this case, one reviewer has queried the absence of parametric analyses. Therefore, analyses were also carried out using parametric tests (ANOVA), and revealed a 10 Simon Chu very similar pattern of results with the exception that between-groups differences on the vivid scale only reached borderline significance (p = 0.065). Post hoc multiple comparisons between groups (Siegal & Castellan, 1988) showed that the mean ranks of change scores for the label group were no different than those for the incongruent odour group on all but the painful scale, for which the mean rank for label group was significantly lower (where smallest changes rank highest). More importantly, the mean of the change score ranks for the congruent odour group were significantly lower than the incongruent odour group on all scales, and significantly lower than the label group on four (pleasant, anxious, painful, and vivid) of the six scales for which there was a significant main effect. Furthermore, in the case of the vivid scale, the change was in the predicted direction such that retrieved memories were rated as more vivid in the congruent odour condition. On only one scale did the congruent odour group not show a significantly lower mean rank (unique, χ2 = 4.95, df = 2, p=.0842). All shifts in ratings in the incongruent cue and label cue conditions are small or negligible but one interesting result is the negative shift in the unique ratings in the label condition. Encouragement to retrieve additional details concerning the remembered episode in the label condition, when little or no event-specific detail is forthcoming, may only serve to yield more non-specific, schema-based knowledge. The activation of such knowledge might serve to alter the perception of the remembered episode, such that it would appear less unique in the light of schematically similar episodes. (D.C. Rubin, personal communication). As noted earlier, use of the single cue comparison method could mean that any difference between olfactory and other cues resulted from the retrieval of more complex memories, as postulated by the differential encoding bias hypothesis. However, because direct cueing is prevented by the double cueing methodology, we can be confident that the results obtained reflect the fact that odours are more potent retrieval cues. A similar 11 Simon Chu argument applies to mood-congruity retrieval effects. Finally, the fact that a similar effect was not observed for the incongruent odour condition rules out another alternative hypothesis, that odours bias participants’ responses as a result of the emotional state they tend to induce. Thus, the effects observed are memory-specific. The present results therefore provide the first unambiguous evidence supporting one popular interpretation of the Proust phenomenon, that odours are especially evocative as retrieval cues for the emotional details of autobiographical memory episodes. Four of the scales revealed greater changes in ratings in the congruent odour group but there may be grounds for believing that the two scales for which the pattern was less clear, embarrassing and personal, may be, by their very nature, less valid probes for emotionality. In real life, embarrassment may be experienced only infrequently as an emotional response to a situation, such that few naturally occurring autobiographical memories actually represent it. There is some support for this assertion in the finding that participants’ ratings on this scale were uniformly low. Of the 210 initial ratings taken of embarrassment experienced in the retrieved episode, 89.5% were 2 or less on the 7-point scale, with 80.1% being rated 1, the lowest point on the scale. With respect to the personal scale, it can also be argued that once retrieved, any memory will be experienced as personal, and additional details which are recovered will not affect the initial impression. Although we are aware that these arguments are post hoc, it is worth pointing out that the selection of scale dimensions was largely dictated by reference to the methodologies of previous studies (Rubin et al, 1984; Herz & Cupchik, 1992), without any serious consideration being given to the validity of the selected dimensions. Despite this shortcoming we believe the results are, nevertheless, convincing. Experiment 2 12 Simon Chu In Experiment 1 the emotional aspects of a remembered episode were treated as one class of event detail. In Experiment 2, the issue is widened by asking if the observed effect applies more generally to all types of event details. One of the rating scales used in the first experiment, vividness, suggests that this might be the case. However, participants may well have felt that the memory was more vivid because their memory for the emotional aspects of the recalled episode, the main focus of the rating scales, was clearer. Furthermore, on another of the scales which might be expected to be influenced by greater detail, uniqueness, the effect of condition was (marginally) non-significant. In measuring emotional versus other details, different approaches may be necessary. In the case of emotional detail, quantification requires the measurement of emotional states remembered as having been experienced at the time of the episode. Hypothetically, an emotional state might be represented mnemonically by values on a limited number of inter-related emotional dimensions, and cues might be assumed to differ in terms of the precision with which those scale values can be ‘read’. Thus, at the risk of stating the obvious, when we say that participants remember more emotion detail this does not mean that they are remembering that a greater number of (different) emotions were experienced, but that they have access to a more veridical record of the limited number of emotions which were experienced. This is why rating scales, rather than simple counts, are more appropriate for this type of event detail. In Experiment 2, more general event details were measured using an approach which entailed transcribing participants’ recollections of events and comparing the number of new event details recalled after the second cue across different cue types. An alternative explanation for the finding that congruent odours lead to enhanced recollection of event details is that it results from the introduction of a qualitatively different cue in the second retrieval phase, as opposed to the same repeated cue in the label condition. Thus, there may be nothing special about odours; any cue which differs from the original cue and is congruent with the content of the recollected episode may produce the same 13 Simon Chu result. To address this additive cue hypothesis, a visual cue condition, in which the participant is presented with a colour photograph in the second retrieval phase, is included in Experiment 2. The photographs depict, in each case, the source object of the label used as the first cue. This, then, represents a visual analogue of the congruent odour condition, and is used to directly test the alternate hypothesis that any secondary cue which is congruent with the recollected episode will significantly enhance recall relative to the label condition. In addition, the incongruent odour condition was retained as a control for the possibility that arousal levels may affect retrieval of event details. Method Participants Forty volunteer participants, comprising a mixture of undergraduate students and people in full-time employment, were assigned to one of four conditions (label, incongruent odour, visual and congruent odour). Materials A reduced set of 6 odours were used: coffee, cinnamon, disinfectant, baby powder, paint, and cigarette. They were presented in a manner identical to that in experiment 1. In addition, 6 x 4 inch colour photographs of each of the stimuli were used as cues in the visual cue condition. Procedure Participants were tested in quiet surroundings which were free of distractions and strong odours. All participants were provided with the same instructions as in Experiment 1. As a second cue in the extra visual condition, participants were presented with a photograph of the stimulus given as a label. Each participant took part in 3 trials, and the 14 Simon Chu same 3 minute time limit was allowed for the recollective stage of each trial. The reports of the recollected events were tape recorded and transcribed for content analysis. Results and Discussion The contents of the recollected episodes were transcribed and content analysed, using single sentences as the unit of analysis. Given that the source text was transcribed speech rather than written word, and that normal speech tends not to involve strict discrete sentences, sentence form was imposed on the text by dividing long utterances into discrete sentence units where it seemed appropriate i.e. where speech appeared to run into a new sentence. The mean number of sentences vocalised after the first and second cue in each condition were as follows (standard deviations in parentheses): label condition, cue 1 - 25.4 (13.2), cue 2 - 15.2 (11.1); visual condition, cue 1 - 24.6 (4.76), cue 2 - 9.6 (3.6); incongruent condition, cue 1 - 24.8 (6.9), cue 2 - 9.2 (7.2); congruent condition, cue 1 - 22.7 (8.1), cue 2 - 19.9 (11.6). Total number of sentences uttered after the first cue was tallied for each of the four conditions (label, 254; incongruent odour, 248; visual, 246; congruent odour, 227). Chi-square analysis revealed no differences between the four conditions (χ2 = 1.68, df = 3, ns.) suggesting that the groups were evenly balanced in verbosity. The experimenter then rated the recollections taken after the second cue, and assigned each sentence to one of two simple categories: new for sentences which contained information relating to the episode which did not appear previously in the recollection after the first cue, and irrelevant for sentences which did not contain new information (invariably, statements which did not contain new information were irrelevant to the retrieval of the episode). A subset of recollections was then independently analysed by two blind raters (following Holsti, 1969), and the inter-rater reliability between all three was calculated (Krippendorff, 1980). The inter-rater agreement was found to be strong (Krippendorff’s α = .717). Insert Table 2 around here please 15 Simon Chu The number of sentences in each category from each condition is detailed in Table 2. As can be seen, the total number of sentences produced for the incongruent and visual conditions is approximately matched (92 and 96 respectively), and much lower than that observed for the label (152) and congruent (199) conditions. For these latter two conditions, the difference in total number of sentences is explained by the increase in the number of new detail which emerges in phase 2 (71 and 120 respectively), with the number of sentences containing irrelevant detail approximately matched (81 and 79 respectively). The relevant proportions of sentences containing new and irrelevant detail show a similar pattern: For the congruent condition, the majority of sentences are those containing new detail; for the label and visual conditions, the proportions are approximately matched; whereas for the incongruent condition, the majority of sentences are those containing irrelevant detail. Given that the data took the form of frequency categorisations of a limited number of statements, Chi-square analyses were deemed the most appropriate method of statistical analysis (Siegal & Castellan, 1988). Data were subjected to a 2 (new vs. irrelevant) x 4 (label vs. incongruent odour vs. visual vs. congruent odour) Chi-square analysis, and the result proved significant (χ2 = 16.72, df = 3, p<0.001). Further partitioning of the 2 x 4 contingency table and subsequent partition Chi-square analysis (Siegal & Castellan, 1988) revealed that the congruent condition was significantly different from the other three conditions (χ2 = 13.44, df = 1, p<0.001), indicating that episodes retrieved in response to the congruent odour cue included more new detail than those retrieved in response to the label, picture or incongruent odour cues. These results provide clear support for the differential cue affordance value hypothesis. Retrieval was anchored to specific episodes in phase 1 and the addition of a congruent olfactory cue in phase 2 produced a significant increase in new details. That this effect resulted simply from the use of two different cues, the additive cue hypothesis, can be 16 Simon Chu ruled out because there was no evidence of a similar enhancement for the visual cue condition. Interestingly, the visual condition yielded a much lower total number of sentences than the label condition. There are probably two (related) reasons for this. First, although not explicitly instructed to do so, participants tended to focus on the visual details of the target cue as represented in their autobiographical recollection. In other words, the visual cues induced a selective search strategy. Second, in the majority of cases, the visual cues did not match those which featured in individuals’ recollections, which led participants to comment on the differences. One reason why the additive cue hypothesis was not supported here, therefore, may relate to the fact that the visual cues employed did not sufficiently map the visual appearance of the items in the recollective episode. Of course, exactly the same argument could be made about the (congruent) olfactory cues because it is extremely unlikely that these matched exactly the sensory experiences of the original autobiographical episodes. Despite this, the overall proportion of new detail which emerged in the visual condition was no different from the label condition. The inclusion of the incongruent odour condition in the present experiment permitted a test of one further hypothesis, that enhanced retrieval results from increased arousal levels associated with ‘sensing’ odours. This hypothesis can also be ruled out because there was no evidence of enhanced retrieval for the incongruent odour condition. In fact, as with the visual cues, there was evidence that relative to the label condition, the total number of sentences produced in phase 2 was attenuated. Also, whilst the visual and incongruent label conditions led to approximately equivalent sentence production in phase 2, the relative proportions of sentences containing new and irrelevant detail differed, with fewer new details being given in the incongruent odour condition. This echoes the pattern observed in Experiment 1 in which the change in ratings for the incongruent odour condition was less than that in the label condition on every scale (see Table 1). One reason for this may relate to the differential cue affordance hypothesis we set out to test. That is, odours 17 Simon Chu are indeed especially potent as retrieval cues, to the extent that they automatically induce the retrieval of unrelated and irrelevant detail in the incongruent odour condition. General Discussion Two sources of data, participants’ ratings of their memory quality and content analysis of the recollective episodes recounted, provide converging evidence that odours are especially potent reminders of autobiographical experiences. Several plausible alternative hypotheses which might underlie the phenomenon were ruled out, through the use of the double cueing methodology and specific control conditions. This permitted a test of what we believe represents the more popular interpretation of the Proust phenomenon, the differential cue affordance hypothesis, which we can now confidently say is unambiguously supported. The findings reported here do not represent the only distinctive features of odourcued memories which are relevant to the Proust phenomenon. As noted in the introduction, odour-cued memories are postulate to be more detailed, more emotionally loaded, and more distant in time. Although the single cue comparison method is unsuitable for testing particular hypotheses concerning the quantity and quality of detail retrieved, we have successfully applied the technique to evaluate the third of these memory attributes, namely, the age at which autobiographical episodes were originally experienced. In this case, the age of a memory, unlike quality or quantity, is a fixed property anchored in time and therefore not open to bias. When plotting the distribution of autobiographical memories across the entire lifespan in a group of elderly participants, we found that the peak number of memories retrieved (the so-called reminiscence bump, Rubin & Schulkind, 1997) appeared at an age earlier for odour cues than the peak for label cues (Chu & Downes, 2000). That is, odour-cued autobiographical memories are older. Taken together, therefore, our 18 Simon Chu experiments employing two complementary methodologies show that odour-cued memories tend to be more emotional, more detailed, and more aged. Thus, we can state with some confidence that the phenomenon attributed to Proust is more than a literary speculation. In common with the vast majority of studies examining autobiographical memories for personal events (rather than knowledge), our experiments present no means of assessing the accuracy of the details retrieved by our participants. The crux of Banaji and Crowder’s (1989) controversial attack on the field of everyday memory followed a similar line of argument and the indignant rebuttals from prominent researchers in the field (e.g. Conway, 1991; Morton, 1991; Neisser, 1991) will not be discussed here. Nevertheless, one method of achieving the form of empirical control which Banaji and Crowder would demand (and one which we are currently pursuing in our laboratory) would be to arrange a series of naturalistic events for participants to experience. This series would involve a range of sensory elements (including odours) such that, after a long delay (i.e. months), detailed retrieval could be examined (and verified) in response to an array of different cue-types. However, we feel that the absence of such a methodology in the present study does not limit the value of our findings, particularly given that we have no reason to presume that the proportion of inaccurate detail in one condition should differ from that in any other. Therefore, even allowing for the inaccuracy of a proportion of retrieved information, if more absolute detail is retrieved in the congruent-odour condition, this implies that more accurate detail is also retrieved. Why should odours differ from other cue types in these ways? One class of explanation draws on knowledge of the neuroanatomical projections to and from the olfactory cortex, and the intrinsic links with circuits known to mediate episodic memory. Thus, the olfactory bulb projects to the amygdala, the fimbria of the hippocampus and the dorso-medial nucleus of the thalamus (Dodd & Castellucci, 1991; Nieuwenhuys, Voogd & van Huijzen, 1988), each of which has some involvement in memory consolidation 19 Simon Chu processes (see Aggleton & Brown, 1999; McGaugh, Roozendaal & Cahill, 2000). The work of McGaugh and colleagues in particular has shown that the amygdala plays a specific role in modulating hippocampal-mediated consolidation processes. Furthermore, in rats, this modulatory function is specifically related to appetitively or aversively motivated learning (McGaugh et al, 2000). Thus, the amygdala may be directly involved in the formation of emotional memories, a view supported by recent functional imaging studies of human participants (e.g. Hamman et al, 1999), and its modulatory role in memory consolidation will lead to the creation of more stable memory representations. Because of the neuroanatomical projections between olfactory processing regions and the limbic structures, odours may be especially privileged in their ability to directly influence the modulatory functions of the amygdala. A second class of explanation is grounded in cognitive psychological theories of autobiographical memory retrieval. Conway’s (1992, 1996; Conway & Pleydell-Pearce, 2000) model of autobiographical memory structure proposes that autobiographical memory retrieval processes are cyclical in nature, such that, given a single cue word, memory probes become more refined, and thus more specific, with each retrieval cycle, eventually leading to the retrieval of details from an event-specific knowledge base. In contrast, highly specific cues, such as sensory stimuli, may be able to bypass the normal cyclic retrieval process and directly access the event-specific knowledge base (Conway, 1992). The fact that there exist independent top-down (thematic) and bottom-up (sensory) retrieval processes, though, does not necessarily lead to the prediction that olfactory cues are selectively advantaged relative to other cue types (although it may predict that retrieval is faster for olfactory cues). This is because both provide access to the same event-specific knowledge base. Possibly, the way in which the different types of cue map onto the knowledge base may differ for thematic and sensory information, such that the latter provide more diffuse links with event details. 20 Simon Chu However, without further independent specification, there is a danger of circularity in adding these speculative assumptions to the model. 21 Simon Chu References Aggleton, J.P., & Brown, M.W. (1999) Episodic memory, amnesia, and the hippocampalanterior thalamic axis. Behavioural and Brain Sciences, 22, 425-489. Baddeley, A.D. (1982). Domains of recollection. Psychological Review, 89, 708-729. Banaji, M.R. & Crowder, R.G. (1989). The bankruptcy of everyday memory. American Psychologist, 44, 1185-1193. Baron, R.A. & Bronfen, M.I. (1994). A whiff of reality: Empirical evidence concerning the effects of fragrances on work-related behaviour. Journal of Applied Social Psychology, 24, 1179-1203. Chu, S. & Downes, J.J. (2000). Long live Proust: The odour-cued autobiographical memory bump. Cognition, 75, B41-B50. Clark, D.M. & Teasdale, J.D. (1982). Diurnal variations in clinical depression and accessibility of memories of positive and negative experiences. Journal of Abnormal Psychology, 91, 87-95. Conway, M.A. (1991). In defense of everyday memory. American Psychologist, 46, 19-26. Conway, M.A. (1992). A structural model of autobiographical memory. In M.A. Conway, D.C. Rubin, H. Spinnler & W.A. Wagenaar (Eds.), Theoretical perspectives on autobiographical memory. Dordrecht: Kluwer Conway, M.A. (1996). Autobiographical knowledge and autobiographical memories. In D.C. Rubin (Ed.), Remembering our past: Studies in autobiographical memory. Cambridge: Cambridge University Press. Conway, M.A. & Pleydell-Pearce, C.W. (2000). The construction of autobiographical memories in the self-memory system. Psychological Review, 107, 261-288. 22 Simon Chu Dodd, J. & Castellucci, V.F. (1991). Smell and Taste: The Chemical Senses. In E.R. Kandel, J.H. Schwartz & T.M. Jessell (eds.) Principles of Neural Science, 3rd ed. Norwalk: Appleton Lange. Ehrlichman, H. & Bastone, L. (1992). Olfaction and emotion. In M.J. Serby & K.L. Chobor (Eds.), Science of Olfaction. New York: Springer Verlag. Ehrlichman, H. & Halpern, J.N. (1988). Affect and memory: effects of pleasant and unpleasant odour on retrieval of happy and unhappy memories. Journal of Personality and Social Psychology, 55, 769-79. Engen, T. & Ross, B.M. (1973). Long-term memory of odors with and without verbal descriptors. Journal of Experimental Psychology, 100, 221-227. Goldman, W.P. & Seamon, J.G. (1992). Very long-term memory for odors: Retention of odor-name associations. American Journal of Psychology, 105, 549-563. Hamman, S.B, Ely, T.D., Grafton, S.T., Kilts, C.D. (1999) Amygdala activity related to enhanced memory for pleasant and aversive stimuli. Nature Neuroscience, 2, 289-293. Herz, R.S. (1996). A comparison of olfactory, visual and tactile cues for emotional and nonemotional associated memories. Chemical Senses 21, 614-615. Herz, R.S. & Cupchik, G.C. (1992). An experimental characterization of odor-evoked memories in humans. Chemical Senses, 17, 519-528. Herz, R.S. & Cupchik, G.C. (1995). The emotional distinctiveness of odor-evoked memories. Chemical Senses, 20, 517-528 Herz, R.S. & Engen, T. (1996). Odor memory: Review and analysis. Psychonomic Bulletin and Review, 3, 300-313. Holsti, O.R. (1969). Content Analysis for the Social Sciences and Humanities. Reading, Massachusetts: Addison-Wesley. Krippendorff, K. (1980). Content Analysis: an introduction to it’s methodology. Beverley Hills, CA: Sage. 23 Simon Chu Laird, D.A. (1935). What can you do with your nose? Scientific Monthly, 41, 126-130. Lawless, H.T. & Cain, W.S. (1975). Recognition memory for odors. Chemical Senses and Flavor, 1, 331-337. Lehrner, J.P., Walla, P., Laska, M. & Deecke, L. (1999). Different forms of human odor memory: A developmental study. Neuroscience Letters, 272, 17-20. Levine, L.J. (1997). Reconstructing memory for emotions. Journal of Experimental Psychology: General, 126, 165-177 Lorig, T.S. & Schwartz, G.E. (1988). Brain and Odour I: Alteration of human EEG by odour administration. Psychobiology, 16, 281-84. McGaugh, J.L., Roozendaal, B., & Cahill, L. (2000) Modulation of memory storage by stress hormones and the amygdaloid complex. In M.S. Gazzaniga (Ed.) The New Cognitive Neurosciences – 2nd Edn. Cambridge, MA: MIT Press. Morton, J. (1989). The bankruptcy of everyday thinking. American Psychologist, 46, 3233. Neisser, U. (1989). A case of misplaced nostalgia. American Psychologist, 46, 34-36. Nieuwenhuys, R., Voogd, J. & van Huijen, C. (1988). The Human Central Nervous System: A Synopsis and Atlas. NewYork: Springer-Verlag. Proust, M. (1922). Swann’s Way. Trans. C.K.Scott Moncrieff, publ. 1960, London: Chatto & Windus. Rabin, M.D. & Cain, W.S. (1984). Odor recognition: Familiarity, identifiability and encoding consistency. Journal of ExperimentalPsychology: Learning, Memory and Cognition, 10, 316-325. Roberts, A. & Williams, J.M.G. (1992). Effects of olfactory stimulation on fluency, vividness of imagery and associated mood. British Journal of Medical Psychology, 65, 197-199. 24 Simon Chu Rubin, D.C., Groth, E. & Goldsmith, D.J. (1984). Olfactory cuing of autobiographical memory. American Journal of Psychology, 97, 493-507. Rubin, D.C. & Schulkind, M.D. (1997). The distribution of autobiographical memories across the lifespan, Memory & Cognition, 25, 859-866 Schab, F.R. (1991). Odor memory: Taking stock. Psychological Bulletin, 109, 242-251. Siegal, S. & Castellan, N.J. (1988). Nonparametric statistics for the behavioral sciences. (2nd edition). New York: McGraw-Hill. Tulving, E. & Bower, G.H. (1974). The logic of memory representations. In G.H. Bower (Ed.), The Psychology of Learning and Motivation: Advances in research and theory (Vol. 8). London: Academic Press. White, T.L. (1998). Olfactory memory: the long and the short of it. Chemical Senses, 23, 433-441. 25 Simon Chu Table 1. Mean initial and subsequent ratings, mean change in ratings (standard deviations in parentheses) and median change in ratings for each scale in each condition incongruent cue condition Mean R1 Mean R2 Mean Change Median (SD) (SD) (SD) Change Pleasant 4.04 (2.08) 4.03 (2.07) 0.07 (0.16) 0 Personal 2.54 (1.90) 2.53 (1.88) 0 (0) 0 Painful 1.56 (1.33) 1.56 (1.33) 0 (0) 0 Anxious 1.47 (1.28) 1.47 (1.28) 0 (0) 0 Embarrassing 1.34 (1.27) 1.36 (1.27) 0.01 (0.05) 0 Vivid 4.50 (1.82) 4.50 (1.82) 0 (0) 0 Unique 3.31 (1.65) 3.21 (1.62) -0.10 (0.17) 0 label cue condition Pleasant 3.24 (3.00) 3.23 (3.00) 0.16 (0.21) 0 Personal 2.56 (2.00) 2.46 (1.74) 0.19 (0.23) 0.10 Painful 1.56 (1.09) 1.60 (1.18) 0.07 (0.13) 0 Anxious 1.77 (1.28) 1.79 (1.31) 0.04 (0.08) 0 Embarrassing 1.57 (1.25) 1.51 (1.15) 0.06 (0.14) 0 Vivid 5.14 (1.63) 5.11 (1.66) -0.03 (0.07) 0 Unique 3.93 (2.07) 3.73 (2.01) -0.20 (0.19) -0.20 congruent cue condition Pleasant 4.13 (1.74) 4.20 (1.86) 0.47 (0.34) 0.40 Personal 3.07 (2.07) 3.06 (2.12) 0.48 (0.50) 0.22 Painful 1.84 (1.45) 1.97 (1.65) 0.44 (0.50) 0.20 Anxious 1.99 (1.64) 1.96 (1.65) 0.34 (0.32) 0.20 Embarrassing 1.63 (1.48) 1.63 (1.43) 0.17 (0.21) 0.10 Vivid 5.09 (1.56) 5.29 (1.67) 0.20 (0.50) 0.20 Unique 3.97 (1.99) 4.00 (1.94) 0.03 (0.35) 0 26 Simon Chu Note: All changes are unsigned apart from vivid and unique which incorporate direction of change. Table 2. Number of sentences (proportion of the condition in parentheses) in each content analysis coding category and total number of sentences from each condition condition 1 new irrelevant total label 71 (.47) 81 (.53) 152 incongruent 33 (.36) 59 (.64) 92 visual 44 (.46) 52 (.54) 96 congruent 120 (.60) 79 (.40) 199 One reviewer queried whether the different instructions used in the experimental conditions may have introduced an unintended biasing effect. It is possible that explicitly stating that “odours might stimulate memory and help remember details about past events” could induce additional motivation to retrieve detail in the odour conditions, motivation which would be somewhat diluted in the incongruent cue condition when participants perceive the mismatch between the cue and the retrieved episode. While we acknowledge that the differences in instructions make this a possibility, we believe it highly unlikely that these differences would influence the resulting data under this particular design. Given that we employ condition as a between-group factor, participants in the incongruent cue condition are not aware that there are other participants in the study for whom the odour cue is congruent with the retrieved episode, and the fact that they experience a mismatch between odour and retrieved episode should not reduce their motivation to retrieve event detail. In addition, all participants are told that “additional effort may be a factor in improving memory performance” and, given that participants know that improved memory performance is expected after the second cue, specific mention of “detail” in the odour conditions should not serve to dramatically influence responses. 27