RESISTANCE TO INTERFERENCE OF OLFACTORY PERCEPTUAL LEARNING

：芭芭拉

RESISTANCE TO INTERFERENCE OF OLFACTORY PERCEPTUAL LEARNING
Richard J Stevenson, Trevor I Case, Caroline Tomiczek. The Psychological Record. Gambier: Winter 2007. Vol. 57, Iss. 1; pg. 103, 14 pgs

Abstract (Summary)
Conceptually similar findings support this notion of persistence, including the ability of adults to name odors not smelled since childhood (Goldman & Seamon, 1992), and expression of a preference for vanillin in adults, who as babies were bottle-fed with vanillin flavored milk (Haller, Rummel, Henneberg, Pollmer, & Koster, 1999).

：芭芭拉

Full Text (5945  words)

Copyright The Psychological Record Winter 2007

[Headnote]
Olfactory memory is especially persistent. The current study explored whether this applies to a form of perceptual learning, in which experience of an odor mixture results in greater judged similarity between its elements. Experiment 1A contrasted 2 forms of interference procedure, 'compound' (mixture AW, followed by presentation of new mixtures each containing 1 of its elements, AX and DW) and 'elemental' (mixture CY, followed by presentation of its elements C and Y) against a nonexposed control. Learning was evident in both interference conditions to the same degree, relative to the control. Experiment 1B established that the interference conditions did not significantly differ from uninterfered paired controls. Experiment 2 compared the 'compound' procedure with 2 exposed control conditions and assessed whether participants had acquired the interfering mixtures too (AX and DY). Learning was evident in the 'compound' treated pair (AW) and also for the mixtures AX and DW that made up the interfering compounds. These results are problematic for configural explanations and a new formulation is suggested.


The claim that human olfactory cognition is qualitatively different from other sensory domains has been made frequently and relies to no small extent upon the finding that odor memory is especially persistent (Herz & Engen, 1996; Richardson & Zucco, 1-9-8-9; Schab, 1991). Evidence accumulated over the last four decades has, in the main, been largely supportive of this assertion and it is confirmed by two related, though procedurally separate, lines of evidence. The first is that recognition memory for odors is initially unimpressive, in contrast say to that for pictures, but that it decays very little over time intervals measured in months (e.g., Lawless, 1978). Conceptually similar findings support this notion of persistence, including the ability of adults to name odors not smelled since childhood (Goldman & Seamon, 1992), and expression of a preference for vanillin in adults, who as babies were bottle-fed with vanillin flavored milk (Haller, Rummel, Henneberg, Pollmer, & Koster, 1999). In all of these cases participants appear to have encoded and retained olfactory information over lengthy periods of time.

The second line of evidence, and the focus of this current pa-per, is the finding that interference procedures generally exert little influence on the retention of olfactory information. Given that theories of forgetting have been heavily dominated by interference-based accounts, it is perhaps not surprising that olfactory memories should be so long-lived, just as we described above. A seminal demonstration in this respect was Lawless and Engen's (1978) finding that associations between pictures and odors are resistant to retroactive interference. That is, once a participant has learned that one odor is associated with one picture, pairing that odor with a different picture does not affect retrieval of the original odorpicture association. Not all findings, however, agree with this conclusion. Koster, Degel, and Piper (2002) reported that retroactive interference affected an implicit odor memory task, in which participants were asked to judge the 'goodness of fit' between pictures of a room and particular odors. Participants who had experienced an odor in one room, and then later received the same odor within a different room, experienced retroactive interference, but only if they could not identify the odorant. A further discordant finding was made by Walk and Johns (1984), who observed that interpolating a similar odor between presenting the target, and the target and distracters in a delayed matching to sample task, had a deleterious effect on subsequent recognition performance. Both of these contradictory findings might be considered atypical. The first relied upon a correlational approach, in that a post-hoc split into identifiers and nonidentifiers was necessary to demonstrate the effect, while the second involved only a 26-s delay between the target, and the target and distracters. Thus in the first case it could be argued that some variable other than identification might have accounted for the effect, while in the second, it could be argued that this finding might not generalize to effects involving longer term memory processes.

Although it is clear that resistance to interference in olfactory memory experiments is not absolute, two more recent lines of evidence do support Lawless and Engen's (1978) original findings, although they are procedurally very different. The first comes from the study of odor-taste learning, in which an odor mixed with a taste results in that odor acquiring the properties of that taste (e.g., Stevenson, Prescott, & Boakes, 1995). This type of learning is apparently resistant to an 'extinction-like' interference procedure. Presenting the odor and the taste alone, after their presentation as a mixture, exerts no detectable effect on learning, while a nonolfactory control paradigm (color-taste) is sensitive to this manipulation (Stevenson, Boakes, & Wilson, 2000a). Similarly, after pairing an odor with a taste (citric acid), then presenting the same odor paired with another taste (sucrose), learning is again unaffected; that is, the odor still smells sour. This procedure, which resembles counterconditioning, does, however, affect a nonolfactory analogue (color-taste), suggesting that it is not a consequence of lack of experimental sensitivity (Stevenson, Boakes, & Wilson, 2000b).

A second line of evidence comes from an even more recently explored form of olfactory task, odor-odor learning, in which sniffing an odor mixture results in the elements later being judged as qualitatively more similar (e.g., Stevenson, 2001a) and less discriminable (Case, Stevenson, & Dempsey, 2004; Stevenson, 2001b), than equally exposed, but unpaired, control odors. Recently, we tested whether this form of perceptual learning is sensitive to an 'extinction-like' procedure, in which the component elements of an odor mixture, experienced earlier, are later presented alone. When compared to a nonolfactory control (color-odor learning) which is affected by this extinction-like procedure, no detectable effect on participants learning for the odor-odor pairs was found (Stevenson, case, & Boakes, 2003). Moreover, even when many more 'extinction' trials than training trials were added, along with more extensive control conditions, the same conclusion still held true (Stevenson, Case, & Boakes, 2005). Rather than refer to this interference procedure as extinction-like, which implies both a conditioned stimulus (CS) and an unconditioned stimulus (US) and a traditional Pavlovian paradigm-features clearly absent in this type of learning-we use the term elemental interference instead.

In the present series of studies, we set out to examine whether odor-odor learning is resistant to a new interference procedure, in which following exposure to an odor mixture (AW), each element is then mixed with another odor (i.e., AX, BW). Appropriate terminology is again not readily available to describe this type of approach. Counterconditioning may be inappropriate because there is no clear US or CS. Retroactive interference also appears a rather poor fit because of its history of use in paired-associate learning. Consequently, and in keeping with calling the other procedure elemental interference, we refer to this new procedure as compound interference.

：芭芭拉

☆Experiment 1A

As in all of the studies reported here, Experiment 1A employed a three-phase within-participant design. In the first phase participants received two odor mixtures (see Table 1 ; AW and CY), each composed of a more familiar and a less familiar element. This combination appears to produce optimal conditions for learning, however, this has not as yet been directly tested (see Stevenson, 2001a, for discussion). We have adopted the convention here of referring to the less familiar odors as A, B, C, or D and the more familiar odors as W, X, Y, or Z. In the second phase of the experiment participants received the interfering stimuli. For one odor pair, AW, this consisted of exposures to two new odor mixtures, each of which contained an element from AW (i.e., AX and BW), the compound interference condition. For the other odor pair, CY, the elements C and Y were presented on their own (elemental interference) as used before in our previous studies of interference. The third phase of the experiment consisted of a similarity test, in which participants sniffed three pairs of odors, the two pairs from the training phase (i.e., A vs. W and C vs. Y) and a novel unexposed pair (D vs. Z).

Method

Participants

Thirty-six Macquarie University students (29 female, 7 male) participated as part of their course requirements. Mean age was 23.8 (range = 17-51; SD = 9.5). In both experiments reported here, no participant had served in an olfactory learning experiment before and no one had an upper-respiratory tract infection.

Materials

The stimuli in this experiment were drawn from eight odorants, which were arranged as two sets of four (see Table 2). Each odorant was first dissolved in propylene glycol and then made up to volume with water. For the first set of four odorants, two mixtures were formed by adding together various quantities of the stock solutions detailed in Table 2; PA-BZ (10% PA, 23% BZ, and 67% water) and AL-CG (25% AL, 25% CG, and 50% water). For the second set of four odorants, four mixtures were formed; MR-CH (60% MR and 40% CH), MR-LN (40% MR and 60% LN), AN-CH (50% AN and 50% CH) and AN-LN (40% AN and 60% LN). Mint odor (L-Carvone, Dragoco; 0.3g/L) served as a filler stimulus to equate the ratio of odorous to water trials during the training and interference phases. When odors were presented alone, concentrations were adjusted so that they approximated that of the mixture components. This was achieved in prior pilot work by progressively changing the concentration of the target odor alone, until it matched the judged intensity of the target odor in the mixture. Concentrations for each stock solution are detailed in Table 2. Each was diluted with differing quantities of water (in parentheses); CG (67% water), AL (67% water), PA (90% water), BZ (80% water), MR (70% water), LN (50% water), AN (60% water), and CH (60% water). All stimuli used here were colorless and presented in aliquots of 50ml in 250ml plastic transparent screw-top jars.

Procedure

The experiment employed a within-participant design (see Table 1). It consisted of three phases (training, interference, and test), all completed in a single 1-hr session. During training participants sniffed two odor mixtures, AW and CY. During interference, they smelled two further mixtures, each of which contained an element from the AW mixture, namely AX and BW, along with the elements taken from the CY mixture, namely C and Y alone. In the test phase, participants rated the similarity of three pairs of odors, A and W, C and Y, and an unexposed control pair D and Z.

During the training phase, participants completed 12 trials, with each trial composed of sniffing one stimulus. In total there were three AW, three CY, three mint odor, and three plain water trials. Presentation order of trials was random for each participant. On each trial, participants took as long as they wished to sniff it (on average this was 3 s) after which they rated the strength of the smell on a 7-point category rating scale (anchors, not at all = 1, very = 7). Although participants had numerous odors to sniff overall, we have explored in previous studies whether this might ultimately affect their discriminative ability. This does not appear to be the case (see Stevenson et al., 2003). The allocation of odor mixtures to conditions was partially counterbalanced. The odor mixtures (or elements) for AW, AX, and BW were always drawn from the second set of stimuli (requiring four mixtures; see Table 2), and the CY and DZ pairs were always drawn from the first set of stimuli (requiring two mixtures; see Table 2). Within each of these sets, an odor combination was equally likely to occur in any of the designated conditions. We were not able to detect any systematic effects of mixture type on learning in this or any other experiment reported here.

Following a 4-min break participants received a further 36 trials (as three sets of 12), each presented in an identical manner to the training phase. A 4-min break separated the second 12 from the third set of 12 trials. In total, participants sniffed six AX, six BW, six C, six Y, and 12 of plain water. The ratio of odorous to water stimuli was the same as for the training phase and trials were again presented in random order for each participant.

After a further 4-min break participants received the test phase. This was composed of three pain/vise comparisons, A vs. W, C vs. Y, and D vs. Z. Order between and within pairs was counterbalanced. Participants were asked to smell the first member of the pair, replace the lid, then smell the second member of the pair, replace the lid, and then rate the similarity of the pair on a 7-point category scale (anchors, 1 = identical, 4 = moderately different, 7 = very different). For clarity of exposition in all the experiments reported here, similarity ratings are presented in reversed format, so that larger numbers reflect greater similarity.

Results and Discussion

Table 1 details mean similarity ratings (and SDs) for Experiment 1A. The AW pair which had received the compound interference procedure and the CY pair which had received the elemental interference procedure, were judged as more alike than the novel test stimuli DZ. Paired t tests revealed no significant difference between the compound and elemental interference procedure pairs (t < 1 ), but both of these significantly differed from the nonexposed control, (respectively, t(35) = 2.98, p < 0.005; t(35) = 2.42, p < 0.025. These findings suggest that the compound interference procedure was no more successful at affecting learning than the elemental procedure, relative to a pair of odors not previously encountered in the experiment.

One pertinent criticism of this result is that no control condition was run in which participants were exposed to an odor mixture in the absence of interference. Consequently, can we be sure that the findings here do not in fact reflect some degree of interference in both conditions that simply went undetected? This appears very unlikely based upon our extensive use of the elemental interference procedure. In three previously published studies (Stevenson et al., 2003, 2005) we have consistently found that even with extensive presentation of the elements alone, no evidence of interference was obtained when contrasted to a within-participant control condition in which the paired (but noninterfered) stimuli were also presented. It would, therefore, appear unlikely that the elemental interference condition would behave differently in this experiment from in the ones reported before. However, to eliminate this possibility we ran a further control experiment (1B; see Table 1 ). In Experiment 1B participants were exposed to the same training phase, followed by a delay equivalent to the time taken to normally complete the interference phase, after which they completed test phase identical to that of Experiment 1 A.

☆Experiment 1B

Method

Participants

Twenty-one Macquarie University students (17 female, 4 male) participated as part of their course requirements. Mean age was 22.8 (range = 17-42; SD = 6.99).

Method

Training was identical to Experiment 1 A. Participants were then asked to read magazines for 22 min (the average time taken to complete the interference phase in Experiment 1A). This was followed by the similarity test, again identical to Experiment 1A.

Results and Discussion

Table 1 details mean similarity ratings (and SDs) for Experiment 1B. The AW pair and the CY pair were both judged as more alike than the novel test stimuli DZ. Paired t tests revealed no significant difference between the AW and CY pairs (t = 1.4), but both of these significantly differed from the nonexposed control, respectively, t(20) = 2.78, p < 0.025; t(20) = 4.74, p < 0.001.

To establish whether or not the magnitude of the perceptual learning effect for the AW and CY pairs, relative to the DZ control pair, differed from that in Experiment 1A, a two-way repeated measures ANOVA was conducted. The ANOVA had one between factor (experiment) and one within factor (pairing type; AW, CY, DZ). The ANOVA revealed only one significant effect, pairing type, F(2,110) = 13.18, MSE = 2.06, p < 0.001. Crucially, there was no significant interaction between pairing type and experiment, F = 2.2, suggesting that the perceptual learning effect in Experiment 1A was of a similar magnitude to the effect in Experiment 1B. Consequently, it appears unlikely, given this and the previous findings alluded to above, that either compound or elemental interference affects this form of learning.

：芭芭拉

☆Experiment 2

Two further concerns can be raised about the design of Experiment 1A and hence its conclusion. First was the absence of full counterbalancing of the assignment of odor pairs to experimental conditions. Although we know that the elemental interference procedure can not affect learning under conditions of full counterbalancing (Stevenson et al., 2003, 2005), we can not be sure in Experiment 1A that resistance to interference in the compound interference procedure might have resulted from some inherent difference in the stimuli themselves (i.e., uniformly higher similarity ratings). Although we were fairly confident this was not the case, based upon our experience with these stimuli in previous studies (e.g., Stevenson, 2001 a 2001 b), it appeared prudent to eliminate this possibility in Experiment 2. This necessitated some pilot work so that we could have full counterbalancing of odor pairs to treatment conditions. A second concern relates to the control condition and in particular the merits of using a nonexposed control, rather than one exposed the same number of times during the training and interference phases, as the condition with which it is to be compared. The latter, to us, seemed to offer a more conservative control, as it would ensure that any nonassociative effects that exposure alone might produce were equated between the control and interference conditions. In sum then, the first aim of Experiment 2 was to assess under more stringent conditions, whether learning could still be detected following a compound interference manipulation.

The second aim of Experiment 2 was to determine whether participants demonstrated any perceptual learning for the mixture elements used in the compound interference condition (i.e., A and X and B and W). In Experiment 1A no attempt was made to assess whether participants had learned anything about these mixture pairs. This is of particular interest for two reasons. First, it may inform us as to why the compound interference procedure does not apparently affect implicit memory for AW (e.g., if there was no evidence of learning). Second, there may be a primacy effect, in that the first learned combination is better retained than any subsequently exposed combinations. Testing for this, along with the other previously mentioned changes, required four modifications to the procedure (see Table 1). First, assignment of odor pairs to treatment conditions was fully counterbalanced. Second, to ensure equivalent levels of exposure of the control odors and the compound interference condition, the control odors were presented during the training and interference phases and in addition, a new pair of control odors was presented alongside them, but only during the interference phase, so that control and treatment conditions were closely matched (see Table 1). Third, a second similarity test was introduced to assess whether learning had occurred for the interfering mixture pairs (AX, BW). Fourth, so that we could better ascertain the degree to which the strength of the original conditioning to AW was similar to or different from that of the two interfering mixtures AX and BW, we equated the number of conditioning trials of AW to the number of interference trials of AX and BW.

Method

Participants

Twenty-five Macquarie University students (18 female, 7 male) participated for course credit. The mean age of participants was 19.6 years (range 18-41; SD = 4.6).

Stimuli

Based upon pilot work, we established that two further combinations of the first set of odors could also support learning. Consequently, in addition to the six mixtures used in Experiment 1A and B, this experiment employed a further two from set 1, namely; BZ-AL (30% BZ, 20% AL, and 50% water) and CG-PA (20% CG, 20% PA, and 60% water). Thus in this experiment each set of odors now yielded four mixtures allowing for more complete counterbalancing than in Experiment 1A. Every other aspect of the stimuli remained the same.

Procedure

The experiment employed a within-participant design, again composed of three phases (see Table 1). The training phase consisted of smelling three types of stimuli, an AW mixture and the elements C and Y. The second phase, interference, involved smelling six different types of stimuli, the interfering mixtures, each of which contained an element common to the AW mixture (i.e., AX and BW) and the elements C, Y and D, Z. This was followed by a two-part test phase. In the first part participants made similarity judgments for three odor pairs, A and W, C and Y, and D and Z. In the second part, they rated the similarity of four further pairs, namely the constituents of the interfering mixtures, A, X and B, W, and the two exposed control conditions C, Y and D, Z.

The training phase involved sniffing 24 stimuli, presented in two sets of 12. In total participants smelled four AWs, four Cs, four Ys, and 12 water only stimuli. Presentation order was random for each participant and as before involved sniffing and rating the strength of each stimulus. Allocation of odor pairs to conditions was as described for Experiment 1 A, except here interference and control stimuli were drawn approximately an equal number of times from each set. Following a 4-min break, participants started the interference phase, in which they sniffed 48 stimuli, composed of four AXs, four BWs, four Cs, four Ys, four Ds, four Zs, and 24 water stimuli. These were presented in sets of 12, with stimuli arranged in random order and a 4-min break separating the second and third sets of 12. After a further 4-min break participants completed the first of a two-part test phase, in which they were presented with three pairs of odors, A and W, C and Y, and D and Z with participants asked to judge the similarity of each pair using the same procedure as described for Experiment 1A. This was immediately followed by a second similarity test in which participants judge the similarity between four pairs of odors, the two mixtures used during the interference phase, A and X and B and W, and the two control pairs, C and Y and D and Z.

Results and Discussion

On the first similarity test participants judged the compound interference pair A and W as most alike and both exposed control conditions as less alike (see Table 1). This difference was confirmed statistically, with A and W being judged more alike than C and Y, t(24) = 4.38, p < 0.001, and D and Z, t(24) = 2.80, p < 0.01. The two exposed control conditions, C and Y and D and Z, did not significantly differ, t < 1.

On the second similarity test, the two interfering mixture condition means were pooled, as they did not significantly differ, t < 1. The two exposed control conditions, CY and DZ were also pooled because, as in the first similarity test, they did not significantly differ either, t < 1. The pooled interfering mixtures were judged as significantly more alike than the pooled exposed controls, t(24) = 2.62, p < 0.02. Evidence of learning was thus obtained for the interfering mixture pairs.

To test whether learning for the interfering mixture pairs (i.e., A vs. X & B vs. W) was similar to or different from that of the compound interference pair (i.e., A vs. W), we compared the size of the learning effect between A and W, mean of C and Y + D and Z on test 1, minus A and W) with that of the interfering pairs, mean of C and Y + D and Z on test 2, minus mean of A and X + B and Y. These two scores significantly differed, t(24) = 2.15, p < 0.05, with a larger learning effect in the compound interference pair A and W relative to the interfering mixture pairs.

：芭芭拉

General Discussion

The main finding from these two studies is that presenting an odor mixture to participants followed by presenting two further odor mixtures, each of which contain an element drawn from the original mixture, does not affect retention of that original encoding, any more than an elemental interference procedure. The elemental interference procedure, as we know from two previous studies, has no effect on learning when compared to a paired control (Stevenson et al., 2003, 2005) and Experiment 1B confirmed this conclusion. In addition, Experiment 2 found evidence that participants had acquired some memory for the interfering mixtures albeit to a lesser extent than for the compound interfered pair. These findings are largely consistent with the general body of results on the persistence of olfactory memory described above but, as we describe below, present some significant problems for the explanatory framework that we have used before to account for this type of perceptual learning.

In previous examinations of odor-odor and odor-taste learning, we have argued that configurai encoding of the respective mixtures is responsible for this effect (e.g., Stevenson & Boakes, 2004; Stevenson, Boakes, & Prescott, 1998). Configural encoding presumes that the combination possesses a quality which is different from, but similar to, the elements from which it is constituted. In olfaction, evidence for encoding as a configuration appears to be the rule rather than the exception, and the strongest support for this idea comes from examining the ecology of olfactory perception. Most olfactory stimuli are composed of 10s or 100s of chemical constituents (e.g., Maarse, 1991), yet our experience of odors is unitary; we smell strawberry, not the 360 or so volatile chemicals which have been identified to constitute strawberry aroma (Latrasse, 1991).

If then, as we suspect, odors are encoded as configurations, what are the implications of this for resistance to interference? Under the elemental interference procedure, as examined in Experiment 1A and reported previously, participants encode the mixture CY as a configuration. Before, we have argued that when the elements C and Y are presented in the interference phase, no further learning takes place; that is, C and Y are not encoded. We suggested this was because the elements redintegrate a memory of the mixture (e.g., C recovers CY) which prevents their encoding taking place. This appears a reasonable enough proposition, because in previous experiments (Stevenson et al., 2003, 2005) the interfered odor pair (e.g., CY) always evidenced the same level of conditioning as an odor pair (e.g., EV) which had been experienced as a mixture, but never with the elements presented alone.

If, however, encoding of the elements C and Y in the example above had taken place, then when C and Y were evaluated for similarity, C should have activated memories of CY and C, while Y should have activated memories of CY and Y. What about the paired control EV in the above example? On the similarity test E should evoke only EV, likewise V should evoke only EV, as the elements were never exposed and thus never learned (or at least not to the same degree). Consequently, E and V should be judged as more similar to each other than B and Y, because the latter have fewer mnemonic features in common. As we noted above, empirically this is not the case. Therefore the finding in Experiment 2 that the interfering mixture pairs (AX and BW) were encoded is a cause for some serious reflection, because it suggests that in the elemental interference procedure (i.e., CY in the example above), the elements should have been encoded and yet the empirical evidence does not appear to back this up.

One resolution to this problem is to assume odor elements are a special case. The elements themselves are likely to be more familiar to participants than their mixtures. Thus participants, arguably, have already encoded, prior to the experiment, the elements and so little further encoding is possible during the presentation of the elements alone during the interference procedure. However, if we assume that elements are already encoded, we then have to accommodate the findings from Experiment 2, which clearly indicate that the interfering mixtures were encoded too. If we include this, as we detail below, then our assumptions about what perceptual information is present during the similarity test must be incorrect. On the similarity test, A should now evoke memories of A, AW, and AX; and W should evoke memories of W, AW, and DW. Now compare this to the elemental interference pair CY, where C should evoke memories of C and CY; and Y should evoke memories of Y and CY. In the elemental interference example the ratio of common to unique elements is 1:1, and in the compound interference example the ratio is 1:2. Thus we would predict greater interference in the compound interference condition and the data from Experiment 1A does not support this.

We are loath to abandon the configural model. Not only are there good grounds for its generality in olfactory perception, as we considered above, its most obvious theoretical alternative, within-event associations, appears an even poorer fit. First, there are no theoretical reasons why within-event associations are not subject to the same constraints as between-event associations. Both should be equally sensitive to interference, and as should now be clear this is not the case for either odor-odor or odortaste learning. Second, empirical evidence also argues against this. Rescorla and Freberg (1978) and Westbrook, Duffield, Good, Halligan, Seth, and Swinborne (1995) have both demonstrated that preexposure (sensory preconditioning) between tastes, or odors and tastes, can be extinguished. Obviously, drawing inferences across procedurally diverse manipulations requires some caution, but within-event associations, as an account, would seem to predict interference here and we do not observe this. For this reason and for the fact that olfactory processing appears to rely upon synthesis rather than analysis, salvaging the configural account appears worthwhile.

One solution is to assume: (1) that all odors are encoded as configurations; (2) that encoding of stimuli presented in the interference phase does occur; (3) that it is less efficient because of a tendency to retain earlier related encodings (primacy) as indicated by previous findings of odor memory persistence (e.g., Lawless & Engen, 1978); (4) that memories can be evoked by similarity (e.g., smelling A evokes a representation of AW); and (5) that memories can also be evoked by association, that is independent of similarity (e.g., smelling A evokes a representation of W, by virtue of an association between AW and W). The representation that results when an odor is smelled is then based upon evocations dictated by perceptual similarity and association. In Experiment 1A then, A from the compound interference condition would strongly evoke A (high similarity match), moderately evoke AW and AX (moderate similarity match), and evoke by association DW and W (by association with AW). C from the elemental interference condition, on balance, would strongly evoke C (high similarity match), moderately evoke CY (moderate similarity match), and evoke by association Y (by association with CY). On this basis, the number of shared similarities would in fact be greater in the compound interference condition [i.e., A vs. W (not shown above)] than in the elemental interference condition [i.e., C vs. Y (not shown above)]. Interestingly in this respect the data in Experiment 1A (although not significant) are in accord with this.

In conclusion, the key findings of this pa-per are that a compound interference procedure is no more effective at altering the perceptual changes induced by presenting an odor as a mixture than the elemental procedure which we have previously explored in depth. Theoretically these findings prove difficult to accommodate within the type of configural framework that we have presented before. Here we offer one possible alternative that at least predicts, post hoc, the type of results that we have obtained.

[Reference]
References
Case, T. I., STEVENSON, R. J., & DEMPSEY, R. A. (2004). Reduced discriminability following perceptual learning with odours. Perception, 33, 113-119.
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.
HALLER, R., RUMMEL, C., HENNEBERG, S., POLLMER, U., & KOSTER, E. P. (1999). The influence of early experience with vanillin on food preference later in life. Chemical Senses, 24, 465-467.
HERZ, R. S., & ENGEN, T. (1996). Odor memory: Review and analysis. Psychonomic Bulletin and Review, 3, 300-313.
KOSTER, E. P., DEGEL, J., & PIPER, D. (2002). Proactive and retroactive interference in implicit odor memory. Chemical Senses, 27, 191-206.
LATRASSE, A. (1991). Fruits III. In F. Maarse (Ed.), Volatile compounds in foods and beverages (pp. 329-376). Marcel Dekker: New York.
LAWLESS, H. T. (1978). Recognition of common odors, pictures, and sim-ple shapes. Perception and Psychophysics, 24, 493-495.
LAWLESS, H. T., & ENGEN, T. (1978). Associations to odors: Interference, mnemonics, and verbal labelling. Journal of Experimental Psychology: Human Learning and Memory, 3, 52-59.
MAARSE, H. (1991). Volatile compounds in foods and beverages. New York: Marcel Dekker.
RESCORLA, R. A., & FREBERG, L. (1978). The extinction of within-compound flavor associations. Learning and Motivation, 9, 411-427.
RICHARDSON, J. T. E., & ZUCCO, G. M. (1-9-8-9). Cognition and olfaction: A review. Psychological Bulletin, 105, 352-360.
SCHAB, F. R. (1991). Odor memory: Taking stock. Psychological Bulletin, 109, 242-251.
STEVENSON, R. J. (2001a). Associative learning and odor quality perception: How sniffing an odor mixture can alter the smell of its parts. Learning and Motivation, 32, 154-177.
STEVENSON, R. J. (2001b). Perceptual learning with odors: Implications for psychological accounts of odor quality perception. Psychonomic Bulletin and Review, 8, 708-712.
STEVENSON, R. J., & BOAKES, R. A. (2004). Sweet and sour smells: Learned synesthesia between the senses of taste and smell. In G. Calvert, C. Spence, & B. Stein (Eds.), The handbook of multisensory processes (pp. 69-83). Massachusetts: MIT Press.
STEVENSON, R. J., BOAKES, R. A., & PRESCOTT, J. (1998). Changes in odor sweetness resulting from implicit learning of a simultaneous odor-sweetness association: An example of learned synesthesia. Learning and Motivation, 29, 113-132.
STEVENSON, R. J., BOAKES, R. A., & WILSON, J. T. (200Oa). The persistence of conditioned odor perceptions: Evaluative conditioning is not unique. Journal of Experimental Psychology: Learning Memory and Cognition, 26, 423-440.
STEVENSON, R. J., BOAKES, R. A., & WILSON, J. T. (200Ob). Counterconditioning following human odor-taste and color-taste learning. Learning and Motivation, 31, 114-127.
STEVENSON, R. J., & CASE, T. I. (2003). Preexposure to the stimulus elements, but not training to detect them, retards human odour-taste learning. Behavioural Processes, 61, 13-25.
STEVENSON, R. J., CASE, T. I. & BOAKES, R. A. (2003). Smelling what was there: Acquired olfactory percepts are resistant to further modification. Learning and Motivation, 34, 185-202.
STEVENSON, R. J., CASE, T. I, & BOAKES, R. A. (2005). Implicit and explicit tests of odor memory reveal different outcomes following interference. Learning and Motivation, 36, 353-373.
STEVENSON, R. J., PRESCOTT, J., & BOAKES, R. A. (1995). The acquisition of taste properties by odors. Learning and Motivation, 26, 433-455.
WALK, H. A., & JOHNS, E. (1984). Interference and facilitation in short-term memory for odors. Perception and Psychophysics, 36, 508-514.
WESTBROOK, R. F., DUFFIELD, T. Q., GOOD, A. J., HALLIGAN, S., SETH, A. K., & SWINBOURNE, A. L. (1995). Extinction of within-event learning is contextually controlled and subject to renewal. Quarterly Journal of Experimental Psychology, 48B, 357-375.

[Author Affiliation]
RICHARD J. STEVENSON, TREVOR I. CASE and CAROLINE TOMICZEK
Macquarie University, Australia

[Author Affiliation]
We thank Cassie Brown, Sarah Jacek, and Margery Aylett for their assistance with these experiments and Dragoco and Quest International for kindly supplying many of the odorants used here. This research was supported by the Australian Research Council. Please address all correspondence to Richard J. Stevenson, Department of Psychology, Macquarie University, NSW 2109, Australia. (E-mail: rstevens@psy.mq.edu.au).