Tactile memory Ranschburg effects under conditions of concurrent articulation

Introduction

Immediate serial recall (ISR) requires participants to recall an earlier presented sequence in its order of original presentation. The Ranschburg effect is an ISR phenomenon characterised by the disinclination to recall repeated items from the previously presented sequence (e.g., Crowder, 1968; Henson, 1998a; Jahnke, 1969). Thus, when an item is repeated during the presentation phase of an ISR trial, and that repetition is spaced (i.e., non-adjacent), participants are disinclined to recall the second presentation of the repeated item (e.g., Armstrong & Mewhort, 1995; Crowder, 1968; Duncan & Lewandowsky, 2005; Henson, 1998a; Jahnke, 1969; Maylor & Henson, 2000). The Ranschburg effect has been attributed to a response suppression mechanism acting at test (Armstrong & Mewhort, 1995; Vousden & Brown, 1998), such that, following recall of the first presentation of the repeated item, that item is then suppressed, thereby inhibiting recall of its second presentation. When sequences do not contain a repetition, a response suppression mechanism is of utility because it discourages perseveration at recall, as evidenced by the low proportion of erroneous within-trial repetitions (estimated at between 2% and 5% of all responses; Henson, Norris, Page, & Baddeley, 1996; Vousden & Brown, 1998). In contrast, when an item is repeated in the sequence, response suppression acts to inhibit recall of its second presentation. That the effect reflects the operation of an output suppression process is evidenced by those studies showing an absence of the Ranschburg effect for either probed or partial recall tasks (e.g., Armstrong & Mewhort, 1995; Jahnke, 1970). That is, when a partial recall task requires recall only of the second presentation of the repeated item (and not the first), recall for the second presentation is not impaired (Armstrong & Mewhort, 1995).

Response suppression (from which the Ranschburg effect is thought to be epiphenomenal) is a feature common to several models of order memory (e.g., Burgess & Hitch, 2006; Henson, 1998b; Page & Norris, 1998). This is of particular importance for ordinal models of serial order in which activation along an exponentially declining primacy gradient is used uniquely to determine the sequence recall (e.g., the Primacy Model; Page & Norris, 1998). For such ordinal accounts, in the absence of response suppression, early list items (possessing high activation levels) will be retrieved repeatedly (perseveration), thus preventing retrieval of later list items (possessing lower activation levels). Although such ordinal models were developed within the context of verbal memory, as noted by Hurlstone, Hitch and Baddeley (2014), experimental evidence examining the extent to which these models are applicable to non-verbal stimuli is absent. While a number of order memory phenomena, for example, serial position curves (e.g., Avons, 1998; Guérard & Tremblay, 2008; Parmentier & Jones, 2000; Ward, Avons, & Melling, 2005), error distributions (e.g., Guérard & Tremblay, 2008; Smyth, Hay, Hitch, & Horton, 2005), and the Hebb repetition effect (e.g., Couture & Tremblay, 2006; Horton, Hay, & Smyth, 2008; Johnson, Cauchi, & Miles, 2013; Johnson, Shaw, & Miles, 2016; Page, Cumming, Norris, Hitch, & McNeil, 2006), demonstrate comparable functioning across both verbal and non-verbal stimuli, evidence for a cross-modal Ranschburg effect is limited. Given that the operation of a response suppression mechanism is central to accounts of serial order memory, it is, therefore, of theoretical importance to test for its presence with ISR of non-verbal stimuli.

A recent study (Johnson, Hawley, & Miles, 2018) examined the Ranschburg effect for ISR of visually presented consonant sequences under conditions of both quiet and concurrent articulation (CA). Under the CA condition, the phonological similarity effect (PSE) was abolished—a finding taken to reflect disruption to the process of phonological recoding of the visually presented consonant sequences (e.g., Baddeley, Lewis, & Vallar, 1984). Importantly, despite the purportedly degraded phonological representations of the consonant sequences, Johnson et al. (2018) demonstrated a Ranschburg effect of comparable magnitude under conditions of both quiet and CA.

Notwithstanding the use of CA, and the resilience of the Ranschburg effect in Johnson et al. (2018), their to-be-remembered sequences comprised verbal stimuli. In an attempt to minimise further the probability for verbal recoding of the to-be-remembered sequences, Roe, Miles, and Johnson (2017) tested ISR for tactile sequences. Here, for the presentation phase, visually obfuscated participants received sequences of six-individual stimulations to each of six fingers. For the test phase, participants were required to recall the sequence by lifting individual fingers in the order of original presentation. A Ranschburg effect was evident such that for sequences containing spaced repetitions (i.e., the item in Position 5 was a repeat of that in Position 2), recall for the second presentation of the repeated item was adversely affected. This epiphenomenal evidence for response suppression in tactile memory is supported further by earlier data showing low levels of erroneous within-trial repetitions in tactile order memory (4.2% of responses; Johnson et al., 2016). Moreover, on those infrequent trials in which an erroneous repetition did occur, the interval between repetitions was large (mean interval = 3.34 items)—a finding consistent with attenuation of item suppression.

Roe et al. (2017) argued that their finding reflected processes within non-verbal memory and was, therefore, evidence in support of a cross-modal Ranschburg effect. However, one caveat to their argument is the possibility that their pattern of data merely reflected a process in which the tactile sequences were recoded into verbal representations (as originally supposed by Mahrer & Miles, 2002). Against this verbal recoding interpretation, Roe et al. (2017) point to earlier data from Mahrer and Miles (1999) demonstrating robust tactile ISR under conditions of both backward counting and CA. However, the Mahrer and Miles (1999, 2002) studies were not designed explicitly to examine the Ranschburg effect, and thus, the possibility remains that participants in Roe et al. (2017) were supplementing sequence recall via verbal recoding strategies and that the reported Ranschburg effect was (at least, partially) a product of verbal recoding.

In the light of the above, the current experiment was designed to test directly the possibility that the tactile Ranschburg effect reported by Roe et al. (2017) was a product of verbal recoding. Participants completed two 20-trial blocks, with each trial comprising a sequence of six tactile stimulations, under conditions of both quiet and CA. CA was employed to minimise the possibility for verbal recoding of the tactile stimuli (e.g., Baddeley et al., 1984; Saito, Logie, Morita, & Law, 2008). Each block comprised an equivalent number of repetition trials, containing a repetition (at Positions 2 and 5), and matched non-repetition control trials. To the extent that the tactile Ranschburg effect manifests independently of those processes controlling verbal recoding and rehearsal, then a reduction in correct recall for the second presentation of the repeated item is predicted under conditions of both quiet and CA. In addition, and in response to a suggestion during the reviewing process, we analyse the pattern of error responses to further our understanding of the processes underpinning the Ranschburg effect. First, given the proposal that both the Ranschburg effect and the associated low frequency of erroneous within-trial repetitions are direct products of response suppression (e.g., Vousden & Brown, 1998), we predict a significant correlation between these two effects. That is, to the extent that a participant exhibits high response suppression, then this should be reflected by both a strong Ranschburg effect and a reduction in the number of within-trial repetition errors. Second, we examine the pattern of responses for those trials in which the participant fails to recall the repeated item within a Ranschburg sequence. As response suppression should reduce the likelihood of recalling an item that has already been outputted, we predict that participants will exhibit an increased probability of recalling the item from the stimulus set omitted from that particular sequence. That is, when failing to recall the repeated item, that item will be replaced at recall by retrieval of the finger not used in that specific trial.

Method

Results

Serial position analysis

For the serial position analysis, a strict scoring criterion was adopted such that a response was only recorded as correct if the correct finger was moved at the correct serial position.

Figure 1a and b shows the serial position functions for the control and Ranschburg trials under conditions of both quiet (Figure 1a) and CA (Figure 1b). The functions demonstrate both primacy and recency together with a reduction in recall accuracy for the repeated item (i.e., Serial Position 5) under both conditions.

OPEN IN VIEWER

Figure 1.

Mean proportion correct for the control and Ranschburg trials as a function of serial position under conditions of (a) quiet and (b) CA.

A three-factor (2 × 2 × 6) within-participants analysis of variance (ANOVA) was computed with the factors: concurrent task (quiet and CA), trial type (control and Ranschburg), and serial position (1-6). The main effect of concurrent task was significant, F(1, 19) = 6.938, mean square error (MSE) = .102, p = .016, η p 2  = .267, demonstrating impaired recall for the CA condition (mean proportion correct and 95% confidence interval [CI] for the quiet and CA conditions = .511 [.440, .583] and .435 [.371, .498], respectively). The main effect of trial type was non-significant, F(1, 19) = 1.364, MSE = .057, p = .257, η p 2  = .067. The main effect of serial position was significant, F(5, 95) = 52.036, MSE = .042, p < .001, η p 2  = .733, reflecting strong primacy and a suggestion of recency. The two-factor interactions between concurrent task and trial type (F < 1), and concurrent task and serial position, F(5, 95) = 1.337, MSE = .019, p = .255, η p 2  = .066, were non-significant. Importantly, the predicted interaction between trial type and serial position was significant, F(5, 95) = 3.589, MSE = .023, p = .005, η p 2  = .159. Paired comparisons were conducted comparing the control and Ranschburg conditions at Serial Positions 1 to 6. These comparisons included Bayes factors computed using default priors with JASP (JASP Team, 2018). For these comparisons, BH₁₀ refers to evidence in support of the experimental hypothesis, where BF₁₀ > 3 is taken as substantial evidence in support of the experimental hypothesis and BF₁₀ < 1/3 is taken as substantial evidence in support of the null hypothesis. The control and Ranschburg conditions did not significantly differ at Serial Position 1: t(19) = 0.167, p = .869, d = .037, BF₁₀ = 0.235; Serial Position 2: t(19) = 0.045, p = .964, d = .010, BF₁₀ = 0.233; Serial Position 3: t(19) = –0.509, p = .617, d = –.114, BF₁₀ = 0.261; Serial Position 4: t(19) = –0.828, p = .418, d = –.185, BF₁₀ = 0.315; and Serial Position 6: t(19) = 0.910, p = .374, d = .204, BF₁₀ = 0.336. However, performance was significantly reduced at Serial Position 5 for the Ranschburg condition, t(19) = 3.336, p = .003, d = .746, BF₁₀ = 12.500, consistent with inhibition for the repeated item. Of primary theoretical importance is the finding that the Ranschburg effect was resistant to CA, as evidenced by the non-significant three-way interaction, F(5, 95) = 1.553, MSE = .010, p = .181, η p 2  = .076. This finding is consistent with the view that the Ranschburg effect persists in the absence of verbal recoding of the sequence items.

Discussion

Two important findings are signalled from this experiment. First, it has replicated the only previous demonstration of the Ranschburg effect (i.e., the inhibitive effect of spaced within-sequence repetitions) for recall of non-verbal sequences (Roe et al., 2017). Specifically, using a tactile ISR task, we have demonstrated recall inhibition for a repeated item when that item is separated by two intervening items (the Ranschburg effect). This finding is consistent with a number of studies showing the effect for recall of verbal sequences (e.g., Crowder, 1968; Duncan & Lewandowsky, 2005; Henson, 1998a; Jahnke, 1969; Johnson et al., 2018; Maylor & Henson, 2000). Second, we have reported a Ranschburg effect of equivalent magnitude under conditions of both quiet and CA. The latter finding is important as we argue that, for CA, verbal recoding of the tactile stimulations is disrupted, and, despite such disruption, the Ranschburg effect persisted. Survival of the effect points to the conclusion that the tactile Ranschburg effect cannot be regarded as merely reflecting those processes underpinning verbal memory; rather, we argue, our finding to be bona fide evidence for a non-verbal Ranschburg effect. Moreover, as the present Ranschburg effect matches that found for verbal materials, parsimony would suggest, therefore, that a mechanism common to both tactile and verbal memory underpins these data.

An additional contribution of the current study is the analysis of both response errors and response omissions, and how this informs our understanding of processes underpinning the Ranschburg effect. It has been suggested that both the low frequency of erroneous within-sequence repetitions and the Ranschburg effect reflect the action of a common process (e.g., Vousden & Brown, 1998). Specifically, they are each a function of a response suppression mechanism for which inhibition of an item post-retrieval reduces the likelihood of subsequent recall of that item within a trial. To test this proposal, we correlated the average proportion of erroneous within-sequence repetitions for the control sequences with the average delta score (our measure of the Ranschburg effect). We report a significant positive correlation between these two variables, providing tentative support for supposing that both effects are determined by a common “suppression” mechanism (notwithstanding the finding that the evidence from the Bayes factor was anecdotal). We note, in addition, that the proportion of within-sequence repetitions in the control condition was somewhat greater (9.0% and 10.5% for the quiet and CA conditions, respectively) than that which we reported previously for ISR of tactile sequences (4.2% of all responses; see Johnson et al., 2016). Of course, the current experiment included Ranschburg trials (i.e., the trials containing within-sequence repetitions) which Johnson et al. (2016) did not. It is plausible to suggest that the current inclusion of within-sequence repetitions acted to “prime” participants to output a greater proportion of repetitions at test.

A second error analysis examined what it is that participants recall when there is a failure to recall the repeated item in the Ranschburg trial. We find that for 25.7% of the erroneous responses, participants recall the item (i.e., finger) not included in the sequence. That is, given that we employ a set size of 6 to create all the sequences in the present experiment (i.e., the same six fingers), a Ranschburg trial utilises only five of the fingers (because a repetition is included in those trials). Thus, for those occasions when participants erroneously fail to recall the repeated item, they recall the finger omitted from the sequence in over a quarter of trials. However, as highlighted in the error analysis, the most common response error for the repeated item was erroneous retrieval of Serial Position 6 (28.4% of all Serial Position 5 errors in the Ranschburg trials). Importantly, this recall error was rarely followed by correct (and therefore repeated) recall of Serial Position 6 (on 2.2% of these trials), consistent with response suppression of Serial Position 6. Instead, on 48.9% of these trials, this recall error was followed by either a response omission or retrieval of the item not included in the sequence, that is, a response that was not an erroneous within-sequence repetition. There are a number of reasons why these findings are of consequence with regard to the control processes underpinning the Ranschburg effect. First, if the elevated proportion of errors at Serial Position 5 for the Ranschburg trials was entirely due to participants erroneously repeating another sequence item, then this would contradict the response suppression explanation simply because logic dictates that those erroneous repetitions should also have been suppressed post-retrieval. Our analysis demonstrates that for a substantial proportion of trials, participants recall a non-repeated item (i.e., an item that was not suppressed). In addition, we report that participants failed to recall the entire sequence (i.e., outputting < 6 items) to a greater extent in the Ranschburg trials (9% of trials) compared with the control trials (6% of trials). Although this difference did not reach statistical significance (p = .061), it does plausibly provide another route by which participants avoided repeating items when making an error in recalling the Ranschburg sequence. In summary, while participants do produce erroneous within-sequence repetitions following errors in the Ranschburg trials, there is evidence for some disinclination in the outputting of such repetitions.

Second, a tendency to recall the omitted finger rather than the repeated finger in the Ranschburg trial does lend some support for the guessing account of the Ranschburg effect. Greene (1991) suggested that, rather than reflecting a memory phenomenon, the Ranschburg effect reflects the strategies that participants employ when guessing. Specifically, the idea here is that when a participant is uncertain of the correct response, and, therefore, their response constitutes a guess, that guess tends not to be the recall of an already outputted item, but rather recall of a yet-to-be recalled item from the stimulus set. This strategy acts to greatly reduce the likelihood of recalling the repeated item by chance, and explains why recall of that repeated item is impaired (although see Henson, 1998a, who demonstrated that guessing cannot fully account for the effect).

Third, a tendency to recall the omitted finger rather than the repeated finger in the Ranschburg trial might be taken as evidence for “transfer of learning” of the matched control sequence to the Ranschburg trial. As noted during the review process, for 50% of the trials, the matched control trial precedes the Ranschburg trial (these two trials are identical but for the repetitions at Serial Positions 2 and 5). It is possible that some initial learning of the control sequence occurs following that single presentation (i.e., the tactile Hebb effect, as reported by Johnson et al., 2016), and therefore when outputting the Ranschburg trial, the failure to recall the repeated item is a result of participants falsely identifying the Ranschburg trial as the initial control trial and attempting to recall that sequence. We argue against this explanation for the following reason. In the 50% of trials where the Ranschburg trial precedes the control trial, we would also predict a drop in accuracy for Serial Position 5 due to learning of the Ranschburg sequence. These reciprocal effects should therefore even out any difference between the control and Ranschburg trials at Serial Position 5; however, this difference persists. Moreover, when we look at the errors for the matched control trials that follow the corresponding Ranschburg trial, there is not a single trial in which participants erroneously repeat the Serial Position 2 item at Serial Position 5. This contradicts “transfer of learning” from the Ranschburg list to the matched control list.

That ISR for tactile sequences exhibits a Ranschburg effect mirroring that found for ISR of verbal sequences is consistent with a range of data reflecting functional similarity for verbal and tactile memory processes. First, the serial position function for ISR of tactile stimuli (Johnson et al., 2016; Mahrer & Miles, 1999; Watkins & Watkins, 1974; see also the present experiment) matches closely that reported for verbal sequences (e.g., Bhatarah, Ward, & Tan, 2008; Drewnowski & Murdock, 1980; Grenfell-Essam & Ward, 2012; Spurgeon, Ward, & Matthews, 2014; Tan & Ward, 2007, 2008; Ward, Tan, & Grenfell-Essam, 2010). Second, the distribution of transposition errors in tactile ISR (Johnson et al., 2016) matches closely that observed for verbal and spatial sequences (Guèrard & Tremblay, 2008). Third, both the serial position function and the list-length-related changes in recall strategy for immediate free recall of tactile sequences match closely those observed for verbal sequences (Cortis, Dent, Kennett, & Ward, 2015). Fourth, the Hebb repetition effect is observed for ISR of tactile sequences, mirroring that observed for verbal sequences (Johnson et al., 2016). Taken together, we argue that these findings add to that body of work (Cortis et al., 2015; Johnson et al., 2016; Roe et al., 2017) consistent with the thesis that order memory for tactile stimuli is functionally equivalent to that for verbal memory.

The present data, combined with that of Roe et al. (2017) and Johnson et al. (2018), support the notion that the Ranschburg effect is not confined to phonological memory. As an effect purportedly epiphenomenal to response suppression, demonstrating the Ranschburg effect cross-modally is an important step in testing the applicability of models of order memory beyond the verbal domain. Indeed, Hurlstone et al. (2014) suggest that common sequencing principles operate across different domains of memory, particularly with respect to competitive queuing and response suppression. Notwithstanding the present findings, it is, therefore, important to examine evidence for the Ranschburg effect in other non-verbal domains (e.g., visual/visuospatial memory). Such work provides methodological challenges because non-verbal visual stimuli cannot be applied to ISR, with serial order reconstruction (SOR) used as an alternative (e.g., Avons, 1998; Guérard & Tremblay, 2008; Ward et al., 2005). In SOR, all list items are re-presented at test, and plausibly, such re-presentation may act to prime participant awareness of the repetition based upon the number of items re-presented in the testing array. In addition, item generation is not required at test in SOR, because participants are only required to reconstruct order by selecting the items in the order of original presentation. It remains an empirically testable question the extent to which response suppression only follows item generation/recall at test (as is required in ISR), or whether SOR can induce response suppression for an item.

In summary, the present study shows the Ranschburg effect with non-verbal stimuli. Specifically, we have demonstrated that this response inhibition effect is not an artefact of verbal recoding, and add to a body of work suggesting functional equivalence across verbal and tactile order memory.