Trial-to-trial dynamics of selective long-term-memory retrieval with continuously changing retrieval targets

Abstract

How do we control the successive retrieval of behaviorally relevant information from long-term memory (LTM) without being distracted by other potential retrieval targets associated to the same retrieval cues? Here, we approach this question by investigating the nature of trial-by-trial dynamics of selective LTM retrieval, i.e., in how far retrieval in one trial has detrimental or facilitatory effects on selective retrieval in the following trial. Participants first learned associations between retrieval cues and targets, with one cue always being linked to three targets, forming small associative networks. In successive trials, participants had to access either the same or a different target belonging to either the same or a different cue. We found that retrieval times were faster for targets that had already been relevant in the previous trial, with this facilitatory effect being substantially weaker when the associative network changed in which the targets were embedded. Moreover, staying within the same network still had a facilitatory effect even if the target changed, which became evident in a relatively higher memory performance in comparison to a network change. Furthermore, event-related brain potentials (ERPs) showed topographically and temporally dissociable correlates of these effects, suggesting that they result from combined influences of distinct processes that aid memory retrieval when relevant and irrelevant targets change their status from trial to trial. Taken together, the present study provides insight into the different processing stages of memory retrieval when fast switches between retrieval targets are required.

Introduction

Different items in memory are often associated with the same retrieval cue. For example, when writing this introduction, the contents of many journal papers have to be retrieved in a specific order, but all of these long-term memory (LTM) representations are related to one and the same super-ordinate cue, e.g., “the control of memory retrieval”. How do we manage to retrieve just one currently relevant association without being distracted by other associations triggered by the same retrieval cue? Theories about the organization of LTM assume that information is represented in networks in which representations are connected according to their associative strength and in which retrieval of one item causes a spreading of activation across associated items (Anderson, 1976, Anderson, 1983). Depending on the task, this spreading of activation may lead to a facilitation of retrieval (Rubin & Olson, 1980), but can also result in retrieval interference (e.g., Bäuml, 2008, Bäuml et al., 2010, Ciranni and Shimamura, 1999). Specifically, when selective retrieval of only one target associated with a cue is required, there is evidence for activation spread leading to interference due to the activation of other items associated with the same cue that compete for retrieval (Johnson and Anderson, 2004, Levy and Anderson, 2002). In free recall tasks, in which all items associated with a cue have to be retrieved anyway, the activation spread has been reported to be beneficial as it shortens the RT for consecutive retrieval (Lorch, 1982, Neely, 1976, Posner and Snyder, 1975). However, in other retrieval tasks in which only one retrieval target is relevant cognitive-control processes are assumed to regulate interfering activation patterns by selectively enhancing relevant and/or by inhibiting associated but irrelevant information. As a consequence, relevant associations are strengthened, and thus better retrieved on later occasions, while irrelevant associations are weakened and less well accessed later on (Anderson, 2003, Anderson and Spellman, 1995, Bäuml and Samenieh, 2010, Goodmon and Anderson, 2011, Johansson et al., 2007, Wimber et al., 2008). The retrieval-practice paradigm (Anderson, Bjork, & Bjork, 1994) allows to study such effects. It has shown that memory of non-practiced associations can be impaired just by practicing other related associations after initial encoding (Anderson et al., 1994, Bäuml, 2008, Ciranni and Shimamura, 1999, Levy and Anderson, 2002). This “retrieval-induced forgetting” (RIF) suggests that existing but not-retrieved associations are weakened, possibly actively inhibited, if other related associations are repeatedly activated. On the other hand, it has been shown that, depending on the status of the associations, selective retrieval can also lead to enhanced accessibility of non-practiced associations with a cue through spreading activation, e.g., when previously to-be-forgotten associations need to be remembered (Bäuml and Samenieh, 2010, Dobler and Bäuml, 2012) or when the information is highly integrated and testing is delayed about 24 h (Chan, 2009). To conclude, depending on the retrieval task facilitatory or detrimental effects can be observed.

The present study builds upon these findings. However, rather than studying such effects in distinct practice and retrieval episodes, as has been done in previous studies, we investigated which processes, detrimental or facilitatory, or both, govern retrieval on a trial-by-trial basis, i.e., when irrelevant associations (potential retrieval competitors) in one trial might become relevant in the immediately following trial.

To investigate dynamic changes of retrieval processes, we manipulated trial-to-trial changes and repetitions of cue–target associations. Participants established small associative networks in LTM, each consisting of an animal and three characteristics that were learned as pictures/symbols (i.e., body weight, sociability, and distance of habitat; see Fig. 1A). Participants were informed that a memory test would follow. During the retrieval task (Fig. 1B), the animals’ names served as cues, indicating the relevant associative network. An immediately following stimulus indicated one of the three associated characteristics that was the retrieval target of the present trial.

In each retrieval trial (Fig. 1B), only one target had to be retrieved, the other two had to be ignored. While all trials involved the same task, i.e., to retrieve one of three possible cue–target associations, they differed in their retrieval history, yielding the four different experimental conditions that are depicted in Fig. 2, i.e., the cue and the target could remain the same in two consecutive trials i − 1 and i (condition #1), the cue could be repeated while the target changed (condition #2), the cue could change while the target remained the same (condition #3), or both cue and target changed (condition #4). Condition #4 served as a baseline condition for which no immediate trial-to-trial effect of selective retrieval – be it facilitatory or inhibitory – was to be expected.

The conditions were motivated by studies in other fields that successfully investigated continuously changing demands on interference control by manipulating sequences of trials, e.g., in task-switching and backward-inhibition experiments (e.g., Dreher and Berman, 2002, Mayr and Keele, 2000), or selective-attention studies (e.g., Neill, 1997, Stadler and Hogan, 1996; for a review see Tipper, 2001). These studies proved the existence of highly dynamic trial-to-trial control processes that reduce interference (from competing task sets or visually presented stimuli) by facilitating the repeated processing of the target information or impairing the processing of previously irrelevant competitors for the next trial(s). By transferring this concept of trial-to-trial interference control to LTM retrieval, we wanted to find out whether similar dynamic control processes are involved when the interference arises on the level of LTM representations.

Compared to all other conditions, it can be expected that condition #1 in which everything is repeated would be the easiest condition, leading to the shortest RTs and fewest errors, because it is a typical positive priming situation in which the same cue–target association has to be accessed a second time. Note that with the present paradigm such a facilitation can be interpreted as repetition priming on the level of memory representations instead of a perceived stimulus or motor response, as we especially controlled for stimulus and response priming (for details, see Section 2.3).

Whereas the prediction for the full repetition condition is relatively straightforward, the predictions for the other three conditions strongly depend on the principal process that is assumed to regulate the trial-to-trial dynamics of selective LTM retrieval, i.e., inhibition or facilitation: On the one hand, the retrieval-induced-forgetting effect (Anderson, 2003) suggests that retrieval competition between associations with the same cue triggers inhibitory control. Accordingly, the consecutive retrieval of different targets associated with the same cue (condition #2) should be more effortful and should lead to poorer performance than the consecutive retrieval of completely unrelated targets (condition #4). However, because in our study all associations related to one cue are learned in one learning session to the same extent, the opposite result is also conceivable, i.e., that, due to substantial over-learning, each network will be highly integrated, and thus retrieval competition might be relatively weak while simultaneously evoked spreading activation might lead to a facilitation of the subsequent retrieval of another target associated the same cue (Anderson, 1983, Chan, 2009, Reder and Anderson, 1980). Then, performance should be better when shifting from one target to another associated with the same cue (condition #2) than shifting between unrelated targets (condition #4). Finally, as participants learned distinct cue–target networks, conditions with a cue change should not differ, regardless of whether the target is repeated or not (condition #3 = condition #4). In both cases a new network has to be accessed. However, conceptual priming cannot be excluded completely in the cue-change target-repetition condition (#3) because the target category, i.e., weight, distance, or sociability, is repeated. This could lead to some processing facilitation. To conclude, the obtained pattern of results can help to determine the kind of processes that are related to trial-by-trial control of selective LTM retrieval.

Alongside behavioral data, we recorded event-related brain potentials (ERPs) to study the temporal dynamics and possible qualitative and quantitative differences of the involved retrieval processes that are likely to be reflected in topographical and amplitude differences (Otten and Rugg, 2005, Picton et al., 2000). In a previous study, we found evidence for ERP differences depending on the retrieval demands of successive trials (Kizilirmak, Rösler, & Khader, 2012), i.e., when the number of to-be-retrieved associations changed from one trial to the next. The present study builds upon this finding, but rather than focusing on retrieval load, we here want to delineate the processes involved in successively retrieving associations from LTM belonging to either the same or to different associative networks. In contrast to our previous study, here retrieval load was held constant. Moreover, we now included the repetition of cue and target (condition #1) in order to enable a full factorial design.

Based on our previous study and on general considerations about the temporal evolution of ERP differences related to memory retrieval (e.g., Rugg & Curran, 2007), we expected ERP effects to show up in three different time windows after the onset of the retrieval cue, i.e., 250–500 ms, 500–1000 ms, and 1000–3000 ms. There are two ERP components during the first two time windows that are classically related to memory processes. The first component is usually found around 350–400 ms with a topographical maximum at mid-frontal electrodes. It has been observed for familiar old, i.e., recognized, in contrast to unfamiliar new items, while it does not vary with the degree of recollection of the learning event (Rugg and Curran, 2007, Vilberg et al., 2006). Although there is an ongoing debate whether it reflects unconscious conceptual priming or rather conscious recognition (Bridger et al., 2012, Mecklinger et al., 2012, Paller et al., 2012), this effect is generally assumed to reflect relatively automatic retrieval processes. Our paradigm does not allow dissociating familiarity from conceptual priming effects, however, if an effect in this time window will be found, we can be relatively confident that it reflects automatic processes. The mid-frontal old/new effect has never been investigated on a trial-by-trial basis, but it can be expected to be especially pronounced for the condition in which both cue and target are repeated (cf. Fig. 2: condition #1 > condition #2), because in target-change trials the target is relatively “new” compared to target-repetition trials. Therefore, this ERP effect can be seen as an indicator of processes that facilitate performance in trial-by-trial selective retrieval. If such an effect would also occur for the condition in which the target is repeated although the cue had changed, it would indicate that not only specific cue–target associations are primed/recognized, but that also the target dimension itself can profit from a repetition, suggesting conceptual priming.

The second component is usually observed with a peak around 500 and 800 ms and a topographical maximum over parietal electrodes (Rugg and Curran, 2007, Wilding, 2000). It has been associated with intentional recollection, i.e., not only recognizing an item as being old, but also being able to recall other information associated with the item. The effect has also been shown to be sensitive to the amount of information recollected (Vilberg et al., 2006) and the recency of its last encounter (Grove & Wilding, 2009). Since recollection can only be facilitated for cue and target repetition, we hypothesize that this component should only show an effect for target repetition vs. target change in cue repetition (condition #1 > condition #2), but not in cue-change trials (condition #3 = condition #4). That is, even though target repetition might lead to familiarity effects even for cue change, since only the target category is repeated but not the target representation itself, there should be no benefit in recollection.

Lastly, the third time window is at the latency of slow cortical potentials (SCPs). In our previous study, we had focused the analysis on these ERPs and found that their amplitude was systematically affected by the necessity to adjust retrieval-control processes from trial to trial (see Kizilirmak et al., 2012 for details). Both the topography and amplitude of negative-going SCPs generally show a close correspondence with BOLD effects (e.g., He and Raichle, 2009, Jost et al., 2011; see Khader, Schicke, Röder, & Rösler, 2008, for a review), which is why their latency does not necessarily correspond to the temporal occurrence of the underlying cognitive process. Negative SCP shifts most likely reflect a sustained field potential arising from the depolarization of the apical dendrites of neocortical pyramidal cells (e.g., Speckmann & Elger, 2005). Therefore, negative SCPs can give a rough estimate about cortical structures which are involved in specific processing episodes. One effect in our previous study had a frontal topography and suggested that control mechanisms during selective retrieval affect targets associated with the same cue more than targets that are fully unrelated. We interpreted this effect as reflecting control processes that might facilitate an attentional shift from one target to another associated with the same cue, in line with right lateral prefrontal cortex being associated with the control of response interference and resolution (Aron, 2007, Aron et al., 2004) and with an attention shift facilitated by inhibiting a previously attended object, location, or dimension (Hampshire, Chamberlain, Monti, Duncan, & Owen, 2010).

Another SCP effect had a bilateral parietal topography and emerged when retrieval load decreased from three to one. We interpreted this effect in line with the finding that parietal regions (i.e., intraparietal sulcus, superior parietal lobe) are involved in the modulation of attention, and probably also in the modulation of internal attention towards memory representations (Cabeza et al., 2008, Chun et al., 2011, Ciaramelli et al., 2010, De Brigard, 2012).

Based on these previous observations and on related effects reported in the literature, we arrived at specific predictions with respect to frontal and parietal SCPs: First, we expect a right-frontal ERP effect for shifts from one target to another related to the same cue (compared to repeatedly accessing the same target), because here the previously accessed target might have to be inhibited and/or the focus on the relevant needs to be enhanced due to interference within the cue–target network (contrast of condition #2 vs. #1). If SCP amplitude would be relatively more negative for target change than for target repetition, this would suggest an increased task-related activation when shifting the focus from one association to another within the same associative network, probably reflecting a higher retrieval effort. The increased retrieval effort can be either due to interference from lingering activation of the previous target, necessitating higher cognitive control, or due to increased effort when a previously inhibited retrieval competitor needs to be accessed and therefore be released from inhibition. As a control, we should not see any of such effects when the cue–target network changes (condition #3 vs. #4), or the effects should be at least strongly reduced, with possible residual facilitatory target effects emerging through conceptual priming of the target category across cues.

Second, parietal regions (i.e., intraparietal sulcus, superior parietal lobe) are assumed to be involved in the modulation of attention, i.e., in the case of internal attention towards memory representations (Cabeza et al., 2008, Chun et al., 2011, Ciaramelli et al., 2010, De Brigard, 2012). Thus, we expect to find parietal ERP amplitude differences related to when attention is directed towards the target association. The cue and target repetition condition (condition #1) is expected to involve the least attentional resources, and should thus produce the smallest parietal SCP effect (because attention does not have to be directed towards a different association). In comparison, shifting the attentional focus from one target to another associated with the same cue (condition #2), should be more difficult and evoke higher parietal activation. Again, as a control, we should see no or at least strongly reduced effects when the cue–target network changes.

To conclude, the different retrieval-related ERPs can support the behavioral data in delineating the kind of process that is generally involved in trial-by-trial memory retrieval dynamics, and, in addition, can be used to specify the possible sub-processes that modulate memory retrieval when relevant and irrelevant targets change their status from trial to trial.