Research reportNeural correlates of episodic memory: Associative memory and confidence drive hippocampus activations
Introduction
In standard recognition memory tasks participants judge whether a presented item has been recently encountered during a learning session (OLD), or not (NEW). The literature converges on the view that memory relies on the functioning of a medial temporal lobe (MTL) network comprising the hippocampus and anterior and posterior parahippocampal brain regions [1]. Models of recognition memory propose that two components contribute to successful recognition, namely familiarity and recollection [1], [2]. Whereas familiarity reflects the intuitive knowledge that something has been encountered before without the need to remember specific details, recollection refers to remembering specific contextual details of OLD items [3], [4]. The exact contribution of these processes to recognition memory is still under debate [1], [5], [6]. Most notably, memory strength, i.e. the differentiation between weak and strong memories, has been identified as a likely confounding variable underlying recollection- and familiarity-based decisions in recognition memory [1], [5]. In fact it is becoming hard to differentiate between both concepts and an adequate experimental paradigm is required to advance our knowledge, especially concerning the function of the hippocampus [7], [8], [9], [10], [11].
In this study, we will examine two variables that affect memory strength, confidence assessment and associative memory, and their interaction. First, associative activation can increase the memory strength of an item [12], also termed relational or associative memory processes [13], [14]. Associations to other items in the task context were found to activate a context memory brain network at encoding ([15], see below) and to engage recollection at retrieval [14]. A high amount of associations is expected to drive recollection, because more specific details about the stimulus (i.e., associated items) can be remembered [16], [1]. Such contextual between-item-associations, for example, elicit the ‘false memory effect’ in the DRM paradigm [17], [18]: learning associated items (e.g., “table,” “sit,” “legs”) to a non-learned target item (e.g., “chair”) strengthens the erroneous recall or recognition of a target item. Thus, remembered contextual details and, in particular, a manipulation of the between-item-associations is known to enhance the strength of the episodic memory signal [12] and to drive recollection [4].
Second, memory strength is also indicated by the subjective confidence with which a studied item is remembered [19], [4]. Familiarity- and recollection-based decisions differ with respect to the subjective level of confidence [20]. Recollection-based decisions are generally made with a higher level of confidence. Thus, the standard recognition memory paradigm can be extended by judgments of the confidence with which these decisions are made. For example, using a six-point rating scale ranging from surely NEW (1) over unsure new (3) to unsure (4) or sure OLD (6) provides a more continuous measure of the memory signal. Participants deliver most recollection-based judgments with high confidence, because of the remembered contextual details, whereas lower levels of subjective confidence accompany most familiarity-based judgments [20].
At the neural level, the hippocampus is the target region of a recognition memory network ([21], [22], [1], but see [23]). For example, Yonelinas et al. [21] examined recollection- and familiarity-based recognition memory judgments at a comparable level of confidence by asking their participants to provide either a recollective ‘remember’ response or a non-recollective ‘know’ response (at varying levels of confidence). Recollection-related brain activation was examined by contrasting the neural responses to confidently remembered items and highest confidence familiar items (see [22] for a similar approach). This contrast identified a network of brain regions associated with recollection: bilateral hippocampus, the medial frontal gyrus (MFG), the posterior cingulate gyrus (PCG), and the left superior temporal gyrus (STG; see Fig. 1 in [21]). Still, it is questionable whether memory strength differs between confidently remembered and high-confidence familiar items [5], [24] emphasizing an overestimation of recollection-related activations in these brain regions. Of note is the high overlap between the recognition memory network and the context memory network, i.e. the same brain regions have been identified as supporting the processing of contextual associations [15], [25]. The neural networks supporting familiarity and novelty in recognition memory have also been examined by means of a linear contrast of confidence: An increase in confidence for familiar OLD items is accompanied by activations in the lateral left inferior-frontal gyrus, a superior parietal region and the precuneus. High confidence NEW items, in contrast, elicit activations in the inferior temporal lobe, the anterior cingulate cortex, a left hippocampal region and the cerebellum [21]. Moreover, remembering familiar items has often been associated with brain activation in the anterior parahippoampal gyrus (= perirhinal cortex) [3].
Other recognition memory studies report that high-confidence decisions are associated with activations in the targeted recollection regions when compared to low-confidence judgments, i.e. MFG, PCG, and bilateral temporal lobes including the bilateral hippocampus [26], [27], [28]. Chua et al. [27] point out that little is known about the neural mechanisms supporting confidence judgments despite the fact that they are widely used in recognition memory tasks. They suggest that confidence judgments may involve both, the cognitive process of confidence assessment and the subjective feeling of confidence [27]. Thus, ‘subjective level of confidence’ subsumes a number of factors that may contribute to the differences between high- and low-confidence judgments, which could involve ease of retrieval, specific attributes of the test item, or heuristics about a subject's own memory ([19], cf. [27]). It is evident, that there exists conceptual overlap between recollection, memory strength and a high ‘subjective level of confidence’ and they appear to rely on similar networks. We therefore hypothesized that the ‘subjective level of confidence’ accounts for some of the above findings associated with recollection [21], [22].
In particular, the shape of the ‘oldness scales’ (e.g., [21], [22]) might provide additional evidence to differentiate between confidence-based activations and memory-related processes (recollection or memory strength). Such oldness-scales depict increases and decreases of event-related neural activation as a function of the perceived oldness ranging from ‘sure new’ to ‘sure old’ responses. As explored in [29], some of the oldness-scales follow a U-shaped function. For example, the inferior parietal cortex reveals strong activations for high-confidence OLD and high-confidence NEW decisions and weak activations for low-confidence ratings. Such U-shaped functions (Fig. 1) are difficult to explain with an assumption of pure memory-related processing, because NEW items have not been studied and are therefore less likely to activate memory-related brain regions. The more likely account for U-shaped oldness scales is the view of confidence being a part of the processes supporting memory decisions (so-called metamemory processes [30], [31] and/or attentional processing [29]). Note, however that MFG, PCG, and STG oldness scales are U-shaped in [21] (also [22]). Activations in the hippocampal region, in contrast, follow an L-shaped function [21], [22] that only increases for high-confidence OLD decisions (see Fig. 1). We expect such a L-shaped neural response pattern particularly for regions that specifically process memory strength, as NEW items are not expected to elicit such a signal.
Thus, the present study aimed to combine a manipulation of between-item-associations with a confidence-based approach in a recognition memory paradigm to examine the oldness-scales in the proposed target regions (bilateral hippocampi, PCG, MFG, STG [21]) of the memory network discussed to support recollection/memory strength. We expected to replicate the pattern of confidence-based U-shaped oldness scales in regions of the memory network outside the MTL, and L-shaped oldness scales in hippocampal and parahippocampal regions. Such a pattern would be in line with the notion of a support memory network dedicated to metamemory and attentional processing distinct from the MTL core memory network. In addition, we applied a novel manipulation of associative connections within the stimulus set. To measure the associative structure in the stimulus set, we used co-occurrence-statistics. This rationale is based on Hebbian learning [32]: Items that occur often together are likely to be associated. Thus, the manipulation of between-item-associations differentiates between words with a high (HIGH-A) or low (LOW-A) amount of associated items in the stimulus set. The manipulation of between-item-associations is thought to explicitly enhance episodic memory processes [12]. Using a similar manipulation, Hofmann et al. [12] were able to increase participants’ memory performance in a behavioral and a simulation study. HIGH-A items elicited a higher proportion of ‘OLD’ responses to OLD items and NEW items. Thus, the manipulation of the between-item-associations can inform about the underlying memory functions. The availability of contextual associations during encoding, for example, has been shown to correlate with activations in the hippocampus, and superior and inferior parietal regions [15], i.e. the main target regions of the recognition memory network.
Brain regions where U-shaped oldness scales are not affected by this manipulation are likely to support confidence, not memory itself. In contrast, when a high amount of associations drives the activation of OLD items only, this supports the conclusion of actual memory processes. As HIGH-A items have increased memory strength [12], we expect a higher amount of associations to specifically engage hippocampal activation. This structure is known to bind contextual associations to the item information [33]. As memory strength is indicated by subjective confidence and affected by contextual associations, an interaction between these factors in one brain region and the observation of L-shaped oldness scales would indicate a high memory strength signal. Thus, we predict greater activations for HIGH-A items in the hippocampus, with an L-shaped scale for HIGH-A and a lower or flattend L-shaped scale for LOW-A items.
Section snippets
Stimuli
Stimuli were German nouns with a length of four to eight letters from the Berlin Affective Word List Reloaded (BAWL-R [34]), which includes ratings of emotional valence, arousal, and imageability, while the number of orthographic neighbors was derived from the CELEX Database [35]. Frequency classes were taken from the German Corpus of the Leipzig Wortschatz Projekt (http://corpora.informatik.uni-leipzig.de/) [36]. Token bigram frequencies were calculated with the SUBLEX software [37], using the
Behavioral data
The behavioral results are displayed in Table 2. The three-way ANOVA (comprising the factors associations, confidence, and old–new) of correctly classified answers revealed no significant main effect (all p's > 0.470), but significant two-way interactions between associations and old-new (F(1,19) = 10.919, p = 0.004), as well as between confidence and old-new (F(1,19) = 39.871, p < 0.001). The associations × confidence interaction (p = 0.196) and the three-way interaction (p = 0.914) are non-significant. The
Discussion
In the present verbal recognition memory study we manipulated the amount of reverberant associative feedback from other items in the experimental context to the presented items. In addition, we examined confidence-based neural responses to test the proposal of a memory-processing network [1], [21], a contrast that showed the highest activations in a wide-spread brain network including all target regions of the recognition memory network. Of particular interest were hemodynamic responses in a
Conclusion
The present results support our initial hypotheses that regions of a proposed recollection network outside the MTL most likely process metamemory functions. Contrasting high- and low-confidence decisions in a recognition memory paradigm revealed a high overlap between the OLD/NEW contrast and activations related to high-confidence decisions. An examination of the activation functions based on the subjective level of confidence of the participants showed U-shaped oldness-scales in MFG, PCG and
Acknowledgements
This work was supported by the Deutsche Forschungsgemeinschaft (research unit “Conflicts as signals in cognitive systems,” TP 4 Jacobs, JA 823/4-2).
References (58)
The nature of recollection and familiarity: a review of 30 years of research
Journal of Memory and Language
(2002)- et al.
The medial temporal lobe and the attributes of memory
Trends in Cognitive Sciences
(2011) - et al.
Human recognition memory: a cognitive neuroscience perspective
Trends in Cognitive Sciences
(2003) - et al.
Recollection and memory strength: confusion about confounds
Trends in Cognitive Sciences
(2011) - et al.
Familiarity recollection and medial temporal lobe function: an unresolved issue
Trends in Cognitive Sciences
(2011) - et al.
Confusion about confounds: response to Diana and Ranganath
Trends in Cognitive Sciences
(2011) - et al.
The familiarity/recollection distinction does not illuminate medial temporal lobe function: response to Montaldi and Mayes
Trends in Cognitive Sciences
(2011) - et al.
The role of the human hippocampus in familiarity-based and recollection-based recognition memory
Behavioural Brain Research
(2010) - et al.
Understanding metamemory: neural correlates of the cognitive process and subjective level of confidence in recognition memory
Neuroimage
(2006) - et al.
Neural correlates of memory confidence
Neuroimage
(2006)
Metamemory: a theoretical framework and new findings
The assessment and analysis of handedness: the Edinburgh inventory
Neuropsychologia
Distinct prefrontal cortex activity associated with item memory and source memory for visual shapes
Brain Research. Cognitive Brain Research
Item, context and relational episodic encoding in humans
Current Opinion in Neurobiology
Putting names to faces: successful encoding of associative memories activates the anterior hippocampal formation
Neuroimage
Medial prefrontal cortex plays a critical and selective role in ‘feeling of knowing’ meta-memory judgments
Neuropsychologia
Some surprising findings on the involvement of the parietal lobe in human memory
Neurobiology of Learning and Memory
Focal retrograde amnesia following closed head injury: a case study and theoretical account
Neuropsychologia
Does the cerebellum contribute to cognitive aspects of speech production? A functional magnetic resonance imaging (fMRI) study in humans
Neuroscience Letters
Functional topography in the human cerebellum: a meta-analysis of neuroimaging studies
Neuroimage
Recognition memory and the medial temporal lobe: a new perspective
Nature Reviews Neuroscience
Recognition memory: what are the roles of the perirhinal cortex and hippocampus
Nature Reviews Neuroscience
Receiver-operating characteristics in recognition memory: evidence for a dual-process model
Journal of Experimental Psychology Learning, Memory, and Cognition
In search of recollection and familiarity signals in the hippocampus
Journal of Cognitive Neuroscience
Remembering words in context as predicted by an Associative Read-Out Model
Frontiers in Psychology
Relational information and the context effect in recognition memory
Memory & Cognition
Recognition memory ROCs for item and associative information: the contribution of recollection and familiarity
Memory & Cognition
Associations evoked during memory encoding recruit the context-network
Hippocampus
The medial temporal lobe and recognition memory
Annual Review of Neuroscience
Cited by (21)
Imaging recollection, familiarity, and novelty in the frontoparietal control and default mode networks and the anterior-posterior medial temporal lobe: An integrated view and meta-analysis
2021, Neuroscience and Biobehavioral ReviewsCitation Excerpt :In apparent support for this prediction, these studies reported that a part of the hippocampus is associated with an L-shaped function. On the other hand, the actual curves shown in the figures of these studies were U-shaped rather than L-shaped (see Fig. 1 of Daselaar et al., 2006; Fig. 2 of Johnson et al., 2013; Fig. 3 of Kuchinke et al., 2013), indicating that the data can be fitted more robustly by a U-shaped function than L-shaped function. Thus, despite the statistical significance, the data do not convincingly support the hypothesis that the relationship is L-shaped.
Transcranial direct current stimulation over the parietal cortex alters bias in item and source memory tasks
2016, Brain and CognitionCitation Excerpt :Correspondingly, evidence from fMRI and event-related potentials (ERPs) studies converged to implicate the PPC in memory processes that were supported by attentional functions. For example, because “new” items are unstudied, they should not elicit retrieval-related activity but may elicit attentional processing, and studies have shown greater activation in the PPC for falsely recognized new (unstudied) items compared to correctly rejected new items (Kahn et al., 2004; Wheeler & Buckner, 2004), high confidence compared to low confidence false recognition of strongly related lures (Kim & Cabeza, 2007), high confidence correct compared to low confidence correct old and new judgments (Kuchinke, Fritzemeier, Hofmann, & Jacobs, 2013), and invalidly cued compared to validly cued correct old and new judgments (Jaeger, Konkel, & Dobbins, 2013; O’Connor, Han, & Dobbins, 2010). Also consistent with the idea that the parietal cortex is involved in attentional selection of mnemonic information, ERPs over parietal areas were increased for specific recollections such as when a probe was self-generated compared to imagined (Leynes, 2012), was endorsed as accompanied with greater details (Vilberg & Rugg, 2009), and was identical to what was studied compared to when changed (Ally, Simons, McKeever, Peers, & Budson, 2008).
Expert-novice differences in brain function of field hockey players
2016, Neuroscience