**1. Introduction**

Consider the following scenario: You forgot the password for your online banking account. Fortunately, you left yourself a password hint: "childhood telephone number". You used the number for many years. Yet, even though it felt so familiar, you could recall it only after experiencing a vivid image of yourself dialing it on your old rotary dial phone. This vignette simplifies the vivid incidental recall (VIR) e ffect. This e ffect has been replicated in experiments where participants are tasked with generating the visual mental images corresponding to a list of cuing object-nouns, and, after an intervening task or rest delay, they are surprised with the request to recall the words. The cues which correspond to relatively higher subjectively vivid images generally yield a higher probability of being successfully retrieved during unexpected recall [1–4].

Yet, the neurocognitive processes or variables underpinning e ffects, such as VIR, have not been addressed directly in terms of underlying memory mechanisms. A possible link can be made with dual

process theories of recognition memory, according to which retrieval can take two forms: a recollection of contextual and qualitative details, or a subjective feeling of familiarity about the stimulus in question [5,6]. For stimuli to be recalled during an incidental recall task, the memory of the stimuli must be retrieved with its associated qualitative details, and therefore, must rely on the recollection process of recognition memory. On this account, VIR could occur because more vivid mental imagery strengthens the process of recollection, resulting in higher scores in incidental recall tasks. However, these theories also imply that another key factor should be involved in the vivid incidental recall e ffect, namely, previous and repeated exposure to the given stimulus, in this case, the familiarity associated with word cuing mental imagery and recall [7]. Still, it is unclear whether a level of familiarity strength similar (or congruent) to that of the vividness in the underlying representation may increase or reduce the success of its recollection, thereby modulating the extent of VIR e ffects. In other words: Are two di fferent types of relatively weak encodings better than a single relatively strong one for consolidating the same memory? To rephrase in terms of our opening anecdotal vignette: Would we still be able to recall our password hint successfully if we had a vivid image of dialing it, but the number itself lost its familiarity? Would this incongruence between the level of familiarity and vividness hinder recollection? This question is the focus of this paper.

According to associative theories, such as Paivio's dual coding theory, two memory representations should have cooperative e ffects on the probability of accurate incidental retrieval [8]. That is, the relative strengths of two memory representations should be additive and should always yield a higher probability of reinstating recollection of the original stimulus during incidental recall. Accordingly, having only one strong encoding should be equivalent to having two di fferent weak ones, and two strong encodings will always be better than two weak ones. Hence, the success of incidental retrieval should increase monotonically with the strength of the memory representation.

Most recently, however, accounts based on the neurobiology of learning and memory challenged associative explanations. In particular, the non-monotonic plasticity hypothesis (NMPH) holds that learning e ffects (recall accuracy) vary as a non-monotonic function of the amount of excitation associated with competing memory representations [9]. Specifically, NMPH predicts that strong learning e ffects result from overwhelmingly high levels of excitation in one of the two competing representations; poor learning e ffects result from moderate excitation in both representations, while no learning e ffects result from low levels of excitation overall [10]. That is, two strong encodings or a strong single one will always better than two weak ones, but two moderately intense encodings will result in the worst learning outcome. As a result, the neural strength of memory representations should be non-monotonically related to the probability of accurate incidental retrieval.

While the NMPH proposes a theory of memory formation and learning, it is first and foremost a neurobiological theory that is rooted at processes taking place at the level of the neuron. It is still unclear whether theories of synaptic strengthening, such as the non-monotonic plasticity hypothesis, are useful frameworks to utilize when modeling higher order phenomenon, such as memory consolidation and retrieval. Recent work using large functional brain networks have identified clear functional and spatial di fferential specialization for both memory consolidation and memory retrieval processes. More importantly, clear di fferences in neural specializations are observed when assessing the influence of the accuracy of memory retrieval (performance) as compared to the cortical activation observed at the time of encoding and retrieval [11,12].

A related, known challenge for theories which attempt to reduce the content of recollections to relatively simple neural mechanisms is that they do not adequately describe the relationships between phenomenal consciousness, for example, in the present context, what seems phenomenally available in imagery (i.e., vivid), and what is ultimately accessed in the underlying memory representations (see [13] for the general argument). This is indicative that building robust and replicable models of memory processes at the level of large functional brain networks can provide insightful analysis that may be able to bridge the gap between the phenomenal, the behavioral, and the neural.

Over the last decade, large functional brain networks have been reliably identified across a variety of independent cognitive states using both functional magnetic resonance imaging (fMRI) and Electroencephalography (EEG). These large functional brain networks, coined resting-state networks (RSNs), have been shown to be accurate biomarkers for di fferent states of consciousness, and more importantly accumulating evidence suggests that abnormal RSN activation patterns underly numerous psychopathologies [14,15]. A growing consensus suggests that functional brain activity can be divided into two anti-correlated RSNs: namely, the default mode network (DMN) and the task-positive-network (TPN) [16]. The DMN is generally referred to as a baseline state or a mind wandering state that involves dynamic activity between multiple cortical and subcortical areas that principally involves the medial temporal subsystems (hippocampus, parahippocampal cortex, and retrosplenial cortex), and the dorsal medial subsystems (dmPFC, TPF, and the temporal pole) [17]. Functionally, the DMN is thought to be responsible for the regulation of emotional processing, self-referential mental activity, and the recollection of prior experiences [18]. The TPN is generally considered to reflect externally mediated cognition that involves activity in a number of subsystems, including the dorsal attention network (intraparietal sulcus, sections of the precentral and frontal sulcus, and middle frontal gyrus), the posterior visual network (retinotopic occipital cortex and the temporal-occipital region), the auditory–phonological network (bilateral superior temporal cortex), and the motor network (regions of the precentral, postcentral, and medial frontal gyri, the primary sensory-motor cortices, and the supplementary motor area) [19].

The literature on the role of RSNs in incidental recall paradigms is sparse. One study has investigated the role of RSNs in the incidental retrieval of episodic memories, concluding that relative deactivation of the DMN results in poor incidental recall scores at the time of retrieval [20]. These results confirm the function of the DMN, namely that suppressing it serves to reduce task-irrelevant processing during sensory intensive tasks. While brain activity during incidental recall appears to be TPN dominant (given the high attentional demand of such a task), little is known about how RSNs influence incidental recall at the time of encoding. According to Craik and Lockhart's levels of processing e ffect, accurate incidental recall of semantic memories is strongly influenced by modulatory variables at the time of encoding. For example, the familiarity of a semantic memory will influence the accuracy by which it can be successfully incidentally recalled [5]. Additionally, memories with greater mental imagery vividness also tend to be recalled at a higher percentage as imagery vividness is often tied to emotional salience [21]. Such modulatory variables are thought to influence the dominant RSN activation pattern at the time of memory encoding [22,23].

Despite the generalized anti-correlated nature of the DMN and the TPN, significant variation in task-dependent RSNs have been observed in both experimental conditions involving healthy controls [24,25] and in numerous psychological disorders [26–36]. Task-based RSN variation in healthy controls is thought to represent inherent genetic variability [37], with some individuals demonstrating dominant TPN activation, others demonstrating dominant DMN activation and some reporting mixed RSN activation for a given task.

One such task that elicits large individual variation in dominant RSN activation is visual mental imagery. Very recently, it has been shown that individuals capable of highly vivid imagery visualization capabilities overwhelmingly utilize the TPN, while individuals with low mean vividness ratings overwhelmingly utilize the DMN [38]. Similarly, in memory recognition paradigms the strength associated with stimulus familiarity demonstrates a dichotomous RSN activation pattern. Specifically, individuals with strong familiarity judgments at the time of stimulus presentation demonstrate a dominant TPN activation, and individuals with weak familiarity judgments demonstrate a dominant DMN activation [39]. Research on the relationship between the strength of neuropsychological variables (in this case familiarity and vividness), at the time of stimulus encoding, and their corresponding RSN activation is of critical importance to understand how learning occurs at the level of large-scale neural networks. Our study is among the first to investigate how the strength associated with traditional

modulatory psychological variables a ffects RSN activation and whether di fferential RSN activation has significant e ffects on memory consolidation.

In the present study, we report a secondary data analysis on a corpus of cuing-nouns, which examines how the naturally varying strength of word familiarity and imagery vividness influences performance in an incidental free-recall task. We entertain a novel theory as to how VIR can be explained, incorporating aspects of both dual-rote associative models and neurobiologically-inspired NMPH. We propose that VIR is best explained by a dualistic neuropsychological model mimicking the global activity in two large RSNs, the DMN and the TPN. Our proposed model of memory consolidation posits that it is not solely the strength of neuropsychological variables that influences future memory consolidation, it is the dominant pattern of RSNs at the time of encoding that accounts for most of the variability in memory consolidation scores. We hypothesize that dominant RSN activity plays a critical role in memory consolidation given the substantial findings in the clinical literature which demonstrate that abnormal RSN connectivity, particularly high frequencies of DMN and TPN activation, result in memory deficits [24–34]. As a result of these findings, our model hypothesizes that conditions in which stimulus encoding reflects competing activation of DMN and TPN will result in poor memory consolidation scores. That is, two encodings, even if weak, will result in better learning outcomes compared to a mix of one strong and one weak encoding.

The approach of testing our dualistic RSN neuropsychological model of memory consolation by relying on psychological evidence is supported by previous research which has established that some psychological variables of low strength have the ability to selectively engage the DMN, while other psychological variables of high strength have the ability to selectively engage the TPN. Overall, our study aimed to investigate how RSN variability at the time of encoding a ffects memory consolidation, as measured by scores on an incidental recall task. We expected to find that higher RSN variability/competition, represented by noun-cues associated with mixed scores on our neuropsychological variables (low vividness and high familiarity, or vice versa), would be associated with poor memory recall, whereas congruen<sup>t</sup> levels (high vividness and high familiarity; low vividness and low familiarity) would be associated with higher memory recall (Figure 1).

**Figure 1.** Schematic depicting the dualistic resting-state networks (RSN) neuropsychological model of memory consolidation. The model depicts its predicted memory consolidation score in relation to the three possible combinations of variable strength at the time of encoding. The width of the arrows indicates the strength of the variables at encoding.

The proposed analysis permits us to contrast our predictions using our dualistic RSN neuropsychological model with those derived from Paivio's dual coding theory and the NMPH, as they relate to VIR. Since as mentioned, NMPH proposes to explain (i.e., reduce) retrieval in terms of basic neurobiological mechanisms, such a framework, can be extended by incorporating current brain imaging research which indicates that higher mental imagery vividness ratings are associated with an overall increase in neural excitability in the relevant cortical areas [40]. Similarly, familiarity related cortical regions, such as the left angular gyrus, have been shown to increase in neural excitability when participants report that a memory is experienced as more familiar [41]. Thus, relating the above neural excitability patterns with the observed memory performance, if NMPH were to be correct, we would expect higher recall probabilities from noun-cues associated with high vividness and high familiarity ratings (high neural excitability). We would expect low recall probabilities from noun-cues associated with low vividness and high familiarity ratings or vice versa (moderate excitability), and we would expect average recall probabilities from noun-cues associated with low familiarity and low vividness ratings (low neural excitability). In contrast, if Pavio's dual coding theory were to be correct, we would expect that a higher probability of recall would manifest when both types of memory representations, imagery vividness and imagery familiarity, are strengthened. The recall probability should increase monotonically in relation to the strength of both imagery vividness and imagery familiarity (for comparison of model predictions see Figure A1).
