Introduction

Over the past 15 years, a new area of research has emerged that focuses on spontaneous thoughts, generally defined as mental contents that come to mind with reduced effort, intention and control (Christoff et al., 2016; Cole & Kvavilashvili, 2019). Recently, there has been an effort to systematize the growing evidence on spontaneous thought, in publications suggesting alternative theoretical frameworks (Ciaramelli & Treves, 2019; Cole & Kvavilashvili, 2021; Mildner & Tamir, 2019), reviews examining age-related differences in spontaneous thoughts (Jordão, Ferreira-Santos, et al., 2019; Maillet & Schacter, 2016) and in a handbook entirely dedicated to the area (Christoff & Fox, 2018). The involvement of the episodic and semantic memory systems has been an important part of this discussion, as authors attempt to explain how spontaneous thoughts occur (Berntsen & Hall, 2004; Mildner & Tamir, 2019; Mills et al., 2018) (for proposals less focused on the episodic and semantic memory systems, see also Barzykowski, Radel, et al., 2019b; Mazzoni, 2019; Vannucci et al., 2019). Here, we focus on how research on spontaneous thought about the past and/or futureFootnote 1 (and previous considerations about its reliance on the episodic and semantic memory systems) can contribute to the discussion of the episodic-semantic interaction, namely to the functional independence between systems.

Most of the previous research discussed here has focused on spontaneous thoughts about the past, and to a lesser extent, about the future. Following from the definition of spontaneous thought mentioned above, these are mental contents about the past and future that are experienced as less dependent on effort, intention, and control (Christoff et al., 2016; Cole & Kvavilashvili, 2019). This broad definition includes involuntary autobiographical memories and future thoughts, which are defined as past events and images of the future that come to mind without a previous attempt to recall or imagine (e.g., Berntsen et al., 2015). For that reason, research on involuntary autobiographical memory and future thoughts is considered an important part of this discussion, and henceforth referred to as spontaneous past and future thought.

The current review is organized around four main sections. First, we provide an introduction to the episodic-semantic interaction debate, defining the key concepts, summarizing recent theoretical proposals, and highlighting the question of the functional dependence (or independence) between episodic and semantic memory systems. Second, we show that in spontaneous as in deliberate thought, there is evidence for an interaction between episodic and semantic memory systems. To do this, we review research related to semantic priming and the descriptions of spontaneous thought. Despite the interaction between episodic and semantic systems, we conclude that the overall evidence indicates that direct access to the episodic memory system is at the core of spontaneous past and future thought occurrence. Third, we suggest that studying spontaneous past and future thought is a unique avenue to explore the episodic-semantic interaction. To analyze the functional independence of the episodic memory system in spontaneous retrieval, we review evidence on semantic dementia that suggests past and future episodic thoughts can be accessed spontaneously when the semantic memory system is compromised. In conclusion, we highlight the need for future research and detail ways in which studies on spontaneous thought could provide key information on how episodic and semantic elements intertwine.

Theoretical contextualization

In our daily life, we make use of our memories in different ways. For example, when looking at a cup, most of us can recall what the purpose of a cup is and how to use it. Sometimes, a particular cup may also prompt us to remember the circumstances in which that cup came to be in our possession – a gift from our mother on our first day in a new house of our own – and take us back to that moment in time. The first example highlights that memory reflects knowledge of the world, whereas the second example shows that memory is specific and gives us information about what, where, when, and who, and is likely accompanied by a sense of travelling back in time. These memories are instances of semantic and episodic memory, respectively, and were first captured by the groundbreaking work of Endel Tulving (Tulving, 1983). Semantic memory is defined as knowledge-of-the-world, without a specific time or sense of personal experience associated, whereas episodic memory is defined by a personal experience of mentally travelling in time to a particular situation in the past (Tulving, 2002), and supports mental time travel to the future (Tulving, 2005). Episodic memory is related, but distinct from autobiographical memory, which is defined as being about one’s own experiences, and will sometimes overlap with instances of episodic memory that involve personal experiences (e.g., Conway, 2005; Wheeler et al., 1997). However, autobiographical memory may also take the form of knowledge about oneself, without a sense of travelling in time to a particular experience, being closer to semantic than episodic memory (Wheeler et al., 1997).

Autobiographical memory is an important concept in the current review for several reasons. First, autobiographical memory research has been pivotal in the episodic-semantic interaction discussion, by showing that there are personal memories with both semantic and episodic characteristics (Renoult et al., 2012). Secondly, theories of autobiographical memory have suggested hierarchical structures that describe episodic and semantic memory interactions for personal information (e.g., Conway, 2005). Finally, spontaneous thoughts have been shown to be frequently about self-related content and thus closely linked to autobiographical memory (D’Argembeau, 2018). Throughout the review we develop these aspects, in which episodic, semantic, and autobiographical memory intersect.

The episodic-semantic distinction was first proposed as a heuristic that led to important new avenues of research, but Tulving soon came to believe that these represented, in fact, functionally and structurally distinct systems (for a historical review, see Renoult & Rugg, 2020), distinguishable based on what they do and how they do it (Tulving, 2002), and with particular neural substrates and behavioral or cognitive correlates (Tulving, 1985). The distinction between episodic and semantic memory was initially met with resistance, but found support in evidence from studies of amnesic patients and brain activation in healthy individuals (for a review, see Tulving, 2002). Currently, a renewed interest is emerging in the questions about the episodic-semantic distinction and, more importantly, the interaction between the episodic and semantic memory systems (Irish & Vatansever, 2020; Renoult et al., 2012; Renoult et al., 2016; Renoult et al., 2019) – to what extent do these systems interact? What type of interaction exists? Does the interaction blur the distinction between systems? Is there a continuum between them? “A continuum of what?” (Tulving, 1983, p. 67). In the next section, we explore these questions, focusing first on defining the episodic-semantic distinction, in Tulving’s terms, and his understanding of the interaction between these systems. Then, we highlight new evidence that has emerged about the interaction and new theoretical perspectives developed to accommodate it.

The episodic-semantic distinction: The question of functional independence and a new focus on the interaction

What does it mean to say that the episodic and semantic memory systems are independent? First, Tulving clearly notes what it does not mean – that the two memory systems are cut off from one another. Since the first formulations of Tulving’s theoretical proposal the interaction between systems has been acknowledged: the “[episodic memory] operations require, but go beyond, the semantic memory system” (Tulving, 2002, p. 5). Tulving distinguished the interaction of the two systems at encoding compared to retrieval. At encoding, when we are learning new information, the semantic memory system is necessary for events to then be encoded in the episodic memory system. In simpler terms – if we cannot learn the concept of a train (semantic memory system), how can we learn the experience of travelling in one for the first time (episodic memory system)? At retrieval, however, the relationship between the two systems can be different. When information already learned comes to mind, this can involve both semantic and episodic memory systems, or only one of them (Tulving & Markowitsch, 1998), that is, the systems can function independently. This clearly does not mean that the systems are mutually exclusive and will not interact, but that they do not necessarily have to. Additionally, this means that although one system may function worse when the other is compromised, they should not be affected to the same degree (Tulving, 1983). This prediction was analyzed in case studies of patients who endured damage in the neural structures that support episodic memory (e.g., hippocampal damage), in particular the case of patient KC (Rosenbaum et al., 2005). Of note, KC showed preserved general and personal knowledge acquired before neural damage, but was unable to recall and re-experience personal experiences, in line with a selective damage in the episodic memory system and the functional independence of systems. Functional independence was further supported by evidence from neural activation that points to the brain lateralization of functions associated predominantly with semantic versus episodic memory (Habib et al., 2003, but see more recent evidence below), and other cases of developmental (vs. acquired) amnesia (Baddeley et al., 2001). Recently, alternative findings and ideas have renewed the interest in these systems and focused on the episodic-semantic interaction.

Renoult et al. (2019) recently reviewed the literature on the episodic-semantic interaction and concluded that the systems are "inextricably intertwined" (p. 1052). In particular, they discussed evidence showing (1) the mix of semantic and episodic elements in personal narratives (personal semantics, Renoult et al., 2012), (2) the neural activation overlap shown by functional neuroimaging (e.g., Burianova et al., 2010; Kim, 2016), and (3) the crucial supporting role of semantic memory in episodic future thinking (Irish & Piguet, 2013). It is possible that the two systems work so closely and dynamically that they cannot be distinguished. Alternatively, it is possible that the current evidence reflects episodic and semantic memory systems being independently engaged during the same task.

One particular difficulty in understanding the episodic-semantic interaction relates to the process of selecting a particular memory. Most of the previous research relies on memory tasks in which participants are asked to remember specific episodes and will frequently rely on their deliberate attempt to remember (but, see Ball, 2007; Barzykowski, Niedźwieńska, & Mazzoni, 2019a). Theory and evidence (e.g., Conway, 2005; Conway & Pleydell-Pearce, 2000; Prebble et al., 2013) suggest that retrieval of a particular episode relies on progressively moving through a knowledge hierarchy that starts by accessing more semanticized forms of autobiographical memory, including lifetime periods and general events, before accessing specific episodes that constitute episodic memory. A similar hierarchical process has been suggested to be the basis of episodic future thinking (Conway et al., 2019; Irish, 2016). This contrasts with past and future thoughts that come to mind without previous intent (e.g., Berntsen & Jacobsen, 2008). In the following sections, we argue that considering episodic-semantic memory in past and future spontaneous thoughts can provide a unique way to understand the functional interdependence of these systems, when compared to the large majority of research in this area focused on deliberate retrieval.

Spontaneous thought and the episodic-semantic interaction

Spontaneous past thought was recognized since the earliest experimental research on memory, and was described by Ebbinghaus in 1885 as coming to mind “with apparent spontaneity and without any act of the will” (Ebbinghaus, 2013, p. 1). Only much later, however, the first experimental attempts to understand this phenomenon were developed, based on concepts from involuntary autobiographical memory (for a review, see Berntsen, 2009) and daydreaming research (Singer, 1975); and the discussion about how to better understand spontaneous thought is still ongoing (Christoff et al., 2018; Seli, Kane, et al., 2018; Seli, Kane, Metzinger, et al., 2018). Here, we define spontaneous thought based on the reduced effort and intention with which it comes to mind (in line with Cole & Kvavilashvili, 2019). The less effortful nature of spontaneous past and future thought is supported by evidence demonstrating that (1) spontaneous past and future thoughts come to mind faster than voluntary past and future thoughts (Cole et al., 2016; Schlagman & Kvavilashvili, 2008), (2) children between 3 and 4 years old report spontaneous past and future thoughts, despite strategic cognitive ability not being developed at that age (Caza & Atance, 2019; Hjuler et al., 2021; Krøjgaard et al., 2014; Krøjgaard et al., 2017; Sonne et al., 2019, 2020), (3) older adults with dementia report detailed spontaneous past thoughts, despite significant executive function decline (Miles et al., 2013), and (4) neural correlates of spontaneous past thoughts show a reduced involvement of areas associated with strategic processing, compared to voluntary memory (N. M. Hall et al., 2008; S. A. Hall et al., 2014).

In the absence of an effortful strategic search, several mechanisms have been suggested to explain how spontaneous past and future thoughts occur, including processes of cue-item discriminability (Berntsen, 2009; Staugaard & Berntsen, 2014) and, more recently, semantic to autobiographical memory priming (Mace et al., 2019). These mechanisms emphasize the role of the episodic and semantic memory systems, respectively, and are thus particularly important for the discussion of the episodic-semantic interaction.

Direct access to the episodic memory system: The cue-item discriminability mechanism

Cue-item discriminability refers to the ability of a cue to isolate a specific memory (Rubin, 1995), and is particularly important to explain the emergence of spontaneous past and future thoughts (Berntsen, 2009). This mechanism was first suggested to explain spontaneous past thoughts and builds upon the encoding specificity principle, which states that an episodic memory is more likely to be retrieved in the presence of a retrieval cue with which it overlaps (Moscovitch, 1995; Tulving & Thomson, 1973). This overlap provides direct access to specific events or the episodic memory system (Conway, 2005; Conway & Pleydell-Pearce, 2000). However, the ability of a cue to isolate a specific memory does not depend only on an overlap between the cue and the memory, but also on the unique relationship between them. This has been suggested by the concept of cue overload (Watkins & Watkins, 1975), which refers to a decrease in the probability of a cue to elicit a memory as the number of different memories associated with the cue increases (for a similar demonstration, named as the fan effect, see Anderson, 1974). For example, London bridge may be uniquely associated with the memory of walking around London during a single visit to the city. However, if one lives in London and crosses the London bridge daily, it will overlap with numerous memories. In both situations, there is an overlap between the cue and the memory, but only in the first case is there a unique link between the two. The unique link is likely to isolate a specific memory, which supports the view that cue-item discriminability provides direct access to the episodic system.

The cue-item discriminability mechanism refers to the need for a unique and distinctive link between the cue and the thought that it elicits, and can explain the occurrence of spontaneous past thoughts in both naturalistic (e.g., Berntsen, 1996; Berntsen, 1998) and laboratory settings (e.g., Berntsen et al., 2013; Schlagman & Kvavilashvili, 2008). Supporting this idea, several studies have shown that spontaneous past and future thoughts are frequently associated with a specific cue identified by participants as triggering the thought (e.g., Berntsen & Hall, 2004; Cole et al., 2016; Plimpton et al., 2015; Schlagman et al., 2007; but see Warden et al., 2019), and the thoughts that are elicited are specific in nature (e.g., Berntsen & Hall, 2004; Berntsen & Jacobsen, 2008; Cole et al., 2016; Johannessen & Berntsen, 2010; Schlagman & Kvavilashvili, 2008), suggesting that cues that trigger spontaneous thoughts are highly discriminable. Additionally, the frequency of spontaneous past thoughts has also been successfully manipulated via cue-item discriminability (Berntsen et al., 2013; Staugaard & Berntsen, 2014). For example, Berntsen et al. (2013) manipulated cue-item discriminability by systematically pairing unique or repeated cues (sounds) and scenes (pictures). These stimuli were presented first in an encoding task, which was followed by a separate task in which sounds were presented on their own. According to the cue-item discriminability mechanism, the authors predicted that a unique cue coupled with a unique scene (e.g., the sound of a particular dog barking in an area of the town you visited once) would correspond to the highest level of discriminability, and more frequently result in a spontaneous past thought when the sound cue is presented on its own. Other pairings of unique and repeated cues and scenes would produce lower levels of discriminability and result in fewer spontaneous past thoughts. Consistent with their prediction, Berntsen et al. (2013) showed that spontaneous past thoughts were more frequent for unique cues. In sum, these findings provide strong evidence for cue-item discriminability as a key mechanism to explain the emergence of spontaneous past and future thoughts.

Cue-item discriminability might occur through pattern completion, whereby a part of a memory prompts the emergence of related information and of the memory as a whole, a mechanism that has recently been suggested to explain the emergence of spontaneous thoughts (Mills et al., 2018). Although this evidence indicates that spontaneous retrieval processes favor access to the episodic memory system, research on the impact of semantic priming raises new questions about this conclusion, as we review in the next section.

Semantic modulation: Semantic information in spontaneous thought, priming effects and semantic-to-autobiographical memory activation

Not all spontaneous past and future thoughts are triggered by an identifiable cue and refer to a specific event. For example, Faber and D’Mello (2018) analyzed free descriptions of thoughts coming to mind while reading a text or watching a video, and found that around 14% fell on a semantic memory category, which included general knowledge like facts, meaning and concepts (“I thought about a scene from both “The Mummy” and “The Return of the Mummy”, p. 5). In a recent study, looking at descriptions of spontaneous thoughts, we found a mix of semantic and episodic elements (Jordão et al., 2020). These results mirror evidence from deliberate retrieval showing that descriptions of autobiographical memories involve both semantic and episodic information (Levine et al., 2002). Moreover, we found that, based on inter-coder agreement, some spontaneous thoughts were better characterized as repeated or extended personal events. Thus, we also found evidence for personal semantics in past and future spontaneous thoughts. Personal semantics refer to memories that do not fit in episodic or semantic memory but include variable elements of both (Renoult et al., 2012). If we understand personal semantics as evidence for a continuum between semantic and episodic memory systems (Renoult et al., 2012; Renoult et al., 2016; Renoult et al., 2019), our results suggest that a similar continuum operates in spontaneous thoughts.

Another example of spontaneous thoughts that do not refer to specific events are spontaneous semantic memories, consisting of fragments of semantic knowledge, which have been suggested to rely on semantic priming, in particular, long-term priming (Kvavilashvili & Mandler, 2004). This possibility is in line with semantic priming research, which shows that priming effects can be long-lasting and persist in long-term memory (for a review, see Joordens & Becker, 1997). Several studies have found that priming can influence spontaneous past thoughts both inside (Barzykowski & Niedźwieńska, 2018; study 2, Mace & Petersen, 2020) and outside the lab (Mace, 2005), and can have durable effects that last several days. In these studies, priming was based on voluntarily recalling personal specific memories about a particular time period (e.g., high school). After this, spontaneous past thoughts were recorded and compared between participants who performed the priming or a control task. The results showed no changes in the overall frequency of spontaneous past thoughts (Barzykowski & Niedźwieńska, 2018; studies 2 and 3, Mace, 2005), or spontaneous task-unrelated thoughts (study 2, Mace & Petersen, 2020), but spontaneous past thoughts about the primed time period were more frequent. This effect is thought to rely on spreading of activation within the autobiographical memory system, which would include elements from both episodic and semantic memory systems (Mace, 2010). More recently, it has been shown that spontaneous past thoughts may be primed not only by other autobiographical memories, but also by semantic memory, in a process called semantic-to-autobiographical memory priming (Mace et al., 2019; Mace & Unlu, 2020). Mace et al. (2019) tested the impact of semantic memory priming on spontaneous past thoughts and found an increase in the number of spontaneous past thoughts that were related to the content of the priming. Relatedly, semantic personal knowledge can also prime deliberate retrieval of autobiographical memories (Sheldon et al., 2020).

Semantic-to-autobiographical priming has been demonstrated using a variety of priming tasks, suggesting the robustness and generalizability of this effect. For example, Mace et al. (2019) used a semantic priming task that consisted of presenting single words of different levels of familiarity and asking participants to subjectively rate the degree of familiarity. Some of the words were selected because they are frequently associated with personal experiences (e.g., summer), and thus were expected to be effective primes for autobiographical memories. To elicit and record spontaneous thoughts, they used a standard vigilance task paradigm in which participants were presented with groups of horizontal or vertical lines and asked to respond when the lines were presented vertically (based on Schlagman & Kvavilashvili, 2008). The presentation also included word cues that the participants were asked to ignore but which have been shown to trigger spontaneous thoughts. When spontaneous thoughts occurred, the participants stopped the task and wrote them down. Mace et al. (2019) found an increase in the number of spontaneous past thoughts in which content was related to the primed concepts, and no differences in the number of spontaneous thoughts overall. In a second study, the authors used a lexical decision task as semantic memory priming to minimize the possibility of autobiographical memory activation during the priming task, and replicated these findings. Mace and Unlu (2020) also replicated the priming effect with sentences and pictures, supporting the idea that the effect occurs in various contexts involving language and visual perception. Semantic priming effects are also evident when more abstract concepts are primed. For example, Vannucci et al. (2019) showed a priming effect using left- and right-pointing arrows as priming material. The authors proposed that left-pointing arrows should prime the concept of the past and lead to more spontaneous past thoughts, while right-pointing arrows should prime the concept of the future and lead to more spontaneous future thoughts. To test this idea, the authors embedded left or right-pointing arrows in the vigilance task previously described. They found a greater number of spontaneous past thoughts when participants were primed with left-pointing arrows, and more spontaneous future thoughts with right-pointing arrows. Similar to the studies previously described, the overall number of spontaneous thoughts remained unchanged by the priming procedure. Across all of these studies priming increased spontaneous thoughts related with the content of the primed concepts, but not the overall frequency of spontaneous past or task-unrelated thoughts.Footnote 2

In sum, the evidence reviewed here demonstrates that spontaneous thoughts are also influenced by the semantic memory system. In fact, Mace and Unlu (2020) suggested that semantic-to-autobiographical memory priming is a ubiquitous phenomenon, responsible for the majority of spontaneous past thoughts in daily life. The essential role of semantic memory has been similarly emphasized for spontaneous thoughts (Mildner & Tamir, 2019), where it has been suggested that the semantic memory system is a scaffold for spontaneous thought.

Episodic and semantic elements of spontaneous thought

It is clear from the above evidence that both episodic and semantic memory play a role in spontaneous thought. Namely, the cue-item discriminability mechanism activates episodic memory by eliciting specific spontaneous thoughts, while semantic memory activation changes the content of spontaneous thoughts. Cue-item discriminability seems to be particularly important for determining if a spontaneous past or future thought occurs. In addition, the frequency of spontaneous thought about the past is influenced by other mechanisms such as attention and intention (Barzykowski & Niedźwieńska, 2016; Barzykowski, Niedźwieńska, et al., 2019; Vannucci et al., 2014). The evidence reviewed here also indicates that spontaneous thoughts are frequently related to the primed materials, indicating that semantic priming influences the content of spontaneous thoughts but does not influence the occurrence of spontaneous thoughts.Footnote 3 As we review below, the role of semantic priming and cue-item discriminability in spontaneous thought is better understood as part of an interaction that involves both semantic and episodic memory systems.

We suggest that a parsimonious way of explaining the impact of semantic priming in spontaneous thought is based on how cue-item discriminability interacts with the availability of particular memories through spreading activation. Berntsen (2009, pp. 107–110) suggested that there are several factors that may influence how cues discriminate particular memories, such as motivation and how recent a memory is (i.e., recency). These factors would moderate the availability of certain memories and thus the likelihood these are recalled in relation to environmental cues. For example, if I am motivated to start exercising, seeing an advert about trainers may remind me of an event in which my friend told me about online physical exercise videos. This does not mean that those trainers were particularly related to the memory and discriminated it from all other “trainer-related” memories I have. However, the fact that I am motivated to start exercising activates memories about that goal, such as the event of talking to my friend, which in interaction with the cue, becomes the most distinctive memory. Thus, memories related to personal goals would be more activated and, because of that, more readily available (e.g., Johannessen & Berntsen, 2010). If a cue is related to several competing memories, the availability of the memory would interact with the cue to isolate that memory, rendering its spontaneous retrieval more likely. We found evidence for this mechanism in spontaneous thoughts by priming personal goals (Jordão, Pinho, et al., 2019). After priming, participants performed a vigilance task including common words and, when randomly probed, described their spontaneous thoughts. We found an increase in spontaneous future thought after priming, consistent with the close relationship between personal goals and future thoughts (D’Argembeau, 2016). Importantly, we also showed that the majority of spontaneous thoughts were cued by the words in the vigilance task. Again, this was not due to the content of the cues, which were not specifically about personal goals or the future, but rather the interaction of the priming activation with the cues.

It is possible that semantic priming plays a similar role to motivational priming by making priming-related contents more available and consequently more likely to be distinguished from other thoughts when a cue is encountered (for similar reasoning, see Barzykowski & Niedźwieńska, 2018). Indeed, when studying the effect of priming personal goals in spontaneous thought, we found this exact pattern: the thoughts were related to an identifiable cue in the majority of cases, and priming influenced content but not overall frequency (Jordão, Pinho, et al., 2019). In other studies involving semantic priming (Mace et al., 2019; Mace & Petersen, 2020), the number of primed spontaneous past thoughts for which participants identified a cue was not reported. However, the authors analyzed different types of cues (more or less related with the original priming stimuli), which suggests that a considerable amount of primed spontaneous thoughts were associated with a cue. This would be in line with an interaction between cue and priming in which the cue remains a key element in spontaneous thought occurrence. Moreover, semantic-to-autobiographical priming did not change the specific nature of spontaneous past thoughts – these were still mostly specific, irrespective of priming (with no exceptions in Mace et al., 2019; Mace & Unlu, 2020). Again, this suggests that semantic priming does not change the mechanism by which specific spontaneous past and future thought occurs.

An alternative hypothesis is that the presence of cues might limit the impact of semantic priming on the occurrence of spontaneous thought. In this case, if no cues were present, there might be an increase in the overall frequency of spontaneous thought after priming. This would indicate that semantic priming influences not only the content of spontaneous thoughts, but also prompts the emergence of thoughts that would not be experienced otherwise. However, current research does not support this idea. In particular, Stawarczyk et al. (2011) measured the effect of personal goal priming during a sustained attention task that included only numbers. They found that participants reported more thoughts about the future, relating to personal goal priming, but the overall number of thoughts did not change. This evidence suggests that priming impacts the content but not the overall frequency of spontaneous thought, irrespective of the presence and type of cues used in the elicitation task.

One prediction that follows from our proposal is that if semantic priming influences cue-item discriminability, as one of several factors which include recency or motivational biases, then semantic priming should have a greater effect on spontaneous thought when the cues are more general. When a cue is more general, it has the potential to overlap with a higher number of memories or thoughts such that semantic priming has a greater impact by increasing the availability of semantically-related thoughts and isolating them. When a cue is very specific, it will overlap with fewer thoughts, reducing the likelihood of semantic impact. In other words, if a cue is more strongly linked with a particular thought, the increased availability caused by semantic priming is less important. To our knowledge there is currently no study that has manipulated both priming and cues in this way. This could be accomplished by analyzing priming effects in paradigms that include cues with different levels of specificity, as in Uzer and Brown (2017), or that manipulate cue overlap directly, as in Berntsen et al. (2013). In Berntsen et al. (2013), a combination of features defines a very specific cue, in contrast with the general verbal cues used in the vigilance tasks in which the priming effects have been shown (Jordão, Pinho, et al., 2019; Schlagman & Kvavilashvili, 2008).

In conclusion, the present evidence on the episodic and semantic elements of spontaneous thought demonstrates an interplay between systems. Some spontaneous thoughts are semantic or include semantic elements. At the same time, current evidence underscores the key role of direct access to the episodic memory system in order for a spontaneous thought to occur. The semantic memory system, as in other types of activation, seems to impact the pool of thoughts that may be directly accessed but does not result in more spontaneous thoughts overall. Spontaneous retrieval seems thus particularly reliant on the episodic memory system and, although semantic modulation may be involved, this does not always appear to be required. These characteristics place spontaneous thought research in a unique position to explore the functional independence between episodic and semantic memory systems, as detailed in the next section.

Spontaneous thought as a new approach to study the functional independence between episodic and semantic memory systems

As reviewed in the previous section, behavioral studies indicate that the occurrence of spontaneous thoughts is associated with a quick and non-effortful access to specific events, which involves the episodic memory system. Neuropsychological evidence supports the idea that spontaneous thoughts are a privileged area to study the functional independence between episodic and semantic memory systems. For example, episodic spontaneous mind wandering is disrupted in patients with hippocampal damage in whom the commonly episodic nature of spontaneous thought is replaced by semantic content (McCormick et al., 2018). Similarly, hippocampal atrophy in Alzheimer’s disease has been associated with changes in the normative profile of mind wandering, which, as in spontaneous thought, involves a shift of attention from the environment to internal mental contents (Smallwood & Schooler, 2015). People living with Alzheimer’s disease experience more thoughts about stimuli that are in their environment when the thought comes to mind, indicating that the mental time-travel quality of specific spontaneous thought is not as accessibleFootnote 4 (O’Callaghan et al., 2019). Thus, these findings show the critical role of the episodic memory system in spontaneous thought.

With respect to the semantic memory system, we suggest semantics may modulate the content of spontaneous thoughts or be involved in the voluntary elaboration of a thought that initially comes to mind spontaneously. The semantic memory system should also be involved in understanding certain cues, when these are more general and conceptual. However, if a cue does not require extensive semantic processing (e.g., a personal specific cue, like a personal photograph), we would expect that the semantic memory system would not be necessary for a spontaneous thought to occur. It is in these cases that spontaneous thought can be a unique area in which to test the functional independence between episodic and semantic memory systems. If the episodic memory system is functionally independent, we should find relatively intact spontaneous thoughts when the semantic memory system is compromised such as in semantic dementia. Although no studies have directly analyzed spontaneous thought in semantic dementia, indirect evidence suggests that spontaneous thoughts are preserved, and we review this evidence below.

Voluntary retrieval of specific autobiographical episodes when very specific cues are provided to prompt recall and access the episodic memory system directly, might parallel cue-item discriminability mechanisms during spontaneous retrieval to directly access the episodic memory system. Indeed, personal cues promote fast and direct retrieval in healthy participants as we would expect based on cue-item discriminability (Uzer & Brown, 2017). In semantic dementia, several studies have shown preserved autobiographical memory retrieval for remote events, when very specific cues such as personal photographs (Westmacott et al., 2001), or other personal detailed information provided by a close person (Moss et al., 2003) are used to prompt recall. Based on these results, we expect that spontaneous thought about the past will not be impaired in semantic dementia, which would be in line with the functional independence of the episodic memory system.

Another area of research that provides indirect evidence is prospective memory (PM). Memory for future intentions has been shown to partially rely on spontaneous cognitive processes. More specifically, while encoding a future action one intends to perform is certainly intentional, thoughts about these intentions often come to mind spontaneously in relation to a particular event (event-based PM, e.g., Scullin et al., 2010) or a particular time (time-based PM, e.g., Kvavilashvili & Fisher, 2007). For these reasons, prospective memory has been identified as a key research area to understand spontaneous future thought (Cole & Kvavilashvili, 2019; Kvavilashvili & Rummel, 2020). In semantic dementia, time-based but not event-based prospective memory is preserved (Kamminga et al., 2014). Kamminga et al. (2014) suggested that one reason for this differential pattern of PM in semantic dementia is that event-based PM relies more on semantic processing (e.g., “When the alarm rings, please put this notebook on the floor,” requires an understanding of what an alarm, a notebook and the floor are), whereas in time-based PM the requirement for semantic processing is reduced (the trigger for the action is defined by the sense of time, “in 10 minutes”). This interpretation is supported by a strong relationship between semantic naming and event but not time-based PM (Kamminga et al., 2014).Footnote 5 It is possible that event-based PM tasks using very specific personal cues may bypass these semantic deficits and prompt recall, as in voluntary autobiographical memories. However, this possibility has not yet been explored in semantic dementia research.

The fact that performance in time-based PM seems to be relatively spared suggests that at least some forms of spontaneous future thought are preserved when the semantic memory system deteriorates, again indicating a functional independence of the episodic memory system. Anecdotal descriptions from semantic dementia case studies also supports this empirical evidence, with reports that some participants were able to “keep track of appointments” (Hodges et al., 1992). It should be noted, however, that the possible strategies used by the participants (e.g., checking a clock) were not analyzed in Kamminga et al. (2014), and it is thus uncertain to what extent the preservation of time-based PM may be due to deliberate strategies used by the participants, in addition to spontaneous retrieval. Nevertheless, these findings raise the possibility that spontaneous future thoughts that consist of relatively simple instructions (buying a present for my mum at the market this afternoon) versus more elaborate future scenarios (imagining the family singing Happy Birthday to my mum in her party this evening) may be accessible even if the semantic memory system is compromised. It is less likely that spontaneous future thoughts that involve more complex events will be preserved. In fact, in deliberate retrieval, the semantic memory system has been found to be an essential scaffold to episodic future thinking, as shown by its impairment in semantic dementia (Irish, 2016). According to the idea that spontaneous future thoughts are often based on memories of imagined future episodes that have been previously contructed in a deliberate manner (Cole & Kvavilashvili, 2021; D'Argembeau & Mathy, 2011; Jordão et al., 2020), we would expect similar deficits in spontaneous retrieval in semantic dementia.

Although we found no studies directly analyzing the impact of semantic dementia on spontaneous thought, there are no reasons to believe that such analysis is methodologically impossible. At the moment there are several lab paradigms that have been successfully used to elicit spontaneous thoughts. Some of these paradigms require little semantic processing and should be adaptable to semantic dementia (Berntsen et al., 2013; Laflamme et al., 2018; O’Callaghan et al., 2015). It would also be possible to adapt more naturalistic paradigms that have been used in other contexts. For example, St. Jacques and Schacter (2013) used a museum tour to create a standardized naturalistic experience and collected specific visual cues of this experience using a wearable camera. Later, these cues were used to test deliberate retrieval of episodic memory against similar photographs. It would be possibile to integrate these photographs in a vigilance task in which participants are not asked to voluntarily recall any memories, but are asked to identify occasions in which they come to mind spontaneously. This would allow measurement of the preservation or deterioration of spontaneous thought in semantic dementia compared to controls, and provide key evidence for the present discussion about the functional independence of the episodic memory system.

The potential of spontaneous thought research to inform the episodic-semantic interaction debate is not restricted to studying semantic dementia patients. Another important area of research with healthy participants has to do with the mix of episodic and semantic information. Based on facilitated access to the episodic memory system under spontaneous retrieval, we would expect differences in this episodic-semantic mix, compared to deliberate retrieval. For example, older adults frequently show a distinct pattern of episodic and semantic elements in autobiographical retrieval that includes an increase of semantic information compared to younger adults (Levine et al., 2002). However, when we looked at episodic specificity between age groups in spontaneous retrieval, we found no evidence for differences in specificity (Jordão et al., 2020). Thus, the age differences in the episodic-semantic interaction seem to be related to the deliberate retrieval process itself – both younger and older adults deliberately search for more episodic detail, and younger adults are more successful in accessing it, while older adults access more general (semantic) information. This highlights that at least some of the semantic information included in memory descriptions does not require reinstatement of this information, which would indicate a decreased functional independence between episodic and semantic memory systems (Renoult et al., 2019). Rather, it is attributable to the deliberate process of recalling, which engages additional strategies to access detail, and may be more or less successful.

In the next section we develop suggestions for two areas we think are particularly important to advance the knowledge in this field: the analysis of spontaneous thought content and its evolution over time, from the moment when it first comes to mind.

Future studies

Exploring spontaneous thought content: Open-ended methods and the need for adapted coding systems

As exemplified above, the evidence about semantic and episodic information in spontaneous thought can provide useful contributions to understand the episodic-semantic interaction. However, there is still scarce evidence about how and how much do semantic and episodic elements intertwine in spontaneous thought descriptions. Methodological factors seem to be partially responsible for this. First, it is common to measure the level of specificity of spontaneous thoughts by asking participants to classify them as specific versus general events, or as a specific, repeated or extended event (e.g., Barzykowski & Niedźwieńska, 2018; Cole et al., 2016; Kvavilashvili & Schlagman, 2011; Plimpton et al., 2015; Schlagman et al., 2009; Schlagman & Kvavilashvili, 2008). A minority of studies used participants’ descriptions to determine the specificity of spontaneous thoughts based on categories, such as dichotomously distinguishing episodic versus semantic thoughts (McCormick et al., 2018; Song & Wang, 2012), autobiographical versus semantic memories (Faber & D’Mello, 2018), or a more diverse number of categories from general knowledge to specific events with detail (El Haj et al., 2012; Jordão et al., 2020). In contrast, in deliberate retrieval, much of the research on episodic and semantic information relies on a detailed analysis of participants’ descriptions, based on the autobiographical interview (AI, Levine et al., 2002). The AI provides a reliable method to analyze episodic and semantic elements within one description. In this analysis, an event that is specific can still include a variable proportion of episodic and semantic elements. This level of analysis seems particularly important in spontaneous thought to identify variations of episodic specificity, given that a high number of specific events is frequently reported. The AI and related adaptations (Addis et al., 2008) have contributed, for example, to the understanding of prospection (Miloyan et al., 2019). Some of the disadvantages of the AI, namely the fact that it is time and labor consuming, have now started to be addressed by semi-automatized transcription and coding (Wardell et al., 2020).

We suggest that one of the key developments in understanding spontaneous thought and the episodic-semantic interaction will be in adapting a detailed classification system of episodic and semantic details, such as the AI, to spontaneous thoughts. Firstly, this will depend on using open-ended methods that allow the collection of descriptions of spontaneous thoughts as experienced by the participants. A considerable amount of the research on spontaneous thoughts has used probes with forced-choice response options, which limits the opportunity to analyze episodic and semantic elements (as well as introducing bias; Jordão, Ferreira-Santos, et al., 2019; Weinstein, 2017). Recently, Jordano and Touron (2018) reviewed findings from think-aloud protocols in the context of spontaneous thought monitoring. We suggest that think-aloud protocols could also be an important tool to understand how episodic and semantic elements interact in spontaneous thought. Secondly, the development of a detailed classification system of episodic and semantic details will have to be adapted to include implicit information, inferred from what was described. We found the presence of implicit information to be quite frequent in spontaneous thought descriptions (Jordão et al., 2020), which is reinforced by the fact that inferred information is mentioned in other classifications of spontaneous thoughts (Faber & D’Mello, 2018, p. 12). Establishing a standardized way to deal with implicit information, and testing its reliability, will be fundamental to the advancement of the field.

Spontaneous thought evolving over time: Neural correlates with high time-precision methods

The growing literature on the neural mechanisms of spontaneous thoughts in healthy individuals offers a useful way to understand this episodic-semantic interaction in spontaneous thought (for a recent review, see Andrews-Hanna et al., 2018). To investigate the episodic-semantic interaction in spontaneous thoughts, we suggest that high time- precision methods, such as electroencephalography (EEG) and the analysis of event-related potentials (ERPs) will be particularly important, in two specific ways. Firstly, to distinguish brain activity associated with the emergence versus maintenance of a spontaneous thought. In fact, a very important issue in spontaneous thought has to do with distinguishing the processes associated with its emergence and its maintenance over time (Smallwood, 2013). In the present review, we hypothesized that the semantic memory system modulates the content of spontaneous thought but not its occurrence. This idea could be tested by using EEG along with vigilance tasks that include meaningful cues. In this context, the initiation of spontaneous thoughts is frequently associated with the cues included in the task, and its time of presentation can be precisely determined by the experimenter. Thus, ERPs could be used to assess whether the occurrence of spontaneous thoughts is consistently associated with activation in brain regions related with the episodic memory system, independently of more or less semantic-related activation.

Secondly, analyzing ERPs for spontaneous thoughts would be important to provide an alternative method to examine episodic specificity. In deliberate retrieval, thoughts with different levels of episodic specificity including general facts, personal semantics and unique episodic memories have been shown to have a distinguishable EEG profile (Renoult et al., 2016). Thus, we expect that spontaneous thoughts of different levels of specificity may also be distinguishable based on their EEG profile. Identifying these profiles would allow for the cross validation of episodic specificity between neural data and behavioral data based on participants’ descriptions. From this perspective, ERP analysis and the development of a coding system for episodic specificity in spontaneous thought, mentioned in the previous section, go hand in hand, and could be an essential stepping stone for future research.

Conclusion

In this review we explored the episodic-semantic interaction by focusing on spontaneous thought. We showed that, as in deliberate retrieval, there is an influence of both the episodic and semantic systems in spontaneous retrieval. This indicates that the two systems interact. To define the type of interaction between them, we focused on semantic priming studies, which revealed that the impact of the semantic memory system has been experimentally observed as a modulation of content. On the other hand, the episodic memory system seems to be particularly involved in spontaneous thought occurrence, when a cue leads to direct access of episodic content. Thus, spontaneous thought does not depend on a previous semantic search, and is a privileged research area to analyze the functional independence of the episodic system. We reviewed neuropsychological investigations of semantic dementia, which suggest spontaneous thought still occurs when the semantic memory system deteriorates, and indicates that the episodic memory system can function independently. Our conclusion is in line with recent views on the episodic-semantic interaction, which recognize the existence of an interaction while asserting that a certain degree of distinctiveness between the systems is present (Renoult et al., 2019). Yet, much remains to be understood regarding the forms of spontaneous thought that mix episodic and semantic elements, and are not easily categorized as one or the other. To explore this, we finished by suggesting two concrete avenues for research: developing a coding system for episodic specificity in spontaneous thought and analyzing event-related potentials of spontaneous thought.