Cue generation and memory construction in direct and generative autobiographical memory retrieval

Highlights

• We compared autobiographical memories reported to involve direct vs. generative retrieval.

• Participants rated field and observer perspective, ‘tell-tale’ signs of memory construction.

• We found frequent reported direct retrieval (approximately 50%), replicating recent findings.

• Direct (vs. generative) retrieval led to higher field and lower observer perspective.

• We suggest that direct and generative retrieval involve different content construction.

Abstract

Theories of autobiographical memory emphasise effortful, generative search processes in memory retrieval. However recent research suggests that memories are often retrieved directly, without effortful search. We investigated whether direct and generative retrieval differed in the characteristics of memories recalled, or only in terms of retrieval latency. Participants recalled autobiographical memories in response to cue words. For each memory, they reported whether it was retrieved directly or generatively, rated its visuo-spatial perspective, and judged its accompanying recollective experience. Our results indicated that direct retrieval was commonly reported and was faster than generative retrieval, replicating recent findings. The characteristics of directly retrieved memories differed from generatively retrieved memories: directly retrieved memories had higher field perspective ratings and lower observer perspective ratings. However, retrieval mode did not influence recollective experience. We discuss our findings in terms of cue generation and content construction, and the implication for reconstructive models of autobiographical memory.

Introduction

The assertion that autobiographical memory is ‘reconstructed’ is virtually a truism in cognitive psychology. In a move away from the computer or video tape metaphors which had previously emphasised memory’s role in reproducing the past, current theories of autobiographical memory emphasise its malleability, its selectivity, its broader, meaning-making functions, and its openness to social influences (e.g. Anderson, 1997, Bluck and Levine, 1998, Conway et al., 2004, Drivdahl and Hyman, 2013, Harris et al., 2014, Hasher and Griffin, 1978, Pasupathi, 2001). These ideas are not new, however, and can be dated back to Bartlett’s (1932) observation that “the first notion to get rid of is that memory is primarily or literally reduplicative, or reproductive. In a world of constantly changing environment, literal recall is extraordinarily unimportant… memory appears to be far more decisively an affair of construction rather than one of mere reproduction” (pp. 204–205). However, the concept of memory reconstruction is used in a variety of contexts and may encompass a variety of different processes in autobiographical memory retrieval (see also Michaelian, 2011).¹ In the current paper, we investigate the relationship between two processes – cue generation and content construction – and whether there is a distinction between them, by comparing the characteristics of memories retrieved directly and memories retrieved generatively.

Theories of autobiographical memory suggest that two qualitatively distinct retrieval processes underlie everyday memory: direct retrieval – remembering without an experience of effortful search; and generative retrieval – remembering with an experience of deliberate or effortful search (Addis et al., 2012, Conway and Pleydell-Pearce, 2000, Haque and Conway, 2001, Rubin and Berntsen, 2009, Uzer et al., 2012). According to Conway’s Self-Memory System model of autobiographical memory (Conway, 2005, Conway and Pleydell-Pearce, 2000), these two retrieval processes access the same store of ‘event-specific’ autobiographical knowledge. Direct retrieval is a ‘bottom-up’ process (Williams et al., 2006), that occurs when a cue shares enough features with a memory to trigger direct access to the autobiographical knowledge base (Conway, 2005, Conway and Loveday, 2010). Generative retrieval, on the other hand, is a ‘top-down’ process (Williams et al., 2006), that involves goal-directed, hierarchical searching and spreading activation across the memory network in an attempt to retrieve a memory that meets current goals. In generative retrieval, associative memory networks mean that potential memory cues activate each other, until a cue is activated that successfully accesses the searched-for memory in the autobiographical knowledge base (Conway, 2005, Conway and Loveday, 2010). That is, in generative retrieval, when encountering a memory cue, one has to bring to mind additional, related information and knowledge from one’s life before a specific memory is recalled (Uzer et al., 2012).

Some empirical research has supported this distinction between retrieval processes. Direct retrieval is associated with shorter retrieval latencies than generative retrieval (e.g. median response time of 2–4 s vs. 10–14 s in Uzer et al., 2012, Experiments 1–3; see also Addis et al., 2012). Direct retrieval is more common in response to concrete noun cues, and generative retrieval is more common in response to abstract emotion cues (Uzer et al., 2012). Neuroimaging research has identified strong overlap as well as some interesting differences in brain activity during direct vs. generative retrieval. Specifically, Addis et al. (2012) found that direct retrieval was associated with stronger activation generally across the neural ‘autobiographical memory network’, and particularly, increased activity in the visuospatial areas. On the other hand, generative retrieval was associated with increased activity in regions involved in executive control, memory search, retrieval of semantic information, memory elaboration, and post-retrieval monitoring (Addis et al., 2012). Conway and Loveday (2010) described the case of an amnesic patient who was unable to engage in generative retrieval, but whose memory function was intact for direct retrieval when she encountered an appropriate cue.

A related but not identical distinction has been made between voluntary and involuntary remembering (see Berntsen, 2010). Although direct retrieval and involuntary remembering are sometimes conflated in the literature, the defining feature of direct retrieval is its lack of effort, while the defining feature of involuntary retrieval is its lack of intentionality (see also Uzer et al., 2012). While involuntary memories must involve direct and not generative retrieval (since conscious effort to search for a memory must involve the intention to recall), voluntary remembering can involve either direct retrieval or generative retrieval. In fact, the paradigms that have indexed direct and generative retrieval as described above have involved intentional (i.e. voluntary) retrieval, in which participants are given the intention (via task demands) of recalling a memory in response to each cue (Addis et al., 2012, Uzer et al., 2012). Models of autobiographical memory have traditionally emphasised generative retrieval as more common than direct retrieval. For instance, Haque and Conway (2001, p. 532) stated that “More common than direct retrieval is the effortful repeated cycle of access, evaluate, and elaborate as a memory is constructed during the process of generative retrieval”. However, new research indicates that direct retrieval occurs as frequently as generative retrieval, even in a laboratory-based intentional autobiographical memory task (Uzer et al., 2012). This finding parallels recent research suggesting that involuntary retrieval is a mundane, everyday experience (Berntsen, 2010).

Uzer et al. (2012) suggested that the generation involved in generative retrieval is the generation of cues, rather than the generation of memory content per se. That is, generative retrieval is a backup strategy that is initiated when the presented cue is unsuccessful at directly triggering a memory; in generative retrieval then, people continue to self-generate additional (internal) memory cues until they activate a cue that directly triggers a memory (Uzer et al., 2012). This proposition makes a strong conceptual distinction between cue generation and content construction. An alternative hypothesis – based on Conway, 2005, Conway and Pleydell-Pearce, 2000, Haque and Conway, 2001) hierarchical model of autobiographical memory – is that the generation of memory cues in practice is not separable from the construction of memory content, because the cues in question are linked to the memory in a hierarchy of increasing specificity. As an example, given the cue word ‘dog’, an individual might first think of their childhood dog, before recalling finding a puppy under the tree on Christmas morning. While moving from the generic cue ‘dog’ to thinking of one particular dog is certainly cue generation, it is also constructing part of the content that will make up the specific event (see also Addis et al., 2012). A third possibility lies somewhere between these viewpoints: it is possible that cue generation and content construction are distinct processes but that they both operate during retrieval.

These alternative accounts lead to quite different understandings about when the autobiographical memory process is initiated and to quite different understandings about the nature of autobiographical memory retrieval. On one hand, Uzer et al. (2012) argue that generative retrieval only involves generation of cues and not of memory content, and thus that both direct and generative retrieval access “pre-stored event representations”. In other words, according to Uzer et al., both processes access event representations that are not the result of reconstruction from fragments of personal knowledge. Uzer et al. (2012) therefore state that they present a fundamental challenge to the prevailing view that the content of autobiographical memories is reconstructed during retrieval (p. 1306), suggesting that there is no evidence for the reconstruction of memory content during retrieval and that existing data can instead be explained by differences in cue generation that precedes memory retrieval. Alternatively, based on the opposing view presented above (c.f. Conway, 2005) the processes of cue generation and content reconstruction are likely to in fact overlap, because the “cues” that are activated in the hierarchical memory system also contain memory content. While such an argument is not explicit in Conway’s (2005) model, it is consistent with his view of the hierarchical nature of memory and the processes of cue generation as occurring via spreading activation within that hierarchy (Haque & Conway, 2001). This latter view is also consistent with Addis et al.’s (2012) findings regarding neural differences in direct vs. generative retrieval, which they argued implied that direct and generative retrieval differ not only in reaction times, but also in the content and characteristics of the resulting memories constructed via these two retrieval processes. In the current paper, we empirically test whether directly and generatively retrieved memories differ in terms of the content of memories generated. Such a finding would be more consistent with the latter view that generative retrieval involves different processes of memory construction (c.f. Conway, 2005), and inconsistent with the former view that generative retrieval involves only the generation of cues entirely distinct from the memory itself (Uzer et al., 2012). On the other hand, if direct and generative retrieval are indistinguishable in terms of the memory content they produce (and only distinct in terms of time or effort involved in retrieval), this would support Uzer et al.’s (2012) account.

Thus, in the current study, we tested whether direct and generative retrieval (measured via self-report) differed in terms of the characteristics of the memories recalled, as well as in terms of reaction time. If direct and generative retrieval differ only in terms of cue generation, we should expect reaction time differences but no differences in the characteristics of memories elicited via these two processes. Alternatively, differences in characteristics of directly and generatively retrieved memories would imply that these two retrieval processes involve differences in the construction of memory content as well as in cue generation. We focused on two particular memory characteristics which have been linked to construction during memory retrieval; we reasoned that differences in these variables might reflect differences in the processes of memory construction. The first characteristic was the perspective or point of view from which the memory was visualised as it was recalled (‘visuo-spatial perspective’), and the second characteristic was the phenomenological experience of remembering as the memory was recalled (‘recollective experience’). We discuss each of these in turn.

Following Freud (1899), Nigro and Neisser (1983) distinguished between two perspectives or vantage points from which an experienced past event may be recalled. The first is the first-person perspective, where the memory involves visual imagery from the same vantage point as at encoding; that is, as they recall it the rememberer sees the event as from their own eyes. Nigro and Neisser (1983) called this “field perspective”. The alternative is the third-person perspective, where the memory involves visual imagery from a different vantage point than at encoding; that is, as they recall it the rememberer sees the event from some other perspective than that of their own past self, and sees themselves in the memory. Nigro and Neisser (1983) called this “observer” perspective. In studying the characteristics of memories recalled from field vs. observer perspective, Nigro and Neisser (1983) found that field perspective was more common in general, but that observer perspective was more likely for older memories, memories that were less emotionally intense, and memories that were higher in self-consciousness (see also Rice, 2010, for a review of the literature on perspective, emotion, and self-consciousness).

Of most interest to the aims of the current paper, visuo-spatial perspective is one characteristic of memories that has been linked to reconstructive processes during memory retrieval. Nigro and Neisser (1983) asked, based on an earlier argument made by Freud (1899), whether memories recalled from an observer perspective may have been subject to more (or at least different) constructive processes during memory retrieval, since the recollection is less similar to the original image than for memories recalled from a field perspective. They suggested that “some types of reconstruction may leave tell-tale signs in phenomenal experience itself” (Nigro & Neisser, 1983, p. 468, our emphasis), and that differences in visual imagery associated with the memory might be one such tell-tale sign of different constructive processes. Other researchers have made similar arguments. For example, Conway (2009) stated that observer perspective may “indicate more memory construction” (p. 2306, our emphasis) and, even more strongly, Siedlecki (2014) argued that “generally speaking, all observer memories are essentially false (inaccurate) memories because unless the individual had an out of body experience during the event, only a field perspective memory is possible” (p. 2). On the other hand, some researchers have defended the possibility that remembering from an observer perspective can in principle be just as accurate as from a field perspective (Debus, 2007, Sutton, 2010). While some previous research found that true childhood memories were predominantly remembered from a field perspective and false childhood memories were remembered from an observer perspective (Heaps & Nash, 2001), other researchers have found no difference in perspective of true and false memories (McIsaac and Eich, 2002, Porter et al., 1999, Siedlecki, 2014).

If retrieval mode reflects only cue generation and not the processes by which the memory content was constructed, we might expect no systematic differences in visuo-spatial perspective for directly vs. generatively retrieved memories. However, if retrieval mode reflects both cue generation and memory content construction, we might expect a tendency for directly retrieved memories to involve a field perspective and for generatively retrieved memories to involve an observer perspective, since generative retrieval might indicate more effortful construction of memory content as well as generation of memory cues. Thus, in the current study we compared the visuo-spatial perspective of directly vs. generatively retrieved memories.

Tulving (1985) introduced the distinction between memory retrieval associated with a feeling of consciously re-experiencing the past, or “autonoetic awareness”, and memory retrieval not associated with a feeling of re-experiencing, or “noetic awareness”. This distinction in the nature of recollective experience is typically operationalized as an individual’s judgement of whether they ‘remember’ that something happened or simply ‘know’ that something happened (Tulving, 1993; see also Gardiner, 1988). In autobiographical memory research, participants report both ‘remember’ and ‘know’ judgements for their own life events, and ‘remember’ judgements are associated with higher ratings on a range of other phenomenological characteristics, including vividness, sensory perceptual detail, emotion, and confidence in accuracy (Hyman et al., 1998, Rubin et al., 2003).

In the recognition memory literature, there is some evidence to suggest that recognition driven by rich recollective experience (i.e. ‘remember’) is significantly faster than recognition driven by mere familiarity (i.e. ‘know’; Dewhurst, Holmes, Brandt, & Dean, 2006) and is both more accurate and expressed with greater subjective confidence (Yonelinas, 2002), paralleling some of the characteristics associated with direct retrieval (Uzer et al., 2012). However, other work has suggested just the opposite; that a remember judgement is an effortful disambiguation process used when familiarity is not diagnostic. For instance, Mandler (1980) described the experience of sitting next to a man on a crowded bus and having a strong sense that you have seen him before. This instant sense of familiarity is “knowing”, but it is only through an effortful search process that you “remember” the last time you saw him and are able to identify him as your butcher (see also Hintzman, 1988, Huppert and Piercy, 1978). This effortful process mirrors the process of cue generation involved in generative retrieval. However, it is currently unknown whether there is a link between recollective experience and retrieval mode: can both direct and generative retrieval involve both remembering and knowing?

Previous research in autobiographical memory has also linked these two characteristics of retrieved autobiographical memories, connecting recollective experience with visuo-spatial perspective. Crawley and French (2005) asked participants to describe childhood memories that they ‘remember’, childhood memories that they simply ‘know’, and childhood memories which they were unsure if they ‘remember’ or ‘know’. They found that field perspective dominated ‘remember’ events, while observer perspective dominated ‘know’ events (see also Libby, 2003). Piolino et al. (2006) found that older adults (compared to younger adults), showed decreased ‘remember’ judgements and increased ‘know’ judgements, as well as parallel decreased field perspective and increased observer perspective. Thus, there is consistent evidence that remember judgements are linked with field perspective and know judgements are linked with observer perspective.

As with visuo-spatial perspective, we used recollective experience to index differences in memory characteristics following different retrieval processes. If retrieval mode reflects only cue generation and not the processes by which the memory content was constructed, we might again expect no differences in recollective experience for directly vs. generatively retrieved memories. However, if retrieval mode reflects both cue generation and memory content construction, we might expect a tendency for directly retrieved memories to be judged ‘remembered’ and generatively retrieved memories to be judged as ‘known’ (consistent with the links between recollective experience and visuo-spatial perspective described above). Thus, in the current study we compared the recollective experience associated with directly vs. generatively retrieved memories.

In the current study, we examined whether directly and generatively retrieved memories differ in their characteristics, particularly in terms of their visuo-spatial perspective and recollective experience. Participants elicited memories in response to cues, and reported their retrieval mode, visuo-spatial perspective, recollective experience, and a range of other phenomenological characteristics. If reported direct vs. generative retrieval only reflects cue generation as suggested by Uzer et al. (2012), then we should see no differences in the characteristics of memories retrieved via these two different retrieval modes. But if generative retrieval reflects different processes of memory content construction from direct retrieval as suggested by Addis et al. (2012) – perhaps more effortful construction of memory content as well as effortful generation of cues – then direct retrieval may be associated with field perspective and with ‘remember’ judgements, while generative retrieval may be associated with observer perspective and with ‘know’ judgements.