The Architecture of Autobiographical Memory: Hierarchical Organisation and Sources of Fragility

Abstract

Autobiographical memory is often treated as a store of discrete personal episodes that can be retrieved and copied. This review examines whether the evidence supports that assumption. We consider the hierarchical self-memory system of Conway and Pleydell-Pearce (2000), Tulving’s distinction between episodic and semantic memory and his account of autonoetic consciousness, Rubin’s basic-systems model, and the reconstructive tradition initiated by Bartlett (1932). We then review neural evidence from neuroimaging meta-analyses and behavioural evidence from the Autobiographical Interview, lifespan studies of episodic and semantic autobiographical recall, childhood amnesia and the reminiscence bump. The convergent conclusion is that a personal memory is a transient construction assembled across several levels of knowledge and several neurocognitive systems under the control of current goals. The components that carry emotional and experiential quality, namely event-specific sensory-perceptual detail and autonoetic re-experiencing, are the least stable across time, age and pathology. A programme seeking to preserve human emotional memory must therefore target the conditions of reconstruction rather than a putative stored trace, and must treat the act of elicitation as part of what is preserved.

Keywords: autobiographical memory; episodic memory; autonoetic consciousness; self-memory system; reconstruction; reminiscence bump

1. Introduction

Autobiographical memory refers to memory for the events and facts of one’s own life. It is among the least tractable forms of memory, because encoding conditions are uncontrolled, retention intervals span decades and accuracy is often unverifiable. Two developments made it tractable. The first was a set of theoretical proposals treating autobiographical memory not as a single store but as an organised system with identifiable levels and components (Conway & Pleydell-Pearce, 2000; Rubin, 2005, 2006; Tulving, 1985, 2002). The second was a set of methods that made the phenomenon measurable: cue-word and lifespan-distribution paradigms (Rubin & Schulkind, 1997), structured interviews separating episodic from semantic content in free narratives (Levine et al., 2002), and neuroimaging protocols capable of isolating retrieval of personal events (Cabeza & St Jacques, 2007; Svoboda et al., 2006).

The present review draws these strands together around one question: what is the structure of autobiographical memory, and which parts of it are fragile? Emotion in memory is not a free-standing item; it is bound to particular events, to particular levels of representation and to a particular mode of conscious access. Which bindings survive time, ageing and disease determines what “preserving” an emotional memory could mean.

2. Architecture: levels, systems and modes of awareness

2.1 The self-memory system

Conway and Pleydell-Pearce (2000) proposed that autobiographical knowledge is organised in three levels of specificity, integrated with a control structure they termed the working self. At the most abstract level are lifetime periods: thematically defined spans, such as a period of study or of living in a particular city, with fuzzy boundaries and associated knowledge of typical people, places and goals. Below them are general events: repeated events, extended events lasting days or weeks and, notably, first-time experiences, treated as privileged access points because they are densely linked to the goals of their period. At the lowest level is event-specific knowledge: sensory-perceptual detail, imagery, fragments of speech and affect specific to a single occurrence.

The model’s central claim is that memories are not stored as units at any level. A specific memory is a temporary pattern of activation spanning all three, produced by a generative retrieval process in which cues are elaborated, candidate knowledge is accessed and evaluated against the current goals of the working self, and the search iterates until a stable pattern emerges or is abandoned. Retrieval therefore reflects a compromise between what the authors call correspondence (fit to what happened) and coherence (fit to the current self). The hierarchy was supported by retrieval-time studies in which general-event cues produced the fastest and most specific recall, and by over-general memory in depression, interpreted as search terminating at the general-event level.

2.2 Episodic and semantic memory; noetic and autonoetic awareness

Tulving (1972) distinguished episodic memory, defined as memory for temporally dated, spatially located events experienced by the rememberer, from semantic memory, organised knowledge detached from the circumstances of acquisition. Tulving (1985) reformulated the distinction in terms of the accompanying consciousness. Semantic retrieval is accompanied by noetic awareness: one knows a fact. Episodic retrieval is accompanied by autonoetic awareness: one is aware of oneself as the person to whom the event happened and has the subjective sense of re-experiencing it in its original temporal context. Tulving (2002) further argued that episodic memory is ontogenetically late, depends on but is not reducible to semantic memory, and is defined by subjective mental time travel.

This partition cuts across the Conway hierarchy. Knowledge of one’s own life can be held semantically (that one attended a particular school) without any capacity to re-experience the episodes. Such personal semantic memory is well retained; episodic autobiographical memory, which requires autonoetic access to event-specific knowledge, is not. Renoult et al. (2019) argue that the two forms are ends of a continuum with intermediate forms (autobiographical facts, repeated events), a reframing consistent with the architecture reviewed here.

2.3 The basic-systems model

Rubin (2005, 2006) proposed that autobiographical memory is not a memory system in its own right but the coordinated product of several basic systems, each with its own neural substrate and representational format: vision, audition, olfaction and other senses, spatial imagery, language, narrative, emotion, and an explicit-memory system that binds them. A vivid memory is one in which these systems are jointly active; a faded memory is one to which only narrative and language still contribute. The model predicts that visual imagery should be the strongest correlate of the sense of reliving, as Rubin and colleagues reported in questionnaire studies, and that damage to a single system should partially impair recall.

2.4 Reconstruction

The constructive framing of all three models has an explicit ancestor in Bartlett (1932). In repeated reproduction of an unfamiliar narrative, Bartlett observed shortening, omission of unfamiliar elements, rationalisation and substitution of familiar items, and attributed these to schemata, organised bodies of past reactions shaping encoding and recall. His conclusion, that remembering is an imaginative reconstruction from a general attitude and a few salient details, rested on demonstrations that were informal by later standards, but the qualitative pattern has since been reproduced in controlled designs, and the schema concept is the direct antecedent of the coherence constraint in Conway and Pleydell-Pearce (2000).

3. Neural correlates

Svoboda et al. (2006) meta-analysed 24 PET and fMRI studies of autobiographical retrieval and identified a core network comprising medial and ventrolateral prefrontal cortex, medial and lateral temporal cortex including the hippocampus, temporoparietal junction, posterior cingulate and retrosplenial cortex, and cerebellum. Heterogeneity in cueing procedures and control conditions limits the specificity of the core-network claim, but the network has been reproduced in later syntheses.

Cabeza and St Jacques (2007) organised the same literature by component process: controlled retrieval in lateral prefrontal cortex, self-referential processing in medial prefrontal cortex, recollection in the hippocampus, visuospatial imagery in occipital and parietal regions, and emotional processing in the amygdala. The network is thus a set of regions whose joint engagement parallels that of Rubin’s basic systems. Two findings bear on fragility: hippocampal activity tracked the vividness and detail of recall rather than the age of the memory in several studies, although whether remote memories become hippocampus-independent remains contested; and medial prefrontal activity was associated with self-relevance, consistent with working-self control. More recent work argues that the subjective sense of remembering depends on interactions among medial temporal, prefrontal and parietal regions rather than on any one (Simons et al., 2022).

Two cautions apply: imaging studies rely on memories of unknown accuracy, and fMRI cannot separate the iterative phases of generative retrieval. The neural evidence constrains, but does not by itself establish, the constructive account.

4. Sources of fragility

4.1 Episodic and semantic content across the adult lifespan

Levine et al. (2002) developed the Autobiographical Interview to quantify the episodic and semantic content of freely recalled events. Transcripts are segmented into details classified as internal (event-specific happenings, perceptions, thoughts and emotions tied to the target event) or external (semantic facts, generic information, repetitions, other events), and a probe phase tests whether further episodic detail can be elicited. In the original study, older adults (mean age about 74) produced fewer internal and more external details than younger adults (mean age about 26) across all lifetime periods, in both free and probed recall. Older adults’ narratives are often longer, but the additional material is semantic. Whether it reflects loss of episodic representations, altered retrieval strategy or altered narrative goals cannot be decided from the interview alone.

Piolino et al. (2009) reviewed cognitive, neuropsychological and neuroimaging findings on episodic autobiographical memory across the lifespan, using tasks that separately score specificity, detail and sense of reliving. Three conclusions follow. The capacity to retrieve specific, detailed, autonoetically experienced episodes declines with normal ageing, whereas personal semantic memory is largely spared. Remote memories are the most semanticised, consistent with a gradual transformation in which repeated retrieval strips specific detail and leaves a schematic version. And in Alzheimer’s disease and semantic dementia the two forms dissociate in opposite directions, episodic components being lost early in the former and recent episodic memories relatively preserved alongside impaired personal semantics in the latter. These double dissociations are the strongest evidence that the partition is real rather than a scoring convention.

4.2 The lifespan distribution: childhood amnesia and the reminiscence bump

When adults date memories elicited by cue words, the distribution across age at encoding has a characteristic shape. Rubin and Schulkind (1997) reported it in 20-, 35- and 70-year-olds. Very few memories were dated to the first three years of life, with a gradual increase to about age seven; participants aged 35 and 70 produced a disproportionate number of memories from roughly ages 10 to 30, a reminiscence bump absent in the 20-year-olds; and all groups showed a recency component. The bump was present for memories rated unimportant as well as important, which argues against an explanation by a few landmark events. Explanations are not mutually exclusive: a cognitive account emphasises the density of first-time experiences in adolescence and early adulthood, which the self-memory system treats as privileged cues; an identity account holds that the bump coincides with the formation of a stable adult self; life-script accounts invoke normative expectations about when important events occur. The effect itself is among the best-replicated in the field.

Childhood amnesia is the other stable feature. Bauer (2015) proposed a complementary-processes account in which encoding and consolidation are less effective in early childhood, because of protracted maturation of hippocampal and prefrontal circuits and because the child’s representations lack the temporal, spatial and self-referential organisation that later supports retrieval, while forgetting is also faster, so that memories which are formed are lost before rehearsal can stabilise them. Longitudinal evidence that children retain early memories for years and then lose them supports this account, which locates childhood amnesia in accelerated forgetting rather than failure to encode.

4.3 The general pattern

Across these literatures the fragile components are the same. Event-specific sensory-perceptual detail is the first to be lost with time and age; the autonoetic sense of reliving is the first to be lost in pathology; both depend on the coordination of multiple systems. Personal semantic knowledge, the fact that an event occurred, is comparatively robust. The emotional quality of a memory tracks the fragile components: the emotion is not the fact that one was afraid but the reconstructed experience of that fear, which requires the sensory and autonoetic contributions lost first.

5. Implications for emotion preservation

The evidence reviewed imposes several constraints on any programme whose object is the preservation of human emotional memory. We formulate them as requirements and testable propositions.

First, the object of preservation cannot be a stored trace, because on all current models no such trace exists in a form that could be copied. We propose that the appropriate target is the set of conditions under which a specific memory can be reconstructed with its sensory-perceptual and autonoetic components intact.

Second, because event-specific knowledge and autonoetic access decay first, they must be captured close to encoding. A testable proposition follows: material recorded within days of an event should support later reconstruction with higher internal-detail scores on the Autobiographical Interview than material recorded months later, with a larger difference for sensory and emotional than for factual details.

Third, elicitation is not neutral. Generative retrieval is guided by the working self, and repeated retrieval semanticises. Any procedure that asks a person to recount an emotional event participates in the reconstruction it records and alters the memory for later retrievals. A programme should document its elicitation procedure as part of the preserved record, and should test whether multi-modal cueing, as predicted by the basic-systems model, elicits more internal detail with less schematic intrusion than free narrative.

Fourth, the lifespan distribution implies unequal opportunities. Emotional memories from the reminiscence-bump years will be over-represented in any retrospective collection, those of early childhood under-represented regardless of effort. A programme should decide whether it aims to preserve what is accessible to the person or what occurred.

Finally, several questions remain open: whether a reconstruction supported by external records preserves autonoetic quality or merely substitutes personal semantic knowledge; whether self-relevance, which governs retrieval, can be preserved when the self that assigned it is no longer available; and what degree of reliving is required for an emotional memory to count as preserved rather than merely recorded.

6. Conclusion

The models of Conway and Pleydell-Pearce, Tulving, Rubin and Bartlett differ in their units of analysis but converge on a single description: an autobiographical memory is assembled at the time of recall, from knowledge held at several levels of specificity and from the outputs of several neurocognitive systems, under the guidance of present goals. Neuroimaging supports this picture without yet resolving the mechanism of construction. Behavioural evidence from the Autobiographical Interview, lifespan studies and the cue-word distribution identifies the same fragile components in every case: specific sensory detail and the sense of reliving. Personal semantic memory endures, but the fragile components carry the emotion. A science of emotion preservation must therefore be a science of reconstruction, concerned with what allows a person to re-experience an event, rather than a science of storage.

References

Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge University Press.

Bauer, P. J. (2015). A complementary processes account of the development of childhood amnesia and a personal past. Psychological Review, 122(2), 204–231.

Cabeza, R., & St Jacques, P. (2007). Functional neuroimaging of autobiographical memory. Trends in Cognitive Sciences, 11(5), 219–227.

Conway, M. A., & Pleydell-Pearce, C. W. (2000). The construction of autobiographical memories in the self-memory system. Psychological Review, 107(2), 261–288.

Levine, B., Svoboda, E., Hay, J. F., Winocur, G., & Moscovitch, M. (2002). Aging and autobiographical memory: Dissociating episodic from semantic retrieval. Psychology and Aging, 17(4), 677–689.

Piolino, P., Desgranges, B., & Eustache, F. (2009). Episodic autobiographical memories over the course of time: Cognitive, neuropsychological and neuroimaging findings. Neuropsychologia, 47(11), 2314–2329.

Renoult, L., Irish, M., Moscovitch, M., & Rugg, M. D. (2019). From knowing to remembering: The semantic–episodic distinction. Trends in Cognitive Sciences, 23(12), 1041–1057.

Rubin, D. C. (2005). A basic-systems approach to autobiographical memory. Current Directions in Psychological Science, 14(2), 79–83.

Rubin, D. C. (2006). The basic-systems model of episodic memory. Perspectives on Psychological Science, 1(4), 277–311.

Rubin, D. C., & Schulkind, M. D. (1997). The distribution of autobiographical memories across the lifespan. Memory & Cognition, 25(6), 859–866.

Simons, J. S., Ritchey, M., & Fernyhough, C. (2022). Brain mechanisms underlying the subjective experience of remembering. Annual Review of Psychology, 73, 159–186.

Svoboda, E., McKinnon, M. C., & Levine, B. (2006). The functional neuroanatomy of autobiographical memory: A meta-analysis. Neuropsychologia, 44(12), 2189–2208.

Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381–403). Academic Press.

Tulving, E. (1985). Memory and consciousness. Canadian Psychology / Psychologie canadienne, 26(1), 1–12.

Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25.