1 Introduction

Patients with Korsakoff’s Syndrome (KS) often confabulate to cope with memory loss. I analyze their confabulations using a representational approach that emphasizes error detection. I extend the teleosemantic view, typically used to explain psychopathological representations like delusions (Miyazono, 2019), to include a novel account of memory in communicative terms. This view, suggested by teleosemantics proponents (Godfrey-Smith, 2012, 2014), allows for analyzing various types of memory malfunction. Clinical confabulations are seen as representations in a malfunctioning communication system.

I present a unified framework for understanding confabulations in KS, showing how they arise from failures in internal error-detection mechanisms and disrupted feedback loops involving social environments. By integrating individual and interpersonal levels of analysis, this account illuminates the complex nature of clinical confabulations.

Error detection is not only a plausible necessary condition for the cognitive use of representation but also yields specific heuristics that are informative for the study of psychopathology, particularly when it is related to memory malfunction. In this way, the error condition contributes to a systematic inquiry into their nature in a unified manner. This makes the emerging error-based approach a potential foundation for a progressive research program.

I will proceed in the following order. The second section introduces the error-based approach to mental representation. The subsequent section focuses on KS and describes the nature of clinical confabulations in KS, giving a short overview of provoked and spontaneous confabulations based on examples. The fourth section provides a teleosemantic account of memory, which is used in the following section to analyze confabulations in KS from the error-based perspective. After discussing two possible objections, the paper concludes by arguing that the error-based perspective provides useful insights into the nature of psychopathology by intertwining individual and sociocultural representations and offering guidance for their systematic study.

2 Mental Representations and Error Detection

Mental representations are intentional mental states or processes, as they are about something (Brentano, 1900; for a review of contemporary positions, see Schulte, 2023). They have vehicles (e.g., a proposition in the language of thought or a neuronal process of the central nervous system), might refer (to an object, a feature of an object, state of affairs), and hold contents, usually equated with satisfaction conditions, which indicate when the representation is accurate (e.g., satisfied by entities in the case of entity presence detectors or true in the case of propositional content).

The perspective taken in this paper derives from teleosemantic theories of mental contents, which appeal to biological functions. Mental representations are cognitive states or processes of a cognitive system whose ability to represent is explained functionally: They have a representational biological function, meaning that they are biologically endowed with the capacity to represent. Many supporters of teleosemantics follow Millikan in endorsing her account of biological proper function (Millikan, 2002), but other approaches to function are also found in the literature (e.g., Dretske, 1986; Neander, 1995, 2017). Representations are tokened by producer mechanisms and consumed by (possibly many) consumer mechanisms. For example, a perceptual representation could be produced by the dorsal stream—one of the vision pathways—and consumed by a motor control subsystem that exploits the contents of the perceptual representations that specify the location of a perceptual object.

In the rest of the paper, I develop the emerging position among defenders of teleosemantic representationalism, which holds that accounting for misrepresentation is not only necessary for a satisfactory account of representation, but that mental representations owe their representational status to their capacity for error detection (Bielecka & Marcinów, 2017; Bielecka & Miłkowski, 2020; Buckner, 2022; Gładziejewski, 2015; Lee, 2018; Miłkowski, 2013). The error detection requirement makes it clear that the functional role of representation in the cognitive economy involves at least the capacity to detect one's own representational error. The content of a representation need not be actually exploited, but must only be available for further exploitation. Similarly, the error need not be detected at all times, but there should be an error detection consumer mechanism for all purported representation producer mechanisms.

The general intuition behind the idea that error detection is necessary for mental representation can be sketched as follows: Only if the agent cares about the accuracy of its own representation R, can we ascribe the representational role to R. This is because R apparently plays a representational role in this agent: Responding to the inaccuracy of representation is a crucial precondition for its representational use (for a similar argument, see Bickhard, 2008; Fodor, 1992, p. 107).

Arguably, if the function of error detection is a necessary feature of any cognitive system that relies on cognitive representation, it can not only justify the assignment of the representational status to a given piece of semantic information, but also make content ascriptions more precise. Take the standard fictional example of kimus and snorfs. Kimus have the function of responding selectively to occasions in which their predators, snorfs, can be found. Yet, they apparently have it without really representing snorfs but merely detecting the color of the sunset and following it to the hilltop, where snorfs are not found (Pietroski, 1992). On the error-based account, kimus cannot be ascribed elementary representational contents that outstrip their capacities for possible error detection. If kimus cannot ever detect that their selective response to the occasions in which snorfs (and nothing else) are found is in error, they cannot be ascribed contents in which snorfs figure. This is because they do not respond to representational errors related to snorfs. Then they only selectively respond to physical occurrences in which snorfs can be found. Or, to take a real example (Lettvin et al., 1959), frogs can be ascribed bug detectors (in contrast to bee-bee detectors) only if they could ever detect that they were mistakenly eating bee-bees. Without additional evidence, we can never know (the original 1959 article does not mention it). In short, on this account of contents, we can legitimately ascribe it to cognitive systems only in a way that respects their abilities to respond to appropriate semantic properties. This implies that at least some indeterminacy of content ascriptions can be mitigated.

Additionally, in this approach, the contents, or satisfaction conditions of representation, necessarily have causal powers (Gładziejewski & Miłkowski, 2017; Miłkowski, 2015), which stands in opposition to claims that only syntax and not meaning can have causal powers. For example, Dennett claims that responding to contents is a “mechanistic impossibility” (Dennett, 1991, p. 119). However, error-detection mechanisms inherently causally respond to the contents. In other words, content is not a mere gloss ascribed by a theorist. At the same time, content is accounted for in terms of semantic externalism, which implies that the causal role of content need is not reducible to internal mental causation. This is because content does not supervene locally on the structure of a single cognitive system.

Confabulations are naturally viewed as representations, typically erroneous, and teleosemantics addresses the naturalization of misrepresentation, which is a key requirement for a satisfactory theory of mental representation (Dretske, 1986). Autobiographical confabulations, such as those experienced by KS patients, are a prime example of “representation-hungry” phenomena (Clark & Toribio, 1994) because they often describe what is not perceptually available (in contrast to perceptual illusions and hallucinations).

Supporters of error-based teleosemantics offer various accounts of error detection. For example, Buckner (2022) insists that there should be predictive learning processes that respond to discrepancies between predicted and actual values in a predictive processing system, while in our previous work (Bielecka, 2019); Bielecka & Miłkowski, 2020; Miłkowski, 2013), we require only that there is some inconsistency detection between two tokens of semantic information. Gładziejewski (2015) seems to suggest that error detection happens only for structural representations.

In all these flavors of error-based teleosemantics, a meta-representation of error is not required in all instances. Notice that this requirement could be a major disadvantage of this approach. Artiga argues against presupposing metarepresentational abilities for all kinds of representation: “the ability to detect its own mistakes seems to require some sort of metarepresentational ability, and it is unclear why we should require metarepresentations in order to possess first-order representations” (Artiga, 2021, p. 662). Clearly we should not, as this would lead to a vicious regress, because these metarepresentations should also be checked for, and that would involve a meta-representation for the meta-representation, and so on (Bielecka, 2019). My solution is simple: The error-detecting mechanism does not require yet another full-blown representation. Instead, mere semantic information is required to check for consistency. Similarly, Buckner’s approach does not seem to require any meta-representation in any strong sense for dealing with discrepancies between prediction and actual values.

In contrast, Shea (2014) argues that error detection requires meta-representational capacities but does not make them a necessary condition for representation. Thus, while some learning processes rely on meta-representation, representation itself may not always involve accuracy checks. The error-based approach, however, views error detection as a necessary condition of representationality, not just an additional feature of producer–consumer systems. In this paper, I follow my previous approach to error-based teleosemantics, which avoids positing meta-representational capacities.

In the subsequent sections of the paper, I argue that the error-based view leads to plausible hypotheses about the nature of confabulation in KS and points out what we might be missing in the existing empirical evidence.

3 Korsakoff's Syndrome and Clinical Confabulations

The primary rationale for focusing on KS in this paper is based on three key considerations. First, the distinctive features of the confabulations in KS are particularly relevant to teleosemantic discussions and will be useful for illustrating the account of memory introduced later (Sect. 4). Second, the well-documented influence of caregivers and clinicians on Korsakoff patients provides concrete examples of how external consumers can incite or mitigate confabulations through coherence cues. This dynamic offers a rich context for understanding confabulation, which is also part of our teleosemantic account and can be extrapolated to other disorders. Third, the extensive documentation of patient histories in KS offers a wealth of qualitative data. This allows for a detailed analysis of how confabulations develop and evolve over time, supporting robust teleosemantic analyses. The depth of these case histories makes KS an ideal model for studying the mechanisms of confabulation.

In a recent review, KS is described as

… a largely irreversible residual syndrome, caused by severe thiamine deficiency and occurring after incomplete recovery from a Wernicke encephalopathy, predominantly in the context of alcohol abuse and malnutrition, characterized by an abnormal mental state in which episodic memory is affected out of all proportion to other cognitive functions in an otherwise alert and responsive patient, whose psychological make-up may be further distinguished by executive dysfunction, flattened affect, apathy, lack of illness insight, and possibly by fantastic confabulations in the early stage (Arts et al., 2017).Footnote 1

Patients with KS have severe memory loss, especially for episodic memory (their procedural memory is rather intact). Executive dysfunction and other cognitive or behavioral disorders may be absent in less severe forms of KS. If present, patients have problems with many of their cognitive skills such as time orientation. This results in lack of interest in their surroundings, difficulties in keeping social interactions, suggestibility, and problems with information detection and suppression. Some may also have perceptual deficits (Berlyne, 1972; Talland, 1961). They may also have severe problems in planning and task organization, i.e., organizing their memories in the appropriate temporal sequence (Arts et al., 2017; Grünthal, 1924).

3.1 Clinical Confabulations in KS

As individuals who communicate by sharing personal and other narratives, we often inadvertently tell stories that are either inaccurate or based on our current emotional state and perceived truth. This tendency to confabulate can be observed when, for instance, recounting a childhood experience such as a relative's birthday party while also experiencing trauma. I can say very unpleasant things about my aunt’s behavior even though I simply can’t remember everything in detail (sometimes, I unknowingly add elements that are entirely false). I could also mix in my mind the behavior of my aunt with the conduct of one of my teachers in primary school. Sometimes we tend to construct ad hoc or even complicated explanations of what we have just done, rationalizing our (sometimes others’) actions or decisions. Nonetheless, it was argued that rationalizing is typically rational (Cushman, 2020), and some forms of confabulation may bring benefits to one’s self-identity (Bortolotti & Cox, 2009).

Confabulations also appear in clinical contexts. People lack uncertainty about something one should doubt, such as one’s impaired memory, one’s ability to move an arm, or one’s ability to see (Hirstein, 2005, p. 4). Prima facie, we see a difference between someone’s almost blind telling about what she is seeing and a person telling a colorful story about how beautiful she was 50 years ago. However, to justify an epistemological distinction between clinical and non-clinical confabulations is an arduous task; it seems to be impossible to draw it without incorporating further clinical facts, thus requiring us to notice or receive information about their other impairments. According to Hirstein (2005) and Coltheart et al. (2011), confabulations involve malfunctions in various knowledge domains, coupled with executive system damage.

Confabulations in KS are typically classified as belonging to the group of so-called memory-type confabulations, which are reports about one's memories. They are caused by memory loss and their content is related to memories.Footnote 2 Because of anterograde amnesia, KS patients cannot remember any new information and cannot keep memories, especially of their life's events, both chronologically and contextually. They tend to confabulate: They report events they didn't even experience, or recollect their actual experiences and thoughts from the past, reporting events they did have but often thinking those events happened further in the past than they actually did. Having also a kind of retrograde amnesia, such patients can recall events that occurred even further in the past. KS patients can also confabulate in such a way that they sometimes make up their statements on the spot. According to Sergei Korsakoff:

Thus, when asked to tell how he has been spending his time, the patient would very frequently relate a story altogether different from that which actually occurred, for example, he would tell that yesterday he took a ride into town, whereas in fact he has been in bed for two months, or he would tell of conversations which never occurred, and so forth (Korsakoff, 1955, p. 399).

Memory-type KS confabulations typically involve autobiographical narratives when a patient is recounting stories about their own life. According to Arts et al., they are ‟false or erroneous memories that arise involuntarily (i.e., non-deliberately) in the context of a neurological amnesia” or as failures in context reconstruction, traditionally dubbed ‟alignment disorder” (Arts et al., 2017, p. 2883).

However, as Bortolotti and Cox (2009) stress, defining confabulation remains a challenge. They provide an admittedly vague characterization: “Most typically, people confabulate when they make statements or tell stories which might be either inaccurate or badly supported by the available evidence. The ‘story’ is genuinely believed by the subject reporting it” (p. 952). While clinical literature typically assumes that confabulations are false, Bernecker (2017) points out that they could be true due to sheer luck, just like a broken clock telling the correct time twice a day. He also criticizes Hirstein’s (2005) approach to defining confabulation as ill-grounded. For instance, one’s true report of a memory can become ill-grounded because of some external manipulation, such as gaslighting. Bortolotti and Cox's characterization allows for the classification of true verbal reports related to memories as confabulations. However, it fails to distinctly separate confabulations from persistent delusions or genuine memory reports, given the vague condition implied by ‘might be’. As a lexical definition, it is simply too broad.

The recent decade witnessed a renewed interest in defining confabulation, as a focal issue in the debate between causal and simulationist accounts of episodic memory (for a recent overview of positions, see Sant’Anna et al., 2022). The causal theorists tend to account for confabulations and other memory errors in terms of inappropriate causal connection between the original event and the verbal memory report (Bernecker, 2017; Robins, 2016, 2020). Simulationist accounts frame confabulations in terms of unreliable acts of imagining the past (Michaelian, 2016, 2020, 2021), attempting to refine the conditions provided originally by Hirstein (2005). In Sect. 4, a teleosemantic alternative will be proposed, which accounts for memory in terms of producer–consumer networks.

3.2 Provoked and Spontaneous Confabulations in KS

Confabulations in KS appear in two formats: provoked and spontaneous. Provoked confabulations appear in response to direct questioning or probing a faulty memory. The content of provoked confabulations is traditionally believed to be rather plausible and connected much better to a patient’s personal habits. It seems stable and is likely to have occurred. In contrast, spontaneous confabulations appear without direct prompting.

It is often believed that provoked confabulations occur in patients with moderate memory loss and moderate cognitive skills, while spontaneous ones occur in patients with substantial memory loss and strong cognitive skills. Despite being controversial and debated (Coltheart, 2017; Metcalf et al., 2007), mostly because of the lack of standardized definitions, the overlap in symptoms, the varied clinical presentations, and inconsistent neuropsychological findings, I will preserve the provoked/spontaneous distinction to emphasize differences in how these confabulations arise. This distinction need not be viewed as a dichotomy, but as a spectrum. Following Coltheart, I simply assume that provoked confabulations occur only when people are questioned. Spontaneous confabulations require no probing, but they are present in amnesic patients, such as KS sufferers, much less frequently.

To showcase the distinction between provoked and spontaneous confabulation in KS patients, I will introduce two psychiatric cases: C.A. (Dalla Barba et al., 1990) and Mr. Thompson (Sacks, 1985).

3.2.1 Patient C.A.

C.A. was a 67-year-old woman who started drinking and, at the same time, reducing food intake after her husband's death, which resulted in severe memory disorder.

When asked about her whereabouts, she refused to acknowledge that she was in a hospital, claiming to be at home; alternatively, she said that she was at a hospital, but claimed to be only visiting a friend. Questioned about wearing a nightgown, she answered that she had to come to the hospital in a hurry and had no time to change in a more appropriate clothing. However confabulation was never produced spontaneously and appeared only when provoked by specific questions by the examiner. (Dalla Barba et al., 1990, p. 527)

Although neurologically almost intact, she was recognized as engaging in provoked confabulating. It seemed that C.A. primarily had a problem in finding herself in a new context. The researcher’s summary noted three separate types of confabulations: wrong responses (putting events in the wrong temporal context), invented responses (different whenever asked), and bizarre responses (enormous deviations from the factual truth). They attribute this behavior to a degraded representation of autobiographical memories (Dalla Barba et al., 1990, p. 533).

3.2.2 Mr. Thompson

The second example is taken from well-known book by Olivier Sacks “A man who mistook his wife with a hat” (Sacks, 1985). One of his patients, Mr. Thompson, suffers from KS and confabulates spontaneously as follows:

'What'll it be today?’ he says, rubbing his hands. ‘Half a pound of Virginia, a nice piece of Nova?’

(Evidently he saw me as a customer—he would often pick up the phone on the ward, and say ‘Thompson’s Delicatessen’.)

‘Oh Mr Thompson!’ I exclaim. ‘And who do you think I am?’

‘Good heavens, the light’s bad—I took you for a customer. As if it isn’t my old friend Tom Pitkins... Me and Tom’ (he whispers in an aside to the nurse) ‘was always going to the races together.’

‘Mr Thompson, you are mistaken again.’

‘So I am,’ he rejoins, not put out for a moment. ‘Why would you be wearing a white coat if you were Tom? You’re Hymie, the kosher butcher next door. No bloodstains on your coat though. Business bad today? You’ll look like a slaughterhouse by the end of the week!’ (Sacks, 1985, p. 108)

Let us now turn to the question of how to account for memory-type confabulation in an error-based framework.

4 A Teleosemantic Model of Memory

This section introduces a teleosemantic model of memory, conceptualizing it as a channel for encoding, storing, and retrieving information. Inspired by Peter Godfrey-Smith (2012, 2014), this approach analyzes memory faults, including KS confabulations, using a producer–consumer framework.

The model surpasses individualistic accounts by incorporating two key features: an evaluation subsystem for internal error detection and external consumers, such as caregivers or doctors, who can reinforce or correct memory reports. This distributed cognition perspective is crucial for understanding spontaneous versus provoked confabulations.

In the teleosemantic framework, memory functions like a communication channel for remembering, extending to hypothetical thinking and planning (De Brigard, 2014). The model comprises three core components. The producer encodes the initial experience, such as witnessing a vivid sunset, into a memory trace. The channel processes and transfers these memory traces, which may degrade over time. The model includes two crucial consumer subsystems. This is inspired by William Hirstein’s account of confabulation, in which he proposes a two-stage model:

  1. (1)

    Retrieval Failure : Errors or distortions within the memory traces arise from factors like decay, interference, or neurological damage, leading to distorted or fabricated memories.

  2. (2)

    Evaluation Failure: The evaluation subsystem fails to detect inconsistencies or implausibilities in retrieved information due to subsystem damage or lack of relevant information, resulting in accepted confabulations.

While causality is essential for information flow, it alone does not suffice for knowledge (Dretske, 1981). Memory errors are also causally driven, necessitating a specification of what makes the causal connection suitable for genuine memory. Prominent causal accounts (Bernecker, 2010; Martin & Deutscher, 1966; Robins, 2016) outline these additional conditions. Such an appropriate connection ensures that the current memory state counterfactually depends on the past representation. Memory traces serve as crucial intermediaries, preserving mental content and facilitating recall. Memory-type confabulations may seem accurate but lack the requisite causal connection.

Drawing from causal and simulationist perspectives, the teleosemantic account identifies five conditions for reliable information flow from past experiences to present memory.

  1. (1)

    Origination: The past representation must originate from the individual’s own experience. Information acquired through other means, such as testimony or externally implanted signals or representations, does not qualify as genuine memory. This condition emphasizes the personal and experiential nature of memory.

  2. (2)

    Authenticity: The memory must accurately reflect the individual’s past viewpoint. This ensures that the memory is a faithful representation of the individual’s past experience.

  3. (3)

    Veridicality: The memory should correspond to objective reality, ensuring it is a reliable source of information about the past.

  4. (4)

    Absence of External Manipulation: The memory should not result from external manipulation or implantation. Genuine memories arise from the individual’s cognitive processes, free from external influences that could distort or fabricate the memory.

  5. (5)

    Continuity of Memory Trace: A continuous memory trace, internal to the individual’s cognitive system, must link the past representation to the present memory state. This trace, often conceptualized as a series of neurophysiological changes, ensures the preservation and transmission of information over time.

  6. (6)

    Integration with Existing Memory System: The memory should integrate coherently (through coherence-based error detection) with the individual's existing memory system, supporting a consistent and reliable narrative of past experiences.

The first condition draws from Martin and Deutscher (1966), while the other conditions are inspired by Bernecker (2010. These conditions ground memory in past experience and ensure reliable information flow (see Fig. 1). Memory-type confabulation in the teleosemantic account can be thus defined as a memory state that is inauthentic, violating at least Authenticity or Origination. Although discussed in the context of episodic memory, this model plausibly applies to any memory system.Footnote 3

Fig. 1
figure 1

A teleosemantic model of memory. Upper legend boxes indicate parts of the information channel ensuring conditions are met (linking conditions to particular parts of the information flow system with dotted lines). Arrows show the direction of information flow. Confabulation occurs in inauthentic memories, while misremembering (Robins, 2016) occurs in authentic but non-veridical memories. The diagram does not depict the distributed architecture of memory traces (De Brigard, 2014). Retrieval subsystem and evaluation subsystem likely involve a heterarchical network of processes, idealized here for simplicity

Full size image

Simulationist accounts, like those of Kourken Michaelian (2016, 2020, 2021) emphasize the constructive nature of remembering, viewing it as a process reliant on reliable imaging of the past. However, reliability requires more than construction or internal coherence; without a connection to the past, purely constructive processes can produce inaccurate memories. The teleosemantic model's five conditions for appropriate information flow and evaluation subsystems ground constructed memories in individual experiences, preventing mere fabrications.

The crucial question is what makes episodic constructions reliable in the simulation account. Michaelian accounts for reliability in terms of proper functioning and appeals to meta-cognitive monitoring, but he does not explicitly provide further conditions, denying only that the appropriate causal connection is necessary. He argues that empirical research on the constructive nature of remembering and memory as mental time travel suggests that appropriate causation is not necessary for remembering (Michaelian, 2021, p. 7483). Remembering can occur even when a minority of the retrieved content is new.

The teleosemantic model does not need to reject this idea, as doing so would imply our memory system is highly unreliable. Memory traces need not be completely preserved; cognitive processing may involve complex transformations of various representations. Generalization from experience is crucial but fallible. Adaptive considerations involve trade-offs between memory accuracy and applicability in guiding actions. Metacognitive monitoring subsystems exist because substantial cognitive processing ensures recollection accuracy.

These subsystems need not be internal. The model may include the social dimension of memory by incorporating external consumers such as caregivers, doctors, and family members. This integration of metacognitive monitoring and distributed cognition (Hutchins, 1995) is supported by the work of Nelson and Fivush (2004), who emphasize the social construction of autobiographical memory. Interactions with others shape, reinforce, and correct our memories. For instance, caregivers and doctors can validate or challenge KS patients' memory reports, impacting their memory processes.

External consumers play a dual role. They can inadvertently reinforce confabulations by accepting and validating erroneous memory reports. Cotard’s syndrome case studies show that caregivers who consistently accept a patient's confabulated account can strengthen that false memory, making it more resistant to correction (Berrios & Luque, 1995). For example, if a KS patient repeatedly mentions a nonexistent family gathering and the caregiver does not challenge this, the patient may believe the confabulation more strongly.

Conversely, skepticism or corrective feedback from external consumers can trigger the patient's internal error detection mechanisms, potentially leading to a revision of the confabulation. Clinical observations show that when a doctor gently challenges a patient's confabulated memory with contradictory evidence, it can activate the patient's sense of inconsistency and prompt re-evaluation. This corrective function highlights the importance of metacognitive prompts and social feedback in therapeutic interventions for KS.

The interplay between internal cognitive processes and external social interactions is crucial for understanding memory formation and correction, especially in KS, where memory systems are compromised. Incorporating external consumers into the teleosemantic model provides a more comprehensive account of memory and confabulation than individualistic models.

Beyond KS, the teleosemantic model holds potential for understanding other memory distortions, such as misremembering and rationalization, and various psychopathologies. Its emphasis on reliable information flow and conditions for genuine memory sets the stage for exploring how representation errors lead to confabulation, as detailed in the following section.

5 Heuristics of Error-Based View on Confabulation

The error-based perspective yields powerful heuristics by requiring that mechanisms for error detection be specified. This usually necessitates further research into the representational phenomenon.

KS confabulations are linguistic expressions with accuracy or satisfaction conditions. There is an entrenched analogy between linguistic meaning and representational content (Pitt, 2013). Traditionally, both mental representation and linguistic expressions have intension (meaning) and extension (truth conditions). This analogy is justified by the correspondence of language and thoughts to the world; both relate to reality and describe it. The simplest explanation for a given narrative being representational is that it was produced by an agent with mental representations. The received view is that people produce narratives because they have internal cognitive representations. Similarly, producing false narratives can be explained as the result of false mental representations.

The coherence-based account of a system-detectable error (Bielecka, 2019; Bielecka & Miłkowski, 2020) can further enrich the account of confabulations. The adequacy of mental representation is available to the cognitive system as long as it can detect informational inconsistency. To detect that representation A is inconsistent with B, a cognitive system must register A's inconsistency with B, and B must be available to the system. For example, in the case of Mr. Thompson, the cognitive system recognizes the incoherence between the concepts ‟doctor” and ‟butcher” and tries to correct for the concept ‟butcher” by relying on perceptual information and memory.

Error detection can fail due to malfunctioning consumer mechanisms. The biological function of the consumer mechanism—including in KS patients—is to detect the error, for example, by comparing representational vehicles. Some KS patients can recover from error and do not confabulate on all occasions, showing that the malfunction is often incomplete. Representations are still tokened because their producers have the biological function even when the function can no longer be appropriately served.

The error detectability condition seems plausible and provides additional heuristic guidance for developing the account of KS confabulations. It leads to hypotheses regarding cognitive processing failures because it requires specifying the roles of particular cognitive mechanisms involved.

Teleosemantic heuristic questions.

  • What is the producer?

  • What is the consumer?

  • Did they evolve to cooperate?

  • What is the mapping function?

Heuristic questions of the coherence-based approach.

  • What is the mechanism of coherence checking?

  • When does the mechanism fail?

  • How reliable is coherence checking?

  • Is error detection biased in favor of some source of information?

The model introduced in the previous section leads to two possible scenarios for error detection failures. One scenario is that the consumer part of the cognitive system of the patient has a strongly impaired capacity to evaluate the satisfaction conditions of representation and correct for the error. Confabulations fill in retrieval gaps, and inconsistencies remain unnoticed by the patient. This is still consistent with an individualist perspective on confabulations.

A non-individualist extension is however more plausible, where the consumer is not entirely part of the patient but could be another person or source of information (e.g., a newspaper). This leads to an account of confabulation along the lines of distributed cognition. Both scenarios are possible and correspond to traditional conceptual distinctions in confabulation research:

Internal Scenario (fits spontaneous confabulations):

  • A patient spontaneously tells stories that are not further evaluated or acted upon.

  • Mr. Thompson quickly discards previously concocted stories as untrue when probed.

Distributed Scenario (fits provoked confabulations):

  • Triggered by patient situations or external help.

  • External consumers, like doctors, become active components of the remembering process.

Let us focus on the internal scenario. A patient can spontaneously tell stories that are not further evaluated and often not acted upon. If no sign is given by others that helps the patient momentarily recognize and momentarily forget that something is wrong, then stories will be continuously fabricated. These are usually not entirely inconsistent with the patient’s knowledge, but as noted by Dalla Barba et al. (1990), they are at times quite bizarre. Although his capacity to evaluate memory representations is impaired, most confabulations of Mr. Thompson are put in an incorrect spatiotemporal context (butchers do exist, but they do not work in hospitals, etc.). When probed, however, Mr. Thompson quickly discards previously concocted stories as untrue (“Good heavens, the light’s bad…”).

The distributed scenario seems to fit both provoked confabulations triggered by the patient situation (such as C.A.) and patients with spontaneous confabulations who obtain some help from others (as was shown in Mr. Thompson's example). The external consumer should be understood as encompassing the other people in the patient's environment, such as the doctor in the case above. The doctor is the external consumer in this scenario. The idea is that the doctor is not just another stimulus for the patient (like the doctor's coat, the hospital corridor). Signalization of the doctor in response to the lack of stimulus or patient's erroneous representing is crucial for the patient's attention, and attention allows him to respond to his problems, in turn allowing for the doctor to serve the function of his external consumer. The doctor becomes an active component of the remembering process.

Both individual and distributed processes, just like provoked and spontaneous confabulations, can occur in the same patient. For example, Korsakoff described a patient who could correct herself but still experienced memory loss leading to confabulations. This case deserves citing at some length:

However, in a long conversation one could note anomalies in the psychic sphere. These anomalies chiefly concerned the memory. Thus, in telling of something about the past, the patient would suddenly confuse events and would introduce the events related to one period into the story about another period. For example, telling of a trip she had made to Finland before her illness and describing her voyage in fair detail. The patient mixed into the story her recollections of Crimea, and so it turned out that in Finland people always eat lamb and the inhabitants are Tatars. When I objected, however, she promptly agreed that she had confused the facts. Such muddling of facts not infrequently occurs in this patient, but in most instances she herself notices them. Furthermore, she frequently gives false information in regard to her past illness. Thus, she assures me that she well remembers a physician who came to see her in consultation and describes him as having black hair, whereas he is completely gray. In general the patient does not remember clearly that which occurred during the time of the illness, although gradually the events of her illness become restored in her memory. After a conversation lasting an hour, the patient becomes very listless; one can see that she becomes extremely fatigued mentally. Fatigue in general has a marked effect on the patient, increasing the disorder of memory (Korsakoff, 1955, p. 404).

In contrast to Mr. Thompson, as described by Sacks, this patient was often able to correct herself—or at least notice inconsistencies (“in most instances she herself notices them”). Unfortunately, the memory loss made her incapable of recovering from error and this is where confabulations started.

Note one striking feature of the cases of Mr. Thompson and Korsakoff's patient: they both “promptly agreed” when corrected. This is rarely the case with our everyday cognitive functioning, where an external correction may be treated with suspicion or even outrage. KS patients may treat other people as reliable parts of their cognitive processes due to their suggestibility and inability to orient themselves in reality otherwise, contrasting with delusions typically characterized by persistence. However, external correction may be the only way to compensate for the fallibility of our individual memory processes. Our own memory processes may be simply insufficient to detect how inaccurate our representations are.

The error-based account can make sense both of extreme memory disorder such as in Mr. Thompson's case, and less severe cases as C.A., not to mention one of the original examples provided by Korsakoff. This shows the flexibility of this kind of representational explanations of KS. Both internal cognitive dynamics and reliance on social interaction can be explained similarly. This is because a teleosemantic approach to representation need not be confined to a single individual. After all, distributed systems, such as a bee signaling the location of the nectar to other worker bees, are classical cases for teleosemantics (Millikan, 1984). In other words, one of the strengths of teleosemantics is that it is scale-free, i.e., it need not be restricted to a particular kind of entity in order to claim that it has a representational function. For example, producers and consumers could be posited at a sub-personal or supra-personal level. This increases the flexibility of representational accounts of psychopathologies. Nonetheless, the error-based perspective naturally aligns with the distributed account of cognition, which also includes possible failures and disorders: There should be some mechanisms of error detection. In our cognitive processing, including the development of episodic memory (Nelson & Fivush, 2004), these frequently rely on the feedback from other agents or artifacts.

To summarize, the error-based perspective offers a coherent picture of what needs to be studied to fully comprehend confabulation. By relying on a teleosemantic model of memory, it provides a novel representational account of confabulation and suggests that the same model could be applied to other kinds of cognitive information flow in psychopathology.

6 Possible Objections

In this section, I briefly discuss two possible objections.

How are mind and language linked exactly? The take on KS in this paper provides both an internal and distributed account of confabulation. One could object that the internal processes were not integrated with social ones and that building the bridge is challenging.

While it is true that recent work on teleosemantics focuses primarily on sub-personal representations (Shea, 2018), the classical position is not committed to mental representation (Millikan, 1984). Teleosemantics does not posit a specific entity or spatiotemporal scale solely to provide representational contents, making it flexible and highly generalizable.

At the same time, the interaction of various communication systems (or producer–consumer systems) is both a theoretical and empirical issue. Hopefully, future empirical evidence will go beyond mere analogy between linguistic and mental contents and elucidate more deeply how these are related. It is also important to note that non-linguistic external representations play a significant role in autobiographic memory. This is particularly striking for people with severe dementia, who rely on external artifacts (called ‟evocative objects") to compensate for memory loss (Heersmink, 2022). However, intertwining various kinds of representation is beyond the scope of this paper.

Admittedly, the account presented here is only a brief outline of a representational approach to confabulation. Much additional empirical work is necessary to establish comprehensive cognitive theories explaining why specific confabulations occur, from ordinary to unusual ones. The established differentiation between spontaneous and provoked confabulations appears to correspond to two kinds of error-detection mechanisms, but we lack evidence to explain why KS patients are susceptible and readily accept corrections.

Existing accounts of memory should not be hastily discarded. My intention in this paper is not to offer a detailed critique of well-developed philosophical accounts of memory. The teleosemantic model takes inspiration from earlier work on sender-receiver dynamics as well as causal and simulationist accounts, identifying shared assumptions. The core arguments for the view are (1) that causality alone is insufficient to specify what makes memory accurate (memory errors are caused, after all), and (2) that the reliability of simulation must also be explained. The only plausible explanation seems to require a reliable information flow. Surely, some philosophers may find these arguments implausible and the representational framework too traditional.

7 Conclusions

The error-based teleosemantic approach to confabulation presented in this paper offers several unique insights and benefits compared to causal and simulationist theories of memory:

  1. (1)

    It provides a unified framework for understanding both internal cognitive dynamics and the role of social interaction in shaping confabulations. By conceptualizing memory as a communication channel with producers, consumers, and error-detection mechanisms, the model can flexibly account for spontaneous confabulations arising from individual memory deficits as well as provoked confabulations influenced by feedback from others. This integrated perspective advances beyond the focus of previous theories on purely internal memory processes.

  2. (2)

    The error-based approach inspires novel hypotheses about specific breakdowns in cognitive processing that give rise to confabulations. It predicts that confabulations occur when error-detection mechanisms fail, either due to individual impairments in evaluating retrieved information or disrupted social feedback loops. These hypotheses provide clear direction for future empirical research to pinpoint the nature of the underlying deficits. In contrast, causal and simulationist theories offer vaguer explanations in terms of inappropriate causation or unreliable simulation.

  3. (3)

    By grounding the analysis of confabulation in a systematic theory of mental representation, the error-based view connects the phenomenon to broader issues in philosophy of mind and cognitive science. It shows how confabulations, as representations, depend on the proper functioning of interconnected producer–consumer systems. This allows for principled criteria to distinguish confabulations from accurate memories.

  4. (4)

    The error-based perspective highlights the indispensable role of metacognitive monitoring and social interaction in regulating memory. Its detailed analysis of error-detection dynamics in Korsakoff patients demonstrates how self-monitoring and responsiveness to correction crucially influence the severity of confabulation. This underscores the importance of these regulatory processes, an insight downplayed in alternative theories focused narrowly on encoding and retrieval.

  5. (5)

    Methodologically, the error-based approach equips researchers with powerful heuristics to guide the empirical study of confabulation across clinical contexts. By mapping the components of the teleosemantic model onto real-world cases, it reveals which aspects of cognitive processing to target for further investigation. This systematic research framework lends itself to discovering commonalities and differences in the mechanisms of confabulation across neuropsychological conditions.

In summary, the error-based teleosemantic account provides a theoretically principled, empirically fruitful, and unifying perspective on confabulation. It integrates internal and social levels of analysis, inspires testable hypotheses, connects to broader issues in philosophy of mind, highlights key regulatory processes, and offers a powerful framework for guiding research. These features mark significant advances beyond the scope and explanatory power of previous philosophical theories of confabulation.