How to define metacognition?
Representational dimension: Meta-level and object-level
The first models of metacognition focused on the idea that an internal observer allows the self-perception of performing a cognitive activity. Therefore, they assumed that metacognition has a representation dimension: the observer has a “model” of the current activity supposed to be “beyond” this activity. Nelson and Narens (1994)’s model of metacognition (1990) developed in the context of memory functions is a good example of such a representational view of metacognition. These authors distinguish two levels of processing. The object-level refers to the level of the cognitive activity (e.g., learning a maths lesson) whereas the meta-level is a representation of this activity which can be formulated in a propositional and self-referenced way (e.g., “I am very bad at maths”).
To communicate, these two levels are linked by two types of metacognitive processes.
Monitoring processes stem from the object-level and inform the meta-level about the actual state of the cognitive activity (e.g., ‘This particular equation is difficult to understand”).
Once the meta-level has been updated, control processes have a direct action on the cognitive level by the use of adaptive strategies to improve the cognitive activity (e.g., “I will spend more time on this problem”). According to such a view of metacognition, there is a behavioural control loop (the monitoring-affects-control hypothesis (Nelson & Leonesio, 1988)) that involves an explicit evaluation of the cognitive activity and a declarative implementation of strategies. An important feature of this model is that the meta-level and the object-level can operate simultaneously, which proposes that metacognition is independent from cognitive processes.
Astriking example of this representational and dissociative structure is a phenomenon known as the Tip-Of-the-Tongue (TOT; Brown & McNeill, 1966). The TOT is a state in which one cannot quite recall a word but has the knowledge that he or she knows the particular word. This phenomenon is therefore an experience of dissociation between the object- and the meta-level: the self-knowledge (i.e., “I know what the capital of Australia is”) is in contradiction with the cognitive performance (i.e., not recalling the answer). Other dissociations are known in the memory literature such as the déjà vu phenomenon mentioned above and where there is a “phenomenological experience of recognizing a current situation and the awareness that this feeling of recognition is inappropriate” (p.2; O’Connor & Moulin, 2010). Another practice example of the objectand the meta-level is given by Nelson (1996). In the sentence “Thiss sentence contains threee errors”, the spelling of “this” and “three” refer to two object-level errors and the meaning of the sentence represents a meta-level error. Thus, to be aware of the meta-level error, one must be aware of the object-level error suggesting a hierarchical structure in metacognition. However, as mentioned by Nelson (1996) after realizing the meta-error, one can realize that “there is a total of three errors after all” leading to the idea of a structure containing more than two levels with different monitoring-control loops. Metacognition has therefore a representational function that paradigms tend to measure using behavioural observation. The majority of research has been conducted on monitoring processes probably as they are seen as a starting point of the loop: there is a need for a self-evaluation to perform an adaptive behavioural control.
Behavioral dimension: First-order and second-order
Although a classical definition of metacognition involves a representational dimension of a cognitive activity, research has also shown the involvement of information directly stemming from cognitive activity (i.e., the object-level) without necessarily involving another level of representation. Traditionally, inferences about monitoring processes have been measured using subjective reports of introspection known as metacognitive judgments Nelson, 1990. These judgements are extensively used in the laboratory to study metacognition and refer to “decisions about decisions” (or “behaviour about behaviour, see Fleming et al., 2012). Experimentally, metacognitive judgements occur when two decisions are made: one known as first-order response refers to an answer in a cognitive task (e.g., to select with stimulus in the brightest) and the other or secondorder response is a metacognitive judgement related to the first-order response (e.g., was the previous response correct?). Several metacognitive judgements1 exist in the literature but the most commonly used are Retrospective Confidence Judgements (RCJs) which are an evaluation of the level of confidence in a previous decision.
In the memory literature, these judgements are seen as a measurement of monitoring processes. However, the “second-order” terminology is now preferably used since metacognition is studied in various cognitive domains beyond the memory domain (e.g., visual perception, decision making, reasoning, motor function, etc.). Moreover, it is assumption-free as second-order judgements can be performed using a range of processes (see Chapter 2). On one hand they can be the result of introspection by the use metacognitive representations as a model of the current cognitive activity. Conversely and on another hand, they can use information directly stemming from the object-level without necessarily involving another level of representation. A good example of such absence of meta-level representation derives from the notion of uncertainty. Uncertainty can be defined as the variability or the reliability of a representation, a stimulus, or an outcome (Bach & Dolan, 2012). A certain amount of work has shown that uncertainty can be used in order to control behaviours such as modifying decision-making therefore acting as a metacognitive component. In some case, these adaptive behaviours do not necessarily imply the existence of a meta-level in both human and non-human animals2 (Kepecs & Mainen, 2012).
From this perspective, the existence of metacognition is suggested by the fact that one is able to produce accurate decisions. Metacognition has a function of evaluation of cognition with the idea that an appropriate evaluation will lead to an adaptive behaviour.
In this sense, another important aspect is error detection; the ability to be aware of our own mistakes. This has been especially studied in the context of reaction time
tasks when participants have to make decisions under time pressure which leads to an increase the number of errors. Although initially studied separately, models and methods of confidence formation and error detection shared similarities leading to the idea that they are two outcomes underpinned by common metacognitive processes (Yeung & Summerfield, 2012). Here again, quantifying confidence or detecting an error have been explained as being based on the reliability of the signal during the decision process (akin to uncertainty) which does not necessarily need the involvement of a meta-level representation (see Section 2.1.2 “Low-level metacognition and implicit knowledge” and for a philosophical about the involvement of meta-representations see Proust, 2007, and see section 2 of this Chapter).
Consciousness dimension: State of awareness
Nelson and Narens (1994) suggest that the main method for investigating monitoring is to ask people to give the output of their introspection. Although metacognition and introspection can overlap, they also differ to some extent. Overgaard and Sandberg (2012) propose a useful distinction between the two: a classification from the subjective point of view or from the functional point of view. Whilst both are second-order processes, metacognition is about cognition (as a function that can be either conscious or unconscious) and introspection is restricted to a conscious state. In this sense, they both have a higher-order function and one can see introspection as “a special case of metacognition” (p.1288, Overgaard & Sandberg, 2012). This suggests that metacognition involves conscious self-reflected mechanisms but also others processes that are not present during introspection.
Whilst it is beyond the scope of this thesis to propose explanations and a complete account of consciousness, it is important to note that the concepts of consciousness and metacognition are relatively close. For the purposes of this thesis, consciousness has been conceptualised as the classical notion of “having subjective experiences” (Frith, 2019). Similarly, metacognition and particularly metacognitive judgements (see next section and Chapter 2) are self-evaluationwhich also has a subjective value. Thus, these judgements can be seen as reports of the content of subjective experiences and are even used as a measure of consciouness (e.g., Seth, Dienes, Cleeremans, Overgaard, & Pessoa, 2008). Some implications for our understanding of consciousness will be discussed in Chapter 7, however it is not a direct aim of this thesis to use metacognition to examine the nature of consciousness.
Although there is a close relationship between metacognition and consciousness, these are nonetheless dissociable to some extent. Metacognition refers to consciousness in terms of the “access of consciousness” that is the ability to be aware of conscious states as a knowledge but do not account for “the phenomenology of consciousness” that refers to subjective experience as a feeling (Block, 2011). It suggests that one can be aware of a state (e.g., I am in a sad mood, I am seeing a green colour) whilst having no metacognitive process involved in this state of awareness (e.g., Am I sure this is green?). Similarly, if consciousness does not necessary imply metacognition, several works have also shown that metacognitive behaviours can occur at the fringe of consciousness.
To investigate this issue, an interesting study case is blindsight. The blindsight phenomenon occurs when patients are able to detect and identify visual stimuli during forced-choice experiments in the absence of conscious awareness of these stimuli when they are presented in the blind visual field (e.g., Poppel, Held, & Frost, 1973).
Table of contents :
1 The diversity of metacognition
1.1 How to define metacognition?
1.1.1 Representational dimension: Meta-level and object-level
1.1.2 Behavioral dimension: First-order and second-order
1.1.3 Consciousness dimension: State of awareness
1.2 Different forms of metacognition
1.2.1 Direct tests of metacognition
1.2.2 Indirect tests of metacognition
Evidence in human animals
Evidence in non-human animals
2 A comparison of the field of metaperception and metamemory
2.1 Methodological comparison
2.1.1 Measures of metacognition: direct tests
2.1.2 The quantification of metacognition
2.2 Theoretical comparison
2.2.1 Relationship between metacognition and task performance
Models of confidence formation
Theoretical accounts of metamemory
2.2.2 Dissociation between metacognition and task performance
Behavioral and neuronal evidence
Post-decisional and second-order models of confidence
Cue-utilization in metamemory judgements
3 Introduction to empirical chapters
3.1 Investigating metacognition across domains
3.1.1 The domain-generality and domain-specificity debate
3.1.2 Domain-generality in metacognition
3.2 Methodological considerations
3.2.2 Material selection
3.2.3 Data and statistical modelling
3.3 Aims of the thesis
4 Is there aGfactor for metacognition? Correlations in retrospective metacognitive sensitivity across tasks
4.2.2 Materials and procedure
4.2.3 Data and statistical analyses
4.3.1 Type 1 performance
4.3.2 Confidence level
4.3.3 Metacognitive efficiency
4.3.4 Supplementary results
Comparison between hierarchical and non hierarchical models .
5 Metacognitive domain specificity in FOKs but not RCJs
5.2.2 Materials and procedure
5.2.3 Data and statistical analyses
5.3.1 Type 1 performance
5.3.2 Metacognitive bias
5.3.3 Metacognitive efficiency
5.3.4 Exploratory analyses
6 Investigate ease-of-processing in metacognition formemoryand visual perception
6.1 Experiment 1: Perceptual fluency induces by prior exposure effects on metacognition for visual perception
Material and procedure
Data and statistical analyses
Relationship with recognition memory
6.1.4 Brief discussion
6.2 Experiment 2: The effect of conceptual fluency on metacognitive judgements
for visual perception
Material and procedure
Data and statistical analyses
6.2.4 Brief discussion
6.3 Experiment 3: Theword heuristic as conceptual fluency effect on metacognition
for episodic memory
Material and procedure
Data and statistical analyses
6.3.4 Brief discussion
6.4 General discussion
7 General discussion and perspectives
7.1 Summary of results
7.2 Implications for research on metacognition
7.2.1 Experience-based and information-based metacognition
A proposal for metacognitive judgement formation
7.2.2 Metamemory and metaperception
Using the meta-d framework in memory and perception tasks
7.2.3 Implications for dimensions of metacognition
The representational dimension
The behavioural dimension
The consciousness dimension
7.3 Other implications
7.3.1 Implication for memory research
7.3.2 Implication for neuropsychology and psychiatry
A A review of metacognition and self-awareness in Multiple Sclerosis
B A multidimensional assessment of metacognition across domains in Multiple Sclerosis
C In the here and now: Short term memory predictions are preserved in Alzheimer’s disease