Reviewing a Model of Metacognition for Application in Cognitive Architecture Design

Abstract

This systematic review answers questions about whether or not a model of metacognition is well accepted and if it can be used in cognitive architecture design. Self-planning, self-monitoring, and self-evaluation are the model concepts, which are viewed as metacognitive experiences. A newly formulated theoretical approach named Attention as Action was targeted, as it is shown to be used in cognitive architecture design. In order to link the model to the theoretical approach, specific concepts like mental imagery and learning experience were researched. The method includes the statistical analysis of key phrases in articles that were collected based on a system of criteria. Data were retrieved from 91 scientific papers to allow statistical analysis of the relationship between the model of metacognition and the theoretical approach to cognitive architecture design. Several observations from the data show that the model is applicable for designing cognitive monitoring systems that depict experiences of metacognition. Furthermore, the results point out that the researched fields require explanations about the concepts defined in the theoretical approach of Attention as Action. Systematically formulated as types of internal attentional experiences, new relations are provided for researching cognitive and metacognitive concepts in terms of the cognitive cycle.

1. Introduction

The initial idea of this work was to investigate studies that support a model of metacognition that can be used as a submodule of a cognitive system. A recently emerged theoretical approach was formulated via a semiotic technique for designing cognitive architectures [1]. It is based on the aspect of personality as an internal agent that executes internal actions simultaneously with the information provisioning of an automatic unconscious process. The approach follows established theories about the timing of the cognitive cycle [2], mental and perceptual imagery experiences [3], and action schemas [4]. As a guiding cognitive architecture, the Internal Decision Model of Attention [1] defines four production rules in terms of three process layers: perceptual–motor, interpretational, and metacognitive. With the requirement of designing a specific, detailed cognitive architecture, our team targeted the metacognitive process layer to produce a model of metacognition. Such a model needed to be definable in terms of the theoretical approach of Attention as Action. More precisely, a combination of definitions of metacognitive experiences were required that corresponded to the formulated concept of internal action and the production rules of the guiding cognitive architecture [1].

A target was set to investigate a model that defines three dimensions of metacognition: self-planning, self-monitoring, and self-evaluation. These have been defined as aspects of metacognitive regulation in the science of learning [5], and have also contributed to developing metacognitive strategies in game play [6]. In addition to its variety of existing applications, our team suggests another utilization of this model—for designing cognitive architectures.

In this systematic review, we address studies on metacognition that satisfy a set of requirements suggesting the possibility of a model for the three defined dimensions. It was accepted that the approach of searching by specific keywords and key phrases could confirm the establishment of the model in scientific fields. The content of this article first presents the theory and the questions that need to be confirmed or rejected. Next, the method and the rules are explained. After that, the results are provided and analyzed. Finally, the results are used as a base to generate a discussion, with the aim of providing accumulative definitions and a conceptual model based on the adopted theoretical approach.

2. Theoretical Views

The three aspects of metacognition—planning, monitoring and evaluation—appear in studies from different scientific spheres [5,6,7]. If a clear understanding of them is formulated in systematic terms, directions might be provided for designing conceptual models of cognitive architectures that explain experiences of metacognition. Furthermore, these models would serve to build a common understanding about learning and self-awareness that would be applicable in cognitive science and computer systems to mimic cognition. A general conceptual model can be established to serve as a framework, upon which hypotheses can be built about types of system states, modules, and relations between them. This way, cognitive systems can be designed that can monitor and predict possible cognitive experiences. Also, computer systems could be created to beneficially prompt users towards metacognition in order to mitigate cognitive biases in crisis situations [1]. The initial question that has to be accepted or rejected is as follows: is the defined model of metacognition well established? This is investigated by searching for the three model keywords and key phrases, and their appearance together in sentences. However, a more complex task is to confirm whether or not the model can be applied in cognitive architecture design.

A recent article presents a new theoretical approach as an application for explaining different cognitive phenomena, like action outcomes, conflict processing, and crisis experiences [1]. What is more, the theoretical approach, namely Attention as Action, is shown to be suited for cognitive architecture design via a semiotic representation technique. The study provides explanations for how a metacognitive experience (ME) is related to other cognitive phenomena, but does not specify types of MEs.

An adopted understanding of online and offline task awareness [8] provides explanations for metacognitive knowledge as opposed to MEs. This understanding positions metacognitive knowledge as a personal comprehension of one’s own skills and thinking, associated with experiences that occurred in the past. On the other hand, MEs are viewed as experiences of self-awareness that occur in the now. Such experiences include feelings of knowing, of familiarity, or of confidence [8]. Other examples of MEs are self-judgements of learning: the phenomenon of the personality, thinking about how well it knows a targeted study material, or experiences of self-evaluation about how well an ongoing task is being performed [8]. This dichotomy provides an understanding about how the information that is produced by metacognitive phenomena is related to time. This means that if this understanding is linked to the targeted model of metacognition, a more specific classification could be formulated of types of MEs.

Metacognitive planning, monitoring, and evaluation are most often viewed as regulation skills [5,9]. On the other hand, metacognitive knowledge and metacognitive regulation exist in dichotomic interrelation [5,10]. Therefore, if an ME is viewed as a dichotomic opposition to metacognitive knowledge, then an ME can be addressed to metacognitive regulation. Thus, the three target model definitions can be suggested as types of MEs. If this view finds support by the results of this systematic review, the target model would fit with the idea that a consciousness is achieved by a consistent sequence of conscious experiences [11]. This idea is further explained in terms of the cognitive cycle [2], which includes several types of learning, all of which are viewed as conscious experiences. Envisioned as an act of self-awareness, in terms of the cognitive cycle, a metacognitive process that gives rise to an ME should be viewed as a conscious process. Therefore, the concept of conscious experience [11], which is tightly related to the cognitive cycle, corresponds to an ME.

Although an ME might be described as unconscious in some rare contexts [12], the adopted theoretical views of this work define a conscious experience as an event of information learning, of which the personality is aware is occurring in the now. Furthermore, the personality is aware of the information that is being learned during a conscious experience. This implies that unconscious information is retained by the personality, but the latter is not aware of it in the current moment. In this context, a new concept emerged in order to refer to processes that take place in cognition without the awareness of the personality: the automatic unconscious process [1]. This concept, along with related ideas, will be further elaborated in the next subsection.

Another demanded confirmation is required to link the target model of metacognition to the concepts of perceptual and mental imagery experiences [3]. This means that specific keywords need to be included in the investigation in the search for relations between metacognition and imagery experiences. Some studies exist that build strong relations between metacognition and mental imagery experiences [13,14]. Described as “the retrieval of perceptual information from memory, and the subsequent examination of this information”, the metacognitive understanding of mental imagery is reported to be a subject that can be improved with practice [14].

An emerging theoretical approach, namely Attention as Action, has been shown to be used for explaining short cognitive phenomena, like crisis phenomena, via systematic conceptualizing techniques [1]. The approach provides explanations of cognitive experiences that are structurally described in the following subsection. Attention as Action defines the concept of internal attentional experience as a short conscious experience that is emitted by the execution of a conscious internal processing of information. This is the main concept that links the theoretical approach to the understanding of ME as an occurrence of metacognitive regulation.

Together with the presented scientific topics, the theoretical approach is required to explain the concept of ME via its semiotic technique for the conceptualization of conscious internal processes. Conceptually, the idea of linking the presented topics in order to achieve a cognitive architecture design with the current approach is presented in Figure 1. In terms of Attention as Action, an internal attentional experience is used to explain varying cognitive phenomena; however, the concept will find application in this work only for explaining the concept of ME. Nevertheless, this systematic review provides new ways to explain MEs in relation to other cognitive phenomena, such as imagery of objects; affect and schemata [15]; perceptual, episodic, and procedural learning [2]; action–outcome learning [16]; conflict processing [17]; and the crisis phenomenon [1]. What is more, the main reason for conducting this work is to provide knowledge about whether or not the listed cognitive phenomena can be explained in terms of the targeted theoretical approach.

2.1. Applied Theoretical Approach

Used as a recipe, the conceptual model in Figure 1 suggested the keywords and key phrase strings on which the conducted research should be based. Further using the conceptual model for guidance, inclusion and exclusion criteria of this systematic review were defined according to the following three specified demands. Firstly, it is important to gain knowledge from the contributing fields that are most fruitful in terms of research on metacognition; these are claimed to be educational psychology and the science of learning. Secondly, the growing fields of mental imagery and research on the cognitive cycle must be related to the model’s describing terms. Thirdly, together, the established knowledge (Figure 1) must be associated with the encapsulated understanding of the theoretical approach of Attention as Action (AaA). A guiding principle of the third demand is to explain cognitive phenomena as short experiences of internal attention produced by an internal agent. This way, they can be related to different system modules of cognitive architectures, and thus be combined in a hypothetical general conceptual model. What is more, several general models can be designed in search of the one that fits best with the understandings of modern science.

2.1.1. Automatic Unconscious Process

The referenced article hypothesizes that cognition collects and organizes relevant information that is “unconsciously produced” by an automatic unconscious process [1]. The proceeding of a process of gathering or organizing information is unknown to the personality. At the time of that process’s execution, the targeted information is outside of the conscious awareness. At some point in time, an automatic unconscious process (AUP) produces an internal event [1], signaling to the personality that relevant or urgent information is available for conscious processing. This is supported by the concept of competition for consciousness that occurs in the unconscious phase of the cognitive cycle [2]. It is explained that “novel, relevant, urgent, or insistent events” [2] (p. 5) are brought to consciousness. In the context of AaA, the competition of consciousness is explained by the phenomenon of several AUPs competing to provide gathered information to the personality. In a rapid manner (within milliseconds), the personality has to target an internal event. That is why, in terms of AaA, this process is referred to as internal decision-making and the personality is viewed as an internal agent [1].

2.1.2. Internal Action

The moment in which the personality targets an internal event, the AUP that produced it begins information provisioning. Simultaneously with this process, the internal agent begins the execution of an internal action, which produces a conscious attentional experience [1,2,11]. This follows the idea that an autonomous agent frequently makes sense of its environment and selects “an appropriate response (action)” [2] (p. 1). However, in terms of AaA, the internal agent operates in an internal environment and produces a response (internal action) as a reply to an internal event—one that is produced by an AUP.

The presented theory is depicted in Figure 2. An important note is that the conscious experience that occurs after targeting an AUP is happening in parallel to the execution of the internal action. From here on, new relations can be built by linking processes recognized as automatic and unconscious from cognitive architectures like LIDA [2]. Also, types of visual mental imagery experiences [15] can be linked to the process of internal action execution (Figure 2). This way, new explanations will be provided for how a specific AUP is related to a mental imagery type. What is more, interconnections can be sought between types of mental imagery, for example, by following the model of the Action Cycle Theory [3]. Such an example can already be observed in another work that applies the AaA approach to explain such interconnections via cognitive architecture design [18].

A degree of conscious awareness could be explained by reasoning about internal actions that can be either automatic or volitional. This leads to the concept of automatic conscious processes being related to volitional conscious processes, both of which are viewed as internal action modes. Such concepts can construe mental states by explaining that one has limited concentration when the internal agent targets internal events automatically. Opposingly, a high concentration can be explained by targeting internal events volitionally in the internal decision-making process.

2.1.3. Cognitive Architecture Design

A model has been provided that explains how internal action (IA) types can be classified according to metacognition [1]. Namely, the Internal Decision Model of Attention (IDMA) presents three layers that correspond to the levels presented in the model of metacognition [6]. The latter defines an object level, a meta level, and a level of flow of information. The meta level conforms to layer 3 in IDMA, which expresses that when the internal agent is in this layer, no volitional motor operation can be executed [1]. A suggestion of this work is that a volitional motor operation is viewed as a body action signal that is emitted from the brain in order to produce a body action. Such a signal corresponds to the propagation of action potentials from the motor cortex, which evokes hand muscles for 19–24 milliseconds [2] (p. 9). In terms of AaA, this work proposes that an emission of a volitional body action signal starts with the execution of a particular IA—the motor IA. This IA also corresponds to the deed IA used to explain volitional motor learning [18].

Another element of the required knowledge for achieving a cognitive architecture design with the AaA approach is the understanding of the stream of incoming sensory information [19]. The fact that the referenced article discusses this concept in terms of metacognition and decision-making supports the application of it in the AaA approach. The IDMA model explains perception by viewing this concept as a phenomenon that produces sensory events in the current [1]. More specifically, it is explained that a significant change in the continuous sensory input drags the attention of the internal agent [1], which makes the agent experience visual perceptual imagery [3]. Viewed as a conscious experience, the perceptual image is occurring in the internal agent’s conscious awareness in parallel to the execution of an IA (see Figure 2). This IA was decided to be simply referred to as the perception IA.

The concepts presented in this subsection reference knowledge that researchers may apply to design their own cognitive architectures. They are represented in Figure 3, where basal directions are provided for how to apply the semiotic technique adhering to the principles of the approach. Cognitive architecture designers may use this technique as a new knowledge representation method for linking a number of concepts from cognitive science. For example, the perception IA can be related to perceptual learning in LIDA, which is associated with the perceptual associative memory [2]. The latter therefore corresponds to AUP x in Figure 3.

The arrow that represents a state change link (Figure 3) expresses the idea that an IA is part of a single iteration of the cognitive cycle and that after it ends, the internal agent may switch to a different one. Therefore, in this context, a state is represented only by the conscious phase of the cognitive cycle (Figure 2). On the other hand, the unconscious information transition (Figure 3) expresses the idea that the cognitive system transmits information without the conscious awareness of the internal agent. For example, the internal agent can be aware of the body action that it is undertaking, but is not aware of the action potentials that evoke hand muscle responses that produce a chosen behavior scheme [2].

2.1.4. Relations with Other Cognitive Architecture Approaches

A related cognitive architecture design approach is the ACT-R [20], which can reasonably be recognized as having a big influence in designing system models from different application spheres [20,21,22]. The ACT-R approach also views a type of cognitive operation as an interrelationship of two processes: a module and a buffer. The AaA approach investigates AUP types that correspond to buffers and internal action types, corresponding to modules.

Another relation can be found in the design of the H-CogAff architecture [23]. The meta management layer corresponds to the metacognitive layer in IDMA, which is represented by layer 3 in Figure 3. The other two H-CogAff layers can also be related to layers 1 and 2 in IDMA, but their deeper discussion exceeds the frames of this work.

Several relations between LIDA and IDMA were already presented in the previous subsections. However, a clear difference between AaA and LIDA is that AaA clearly states a system of five explanatory rules of an internal attentional experience, described in the following way.

The occurrence of an internal attentional experience

• Corresponds to a short conscious experience [11] (attentional experience [1]);
• Is due to the simultaneous conduction of IA execution and the information provisioning of an AUP;
• Occurs without other internal attentional experience taking place during its course;
• Is preceded by other internal attentional experiences;
• Is followed by other internal attentional experiences.

This formulation of an internal attentional experience points directly to the statement that such an occurrence is related to the action selection phase of an iteration of the cognitive cycle [2]. Also defined as the start of the conscious broadcast [2], this phase corresponds to the internal action execution stage in Figure 2. This view allows the concept of a deliberate cognitive process to be defined as a sequence of internal attentional experiences separated by pauses of unconscious phases. Therefore, the duration of a deliberate conscious process is expressed in the following way:

$$\mathrm{D}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{d}}_{\mathrm{i}},$$

(1)

where d is the duration of a single iteration of the cognitive cycle in milliseconds [2], n is the sequence count, and D is the duration of the deliberate conscious process.

Adhering to the idea of the cognitive cycle [2], a deliberate conscious process includes unconscious and conscious phases in every iteration. In order to express the total duration of consciousness (C) in a deliberate conscious process, the following equation emerges:

$$\mathrm{C}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{d}}_{\mathrm{i}}-{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{u}_{\mathrm{i}},$$

(2)

where u is the duration of an unconscious phase (perception and understanding [2]) corresponding to a single cognitive cycle iteration.

2.1.5. Integrating the Model of Metacognitive Regulation

In order to apply the targeted model of metacognition for cognitive architecture design with the AaA approach, the concepts of metacognitive planning, monitoring, and evaluation need to be viewed in a specific way. The main idea is that each of them is perceived as an ME that is occurring as an act of metacognitive regulation (see Figure 1). Therefore, if research texts discuss the model concepts together with concepts from AaA, such as ME, it could be stated that the research fields produce explanations that link the AaA and the model concepts. Guided by Figure 1, keyword phrases like ‘mental imagery’ and ‘learning experience’ were investigated if they occurred in manuscripts that also discussed the model concepts. If such links exist, theoretically this would mean that established knowledge of the cognitive cycle [1,2,11,18] and mental imagery [3,15] can be linked to the model of metacognition, explaining the metacognitive phenomena of planning, monitoring, and evaluation as types of internal attentional experience (Section 2.1.4).

An explanatory relation can be established by viewing the model phenomena as IAs. This is represented by metacognitive layer 3 (see Figure 3). This way, by following the rule set of IDMA [1], new explanations would be provided about how the target model concepts are related to other concepts from cognitive science. Also, hypotheses would emerge about different state changes between IAs and unconscious information transitions (from Figure 3). A objective can be set towards designing a comprehensive general model of internal attention that includes the meta level, the object level, and the layer between them [6]. The guiding conceptual model in Figure 3 is an update of the IDMA model [1] and can serve to achieve this goal.

3. Research Methods

The method applied in this review can simply be described as searching for specific keywords or key phrases in articles that satisfy the adopted inclusion and exclusion criteria. The term concept token was adopted in this review as a descriptor of a key phrase or a keyword that refers to a concept in science. Two or more corresponding concept tokens can be accepted as referring to the same concept, which allowed the actual counting of occurrences of a concept in the research texts. Thereafter, a proper statistical analysis was conducted to analyze the counts of concept occurrences in the articles. Relations were researched by analyzing whether or not specified concepts had high numbers of occurrences in a single article. On the other hand, articles were counted that included the specified relation in order to display the solidness of it in the scientific fields.

3.1. Concept Tokens

The model keywords—planning, monitoring, and evaluation—are used in different scientific spheres and articles, but their specific understanding as metacognitive phenomena has to be addressed. Therefore, a strong confirmation of the existence of the model would be if the targeted keywords were found to be preceded in the text by ‘self-’ or ‘metacognitive’. However, sometimes authors may use the keywords without ‘self-’ or ‘metacognitive’, but may clearly state that they are metacognitive phenomena, e.g., [5,9]. That is why the model keywords (planning, monitoring, and evaluation) were also counted as separate concepts alongside self-planning, self-monitoring, and self-evaluation. For example, it is accepted that the concept token ‘metacognitive planning’ claims the occurrence of the concept “self-planning”. Also, that the concept token ‘planning’ claims the occurrence of another concept, “planning”. Then, as an example, in the text “they performed self-planning”, the occurrences of two concepts are counted—“planning” and “metacognitive planning”.

Generally speaking, the method dictates that a concept occurrence in an investigated article can be claimed by several concept tokens, and also that a single phrase may contain concept tokens that claim the occurrence of several concepts.

Table 1. These are the defined concepts and their corresponding concept tokens that were applied in the search. The right column explicitly presents the words and phrases (concept tokens) that were searched in the accepted articles in order to claim occurrences of their corresponding concepts.

Concept	Concept Tokens
Self-planning 1	self-planning, metacognitive planning
Self-monitoring 1	self-monitoring, metacognitive monitoring
Self-evaluation 1	self-evaluation, metacognitive evaluation
Planning 1	planning
Monitoring 1	monitoring
Evaluation 1	evaluation
Metacognitive regulation 2	self-regulation, metacognitive regulation
Self-assessment 2	self-assessment, metacognitive assessment
Metacognitive knowledge 3	metacognitive knowledge
Metacognitive experience 3	metacognitive experience
Experience 3	experience
Mental imagery 3	mental image, self-image, mental representation, mental model, mental picture, imagery
Learning as experience 3	learning experience

¹ Concepts of the targeted model of metacognition. ² Concepts that support the model. ³ Concepts that link the model to the theoretical approach for cognitive architecture design.

is presented in order to gain a clear perspective of all the concepts, the occurrences of which were counted in the selected articles. The targeted model of metacognition is most specifically defined by the three key phrases: self-planning, self-monitoring, and self-evaluation. As a concept token, each of them corresponds to another concept token that is formed by ‘metacognitive’ and the target model keyword, for example, ‘metacognitive monitoring’ (Table 1).

In order to confirm that the targeted model of metacognition is established in the scientific fields, more related concepts were required. Planning, monitoring, and evaluation as metacognitive phenomena can be referred to as regulation skills [9,24], metacognitive strategies [25], or dimensions [6], but they are conceptually defined as derivatives of metacognitive regulation [5,9]. This led to defining the concept token ‘self-regulation’ as corresponding to ‘metacognitive regulation’ (see Table 1).

3.2. Phrase Extraction

For this research method, it was considered important to investigate the order of the model keywords in a phrase. It was found that quite often authors listed them in the following order: (1) planning, (2) monitoring, and (3) evaluation [6,9,10,24]. If authors were found to use the same order of listing the model concepts, then this meant that they may have read it from the same source and had the same understanding of it. The more this order of mentioning is found to occur, the stronger the support for the model’s existence and establishment is.

3.2.1. Leading Key Phrase Extraction

A manual annotation technique was designed and applied for every article in order to investigate the order of mentioning of the model concept tokens. It was required that each article had to be investigated by a researcher, who had to find a specific, as short as possible, leading key phrase that was most relevant to the targeted model. A striking leading key phrase was one that included the model concept tokens ‘self-planning’, ‘self-monitoring’, and ‘self-evaluation’. However, a good hit was considered a phrase that included the model keywords without the ‘self-’ prefix (‘planning’, ‘monitoring’, ‘evaluation’). For example, if an article contained the text “self-planning, self-monitoring and self-evaluation”, then this was considered a striking leading key phrase. An example is the article by Fauzi’ah and Khaliyah, who include a keyword phrase that fully corresponds to the model in their abstract [26].

There was a demand for the researchers to provide in their reports the string of the leading key phrase, which had to be searchable in the text. In articles where the model concepts were not mentioned together, a phrase was picked that the researcher considered valuable. Such key phrases were not considered important as they did not include the model concepts. However, this technique also helped to provide directions for other concept tokens that might be relatable to the model concepts. No reliability measures, such as inter-coder agreement, were followed, as the main target was to collect phrases that included concepts of the targeted model, and no specific validation was needed.

3.2.2. Categorical Variables

In different cases, the striking model keywords might have been included but might not have been the only keywords in the phrase [27]. This is considered a small negation of the model, as it is demanded that only the three concepts should be used as standalone concepts. To address this issue, two categorical variables were defined: (1) the three concept tokens are mentioned as standalone concepts in a phrase, and (2) the three concepts exist together in a phrase but are not the only ones. For example, in the article by Sun, Zhang, and Lei, the leading key phrase is “self-planning, self-monitoring, self-regulation, and self-evaluation” [27]. In this case, the concept token ‘self-regulation’ is listed between monitoring and planning, which makes it the leading key phrase of category (2).

By applying this technique, we were able to be statistically specific about which articles absolutely defined the model (1) and which defined it as a linking part with more concepts (2). However, when evaluating the categorical variables, a lenient rule is applied—it is not demanded that the keywords be preceded by ‘self-’ or ‘metacognitive’. For example, McMahon defines “planning, monitoring and evaluation” [28] in a phrase without any other concepts, so this is considered a standalone occurrence (1), even though ‘self-’ and ‘metacognitive’ are not included in the keyword phrase. Some may argue that this allowed articles to be considered as phrasing the model keywords without them referring to metacognitive phenomena. This is mitigated by the first criterion described in the next subsection.

3.3. Criteria for Articles

Each of the articles presented in the final results in the spreadsheet comply with the defined criteria in this section. Some of them were considered inclusion criteria and are simple rules that can easily be checked at the beginning of the investigation of an article. The others (exclusion criteria) were viewed as a system of rules based on requirements for the counted occurrences of the accepted concept tokens. The latter serves as the main guidance for conducting the review, as formulated in Table 1.

3.3.1. Inclusion Criteria

There was one important basal rule, which served as a gate check for the investigation of the articles—the text had to contain the word ‘metacognition’ or its derivatives (‘metacognitive’ or ‘meta-cognitive’).

Another inclusion criteria rule was that the researched works had to be published after the year 2000 and that they had to be of the following types: journal article, conference paper, or book chapter. If the paper was a book chapter, other papers from the same book could not be included in the report. This rule was accepted because a book might be specifically about self-regulation, which is the main indication of the model. If all of the book chapters were included, the spread of the model would not be properly statistically investigated. There were no criteria that demanded scientific fields—any field was accepted in the research.

The last inclusion rule applied when a manuscript was not available to the researcher, but the abstract was. In this case, the article was accepted in the research report only if the concept tokens of the model were listed together in a phrase in the abstract. Other concepts were not counted, as the abstract was not considered a source of proper statistical data.

3.3.2. Exclusion Criteria

The criteria described in this subsection represent a system of rules based on how many of the concepts formulated in Table 1 occurred in the text of each work. They are viewed as exclusion criteria because the text was analyzed (concept tokens were counted), after which the system of rules was applied. If one of the rules in the system was violated, the article was excluded from the report. The system of rules is represented in the following way:

• If the text does not include any occurrences of the ‘self-’ model concept (self-planning, self-monitoring, or self-evaluation), it must include at least one occurrence of each of the model keywords (planning, monitoring, and evaluation);
• If the text includes occurrences of only one of the ‘self-’ model concepts, it must include occurrences of at least two of the model keywords.

It is important to note that model occurrence is viewed as the appearance of one of its corresponding concept tokens in a text. For example, a single occurrence of the concept of self-planning can be claimed by the existence of the concept token ‘metacognitive planning’ in the text.

3.4. Linking to Cognitive Architecture Design

The subsections presented thus far were dedicated to investigating whether or not the model of metacognitive regulation is established in the scientific fields. A systematic review can be conducted only by applying the methodology from Section 3.1 to Section 3.3. The research methods in this subsection serve as an extension of the model investigation and may or may not be applied in a future reproduction of this research. The guiding idea for linking the model to cognitive architecture design is presented in Section 2.1.5. Here we describe the methods for linking the statistical results of the AaA concepts to the model of metacognitive regulation. Also, due to the close relation between the targeted AaA theory and the Global Workspace Theory of consciousness [29,30], it can be stated that these methods provide associations between the concepts of ‘learning experience’ and conscious experience [30].

The goal of these research methods that extend the systematic review is to investigate whether or not the model of metacognition can be integrated into a cognitive architecture designed with the targeted AaA approach. Therefore, by finding articles that include occurrences both of the model and the AaA concepts, a claim can be affirmed that the AaA approach is applicable in the context. If such links between the concepts of the AaA approach and the model are presented, the semiotic technique (Section 2.1.3) can be claimed as trustworthy.

3.4.1. Model Occurrence and Attention as Action Concepts

One of the methods for linking expresses the idea that if one of the categorical variables (Section 3.2.2) is true, then there is an occurrence of the model of metacognition. If such an occurrence in a specified article exists together with occurrences of any of the AaA concepts—ME, mental imagery, or learning experience (Table 1)—then this article is evaluated as linking to the approach. Applying this method first requires establishing the phrase extraction methods (Section 3.2) and counting the occurrences of the tokens corresponding to the AaA concepts. Being specific in meaning, the concepts of ME, mental imagery, and learning experience were demanded to have at least one occurrence in the text in order for them to be linked to the model. Of course, it is considered that a higher occurrence of the AaA concepts strengthens the link to the model.

In general, this method implies that if any of the AaA concepts occur in an article that has a model occurrence, then this concept is linked to the model. Therefore, in terms of this method, categories are formed based on a specified AaA concept. The more articles with combined occurrences of the model and an AaA concept, the stronger the link to that AaA concept.

3.4.2. Interconnecting Attention as Action Concepts

Due to the provided exclusion criteria (Section 3.3.2), it can be stated that all of the accepted articles include a significant discussion of the model concepts. This can be interpreted in a way that, if AaA concepts occur in any of the texts, then it is plausible that these concepts can be explained in terms of the model concepts. That is why a method was formulated that addresses the occurrences of two AaA concepts in a single article. This method is applied by evaluating articles that have any occurrences of AaA concepts, after which categories of articles with interconnected AaA concepts are formed. In this review, of great interest were articles that included occurrences of the following combinations of concepts:

• ME and mental imagery;
• Mental imagery and learning experience;
• Learning experience and ME.

The higher the count of articles in a given category, the higher the link between the AaA concepts in this category. A category having many articles would suggest that its corresponding concepts have a plausible relation to the model concepts. Also, if a high number of occurrences of the AaA concepts are found, then this is interpreted as a demand for explaining the model concepts in terms of AaA.

3.4.3. Model and Interconnected Linking Concepts

This method is the strictest compared to the other linking methods, and therefore provides strong links to the AaA concepts. It is applied by counting articles where references to the model and interconnected AaA concepts occur together. It was expected that few of these articles would be counted as the AaA concepts were quite specific in meaning and it would be unlikely that a striking leading phrase like “planning, monitoring and evaluation” would occur in text that also discusses both mental imagery and ME.

Additionally, this method can be applied by investigating the occurrence of the model combined with occurrences of the three AaA concepts that are of great interest (ME, mental imagery, and learning experience). Such articles are even less likely to exist.

3.5. Research Tools

The gathered data on concept token occurrences, leading key phrases, model factors, year of publishing, and the name of each article were stored in a spreadsheet. The Microsoft Excel software application with its data analysis tools was useful, as it provided a quick generation of plots and other data reports that were helpful for monitoring the results during and after the data collection.

The Scopus database was used for searching for articles by keywords. Google Scholar was also used for finding articles that satisfied the criteria via searching by keywords and specific sentences. On the other hand, the platform Academia.edu was useful for obtaining information on articles that might be relevant to the research topic. This was accomplished by applying for a subscription, which included receiving emails of articles that were similar to previously read papers on the platform.

For chart generation, the Plotly app by Chart Studio: https://chart-studio.plotly.com/ (accessed on 3 March 2025) and the online Graph maker by Image Online: https://graphmaker.imageonline.co/ (accessed on 3 March 2025) were used. Image editing was performed with Microsoft’s Paint.NET software.

4. Results

Initially, the distribution of articles over the years is presented in order to achieve an understanding of the interest of the science community in the model of metacognitive regulation. Next, the model concepts are analyzed via charts of percentages of occurrences. The reported articles are addressed by groups based on scientific fields and how the authors describe the model concepts. Indications are provided for whether or not the model is well established in the fields. Also, analysis is provided on the links between the concepts of the theoretical approach and the ones of the targeted model of metacognitive regulation.

4.1. Distribution in Years

The year of publishing of each article marks the time period in which science groups completed research work with the model of metacognition. Assuming that the criteria are strict enough, it can be said that every article in the data report uses the concepts of the targeted model to an extent.

The area chart in Figure 4 suggests that there has been a solid increase in interest in the self-regulation model since the beginning of the 2020s. In the year 2009, six reported articles were published, which can be interpreted as the peak of the initial leap in interest in the model. In the year 2017, another peak of seven articles is observed. The chart also implies that the model would have a growth of interest in upcoming years, which means that cognitive architecture models that apply the model would be of help for researchers.

4.2. Model Concepts

The model concepts, marked with ¹ in Table 1, are analyzed in this section with the aim of providing directions about which concepts require more research. The ones that are less frequently used in the papers are viewed as the ones that require more investigation.

Chart (a) in Figure 5 clearly shows that the concept token ‘monitoring’ is most frequently found in the reviewed papers. Self-monitoring is also dominant compared to self-evaluation and self-planning (a). This indicates that the understanding of self-regulation is mostly associated with self-monitoring. This statement is supported most evidently with works like [31,32,33,34,35,36]. Even though other model concepts are also mentioned, in these papers, researchers describe self-monitoring as the main derivative of self-regulation.

A strong indication is that the concept of self-planning is not well established in the targeted scientific spheres. However, planning, as a standalone concept, has a closer number of occurrences than the concept of evaluation (21 and 25.8), as shown in Figure 5. Many articles that refer to self-regulation mention planning alongside monitoring and evaluation. This can be seen in research papers related to the sciences of learning and education [37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65]. The commonality between these texts is that the exact key phrase ‘planning, monitoring and evaluation’ occurs at least once. Sometimes authors use the word ‘evaluating’ instead of ‘evaluation’, but because of the other two words listed alongside it, the token ‘evaluating’ is also considered as corresponding to the model keyword. This is striking evidence that the science fields of learning and education have adopted the metacognitive model of self-regulation. A more specific science direction that adopts the general model concepts includes the investigation of the verbal imagery style of learning [65]. This is an important observation as it is related to the concept of imagery experience that is adopted by the AaA approach—the framework that is targeted for cognitive architecture design [1,18].

The general model keywords are also mentioned together in reported studies from scientific fields different than learning and education. Some distinctive topics include human–AI interaction [66], large language models [67], and artists’ creative problem solving [31]. Some of the reported papers contain phrases that imply the general model keywords, but do not specifically state them [68,69,70,71,72]. For example, in the work by Williamson [70], ‘goal setting’ is implied as corresponding to planning, as it is used together with self-evaluation and self-monitoring. Also, in the paper by Rivas [72] (p. 5), ‘supervision’ seems to take the place of the concept token ‘monitoring’, as it stands between ‘planning’ and ‘evaluation’.

Some studies that solidly support the establishment of the model suggest the keywords indirectly by describing self-regulation and mentioning the keywords, but not together in a single phrase [73,74,75,76]. An example is the phrase “in self-regulation, goal setting produces an explicit feedback loop that requires self-evaluation” [73] (p. 8).

There are studies that mention two of the three model keywords in a phrase, like monitoring and evaluation [77,78], planning and monitoring [79,80,81,82,83,84,85], and planning and evaluation [81,82,83,84,85,86,87,88,89]. Others mention two of the concept keywords in different places but concentrate on one of them, like monitoring [90] or evaluation [91].

Another group of studies was addressed that did not solidly support the model. However, their inclusion in the report is useful, as it shows that there is interest in specific concepts that link self-regulation processes to the theoretical approach for cognitive architecture design (AaA). These studies describe concepts like mental imagery and learning experience [92,93,94,95]. The article by Katyal and Fleming refers to “metacognition about mental imagery” [94] (p. 230).

A group from the reported works was formed related to neuroscience and medicine [95,96,97,98,99,100]. One of them [99] included the general concept keywords as standalone words in a phrase, and another strongly referred to the specific ‘self-’, containing concept tokens [96]. The others from this group more scarcely refer to the model concepts, but still refer to self-regulation and self-assessment [96,97].

Lastly, a very specific small group of studies must be mentioned, as they provide directions for how self-regulation processes are viewed as related to computational models and mechanisms [101,102]. The two of them significantly apply the model concepts, but do not clearly state the three of them in a single phrase.

4.2.1. Categorical Variables

This subsection analyzes the results of the categorical variables explained in Section 3. Three groups based on the results of the categorical variables are formulated. They and their results are presented in Figure 6. The two categorical variables (Section 3.2.2) are simply Boolean variables:

• The three model concepts are mentioned together and as standalone words in a phrase;
• The three model concepts are mentioned together in a phrase.

Combined, these variables form three groups, because if the categorical variable (1) is true, then (2) is also true. This way, three types of articles are categorized: articles in which (1) and (2) are true, in which only (2) is true, and where neither are true.

In Figure 6, the pie chart presents the statistical results of the three groups formed by the defined categorical variables. The collected research was what could possibly be found by our team that complied with the criteria defined by the method (Section 3.3). This means that many articles from the red group (Figure 6) might refer to the model concepts but may not clearly state them in a single phrase or a conceptual model. That is why the method demands a leading key phrase to be reported for each paper, in order to closely reflect the model.

4.2.2. Leading Key Phrases

The shortest possible key phrases were reported, but it was also required that the key phrases’ structure not be changed so that they could easily be found in the work they came from. This technique allows not only the model concepts to be analyzed more specifically, but also the investigation of articles in the red group (Figure 6), for which none of the categorical variables are true.

The reported key phrases that included three of the model concepts are presented in

Table 2. This table shows the actual leading key phrases from the reported texts that were associated with the target model and the number of articles in which they were found.

Association	Leading Key Phrase	Articles
Target model	self-planning, self-monitoring and self-evaluation	7
	planning, monitoring and evaluation <or evaluating>	32
	model + self-selection, self-reflection	1
Alternative model *	self-planning, self-monitoring, self-regulation, and self-evaluation	1
	planning, monitoring, regulation, and self-evaluation	1
	planning, monitoring, regulation, and evaluation	1
	planning, monitoring, regulating, and evaluating	1
The three combinatory and standalone	planning, self-monitoring, and evaluation	1
	planning, self-monitoring, self-evaluation	1
	planning a certain task, monitoring and comprehending its progress and evaluating	1
	planning (e.g., advance organizers), monitoring (including self-monitoring), and evaluating (including self-evaluation)	1
The three combinatory and with others	planning, monitoring, information management and evaluation	1
	planning, organizing, self-monitoring and self-evaluating	1
	planning, strategies, knowledge, monitoring, evaluating	1
	monitor their knowledge, decisions, and actions to promote conscious planning, supervision and evaluation	1
	planning, monitoring and control, evaluation	1
	planning, monitoring, information management, debugging, and self-evaluation	1
	planning, implementing strategies, monitoring, and evaluating self-learning	1

* A model that was found during the research and that is tightly related to the target model.

. During the research process of this systematic review, an alternative model appeared. The target model concepts of planning, monitoring, and evaluation are usually viewed as types of metacognitive regulation. In four articles that correspond to the key phrases of the alternative model (Table 2), the target model concepts are clearly stated alongside self-regulation [27,31,62,71]. This means that some authors view self-regulation as an additional concept related to the three others. The alternative model suggests a different understanding, in which self-regulation is not defined as a classifying type of the processes of planning, monitoring, or evaluation, but as a process itself.

The specifying model concepts of self-planning, self-monitoring, and self-evaluation are not that often found together in a single phrase compared to the general model concepts (planning, monitoring, and evaluation). It can be argued that in the context of metacognitive regulation, the general concepts directly correspond to the specifying ones. On the other hand, seven articles of those reported contain key phrases that include the specifying model concepts [6,103,104,105,106,107,108]. This shows that researchers still require the use of the specifying model concepts. This is supported by analyzing the counts of the occurrences of the specifying and general model concepts. In 30.96% of the cases, researchers use self-planning instead of planning, in 30.7% they use self-monitoring instead of monitoring, and in 6.2% self-planning is used instead of planning. This information is presented in the Excel file attached to this work containing the results of the research.

Of the 91 articles collected in the report, 36 of them do not have the three model keywords used together in a phrase. Their phrases can be analyzed on the sheet “key phrases” in the results file. Of the 36 weakly supportive articles, 18 have a leading key phrase that contains two of the model concepts. What is more, in these two-concept phrases, other concepts are found that are related to the third missing concept. For example, in the phrase “self-evaluation, self-monitoring (Watson, 2004), goal setting” [70], goal setting can be related to planning. The tokens “goal setting” and “goals” are found in the other group, in which the reported leading key phrases contain one of the model concepts. The facts presented in this paragraph suggest that some of the weakly supportive articles have significant hints towards the model.

4.2.3. Model Concept Distribution

It was considered important to analyze the reported articles in terms of concept occurrences and the distribution of the model concept occurrences. The criteria presented in Section 3.3 allow papers to be included in the report that contain in their text only two of the three general concept tokens. This means that three types of works exist that are formed based on combinations of the missing concept.

Table 3. Article types based on missing model concepts.

Planning	Monitoring	Evaluation	Count
Missing	-	-	13
-	-	Missing	1
-	Missing	-	0

shows how many articles have a missing model concept. It is clear that there are no papers in the report that are missing the concept token ‘monitoring’. It was demanded that the 13 articles that are missing “planning” should be analyzed more deeply. By investigating the articles in the results table in the report file, it was found that three of them [70,93,96] had “goal” or “goal setting” in their leading key phrase. The other three included other considerable goal concepts, like “regulatory goals” [77], “learner develops goals and plans” [90], and “evolving behavioural goals” [91]. This suggests that six articles of the ones that are missing “planning” discuss concepts that are related to it. On the other hand, the one that is missing “evaluation” discusses “self-assess” as a metacognitive process [25], which can be related to metacognitive evaluation.

Seven articles that are missing “planning” and that do not have planning-related concepts are weakly supportive of the model. However, there are parts of these works that point out how to discuss concepts that are part of the theoretical approach of AaA in terms of metacognition.

4.3. Links with Attention as Action

In Table 1, five concepts are presented that are systematically defined in the theoretical approach of AaA [1,18]. Their occurrences in the texts were analyzed according to the categorical variables to answer the question of whether the model concepts can be applied in designing cognitive architectures using the theoretical approach as a framework. Here, we present the results produced by the application of the methods described in Section 3.4.

4.3.1. The Model and Attention as Action Concepts

Of all the reported 91 works, 25 had occurrences of learning experience, 26 had metacognitive experience, and 27 contained any mental imagery in their text. Now, the question is how significantly does each group refer to the targeted model of self-regulation? This is analyzed by investigating how many articles that include the linking concept also have the categorical variable of “The three mentioned together” (Section 4.2.1) with a true value. Because of the fact that 55 out of the 91 works mention the concepts “planning”, “monitoring”, and “evaluation” in a single phrase, the occurrence of the categorical variable with true value is referred to as a model occurrence phenomenon (MOP).

A striking finding was that 21 out of the 25 papers containing any “learning experience” showed an MOP (Figure 7). What is more, 20 out of 26 papers that discussed metacognitive experience (ME) also showed an MOP. This means that researchers relate to the model quite often when discussing these two linking concepts. The theoretical views presented earlier (Section 2) show the orientation towards the concept of ME. The latter is viewed as an experience of self-regulation, and thus an experience of planning, monitoring, or evaluation. Following this understanding, the concept of metacognitive knowledge should be considered as opposed to metacognitive regulation [5,103], and therefore its occurrence in the texts further supports the linking between the theoretical approach concepts and the targeted model. By observing the results sheet, it turned out that all of the 26 papers that included the concept of ME also included metacognitive knowledge. Together with the other 77% of the 26 papers that showed an MOP, these findings strongly support the idea that self-regulation can be viewed as an ME.

The 25 works in which the concept of learning experience occurs include 18 that mention metacognitive knowledge and 19 that discuss metacognitive regulation. This links the learning experience concept to the targeted model, but only in matters of occurrence. The concept of learning experience can presumably be considered quite a general phrase, which might be used by authors for referring to different ideas. Learning, viewed as an experience that occurs in the cognitive cycle [18], is described as a short (in terms of the timing of the cognitive cycle [2]) internal experience that can be either an ME or another conscious cognitive phenomenon [1,11]. The concept of experience is certainly also general, but it is plausible that statistical analysis is mentioned. The number of articles that contain the concept token ‘experience’ at least once is 75, which is 82% of the total count of reported articles. Of them, 46 (61%) show the MOP, and by considering all 55 that show the MOP, it is persuasive to state that most (83%) of the works that mention the model of metacognitive regulation [5] use the word experience.

It can be said that mental imagery [15] is a relatively new concept emerging in different scientific fields, and it is rarely addressed in relation to metacognition. However, some of the reported articles are concentrated on topics that relate metacognition to mental imagery [13,14]. In the chart in Figure 7, the concept is not that often accompanied by the MOP. Yet, 13 out of 27 works mention the three model keywords and discuss mental imagery. It is important to note that the mental imagery concept corresponds to several concept tokens (see Table 1). Verbal imagery [102], mental representation [7], and mental model [82,92] were all counted as concept tokens corresponding to mental imagery, seemingly together with the tokens ‘self-image’ [84,95], ‘mental image’ [6,13,14], and ‘imagery’ [73].

Another statistical characteristic of the mental imagery concept is its compliance with metacognitive regulation and knowledge [5,103]. Of the 27 articles that include the mental imagery concept, 16 discuss metacognitive knowledge and 14 metacognitive regulation. Some thoroughly discuss the latter [73,82,84,97], which links it to mental imagery.

4.3.2. Interconnected Attention as Action Concepts

Here, we present the results produced by the method for finding interconnected AaA concepts (Section 3.4.2). In order to investigate articles that significantly discuss the model concepts, exclusion criteria were considered. This method is stricter than the one that was applied in the previous subsection, as it involves linking AaA concepts that are specific in meaning.

In

Table 4. Count of articles that include occurrences of two important AaA concepts.

Interconnected Concepts	Article Count
Mental imagery and learning experience	9
Learning experience and metacognitive experience	9
Metacognitive experience and mental imagery	7

, counts of the articles that include interconnected occurrences of two linking concepts are shown. These concept relations suggest that researchers may benefit from a unifying theoretical approach that incorporates these concepts towards the explanation of metacognition.

A type of mental representation is discussed as ‘possible selves’, which are expressed by students’ imagery of their goal attainment [51]. This is quite relatable to the mental imagery of goals as explained by Marks [3,15]. The referenced study [51] thoroughly discusses the targeted model and the concept of ME. This makes it a strongly supportive work, showing that the theoretical approach is suitable for adopting the targeted model of metacognition.

Another study from the field of educational science discussed MEs as “conscious experiences involving thinking and feelings associated with learning” [86] (p. 4). The article uses the concept token “learning experience” several times, relating it to ME. Another view from a different author explicitly links the concept of learning experience to metacognitive regulation [61] (p. 5). This author discusses the concept of ME as an online [8] cognitive endeavor of a feeling or a judgement. The referenced study also practices the concept of mental representation to refer to object-level and meta-level information [61], further expressing the demand for a unifying approach for cognitive architecture design.

4.3.3. Strict Links to Attention as Action Concepts

Tables of the articles that belong to this method are presented in the spreadsheet associated with this systematic review, which is publicly available for access. The results suggest that most of the articles that have interconnected occurrences of the AaA concepts shown in Table 4 also include the MOP. Below, the interconnected concepts corresponding to Table 4 are presented, as well as the articles in which they are found that also include the MOP.

• Mental imagery and learning experience: [7,9,41,61,63,66];
• Learning experience and ME: [38,42,43,54,61,63,66,68];
• ME and mental imagery: [32,51,61,63,65,66,72].

In the case of articles that have ME and mental imagery interconnection, one hundred percent of them include the MOP as well. This is considered a strong indication that ME and mental imagery are associated with the model. Of the nine investigated articles that have occurrences of learning experience and ME (from Table 4), eight of them also include the MOP. As for the mental imagery and learning experience interconnection, six out of nine include the MOP as well.

An additional strict investigation method was to count any articles that had occurrences of all three targeted AaA concepts and had the MOP. Although the expectations were that no such articles would be found (Section 3.4.3), there were three works that satisfied these strict conditions [61,63,66]. All of them include an occurrence of the MOP as well.

5. Discussion

The targeted model of metacognitive regulation is well established in the scientific fields of education and learning, as many articles explicitly refer to planning, monitoring and evaluation (Figure 6). A significant number of neuroscientific studies [96,97,98,99,100] complied with the stated criteria rules (Section 3.3), showing that the model is applicable in this field as well [100]. The reported studies on computational and language models [66,67,101,102] apply the model in their explanations, providing directions for the conceptualization of process systems. The fields of cognitive training and experimental psychology were also shown to practice the model of metacognition [6,81,88,106]. The model is well recognized in studies related to serious games and video game training [6,106,107,109]. The chart (Figure 4) that presents the years of publication of the reviewed works strongly suggests that new applications of the targeted model will exist in the future (Section 4.1).

It was a supportive finding that the negative expectations for the strict method (Section 3.4.3) were refuted in Section 4.3.3. What is more, a significantly high percentage of the articles that had occurrences of several AaA concepts also had occurrences of the model. This suggests that if scientific communities discuss AaA concepts in terms of model concepts, then they deliberate the model concepts together in a phrase. These observations were achieved via the method for leading phrase extraction presented in Section 3.2.

The concepts that link the targeted model to the theoretical approach of AaA appeared to be in close relation with the comprehensions of the concepts of metacognitive regulation [51,61,82,84,86,97]. As a framework for designing cognitive architectures for monitoring learning [18], the theoretical approach of AaA appears to be a tool that will fit with researchers’ demands. Useful conceptual models could be designed to explain conscious experiences as short internal occurrences of learning in terms of the cognitive cycle (Section 2). This would be useful for digital information systems that are required to prompt towards metacognition [1], systems for cognitive monitoring, and the conduction of different digital cognitive tests [18].

5.1. Critiques

A criticizing observation is that the research methodology targets a range of scientific spheres. Different scientists may understand self-planning, self-monitoring, and self-evaluation in different ways. The fact that an alternative model was found (see Table 2) that is tightly related to the target one shows that authors do not always see the target model concepts as types of metacognitive regulation. This critique is supported by the fact that four articles were found [27,31,62,71] that specified ‘self-regulation’ or ‘regulation’ alongside the model concepts, indicating them as separate phenomena.

Another observation on the methods linking the articles to cognitive architecture design (Section 3.4) is that concept tokens claim occurrences of AaA concepts, but no investigation was undertaken to confirm whether or not these concepts were linked to the model concepts in some way.

Some critiques can be identified in the cognitive architecture design approach. Being relatively new, it lacks a significant amount of research. It is a framework for designing cognitive architectures that can present different hypotheses, and its basal understandings act as an underpinning for them. As such, the two only articles [1,18] that covered AaA were criticized to be insufficient.

5.2. Striving for a General Internal Model of Attention

The updated version of IDMA [1] presented in Section 2.1.5 implies that a comprehensive general model that includes many types of IAs can be designed. Such a model would be dedicated to providing explanations about IA states changing in terms of a deliberate conscious process (defined in Section 2.1.4). We strive to design a cognitive architecture that is able to classify and systematically explain all IAs that humans may have.

Considering the findings of this systematic review, an initial attempt was decided to be undertaken towards designing a cognitive architecture that implements the model concepts investigated in this work. As a cognitive system, the conceptual model of the architecture can figuratively explain how different cognitive phenomena lead to metacognitive processes, resulting in processes occurring within seconds as a response to one or several consecutive sensory events.

The designed cognitive architecture is referred to as the General Internal Model of Attention (GIMA), as it represents a map of potentially all internal attention states that are represented conceptually and systematically. The latter, as explained in Section 2.1, are viewed as short conscious processes (IAs) that are executed by an internal agent with the aim of acquiring knowledge from an AUP. Now, after finding evidence for the establishment of the targeted model, the concepts of self-planning, self-monitoring, and self-evaluation can find their place in the metacognitive layer.

In Figure 8, an iteration towards achieving a comprehensive GIMA that includes the metacognitive model concepts is presented. The GIMA is a model of AaA that applies a knowledge representation technique (semiotic technique) [1,18] for expressing links between comprehended concepts from cognitive science in a systematic way. It can be observed that the GIMA is an extension of the conceptual model shown in Figure 3 (Section 2.1.3). Interconnections are represented by linking lines between the different IAs. These lines signify the state change links between the IAs. The interconnections between the IA concepts are based on established understandings of the General Model of Attention as Action [1,18], as well as other studies on mental imagery [3,15] and automatic unconscious processes [2,109]. Scientific support for the underpinning components is presented in Section 2.1. The IAs a and b and their corresponding AUPs are hypothesized and are not discussed in this work.

As a cognitive system, the GIMA follows the idea of IDMA [1] (pp. 11–12) by defining a rule set for state switching based on the three layers. An IA is viewed as a short current mental state that is rapidly changed (Figure 2) in the framework of the cognitive cycle [2]. The rule set provides information about the possible internal decisions that the internal agent may take from a starting point (an IA) in a layer with index number i. This is expressed via the following simple system of rules applied for every internal decision.

After finishing the performance of IA x in layer i, the internal agent

• Can perform an IA in layer i that is not x;
• Can perform an IA in layer i + 1;
• Can perform an IA in layer i − 1;
• Can deliberately evoke body action signal only if x is the motor IA;

The relations between the Action Cycle Theory (ACT) [3] and LIDA [2,109] have been studied, suggesting that LIDA conceptual model types of learning correspond to types of mental imagery experiences [18]. Depicted with orange arrows in the first figure in the article by Kugele and Franklin [109], specific types of learning are linked with memory modules in the conceptual model of LIDA. The AaA approach views these modules as AUPs because they automatically produce information. Applied in the GIMA, the AUPs represent knowledge about the layer to which the memory modules belong (Figure 8). On the other hand, an IA type has been recognized as a learning type in LIDA [18]. This led to the formulation of IA types as defining mental imagery types (modules) in ACT [3] and the learning types from LIDA together [2,109]. These indications support the correspondence proposed in Section 2.1.3 between the perceptual associative memory and the perception IA. Additionally, the evidence from Section 2.1.3 implies that the motor IA corresponds to the sensory motor learning in LIDA that is linked with sensory motor memory [109]. This knowledge is represented by the confluence of the motor IA and the sensory motor memory AUP, as shown in Figure 8.

Critiques

The GIMA model is still in its initial phase of research. It has underpinning knowledge [1,18] backed up by LIDA and the ACT, but the AUPs associated with the new self-planning, self-monitoring, and self-evaluation IAs are yet to be studied and related to processes occurring in human cognition.

Simulations of the GIMA are planned but not yet conducted. A software that applies the ideas of the AaA approach is required in order to reproduce the traversing of IA states based on sensory events produced by the stream of incoming sensory information (SISI). This can be achieved by acquiring knowledge from the understanding of Event Cognition [110] and computational models related to decision-making in terms of the SISI [19].

5.3. Directions for Applications

Occurring in parallel to an IA execution, an internal attentional experience (defined in Section 2.1.4) arises for a short duration in the framework of the cognitive cycle [2]. This means that the GIMA is useful for explaining internal experiences occurring in short periods of time and its application is efficient when it comes to monitoring short volitional body movements and sensory changes.

A digital information system (DIS) can be developed that aims to achieve metacognition or other cognitive phenomena prompts. This can be realized by formulating rules that define a mental imagery experience as a consequence of a digital stimulus, such as the occurrence of a specific sound, image, or a text of information. In this manner, the interconnections between the IA states of the GIMA can serve to guide the system engineer to design such features. This way, in significant circumstances, which demand accurate decision-making, the DIS can help with mitigating cognitive biases by digital prompts towards metacognition [1].

By observing Figure 8, it can be detected that the GIMA can be represented by a weighted bidirectional graph that provides IA knowledge. The IAs can be signified by the nodes in the graph, and their state change links are represented by extraverted (two-directional) edges that have different weights for each direction. This allows the GIMA to be viewed as a data structure that can be applied for the cognitive monitoring of the user of a DIS. Each weight that links two IAs can be a value between 0 and 1, representing the probability of mental state switching. This way, the DIS can alternate the user interface by showing a specified type of information as a digital stimulus. It is important to note that a change in the user interface triggers a change in the current IA state of the user.

Figure 9 depicts the GIMA as a data structure and an example of IA state transitioning. Whenever new information is depicted on the user interface, the transition weight is updated. This approach views the cognitive monitoring system as a separate module that surveils the information that the DIS shows on the user interface. In the example in Figure 9, state transitioning from the perception IA to IA ‘a’ leads to an increase in the probability of weight 22. In the case of iteration ‘i + 1’, the cognitive monitoring module can provide predictions for which internal action is most probable to occur in iteration ‘i + 2’. Based on the prediction, the DIS can provide a digital stimulus that prompts the user towards a beneficial cognitive state.

An example application of the GIMA as a data structure for cognitive monitoring can be provided in the case of a DIS for crisis management [1,111]. Such implementation requires the software of the DIS to have a particular user interface that includes several sections that show different information based on the new incoming data. A suggested user interface is depicted in Figure 10. The top section presents information about the environment that is delivered by other systems or sensors, while the management section provides a user interface via which the user takes critical decisions. The guiding information section is used to accomplish prompting via a digital stimulus.

This design requires production rules, each of which is defined as follows:

• Condition: The current IA state;
• Digital stimulus: An appearance of information shown in the guided information section (Figure 10);
• Product: The IA state to which the digital stimulus prompts.

Some directions are presented below when the condition is the crisis phenomenon [1]. The condition of the crisis phenomenon can be addressed to IA ‘a’, and the IA transitions to the weights 6, 12, and 14 (Figure 9). In this case, in the guiding information section, digital stimuli can be shown as follows:

• Information about methods to be applied: Prompts towards self-planning IA;
• Information about the current cognitive monitoring state of the user: Prompts towards self-monitoring IA;
• Information about what has been achieved by undertaking management operations: Prompts towards self-evaluation IA;

This design approach only provides directions for future work. Its beneficial application in metacognitive prompting is not yet proven, but it serves to provide ideas for applying the concepts investigated in this systematic review. It can be used in planned future works related to this one. It presents a major challenge that demands the investigation of applicable digital stimuli that prompt towards an IA state.

Ideas for Educational Digital Information Systems

By applying the theoretical approach of AaA for theorizing and the knowledge representation technique for conceptualization, systematized knowledge can be generated. This way, this knowledge can be integrated into a DIS that can store and provide information in ways that closely resemble human interpretation, which is an advantage in education, as well as learning and knowledge synthesis.

On the other hand, the AaA approach of explaining internal conscious actions of learning as parallelly occurring along the information provisioning of unconscious processes reveals new alleys for experimental research. DISs can be developed that generate scenario models of cognitive monitoring [18] during the conduction of digital cognitive tests, with paradigms such as Multiple-Object Tracking [112], or others that include conflict processing tasks [113].

6. Conclusions

This systematic review presented the methods and results of two explorations. First, it researched a model of metacognition that defines self-planning, self-monitoring, and self-evaluation as phenomena of metacognitive regulation. Second, it investigated the relations between the model and the approach of AaA for cognitive architecture design.

The reported results showed that the model is well established in the scientific fields of education and learning. However, some studies on neuroscience, experimental psychology, video game training, cognitive architectures, and language models are also shown to apply the model. Significant relations were found that link the model to the AaA approach. The latter, as an emerging framework that strives to design a GIMA, demands scientific exploration and is not yet considered an absolutely solid approach.

Based on the results, it was decided that an initial step should be taken in integrating the targeted metacognitive concepts into a cognitive architecture. A conceptual model of such is presented in this work by applying the knowledge representation technique of the AaA approach. Directions are provided for designing a digital information system that applies the designed cognitive architecture for cognitive monitoring and cognitive prompting.