Episodic Model Imprinting (EMI): A Tripartite Framework for Mental Model Processes  

2.  Support for Three Processing Systems

-top-

Within the last 30 years, dual-process theory has received growing attention, and subsequently has evolved into a significant theoretical and empirical contribution to cognitive psychology (see Table 1). For example, the dual-process distinction has been made between automatic and controlled processes (Schneider & Shiffrin, 1977), associative and rule-based reasoning (Sloman, 1996), implicit and explicit learning (Reber, 1993), recognition and recall (Yonelinas, 2003), recognition-primed decisions and rational choice strategies (Klein, 1998), and heuristic processing versus analytic processing (Evans, 1984). To facilitate discussion of related concepts between dual-process theorists, Stanovich and West (2000) proposed the terms System 1 and System 2. Evans (2003) thus explains that System 1 is old in evolutionary terms and shared with other animals and operates with autonomous subsystems and more associative domain-specific knowledge input modules, while System 2 is more distinctly human and permits goal-directed abstract reasoning and hypothetical thinking. Both systems are supported by neuropsychological evidence (e.g., Goel, 2003). There is also evidence that System 1 belief-based reasoning can bias System 2 logical reasoning (e.g., Quale & Ball, 2000). Recently, Goel and Dolan (2003) have provided fMRI evidence that System 2 processes can intervene or inhibit System 1 belief-biased reasoning when deductive reasoning is required. Schneider and Chein (2003) have further extended dual-process theory by proposing an associative learning mechanism and episodic store for recording associations to their CAP2 WM architecture. In associative learning, output signals from distributed data modules cause changes in the connection weights within and between modules. Learning within each module can take two forms. In associative learning, stimuli can proceed along mental pathways without support from the attentional system. Priority learning occurs within a module when stimuli are consistently attended to, resulting in changing the weights between the input and a priority unit.

Table 1. Terms for the Two Systems in Dual-Process Theory as Used by a Variety of Theorists

Source: Stanovich & West (2000, p. 659).
* Added by author of this research.

2.1   Proposing a System 3 Domain-General Learning Mechanism

In addition to System 1 and System 2, it is now being proposed that a third system shares enough distinct central executive processes, support systems, and localized brain activity areas, as well as resource competition with System 2, to be dissociated from dual-process systems. This dissociation is particularly relevant for the production of episodic models in working memory when users encounter novel information for which they have little LTM support (i.e., no conceptual model). A conjectured System 3 is targeted towards processes and support systems that facilitate domain-general goal-oriented chunking (i.e., the coordination of associations made into coherent structures) and domain-general statistical learning or perceptual chunking (i.e., the automatic segmentation of data streams into meaningful units or chunks). Evidence for a domain-general statistical learning mechanism in infants and adults that facilitates the extraction of patterns from data streams across both auditory and visual domains is growing (e.g., Kirkham, Slemmer, & Johnson, 2002), whereas goal-oriented chunking needs more research (Gobet, et al., 2001).

In an fMRI study investigating verbal and nonverbal paired-associates, Chein and Schneider (submitted) have recently found evidence of a domain-general learning network congruent with their description of controlled priority learning in CAP2. A proposed System 3 builds on Schneider and Chein’s (2003) concept of associative and priority learning. It is further conjectured that the phonological loop and visuospatial sketchpad buffers, slave systems of the Baddeley and Hitch model (1974) of WM, may also play a crucial role in facilitating the maintenance of aural and visual information, thereby increasing WM capacity and the possibility for both implicit perceptual chunking, and more importantly, explicit domain-general goal-oriented chunking. These support systems may have evolved for schema acquisition related to auditory and visual data streams. Support for System 3 can also be found in O’Reilly et al.’s (1999) biologically inspired connectionist model, where control of WM is not centralized, and instead emerges from interactions between different brain systems that include the prefrontal cortex (PFC), hippocampus and related structures (HCMP), and the posterior perceptual and motor cortex (PMC). Of these systems, the PFC plays the most important role in the maintenance and updating of representations, by recurrently activating relevant items. The HCMP system rapidly learns arbitrary associations and is responsible for the rapid binding of novel elements. PMC representations are distributed and embedded in slowly acquired domain-specific processing systems that house long-term learning and skills. In EMI, System 1 can be represented in the PMC, System 2 can be represented in the PFC, and aspects of System 3 can be represented in the HCMP. O’Reilly et al. (1999) view interactions between automatic and controlled processing as occurring along a continuum. Controlled processing arises out of PFC biasing and/or HCMP binding. System 3 coordination of associations made into coherent structures would exist at the end of the automatic and controlled processing continuum, as shown in Fig. 2, where both PFC and HCMP activation are at a maximum.

Fig. 2.  Ways in which the HCMP and PFC contribute to the automatic vs. controlled processing distinction (O’Reilly et al., 1999, p. 387). System 1, System 2, and EMI’s conjectured System 3, have been projected onto the automatic and controlled processing continuum. Automatic processing from the PMC is PFC and HCMP independent.   

2.2   Germane Cognitive Load

The strongest evidence for the dissociation of System 3 domain-general learning processes from System 2 can be found in research on cognitive load theory (Sweller, 1988), which posits that “goal attainment and schema acquisition may be two largely unrelated and even incompatible processes” (p. 283). In other words, the demand for resources in performance type reasoning activities in System 2, compete with resources available for learning. According to CLT, there are three types of cognitive load: (1) intrinsic load, which is the load inherent in the learning material itself, (2) extraneous load, which is the load associated with features of the learning material and other processes, and (3) germane load, which is the load associated with combining elements in working memory to create schema (Sweller, van Merriënboer, & Paas, 1998). The relationship between CLT and EMI can be conceptualized as shown in Fig. 3. All three types of cognitive load are additive. If intrinsic or extraneous load is too high, germane load decreases leaving less room or no room for schema acquisition. A core assumption in CLT is that learning can be increased by reducing extraneous load and increasing germane cognitive load. Research on “germane cognitive load” in instructional design (Sweller et al., 1998) has demonstrated that particular types of instructional designs strongly influence learning outcomes. Gerjets, Scheiter, and Catrambone (2004) maintain that germane cognitive load goes beyond the “mere simultaneous activation of elements in working memory” by adding higher-level cognitive processes that “integrate the elements into a schema” (p. 39).

The coordination of elements into structures has also been researched by Oberauer, Süß, Wilhelm, and Wittmann (2003), who suggest that one of the primary functions of WM is  “to build new relations between elements and to integrate relations into structures” (p. 169). Coordination requires simultaneous access to several information elements in order to construct new relationships, however, it does not imply that the elements are manipulated or processed in any way. Like supervision, coordination is not so much a memory function but an attentional function of WM that requires simultaneous access to several distinct information elements without fusing them into a single chunk. This key central executive process is critical in facilitating goal-oriented chunking that builds strong associations between elements. 

Fig. 3. Relationship between dual-process theory (DP2), cognitive load theory (CLT), and EMI

Inherent in the proposal of a third system is also the notion that “goal-oriented chunking” is based on reasoning process initiated by conscious analogies of new information with old information, and as such, is just as theoretically and empirically distinct as strategically controlled “rule-based” reasoning and automatic “associative reasoning” as differentiated by Sloman (1996). In connectionist frameworks, processes for memory access are determined by the similarity between input patterns and stored patterns in LTM. Hence, if information is unfamiliar, System 2 recall processes are triggered to search LTM for related schemata. Rumelhart (1989) describes that “when part of a familiar pattern is presented, the system responds by ‘filling in’ the missing part” (p. 302). The missing part creates a focal point for the attentional system. Occasionally, however, when there is no existing schemata to interpret new information, Rumelhart and Norman (1981) assume that the “next best schemata” are found using analogical processes. Features that are not present when matching the “old” schemata with the “new” information may then server as a trigger for the creation of a new schema. At this point, the type of controlled processing used to associate, update, and coordinate elements of the “new” information with the “old” schemata will have a crucial impact on learning.

3.  A Framework for Processing Episodic Models

-top-

The Episodic Model Imprinting (EMI) theoretical framework proposes three non-unitary processing systems to describe and predict the episodic model processes users go through as they encounter and respond interactively to new information. In EMI, a system is defined as a collection of WM processes, subsystems, and/or support systems that facilitates a general cognitive function necessary for the production of episodic models in working memory. Controlled by distributed and focussed attention, the processes within EMI’s systems are referred to as consistency, availability, and learnability processing, or more simply: System 1, 2, and 3. Consistency and availability processing systems resemble Stanovich and West’s (2000) “System 1” and “System 2” dual-process accounts of reasoning, with System 1 consisting of more domain-specific knowledge and skill subsystems, and System 2 consisting of more domain-general subsystems that permit problem solving and the manipulation of information, constrained by WM capacity. System 3 is a collection of subsystems and WM processes dedicated to domain-general chunking, and more specifically, making associations in WM to reduce cognitive load in System 2. System 3 also facilitates schema acquisition in LTM, and overtime, the development of domain-specific processing and learning mechanisms for System 1 and System 2. Each system can also be characterized by a dominant learning mode (Rumelhart & Norman, 1978): accretion, tuning, and structuring. Rumelhart and Norman (1978) explain that “memory accretion is most efficiently done when the incoming information is consistent with the schemata currently available” (p. 50). Minor changes to schema are made. If new information is “mildly discrepant,” more significant changes in the form of tuning of the schemata will be necessary. New information that is “more discrepant,” may require restructuring (i.e., major changes in schema) or structuring (i.e., the creation of “new schema”). Table 2 summarizes this and other characteristics of each system.  

Table 2. Characterization of  Processes Within and Between EMI Systems

3.1   Episodic Model Processing Sequence and Cognitive Cycle

To generate a coherent episodic model, the three processing systems of the EMI framework are accessed sequentially. At any point during the EMI processing sequence, if the user is satisfied with the functionality of their episodic model, they can proceed to goal execution. The processing sequence itself is a continuing iteration of recognition, recall, and chunking processes. The sequence iterates until all desired interaction goals are executed. 

Cognitive Cycle

It has been conjectured that a full WM processing sequence from input to output, called a “cognitive cycle,” takes a minimum of 200 msec (Baars & Franklin, 2003). This is equivalent to a maximum of about four or five decisions or goal formations or executions per second. In EMI, during episodic model generation, users may be able to refresh, change, update, and run their episodic model about four to five times a second. Baars and Franklin (2003) have explained that “more than one cognitive cycle can proceed at any given time” (p. 167) and that because of overlapping and automaticity as many as twenty cycles might be running per second. However, consciousness imposes seriality on otherwise concurrent processes thus only one single cycle can be conscious at any given instant. 

Gislason (2004) visualizes cycles of consciousness depending on spontaneously emitted pulses from pacemaking neurons in the brain stem that rise up to awaken neurons in the limbic system and thalamus, which then disperse into all areas of the cerebral cortex like a fountain with the brain stem acting as the pump. Without this rhythmic cyclical ascending activation, humans lapse into a coma. Anesthetics that interrupt consciousness, interfere with this cortical activation. 

Processing Sequence

The EMI processing sequence can be broken down into three stages, as shown in Fig. 4: (1) When a user encounters new information, they automatically create an episodic model. If the new episodic model matches previously stored representations, the user processes it rapidly, and either proceeds to create an interaction goal, if they did not have one initially, or executes their initial goal. If encountering interference, the user defaults to System 2 to process the increased load. (2) If the newly created episodic model partially matches stored representations, the user will attempt to update the missing parts by recalling easily available information from LTM. If interference is encountered, or if the load to WM is too high, the user may access System 3 to chunk information. (3). If recognition of the new episodic model is low, the user will attempt to chunk new information with strong associations. However, if interference is encountered, or if the load to WM is still too high, the user may be unable to generate a coherent episodic model, and performance and retention will suffer. The user may then abandon the task or create a new interaction goal.

• In System 1 processing, there are strong associations between the episodic model created in WM and chunks stored in LTM, hence the episodic model may be immediately usable complete with built-in interaction goals. 

• In System 2 processing, there are moderate associations between the episodic model created in WM and chunks stored in LTM. System processing and several iterations of the sequence may be necessary to generate a coherent episodic model usable.

• In System 3 processing, there are weak associations between the episodic model created in WM and chunks stored in LTM. Substantial processing and iterations of the sequence will be necessary to generate a coherent episodic model. 

Fig. 4. Lower-order and higher-order processing continuum and sequence 
for episodic model production within the three processing systems of the EMI framework. 

Figure 4 also distinguishes between lower-order and higher-order working memory processes. This is in line with Levels of Processing theory (Craik & Lockhart, 1972) which proposed Type I storage and Type II elaboration and organization processes. As discussed earlier, Oberauer, et al. (2000; 2003) have also proposed a distinction between supervision, storage, and processing, and coordination. Ruff et al. (2003), Petrides (1995), and Owens et al. (1999) have also found neurological support for a distinction between maintenance and integration (i.e., storage and coordination).

3.2   EMI Cognitive Model

While the processing sequence model shown in Figure 4 illustrates automatic, lower-order, and higher-order executive processes, as well as the sequential nature of processing within the EMI framework, the EMI Cognitive Model shown in Figure 5, illustrates in greater detail, relationships between each system and related learning mechanisms that include the conjectured domain-general chunking and statistical learning mechanism, along with System 1 implicit associative learning and System 2 explicit priority learning as conceptualized by Schneider and Chein (2003). The purpose of the EMI Cognitive Model is to furthermore represent how Rumelhart and Norman’s (1978) three modes of learning function within each of the three systems. Each of the three system processes and related interactions are explained in the following sections along with simplified flow diagrams of the model.

Fig. 5. The EMI cognitive model. 
Three non-unitary systems for processing episodic models in WM.

3.2.1  System 1: Consistency Processing – Automatic Recognition

System 1 is the “default recognition stage of cognition” demanding limited attentional resources. It operates at a low WM load. Consistency processing is the process of reconstructing a single “familiar” episodic model in the user’s WM of a recognizable event or problem situation through the activation of an existing “retained” episodic model. A retained episodic model is a stored representation of a similar event or problem situation, along with plausible goals, relevant cues, expectancies, and typical actions. A retained episodic model is akin to a user’s mental model, commonly defined in the HCI literature, as the mental model developed through interaction with a system (Norman, 1988). Consistency processing is the most frequently used type of processing for everyday actions and decisions made under time pressure. Klein (1998) defines this type of rapid decision making as recognition-primed decision making. Sloman (1996) would characterize S1 as parallel associative-based processing responsible for intuition, imagination, and creativity. In S1 processing, there are strong associations between the episodic model created in WM and schemata in LTM, hence the episodic model may be immediately usable complete with built-in interaction goals. S1 is further distinguished by frequent activation of procedural knowledge in carrying out familiar tasks.

As shown in Fig. 6 and Fig. 7, consistency processing generates a familiar episodic model with large amounts of input automatically activated from LTM. Familiar episodic models are immediately coherent. A memory record of the interaction may be stored by means of domain-specific implicit learning, as shown in Fig. 6, or by domain-general statistical learning (automatic coordination of associations made into coherent structures), as shown in Fig. 7. In consistency processing, implicit learning is minimal, unless attention is directed to coordinate associations made into coherent structures, as shown in Fig. 11. In consistency processing, learning usually occurs in the form of schema accretion, in which new associations are attached to existing schemata, thereby enriching a user’s mental model.

Fig. 6. Consistency processing with implicit associative learning. 
Episodic model generated in WM, matches previously retained conceptual model.

Fig. 7. Consistency processing with domain-general statistical learning.

3.2.2  System 2: Availability Processing – Controlled Recall and Manipulation

If S1 fails to create an adequate “familiar” representation of an event or problem situation to enable the user to operate at a desired level of performance, S2 takes over to facilitate availability processing. System 2 is the “overriding capacity-limited stage of cognition” that under executive control can inhibit and modify S1 outputs. If overloaded, it can also inhibit learning. Availability processing is the process of updating a previously constructed “familiar” episodic model or building a new “functional” episodic model using other “retained” episodic models or LTM chunks, all of which can be easily “cued” or accessed, as well as more readily maintained and manipulated in WM. Availability processing is slower than consistency processing, frequently involves “recall” rather than “recognition,” is limited by WM capacity, and is the stage where most rule-based or abstract reasoning takes place, as well as goal-oriented decision-making and problem solving. S2 also facilitates dual-task interactions and task switching. In S2, an episodic model starts off as an incomplete representation of an event or problem situation but can usually be updated sufficiently to allow a user to operate at a desired level of performance, and hence become “functional.” In S2 processing, there are moderate associations between the episodic model created in WM and schemata LTM. System processing and several iterations of the sequence may be necessary to render a functional episodic model coherent. S2 is further distinguished by frequent activation of declarative knowledge in order to update episodic models.  

As shown in Fig. 8, availability processing generates a coherent functional episodic model out of domain-specific automatic LTM activation from S1, and conscious LTM activation from S2. A memory record of S1 and S2 processing is most likely to by stored by means of explicit learning, as shown in Fig. 8, such as “priority learning” as described by Schneider and Chein (2003). In availability processing, learning is greater than in S1, but is susceptible to extraneous cognitive overload. S2 learning usually occurs in the form of schema tuning, in which old schemata are modified, thereby updating a user’s mental model.

Fig. 8. Availability processing with explicit priority learning. Episodic model generated in WM,
partially matches previously retained conceptual model. Updating required.

3.2.3 System 3: Learnability Processing – Goal-oriented Chunking

If a novel situation or problem is encountered, in which both S1 and S2 fail to create a “familiar” or a “functional” episodic model, S3 takes over to facilitate learnability processing. System 3 is the “chunking stage of cognition” that demands a restricted goal-oriented attentional focus to coordinate associations made in WM into coherent structures. In S3, these structures are referred to as “structured” episodic models. Strong goal-oriented associations within these structures lead to better long-term retention. Demanding more attentional resources and WM capacity, learnability processing can be the slowest and most deliberate type of processing. System 3 is also the key system responsible for creating long-term working memory (Ericsson & Kintsch, 1995), and skilled learning, which when acquired becomes part of System 1 processes. In S3 processing, there are initially weak associations between the episodic model created in WM and schemata in LTM. Substantial processing and iterations of the sequence will be necessary to generate a coherent structured episodic model. S3 is further distinguished from S1 and S2 by frequent activation of structural knowledge, in order to facilitate the coordination of associations made into structures.

Fig. 10. Learnability processing with domain-general goal-oriented chunking.

Fig. 11.  Metacognitive domain-general goal-oriented chunking after consistency processing, 
in which an old conceptual model is significantly restructured. 

3.2.4  EMI's Theoretical Assumptions

The foundations of the EMI framework are based on the following seven assumptions, derived from important conclusions made in the above discussions: 

1. Dynamic Representations 

An “Episodic Model” is a dynamic representation created in WM by combining information stored in LTM and characteristics extracted from the environment.

2. Coherent Episodic Models

Information processing in working memory is directed towards creating a single coherent episodic model that exerts a low cognitive load on the processing system. A coherent episodic model is a mental representation of an event or problem situation perceived by the user to be sufficiently complete to facilitate the execution of a chosen interaction goal.

3. Tripartite Framework

There are three non-unitary systems for creating and processing episodic models in working memory. These systems are sequentially and iteratively utilized. 

4. System 1: Consistency Processing

System 1 is the “default recognition stage of cognition” made up of numerous domain-specific subsystems that demand limited attentional resources to process familiar information. It operates at a low cognitive load when no interference is present or when no supervision is required. 

5. System 2: Availability Processing

System 2 is the “overriding capacity-limited stage of cognition” that under executive control can inhibit and modify System 1 outputs, and if overloaded can interfere with System 3 processing.

6. System 3: Learnability Processing 

System 3 is the “chunking stage of cognition” that demands a restricted goal-oriented attentional focus to coordinate associations made in working memory into coherent structures. 

7. Episodic Model Usability (Performance and Retention Optimization) 

Performance can be optimized if the episodic model generated by System 1 processing is consistent with information already known and if any elements required for updating or problem solving in System 2 processing are easily available. Retention can be optimized in System 3 processing if the episodic model is learnable, that is easily structured (i.e., chunked) from component parts.

4.  Associations Matrix

-top-

Episodic model usability can be further operationalized using a WM and LTM associations matrix (see Fig. 12, 13, and 15). Cowan (2001) defines a chunk in WM as having strong intra-associations within a collection of items and weak inter-associations between items. He also distinguishes between pure STM capacity limit expressed in chunks from compound STM limits obtained when the number of separately held chunks is unclear. He claims that Miller’s (1956) “seven, plus or minus two” limit should be more precisely defined as a “compound STM limit,” in which support structures like Baddeley and Hitch’s (1974) phonological loop can assist individuals in holding more items in short-term memory. In extending these concepts to EMI, cognitive load can be classified according to stronger, moderate, or weaker associations between the episodic model in WM and conceptual model LTM, as well as stronger, moderate, or weaker associations between elements  within working memory (i.e., new chunks) to yield six levels of associations (see Fig. 13). The upper end of each of the three systems with strong inter-associations between items in WM, represent episodic models that are more usable than those with moderate or weak inter-associations. Such a matrix also provides a way to classify and test different types of interactions using interfaces or instructional materials with quantitative and qualitative differences in working memory loads. 

Fig. 12. Aggregate cognitive load for strong, moderate, and weak associations between information stored in LTM and the user’s episodic model, and associations between new chunks created dynamically in WM.

• Baseline Recognition – When a user encounters new information, such as an interface navigation menu, a minimum amount of strong associations with information stored in LTM is needed to process basic inputs such as letters of the alphabet, font faces, and the meanings of words. This can be called baseline recognition and is generated automatically via perceptual chunking.

• LTM Associations – LTM associations are associations between the user’s conceptual model and the new information (i.e., the “LTM match”). If a new episodic model closely matches previously stored conceptual models, there is a low cognitive load. If it does not match previously stored representations (i.e., moderate or weak LTM-associations), cognitive load is higher. 

• Inter-Associations – Inter-associations are associations between chunks in WM. For S1, inter-associations that make up the user’s episodic mental model in WM will be strong. In S1, it is impossible to have episodic models with moderate or weak inter-associations as there is a full LTM match. In S2, new information extracted from the environment that is not familiar may be easy or moderately difficult to chunk. In S2, activated LTM chunks with strong or moderately strong inter-associations can be processed into coherent episodic models. However, if inter-associations between activated LTM chunks are weak, processing defaults to S3. In S3, new information that does not match previously stored representations and is furthermore difficult to chunk in WM will generate a very high cognitive load. On the other hand, if it is relatively easy to make new associations, the newly chunked information will reduce cognitive load. Intra-associations are associations within chunks. In all systems, intra-associations can be weakened by interference or distractions to attentional focus, thus increasing cognitive load.

Applying the Episodic Model Association Matrix to Gas Burner Stovetop Design

Proper gas stovetop design is important for safety and usability. The EMI Association Matrix can be applied to various designs for gas burner stovetops, as shown in Fig. 13. For further explanation and examples, click on each of the six categories of associations between LTM and within WM .  

C1 ~  A2 ~ A3 ~ L4 ~ L5 ~ L6

Fig. 13. Episodic Model Association Matrix for gas stovetop design. 

5.  Experimental Design

-top-

Up to this point, developing EMI has been primarily an analytical task involving the synthesis of a great number of important relationships and processes as outlined in the cognitive psychology and cognitive science literature. Nevertheless, the scope of applications to HCI and instructional design is promising. In applying EMI to instructional design, the following assumptions can be made: (1) learning is based on experience; (2) goal-directed chunking is the fundamental mechanism for binding experiences to LTM in a retrievable format; and (3) by striving to create usable familiar, functional, and structured episodic models in learners, the capacity for goal-directed chunking is optimized. 

Currently, the EMI framework is being applied to Web-based navigation menus to obtain qualitative and quantitative experimental evidence to verify performance and retention characteristics of each of the three proposed processing systems. One of the central hypotheses being tested is that strong inter-associations in WM lead to better retention in LTM (i.e., better conceptual models). A pilot study has been conducted to test a repeated measures experimental design that uses two-level navigation menus to facilitate or interfere with S1, S2, or S3 processing. Preliminary results from retention scores and think-aloud feedback are promising. In the future, relationships between chunking, WM capacity, and retention, and how and to what extent the contents of episodic models can be manipulated by the central executive, will be examined more closely to provide further conceptual clarity to the EMI framework, as well as processing channels in the EMI cognitive model.

Experimental Design Overview

A within-subjects balanced Latin Square design has been chosen as an appropriate method for testing seven web-based menu design treatments and their impact on performance and retention during two navigation search tasks. A total of nine menu designs were developed including six experimental treatments, one control treatment, and two practice treatments (see Fig. 14). 

The basic experimental design will be conducted over two sessions: a one hour web-based navigation session followed by a two hour WM span and mental ability testing session.  

Fig. 14. Within-subjects experimental design followed by control interface (X0)with consistency design treatment, and screen shot of HTML form used to record retained episodic models.

Association Matrix of Experimental Design Treatments

To facilitate analysis of important relationships within and between systems conjectured in EMI framework, each of the experimental treatments were designed to correspond to at least one of the aggregate cognitive load levels identified in the “EMI Associations Matrix” as shown in Figure 15. For the following experiment, baseline recognition common to all eight menu designs was quite high, with modifications made only to the structure of the first- and second-level menu items. High baseline recognition is based upon the assumption that the common nouns used for each menu are equally familiar and memorable.

Fig. 15. Relationship of experimental treatments to the EMI Association Matrix of aggregate cognitive load for strong, moderate, and weak associations between information stored in LTM and the user’s episodic model, and inter-associations between new chunks created dynamically in WM.

Cognitive Skills Tests

For the follow up session after menu navigation, a battery of fourteen cognitive computer- and paper-based tests were chosen to measure either WM span and related executive processes or mental ability performance (see Fig. 16). These psychometric tools were chosen from previous research in the field and where necessary adapted as psychometric tools for measuring abilities and processes most likely required for the retention of Web-based navigation menus. In general, the objective of such tests are to determine what kinds of working memory functions, specifically supervision, storage and processing, and coordination, might correlate highly with episodic model retention scores (examples of WM tests are shown in Fig. 17).

Fig. 16. Classification of 9 WM tasks and 5 mental ability tests based on Oberauer et al.’s (2000) three overlapping functional categories of supervision, storage and processing, and coordination (with the added category of “association"). 

Fig. 17. Example for sequence of screen displays for selected WM tasks and sample mental ability tests.

Main Hypothesis and Research Questions

System 1 and System 2 (i.e., the “dual-process” distinction) are well supported by current research. However, although research in this area is growing, the conjectured System 3 processes are less well understood. The following hypothesis is directed towards the study of System 3 domain-general chunking. 

H_R The coordination of associations made into coherent structures facilitates retention.

The following hypothesis seeks to find evidence to confirm characteristics of learnability processing. Such evidence may provide support for a domain-general learning mechanism involving higher-order central executive processes. General speaking, it is hypothesized that design treatments that facilitate the conscious coordination of associations made into coherent structures (i.e., goal-oriented chunking) will increase episodic model retention. Likewise, design treatments that do not facilitate such processes should result in lower retention.

Furthermore, if in fact association and coordination abilities do foster retention, individuals with superior association and coordination WM and cognitive skill abilities should have better overall retention scores than individuals who do not, especially on menu design with learnability-based treatments. Individuals with a high WM span may also show superior retention of menu designs, which may be particularly evident on tasks the require retention of unstructured menu designs. It important to distinguish between the contribution of  “maintenance and storage” abilities (lower-order WM processes) versus the contribution of “association and coordination” (higher-order WM processes) to episodic model retention.  

R_(Q)1 During elicitation tasks, if the contents of a retained episodic model are unpacked in large chunks, is retention higher? 

The generally accepted “mental model hypothesis” (Kellogg & Breen, 1990) suggests that within any task domain, the richness of a person’s mental model has a critical impact on the person’s task performance in that domain. In the following study, it is thus assumed that in order for participants to navigate menu designs effectively, they must be able to build an accurate episodic model of the system and retain it. Episodic models that are coherent (i.e., help them complete goals) and usable (i.e., exert a low cognitive load) should be retained longer. 

A usable retained episodic model (i.e., the user's conceptual model) would likely consist of relatively large chunks stored in LTM (possibly a single chunk). By examining how participants unpack information during elicitation tasks, and noting pauses (greater than 2 secs), the number of chunks and chunk sizes can be estimated (Chase & Simon, 1973). Participants with larger chunks sizes should have superior retention. The following research question seeks evidence to confirm assumptions made regarding the importance of “coherence” in episodic model production. Such evidence should also lend support to the main hypothesis (H_(R)1). 

R_(Q)2 What effect will working memory interference have on menu navigation performance and accuracy, as well as menu retention scores? 

The following research question addresses theoretical assumptions regarding distinctions between System 2 and System 3. How will lower-order interference, such as articulatory suppression (“the”) and higher-order interference such as “counting,” affect schema acquisition? Counting may place a greater demand on central executive coordination resources (e.g., Towse & Hitch, 1997).

R_(Q)3 Is there a correlation between analogical reasoning ability and retention scores? 

Patterned generation involves the creation of new schema by generalizing or copying an old one with a few modifications (i.e., interpreting new information based on old information). This type of restructuring is by far the most frequent type of new schema generation. Rumelhart (1980) explains that “such learning is, in essence, learning by analogy” (p. 54). The following research question seeks to address theoretical assumptions regarding characteristics of learnability processing, specifically the importance of the conscious association of old knowledge with new knowledge to create strong episodic model coherence. Analogical reasoning may be at the core of logical reasoning (Sowa & Majumdar, 2003), problem solving, (Goswami, 2001), and general cognition (Hofstadter, 2001). It is a good measure of intelligence (Sternberg, 1977). Individuals with superior analogical reasoning skills may have higher retention scores on menu designs that facilitate structuring.

6.  Conclusion

-top-

Barnard’s (1999) Interacting Cognitive Subsystems (ICS) framework describes nine cognitive subsystems for creating mental representations in WM. A major advantage of the ICS architecture is that cognition can be considered within an overall psychological and physiological context, making it ideally suited for complex real-world tasks. It also provides a rich structure for examining cognitive resources required to operate user interfaces. However, it seems to have a limited ability to predict how a user responds when the behavior of a system is not what the user expects, in other words, when the actual behavior of the system departs from the user’s mental model (Rushby, 2001). The EMI framework hopes to provide designers with an understanding of how users react to novel information in which they have no previous mental model, as well as how designers might work to create episodic models with strong associations in WM, which can subsequently be more easily retained. From an HCI perspective, EMI attempts to answer the question: How can designers rapidly create conceptual models in users?

Schneider and Chein’s (2003) Controlled and Automatic Processing version 2 (CAP2) architecture provides a model for skilled processing and learning, and like O’Reilly, Braver, and Cohen’s (1999) Biologically Based Computational Model (BBCM) of WM, is theoretically based on dual-process theory. Both CAP2 and BBCM have also taken advantage of growing neuroscience data in modeling WM processes representing a positive leap in current WM research. However, from a designers perspective, CAP2 and BBCM are difficult to apply to instructional and interface design. The EMI framework hopes to provide what Sutcliffe (2002) refers to as “cut-down versions of theories to promote better understanding of usability issues” (p. 24). Sutcliffe (2002) describes such theories as “digestible chunks of knowledge that can be applied to design” (p. 25), in which the complexity of a theory is hidden from designers and the advice contained in the theory generalized so it can be re-used in a wide range of contexts.

References 

-top-

Baars, B.J., & Franklin, S. (2003). How conscious experience and working memory interact. Trends in Cognitive Science, 7, 166-172.

Baddeley, A.D. (2000). The episodic buffer: a new component of working memory? Trends in Cognitive Science, 4(11), 417-423.

Baddeley, A. D. (2002). Is working memory still working? European Psychologist, 7(2), 85-97.

Baddeley, A.D., & Hitch, G. (1974). Working memory. In G.H. Bower (Ed.), The psychology of learning and motivation, Vol. 8 (pp. 47-89). New York: Academic Press.

Barnard, P.J. (1999). Interacting cognitive subsystems: Modeling working memory phenomena within a multiprocessor architecture. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 298-339). New York: Cambridge University Press.

Brewer, W.F. (1987). Schemas versus mental models in human memory. In P. Morris (Ed.), Modeling cognition (pp. 187-197). Chichester, UK: Wiley.

Cañas, J. J. & Antolí, A. (1998). The role of working memory in measuring mental models. In: T.R.G. Green, L. Bannon, C.P. Warren & J. Buckley (Eds.). Proceedings of the Ninth European Conference on Cognitive Ergonomic-Cognition and Cooperation (pp. 139-144).

Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55-81.

Chein, J.M., & Schneider, W. (submitted). Evidence of a domain general learning network: An FMRI investigation with verbal and nonverbal paired-associates. Nature Neuroscience.

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87-186.

Craik, F. I. M., & Lockhart, R.S. (1972). Levels of processing. A framework for memory research. Journal of Verbal Learning and Verbal Behaviour, 11, 671-684.

Epstein, S. (1994). Integration of the cognitive and the psychodynamic unconscious. American Psychologist, 49, 709-724.

Ericsson, K.A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211-245.

Evans, J. St. B. T. (1984). Heuristics and analytic processes in reasoning. British Journal of Psychology, 75(4), 541-568.

Evans, J. St. B. T. (2003). In two minds: dual-process accounts of reasoning. Trends in Cognitive Science, 7(10), 454-459.

Evans, J. St. B. T., & Over, D. E. (1996). Rationality and reasoning. Hove, England: Psychology
Press.

Gerjets, P., Scheiter, K., & Catrambone, R. (2004). Designing instructional examples to reduce intrinsic cognitive load: Molar versus modular presentation of solution procedures. Instructional Science, 32, 33-58.

Gislason, S. J. (2004). The book of existence and the human mind. Sechelt, B.C.: Environmed Research Inc. (Extract from chapter on “Neurophysiology of Consciousness” available: http://www.nutramed.com/Philosophy/consciousness_3.htm)

Gobet, F., Lane, P. C. R., Croker, S., Cheng, P. C-H., Jones, G., I Oliver, & Pine, J.M. (2001). Chunking mechanisms in human learning. Trends in Cognitive Sciences, 5(6), 236-243.

Goel, V. (2003). Evidence for dual neural pathways for syllogistic reasoning. Psychologica, 32, 301-309

Goel, V., & Dolan, R. J. (2003). Explaining modulation of reasoning by belief. Cognition, 87, 11-22.

Hammond, K. R. (1996). Human judgment and social policy. New York: Oxford University Press.

James, W. (1890). The principles of psychology (Vol. 2). New York: Holt.

Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language, inference and consciousness. Cambridge: Cambridge University Press.

Kellogg, W. A., & Breen, T. J. (1990). Using pathfinder to evaluate user and system models. In R. Schvandeveldt (Ed.) Patherfinder associative networks: Studies in knowledge organization (pp. 179-195). Norwoord, NJ: Ablex.

Kirkham, N.Z., Slemmer, J.A., & Johnson, S.P. (2002). Visual statistical learning in infancy: Evidence for a domain general learning mechanism. Cognition, 83, B35-B42.

Klein, G.A. (1998). Sources of power: How people make decisions. Cambridge, MA: MIT Press.

Levinson, S. C. (1995). Interactional biases in human thinking. In E. Goody (Eds.), Social
intelligence and interaction (pp. 221-260). Cambridge: Cambridge University Press.

Miller, G. A. (1956). The magical number seven plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 81-97.

Norman, D.A. (1988). The psychology of everyday things. New York: Harper & Row.

O’Reilly, R.C., Braver, T.S., & Cohen, J.D. (1999). A biologically based computational model of working memory. In A. Miyake, & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control (pp. 375-411). NY: Cambridge University Press.

Oberauer, K., Süß, H.-M., Schulze, R., Wilhelm, O., & Wittmann, W. W. (2000). Working memory capacity—facets of a cognitive ability construct. Personality and Individual Differences, 29, 1017–1045.

Oberauer, K., Süß, H.-M., Wilhelm, O., & Wittmann, W.W. (2003). The multiple faces of working memory: storage, processing, supervision, and coordination. Intelligence, 31, 167-193.

Quayle, J.D., & Ball, L.J. (2000). Working memory, metacognitive uncertainty, and belief bias in syllogistic reasoning. The Quarterly Journal of Experimental Psychology, 53A(4), 1202-1223.

Owen, A. M., Herrod, N. J., Menon, D. K., Clark, J. C., Downey, S. P. M. J., Carpenter, T. A., Minhas, P. S., Turkheimer, F. E., Williams, E. J., Robbins, T. W., Sahakian, B. J., Petrides, M., & Pickard, J. D. (1999). Redefining the functional organization of working memory processes within human lateral prefrontal cortex. European Journal of Neuroscience, 11, 567-574.

Pashler, H., Johnston, J. C., & Ruthruff, E. (2001). Attention and performance. Annual Review Psychology, 52, 629–651.

Petrides, M. (1995). Functional organization of the human frontal cortex for mnemonic processing: evidence from neuroimaging studies. Annals of the New York Academy of Sciences, 769, 85–96.

Pollock, J. L. (1991). OSCAR: A general theory of rationality. In J. Cummins & J. L. Pollock (Eds.), Philosophy and AI: Essays at the interface (pp. 189-213). Cambridge, MA: MIT Press.

Posner, M. I., and Snyder, C. R. R. (1975). Attention and cognitive control. In R. L. Solso (Ed.), Information Processing and Cognition (pp. 55–85), Hillsdale, NJ: Erlbaum .

Reber, A. S. (1993). Implicit learning and tacit knowledge: An essay on the cognitive unconscious. New York: Oxford University Press.

Ruff, C. C., Knauff, M., Fangmeier, T., & Spreer, J. (2003). Reasoning and working memory: common and distinct neuronal processes. Neuropsychologia, 41, 1241–1253.

Rumelhart, D.E. (1989). Toward a microstructural account of human reasoning. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 298-311). Cambridge: Cambridge Uni. Press.

Rumelhart, D.E., & Norman, D.A. (1978). Accretion, tuning and restructuring: Three modes of learning. In J. W. Cotton and R. L. Klatzky (Eds.), Semantic factors in cognition (pp. 37-53). NJ: Erlbaum.

Rumelhart, D.E., & Norman, D.A. (1981). Analogical processes in learning. In J. R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 335-361). Hillsdale, NJ: Lawrence Erlbaum Associates.

Rushby, J. M. (2001). Modeling the human in human factors. Proceedings of the 20th International Conference on Computer Saftey, Reliability and Security, 86-91.

Schneider, W., & Chein, J.M. (2003). Controlled & automatic processing: Behavior, theory, and biological mechanisms. Cognitive Science, 27, 525-559.

Schneider, W., & Shiffrin, R.M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1-66.

Sloman, S.A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119, 3-22. (In Gilovich, T., Griffin, D., & Kahneman, D. (2002). Heuristics and biases: The psychology of intuitive judgment (pp. 379-396). New York: Cambridge University Press.

Stanovich, K.E., & West, R.F. (2000). Individual differences in reasoning: Implications for the rationality debate? Behavioral & Brain Science, 23, 645-726.

Sutcliffe, A. G. (2002). On the effective use and reuse of HCI knowledge. In J. M. Carroll (Ed.), Human-computer interaction in the new millennium (pp. 3-29). New York: ACM Press.

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12, 257-285.

Sweller, J. van Merriënboer, J.J.G., & Paas. F. (1998). Cognitive architecture and instructional design. Ed. Psych. Rev., 10(3), 251-296.

Towse, J. N. Hitch, G. J. (1997). Integrating information in object counting: a role for a central coordination process? Cognitive Development, 12, 393-422.

Yonelinas, A.P. (2002). The nature of recollection and familiarity: A review of 30 years of research. Journal of Memory and Language, 46, 441-517.

-top-