REPRESENTACIONES EN EL CONTEXTO EDUCATIVO Para entender el contexto educativo como el espacio social en que
C. Sobre los hallazgos
Because Section 1.2 provided an outline of Oberauer’s (2009a) model, this section is devoted to additional theoretical details. As a reminder of Section 1.2, Oberauer (2009a) states that representations in memory exist in two types of nodes: content nodes (including representations of objects, words, or events, such as elements in a series) and context nodes (including representations of roles or positions, such as place within a series). Content nodes are bound to context nodes to signify an element’s role in the relation. Together, a set of content-context bindings constitute a relation with each binding representing one argument in the set. Instantiating and comprehending a relation thus means that the bindings have been integrated. Because any content node can be bound to any context node, an infinite
combination of relational structures can be constructed, with the only limitation being the number of bindings that can be held active at any one time.
Like Cowan’s model and the attentional control view, Oberauer’s concentric model (Oberauer, 2002, 2009a; Oberauer et al., 2007) centres on attention providing temporary access to memory. The concentric model divides memory into tiers based on the level of attentional activation. Figure 2.3 demonstrates this model visually (Oberauer, 2009a). The highest level of activation is the focus of attention, where a single binding (i.e., a tethered content-context dyad) is held in immediate conscious awareness. The binding in the focus of attention is the strongest, shielded from interference by conscious attention. It is also the most flexible, as binding and unbinding of these representations can occur most freely at this level.
The next level is the region of direct access. At this level, a set of related
representations are activated above threshold, granting a privileged status where they are available to immediately be brought into the focus of attention when demanded. To be activated to this level, content representations are bound to contexts that signify a shared relation. Rather than decaying over time, elements are unbound when they are no longer
related to the currently activated relation. This is because new, unrelated representations may be brought into the direct access region and cause attentional interference (Oberauer,
Lewandowsky, Farrell, Jarrold, & Greaves, 2012), causing the focus of attention to struggle to maintain the most activated element. This shares Cowan’s teleological argument for capacity limits. However, unlike Cowan’s theory, the extra layer (the direct access region) means that unwanted representations may cause interference, but not necessarily will cause interference. Novel relations are also generated in this direct access region, with binding occurring serially on each element until a relation is constructed. In a problem-solving
situation, this novel relation may be congruent with the problem solution (in an analogy task), or it may produce a novel element congruent with the problem solution (in most other tasks).
Below this level (outside the direction access region) are representations that have some activation above baseline. This could be because they were recently activated to a higher level or because they are implicitly associated with representations currently active in the direct access region. For instance, a fire truck representation in the direct access region may provide some activation above baseline to firefighter, ambulance, and red even though none of those elements have been within the direct access region or central attention recently. This associative activation allows for quicker and easier access (as in priming), though aside from this associative activation, there is nothing qualitatively separating these activated elements in LTM from elements in LTM with no activation. However, importantly, activated representations outside the direct access region are qualitatively distinguished from elements within the direct access region in that they are not bound to contexts (roles). Well-established relations propagate activation between common representations (content and/or contexts) but representations are not bound until they reach the region of direct access. This allows these above-baseline representations to still be recalled (as in a span task) through cued retrieval by associating activation of targets with cues, or as a list by cascading gradients of activation
levels through the list items (Oberauer, 2009a). Crucially, these methods are not as reliable as preserving the original bindings because the associations may be too weak to spread
activation, but they may be the only option when processing is significantly diverted (as in complex span tasks).
Figure 2.3. An architecture of declarative working memory from Oberauer (2009a), with
labels added. Circles represent individual representations (/elements) in long-term memory and bidirectional arrows representing associations between representations. Currently, elements A, B, and C (represented within content nodes) are all bound (dashed lines) to roles within an interconnected relation (represented within context nodes; lined triangle) within the direct access region (the rectangular frame). Element B is currently within the focus of attention (cone). The bound elements propagate activation to associated elements, though these elements may or may not necessarily be activated above baseline (depicted by shading).
A final point to consider in Oberauer’s model is dimensionality. Like the concept of dynamic chunking in Cowan’s model, memory elements can be represented in several dimensions. For instance, fire truck could be represented on dimensions for physical space (next to us or far from us), hypothetical space (within a fire station or on a road), colour (red or white), as a category (emergency vehicles or land transport), or a theoretically infinite other number of dimensions. LTM stores associations on all these dimensions but when the element is represented in the direct access region (i.e., when it is bound), only a limited
subset of these dimensions is activated. For instance, fire truck could be bound to the vehicle category in a relational instance of drives(driver,vehicle) and, although associations on many other dimensions are tangentially activated (e.g., red, ambulance), the dimension of the current binding promotes activation of elements that share the dimension of the active relation like firefighter to fill the driver role. One merit to our highly flexible generative memory system is that any element could fill the driver position, but it may require some creative and effortful thinking to construct and comprehend the relation. For instance, dog could fill the driver role, but it would not make much sense unless we also manipulate the dog to take on unnatural attributes like paws that can reach the pedals or work a gear shift. Dimensionality allows the memory system to be capable of generative thought while also promoting common, logical declarative thought.