Representation Systems

Recently McNamara and colleagues proposed a model (McNamara, 2003; also see Mou and McNamara, 2002; Mou et al., 2004) to account for new findings on the relationship between studied view, canonical view and updated view representations. McNamara (2003) wrote,

Since participants were disoriented and unaware of their true physical heading, the analysis of interest was whether they retained the "updated view" representation corresponding to their heading before disorientation.

" ... the spatial structure of the environment is represented in terms of an intrinsic reference system; one defined by the layout itself (e.g., the rows and columns formed by chairs in a classroom). Intrinsic directions or axes are selected using cues, such as viewing perspective and other experiences (e.g., instructions), properties of the objects (e.g., they may be grouped together based on similarity or proximity), and the structure of the environment (e.g., geographical slant)."

According to this model, 1) the representation encodes allocentric object-to-object relationships; 2) the direction reference can be selected in various ways, either according to the viewer, the object array itself, or the environment geometry; 3) the spatial working memory system which is primarily for guiding actions updates the viewer's position and orientation relative to the same reference frame, e.g., treating the "self' as just another object, and the egocentric self-to-object relations are then computed from the object-to-object relationships in LTM. This model is similar to Sholl's model (Easton & Sholl, 1995), which also includes an allocentric LTM system and an egocentric working memory system.

We agree that there are LTM representations which are typically not updated, and working memory representations that are updated as the viewer moves. However, we disagree on the nature of the LTM and the nature of updating in the working memory system. We believe that 1) the working memory system encodes egocentric, self-to-object relationships and these relationships are updated directly based on self-motion cues; 2) the LTM system either takes an instance of the working memory representation (i.e., the studied-view), or rotated versions of this representation if rotation leads to a more symmetric, simpler form (i.e., canonical-view). In either case, the representation is of the same, or very similar, nature as the working memory representation. 3) the updating system does not typically consult the LTM representation and can operate independently. In fact, while the updating system is in operation, access to the LTM may be limited or even completely eliminated. 4) when updating is disrupted, LTM may be retrieved to reinitiate the updating process. 5) the updated representations are transient and are typically not preserved when updating is disrupted.

3.1 Two critical issues

There are several critical issues that can potentially distinguish between these models. Here we discuss two of them.

3.1.1 The nature of updating

My colleagues and I conducted a series of studies to examine the nature of spatial updating. There are three pieces of evidence supporting the hypothesis that spatial updating is an independent process that does not rely on another LTM representation. First, people's knowledge of the object-to-object relationship is disrupted by disorientation per se, above and beyond normal memory decay (Wang and Spelke, 2000), suggesting that the updating system relies on representations that are dynamic, or process-dependent, not "static" type of memory such as LTM. Second, a recent study showed that the efficiency of updating depends on the number of objects being updated (Wang et al., 2006). This effect is a direct consequence of updating self-to-object relationships, but is difficult to explain if people treat "self' as just another object and simply update its position relative to the external reference frame. Third, the studies discussed earlier in this chapter showed that updating can operate normally while people had little access to the LTM, again cast doubt on its dependence on LTM. Based on the converging evidence, we believe the updating system is an egocentric system that represents self-to-object relations directly.

3.1.2 The nature of LTM

There is little direct evidence supporting the object-to-object relationship coding, neither is there conclusive evidence against it. Mou and McNamara (2002) showed that a regular array of objects can induce view-dependent representations corresponding to its axis even if that view was never experienced. However, selection of a direction does not imply which relationships are coded. McNamara and colleagues used a judgment-of-relative-directions task to measure knowledge of the object-to-object relationship. However, relative direction judgments can also be made using representations of self-to-object relations, and there is no direct evidence that shows what the task actually addresses.

There are two important issues related to the nature of the LTM. First, the number of object-to-object relations increases exponentially as the number of objects increases. If people only represent a certain-sized subset of these relations, then the chance a specific relation is represented should decrease exponentially. Thus, performance should decrease exponentially as the number of objects increases. In contrast, the number of self-to-object relations is a linear function of the number of objects. Thus, egocentric coding predicts linear relationship between performance and setsize. Second, is the representation position-dependent or position-independent? Object-to-object relationship coding is typically considered position-independent, while self-to-object coding is position-dependent. A position-dependent representation should show cost when the perspective change involves viewer translation, while a position-independent representation should not. Answers to these questions can potentially shed light on the nature of the LTM.

0 0

Post a comment