Recall that we began this chapter with a series of pithy comments about the ways in which cognitive psychologists derive evidence for theoretical entities. That task begins with an analysis of empirical data and proceeds to a theoretical interpretation only through the lens of a particular model. Although we have not emphasized it here, it is important to remember that any comparison of conditions or measures assumes some underlying model, and that those comparisons that are simple do not necessarily reflect simplicity in that underlying model.
Through our short tales in the first section, we discussed the theoretical interpretations of modelbased analysis only as necessary. In this section, I outline rules that other researchers have used to guide the relation between theoretical parameters and theoretical entities. Consider the final example from the first section, in which performance from multiple recall tasks was combined to yield estimates of the contribution of deliberative recollection (R) and automatic memory retrieval (A) to cued recall (Table 24.1). The manipulation of attention had opposite effects on inclusion and exclusion probability, which made the raw data difficult to interpret. However, the model parameters told a very clear story: Attention affects recollection, but not automatic memory. This dissociation provides a first step toward the postulation that these two bases for responding actually represent different memory systems or different memory processes. What else is necessary?
The primary basis for such postulation is the existence of converging multiple dissociations (Schacter & Tulving, 1994). The evidence that aging, for example, selectively impairs recollective but not automatic memory strengthens the case that the two are separate entities (e.g., Benjamin & Craik, 2001; Jacoby, 1999). In the context of animal learning, Lorenz (1970) argued that imprinting was a fundamentally different process than that of normal learning and pointed to various dissociations between the two, such as the presence of a critical period for the former, but not the latter (cf. Shettle-worth, 1993).
Tulving (1984) argued that memory systems should be distinguished in large part on the basis of the information they store and the operations they perform on that information. Thus, procedural memory, which governs the executions of actions and skilled performance, can be distinguished from declarative memory, which contains verbalizable knowledge. Procedural memory contains information about the rapid coordination of limb movements and thus maintains a unique information store. Declarative memory maintains information in sufficiently flexible form to allow inferential processes to act on propositions in memory and thus allows unique operations unavailable to procedural memory. These differences do indeed play out as a number of dissociations in both animals and humans (Squire, 1992).
In addition, Tulving (1984) suggested that memory systems be defined in part by their neural substrates. This is an important point, given the renaissance of cognitive neuroscience briefly remarked on earlier, and I wish to offer an alternative viewpoint as a final remark. The denouement of the argument is that there is no reason why brain systems and cognitive systems should be one and the same.
But do not all the functions of cognition lie in the brain, and therefore shouldn't the structure of the brain be a reasonable playground for the construction of cognitive theories? The answer is no, for the same reason that neither protein strings, nor molecules, nor atoms, nor quarks should be the building blocks of a cognitive theory. Theoretical entities in cognitive psychology are only useful insofar as they allow a handy categorization of experimental results. Thus, despite the fact that habituation in the eye and in the ear take place in different brain regions, we nonetheless recognize a unifying concept that unites the two forms of learning.
A trickier question, however, is whether we are justified in postulating multiple cognitive components that exist in a single brain region. Consider the granddaddy of all distinctions in human memory, that between episodic and semantic memory (Tulving, 1983). Episodic memory stores events from an autobiographical perspective; semantic memory stores facts and knowledge and contains no information about specific past episodes. This distinction has been among the most useful in modern memory research and makes sense out of a huge number of empirical phenomena. Yet, numerous influential theories propose that the information underlying these two memory "systems" is one and the same. For example, Hintzman (1986) showed that a memory system that stored nothing more than specific individual events—in other words, its memory was exclusively episodic—could yield behaviors that were hallmarks for the postulation of semantic memory. Does such a demonstration imply that the distinction is no longer useful? Of course not. Although it may well turn out the brain does not honor this distinction, there is no reason why a cognitive theory should not. Similarly, we can build a reasonable model out of integers and logic components of the way in which our desktop computer performs some computational task, despite the fact that the computer's own representation is binary, and its logic components are nothing more than the arrangements of binary operators. There is no doubt that knowledge about the structure and function of brain regions can and should inform cognitive notions about memory, but there is a danger is failing to recognize additional appropriate levels of abstraction beyond the physical substrate and inappropriately besmirching theories that have desirable qualities.
Was this article helpful?