The memory systems approach has frequently been criticized by investigators who defend a focus on processes or procedures rather than systems. These are roughly equivalent to what I call operations and, like operations, might be identified only at a very general level early in inquiry (e.g., the process of retrieving a memory) and at a finer grain later (e.g., two of the processes by which a memory is retrieved, given a cue, might be spreading activation from the cue and assessment of the strength of a retrieved item). Roediger, Buckner, and McDermott characterized the process approach as follows, highlighting Paul Kolers' role in its development.11
The hallmark of the procedural approach, harking back to Bartlett and Neisser, was that performance on memory tasks could be described as skilled performance and that one should look to the procedures of mind to explain cognitive performances. Many experiments can be interpreted as supporting the procedural approach, including several revealing dissociations in performance on tasks that all measured recognition of words. In particular, Kolers' experiments showed that transfer from one task to another benefited to the degree that the procedures underlying performance on the two tasks were similar (Roediger, Buckner, & McDermott, 1999, p. 42).
The challenge for the process approach is how to identify processes (operations). Much of the research focuses on establishing that performance in different
11 In large part, Kolers construed his process approach as opposed to a structuralist approach in which memory was viewed as involving stored representations, and the mind was populated with different structural components. Kolers and Roediger (1984) stated: "We will show in some detail that distinctions between mental representation and mental process, between 'symbol' and 'skill,' are of questionable worth for psychology and may indeed actually misrepresent psychological processes" (p. 429). Instead of drawing a dichotomy between structuralist and proceduralist accounts, however, it is perhaps more fruitful to construe Kolers' project as recommending a corrective in the symbolic/linguistic account of mental representations and recognizing a broader range of mental representations and processes on them. Kolers and Roediger's positive statement of their view, in fact, maintains the distinction between representations and operations, but shifts the focus to operations (skills): "We will argue that knowledge is a matter of skill in operating on symbols, that the latter are of many kinds, that the kinds are not perfectly correlated, and that knowledge is, as a consequence, means dependent" (p. 430). One feature of the corrective, though, may be the recognition that there is not just one set of representations on which different operations are performed, as suggested for example by Anderson: "Representations that do not preserve the exact perceptual structures of the events remembered are the mainstay of long-term memory. It is important to appreciate that these meaning representations are neither linguistic nor pictorial. Rather they encode the meaning of pictures and linguistic information" (Anderson, 1990, p. 122). One alternative to consider is that in many cases representations may be grounded in the sensory modalities through which we acquire knowledge of the world; as discussed in chapter 5, there is growing evidence that modality-specific representations are reactivated in thought processes.
tasks relies on different processes rather than characterizing the processes themselves. Kolers, for example, developed a strategy for establishing that different processes were involved by examining whether learning one kind of task (reading inverted text) transfers to performance on another (reading reversed text). If performance of the two tasks drew upon similar processing operations, then there should be transfer from learning to perform one to learning the other, but not otherwise (Kolers & Ostry, 1974; Kolers & Perkins, 1975).12
A similar approach was employed by John Bransford and his collaborators (Bransford, Franks, Morris, & Stein, 1979) in advancing the idea of transfer-appropriate processing. The idea is that recall performance would be affected by similarity between the processes used in acquiring information and those required in recall. Performance would be better if the learning and recall tasks both involved semantic processing than if learning involved phonemic processing whereas recall emphasized semantic processing. If the recall task involved a phonemic judgment (e.g., "Does this word rhyme with a word seen during encoding?"), then performance would be better if the learning also required phonemic processing.
As in the memory systems literature, advocates of processing approaches relied on findings of dissociations. The difference is that they used dissociations to distinguish, not entire systems, but component processes involved in one task from those used in another task. Roediger and his colleagues (Roediger, Weldon, & Challis, 1989) developed a set of principles that they hoped would point the way to a processing-based account of a wide range of dissociation data:
1. Performance on memory tests benefit to the degree they invoke the same processes required in encoding
2. Explicit and implicit memory tests tend to invoke different processes
3. Most (but not all) explicit memory tests (free recall, cued recall, recognition) depend primarily on semantic and conceptual processing
4. Most (but not all) implicit memory tests depend primarily on perceptual processing13
12Kolers and Roediger described the approach: "Our view, transfer of training and savings methods, properly applied, constitutes a fundament upon which to construct an empirically based cognitive psychology. The techniques would be applied as measures of skills acquired in one cognitive task and expressed in performance on another. Degree of transfer from one task to the second or, as in Nelson's (1978) work, the more subtle measurement of savings, can aid in diagnosing the underlying cognitive operations. The idea is that any complex event is composed of a number of component activities, and the more alike they are, the more alike the behavior will be (Kolers & Perkins, 1975). Judicious experimentation may allow one to infer their identity" (p. 443). They proceeded to talk of developing a taxonomy of "trainable capabilities" that can be organized in the performance of different tasks.
13 As Roediger et al. acknowledged, there is not a sharp distinction between perceptual and conceptual processes; they proposed, instead, a continuum.
By comparing performance on implicit memory tests that required semantic vs. perceptual processing, Blaxton (1989) obtained dissociations within implicit tasks. Specifically, she showed that generating words from conceptual cues as opposed to simply reading the words led to better recall on conceptual tests than on perceptual tests, whether episodic or semantic, whereas reading produced better recall on perceptual tests than conceptual ones, whether episodic or semantic. However, researchers also found dissociations between tasks thought to employ conceptual processes (Cabeza, 1994), evidence that did not fit well with the existing procedural accounts. In response, advocates of the procedural approach proposed subdividing the perceptual and conceptual processes into finer-grained processes (Roediger, Gallo, & Geraci, 2002).
The advocates of the processing approach viewed themselves as providing compelling evidence against the systems approach. Kolers and Roediger (1984) reasoned: "If dissociations are found among tests tapping the same memory system, then the discovery of dissociations between tasks cannot be taken as evidence for different memory systems" (p. 438). Nonetheless, advocates of the systems approach persisted in developing and refining their view. For example, in response to Blaxton's results noted above, Tulving and Schacter (1990) proposed their fourth memory system: the perceptual representation system.
One advantage the memory systems approach seemed to offer was an account of the overall capacity to perform a task. Performance on a given task was simply the result of the application of a whole memory system. The process approach seemed to offer only a hodgepodge of processes. Morris Moscovitch (1994) offered an alternative perspective that emphasized the integration of operations, an approach he termed components of processing. As the term processing in its name suggests, the components of processing framework is a descendent of the memory processes approach. It adds the idea that different tasks may draw differentially upon memory processes to create a processing network. If two tasks can be dissociated (by showing that a manipulation affects performance on one task but not on the other), then there must be at least one component process that figures differently in the two tasks (Hintzman, 1990). Within this framework, dissociations are no longer used to tease apart whole systems, but only differences in reliance on components.
Some researchers have denied that there is opposition between the memory systems and the components of processing approaches. Tulving, for example, maintains that "the classification approach complements the process-oriented approach to memory; it is not an alternative to it" (1991, emphasis in original). Indeed, when the components of processing advocates talk of recruiting different processes in performing a particular memory task, they can be construed as describing the building up of a memory system or mechanism. But the emphasis that components of processing researchers place on constituent operations constitutes an important difference in research objective. Their project is directed at mechanistic decomposition, not just phenomenal decomposition. Even so, however, the components of processing approach has provided little in the way of a specific account of what operations are involved in performing various memory tasks. This is not very surprising because, as I argue in the next section, this is a generic problem for cognitive science.
Was this article helpful?