In his recent article in the Psychonomic Bulletin & Review that stimulated this digital event, Nelson Cowan undertakes the impressive endeavor to disentangle the various definitions of working memory (WM) that have been around in the literature since the term WM was coined. He does so to counter the confusion that the use of different definitions has entailed.
It is a confusion I have felt myself, both when first confronted with the concept of working memory coming from a psycho-/neurolinguistics background, and when struggling to answer questions on what exactly working memory is in discussions with researchers from the learning sciences. My thoughts on Cowan’s review keep circling around two or three points mainly. Having failed to combine them into a coherent text, I will just list each of them in turn.
First, Nelson Cowan makes it very clear that there is a problem when the definition of a construct is almost equated with the theory that seeks to explain it. What I also read between the lines is that there is another danger in that a construct is almost equated with the paradigms applied to measure it. In working memory research, as in much of cognitive psychology, there is always a slight risk to focus on investigating and explaining participants’ behavior in our most favorite paradigm, like immediate serial recall or complex span, as an end in itself rather than to focus on the construct this paradigm is supposed to measure. Our theories may then perfectly account for why participants are able to recall a list of words interspersed with reading a list of sentences as in the reading span task, but cannot account equally well for the role of working memory in discourse comprehension or other complex cognitive tasks. There is a very good case for using paradigms that are as process-pure as possible. Nonetheless, there is also a danger in purposefully ignoring the peculiarities of the complex cognitive functions working memory is used for when this goes along with a concept of working memory that does not include these mental activities, as in some instances of the storage-and-processing definition (#5). Notably, such a storage-and-processing concept is quite different from what Daneman and Carpenter intended when they introduced the reading span task to investigate the role of working memory in reading comprehension in 1980. If we want working memory to remain useful, we may not lose the perspective of what cognitive functions it is working for or, crudely borrowing a phrase from a very distinct domain, we need to avoid alienating working memory from its work and its products…
A further problem that Nelson Cowan highlights is the distinction between the meaning of working memory and that of alternative terms such as short-term memory or immediate memory. One of the problems here might be that in some cases the usage of these alternative terms would also qualify them as working memory concepts. When reading Cowan’s article, I kept thinking of the work of Robert Jarvella in the 1970s and 1980s, whose primary interest was in establishing a theory of discourse processing and the role of immediate memory (Jarvella, 1979) or short-term memory (Jarvella, 1980) therein. Even though he did not use the term working memory, he exhibited a strong processing focus and used the running memory span paradigm (with connected discourse). It might therefore be of interest to categorize Jarvella’s understanding of immediate or short-term memory along the lines suggested by Cowan. Jarvella’s view is characterized by the following assumptions: (1) “In continuous listening and reading situations, it is common for us to remember exactly only the most recent bits of the message to which we are attending” (Jarvella, 1979, p. 379); (2) the forgetting of a previous sentence or clause is not “simply a consequence of hearing further speech” but rather “seems to depend on how actively the following segment is processed” (Jarvella, 1979, p. 382); and (3) “immediate memory for clauses and sentences heard in discourse depends partly on what happens in the just following context […] such that if “a following segment is largely independent and explicitly expresses primarily new or unrelated information, the preceding one’s form is more quickly forgotten“ (Jarvella, 1979, p. 383). I am not sure where this construct fits in best – maybe a combination of generic working memory (#6) and recent-event working memory (#4)? Nelson, I need your help here.
What are further implications of the issues Nelson Cowan discusses? There is a lot to learn for theoretical and empirical investigations on the nature of working memory, from just being reminded about the need to make clear what it is we want to explain, over the aim to agree on this definition, to a stricter separation of definitions, theories, and paradigms applied to test these theories. However, my worry is that the issue of different definitions of WM with different theories in tow is not restricted to the field of working memory itself but goes a lot further and is even more problematic when research in other areas builds on these definitions. This is, for instance, the case in applied research on instructional design, in which working memory and its limitations are assigned a critical role for learning. If you want to acknowledge the role working memory plays in learning from a multimedia presentation, a reasonable approach would be to rely on agreed-upon knowledge in the working memory literature and elaborate on that in your own area of interest. This is also the approach we as working memory researchers often take when we are interested in relations between working memory and attention, long-term memory, or language processing. Cowan points out that, within the field of working memory, we haven’t even agreed upon what we think of as working memory. So what is the agreed-upon knowledge that others can rely on? Given that so many definitions are circulating, it is almost impossible to adopt a theory-neutral definition to a different research area. Therefore, the definitions that are given in instructional design theories are often based on more than one theory and definition – and, to complicate matters, do not necessarily specify which part comes from which. As two examples, I have chosen the following definitions from two influential theories of learning and instruction, the Cognitive Theory of Multimedia Learning (e.g., Mayer, 2009) and the Cognitive Load Theory (e.g., Paas & Sweller, 2014).
“Working memory is used for temporarily holding and manipulating knowledge in active consciousness.” (Mayer, 2009, p. 62).
“Working memory: The cognitive structure in which we consciously process information. Notable for its severe capacity and duration limits when dealing with new information.” (Paas & Sweller, 2014, p. 41).
The fusion-like character of the adopted working memory construct that can already be seen in these definitions looks like a real mess when it comes to more specific aspects such as the nature of capacity limits. To me, a particularly telling example is the following detailed description within the Cognitive Load Theory:
“When dealing with novel, biologically secondary information, human working memory has two severe limitations. Miller (1956) indicated that working memory is able to hold only about 7 elements of information. It can probably process in the sense of combine, contrast or manipulate no more than about 2-4 novel elements. On these numbers, the capacity of working memory when dealing with new, biologically secondary information is severely constrained. The duration of working memory is also constrained. Peterson and Peterson (1959) found that, without rehearsal, almost all the contents of working memory are lost within about 20 seconds.” (Paas & Sweller, 2014, p. 33).
Reading these adaptations of the basic construct reminds me of Sir Frederic Bartlett’s participants in the serial reproduction paradigm, who had to recall the Indian folk story “The War of the Ghosts”, which originated from a cultural background distinct from their own. The clue of the paradigm was that only the first participant in a reproduction series read the original story, while the second participant read the first one’s recall of the story, the third participant read the second one’s version, and so on. In the end, just like in the children’s game of telephone (or Chinese whispers in British English and ‘Stille Post’ in German, respectively), the original meaning was significantly altered and the recalled story was much more in line with participants’ prior knowledge and beliefs than the original one. When I stumble upon distortions of the working memory construct in other research areas, I tend to blame these entirely on those who are borrowing, but upon reading Nelson’s paper, I couldn’t help but notice that I and others like me are just as much to blame.
Jarvella, R. J. (1979). Immediate memory and discourse processing. Psychology of Learning and Motivation, 13, 379-421.
Jarvella, R. J. (1980). Short-term memory during discourse processing. Advances in Psychology, 5, 221-225.
Mayer, R. E. (2009). Multimedia learning (2nd ed.). Cambridge: Cambridge University Press.
Paas, F. & Sweller, J. (2014). Implications of cognitive load theory for multimedia learning. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning: Second edition (pp. 27-42). Cambridge: Cambridge University Press.)