Notes On Jonides "Working Memory And Thinking"

Thinking recruits a complex set of mental processes. An underlying commonality, though, is the use of memory. "To see this extensive influence, let us examine three quite different cases of human reasoning to illustrate not only the critical role of memory in many complex cognitive tasks, but also the varied forms of memory that are required."

7.1 Thinking And Memory

The first task that Jonides considers is the Raven Progressive Matrices test. To do this task, you have to abstract rules governing patterns, and choose the missing pattern that fits these rules. What processes are required to do this? You have to identify the elements in each cell of the matrix, and you have to keep track of the characteristics of figures in each problem and of your best guess for the rules. What is meant by "keeping track"? -- the use of working memory. (Long term memory is going to be required too, but Jonides will not discuss it in any detail.)

The second kind of task that is considered is solving word problems that involve spatial relations among items. Common strategy to solve this kind of problem is to build a mental image -- which will require memory, not only to hold the image, but to let processes work to build it in the first place. In this sense, the use of memory is very similar to that required by the Raven task.

The third example is mental arithmetic. To solve such problems, a common strategy is the use of stages of computation. Working memory is a key resource for this type of strategy. "For a strategy like this to be effective, a memory system that can hold various pieces of information must be involved." Long term memory will be required too; this raises the question about how long term memory and working memory interact. "Long term memory supplies the knowledge, strategies, and skills that are needed to execute a solution. The actual computation is done by an execuitve processor, making use of partial information stored in a working memory buffer." Key features of this position: (1) two qualitatively distinct memories; (2) thinking is driven by interactions among these memories; (3) working memory includes a processing capability too.

7.2 Working Memory And Long-Term Memory

"Thinking requires memory. It is also compelling that the type of memory required is not unitary in character. There is need for both a long-term memory system that holds knowledge and skills and a working-memory system that can hold information briefly for present purposes." What other evidence supports this kind of distinction?

First, brain injured patients provide evidence for dissociations among memories. For instance, H.M.'s LTM was intact, but he could not acquire new information. Working memory, though, was normal. In contrast, K.F., with a different injury, has normal LTM (can learn new info), but has abnormal working memory as assessed by digit span. (In other words, Jonides is used a double dissociation from clinical neuropsychology to argue for distinct memory systems. It is important to note that the logic for doing this is not so cut and dry; see Shallice for an example of issues that have to be considered for this to work.)

A second type of evidence comes from the functional dissociation of memories from serial position curve evidence; in particular, the dependence of the primacy and recency effects on different independent variables. "Notice that the pattern of behavioral evidence just reviewed constitutes another double dissociation, similar to the one that was suggested by examining patients with amnesias. The double dissociation comes about because different behavioral variables have different effects on the serial position curve."

7.3 Working Memory In Thinking

What is the role of working memory in thinking? Working memory is required to manage goals and subgoals that appear during problem solving. So, individual differences in problem solving may be due to individual differences in working memory abilities. For example, some evidence suggests that age-related deficits on the Ravens test are due to working memory impairments.

One aspect of working memory ability is speed of processing. "That is, success in tasks that rely on working memory may depend on how quickly subjects can process the information involved in the task. Perhaps this is because the faster the processing, the less likely it is that the material in working memory will be forgotten."

From all of this, a reasonable prediction is that interfering with working memory will result in decreased ability to reason. This is difficult to test; one approach is a dual task methodology, where one task is of interest, and the other is designed to impair working memory. The evidence suggests that the prediction is likely true. Given that working memory is important in thinking, we might also ask whether it is important for anything else.

Working Memory In Language Comprehension

"Working memory must be recruited in the service of language comprehension." What is its role here? (1) "it might store partial informaiton about an utterance or a piece of printed text while the remainder of that utterance or text was encoded." (2) "comprehension processes might work with the information being temporarily stored to produce a coherent meaning for an entire utterance or a piece of text." Why? Ambiguous words might need to be reinterpreted as more information is supplied; this requires that the ambiguous words be held in memory.

There are indeed some high correlations between working memory span and tests of reading comprehension. Other results show that "suppressing working memory causes a reduction in the comprehension of text that is relatively difficult to comprehend under ordinary circumstances, but not text that is easy to understand. Thus, it may be that working memory is engaged when significant effort is required for language comprehension to be successful."

Consistent with this claim is evidence suggesting that brain-damage to working memory causes problems with the comprehension of complex sentences, but does not appear to affect the comprehension of simple sentences. "Working memory is involved in sentence comprehension, but only when sentence structure is sufficiently complex that many owrds have to be held in memory while the remainder of the sentence is perceived." But not all the evidence supports this view -- for instance, some brain damage to working memory has no noticeable effect on sentence comprehension!

A Working-Memory Theory

In this section, Jonides starts by describing Baddeley's theory of working memory, and then by elaborating this theory on the basis of new evidence. (As far as PSYCO 354 goes, the remaining material in the chapter provides a solid update on the use of functional dissociations by cognitive psychologists!)

"A proper theory of working memory must include both storage and processing components." In Baddeley's theory, this was evident in the proposal of a central executive responsible for operating on information and scheduling the allocation of attention, a phonological loop to store verbal information, and a visuospatial buffer to encode visual information. Note that semantic information is not explicitly encoded in this model.

Mental arithmetic provides one example of the theory in action: "the central executive is the seat of mental-arithmetic operations, drawing on information stored temporarily in a buffer that holds the numerical facts of the problem at hand."

What evidence supports the notion of the phonological loop? This loop, BTW, consists of two different components -- a phonological buffer, and a rehearsal process that recirculates (and therefore refreshes) the contents of the phonological store.

The phonological store is assumed to use a phonological code for items. This hypothesis is confirmed by work like Conrad's collection of confusion matrices, which "indicate that the code in which information is held in working memory must have some property of the sound of the letters. If so, then partial forgetting of a letter leaves some trace of part of its sound, leading to possible confusion with a similar-sounding alternative letter."

Other results show that "hindering someone from creating an articulatory code [causes] a decline in working-memory performance if the person tried to use an articulatory representation even in the face of interference."

How much information can be stored in the phonological loop? The old view, from Miller, is about 7 "chunks". This view needs updating, though. For example, words that take longer to pronounce are more difficult to recall. "An important implication of the effect of pronuciation lenght on recall is that the capacity of the phonological loop cannot simply be measured in chunks (for example, words if the stimuli are words), otherwise it would have been constant regardless of the lenght of pronunciation of a word." So there appear to be two limits on capacity: (1) Chunks, (2) limits imposed by rehearsal process; stuff too long to fit into the rehearsal "time slice" will not be stored properly. Rehearsal capacity is likely time based.

"Rehearsing verbal material from the buffer is assumed to refresh the strength of the memory trace." So, language centers in the brain must be very active in working memory processing. Brain imaging studies support this view. E.G., left hemisphere centers, in particular Broca's area, are very active during rehearsal. "Thus, rehearsal can be characterized as inner speech in the service of memory maintenance."

Is rehearsal distinct from the phonological buffer? Approach to this issue is functional dissociation again, nicely described by Jonides: "The logic is this: If there are two working-memory components (a buffer and a rehearsal process), one might be able to identify two experimental variables, one of which influcences storage of infomraiton in the buffer but not rehearsal, and the other of which influences rehearsal but not storage in the buffer." One study by Longoni, Richardson, and Aiello gets particular mention here -- they find memory is better for short words vs. long words, and that memory is better for lists of dissimilar sounding words vs. lists of similar sounding words. Importantly, though, "articulatory suppresion abolished the effect of word length, but had no such influence on the effect of phonemic similarity. This result suggests that articulatory suppression and word length both affect the same psychological process, different from the one affected by phonemic similarity."

Similar suport is found with brain injured patients. E.G., PV shoes a number of specific impairments indicating a special problem with rehearsal, and not specifically with the phonological buffer. Conclusion: rehearsal and the phonological buffer are distinct, functional components.

Some problems appear to require the construction of representations that preserve spatial relations; phonological code is not well-suited for this. Instead, the visuospatial buffer is assumed to be used for this kind of information. It is similar to Kosslyn's use of executive processing to convert propositional information into spatial. Is there evidence for its existence independent of the components that have been described so far?

Again, the move is to functional dissociations. Visual information interferes with spatial task, but not verbal; the reverse is true for verbal information. PET studies offer support too. For example, PET studies have been done of the "two-back" task, in which subjects have to say "yes" if a letter being presented now is the same as the letter presented two letters ago. PET reveals left hemisphere sites, including Broca's area, the frontal lobe, and the left parietal lobe, light up for this task. A similar task for visual information, though, lights up different areas -- this time in right hemisphere, 2 frontal lobe sites, a parietal lobe site, and an occipital lobe site. "One hypothesis consistent with these activation sites is that subjects create a mental image of the dot locations, using processes of the occipital lobe. They encode the locations of the dots from this image using processes of the parietal lobe, and they then store these encoded locations using frontal mechanisms."

Single cell recording techniques have also been used successfully. Eg, monkeys are trained to mve their eyes back to a remembered location; neurons in frontal cortex are wiretapped during this. Evidence suggests that individual cells appear to be used as memories for specific spatial locations.

Now, another elaboration of Baddeley's model is in order. Not all visual information is spatial (e.g., colour and shape). Is such information encoded in the spatial buffer, or in a separate store? Single cell recording techniques have been applied to this question. "In the first task, they [monkeys] must maintain memory of the spatial location of the stimulus, but in the second, they must maintain a representation of the stimulus' shape. In both cases, though, they are required to make a left or right eye movement as the response, thus equating the motor task." The result -- active neurons are found in different locations, depending on the task. Analogous results are found with PET studies of humans. The bottom line is that Baddeley needs to talk about at least two qualtiatively distinct visual buffers.

"This brings the number of buffers involved in working memory to at least three: a phonological loop, a spatial buffer, and a visual buffer." But this still must be incomplete. Sematnic or conceptual information must be involved too. But how? One proposal is that the central executive may translate codes for the three buffers into semantic information when this is required. A second proposal is for a conceptual buffer. "Compelling as it seems that there must be some conceptual code for informaiton in working memory, there has been surprisingly little research to uncover how this information is stored and accessed by central executive processes."

What about the central executive? It is likely a set of processes! "Understanding the nature of central executive processes will involve studying myriad processes in myriad tasks. It is perhaps for this reason that it is widely recognized that progress in understanding central executive processes has lagged behind progress in understanding the buffer systems that report to these processes.

But some characteristics can still be proposed. The central executive has to do goal management; "to keep the goals of the current situation in memory and to allow each to control behaivor in a manner that is coordinated with each of the other goals." A second function is to schedule the subcomponent processes that are required (in a specific order) to accomplish some goal. Scheduling, with training, can become automatic for some tasks. Controlled processes (in the Schneider/Shiffrin sense) tax the scheduling function. If scheduling is done, then the central executive must be involved in planning and in selective attention. (The latter because selective attention is used to focus on some operations, and ignore others, to allow process scheduling to come to pass.) This kind of processing is controlled by frontal lobes. Indeed, frontal lobe damage leads to problems with such behaviors -- high susceptibility to interruption, tendency to perseverate (i.e., to have troubles switching from an old goal to a new goal). "The results of studies of central executive processes paint a picture of a coordinating mechanism."


"Working memory is the engine of cognition."

Pearl Street | "An Invitation To Cognitive Science" Home Page | Dawson Home Page |