So what’s this שטיק about working memory? Out of 71 posts on this site that we have published since starting out 9 months ago, 7 were about working memory or have at least mentioned it (here, here, here, here, here, and here). This is indicative of the importance of working memory to human cognition—and indeed, we have noted repeatedly that working memory explains roughly half the variance in IQ.
One thing that we have not covered before is that for working memory to be effective, it has to meet two competing and contradictory demands: On the one hand, working memory (WM) has to maintain information that is required for the task at hand, and it has to do so without distraction or inopportune obliviscence. On the other hand, working memory must get rid of information once it is no longer needed and replace it by new information. There is nothing worse than an intermediate result that is unnecessarily cluttering your memory while you are computing the sum of 327 and 756. Once you have carried the first “1” you want to forget about it as quickly as possible lest it interfere with the next carry (if there is one) or retention of the sum of the “ones”.
So working memory must be both static and dynamic at the same time.
How does our cognitive apparatus achieve those two conflicting goals?
There has been a fair amount of research activity directed at this question (full disclosure: I have contributed to that research). One idea that has emerged from this work is the notion of a “gate” that shuts out incoming perceptual stimuli when necessary to preserve WM contents, but that can be opened when new information is available and needs to be updated into memory.
A recent article in the Psychonomic Bulletin & Review provided another test of this “gating” idea by exploiting the fact that Hebrew is read (and written) from right to left, whereas English is read (and written) from left to right.
Researchers Kessler and Oberauer presented people with a WM updating task, in which participants had to keep track of the contents of 4 boxes on the screen. The contents were briefly presented for study and then disappeared. The basic idea is depicted in the figure below which shows the time line of a single trial:
Participants first memorize “P G S R”, which then disappears and is replaced by a sequence of updating steps. On each step, some letters appear and the participant has to replace their memorized content of the corresponding box with the new information. An asterisk denotes that the particular box need not be updated, and that the initial information needs to be retained. At each step, the participant presses a key when they have completed memory updating and then proceed to the next step. When all update steps are completed, people recall the letter in each box in response to the “?” probe.
For the above example, incidentally, the final answer is “Q J L F”.
In their most recent study, Kessler and Oberauer presented the letters either in English or in Hebrew to bilingual participants. The rationale of this manipulation was that participants would scan the boxes from left to right for [P G S R], and right to left for [ג ל ט מ].
How would this manipulation reveal the role of a gate that switched between protecting and updating of WM contents?
Kessler and Oberauer inferred the gate by considering the time it took people to process each updating step. Recall that participants had to press a key when they had updated their WM at each step, and that this updating involved a display that contained some new letters as well as some asterisks, such that all “*” boxes retained their previous content and only boxes that contained a letter were updated. In previous experiments by the same authors, participants’ response time could be explained by a simple equation that contained a penalty term for any switch between an item that didn’t require updating and one that did (and vice versa). This penalty was interpreted to represent the switch of the gate from protecting WM to permitting the intake of new information.
However, that interpretation was tied to the assumption that people processed the stimulus from left to right—thus, when shown [Q * L *] there would be two presumed gate switches before people updated the third box with “L”, whereas [* J B F] would require only one switch before the letter in the same third position (“B”) was processed.
In this latest study, the idea was put to a more precise test by presenting either English or Hebrew stimuli, grouped together in blocks so participants knew what language to expect. The idea was that if people scanned in the language-appropriate direction, then the updating step [* * F G] and the step [ג ל * *] should be identical in terms of gate switches.
The results were both quite complex but also pleasingly straightforward. The complexity arises when the data are considered at the level of mean response times for each of the 20 conditions in the study: For each language, there were 10 ways in which the updating steps could be performed, from [* * * *] to [A G J K] with 8 other permutations in between. Fortunately, that complexity can be ignored because of the fact that all 20 conditions were explained by a straightforward equation (a so-called “measurement model”) that assumed scanning in the language-appropriate direction and a penalty term for each switch of the presumed gate. Intriguingly, the penalty term (and other terms in the model) did not differ between languages; the only difference between languages was whether scanning proceeded from the left or from the right.
The study thus provides the explanation for several everyday phenomena: If your toddler ignores your calls to come to the dinner table, it is because she is protecting the contents of her WM from interfering stimuli.