share interactive transcript request transcript/captions live captions
download
|
MyPlaylist
VALERIE CAMOS: [INAUDIBLE] told you that idea is that, especially for verbal material, we will have two mechanisms of maintenance or two systems of maintenance. And so one will be more an intentional demanding, domain general mechanism that, because it's domain general, it could also be used for verbal material and other domain-specific material. And usually, when you think about working memory, especially from the European/British tradition, you mostly speak about the phonological loop. And this is-- oops, this is not it. OK. Which way is it?
SPEAKER: The right one.
VALERIE CAMOS: OK. So among the different studies we did, what we tried to do is tried to understand how the two systems coexist, how they're going to interact. And so one of the first experiments we did was using this kind of device. Similarly, the same complex span task, in which participants maintain a series of letters. And between those letters, they've seen some digits, six digits.
And they have two types of tasks to do on these digits. Either they have to do a relatively low-demanding task which is a target detection. I mean, they just press a key when they see 5 on screen, which is relatively low demanding. Or they have a much more demanding task, which is a sum verification, verifying that this third digit, and the six digit is the addition of the two previous ones.
So this is the way we're manipulating, actually, the cognitive load. This is another way that we did not present it yet, but it's another way to manipulate the cognitive load. And in a cross design, orthogonally, we either ask participants to do this task silently or to read the digit on screen, which is going to induce an articulatory suppression and block a possible phonological loop. We also covaried the pace to verify if this could affect the trend or the relationship between the two systems.
And what we observed is that here, you've got the number of letters they've been able to recall. And you have for slow and fast pace. We have exactly the same pattern, so it's quite resistant for that.
And first, you have the classic effect on recall of the articulatory suppression. When people have to do something aloud, you have a reduction of recall. You can see in both. But also, we replicate another way of doing the cognitive load manipulation.
I mean, when you do a task that is more demanding, you have this decline in performance at any time. This is significant. This is significant. But never, never, no interaction between the two manipulations. So this was the first idea that the two systems are really independent, and work independently on the memory traits.
So if we have two systems, a central system that we know very well now is attentional demanding, and another a system which is kind like a phonological loop, this phonological loop is well-known to be affected by the characteristics, phonological characteristics of the memory item to maintain. And one of the two big effects of the phonological loop is the phonological similarity effect and the word length effect. The phonological similarity effect is that it's more difficult to maintain series of words when they are phonologically similar than dissimilar, because it induces some phonological confusion if they are phonologically similar.
The other effect is the word length effect. If you have longer words to maintain, it's longer to rehearse. And then on a constant duration, you are going to use less words than for short words. And then in this case at the end, this phonological loop was only allowed to maintain less long words than short words.
So in the similar design, we also manipulate these two different factors by presenting either list-- for example, in the first experiment, list of similar words versus list of dissimilar words. And we, once again, orthogonally vary the implication of the central system and of the phonological loop. We vary the implication of the attentional system by either having nothing to do, which is really the minimum cognitive load, the no cognitive load condition, or to ask our participants to do a location judgment task.
Here, it's pretty difficult, but the squares are either up or down on screen. And it's very subtle difference. It really has participants to be very careful in their response by pressing keys. This is the manipulation of the cognitive load.
And on the top of that, either they have a silent task-- here it was silent, here it was silent-- or would have to do an articulatory suppression. And to really control the pace of this articulatory suppression then to ensure that it's exactly the same amount in the different condition, they have to say yes, yes to a beep that gives them the rhythm to follow that. So here, they only have to do an articulatory suppression. And then the last condition, they have to do the joint manipulation of concurrent articulation and a cognitive load effect, or distraction of attention.
So what we see is we replicate the previous finding that is both the articulatory suppression induces reduction of recall performance in both cases. Introducing a concurrent task is going to reduce recall performance. You have this difference is significant and is different is also significant.
But the most interesting thing in this experiment is that the phonological similarity effect-- the difference between dissimilar and similar words-- occurred only when there was no articulatory suppression. I mean, in both, when there was nothing to do or even if they have something to do, but silent, you can observe this phonological similarity effect. As soon as you introduce an articulatory suppression, the phonological similarity effect disappears.
But it disappears independently-- you have still an effect of the attention of demand, which fits with the idea that you have two systems that are independent. And when you can use the phonological loop, you are sensitive to the phonological similarity on the top of the effect of the attention on demand. And actually, when we did the same experiment manipulating the word length, presenting a list of short words or list of long words, I actually really changed the slide. It's exactly the same findings.
So the word length effect occurred only when the task was silent on the top, and additively to the manipulation of the attentional demand of the concurrent task, which fits also nicely with this idea of having two independent systems, and also with our idea that maintenance of verbal information is not just the job of the phonological loop of a domain-specific system, but there is really other way to maintain verbal information, which is more central.
So we did a lot of things, mostly in my lab, about this, showing that it's independent or that people can jointly use this. Adults can make, also, adaptive choice. Depending on the type of material you give an adult, you will either use-- he or she will either choose to use phonological loop or the attentional system.
For example, if you gave an adult a list of phonologically similar words, it's pretty stupid to use your phonological loop, because you're going to introduce confusion. And you can really see in the data, they backup [INAUDIBLE] attentional system. And then suddenly, they become sensitive to concurrent attentional demand.
We also show there is also lead by different type of representation. We're just seeing for the phonological loop, I have some data showing that the central system is more in the semantic-- they store more semantic representation. And others than us have shown there is this distant brain system, brain networks, mostly in the [INAUDIBLE] Johnson lab. [INAUDIBLE] a few things about that. There's two different networks that sustain this. So we have a few bits of evidence about this.
So collecting all these things that we've been doing this past 10 years or so, we try to integrate all this in a cognitive architecture. And this is the last version of a model. Try to figure out how we can understand all this. And the idea is that "all this" is the working memory, kind of a center of the cognition for us.
So you have here the representation, working memory representation as PSA, built up like mountain model. The idea is a transient representation built at the moment. And representation files are only stored in this, in an episodic working memory buffer.
And this system can only maintain a short amount of information representation, probably around four, if we trust [INAUDIBLE] Cohen. But we quite trust him sometimes. Then you have here four representations stored here.
And those representations could be manipulated through a production system. We are very inspired by ACT-R, John Anderson model. And so this production system could either rebuild, reconstruct the representation. This is the so-called "storage function," which is actually a processing of reconstruction, or you process the information which is really transforming, changing the representation.
And it's all done through this loop that the working memory representation will be manipulated by the prediction system. And this loop is what we call an "executive loop." That's goes into many-- the central system is here.
And this is the central bottleneck. Because when this loop is used for storage, it cannot be used for processing, and vice versa. Besides that, you have what I just described, a phonological loop with a buffer and an articulatory rehearsal system, which is independent.
We also believe there is a visual spatial buffer, but without any domain-specific maintenance of the information. [INAUDIBLE] buffer, we are quite unclear on that. It's really inspired by ACT-R. And goal model-- sorry it's not very clear, but the goal model is to direct the activity of this executive loop, and also the idea that declarative long-term memory are going to feed to build up this representation. And it's something that we're currently working on.
So to just describe more the functioning of this, if you more or less familiar with ACT-R model is really based on that. The idea is that the current content of the working memory representation-- this is two ways of representing more or less the same things with two different designs-- the working memory representations are actually the condition of firing of the production rule. And the production rule will fire some executive function. Actually, executive functions are tools to transform the working memory representation, transform or keep it as same, but just like in processing.
So you have different [INAUDIBLE] in the representation of the working memory that will feed the condition of the prediction rule. The conditioned rule will just fire. And then you have some action. And this action is made true executive function.
So just another way, once again, to represent that-- you can see that through time, what's going on? You have a representation. It's going to be the condition of this prediction rule, firing an action.
This action will act on the-- will trigger an executive function, and act on the working memory representation, which means that we have a new working memory representation. This new working memory representation will be a condition of a new potential production rule that's going to fire another executive function that transforms or rebuilds the working memory, and so on, and so forth.
And this is why we call it an "executive loop," because it's just like a loop. But at the same time, it's using the executive function. And it's the idea that we do not have a central executive, as in [INAUDIBLE] as most people know since undergrad the way it works. But here, you really don't have a [INAUDIBLE]. But the executive control is emergent from the functioning of this loop. That's the idea we come up with. We try to proceed.
So that means that executive functions, they are currently defined mostly in the role psychology literature, updating, shifting, inhibition, should be especially important in-- or should mostly use this central, this executive loop. They're going to impaired-- I mean, no, no, I can't find an English word for that. They mostly involve this executive loop.
So if you really want to embed this central bottleneck within a model, introducing executive function within a complex span task will be the best, and a neat measure of the functioning of the executive loop. Because most of the function-- each time you do-- how do I explain this correctly? But each time you have to do a processing, there is other-- you know you have some multi activity that probably do not require attention so much.
What is going to require? Mostly, this central executive loop is the executive function. So if you want to have the cleanest measure of the functioning of the executive loop, it's better to use a task that will involve a lot of executive function. So this is exactly what we did.
In this series of experiments, I'm going to wrap up a large series of study very quickly. But the idea here was always-- so we have several experiments. And each time we have two conditions to compare, one that will involve an executive function, and a control one, very similar, but without executive control.
And so this is going to help us to vary the implication of the central bottleneck on the different conditions. I'm going to explain this with an example. For example, this Stroop-- everybody knows this one. You introduce a Stroop task within a complex span task.
Participants have to maintain digit while they have to name the color of the ink. Either you have this condition that would require to inhibit the reading of the words, which is inhibition one of the executive functions, or you have a control condition, in which we presented adjectives. So in this case, they do not have to inhibit anything. They just name the color.
So on each experiment, this is what we did. Another experiment with inhibition, you have, once again, participants maintain words. This is words in French, are words in French, triage.
And here, you have to name or to enumerate how many objects on screen. But here, it's digit. And so it's going to be more difficult. You have to inhibit to read the digit. And here it's letter, so you just have to say how many letters you see on screen.
And so this is what we did for different tasks. For this one, for example, what the model predicts is that if you look at the bar, it's the response time on the first that I showed you. When it's with adjective, it's shorter to response that color. It's the classic Stroop effect.
And conversely, the number of items that participants can maintain is reduced in this condition, because we introduce the use of the executive function inhibition, and then impaired the functioning of the center or this-- not distract attention, yeah, distract attention, or use, for longer duration, the central bottleneck of the executive loop. We have the same for this second experiment.
And so what we did in this large series of experiments, we each time used this response time as an evaluation of how much does this loop, the executive loop, was involved in the task. It's an estimation of this to have an approximation of the cognitive load. And our idea is that we should observe that the recall performance is directly predicted by this cognitive load.
And actually, you have the two experiments that I showed you were in green. They are here. So we did this with several experiments involving updating with response selection, with retrieval.
And all these different tasks, if you bring this back to the same measure of a cognitive load, evaluated by the sum of the response time divided by the total time, you can see that they nicely plot on the same line. And we still have 2% missing where it goes nowhere.
But it's quite nice. And it's exactly what we were predicting. And we think that we had such a nice result because we mostly use-- these executive function were especially used in the executive loop. If I have a bit of time, can I just-- yeah?
I don't know what time does it finish? Oh.
SPEAKER: [INAUDIBLE]
VALERIE CAMOS: Yeah. OK, so just a few words about development. Because at the beginning, all this start by development. That was the question of how we can understand the development of working memory.
So the first thing is are we going to observe the same kind of relations between the cognitive load and the recall performance in children, which index this functioning of an executive loop. And so as a first experiment, we used the same reading span task as we presented-- Pierre presented at the beginning, with-- we can't really see here, but eight years old, 11 years old, and 14 years old. And we vary the number of digits per time they have to read while they maintain letters.
The first thing that you can see is that at 8, 11, 14, we still replicate this linear relationship between the cognitive load and the amount of information children were able to maintain. The second thing you can see is that 8-year-olds recall less than the others, which is the thing to explain. But you'll see that the slope is going steeper and steeper with age.
And actually, adolescents got about the same relation as what we observe in undergrad, young adults. And the idea here is that because the mechanism of refreshing becomes more and more efficient or more and more used-- because functionally, that leads to the same prediction-- it's mostly the olders, the teenagers, that are more affected by the introduction of higher cognitive load.
It's only those who are going to use the most this tool that refreshing is, that are going to be more impaired when you distract attention, and they can reduce this refreshing. And the younger kids, they're probably less using it or they're less efficient of using it. And this is why they're less impaired in their recall when you distract their attention or when you impaired the mechanism of refreshing.
The ideas, next, is that is there, then, an age? If you start to think that the slope is becoming shallower and shallower when you go down on age, is there an age on which there is absolutely no refreshing? And in this case, we start to study with other type of material, 5 and 7 years old. They have to maintain this time animals presented by picture and names, and they have to name the color of smilings. There were smilings here.
We vary the number of colors presented between each animals, either one color in 2,000 milliseconds or two colors in 2,000 milliseconds or two in 4,000 The idea here is that if you compare the first one and the last, one color in 2,000 and two color in 4,000, it's the same cognitive load. You have to do the same things.
So here and there, there's exactly the same cognitive load. But when you do two color in 2,000, it's a higher cognitive load. And if children use refreshing, they should have lower recall performance here than in this, too.
Conversely, if you start to think of a child who's not going to have a refreshing mechanism, what would happen for our model? It will maintain the information. But they have nothing to refresh this information.
Because we predict a time-based decline, they will then forget this information. And more you wait, lower will be the recall performance. So if there is no refreshing in some age group, we should see that the children should be affected by only the duration. And so those two that last the same should not differ. And this one should lead to a lower performance, because it lasts longer.
So here's the result, contrasting 5 and 7 years old. Of course, there is a huge difference on their recall performance. But the pattern of result is really different.
At 7, we replicate a cognitive load effect. The two conditions with the same cognitive load did not vary, and the children had better recall performance than when they have a higher cognitive load. So exactly it has the 8-year-old in the previous experiment.
But for the 5-year-old, it was totally different. The two conditions with the same duration did not vary. And when you increase the duration, you observe a reduced performance, as if-- and this is what we think the children did not have the use of refreshing, did not have or do not use refreshing. Still, easy to understand this question.
SPEAKER: Yeah, it's interesting, if the refreshing is, in some cases, a kind of implicit reversal. This is exactly the age at which you see one and two-word rehearsals spontaneously emerging, like [INAUDIBLE].
VALERIE CAMOS: You mean verbally rehearse?
SPEAKER: Exactly.
VALERIE CAMOS: Exactly. It's exactly at the same age. But I do not think it's [INAUDIBLE] by the same type of mechanism. And also, there's currently a lot of debate about this age of rehearsal, so that could appear before that. So yeah, it's exactly at the same age. And this is the thing we are currently working on, trying to disentangle the effect of the two at the developmental level.
The last one is that our model could also make predictions about what will make the difference between two age groups. If you compare two age groups doing the same task, younger kid, older kid, the older kid, we know they're going to have a better recall performance. But at the same time, we know they are faster.
So the processing is going to be shorter for them than for the younger. And because the processing is also how much they're going to forget, that's for sure one source of difference. And that could explain why the older have better recall performance. They also have shorter period of decay.
The other thing is that because it's faster processing in older children, if you keep the duration constant-- which is always this in experiment design-- that means they're going to benefit, the older children benefit, for a longer period of refreshing. So they definitely benefit of this. They have lower decay, and they have more time to refresh. So at the end, that could explain why they have better recall performance. Overall, they are lower cognitive load.
So what would happen if we are able to maintain and to control and equate these differences? So in this really neat series of experiments, first of all, we contrast two age groups, 8 and 11. And on the first experiment, they did exactly the same task-- baseline maintaining letters.
And they have to add 1 to each digit, say 5, 8, and 3. And they do the same. And we have exactly the same presentation rate, present, and the gap between two digits, the duration, was exactly the same.
We also manipulated the different durations here to vary the pace. We replicated classic developmental differences, that the older children are better recall performance than the younger. And we have a pace effect in both age groups.
Now, when you measure the response time of adding 1 at 11 years old, it takes only-- it takes nearly 400 milliseconds less than at 8 years old. So that means at 11 years old, they will have less decay, as I just say, and more time, also, to reactivate the information. So what we did is that we're going to do, in the second experiment, asking the 11-year-old to do another task that would last the same duration as for the 8-year-old.
And it happens that if you ask 11 years old to do plus 2, it takes about the same duration as at 8 years old to add plus 1. So this way, you have the two tasks with the two processings that last the same. And we replicate the experiment this way.
For sake of comparison, here, you've got the 11 year old in the first experiment. The 8 year olds are here, and the 11 years old of this second experiment. So you can see very clearly that when you start to ask 11 years old to do a test that lasts the same duration as the task for the 8 year old, you have a reduction of the difference between the two age groups, and still pace effect on both.
But there is still an age-related difference between the two. Why so? We just say that 8 years old are slower. They have lower processing speed.
So when they refresh, we can also believe that they refresh less fast, or less efficiently, or they need more time to refresh the same amount of information as the 11 year old. So what we did in the pretest, we used several different tasks to measure processing speed, and to know what was the relationship between the processing speed at 8 years old compared to 11 years old.
That means if you read this graph, that means here when you are 11 years old, things you need 1,000 milliseconds needs 1,400 milliseconds if you're eight years old. That's the way you can read it. Because they are slower.
So we can tailor the time for refreshing to the 8-year-old. We can give 8 year old the same amount of time that they should need to perform the same job as an 11-year-old for the refreshing. And so using this function, you can use this, and then replicate once more this task.
Here 11 years old, they do the plus 2. Here, they do plus 1, so same duration of the processing. But here, the gap that's available or the time that's available to refresh, you give more time to the 8-year-old, and this time is tailored to the processing difference between the two age groups.
Once more, sake of comparison, you have the 8-year-old here from the previous experiment. And as you can see here, you have now no more age effect. It totally vanished between the two age groups, and you still have a pace effect.
So if we recap this, if you ask two age groups to do the same task, obviously, you have an age-related difference that is quite large. When you start, you just first equate the processing time, there's different reduce. But if you equate the processing time, and gave to the younger a timed that is tailored to the processing speed, the difference disappears.
So for us, this really pinpoints on the different source, potential different source of difference on development to understand working memory development. It's probably not the entire story. We don't think that it's so simple.
It's two age group between 8 and 11. Of course, we don't think that if you do it between a 7-years-old and 20, you're going to vanish everything like this. There's also strategy, and type of maintenance mechanism, and the knowledge that people can use that's going to be important.
So if I recap this talk, it's clear-- I mean, I hope we convinced you-- that time is an important parameter to understand the functioning of working memory, and I would even say cognition, but clearly. We show there is a processing storage trade-off that is time based. There's also the idea that there is a central interference.
I mean here, this loop, this executive loop, will be the source of this central interference between the verbal and visual spatial processing. And it's totally different from the classic [INAUDIBLE] idea of domain-specific mechanism. We bring some evidence there is a temporal decay and [INAUDIBLE] in working memory.
And there are two systems of maintenance, specific for verbal maintenance or the maintenance of verbal information. And we don't find any very evidence of this for the visual spatial information. And we also show that times is an important factor if we want to understand working memory development.
And so this is the last version of our model that we published in this book, that recaps and brings small thoughts about how we see the world in general, about working memory. And this could be only done because we have plenty of people around us that do this experiment. And they belong to this different lab.
And in case you don't know, Switzerland is this tiny spot here. And Pierre is in the University of Geneva here, I mean the University of Freiburg. And this is one of our favorite spots. And this is lovely inhabitant of Switzerland. Thanks a lot for the invitation.
[APPLAUSE]
Valérie Camos of the University of Fribourg presents research that elucidates the importance of time in the functioning of working memory. Also, she proposes that the executive-loop is a source of interference between verbal and visuo-spatial processing. Further, she provides evidence for two subsystems specific for verbal maintenance.
Recorded April 8, 2015 as part of the Human Development Outreach and Extension program.
