context

Humans are the animals that manipulate their cognitive environment.

I reported recently on an intriguing study involving an African people, the Himba. The study found that the Himba, while displaying an admirable amount of focus (in a visual perception task) if they were living a traditional life, showed the same, more de-focused, distractible attention, once they moved to town. On the other hand, digit span (a measure of working memory capacity) was smaller in the traditional Himba than it was in the urbanized Himba.

This is fascinating, because working memory capacity has proved remarkably resistant to training. Yes, we can improve performance on specific tasks, but it has proven more difficult to improve the general, more fundamental, working memory capacity.

However, there have been two areas where more success has been found. One is the area of ADHD, where training has appeared to be more successful. The other is an area no one thinks of in this connection, because no one thinks of it in terms of training, but rather in terms of development — the increase in WMC with age. So, for example, average WMC increases from 4 chunks at age 4, to 5 at age 7, 6 at age 10, to 7 at age 16. It starts to decrease again in old age. (Readers familiar with my work will note that these numbers are higher than the numbers we now tend to quote for WMC — these numbers reflect the ‘magic number 7’, i.e. the number of chunks we can hold when we are given the opportunity to actively maintain them.)

Relatedly, there is the Flynn effect. The Flynn effect is ostensibly about IQ (specifically, the rise in average IQ over time), but IQ has a large WM component. Having said that, when you break IQ tests into their sub-components and look at their change over time, you find that the Digit Span subtest is one component that has made almost no gain since 1972.

But of course 1972 is still very modern! There is no doubt that there are severe constraints on how much WMC can increase, so it’s reasonable to assume we long since hit the ceiling (speaking of urbanized Western society as a group, not individuals).

It’s also reasonable to assume that WMC is affected by purely physiological factors involving connectivity, processing speed and white matter integrity — hence at least some of the age effect. But does it account for all of it?

What the Himba study suggests (and I do acknowledge that we need more and extended studies before taking these results as gospel), is that urbanization provides an environment that encourages us to use our working memory to its capacity. Urbanization provides a cognitively challenging environment. Our focus is diffused for that same reason — new information is the norm, rather than the exception; we cannot focus on one bit unless it is of such threat or interest that it justifies the risk.

ADHD shows us, perhaps, what can happen when this process is taken to the extreme. So we might take these three groups (traditional Himba, urbanized Himba, individuals with ADHD) as points on the same continuum. The continuum reflects degree of focus, and the groups reflect environmental effects. This is not to say that there are not physiological factors predisposing some individuals to react in such a way to the environment! But the putative effects of training on ADHD individuals points, surely, to the influence of the environment.

Age provides an intriguing paradox, because as we get older, two things tend to happen: we have a much wider knowledge base, meaning that less information is new, and we usually shrink our environment, meaning again that less information is new. All things being equal, you would think that would mean our focus could afford to draw in. However, as my attentive readers will know, declining cognitive capacity in old age is marked by increasing difficulties in ignoring distraction. In other words, it’s the urbanization effect writ larger.

How to account for this paradox?

Perhaps it simply reflects the fact that the modern environment is so cognitively demanding that these factors aren’t sufficient on their own to enable us to relax our alertness and tighten our focus, in the face of the slowdown in processing speed that typically occurs with age (there’s some evidence that it is this slowdown that makes it harder for older adults to suppress distracting information). Perhaps the problem is not simply, or even principally, the complexity of our environment, but the speed of it. You only have to compare a modern TV drama or sit-com with one from the 70s to see how much faster everything now moves!

I do wonder if, in a less cognitively demanding environment, say, a traditional Himba village, whether WMC shows the same early rise and late decline. In an environment where change is uncommon, it is natural for elders to be respected for their accumulated wisdom — experience is all — but perhaps this respect also reflects a constancy in WMC (and thus ‘intelligence’), so that elders are not disadvantaged in the way they may be in our society. Just a thought.

Here’s another thought: it’s always seemed to me (this is not in any way a research-based conclusion!) that musicians and composers, and writers and professors, often age very well. I’ve assumed this was because they are keeping mentally active, and certainly that must be part of it. But perhaps there’s another reason, possibly even a more important reason: these are areas of expertise where the proponent spends a good deal of time focused on one thing. Rather than allowing their attention to be diffused throughout the environment all the time, they deliberately shut off their awareness of the environment to concentrate on their music, their writing, their art.

Perhaps, indeed, this is the shared factor behind which activities help fight age-related cognitive decline, and which don’t.

I began by saying that humans are the animals that manipulate their cognitive environment. I think this is the key to fighting age-related cognitive decline, or ADHD if it comes to that. We need to be aware how much our brains try to operate in a way that is optimal for our environment — meaning that, by controlling our environment, we can change the way our brain operates.

If you are worried about your ‘scattiness’, or if you want to prevent or fight age-related cognitive decline, I suggest you find an activity that truly absorbs and challenges you, and engage in it regularly.

The increase in WMC in Himba who moved to town also suggests something else. Perhaps the reason that WM training programs have had such little success is because they are ‘programs’. What you do in a specific environment (the bounds of a computer and the program running on it) does not necessarily, or even usually, transfer to the wider environment. We are contextual creatures, used to behaving in different ways with different people and in different places. If we want to improve our WMC, we need to incorporate experiences that challenge and extend it into our daily life.

This, of course, emphasizes my previous advice: find something that absorbs you, something that becomes part of your life, not something you 'do' for an hour some days. Learn to look at the world in a different way, through music or art or another language or a passion (Civil War history; Caribbean stamps; whatever).

You can either let your cognitive environment shape you, or shape your cognitive environment.

Do you agree? What's your cognitive environment, and do you think it has affected your cognitive well-being?

In a recent news report, I talked about how walking through doorways creates event boundaries, requiring us to update our awareness of current events and making information about the previous location less available. I commented that we should be aware of the consequences of event boundaries for our memory, and how these contextual factors are important elements of our filing system. I want to talk a bit more about that.

One of the hardest, and most important, things to understand about memory is how various types of memory relate to each other. Of course, the biggest problem here is that we don’t really know! But we do have a much greater understanding than we used to do, so let’s see if I can pull out some salient points and draw a useful picture.

Let’s start with episodic memory. Now episodic memory is sometimes called memory for events, and that is reasonable enough, but it perhaps gives an inaccurate impression because of the common usage of the term ‘event’. The fact is, everything you experience is an event, or to put it another way, a lifetime is one long event, broken into many many episodes.

Similarly, we break continuous events into segments. This was demonstrated in a study ten years ago, that found that when people watched movies of everyday events, such as making the bed or ironing a shirt, brain activity showed that the event was automatically parsed into smaller segments. Moreover, changes in brain activity were larger at large boundaries (that is, the boundaries of large, superordinate units) and smaller at small boundaries (the boundaries of small, subordinate units).

Indeed, following research showing the same phenomenon when people merely read about everyday activities, this is thought to reflect a more general disposition to impose a segmented structure on events and activities (“event structure perception”).

Event Segmentation Theory postulates that perceptual systems segment activity as a side effect of trying to predict what’s going to happen. Changes in the activity make prediction more difficult and cause errors. So these are the points when we update our memory representations to keep them effective.

Such changes cover a wide gamut, from changes in movement to changes in goals.

If you’ve been following my blog, the term ‘updating’ will hopefully bring to mind another type of memory — working memory. In my article How working memory works: What you need to know, I talked about the updating component of working memory at some length. I mentioned that updating may be the crucial component behind the strong correlation between working memory capacity and intelligence, and that updating deficits might underlie poor comprehension. I distinguished between three components of updating (retrieval; transformation; substitution), and how transformation was the most important for deciding how accurately and how quickly you can update your contents in working memory. And I discussed how the most important element in determining your working memory ‘capacity’ seems to be your ability to keep irrelevant information out of your memory codes.

So this event segmentation research suggests that working memory updating occurs at event boundaries. This means that information before the boundary becomes less accessible (hence the findings from the walking through doorways studies). But event boundaries relate not only to working memory (keeping yourself updated to what’s going on) but also to long-term storage (we’re back to episodic memory now). This is because those boundaries are encoded particularly strongly — they are anchors.

Event boundaries are beginnings and endings, and we have always known that beginnings and endings are better remembered than middles. In psychology this is known formally as the primacy and recency effects. In a list of ten words (that favorite subject of psychology experiments), the first two or three items and the last two or three items are the best remembered. The idea of event boundaries gives us a new perspective on this well-established phenomenon.

Studies of reading have shown that readers slow down at event boundaries, when they are hypothesized to construct a new mental model. Such boundaries occur when the action moves to a new place, or a new time, or new characters enter the action, or a new causal sequence is begun. The most important of these is changes in characters and their goals, and changes in time.

As I’ve mentioned before, goals are thought to play a major role (probably the major role) in organizing our memories, particularly activities. Goals produce hierarchies — any task can be divided into progressively smaller elements. Research suggests that higher-order events (coarse-grained, to use the terminology of temporal grains) and lower-order events (fine-grained) are sensitive to different features. For example, in movie studies, coarse-grained events were found to focus on objects, using more precise nouns and less precise verbs. Finer-grained events, on the other hand, focused on actions on those objects, using more precise verbs but less precise nouns.

The idea that these are separate tasks is supported by the finding of selective impairments of coarse-grained segmentation in patients with frontal lobe lesions and patients with schizophrenia.

While we automatically organize events hierarchically (even infants seem to be sensitive to hierarchical organization of behavior), that doesn’t mean our organization is always effortlessly optimal. It’s been found that we can learn new procedures more easily if the hierarchical structure is laid out explicitly — contrariwise, we can make it harder to learn a new procedure by describing or constructing the wrong structure.

Using these hierarchical structures helps us because it helps us use knowledge we already have in memory. We can co-op chunks of other events/activities and plug them in. (You can see how this relates to transfer — the more chunks a new activity shares with a familiar one, the more quickly you can learn it.)

Support for the idea that event boundaries serve as anchors comes from several studies. For example, when people watched feature films with or without commercials, their recall of the film was better when there were no commercials or the commercials occurred at event boundaries. Similarly, when people watched movies of everyday events with or without bits removed, their recall was better if there were no deletions or the deletions occurred well within event segments, preserving the boundaries.

It’s also been found that we remember details better if we’ve segmented finely rather than coarsely, and some evidence supports the idea that people who segment effectively remember the activity better.

Segmentation, theory suggests, helps us anticipate what’s going to happen. This in turn helps us adaptively create memory codes that best reflect the structure of events, and by simplifying the event stream into a number of chunks of which many if not most are already encoded in your database, you save on processing resources while also improving your understanding of what’s going on (because those already-coded chunks have been processed).

All this emphasizes the importance of segmenting well, which means you need to be able to pinpoint the correct units of activity. One way we do that is by error monitoring. If we are monitoring our ongoing understanding of events, we will notice when that understanding begins to falter. We also need to pay attention to the ordering of events and the relationships between sequences of events.

The last type of memory I want to mention is semantic memory. Semantic memory refers to what we commonly think of as ‘knowledge’. This is our memory of facts, of language, of generic information that is untethered from specific events. But all this information first started out as episodic memory — before you ‘knew’ the word for cow, you had to experience it (repeatedly); before you ‘knew’ what happens when you go to the dentist, you had to (repeatedly) go to the dentist; before you ‘knew’ that the earth goes around the sun, there were a number of events in which you heard or read that fact. To get to episodic memory (your memory for specific events), you must pass through working memory (the place where you put incoming information together into some sort of meaningful chunk). To get to semantic memory, the information must pass through episodic memory.

How does information in episodic memory become information in semantic memory? Here we come to the process of reconstruction, of which I have often spoken (see my article on memory consolidation for some background on this). The crucial point here is that memories are rewritten every time they are retrieved.

Remember, too, that neurons are continually being reused — memories are held in patterns of activity, that is, networks of neurons, not individual neurons. Individual neurons may be involved in any number of networks. Here’s a new analogy for the brain: think of a manuscript, one of those old parchments, so precious that it must be re-used repeatedly. Modern technology can reveal those imperfectly erased hidden layers. So the neural networks that are memory codes may be thought of as imposed one on top of each other, none of them matching, as different patterns re-use the same individual neurons. The strongest patterns are the most accessible; patterns that are most similar (use more of the same neurons) will provide the most competition.

So, say you are told by your teacher that the earth goes around the sun. That’s the first episode, and there’ll be lots of contextual detail that relates to that particular event. Then on another occasion, you read a book showing how the earth goes around the sun. Again, lots of episodic detail, of which some will be shared with the first incident, and some will be different. Another episode, more detail, some shared, some not. And so on, again and again, until the extraneous details, irrelevant to the fact and always different, are lost, while those details that common to all the episodes will be strong, and form a new, tight chunk of information in semantic memory.

Event boundaries start off with an advantage — they are beginnings or endings, to which we are predisposed to attend (for obvious reasons). So they start off stronger than other bits of information, and by their nature are more likely to be vital elements, that will always co-occur with the event. So — if you have chosen your boundaries well (i.e., they truly are vital elements) they will become stronger with each episode, and will end up as vital parts of the chunk in semantic memory.

Let’s connect that thought back to my comment that the most important difference between those with ‘low’ working memory capacity and those with ‘high’ capacity is the ability to select the ‘right’ information and disregard the irrelevant. Segmenting your events well would seem to be another way of saying that you are good at selecting the changes that are most relevant, that will be common to any such events on other occasions.

And that, although some people are clearly ‘naturally’ better at it, is surely something that people can learn.

On a number of occasions I have reported on studies showing that people with expertise in a specific area show larger gray matter volume in relevant areas of the brain. Thus London taxi drivers (who are required to master “The Knowledge” — all the ways and byways of London) have been found to have an increased volume of gray matter in the anterior hippocampus (involved in spatial navigation). Musicians have greater gray matter volume in Broca’s area.

Other research has found that gray matter increases in specific areas can develop surprisingly quickly. For example, when 19 adults learned to match made-up names against four similar shades of green and blue in five 20-minute sessions over three days, the areas of the brain involved in color vision and perception increased significantly.

This is unusually fast, mind you. Previous research has pointed to the need for training to extend over several weeks. The speed with which these changes were achieved may be because of the type of learning — that of new categories — or because of the training method used. In the first two sessions, participants heard each new word as they regarded the relevant color; had to give the name on seeing the color; had to respond appropriately when a color and name were presented together. In the next three sessions, they continued with the naming and matching tasks. In both cases, immediate feedback was always given.

But how quickly brain regions may re-organize themselves to optimize learning of a specific skill is not the point I want to make here. Some new research suggests our ideas of cortical plasticity need to be tweaked.

In my book on note-taking, I commented on how emphasis of some details (for example by highlighting) improves memory for those details but reduces memory of other details. In the same way, increase of one small region of the brain is at the expense of others. If we have to grow an area for each new skill, how do we keep up our old skills, whose areas might be shrinking to make up for it?

A rat study suggests the answer. While substantial expertise (such as our London cab-drivers and our professional musicians) is apparently underpinned by permanent regional increase, the mere learning of a new skill does not, it seems, require the increase to endure. When rats were trained on an auditory discrimination task, relevant sub-areas of the auditory cortex grew in response to the new discrimination. However, after 35 days the changes had disappeared — but the rats retained their new perceptual abilities.

What’s particularly interesting about this is what the finding tells us about the process of learning. It appears that the expansion of bits of the cortex is not the point of the process; rather it is a means of generating a large and varied set of neurons that are responsive to newly relevant stimuli, from which the most effective circuit can be selected.

It’s a culling process.

This is the same as what happens with children. When they’re young, neurons grow with dizzying profligacy. As they get older, these are pruned. Gone are the neurons that would allow them to speak French with a perfect accent (assuming French isn’t a language in their environment); gone are the neurons that would allow them to finely discriminate the faces of races other than those around them. They’ve had their chance. The environment has been tested; the needs have been winnowed; the paths have been chosen.

In other words, the answer’s not: “more” (neurons/connections); the answer is “best” (neurons/connections). What’s most relevant; what’s needed; what’s the most efficient use of resources.

This process of throwing out lots of trials and seeing what wins, echoes other findings related to successful learning. We learn a skill best by varying our practice in many small ways. We learn best from our failures, not our successes — after all, a success is a stopper. If you succeed without sufficient failure, how will you properly understand why you succeeded? How will you know there aren’t better ways of succeeding? How will you cope with changes in the situation and task?

Mathematics is an area in which this process is perhaps particularly evident. As a student or teacher, you have almost certainly come across a problem that you or the student couldn’t understand when expressed in one way, and maybe several different ways. Until, at some point, for no clear reason, understanding ‘clicks’. And it’s not necessarily that this last way of expressing / representing it is the ‘right’ one — if it had been presented first, it may not have had that effect. The effect is cumulative — the result of trying several different paths and picking something useful from each of them.

In a recent news item I reported on a finding that people who learned new sequences more quickly in later sessions were those whose brains had displayed more 'flexibility' in the earlier sessions — that is, different areas of the brain linked with different regions at different times. And most recently, I reported on a finding that training on a task that challenged working memory increased fluid intelligence in those who improved at the working memory task. But not everyone did. Those who improved were those who found the task challenging but not overwhelming.

Is it too much of a leap to surmise that this response goes hand in hand with flexible processing, with strategizing? Is this what the ‘sweet spot’ in learning really reflects — a level of challenge and enjoyability that stimulates many slightly different attempts? We say ‘Variety is the spice of life’. Perhaps we should add: ‘Variety is the key to learning’.

I recently reported on a finding that memories are stronger when the pattern of brain activity is more closely matched on each repetition, a finding that might appear to challenge the long-standing belief that it’s better to learn in different contexts. Because these two theories are very important for effective learning and remembering, I want to talk more about this question of encoding variability, and how both theories can be true.

First of all, let’s quickly recap the relevant basic principles of learning and memory (I discuss these in much more detail in my books The Memory Key, now out-of-print but available from my store as a digital download, and its revised version Perfect Memory Training, available from Amazon and elsewhere):

network principle: memory consists of links between associated codes

domino principle: the activation of one code triggers connected codes

recency effect: a recently retrieved code will be more easily found

priming effect: a code will be more easily found if linked codes have just been retrieved

frequency (or repetition) effect: the more often a code has been retrieved, the easier it becomes to find

spacing effect: repetition is more effective if repetitions are separated from each other by other pieces of information, with increasing advantage at greater intervals.

matching effect: a code will be more easily found the more the retrieval cue matches the code

context effect: a code will be more easily found if the encoding and retrieval contexts match

Memory is about two processes: encoding (the way you shape the memory when you put it in your database, which includes the connections you make with other memory codes already there) and retrieving (how easy it is to find in your database). So making a ‘good’ memory (one that is easily retrieved) is about forming a code that has easily activated connections.

The recency and priming effects remind us that it’s much easier to follow a memory trace (by which I mean the path to it as well as the code itself) that has been activated recently, but that’s not a durable strength. Making a memory trace more enduringly stronger requires repetition (the frequency effect). This is about neurobiology: every time neurons fire in a particular sequence, it makes it a little easier for it to fire in that way again.

Now the spacing effect (which is well-attested in the research) seems at odds with this most recent finding, but clearly the finding is experimental evidence of the matching and context effects. Context at the time of encoding affects the memory trace in two ways, one direct and one indirect. It may be encoded with the information, thus providing additional retrieval cues, and it may influence the meaning placed on the information, thus affecting the code itself.

It is therefore not at all surprising that the closer the contexts, the closer the match between what was encoded and what you’re looking for, the more likely you are to remember. The thing to remember is that the spacing effect does not say that it makes the memory trace stronger. In fact, most of the benefit of spacing occurs with as little as two intervening items between repetitions — probably because you’re not going to benefit from repeating a pattern of activation if you don’t give the neurons time to reset themselves.

But repeating the information at increasing intervals does produce better learning, measured by your ability to easily retrieve the information after a long period of time (see my article on …), and it does this (it is thought) not because the memory trace is stronger, but because the variations in context have given you more paths to the code.

This is the important thing about retrieving: it’s not simply about having a strong path to the memory. It’s about getting to that memory any way you can.

Let’s put it this way. You’re at the edge of a jungle. From where you stand, you can see several paths into the dense undergrowth. Some of the paths are well-beaten down; others are not. Some paths are closer to you; others are not. So which path do you choose? The most heavily trodden? Or the closest?

If the closest is the most heavily trodden, then the choice is easy. But if it’s not, you have to weigh up the quality of the paths against their distance from you. You may or may not choose correctly.

I hope the analogy is clear. The strength of the memory trace is the width and smoothness of the path. The distance from you reflects the degree to which the retrieval context (where you are now) matches the encoding context (where you were when you first input the information). If they match exactly, the path will be right there at your feet, and you won’t even bother looking around at the other options. But the more time has passed since you encoded the information, the less chance there is that the contexts will match. However, if you have many different paths that lead to the same information, your chances of being close to one of them obviously increases.

In other words, yes, the closer the match between encoding and retrieval context, the easier it will be to remember (retrieve) the information. And the more different contexts you have encoded with the information, the more likely it is that one of those contexts will match your current retrieval context.

A concrete example might help. I’ve been using a spaced retrieval program to learn the basic 2200-odd Chinese characters. It’s an excellent program, and groups similar-looking characters together to help you learn to distinguish them. I am very aware that every time a character is presented, it appears after another character, which may or may not be the same one it appeared after on an earlier occasion. The character that appeared before provides part of the context for the new character. How well I remember it depends in part on how often I have seen it in that same context.

I would ‘learn’ them more easily if they always appeared in the same order, in that the memory trace would be stronger, and I would more easily and reliably recall them on each occasion. However in the long-term, the experience would be disadvantageous, because as soon as I saw a character in a different context I would be much less likely to recall it. I can observe this process as I master these characters — with each different retrieval context, my perception of the character deepens as I focus attention on different aspects of it.