working memory

In a 1987 experiment (1), readers were presented with a text that included one or other of these sentences:

or

Both texts went on to say:

After reading the text, readers were asked if the word sweatshirt had appeared in the story. Now here is the fascinating and highly significant result: those who read that John had put on a sweatshirt responded “yes” more quickly than those who had read that he had taken off his sweatshirt.

Why is this so significant? Because it tells us something important about the reading process, at least in the minds of skilled readers. They construct mental models. If it was just a matter of the mechanical lower-order processing of letters and words, why would there be a difference in responses? Neither text was odd — John could as well have put on a sweatshirt before going out for a jog as taken it off — so there shouldn’t be a surprise effect. So what is it? Why is the word sweatshirt not as tightly / strongly linked in the second case as it is in the first? If they were purely textbase links (links generated by the textbase itself), the links should be equivalent. The difference in responses implies that the readers are making links with something outside the textbase, with a mental model.

Mental models, or as they are sometimes called in this context, situation models, are sometimes represented as lists of propositions, but in most cases it seems likely that they are actually analogue in nature. Thus the real world should be better represented by the situation model than by the text. Moreover, a spatial situation model will be similar in many ways to an image, with all the advantages that that entails.

All of this has relevance to two very important concepts: working memory and expertise.

Now, I’m always talking about working memory. This time I want to discuss not so much the limited attentional capacity that is what we chiefly mean by working memory, but another, more theoretical concept: the idea of long-term working memory.

Think about reading. To make sense of the text you need to remember what’s gone before — this is why working memory is so important for the reading process. But we know how limited working memory is; it can only hold a very small amount — is it really possible to hold all the information we need to make sense of what we’re reading? Shouldn’t there be constant delays as we access needed information from long-term memory? But there aren’t.

It’s suggested that the answer lies in the use of long-term working memory, a retrieval structure that keeps a network of linked propositions readily available.

Think about when you are studying / reading a difficult text in a subject you know well. Compare this to studying a difficult text in a subject you don’t know well. In the latter case, you may have to painfully backtrack, checking earlier statements, trying to remember what was said before, trying to relate what you are reading to things you already know. In the former case, you seem to have a vastly expanded amount of readily accessible relevant information, from the text itself and from your long-term memory.

The connection between long-term working memory and expertise is obvious. And expertise has already been conceptualised in terms of retrieval structures (see for example my article on expertise). In other words, you can increase your working memory in a particular domain by developing expertise, and the shortest route to developing expertise is to concentrate on building effective retrieval structures.

One of the areas where this is particularly crucial is that of reading scientific texts. Now we all know that scientific texts are much harder to process than, for example, stories. And there are several reasons for that. One is the issue of language: any science has its own technical vocabulary and you won't get far without knowing it. But another reason, far less obvious to the untutored, concerns the differences in structure — what may be termed differences of genre.

Now it might seem self-evident that stories are far simpler than science, than any non-fiction texts, and indeed a major distinction is usually made between narrative texts and expository texts, but it’s rather like the issue of faces and other objects. Are we specially good at faces because we're 'designed' to be (i.e., we have special 'expert' modules for processing faces)? Or is it simply that we have an awful lot of practice at it, because we are programmed to focus on human faces almost as soon as we are born?

In the same way, we are programmed for stories: right from infancy, we are told stories, we pay attention to stories, we enjoy stories. Stories have a particular structure (and within the broad structure, a set of sub-structures), and we have a lot of practice in that structure. Expository texts, on the other hand, don't get nearly the same level of practice, to the extent that many college students do not know how to handle them — and more importantly, don't even realize that that is what they're missing: a retrieval structure for the type of text they're studying.

Research with children has demonstrated that the ability to learn new words is greatly affected by working memory span - specifically, by how much information they can hold in that part of working memory called "phonological short-term memory". The constraining effect of working memory capacity on the ability to learn new words appears to continue into adolescence.

But, as you grow in experience, building a vocabulary, this constraint becomes less important. Because working memory capacity is measured in "chunks" - and the amount of information contained in a chunk is extremely malleable. To a large extent, developing chunking strategies is what memory improvement is all about.

In terms of learning another language, there are essentially four possible classes of word:

• words that are already familiar because they are the same in your native language (or another known language)
• words that are already familiar because they involve words that you already know in that language (e.g., learning a related verb form, or learning a word made up of two words you already know, such as sweat-shirt)
• words that resemble a known word with similar or related meaning (e.g., Russian garlo means throat, and the word garlo resembles the word gargle)
• words that have no ready association to known words

It appears that in these first three cases, the size of your phonological short-term memory is of no significant relevance. It is only in the last case - where the word cannot utilize any meaningful associations - that your phonological short-term memory capacity becomes important.

Fairly obviously, as your knowledge of language (your own and others) grows, the more meaningful associations you will be able to make, and the fewer new words will fall into this last, difficult, category.

This suggests, of course, the usefulness of a mnemonic strategy (specifically, the keyword strategy) in the last, difficult case.

The importance of phonological short-term memory is also greater for productive learning (learning to produce a language, i.e., speak or write it) than in receptive learning (learning to read or understand a language). For productive learning, the pronounceability of the new words is very important. The more easily pronounced, the more easily learnt.

Many people, particularly as they get older, have concerns about short-term memory problems: going to another room to do something and then forgetting why you’re there; deciding to do something, becoming distracted by another task, and then forgetting the original intention; uncertainty about whether you have just performed a routine task; forgetting things you’ve said or done seconds after having said or done them; thinking of something you want to say during a conversation, then forgetting what it was by the time it’s your turn to speak, and so on.

This is clearly an issue for many of us. Part of the reason, I believe, is simply that we expect too much from ourselves. For example, research has shown that even a very, very short delay between recalling an intention and being able to carry it out is sufficient to dramatically reduce the likelihood that you will remember to do the intended action — we are talking about a delay of only 10 seconds!

The problem is exacerbated by age (I’m not talking about advanced age — I’m afraid certain aspects of cognitive processing begin to decline as early as the 30s).

Part of the problem is also that we tend to believe that we don’t need to do anything to maintain a thought, particularly when it has “popped” into our minds easily. But current estimates are that unrehearsed information lingers in working memory for less than two seconds!

Some of these problems are dealt with in my article on action slips (these problems are not, strictly speaking, a failure of memory, but a failure in attention), and in my book on Remembering intentions.

But in this article I want to talk about another aspect: the relationship between working memory, and attention (and, as it happens, intelligence!).

In my article on working memory and intelligence I talk about the difference between crystallized and fluid intelligence — that fluid intelligence is probably a better measure of what we think of as “intelligence”, and that working memory capacity is often used synonymously with fluid intelligence. A new theory is that the relationship between working memory and fluid intelligence is due to the ability to control attention.

This theory emphasizes the role of attention in keeping information active (i.e. in working memory), and argues that working memory capacity is not, as usually thought, about the number of items or amount of information that can be held at one time. Instead, it reflects the extent to which a person can control attention, particularly in situations where there is competing information / demands.

I have to say that this makes an awful lot of sense to me. I can’t, in the space I have here, go into all the evidence for and against the theory, but here’s one situation which is interesting. The “cocktail party phenomenon” is a well-known method in psychology, whereby people are given two streams of audio, one for each ear, and instructed to listen only to one. At some point, the person’s name is spoken into the unattended stream, and about a third of people pick that up. In a recent take of that classic study, researchers compared the performance of people as a function of their working memory capacity. Only 20% of those with a high capacity heard their name in the unattended channel compared to 65% of low-capacity people. The point being that a critical aspect of good attentional control is the ability to block our irrelevant information.

This ability is one that we already know is worsened by increasing age.

The message from all this, I guess, is that:

• short-term memory problems are, by and large, attention problems.
• attention involves both the ability to keep focused on the information you want to keep active, and the ability to not be distracted by competing and irrelevant stimuli.
• you need to actively attend to keep information active, particularly as you get older.
• many of us over-estimate how much information we can keep active at one time.

And if you want strategies to help you keep more information active, I suggest you look at improving your ability to chunk, condense and label information. If you can reduce a chunk of information to a single label quickly, all you need to do is remember the label. (I explain all this at length in my book The Memory Key, but I’m afraid it needs far too much explanation to go into here).

Anyway, I hope this helps those of you (most of us!) with short-term memory problems.

This article originally appeared in the April 2005 newsletter.