Chunking describes the process by which individual pieces of information are broken down and grouped together. The process is said to make the recall of information easier because it helps to bypass the inherent limitations of working memory. However, chunking also relies heavily upon long-term memory.
Chunking is, therefore, related to another aspect of memory known as memory span or digit span (the latter term derives from tests that measure memory span by asking people to repeat back a list of digits that increase incrementally). In education, we often describe these terms in relation to cognitive load, or the demands learning places on our limited mental resources.
The way instructional design can be adapted to help learners cope with these limitation has been encapsulated into a model known as Cognitive Load Theory (CLT). In turn, CLT draws on aspects of memory research that go back as far as the late nineteenth century, including studies that attempted to discover the limitations of human memory.
This is all well and good, but what exactly is a chunk and how does chunking help us enhance our limited working memory?
Unfortunately, there isn’t a great deal of consensus on this, even Miller in his famous 1956 paper on capacity noted ‘we are not very definite about what constitutes a chunk of information.’ Cowan (2001) has defined a chunk as ‘groups of items that have strong, preexisting associations to each other but weak associations to other items’. Cowan, like many other theorists, emphasise the role of long-term memory, in that a chunk must be representative of something we already know. Anderson’s ACT-R model, however, views chunks as schema-like structures containing pointers that encode their contents. This latter definition does make sense, in that cognitive schemas and chunking can be thought of as similar mental workarounds.
For us to begin to understand the nature of a chunk, we first need to step slightly away and look at the nature of working memory capacity and why it can be so problematic.
Researchers have known for some time that short-term memory is limited. Possibly the first study into such limitation even pre-dates the separation of memory into short-term and long-term components (the first model to incorporate this separation appeared in 1968, although there was general agreement that memory was probably organised in such a way well before Atkinson and Shiffrin’s multi-store model).
A paper published in the January 1887 edition of Mind by Joseph Jacobs details a study of what is now referred to as memory span. Jacobs initially asked his participants (students from North London Collegiate College) to repeat back nonsense syllables (a technique already utilised by experimental psychologist Hermann Ebbinghaus) but found they varied too greatly in relative difficulty of pronunciation and ‘relative facility to rhythm.’ He, therefore, abandoned this idea and instead chose letters (with the exception of W) and numbers (with the exception of seven). These omissions were necessary to ensure equal length and speed of vocalisation. He found that, on average, participants could successfully repeat back around 7 letters and 9 numbers.
Due to the behaviourist ascendency of the early part of the twentieth century, the study of memory fell out of fashion, predominately because of its internal and, therefore, unobservable nature. However, with the shift towards cognitive psychology from around the 1950s, there was renewed interest in mental functions, including memory, attention and perception.
George Miller’s 1956 study remains one of the most cited in the history of psychology, while Jacobs’ has fallen into relative obscurity. Miller found that, on average, people were able to hold between 5 and 9 pieces of information in short-term memory at any one time, or more precisely, 7 +/- 2 chunks. According to Miller, the capacity of verbal short-term memory is determined by the number of chunks that can be stored in memory, and not the number of items or the amount of information. More recently, Cowan has suggested that the capacity of short-term memory is closer to 4 chunks.
As an aside, these results are highly consistent, both in formal and less controlled settings. I’ve been using the digit span technique with A-level psychology students since 2004 and it’s remarkable how averages rarely shift too far from the research literature.
What’s your digit span?
This idea of digit, or memory, span might need a little unpacking, especially as Miller was a little vague about what constituted a chunk.
If I were to read out a series of numbers, lets say 58316724 and ask you to immediately say them back to me in order of presentation (a technique known as serial recall testing), how many you manage to recall before any error occurred would represent your digit span. If you repeated back 583167, your digit span would be six (the actual procedure would consist of several trials). I could also perform the study with letters, for example FBICIATV.
If you have a digit span of 6 and I ask you to repeat back 8 numbers or letters, the first 6 will, in effect, leave no room for the remaining two. However, what if we could recode the information so that it took up less space? Using a computer analogy, could we somehow compress the data? Lets look at those numbers again.
We could group some of these digits together, such as:
58 31 67 24 or
The first option then recodes 8 digits into 4 pieces of information and the second option into 2 pieces of information.
Chunking or Grouping?
These, then, may (or may not) represent chunks, because Miller makes the distinction between grouping and chunking. The numbers are unlikely to be meaningful so there is no representation of them stored in long-term memory. We may remember them better simply due to the pauses we place between the numbers, very much like the way we remember telephone numbers. Often this grouping of numbers is erroneously described as chunking.
Let’s, then, take a look at the letters.
We could group the letters thus:
FBI CIA TV or even
FB ICI ATV
Chances are the first option in represented in the majority of people’s long-term memory, while the second may or may not be (depending on whether your young enough to make Facebook out of FB, know that ICI is a chemical company or are old enough to make Associated Television out of ATV, (or adventurous enough to drive an All Terrain Vehicle).
We, therefore, need the knowledge of these pieces of information already in our long-term memory, otherwise they remain meaningless or, in Cowan’s words, they do not have ‘strong, preexisting associations to each other’ – they are just a string of random letters that have been grouped.
Back to the numbers. In fact, here is a new sequence:
There are twenty numbers here and you need to recall them all in sequence without writing them down. We can, however, group them:
1066 1914 1918 1939 1945
In all probability, these are dates that are represented in your long-term memory (so, yes, they are chunks) and you’ve just increased your digit span from 6 to 20, well, kind of. The thing is that you haven’t actually increased your working memory capacity, you’ve just discovered a way to get around its limitations.
We, therefore, use what we know in order to chunk information together. In a paper from 1980, Ericsson, Chase and Faloon describe the participant ‘SF’(later revealed to be Steve Faloon), an avid runner, who made extensive use of his knowledge of running times to recall up to 79 digits. Interestingly, however, he was unable to increase the number of consonants he was able to recall, probably because he was unable to associate them pre-stored long-term information.
The ability to successfully chunk is therefore related to what we already know. This also fits with the widely held notion that learning is cumulative – the more we learn, the more we are able to learn. Learning also becomes quicker and more efficient.
This deeper understanding of chunking can then be used to better implement learning strategies that take advantage of the way long-term memories can be linked to newly acquired information.
If you enjoyed reading this, you might like to visit my TES page
One thought on “When is a chunk not a chunk?”
[…] What counts as a ‘piece’ of information? […]