Aug 03 2017


brainpuzzleIf I gave you a string of digits to remember, how many do you think you could handle? For example, try to remember the number – 8945557302. That’s 10 digits. Most people can handle only 7, and there is a specific neurological reason for this. Our working memory is wired for about 7 bits of information (give or take 1-2 bits). Now, try to remember the number as 894-555-7302. That is recognizable as a phone number, and despite the fact that the individual digits exceed our bit capacity, most people can remember such numbers.

Grouping bits of information into recognizable patterns in order to make them easier to remember is a phenomenon called chunking, first described by George Miller in his 1956 book, The Magical Number Seven, Plus or Minus Two: Some Limits on our Capacity for Processing Information.

Chunking is best established as a mnemonic device. Miller and others conducted experiments to see how much chunking as a strategy could extend the limits of human memory. For example, the typical number of binary bits, such as Morse code, people can remember is 9. However, someone who understands morse code will chunk the individual bits into groupings of three, each one representing a letter. They will then group letters into words and words into sentences.  This type of chunking extended memory from 9 to 40 binary bits.

Another study involved a runner tasked to remember strings of digits, again starting with a typical digit span of 7. However, he was encouraged to chunk the digits into race times – a meaningful pattern with which he was very familiar. With a little practice he increased his digit span to 80 numbers.

What these and other studies show is that chunking data into groups is a very effective strategy for managing large sets of information. Further, the more meaningful the groupings, and the more familiar the patterns, the more effective this strategy is. Meaningful patterns are important. For example, chess masters can famously memorize entire chess boards, with the position of every piece on the board. However, they cannot remember random placings of chess pieces, only meaningful placements that result from actual game play.

The phenomenon of chunking extends beyond bits of information. We also chunk motor skills, by breaking down a complex task into definable combinations of movements.

The deeper question is – why is chunking so universal and so useful? It seems likely that is results from a basic fact of how our brains process information. This fits with the independent research that shows that our brains are essentially pattern-recognition machines. The neurons of our brains are organized into massive parallel processors that seek recognizable patterns – visual, auditory, numeric, linguistic, temporal, spatial, and conceptual patterns.

Since our brains essentially remember patterns, it makes sense that we would naturally chunk data into groups. Further, we either impose an arbitrary pattern on those chunks, or we relate them to a pattern with which we are already familiar. Any 10-digit number can be remember as a phone number, for example.

Patterns are also nestled. Language is probably the most obvious example. We have individual letters that represent sounds. These are grouped into patterns called words. Words are then grouped into sentences, which are grouped into paragraphs, which are grouped into chapters, poems, essays, or other specific forms. Even pictoral languages follow this overall pattern, with individual characters representing words instead of phonemes. Even then, individual characters are often clusters of several smaller characters.

We also chunk the natural world. We don’t just see a trunk, branches, and leaves. We see a tree – all the parts make a whole pattern that we conceptualize as one thing. Further, we may see a grove, or an entire forest. Similarly, we don’t just hear notes, we hear a melody, a refrain, a movement, a song, or even an entire opera or symphony.

Our brains organize information into nestled chunks of meaningful patterns. This is an adaptive and effective way to deal with massive amounts of information. It can also be constraining, however, since once we are familiar with a pattern we tend to impose it on information. This is why we also need to be flexible – able to learn knew patterns, unlearn patterns that are flawed or inaccurate, and make adjustments as new information is available. That can take a lot of effort, however.

We also tend to make arbitrary chunks – categories that are inherently subjective. The analogue world in which we live isn’t divided into neat categories, but our brains need to break up that continuum into definable chunks we can label and conceptually organize into patterns. That is why we agonize over whether Pluto is a planet, a dwarf planet, or something else. Pluto is what it is, but we feel the need to place it in its proper chunk.

Understanding chunking can be helpful. That is a common theme of this blog – understanding how the brain works so that we can more effectively use it as our primary tool for navigating the world, and avoid falling into the pitfalls derived from the limitations of our brains. Chunking, at the very least, is a useful strategy for memorizing information. But it also helps us understand how our brains deal with information, and avoid becoming a slave to this tool.

Patterns are useful, even necessary, but you need to question them and be willing to abandon or modify them.

16 responses so far

16 thoughts on “Chunking”

  1. TheTentacles says:

    The Gestalt School of Experimental Psychology coined the dictum that the “whole” is prior to the “parts”, that we FIRST perceive the world as a complete whole and that whole is not just some linear addition of lots of bits of information. This led to later theories that can be broadly conceived of as coarse-to-fine, that our brains are well optimised to process the gist quickly, and this gist then informs more detailed scruitiny. This is what you term “chunking”. Navon has a famous 1977 paper “Forest Before Trees: The Precedence of Global Features in Visual Perception” that precisely matches the analogy you gave.

    Our very early visual system has two major channels, magnocellular and parvocellular, which are fast+coarse and slow+fine. Work by a number of theorists of cortical function have built up a theory where these early neural distinctions are built up into several interacting recurrent loops of processing (Jean Bullier in 2001, Hochstein and Ahissar in 2002, Moshe Bar in 2004 and many others) so that the “Chunked” whole always informs the later fine discrimination.

    Our prior knowledge can also be conceived of as whole, and it is “prior” (comes before) sensory input. We know that, for example in early visual cortex only 0.2% of neuron inputs come from the eye, and most are intrinsic/recurrent.

    No matter how much you know a visual illusion is an illusion, you cannot undo the influnce of the global prior, of the chunked pattern as you call it. The cool things is the neural mechanisms that make the whole “prior” to the parts (at least for our visual perception), for chunking and patternising our world, are now being slowly uncovered.

  2. Pete A says:

    The 10-digit number in your example 8945557302:

    894-555-7302 is the worst way to remember it (for most people);

    8945 557 302 is also just three ‘chunks’, but it has the most-complex-to-recall chunk first, followed by two easier-to-recall equally-sized chunks.

    Telephone numbers within the UK are typically 11 digits, having a prefix of 0. Using your example number, it would be 08945557302. The recommended way of writing such a telephone it is:

    08945 557 302

    or for international access

    (+44) 8945 557 302

    The most difficult ways to dial it correctly from a business card / website / e-mail / sheet of paper — let alone remember it — include the following:

  3. Pete A says:

    I forgot to mention that the number in your example, 8945557302, is a 66-bit number:

    10 0001 0101 0011 0010 0101 1111 0011 0110 (big-endian format)

    Obviously, our memory is not capable of storing then recalling 66-bit numbers. 66-bit numbers have a set size of 2^66 ≈ 7.4E+19.

    Furthermore, adding a prefix of 0 to the number does not increase its number of bits, but it does increase the number of decimal digits that need to be stored and retrieved.

  4. Bill Openthalt says:

    Pete A —

    I think Steve is not using “bit” in the CompSci sense.

  5. BillyJoe7 says:

    I can never remember phone numbers – except my own 8 digit phone number, and only because the last 5 digits are identical (and, of course, the first 4 digits are the area code).

    As for my mobile number…well I know it starts with 04** *** ***

    (By the same token, I think 894-555-7302 is easier to remember than 8945-557-302, because the middle 3 digits are practically a given)

  6. Pete A – I disagree. You miss the fact that training and familiarity have a huge effect on chunking efficacy. Phone numbers are familiar, and so using a familiar patterns likely has more of an effect than an unfamiliar pattern even if there is a separate advantage as you say. (My example is specific to the US, of course.)

    I was not using “bit” in the computer context but the neurological context – one piece of information. That is why I specifically used “binary bit” when appropriate.

  7. Pete A says:

    Dr Novella — Yes, the best way to remember things such as telephone numbers is, I think, using the method with which we are already familiar / most familiar. E.g., very few people are able to recall their date of birth in the ISO 8601 unambiguous format YYYY-MM-DD.

    The ways in which people most easily remember chunks of information, and the error probability of their recall, is a core element of human factors and ergonomics.

    I’ve always been fascinated by the question: What constitutes one piece of information, from a neurological perspective? Take for example memorizing our times tables from 1 to 10: how many pieces of information is this?

    It seems that it is much less than 100 pieces or chunks of information, however, I’m completely unable to recall 9×7: I have to mentally swap the numbers to 7×9 then I’m able to instantly recall the answer 63. I’ve tried repeating “nine sevens are sixty-three” several times per day for a week while visualizing the numbers, but it doesn’t work. I find this disconcerting because I learnt all of the entries during my childhood. Some of my friends have the same problem with their times tables so it would be useful to know if there’s a scientific explanation for our embarrassing ineptitude.

  8. BillyJoe7 says:

    “the ISO 8601 unambiguous format YYYY-MM-DD”

    In Australia we use the format DD-MM-YYYY, which is both unambiguous and sensible, rather than the American convention of MM-DD-YYYY which is neither.

    Same as SURNAME-FIRSTNAME. The more recent government forms have changed this to the more sensible FIRSTNAME-SURNAME (though, incredibly, some still say FIRSTNAME-CHRISTIAN NAME/SURNAME! – I have a Muslim friend who bristles when asked for his Christian name: “are you for REAL?”, “I don’t have a CHRISTIAN name!”, “which CENTURY do you live in?”. I just smile and say “ask him what his SURNAME is”).

    Interestingly, the people who run most of Australia’s mountain running events, uses the FIRSTNAME-SURNAME format and their list are arranged alphabetically based on your FIRSTNAME and they print your first name on your race plate rather than your surname. Sounds much friendlier.

  9. Pete A says:


    The problem with the format DD-MM-YYYY is that it is indistinguishable from MM-DD-YYYY format for day numbers that are less than or equal to 12. E.g., if a web form requests the user to enter DD-MM-YYYY, it shouldn’t be a surprise if a US-based user enters instead MM-DD, and vice versa. It is impossible to detect these inadvertent and inevitable errors unless month names instead of numbers are enforced.

    As an aside, I’ve always though that having to state one’s date of birth is very amusing because we are are never asked in which timezone we we born, and whether or not daylight saving time was active at our time of birth. E.g. UK time is UTC time during the winter and UTC +1 hour (BST) during the summer, therefore everyone born during the first hour of a BST day was actually born on the previous day in UTC time. IOW, by referring to our date of birth relative to the timezone in which we were born, our date and time of birth varies as we travel across the globe 🙂

    Regarding Christian names: I’ve always thought that to be offensive, even when I was religious. One time I was asked, I replied “Jesus H. Christ. What’s yours?”

  10. GrahamH says:

    25th of December versus December (the) 25th. Fairly cultural I would expect.
    I prefer some that can be logical, such as increasing units. Days<weeks<years.
    However, much of the world is metric and the US is often non-metric.

    Back to chunking, very interesting. I'm also a dance teacher among other things and I often try to teach groupings of similar moves (e.g. right hand turns using several hand positions from leader) with varying success.

  11. BillyJoe7 says:


    In my country of birth I was born on the 17th, but here it would have been the 18th, in which case my birthday is actually on the same day as my mother-in-law!

    Same problem with the Moon landing. It was on the 20th July, which is also my daughter’s birthday. But here the Moon landing was on the 21st.

  12. JimV says:

    A related issue is the number of objects you can visualize mentally, which is the number of things you can look at and know the number of without counting – about eight.

    Our primate relatives seem to share this ability. I saw an old PBS show on animal behavior once in which a wild monkey was shown eight oranges (or some fruit) in a box, and then the top of the box was covered. With some magician’s trick one of the oranges was removed from the box without the monkey seeing it, and the box was opened again. The monkey made an open-jaw, stretched-neck look of surprise.

  13. Pete A says:


    Many thanks for your reply, they are exemplary practical examples of the difference between using local time and UTC for dates.

    (I shall click the “Submit Comment” button on 2017-08-06T00:00:10Z in order to compare it to the timestamp attached to this comment, because I don’t yet know the timezone that’s being used for the timestamps.)


    The subitizing ability of most humans is limited to four or five items, beyond which we have to enumerate, however, the limit is affected by both the arrangement of the items and the shape of the items. I seem to recall reading that some chimps can subitize up to 40 items, but I can’t find a reference so it’s best to assume that 40 is incorrect.

  14. BillyJoe7 says:


    “The subitizing ability of most humans is limited to four or five items”

    That sounds about right.

    With a larger group, I would tend to subgroup into groups of 3, 4, or 5.
    So I might not be able to see a group of 8, but I might be able to see a subgroup of 4 and another subgroup of 4 and then perform a simple quick addition. And I might not be able to see a group of 9, but I might be able to see 3 subgroups of 3 objects. For 10, it might be 2 subgroups of 5 objects.

    But, as you said, it could depend somewhat on the arrangement of the objects. For 8 objects it might actually be a subgroup of 3 objects and a subgroup of 5 objects, rather than 2 subgroups of 4 objects

  15. Pete A says:


    Yes, it would be easy to subitize 9 items such as 3 groups of 3 similar mobile phones. It would be very difficult or impossible with 9 dissimilar items having a random arrangement. Subitizing is related to attentional blink.

    QUOTE from Wikipedia article Attentional blink:
    The attentional blink can be moderated by changes in visual similarity between targets and distractor stimuli, but it can also be affected by conceptual similarities, suggesting that stimuli are processed to quite a deep level preconsciously, with much of the resulting information discarded before it reaches consciousness.
    END of QUOTE

    A good explanation of subitizing is given by Tom Stafford and Matt Webb in their book Mind Hacks: Tips & Tools for Using Your Brain, Chapter 3 Attention: Hack #35 Count Faster with Subitizing, subsection How it Works.

  16. Newcoaster says:

    Memory is fascinating.

    I took a psychology course in University back in the day, where one of our assignments was to use the “Method of Loci” to remember a long string of words. If I recall, most of us were easily able to remember 25-30 random words or more, after some practice with the method. The key was to have a very distinct place for each number, and to memorize that first. I walked around the campus in a typical pattern and came up with 25 distinct places I could remember and assign a number (bottom of the elevator, coffee shop, video arcade, bookstore..etc) Then one just imagined whatever the word was associated with that location and created a mental image, the more bizarre the better.

    A closely related topic is the “tip of the tongue” phenomena. When you really just can’t remember that word that you know you know. A way to induce this is to read a random dictionary definition to a group and see what kind of words people come up with for the word being defined. Interestingly, a significant number of the wrong guesses either started with the same first letter, or rhyme with the word that was being defined.

Leave a Reply