Re: George Miller [Magical Number 7 +/-2] Part 6

From: Stevan Harnad (
Date: Sun Feb 22 1998 - 19:11:20 GMT

On Sun, 22 Feb 1998, Elizabeth Hocking wrote:

> In the first paragraph of this section, Miller stated: "Since the
> memory span is a fixed number of chunks, we can increase the number
> of bits of information that it contains by building larger and larger
> chunks, each chunk containing more information than before."
> This is the essence of the whole of this section of his article. An
> example could be taking snowdrops, daffodils, crocuses, dandelions,
> daisies, tulips, roses and bluebells and renaming them as flowers.
> These join trees, bushes and grass as vegetation, and so on.


> The example that Miller gives is morse code. He says that after the
> man learning this code organises sounds into letters, "the letters
> organise themselves into words". This is not a particularly hard idea
> to grasp, except that the way Miller says this is rather amusing, as
> it suggests that, put in those words, it is the letters doing the
> work rather than the man who is learning!

Good point. Think of it this way. When you don't know Morse code and
you look at a message line this "._ __. __._ .. ._. _ __ __." You have
to remember a string of 17 .'s and _'s. But if you knew morse code (and
I've forgotten it, the message might be V I G Y A Z N I, so you would only
have to remember 8 letters. Or, if you know the "code" which is the
Hungarian language, you would only have to remember one word: "Careful!"

This is how rechunking works. It may LOOK as if the letters are doing the
work, but of course it's your brain, which has so thoroughly learned what
the letters are that it immediately sees "._" as "V" [just as we immediately
see "V" as the letter "VEE"].

So the letters do the work only after your brain has learned the code so
well that it has become automatic.

> Miller confirms that the jargon for this renaming is "recoding", and
> although there are many methods for this, the easiest is that which I
> illustrated with the names of flowers.
> Miller recounts experiments undertaken by Sidney Smith in the 1950s.
> One task that Smith introduced was recoding 18 digit binary numbers
> (many more than people can recall, which is normally 7 +/- 2) into
> chunks with decimal names. However, after participants said that they
> had learned this recoding system, their recollection of the digits
> were better than before learning, but not as good as Smith expected.
> The conclusion drawn was that the time spent learning (5-10 minutes)
> was not long enough. The switch between codes must be nearly
> automatic. Therefore, Smith drilled himself on the methods and
> enabled himself to recall a 40 digit binary number. In reading this,
> I noted that many things that we do involve automation. For example,
> remembering phone numbers, driving, typing. We learn to automate so
> that remembering things is not so time-consuming.

Yes, and in fact there is a whole line of research on the process by which
we move from conscious, controlled processes (like first trying to ride a
bike) to the automatic ones they become once they are "overlearned" -- which
means once they are learned so well that you can't forget them any more
(unless you get brain damage!): Try "forgetting" that the 3 straight lines
of the pattern "Z" are "ZED."

The stroop colour-word test is a good one to check for overlearning,
recoding and chunking. The classical way it is done is to write the names of
colours, say, "RED" in the wrong colour, say, "RED" written in green. There
is interference: If you are told to say the colour the word is written in,
what the word SAYs interferes and causes errors, and vice versa. That's
because both the code for naming words and for naming colours is so
overlearned that it is automatic. (There is no stroop interference for
colour words written in Hungarian: "VOROS" "ZOLD" "SARGA": You'd have no
trouble saying this or saying what colour they were in, because you don't
know the code. I, on the other hand, would have trouble, because I speak

Here's a good experiment for a practical: Can you CREATE a stroop effect
with new coding? You can get a somewhat weaker stroop effect from words and
shapes. For example, pick three simple shapes (not standard ones like
circles or squares, but new ones, say something that's round on the top and
v-shaped on the bottom and so on). Teach subjects to give them nonce names,
say "dreebles" "grundles" etc. You'll need to use more than 7+/- 2 shapes to
make it hard, but perhaps even 3 would work.

The training goes like this: They have to learn to say each shape's
name as fast as they can when they see it. What they see every time is
a bunch of little dreebles used to spell another plural nonce word (use
other nonsense words that are pronounceable like "quotrips").
But occasionally, make it into a stroop, spelling the word "dreebles"
using "grundles" and vice versa.

You should find that they make more mistakes with the stroops as they get
faster at identifying the dreebles, etc. You should also do some reverse
stroops, asking them to say the word spelled out rather than what little
shapes it's made of. This too should get worse for the stroops as your
subjects get faster at naming the dreebles, etc. What has happened is that
in coding those shapes as dreebles, your brain has made an automatic
dreeble-detector. But the more automatic that detector is, there more
trouble you'll have when the word is a stroop.

That's the negative side of building new "chunks." But mostly it's an
advantage. Not that many things in life are stroop-like. (Can you think of
any others: what you want is examples of things that have two different
names, and one can interfere with the other. Word/name interference seems to
be the strongest, but it could be name/name interference for the same

> Miller goes on to say that "The point is that recoding is an
> extremely powerful weapon for increasing the amount of information
> that we can deal with. In one form or another we use recoding
> constantly in our daily behaviour." This reminded me of the accounts
> that Stevan gave of a person (I can't remember his name or codename)
> that was unable to recode because he had a seemingly limitless memory
> and saw the same dog in different positions as different and
> therefore needing a new name.

Exactly! And that was what I was hoping you'd say! I'll remind you about
Funes the memorious during the seminar...

> "It seems probable that even memorisation can be studied in these
> terms". I have often been told to make memory links between things
> that I need to remember, which works, as long as you don't forget the
> links!

There are lots of ways to "recode" so that a name can pick out a bigger and
bigger chunk. But of course you need to remember the connection with the
name. If you used a memory trick like thinking of a picture or story that
links the chunk with the name, and you forget the picture, you lose the
chunk. That's why overlearning is important (as in the way you know that "V"
is VEE), because overlearnt things you will never forget; weak associative
links you might. If you use them over and over, though, they become
overlearnt too.

> A television show of a few years ago demonstrated this by
> pitting members of an audience who had learnt information through
> memory links using movements of the arms, against experts in the
> field of the information they had learned. Very often, those who had
> learnt the information all at once received the most points.

Kid-sib couldn't follow this...

> In reading this article, I thought of an instance where I learnt
> something, and still remember it 8 years on. When I was 11, a few
> friends and I learnt the order of the books of the New Testament in
> the Bible to help with RS lessons. I cannot remember learning them,
> but since soon after I can rattle off these 27 names automatically,
> with no thought whatsoever. Do I remember the order and link between
> each, or do I recode them into 5 groups of five and the last 2, or
> does the motor movement of my mouth cause them all to be remembered?
> I have more difficulty writing them down, and this is only helped by
> saying them aloud. This suggests that it is the movement of my mouth
> that I have recoded. Does this make sense?

Yes, it looks as if you linked them to saying them rather than just thinking
them. This jingles help. Another one is "Thirty days hath september, april
june and november." Using that (and remembering that February is different)
you can always figure out whether any month has 30 days of 31, but it is not
automatic and overlearned because if I suddenly say "june" or "november" you
can't say how many days it has immediately, only by going through the
jingle. For overlearning the link has to be direct. You need a "30-day-month
detector" that sees it at once. Sometimes you get this if you do it often
enough. You get it even better if your mind can find out some feature or
some simple rule by which you can tell at once which is a 30-day and which a
31-day month:

30 Apr Jun Sep Nov

31 Jan Mar May Jul Aug Oct Dec

Is there any feature that all the 30's have and none of 31's do, or vice
versa, that you could detect directly, without the jingle? I'm not sure
myself, and that's probably why we do it by jingle: There's no easier way
to recode them than by rote memory (using a jingle). Now if all and only
the 30's' names began with consonants, or something like that...

> Obviously recoding is important in language, but Miller says that the
> recoding of images is important in our everyday lives, but this is
> much more difficult to study empirically.

Probably most of image recoding is language based too: That's how I get to see
"dreebles" as recognisable chunks, along with circles, triangles, etc.

After all, it is images that were recoded when a "V" became a VEE.

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:24:19 GMT