# Learning and recalling chunked high order sequences

So, I’ve been thinking a little about chunking of sequences in the brain. And I guess my main thought and question is why does the brain do this?

It is easy to find examples where it feels like my brain is chunking, so I presume it is a real thing. Say you are recalling a password, I seem to break it into subsequences. You start at the first subsequence, then the end of that subsequence prompts the next subsequence, and so on. But you only know your password in sequence. It is impossible to recall the sequence in reverse (without some mental gymnastics), or even predict the element a few steps down from where you are. You only know what immediately comes next.

I guess a couple of easy examples are the alphabet and the digits of pi. The brackets show how my brain chunks them, though other people may have different chunk sizes:

(ABC)(DEF)(GHI)(JKL)…
(3.14)(15)(92)(65)(35)(89)…

So the question is, why does the brain chunk? Why not just store a full sequence such as:
alphabet -> A -> B -> C -> D -> E -> F -> G -> …

alphabet -> alpha1 -> A -> B -> C -> alpha2 -> D -> E -> F -> alpha3 -> G -> H -> …
ie, a sequence of sequences.

The higher order sequence:
alphabet: alpha1 -> alpha2 -> alpha3 -> alpha4 -> …
and the lower order sequences:
alpha1: A -> B -> C
alpha2: D -> E -> F
alpha3: G -> H -> I
alpha4: J -> K -> L

Then to complicate the picture a little, it seems when you spell out the letters of a word, it doesn’t use chunking. Again, if I mentally spell out even long words my brain doesn’t feel like it is chunking. So why for the alphabet, digits of pi, or secure passwords does the brain chunk, but not for spelling words?

BTW, I have working proof-of-concept code for learning and recalling chunked sequences in my notation. So the sequence of sequence idea works. And of course, HTM high order sequence learning is a key part. And it should be easy enough to extend this to the more general idea of a sequence of a sequence of a sequence of a sequence of … but I can’t currently think of a good example to test that idea with.

This is a total guess, but I imagine that the the brain “chunks” whenever it can do so and still be able to accomplish all of the tasks being demanded of it. Abstracting things allows us to categorize and associate similar things (one of the hallmarks of human intelligence), so it seems to me that the brain would be trying to do this whenever possible. However, to accomplish some tasks, we require lower-level elements of an abstraction.

In reading, for example, some people (and typically young children who are first learning to read as well) will do “sight reading”, where they recognize whole words by how they look. If my 6-year old son encounters the word “attach” in a book, he will read it as “attack” (because of all the video games he plays…) and need to be corrected. In school, we are taught phonics, and tend to have this skill drilled into us. We also tend to have spelling tests in school, in which precise spelling is required to receive good grades. We also are taught to draw letters properly, and this is again drilled into us. To perform these task repeatedly and receive positive reinforcement, our brains must break down its “chunk size” into individual letters. When it comes to spelling words, there may be room for some “chunking” of letter groups (for example “QU” vs “Q”,“U”), but not in general, due to extensive training on tasks that requires us to deal with abstractions at the level of individual letters.

Number chunking is much easier to get away with. When drilling math concepts in school, we spend a great deal of time working with 2, and 3 digit numbers. Therefore, we build abstractions for these multi-digit numbers to help simplify those tasks. These abstractions can then be used as features of a sequence. In the example of digits of PI, it makes it easier to remember a lot more digits of PI if we think of them as a sequence of multi-digit numbers, than as a sequence of single-digit numbers (because the sequence being memorized is much shorter in length).

1 Like

BTW, my own “chunking” for the alphabet is like so:

(AB) (CD) (EF) (G) (HI) (JK) (LMNO) §… damn that infernal song!