Chunking is a very important subject, yet it is not sufficiently discussed. It appears to be very simple, misleadingly simple. When used in memory championships, chunking is preset and given as a template. But when we read, the nature of chunking changes. When we multitask it changes again.
The basic chunking
The basic chunking is very simple. Remembering a lot of independent items is hard. Our mind is not built for it. We can distinguish visually up to 8 dots, and then we either count or group the items. There are many experiments with infants and the numbers vary based on experimental settings. One thing is sure. Once we start chunking and counting, we can handle very large numbers with ease.
So if we need to perceive disks, we usually group them into 5×5 squares and then organize these squares in larger 3×3 structures and so on. The actual number is less important. It is usually 3,4, or 5, but can be something else.
When a memory master needs to memorize a set of numbers, his tasks is easy. He trained with PAO of a certain length, and this is the exact amount of digits he takes in each chunk.
Chunking with random words
When we chunk random words there is an extra element of shared motive. We try to find groups of 3-5 words that share some attributes and use those attributes as chunking criteria. This behavior is known from games for small children, where cards are organized by color and by other characteristics, for example, faces vs numbers.
So we look at say 20 words, and try to see the 4 words that share some quality. We encode them as a chunk and move until very few words remain. They may share nothing, but we need to chunk them anyway.
If you do not like chunking by random attributes, you can chunk by geometric location in the text. With random words, it makes no difference, but we should almost never deal with random words.
Chunking while reading is hard, so let us divide it into several steps. In the first step, simply highlights the words that you need to remember. It does not really matter how many words, but try to go for something above 30 and below 100. Maybe you will need more than one article for that, yet select articles about the same subject.
Now, notice that some words appear in the context of the main subject and they are contextually expected. Other words you would never expect to find in the text. They encode innovation and provoke curiosity. And there are also many more words in-between that you can easily ignore. Life is not about random words, it is about finding meaning.
There is a simple and a more complex way to group the highlighted words. The simple way implies grouping the words in groups of 5 until the end of section and then restarting the count.
The more complex way is keeping a matrix of words to remember. Usually, a 3×3 matrix will do. Then we remember not individual words, but groups of 3 words as one semantic unit. For example “big gray wolf” is 3 words long, but one semantic unit, part of larger visualization. Clearly, the words we need to remember do not go together that well. If they are complex, one word will take a whole semantic unit. If they are simple, three words can work.
A still more complex way involves 5×5 matrix. The first row is built from 5 individual words. Then a 3×3 matrix. And finally a row with one word only: the word that is damn hard to put anywhere. While this way appears to be great, it is pretty hard to use – at least for me, as I get tired very fast using it.
So now, take the highlighted words and chunk them together.
Transforming the words
A PAO usually encodes 5 words. If you use a chunk of 5 words, try to build a PAO. You can substitute any word by a word similar in sound or meaning. You can change the role of the word to be a subject, object, verb, adjunctive, or advers. Adjectives are usually very easy to visualize. Subjects are usually clearly visible from the selection of the words. Objects can encode pretty much anything. Verbs that we can visualize are very limited. Adverbs can become messy, and possibly should be avoided.
An example “aspect ratio of multidimensional hyper rectangles”. I caught it just now from reading excerpts. Did not plan to encode. What do I do? There is no verb, no subject. I cannot visualize something multidimensional. So I start with a basic idea that to show something multidimensional we need projections. I take a complex boxy shape and put it in a path of a laser beam, generating 9 rectangles of different aspect ratio. Now I need to visualize that from all those hyperrectangles I actually need an aspect ratio. “Aspect” means to look at in Latin – which is a projection. “Ratio” means reckoning… So I visualize a judge with a projector looking at a screen with rectangles. If I can do a PAO with something like that, I am sure you can do it with pretty much anything.
Moving the words around
Suppose you get your 5 words and cannot form a PAO. Now what? You can lend the bad word to the next chunk, and take the best-looking word from the next chunk. So while usually, we talk about a chunk of 5, we actually work with two chunks of 5 each time.
Working with 3×3 matrix can actually be easier, as we can choose on the fly at which row each next word is most likely to be used. In some way the 5×5 matrix is even better since the 3×3 part can be built nearly perfectly, the 5 words part can be passable, and the annoying extra words will get the last row and be memorized separately.
On the other hand, the larger matrices offer additional complexity to processing and require more working memory – which limits the complexity of the texts we can process. For me, it feels like giving up 20 IQ points to get 30% faster memorization. A hard choice to make.
From words to PAO
A PAO made from 5 words is pretty standard. 1 word for action, 2 words for a person, 3 words for an object with one spare to use as we want – often dual coding or providing a context.
A 3×3 matrix is a bit more complex. We can make two PAO objects, or accessorize our current PAO.
A 5×5 matrix usually hides at least one word that is very hard to visualize and it may require an entire dedicated PAO just for itself.
Putting PAO in mental palaces and mindmaps
In any case, a room in a mental palace has 4 corners which is best fit to 4 PAO objects. A 20 words selection can stand in the corners of one room. A room in a mental palace also has walls, which are great for mindmaps. Just notice that mindmaps are very good for highly hierarchical information. We can introduce 4 branches per node – just like room corners – and at the end 4 leaves, e.g. 4 PAOs.
So while the first layer of chunking follows multiples of 5, the next layer is best suited for multiples of 4. You can use other numbers, but then you will need to make special adaptations in your methdology.
For example, 6 is a fine number for a PAO, and you can put 2 PAOs along the wall where there are no doors, just below the mindmaps. In a mindmap you can use radix 6, usually as two clusters of branches per node. Harder to remember, harder to learn, possibly slower, but definitely doable.
The hard part
Now, this was not even the hard part. The hard part is doing this procedure each section. As we read we collect all words from the section into a chunk of 5 words, or 3×3 matrix, or 5×5 matrix. After reading the section we have the chunk. While performing saccadic eye motion, visualize the chunk and put it into the mental palace – as PAO or mindmap as you wish. And continue. As fast as you can. Without even thinking about the process. This is the challenge. It is hard initially, and usually, it is attempted after mastering reading without chunking, Yet only when you master it, your memorization speed catches up with your reading speed – at high speeds.