(Part 1 can be found here; part 3, here; part 4, here.)
I’m not entirely sure what the point of the first post was. We’ve barely touched any flashcards, and you spent most of your time on Ancient Greek verbs.
I spend a lot of time on them anyways. The point is that a cell is not a chunk, and the human mind stores knowledge in chunks, not cells.
That’s a distinction without a difference.
No. Chunks are recursive, cells aren’t. It makes all the difference.
You’ve invented a strawman of “the mind as spreadsheet cells” that nobody believes in.
Like I said: if people did believe it consciously, what would be different?
Why are you trying to channel—
You’re reading this, which means it’s for you.
IV. Chunk Processing
Here is the first major kernel of insight:
1. Memorization consists of breaking down some mass of knowledge into chunks small enough to fit in your head.
This seems entirely obvious. Of course when you’re taking 10th-grade American history or first-year Ancient Greek or organic chemistry you’re going to have to break down the curriculum into small pieces that you can actually learn. You can’t absorb an entire textbook in a single night, and even if you could, you’d be going through it paragraph by paragraph instead of uploading the whole thing to your brain at once.
But—a chunk is not a cell! And chunks are recursive. Which leads us to the next two principles:
2. If a chunk is not going in, it is probably too large and needs to be broken down.
3. The brain processes chunks one at a time.
Where does this leave us?
Consider the unlucky student of German, Russian, Latin or Greek. All four of these languages are rather notorious for creating new verbs from basic ones by adding a preposition as a prefix:
Latin dormiō ‘I’m sleeping’ → obdormiō ‘I fall asleep’
Greek πίπτω ‘I fall’ → προσπίπτω ‘I attack’
German setzen ‘to set, put’ → übersetzen ‘to translate’
Russian давать ‘to give’ → сдавать ‘to take (an exam)’
(If this seems too esoteric, ask yourself what understanding has to do with standing.)
In the most common 6200-odd words in ancient Greek1, there’s the basic verb πίπτω—and then 18 additional verbs formed by adding some preposition as a prefix (sometimes two, as in the case of συν-εισ-πίπτω ‘I rush in along/together with’). My trusty copy of Jones and Tschirner’s frequency dictionary of German (the top 5000 words) lists setzen along with 15 prefixed derivative verbs. This doesn’t even begin to count further derivations like Zusammensetzung ‘composition’ or πτῶσις ‘a fall, falling’.
Geez, that’s a long one. Uh, ‘out-in-other-placing.’ Does it mean “outsourcing?” The Goethe Institut keeps making us read things about the German economy.
And that’s the 1,390th most common word in the language!
A fluent speaker of German, whether native or foreign-born, has Auseinandersetzung in their head as a single chunk. They aren’t even thinking about what it has to do with setzen in conversation.
But you aren’t there yet. To get Auseinandersetzung in your head as a single chunk, you must first learn its sub-chunks—until they fuse together.
Ah, ja, gestern hatte ich eine…ummm…Ausein…über?…setzung mit meiner Freundin.
Eine Auseinandersetzung? Ah, ja, genau, danke.
Nota bene: the arrows are as important as the chunks themselves. We’ll get to why.
V. Anki Follows the Brain
Reviewing material, whether that’s with Anki or with something more old-fashioned like a homework sheet of practice problems, can be thought of as a process that feeds the user prompts to try and fetch a chunk of information. The ideal prompt gives the user exactly as much information as is needed to fetch the chunk with certainty—and no more. Suppose the pertinent info we need our students to learn is that Eisenhower was president at the time of Brown v. Board.
Good question: Who was president at the time of Brown v. Board of Education?
Bad question: Who was president at the time of an important civil rights decision?
Bad question: Which former WWII general was president at the time of Brown v. Board of Education?
Question 2 doesn’t give us enough information to pin down the answer with certainty. Question 3 gives us too much information—there’s only one president who was a general during WWII and we might very well have fetched the answer that way.
This seems more obvious than it is. The gold standard for review is active recall. Multiple choice tests are common because they’re easy to grade, but they’re fundamentally exercises in recognition2.
Good Anki cards follow a pretty basic principle: each prompt should fetch a single chunk precisely, with as minimal a prompt as is necessary to call it up. This is why cloze cards (essentially, fill-in-the-blank; Anki uses a bright blue […] instead of a blank) work so well.
This card is jogging the same connections as this “basic” card with no cloze:
We can also go the other way around, with a question where the choice of cloze format is somewhat more natural:
Not all of my cards are clozes—most of them are foreign-language vocab for reading, so most of them aren’t—but almost all of my cards that aren’t vocab are clozes, because well-designed cloze deletions are very, very good at inducing the prompt-to-chunk path that characterizes active recall. Make a cloze out of a chunk. If it fails to install, cut up the chunk and cloze it.
To understand why this works so well, it’s worth considering the nature of forgetting for a second.
VI. A Typology of Forgetting
There are at least four (and a half) types of forgetting.3
Full-on hard-drive wipe, as if you’d never seen this before in your life and have to relearn it from scratch. This type of forgetting is pretty rare outside of brain injuries or dementia patients, but it does happen to all of us from time to time (although, in my experience, it’s usually some other type of forgetting in disguise). It’s also not a very interesting kind of forgetting, in that the only “solution” to it is to relearn the material from zero, so we won’t consider it further.
(Type 1.5: the information didn’t install in the first place, and you’re operating under the illusion that it was supposed to. But I studied all the steps of the Krebs cycle! Nope—you studied the first few steps and then let your eyes glaze over the remainder, which never landed).
Production failure. This is probably the most common one, and the bane of language learners everywhere.
Production failures occur because connections and junctures between chunks are directional. It is easier to go from a foreign language to a native language than the reverse. The chunk Auseinandersetzung forms two connections to the chunk argument—one German to English (strong) and one English to German (weak).4 Weaker connections are slower to traverse and may well become dormant, beyond the immediate reach of active recall.
Subchunk failure. This is related to forgetting type 2 and, in fact, may be the same thing, because chunks are recursive and “Auseinandersetzung means argument” is itself a chunk.
In subchunk failure, much of the chunk is retreived, but some is not. The chunk indexing a word in many languages, for example, will include its gender. Suppose we’ve forgotten the gender of Milch (feminine):
We can still be understood because we have other chunks containing heuristics and rules that we fall back on to generate factoids by analogy, which are correct unless they aren’t.
The gender and plural of Zeitung are automatic (dashed boxes), like the formation of amāvī from amō, amāre. The gender and plural of Schwung are proper subchunks of the chunk containing Schwung.5
This doesn’t just apply to languages, either:
(Followers with linguistics degrees may wonder if I think I’ve found some sort of model of language in the brain. Maybe; that will be an appendix to the series.)
Wire-crossing.
This is fundamentally a kind of subchunk failure, but it’s a very common and salient subtype—the one that originally triggered the “aha” moment for me. I will cite two personal examples, one from Gothic and the other from Coptic.
a. One of the two words for ‘to steal’ in Gothic is hlifan (the other is stilan, which is more transparent)6. Due to a sound change in the history of the Germanic languages, there are many words in Gothic that alternate between an f in some forms and a b in other forms.
So let’s fire up the Gothic deck.
Ah, geez, was this…was this hlifan or hliban? Guh…
This occurs because there’s wire-crossing leading to subchunk failure. Directionality applies here: if I am prompted with hlifan I have no trouble fetching the English meaning!
You may already have figured out how to tackle this; I’ll explore it in full in part 3.
b. Another example, this time from Coptic. The Coptic word for (the) famine is (ⲡⲉ)ϩⲕⲟ, where ⲡⲉ- (pronouced pe-) is the masculine definite article, and ϩⲕⲟ (the backwards S thing is pronounced like a strong h) is the root for “famine” or “hunger”. Easy-peasy:
There’s just one issue: the word for ‘horse’ is ⲡⲉϩⲧⲟ.
If there’s only one word in our Coptic vocabulary of the shape ⲡⲉϩCⲟ (where C = a consonant), then we can usually commit it to memory pretty easily, and produce it just fine. If we have two, then we invite wire-crossing.
Importantly, wire-crossing is not type 1 forgetting, where you introduce a new word of a similar shape and the difference just falls out of your head completely. Both words have been learned along with the middle consonant, but they’re similar-looking enough that the wires have crossed. Banging your head against the vocab card can work eventually, in the same way that writing the formula for acetate can work eventually, but it’s painful and inefficient, because the problem is not with the word-chunk as a whole but with a particular subchunk within it.
So: what is to be done? We will take a look at my answer to this question in Part 3.
Chosen as a (somewhat arbitrary) cutoff for the frequency list.
Math tests excepted, so long as you’re careful enough to design plausible distractors. There are people who have used Anki to good effect for math, but I’ve always found it troublesome. I would be curious to hear from people who have had experience with it in the comments.
I say “at least four (and a half)” because there are probably additional types that could be useful models (remember, a typology is a map, and the map is not the territory, especially in metaphors for the workings of the mind). My experience has been that most cases of ‘I forgot this information’ fall into one of these four categories. There’s certainly also “I can’t remember where I put the kids’ quizzes/my wallet/my keys”. The absentminded polymath is a well-justified trope, which surely points to “forgetting type #5” being fundamentally different from the first four types, (presumably because factual information and life experiences are processed separately).
I am well aware, of course, that fluency in a language requires the ability to operate in that language without translating everything from the L1. If you’d like, however, replace “argument” with “the primordial Platonic concept usually represented by argument in English and Auseinandersetzung in German.” In any case, even quite fluent speakers of a foreign language usually do start trying to index from their L1 when they trip up and forget a word they’re looking for—if you aren’t tripping up, the word’s not lost!
If you’re sick of all these foreign languages, consider English plurals for a second. Words like napkins or televisions are automatic (dashed box), because they're generated automatically from the headwords napkin and television plus a rule that plurals in English are generated by suffixing an -s. That rule is a default which is overridden in the case of words like geese and mice.
Hlifan is cognate to Latin clepō and Greek κλέπτω, for the Indo-Europeanists keeping track at home.