How to learn a foreign language

Naturally, the strategy one should employ depends on one's specific goal and motivation. As motivation adds a whole host of complexity to the matter, we will largely ignore it. I will begin by assuming our goal is impersonating a native.

Let's take a look at the history of second language acquisition as a field. It has more or less happened in discrete steps which build off of eachother.

Notice that of these five steps, we are closer and closer approximating the method employed by babies in L1 acquisition. They do not learn with conscious study, but rather learn to extract signal from chaos through mass input.

Limitations of Conscious Understanding

Consider some foreign concept which you aim to learn. I'll pose the following fact as an example: If you use CFOP with edge control or winter variation, you are way more likely to get an LL skip. Chances are you haven't the first idea what it means. I could define all the terms to you, but even if I tell you that CFOP is a strategy for solving a rubiks cube, edge control is used for orienting edges of the top layer during the solution of the second layer, and LL stands for last layer, you still would not be able to explain, or even imagine, what I am really trying to say. Learning languages is like this. You can know have some understanding of all the words present, but without knowing the statistical space they exist within, one cannot perceive them as they are. Only by knowing the full space can one perceive its elements!

Language learning is thus a convergent and imprecise thing. It is often the case for learners to want to learn something down, that is, memorize it wholly and move on to the next thing. By the above logic, this is not actually achievable. You cannot memorize, say, all of the numbers in a language in a meaningful fashion - disembodied and without context, they can only ever exist as the inactionable semantic facts provided by rote memory. It is thus more important to converge diffusely on a holistic understanding of all of the language's features in an unguided fashion, which can be immensely frustrating as progress is virtually impossible to measure.

The same can be said of the nested features of language- language is an integrated multiresolutional stack of subskills. To learn a language, you must acquire its phonology (sounds,) morphology (system of word construction,) and syntax (word ordering.) All along with a large chunk of vocabulary. These equivalently must be learned in parallel.

My Beef With Refold

I actually don't blame Refold at all for not implementing the methodology I am about to describe. It is quite harsh, compared to basically everything else available. Not exactly the basis of a platform for making language acquisition accessible to all. That said, my suspicion is not that they consciously omitted the following details, but are rather unaware of the essence of the acquisition of statistical patterns from noise.

I think refold philosophy is lacking in the earlier stages (substituted with vocabulary memorization and conscious study). The notion of comprehension vs production is present, but I think you can point out a few stages that should take place prior. For example, one acquires the ability to differentiate where one word ends and the next begins, determines the space of realizations of particular phonemes, and things of the sort. I call this "segmentation." With that ability in place it's easy to pick out words but still hard to understand meaning. Picking out words and getting used to them I call "recognition", Matt from Refold calls it mental dictionary entries and uses Anki as a substitute. Beyond that are comprehension and then production, whose differentiation Refold is acutely aware of.

Segmentation, Recognition, Comprehension, Production

I think mangling these first two stages could explain why learners' accents (e.g. pitch accent) usually have more mistakes than other parts of production.

I should be clear that I am not implying that there are different neurological systems for each of these stages. They are just a breakdown of what healthy acquisition tends to looks like. By becoming aware of this process as it occurs naturally, we can use it as a guide, refine our expectations, better measure progress, and not be misled by any linguistic snake-oil salesmen.

In Practical Terms

I think the largest of all sins of Refold's guide is the expectation for learners to jump right into attempting comprehension, or to aid comprehension via Anki. This is immensely frustrating - trying to pick words out of completely slurred gibberish - it's like me asking you to do word-search on an enormous grid of Chinese characters. The only option is to painstakingly go one-by-one and rely directly on sensorial input with no mediating interface. Although Matt has talked about it at points, I haven't gotten a sense from their new leadership that they really understand the gap here.

Begin with an enormous amount of passive immersion. Months of it, for as many hours in the day as possible. This should get you through the segmentation phase. Continue with that schedule, but begin to investigate patterns which call out to you as familiar for a year. Pay attention to them and acknowledge the words that take on an aroma of familiarity. Only then should you begin with Anki to accomplish comprehension (the topic of the next section.) With several years of comprehension training, the productive capability will arise on its own, without much intensive practice.

The Anki Curve

One can think of the acquisition of some particular linguistic feature as located somewhere in this 4-phase timeline. Notice that in intermediate stages of language acquisition, the learner is capable of grammatically producing particular features, whereas sometimes there may be hiccups in recognition or even segmentation. That is to say, this 4-step categorization can be applied not only globally to the language as a whole but also locally to particular features.

Refold suggests using Anki to learn the first bunch of words and springboard from there. I suggest not: the highest level of bang-for-buck with conscious study is at the comprehension stage. You can not learn segmentation through a closed set of memorized vocabulary. Thus, I suggest not even touching Anki to jumpstart knowledge in a new language, or to learn words that don't feel familiar already. Using it for exposure to completely novel features is tedious and barely works. There is similarly no point in studying what has already breached conscious understanding of meaning. That leaves only the in-between area (compare to i+1) to be learned with anki: that which feels familiar but is not consciously understood. Grammatical parsing is a separate system from the conscious knowledge of the interspersed lexemes, and the latter is what Anki is for.

Some other rapid-fire anki tips:

A note on speak-early

It is worth noting that alongside the input hypothesis has developed another school of thought, which can be called "speak early, speak a lot." It motivates most of language-learning pop culture and pop-methodology, from apps to tutoring. The idea is fundamentally that one learns to speak a foreign language through practicing speaking, which seems like a sensible idea at first glance.

A friend and student of psychology, after hearing my opinions on language acquisition, claimed: "To speak a foreign language, the requisite action is practice speaking the language. One learns to play basketball not by watching others play basketball, but rather by playing the game." This is somewhat misguided. Practice is paramount, but allow me to modify the basketball analogy. To practice speaking a language that you cannot understand is to practice throwing basketballs in random directions, blindfolded. You have no means by which to identify when you are wrong. If you think the natives you talk to are interested or even able to correct your mistakes, especially in pronunciation, then you are sorely wrong. To learn to play basketball, one must first learn to see. That turns out to be by far the more difficult task.

Italki, duolingo, and plenty of other services are rooted in this idea of practicing usage from day 1. Although I think that it's not the right approach long-term, it has its place. I have personally had a lot of success with Italki for Indonesian. It took me about 20 one-hour lessons to be able to fumble my way through basic conversations. Here is a guide on how to do this effectively, although I don't espouse using his methodology long-term. That can be a really awesome springboard for starting an immersion approach. It builds motivation and proves to the learner in comparably little time that their goals of chatting with natives are not so unthinkable. Again though, and I can't stress this enough, don't stick with it long term.

How to deviate from this guide

At the beginning of this post I mentioned two disclaimers: goals and motivation. Up until now I have assumed that your goal is effectively becoming a native speaker and your motivation is an unending source. It's most likely that neither of these assumptions are true, although they are valuable for understanding the psychology of language acquisition.

First, to address motivation, it is crucial to make jumps in the methodology when necessary in order to stay motivated. Segmentation is notoriously boring. Attempting to produce (in front of a native) can be fun! Balance a healthy ordering of acquisition with what you find personally meaningful.

Let's break language learners up into 4 types by the proficiency they wish to achieve.

  1. Wants to know some basic phrases and vocabulary just to visit a foreign country and make it around.
  2. Wants to reach a "conversational" proficiency, enough to be able to have some sort of meaningful conversation with native speakers.
  3. Wants to reach a "fluent" proficiency, enough to navigate whatever social scenario they might find themself in.
  4. Wants to impersonate a native.
Goal n roughly corresponds to that which is achievable by skipping all but the last n steps of SRCP. One can learn to parrot memorized phrases, accomplishing goal 1, by only practicing production. Goal 2 is achieved by learning some words and their meanings in various contexts, but without drudging them up from the mess of unpatterned signal by means of raw exposure. Goal 3 can be accomplished by using immersion beginning with reading instead of listening, which is a modality of expression that is inherently self-segmenting by means of punctuation. The fourth can only be achieved through true acquisition.