Senior software engineer at Qualia Labs · Co-founder of Fox.Build Makerspace · Former co-founder of FarmBot

Two years of spaced repetition vocab experiments

Vocabulary Learning Experiments for TOPIK (and other standardized tests)

If you are studying for the Test of Proficiency in Korean (TOPIK), or any other standardized language assessment (TOEIC, JLPT, HSK, DELF, etc.), you will eventually face the unfun reality of memorizing several thousand vocabulary words. You can reach fluency through immersion, conversation, and reading, but if your goal is to pass a standardized exam, you will still need to learn many low-frequency, high-specificity terms like public holiday, dissatisfaction, or lion tamer.

This article summarizes the vocabulary-learning strategies I have tried over the last two years - including a new technique I’m experimenting with now.

Bad Idea: Learning Words in Isolation

The worst (yet most common) approach is to memorize the dictionary.

Example:

염려하다: To be anxious.

This card has problems.

  1. It’s in dictionary form.
    Native speakers rarely say 염려하다 in its unconjugated state. You’re more likely to encounter 염려해요 or 염려했습니다.

  2. It relies on an English translation.
    English translations often sound natural in English but do not reflect nuance in Korean.
    For example, 문득 is often translated as suddenly, but that flattens the meaning. 갑자기 is suddenly. 문득 means when a thought pops into your mind quickly, and in forms like 문득문득, it carries a warm, reflective tone closer to from time to time, a thought comes to me out of nowhere.

Many Korean words have this kind of subtle emotional or situational nuance that translations accidentally erase. This is similar to the difference between famous and infamous in English - same root, different feeling.

  1. You lose all sense of context. Words do not live alone; they live in patterns. Memorizing an isolated equivalence like envy = 부러움 doesn’t show you how the word behaves in sentences.

Exception: Tangible Nouns

I’m more relaxed about memorizing simple physical nouns in isolation.
Bell, carrot, and fridge are all straightforward. A bell is a bell.

But even here, translations sometimes hide nuance:

Cup => 컵 (a noun) vs. 잔 (a counter word)
A word can still sit inside a network of relationships you don’t see from one English translation.

Still, for pure, concrete objects - 고양이, 망치, 바나나 - memorizing a simple definition is fine.

Vocabulary is a numbers game, and tangible nouns are easy wins.

Experiment: Memorizing Full Sentences

To capture nuance, I tried learning full sentences instead of isolated pairs.

Example for 문득문득 (from the official dictionary):

문득문득 생각이 나다.
Suddenly, thoughts come to mind.

This is a big improvement. Now the word is connected to thought, not just a generic suddenly.

But… I ran into problems.

1. Good sentences are long; flashcards must be short.

Finding the perfect sentence - short enough to recall quickly but rich enough to convey nuance - is slow. Multiply that by thousands of words and it becomes unmanageable.

2. ChatGPT tends to recycle certain words.

If you ask for thousands of example sentences, eventually everything involves 느낌, 필요, 계획, 노력, etc.
You end up learning sentences that sound like a motivational poster factory.

3. Dictionary example sentences are surprisingly inconsistent.

After copying hundreds of them, I realized many are too formal, oddly phrased, or not actually representative of modern usage.

4. Full sentences are hard to memorize and hard to self-grade.

Did you forget a particle? Did you accidentally switch a verb ending?
A sentence card feels like juggling five balls just to learn one word.

Even when using an AI-graded app like Koala Cards, the small mistakes add friction. Without feedback, it’s difficult to know whether you actually produced a natural sentence.

Experiment: Chunks and Collocations

To simplify without losing context, I moved to lexical chunks - small word combinations that commonly occur together.

Example for 실질적이다:

실질적인 도움
Practical help

A short chunk like this is easier to memorize than a sentence but still shows how the word behaves.

Why chunks work

  • They mirror how words naturally occur in real speech and writing.
  • They teach you the company a word keeps, which is often the heart of vocabulary meaning.
  • You can combine chunks later to build real sentences.

What is a collocation?

A collocation is a pair of words that frequently appear together.

English:

  • commit a crime (not perform a crime)
  • heavy rain (not strong rain)

Korean:

  • 기분이 상하다
  • 시간을 들이다
  • 관심을 보이다
  • 잔소리를 하다

Words have favorite partners. Learning these is far more powerful than learning translations.

New Idea: Monolingual Korean Definitions

My newest experiment:
Learn the isolated word - but with a full Korean definition, not an English one.

Example (문득):

문득
갑자기 어떤 생각이 떠오를 때 쓰는 말이에요.

Suddenly the meaning feels alive. You don’t just know the translation - you understand the sensation of the word.

This also solves a big problem with flashcards in general, and that is the fact that translation should be avoided in language learning. Language Jones, one of my favorite language learning YouTubers, published a great video explaining why translation is bad for language learning.

The Problem with Translations

To make this method even better, I follow two additional rules:

1. Conjugate verbs.

Never memorize the plain dictionary form, since it is the least likely form to be used.

I use present-tense 요-form:

  • 강조하다 => 강조해요
  • 흔들리다 => 흔들려요
  • 의심하다 => 의심해요

This trains my brain to think in real usage patterns, not grammatical stems.

2. Use noun-modifying forms for adjectives.

Adjectives behave like verbs in Korean, so learning them as modifiers is more natural:

  • 어색하다 => 어색한
  • 조심스럽다 => 조심스러운
  • 귀찮다 => 귀찮은

You’ll use these forms constantly, so you might as well memorize that version.

Why This Might Work Better

This monolingual approach keeps the simplicity of single-word cards while preserving nuance and context. It also forces you to interact with Korean as Korean, not as a shadow of English.

It’s early in the experiment, but so far it feels promising - less friction than full sentences, but far richer than English translations.