Practice Reading Korean Words: Why Re-Reading Isn't Reading, and What Is

The situation arrives on schedule. You learned the hangul chart in a weekend. You can sound out anything — street signs, song lyrics, the side of a snack bag — then you sit down with an actual article and read like a first grader, so slowly the front of the sentence dies before you reach its verb. So you do what everyone does: you read the page again, it feels smoother, and you take that for progress. You shouldn't. Re-reading is the thing you were already doing. Usual disclosure: I build Piccard, a flashcard tool that runs in both directions on English and Korean. My argument below is that reading practice is retrieval practice wearing a different coat — and that the material it works on has to be your own.

What slow reading actually is

Two activities share the name. One is decoding: turning syllable blocks into sounds and holding them in working memory until the sentence assembles. The other is recognition: seeing 그렇지만 and pulling the word out whole, no assembly, the way you pulled every word in this sentence.

Your reading speed is just the ratio between the two piles. Hangul is why the first pile stopped being the problem so quickly: it was built exactly once, on purpose — an alphabet promulgated by Sejong the Great in 1446, featural to the point of showing its work (consonant shapes echo the speech organs that pronounce them), with letters grouped into syllabic blocks (Wikipedia). That regularity is why you could pronounce 오늘 뭐 했어? long before you could retrieve it at speed. The script gave decoding away free; the bottleneck was never the script.

What reading research keeps finding is that fluent readers barely decode: they recognize words and common chunks as units and spend attention on meaning.

What memory research keeps finding is that recognition at speed is built by retrieval, not exposure: producing an answer strengthens memory more than seeing it again — a result old enough to have a name, the testing effect, with a canonical demonstration in Roediger and Karpicke's 2006 experiments. Re-reading is pure exposure: the page keeps showing you the answer, so the retrieving never happens. That's why the second pass feels better than it is and next week's new article doesn't.

Arithmetic, not measurement: call it two seconds to assemble a word you half-know, a quarter-second to retrieve one you own. A 120-line episode with fifteen strangers carries thirty seconds of assembly toll; once those fifteen are cards you've retrieved five times, under four. The page didn't change. The piles did.

The materials, honestly compared

Whatever you read, you practice the words that page happens to contain — the choice of material is the choice of what gets practiced. Each format trades differently:

If you want an external yardstick, TOPIK grades Korean from Level 1 to Level 6 across its two tiers — a ruler, not a method. This post is about the method: which pile each page moves your words into.

Chosen material is what survives

Assigned material dies in week three; chosen material persists, because the words that stick live in what you would have consumed anyway. This is the same fork the Japanese-direction sibling hit while taking apart the 1000 most common Japanese words: a shared core any good list covers, and a personal tail only your material supplies. The core is a list problem. The tail is a capture problem — and reading speed is mostly tail.

Here is the document route as it ships, Korean direction. Paste a Korean article into the app, or photograph a textbook page. Every word comes back as a candidate card tagged high, medium, or low for how much it is worth learning, sorted with the high tier on top and already ticked, and one confirmation turns at most 20 candidates into cards. The 20 is a named constant in our codebase, sized because the model spends roughly 300–350 output tokens per card — not a growth mechanic.

The ranking is a budget device: what you already read fluently sits at the bottom, skipped in one pass, so review time goes to the tail instead of being ground away re-confirming the core.

Turning watching into reading practice

On a YouTube video that has Korean captions, the extension shows the transcript line by line. Click one line — the line you clicked, only; there is no convert-the-whole-video button — and candidate cards come back from it: a word, its meaning, an example sentence. Keep the ones that outran you, keep watching. A video without captions has nothing to click, and the tool says so instead of improvising.

This is reading practice, not just vocabulary collection: a caption line is a complete short sentence at spoken pace, about something you chose, with the scene attached. The click marks the sentence you couldn't parse at speed, and the card it makes carries the sentence and the timestamp.

When that card comes up for review, the video replays from that line: the voice, the scene, the moment the word meant something. The retrieval re-presents the context the word was encoded in.

For text you read rather than watch, the companion is the double-click. Double-click any Korean word on any web page — a news article, a wiki page, a web novel, anything with selectable text — and a popup dictionary opens on it; one more click and it is a card. The point is flow: you don't stop mid-line to assemble a stranger; pop it, keep reading, let the schedule do the retrieving later.

The edges, stated plainly: uploading audio or video files for transcription is built for English audio only, so Korean listening material reaches your deck through the caption route, not file upload. The 20-candidate cap holds on every path — same constant, same arithmetic.

The schedule that turns pages into speed

The fade was measured in the 1880s and named the forgetting curve; the countermeasure, reviewing each item just before it drops with the gap lengthening every time you survive, is spaced repetition.

Scheduling in Piccard is FSRS-6 with zero configuration: a 0.9 target retention, four ratings (again, hard, good, easy), nothing else to tune. Review is where the retrieving happens — the reading you did all week becomes the quiz you take all week.

The plan I would run:

1. Read one thing you chose every day — article or episode — with the double-click companion open. Capture only what outran you; five reviewed cards beat forty abandoned ones.
2. Once a week, paste or photograph the week's material through the document route and confirm the high tier. Twenty at a time is the ceiling anyway, and roughly a week's worth.
3. On captioned videos, click the line you had to re-read to parse — the diagnosis and the cure in one gesture.
4. Empty the review queue daily, no exceptions. When a card replays its line, read the whole line aloud before you rate it.

If your questions run wider than Korean: which words deserve cards at all is argued in Which English Words Are Worth Learning? A Three-Part Filter — the filter is language-agnostic even though its examples are English. The capture-plus-review loop under the caption route is taken apart in How to Save Words from YouTube Subtitles (and Actually Remember Them). The exam-list version of the same suspicion, Japanese direction, is JLPT N5 Vocabulary: The List That Isn't Official.

The short version

Slow Korean reading is not a script problem. The script gave decoding away free; the speed lives in how many words you retrieve whole instead of assembling. Re-reading trains the wrong thing — exposure instead of retrieval — so the second pass always feels better than it is.

Practice reading by retrieving: pick material you would consume anyway, let the clicks mark the lines that outran you, rank the pasted page so review budget goes to the tail and not the core, and let the schedule bring each word back right before it fades.

The free tier is 150 energy (about 150 cards — at two or three captured lines per video, a couple of months of the habit before you pay anything), then prepaid top-ups at $5 / $10 / $20 — pricing here. No subscription, and topped-up energy never expires. To start with the caption route: Piccard on the Chrome Web Store.

All guides: Piccard Blog.