What Is Sentence Mining? The Rules, a Worked Example From a Real Video, and When to Stop
Sentence mining means collecting whole sentences from things you watched or read and understood — each with exactly one piece you didn't know yet (i+1) — and turning them into flashcards you review on a spaced-repetition schedule. Here are the three rules that make it work, a complete example card from a real Spanish video, a sane daily number, the tools that do it, and the point where it stops paying off.
Sentence mining is collecting whole sentences from content you understood — each containing exactly one thing you didn't know yet — and turning them into flashcards you review on a spaced-repetition schedule. The "exactly one" is the i+1 idea: everything else in the sentence is already yours, so the sentence itself teaches the new piece.
It grew up in the Japanese immersion community — AJATT, then MIA, then Refold, each of which describes its own version — and spread from there to every language people learn by watching and reading. This page is the method itself: the rules, one complete example card, how much to mine, which tools do it, and the honest part about when it stops being worth the time.
The three rules
Most of sentence mining's reputation — good and bad — comes down to whether these three are followed:
- One new thing per sentence. One unknown word, or one unknown chunk, or one grammar point you haven't seen used like this. A sentence with three unknowns isn't a card, it's a puzzle — you'll fail it in review without knowing which part you forgot. Skip it, or find the same word in a simpler sentence later. It will come back.
- Mine from content you actually understood. The sentence should come from a video, show or article you were following — your comprehensible input, not something you were struggling through. Comprehension is what makes the rest of the sentence free.
- Keep the sound, or at least the context. Audio of the original line, or a link back to the moment you heard it, turns a reading card into a listening card and lets you re-check what was actually said. Without it, the sentence drifts away from the voice and the scene that made it memorable.
A complete example, from a real video
In a Spanish comprehensible-input video by Español con Juan, Juan describes arriving in London in 1997, barely understanding English, and watching a sitcom anyway. At 4:01 he says:
Yo lo veía porque me ponía de buen humor.
If you're an intermediate learner, you know yo, lo, veía (I watched it) and porque, and humor looks like English. The one new piece is me ponía de buen humor — "it put me in a good mood". That's an i+1 sentence: one unknown chunk, carried by everything you already know. It's also a good chunk to keep: de buen humor and its opposite, de mal humor, are phrases you'll keep meeting.
The card:
| Front | Back |
|---|---|
| Yo lo veía porque me ponía de buen humor. — plus the audio of Juan saying it | "I watched it because it put me in a good mood." · poner (a alguien) de buen humor = to put someone in a good mood · ponía: imperfect, a habit — it did, every time · link back to 4:01 |
Notice what's not on the card: no isolated "humor = mood", no conjugation table, no second unknown. When this card comes back in a week, the job is to hear the sentence and understand all of it. If you can, the chunk has stuck.
How many sentences a day?
We couldn't find a number with research behind it, so here is a suggestion, labelled as one. Start with five to ten new sentences a day, and let your review load tell you when to change it. Each new card comes back several times over the following weeks, so today's number decides how long your reviews take a month from now — and a sentence card takes longer to review than a single word.
For reference, Anki's built-in default is 20 new cards a day (set in its source code), and LingoKeep's default is 10. Either is a ceiling, not a target. If your daily review starts eating the time you had for watching, mine less; the input is the main course.
The tools that do it
The method needs three things: something that makes subtitles or text clickable, a way to grab the sentence (and ideally its audio) into a card, and a spaced-repetition scheduler. Here's what each common option covers — capabilities from each tool's own site or documentation, checked October 2026:
| Tool | What it does in a mining setup | What a card can carry | Where you review | Cost |
|---|---|---|---|---|
| Anki | The scheduler most setups feed into | Whatever note type you build | Anki, on desktop, web or phone | Free on computers and Android; the iPhone app is paid |
| Yomitan | Pop-up dictionary for text in the browser; creates Anki cards through the AnkiConnect add-on | Word, reading, definition, the sentence it appeared in, dictionary audio, a screenshot, the page URL | Anki | Free, open source; you import the dictionaries yourself |
| asbplayer | Clickable subtitles on streaming video (YouTube and Netflix subtitles auto-detected); opens a card editor on the current line | Sentence, word, definition, an audio clip, a screenshot, a link back to the source | Anki | Free, open source |
| Migaku | Browser extension plus its own study app | The word, its context sentence, audio, a screenshot, a definition and an AI explanation | Migaku's own spaced-repetition app, with iOS and Android versions | Paid — $10 a month or $96 a year, after a 10-day free trial |
| LingoKeep | YouTube extension: dual subtitles, click-a-word dictionary, one-click save | Word, dictionary form and meanings, the full sentence, its translation, a link to the exact second | Built-in FSRS review in the extension or your phone's browser; Anki export on Pro | Free for 100 cards with review free forever; Pro $4.99 a month |
The classic setup is the first three together — free, flexible, and a real evening of configuration before your first card. Migaku packages the whole loop for a subscription. LingoKeep packages it for YouTube only, and leaves the review free.
When sentence mining stops being worth it
It's a method for a stage, not for ever, and it's worth knowing the signs:
- Every sentence is i+0. When most of what you watch has nothing new in it, the sentences worth mining get rare and each one costs more searching than it pays back. That's not failure — it's the point at which volume of input does the work cards used to.
- Review is crowding out input. If you're spending more time on cards than on watching or reading, the cards have stopped serving the input and started replacing it. Cut new cards to zero for a week and watch.
- You're mining i+3. Cards from content you didn't understand fail repeatedly and teach little. That's a sign the content is too hard, not that you need more cards.
- It's become a chore you dread. A method you stop doing is worse than a lighter one you keep. Five cards a day you actually review beat thirty you abandon in March.
Where LingoKeep fits
Full disclosure: LingoKeep is ours, so the limits first. It only works on YouTube, in desktop Chromium browsers, for Spanish, Japanese, French and German. It doesn't do Netflix, books or web pages, and it captures dictionary meanings and a link back to the moment rather than an audio file — the review card replays the clip from the video instead. If your mining happens across many sources or in another language, the Anki + Yomitan + asbplayer setup is the honest recommendation.
What it does is the one-video version of the whole loop without the setup: click the word in the subtitle line, save it, and the card keeps the sentence, its translation and the timestamp; drag across a phrase to save a chunk like me ponía de buen humor as one card. Review is free, on an FSRS schedule, and Anki export is there on Pro if your cards belong in a bigger deck.
Where to go next
- How mined cards fit into Anki, and which card types work: Anki for language learning
- The Japanese side of the tradition, with the channels to mine from: comprehensible input for Japanese
- Why the schedule matters as much as the sentence: FSRS explained
- Moving LingoKeep cards into Anki: the Anki export guide
FAQ
Q: What is sentence mining? A: Sentence mining is collecting whole sentences from content you watched or read and understood — each with exactly one new word, chunk or grammar point (i+1) — and turning them into flashcards that a spaced-repetition scheduler brings back over the following weeks. The rest of the sentence is context you already know, so the card teaches the new piece in use rather than as an isolated translation.
Q: Is sentence mining worth it? A: For most learners between the beginner and advanced stages, yes: it turns words you keep meeting and forgetting into words you recognise, at the cost of a few minutes of review a day. It's worth less at the very start, when you can't yet find sentences with only one unknown, and at advanced levels, when new material is rare enough that more input teaches faster than more cards.
Q: How many sentences should I mine per day? A: There's no researched number. As a starting suggestion, five to ten new sentences a day, adjusted by how long your daily review takes — each new card adds reviews for weeks. Anki's default limit is 20 new cards a day and LingoKeep's is 10; treat either as a ceiling rather than a target.
Q: Sentence mining vs vocabulary decks — which is better? A: They do different jobs. A pre-made vocabulary deck is fast to start and covers common words in a fixed order, but the sentences on it aren't ones you met, so there's no memory of a scene to hang them on. Mined sentences come from your own input, so they're the words you actually need, in context you already understood. A reasonable split is a small frequency deck at the very start, then mining once you can follow real content.
Q: Do I need Anki for sentence mining? A: No. Anki is the most common scheduler for it, and the classic toolchain (Yomitan and asbplayer feeding Anki through AnkiConnect) is free. But any spaced-repetition system works: Migaku has its own study app, and LingoKeep has free built-in FSRS review for YouTube sentences, with Anki export on Pro if you want them there later.
Keep reading
Anki for Language Learning: What It's Great At, and the One Place It Breaks
Anki is still the best spaced-repetition tool for language learners, and almost nobody quits because of the algorithm. They quit because of where the cards come from. Here's what Anki does well, the card types that actually work for languages, and how to fix the part that breaks.
FSRS Explained: How the Spaced Repetition Algorithm Actually Works
FSRS is the open-source algorithm that decides when you should review a flashcard so you remember it with the least possible effort. Here's how it models your memory with three numbers, why it replaced the 1987-era SM-2 formula, and how to actually use it — in Anki or anywhere else.
A Japanese Immersion Roadmap: How Many Hours Japanese Takes, and the Six Stages in Between
Japanese takes an English speaker thousands of hours: the US Foreign Service Institute budgets 88 weeks (2,200 class hours) for it, and Refold's immersion estimate for a language that distant is about 2,700 hours of total effort to functional fluency, double its figure for Spanish. Here's what each of those numbers measures, a six-stage roadmap from kana to speaking with rough hours for every stage, the parts only Japanese has (kana, kanji, reading), and the one thing hour-counting doesn't buy you.
The Dreaming Spanish Roadmap, Explained (2026): All 7 Levels, the Hour Numbers, and What They Don't Tell You
The Dreaming Spanish roadmap is seven levels and six hour milestones from 0 to 1,500 hours of input. Here's each level in the roadmap's own words, what each one actually feels like, when the method says to start speaking and reading, and the one thing hour-counting doesn't buy you.