Sentence mining, done well

Sentence mining means collecting real sentences from things you read or watch, each with one new word or structure you want to learn, and turning them into flashcards. Because the word comes with its context, you learn how it’s used as well as what it means, and the card is easier to remember. Done well, it’s one of the most efficient ways to grow your vocabulary between reading sessions.

Done badly, it becomes a review pile that eats the time you meant to spend reading. Here’s how to do it well.

What sentence mining is

The idea spread through online language-learning communities in the 2000s. It was popularized by Khatzumoto’s All Japanese All The Time, and later by communities built around comprehensible-input methods such as Refold. The basic loop is simple:

1. Read or watch something in your target language.
2. Notice a sentence where you understand almost everything except one word or phrase.
3. Save that sentence with the new item marked.
4. Review it later with a spaced-repetition scheduler, so it comes back just before you’d forget it.
5. Meet it again in your reading, which locks it in.

The key constraint is in step 2. A good mined sentence has exactly one unknown thing in it. This is often called an i+1 sentence, borrowing Stephen Krashen’s shorthand for input just beyond your current level (more on that in comprehensible input, explained).

Why sentences beat isolated words

You could just save the word and its translation. Here’s why context is worth the extra effort.

Context carries meaning a translation can’t

A dictionary tells you quedar can mean “to remain,” “to be located,” “to fit,” “to arrange to meet” and more. The sentence ¿Quedamos el viernes? tells you which one you met, and how it’s used. Over time you collect the uses that matter.

Repetition in context is what builds a word

Research on vocabulary learning keeps finding that words are learned through repeated encounters, and that one meeting rarely does it. Stuart Webb’s experiments (”The effects of repetition on vocabulary knowledge,” Applied Linguistics 28(1), 2007) found that learners’ knowledge of new words grew on several measures as encounters went up from one to ten. A meta-analysis by Takumi Uchihara, Stuart Webb and Akifumi Yanagisawa (”The effects of repetition on incidental vocabulary learning,” Language Learning 69(3), 2019) confirmed that repetition helps, though it isn’t the only factor.

Flashcards give you some of those encounters on a schedule. Reading gives you the rest. A mined sentence is a bridge between them: it’s a review that still looks like real language.

Retrieval practice and spacing do the rest

Two of the most robust findings in learning research apply here. Retrieval practice: actively recalling something strengthens memory more than rereading it (Roediger & Karpicke, Psychological Science, 2006). And the spacing effect: reviews spread out over time beat reviews crammed together (Cepeda and colleagues’ meta-analysis, Psychological Bulletin, 2006). A spaced-repetition app does both. Our FSRS explainer covers how a modern scheduler picks the timing.

How to pick a good sentence to mine

Not every sentence with a new word is worth a card. Run each one through a few quick checks.

The checklist

• One unknown. You should understand every other word. Two unknowns make a confusing card. Three make a card you’ll fail forever.
• Short. Under about 15 words is ideal. Long sentences are slow to review and the target gets lost.
• Useful word. Prefer words you’ve seen before or expect to see again. A word that has come up three times this month is a better pick than a rare one from a poem.
• Clear context. The sentence should make the meaning guessable, or at least memorable. “He took the ___” teaches nothing. “He took the umbrella because it was raining” does.
• Natural language. Real sentences from things you enjoyed beat textbook examples. You’ll remember the scene.

What to skip

• Names, places and one-off technical terms.
• Words you can already guess correctly every time from context.
• Sentences that only make sense with the paragraph before them.
• Anything you saved just because it was unknown, not because you’ll need it.

How to build the card

There are two common formats. Both work. Pick one and stick with it.

A few details make cards much better:

• Highlight the target word so you know what you’re being tested on.
• Write the meaning in context, not every dictionary sense. “Arrange to meet (here)” beats a list of eight.
• Add audio if you can. Hearing the sentence connects the written word to its sound, which helps listening later.
• Keep the source. A note like “from chapter 3 of the book” brings back the scene, and the scene brings back the word.
• Grade on the target word only. If you understood the highlighted word, it’s a pass, even if you stumbled elsewhere in the sentence.

Some learners also add an image or a screenshot from the video. That helps if it’s quick. It isn’t worth five minutes per card.

How many sentences to mine per day

This is where most people go wrong. Every new card adds future reviews. Mine 50 sentences in an enthusiastic weekend and you’ll be paying for it for months.

As a rough guide:

• Beginners (under about 2,000 words): 5–10 new cards a day. Many common words will come back in your reading anyway.
• Intermediate learners: 10–20 a day is sustainable for most people with 20–30 minutes of review.
• Advanced learners: fewer, more carefully chosen cards. At this stage most unknown words are rare, and reading volume matters more than any one card.

A useful rule: your daily reviews should take less time than your reading. If flashcards start crowding out the input they were supposed to support, add fewer new cards for a while.

Common mistakes

1. Mining everything. If every unknown word becomes a card, your deck fills with words you’ll rarely see again. Be picky.
2. Mining from material that’s too hard. If a page has thirty unknown words, it won’t have many one-unknown sentences, and the ones it has will be shaky. Mine from text you mostly understand: around 95% coverage or above, the zone in the 98% rule.
3. Letting reviews pile up. A backlog of 400 due cards feels terrible and leads people to quit. Missing a day is fine. Then cut new cards until you’re caught up.
4. Treating the deck as the method. Mining supports reading. It doesn’t replace it. You’ll learn many more words from meeting them in stories than from cards.
5. Grading generously. Pressing “Good” when you half-remembered teaches the scheduler that the card is easier than it is, so you’ll see it too rarely.

Tools for sentence mining

The classic setup is manual: copy a sentence into Anki, add the meaning, and let Anki’s scheduler handle the rest. It’s flexible, free on desktop and Android, and fully under your control, but making each card by hand is slow.

Many tools now automate the capture step. Migaku can make a card from a word in a subtitle or web page with a keystroke, including the sentence, audio and a screenshot, and send it to its own flashcard app or to Anki. Language Reactor lets you save words and phrases from subtitles with their context and export them. These are especially good if most of your input is video.

If most of your input is reading, a reader with built-in saving keeps the whole loop in one place.

How Lumo handles it

Lumo, the reading app we’re building, isn’t available yet, but this is how it will work. Tapping a new word in any text saves it with the sentence it came from, and saved words become flashcards automatically, scheduled with FSRS. You can also save multi-word phrases. The words you save are highlighted when they turn up again in anything else you read, so each new encounter is easy to spot. It doesn’t replace careful card-building for people who love it, but it takes the copying and pasting out of the loop.

A simple daily routine

1. Read first, 20–30 minutes of something at a comfortable level.
2. Save as you go, but only sentences that pass the one-unknown test.
3. Review after, 10–15 minutes of due cards.
4. Watch your ratio. If reviews take longer than reading, mine less for a week.

That’s the whole method. The sentences come from your reading, the reviews keep them alive, and the reading brings them back again.

FAQ

What is sentence mining in language learning?

Sentence mining is collecting real sentences from books, articles or videos, each with one new word or structure, and reviewing them as flashcards with spaced repetition. It teaches words together with their context, which makes them easier to remember and use.

Is sentence mining effective?

It builds on well-supported findings: words are learned through repeated encounters in context, and retrieval practice with spaced reviews improves memory. It works best alongside lots of reading or listening, not as a replacement for it.

How many sentences should I mine per day?

Most learners do well with 5–20 new cards a day, depending on level and available time. Watch your review load: if reviews start taking longer than your reading, cut back on new cards.

Should I mine sentences or individual words?

Sentences usually make better cards because the context shows how the word is used and helps you remember it. A word-only card is faster to review, so some learners use word cards for very common, concrete words and sentence cards for everything else.

What is an i+1 sentence?

It’s a sentence where you understand everything except one new word or grammar point. The term borrows Krashen’s “i+1” idea of input just beyond your level. These are the best sentences to mine, because the context makes the new item clear.