If you’ve ever crammed a vocabulary list, aced the quiz, and then forgotten everything two weeks later — the problem wasn’t your memory. It was the method.

Isolated word lists are the default vocabulary strategy for most language learners. They’re also one of the least efficient ways to build lasting lexical knowledge. The reason has to do with something researchers call word knowledge depth — and it turns out that knowing a word is far more complex than knowing its translation.

Sentence mining — the practice of extracting full sentences from authentic content and turning them into flashcards — is the convergence of nearly everything cognitive science knows about how vocabulary is actually acquired. It’s not new. It’s been a core practice in the immersion learning community for over a decade. But it remains surprisingly unknown outside that world.

This article explains why it works, how to do it, and how to avoid the most common mistakes that derail the process.

Why Vocabulary Lists Are Inefficient

In 2001, Paul Nation published what remains the definitive framework for understanding vocabulary knowledge. His key insight: knowing a word is not a single thing. It’s at least nine things.

Nation identified three broad dimensions of word knowledge — form, meaning, and use — each with receptive and productive aspects. Knowing the word “break,” for instance, means knowing:

A vocabulary list gives you exactly one of these dimensions: a single meaning paired with a single form. Everything else — collocation, register, grammatical behavior, pronunciation, polysemy — is missing. You end up with what researchers call brittle knowledge: a word-to-translation mapping that works on a flashcard but collapses in real use.

This is why learners who study word lists can often recognize vocabulary on a test but can’t use the same words in conversation. They have the thinnest possible version of word knowledge.

There’s a deeper problem. Laufer and Hulstijn’s Involvement Load Hypothesis (2001) demonstrated that vocabulary retention is directly proportional to the cognitive effort involved during encoding. They identified three components of involvement:

  1. Need — is there a genuine communicative reason to learn this word?
  2. Search — did the learner actively look up or figure out the meaning?
  3. Evaluation — did the learner compare the word to other words or decide how to use it in context?

A pre-made vocabulary list scores low on all three. The words were chosen by someone else (no need), the translations are handed to you (no search), and there’s no context requiring you to evaluate usage (no evaluation). The involvement load is minimal — and so is retention.

What Makes a Sentence Card Different

A sentence card is a flashcard built around a complete sentence extracted from real content, with one target word as the focus. Here’s what a well-made sentence card looks like:

Front: 彼女は約束を破ったのに、謝らなかった。

Back: She broke her promise but didn’t apologize. (破る = to break [a promise, a rule])

Compare this to a standard word card:

Front: 破る Back: to break; to tear; to violate

The sentence card is doing several things simultaneously:

1. Context disambiguates meaning. The word 破る has multiple senses. The sentence locks in the specific sense of breaking a promise — not tearing paper or violating a law. Webb (2007) demonstrated that learning words in context leads to significantly greater gains in all dimensions of word knowledge compared to learning from word pairs alone.

2. Collocation is encoded automatically. You’re not just learning 破る — you’re learning that promises are the kind of thing you 破る. This collocational knowledge is almost impossible to acquire from word lists but emerges naturally from sentence-level exposure.

3. Grammar comes free. The sentence exposes you to the ～たのに (“even though”) construction, the past tense conjugation, and the negative form of 謝る — none of which you explicitly studied, but all of which your brain is processing.

4. The generation effect kicks in. When you see the sentence on the front of the card and try to understand it, you’re doing active cognitive work — parsing the grammar, inferring meaning from context, reconstructing the message. This effortful processing is exactly what the testing effect and the generation effect predict will produce stronger memory traces.

5. Emotion and narrative create durable memories. A sentence about someone breaking a promise and not apologizing has a small emotional charge. It’s a story fragment. Emotionally salient content is processed more deeply and remembered longer — this is the “E” (Emotion) in the CREED framework that synthesizes the conditions for optimal acquisition.

Pellicer-Sánchez and Schmitt (2010) found that incidental vocabulary learning from reading — encountering words in meaningful context — leads to robust gains across multiple dimensions of word knowledge, including the kind of deep knowledge (collocation, grammatical behavior) that explicit list-learning rarely produces.

The Complete Sentence Mining Process

Sentence mining is simple in concept and requires discipline in execution. Here’s the step-by-step process:

Step 1: Consume content in your target language

Watch a show, read a book, listen to a podcast, browse a website. The content should be at your level — ideally, you understand 90–98% of the words. This is the “comprehensible input” threshold that Nation and Webb identified as optimal for incidental vocabulary learning.

Step 2: Notice an unknown word in context

You encounter a sentence where you don’t know one word, but you understand enough of the surrounding context to make sense of the sentence with a dictionary lookup. This is the ideal candidate for mining.

Key rule: one unknown word per sentence. If a sentence has three words you don’t know, it’s not a good mining candidate — you won’t have enough context to anchor the meaning. Move on and find a cleaner example.

Step 3: Look up the word

Use a quality dictionary — not just Google Translate. For Japanese, that might be Jisho or a monolingual dictionary. For European languages, WordReference or Linguee. You want the specific meaning that fits this context, plus any collocational or usage information.

Step 4: Create the card

Step 5: Review via spaced repetition

Add the card to your Anki deck and review it using active recall. When you see the sentence, try to understand it — including the target word — before flipping to check. Rate honestly. The SRS algorithm handles the rest.

Step 6: Repeat daily

The process becomes a habit: consume content → notice words → mine sentences → review cards. Over time, your vocabulary grows organically from material you’ve actually encountered and cared about — not from a list someone else curated.

Optimal Sources for Mining by Level

Not all content is equally mineable. The right source depends on your level.

Beginner (0–1,000 words known)

At this stage, authentic native content is mostly incomprehensible. Your mining sources should be:

You may also benefit from a pre-made frequency deck for your first 500–1,000 words. These aren’t sentence-mined cards, but they establish the base vocabulary that makes sentence mining possible. Think of them as scaffolding you’ll eventually discard.

Intermediate (1,000–5,000 words known)

This is the sweet spot for sentence mining. You know enough to understand most of what you encounter, but you’re constantly hitting new words in context. Ideal sources:

Advanced (5,000+ words known)

At this level, unknown words become rarer, and mining becomes more targeted:

Tools That Make Mining Efficient

Sentence mining by hand — pausing a video, typing out the sentence, looking up the word, creating the card — is slow. Several tools dramatically speed up the process:

Migaku

The most comprehensive sentence mining toolkit available. Migaku is a browser extension and Anki integration that lets you mine sentences directly from Netflix, YouTube, and web pages. One click captures the sentence, audio, screenshot, and dictionary definition. It generates Anki cards automatically. For serious miners, this is the gold standard.

Language Reactor

A free Chrome extension that enhances Netflix and YouTube with dual subtitles, word-by-word translations, and the ability to save sentences. Less automated than Migaku but effective and accessible.

Anki Add-ons

If you prefer a manual workflow, several Anki add-ons help:

The Manual Process

No tools? No problem. The manual process works fine:

  1. Keep a note-taking app open while consuming content
  2. When you encounter a mineable sentence, copy or type it
  3. Look up the unknown word
  4. Create the Anki card manually
  5. Add audio via a TTS service if desired

This is slower, but the extra effort of typing out the sentence may actually improve initial encoding — another instance of desirable difficulty at work.

How Many Cards Per Day: Avoiding the Review Tsunami

This is where most sentence miners fail. Not in creating cards — in drowning under reviews.

Every new card you add today generates reviews for weeks, months, and years into the future. The math is unforgiving. If you add 30 new cards per day, within a month you’ll be facing 200+ daily reviews. Within three months, 300+. This is the review tsunami — and it kills consistency faster than anything else.

The sustainable numbers, based on community experience and SRS scheduling math:

New cards/day Daily reviews (steady state) Time required
5 ~35–50 10–15 min
10 ~70–100 15–25 min
15 ~100–150 25–40 min
20 ~140–200 35–55 min
30+ ~200–300+ 60+ min

The recommendation for most learners: 10–15 new cards per day. This is enough to add 300–450 new words per month — a pace that produces real, noticeable progress — without making reviews an unbearable chore.

The critical rule: never skip reviews. Adding new cards is optional on any given day. Reviewing due cards is not. If your review pile grows intimidating, stop adding new cards until it’s under control. A smaller deck reviewed consistently beats a massive deck abandoned after two months.

If you’re using FSRS (Anki’s modern algorithm), set your desired retention to 90%. This gives you the optimal trade-off between retention and review load — the same level of retention with 20–30% fewer reviews than the legacy SM-2 algorithm.

Mining What You Enjoy vs. Frequency-Optimized Content

There’s a tension at the heart of sentence mining: should you mine from content you love, or from content optimized for high-frequency vocabulary?

The frequency argument is straightforward. Zipf’s law tells us that the most common 2,000 words in any language cover roughly 80–90% of everyday speech. If you systematically mine the highest-frequency words first, you maximize comprehension gains per card created. This is the efficient path.

The enjoyment argument is equally compelling. The research on motivation — particularly Self-Determination Theory (Deci & Ryan, 2000) — shows that autonomy and intrinsic interest are the strongest predictors of sustained engagement. If you find your mining content boring, you’ll stop mining. A method you abandon is infinitely less effective than a suboptimal method you maintain for years.

The CREED framework resolves this tension. Emotion is one of the five conditions for optimal acquisition. Content that genuinely interests you — a thriller series, a cooking channel, a history podcast — produces emotional engagement that enhances encoding. You remember vocabulary from scenes that made you laugh, characters you cared about, topics you were genuinely curious about.

The practical resolution:

Pellicer-Sánchez and Schmitt (2010) found that readers who were engaged with their reading material showed greater incidental vocabulary gains than those reading less interesting texts — even when exposure frequency was controlled. Enjoyment isn’t a luxury. It’s a variable that directly affects acquisition.

The Bottom Line

Sentence mining works because it aligns with how the brain actually acquires vocabulary. Not as isolated translation pairs, but as rich, multi-dimensional knowledge embedded in context, connected to emotion, reinforced through active recall, and spaced for optimal retention.

It combines the best of what we know from cognitive science: Nation’s dimensions of word knowledge, Laufer and Hulstijn’s involvement load, Webb’s context effects, the testing effect, spaced repetition, dual coding, and the generation effect. No other single vocabulary method integrates this many evidence-based principles simultaneously.

The process is simple: consume content you understand mostly, notice words you don’t know, extract the sentence, create a card, review it. The tools make it fast. The SRS makes it durable. The content makes it interesting.

Start with 10 cards a day. Mine from something you actually want to watch or read. Review every day without exception. In six months, you’ll have 1,800 words learned in context — with collocations, grammar patterns, and emotional associations attached to each one.

That’s not a vocabulary list. That’s a vocabulary.

This article is part of the series “The Science of Language Learning” — where we break down what research actually says about how adults acquire languages, and how to use that science to learn faster.

Previous in the series: How Your Brain Learns New Sounds (And Why Some Are So Hard)

Next in the series: The Complete Anki Setup Guide for Language Learners

References: