What Sentence Mining Actually Does
Sentence mining takes the unknown words you hit in immersion material and converts them into known words. That single conversion is what moves comprehension and fluency forward. When a word keeps appearing in films, podcasts, books, or games you already care about, your brain is primed to hold onto it, and mining is what captures that moment instead of letting it slip.
The timing rule comes from linguist Stephen Krashen’s input hypothesis: learners absorb vocabulary best from sentences they already understand except for one unknown word. That “i+1” gap is where anki sentence mining pays off fastest.
Immerse First, Then Mine i+1 Sentences
Mining and immersion are two different activities, and running them at the same time wrecks both. Pause to look up every unfamiliar word and the immersion flow dies. Skip lookups entirely and vocabulary you could have captured disappears with the episode.
So split them. Watch the episode, finish the chapter, play the gaming session, all without reaching for a dictionary. Afterward, when you notice a word showed up more than once and you never quite got it, that’s your mining signal.
Aim for i+1 sentences: otherwise comprehensible, with exactly one unknown word. Context carries most of the meaning, so the single gap stands out and retention climbs.
Check for Duplicates Before Making New Cards
Before you build a new card, check whether that word is already sitting in your deck. If it is, reprioritize the existing card rather than hand-making a second one. Duplicates eat review time and muddy your sense of what you actually know, and the fix is just resurfacing a card you already own.
Core Tools for Manual and Batch Mining
Two approaches cover almost everything: single-card lookups for a handful of words, and batch mining that pulls hundreds of cards from one text or video.
Yomitan for Single-Card Lookups
Yomitan is a free browser extension that looks up Japanese words and sends a card to Anki in one click. Install it, highlight a word on any page, click the popup, and the card lands in your deck.
The trade-off is scope: one card at a time, and it cannot record audio on its own. For the handful of words left over after an immersion session, that’s plenty. Larger sessions are where batch tools pull ahead. If you want Yomitan working on local text files opened in Chrome, turn on Incognito and file URL access in Chrome’s Extensions settings.
Migaku for Batch Mining
Migaku is paid, and it batch-mines hundreds of cards at once from text or video while tracking which words you already know. Nothing free matches that combination of batch speed and word tracking, which is the honest case for paying.
The color coding tells you where every word stands: red means unknown and counted for mining, yellow means seen but not yet known and not counted, green means known and excluded from mining, gray means ignored and excluded entirely.
Reach for it when you have a block of text or a video subtitle file to process. Load the text, scan the flagged words by color, export the batch to Anki. Local files need the same Incognito and file URL permissions in Chrome’s Extensions settings.
Textractor for Game Text Extraction
Textractor hooks into video games, visual novels especially, and pulls out in-game text you otherwise cannot copy. Export what it captures, then run it through Yomitan or Migaku like any other text source.
Sourcing Audio and Video for Mining
Text-to-Speech Audio
Gyoza (also written ‘gavo’) is the first choice for generating pitch-accurate sentence audio. It sits above voice synthesizers like Koeiromap because it preserves the intonation patterns that let you recognize a word when someone actually says it.
Video Subtitles
For batch mining from video, use a real SRT subtitle file instead of YouTube’s autogenerated captions. Migaku depends on exact word boundaries to flag unknowns, so accurate manual transcription is what makes the batch reliable.
When there’s no text transcript at all, raw YouTube audio or audio-only material with no script, mining generally isn’t worth it. Checking whether autogenerated subtitles are correct assumes you’re already advanced, and a learner-stage miner burns the session fixing errors instead of studying.
Replacing Low-Quality Audio
Some sources hand you clips that are poor quality or spoken by non-native speakers with wrong pronunciation. Swap those for Forvo recordings or a local native audio bank you’ve built up yourself.
Tools to Avoid and Batch Editing Shortcuts
Skip Jidoujisho, JTAT, and Subs2SRS. The cards they auto-generate arrive without proper definitions and need so much manual repair that the speed advantage of batch mining evaporates.
Once Migaku has produced a batch, Anki’s own note operations finish the job fast. “Pack reading from meaning” fills the reading field across many cards, and “pitch accent bulk add” generates pitch accent graphs in bulk. Between them they replace an evening of card-by-card editing.
Turning a Mining Log Into Reviews Without Rebuilding It Twice
The part that stays manual is everything around the cards. You keep a running log of mined sentences somewhere, you check it against your deck so a word doesn’t get carded twice, and you keep vocabulary from immersion on a different review system than everything else you’re studying.
Picture the same log doing double duty: the sentences you wrote down become review material on their own, one record per word, one place to check before you make a card.
Fluxo organizes your own notes into spaces and topics, generates flashcards, quizzes, and summaries from what you wrote, and schedules spaced-repetition review that resurfaces the material over time. You write the mining notes; the review scheduling runs on top of them at https://fluxo.today.