Sentence mining on a Mac,
in any app.

What sentence mining actually is

You are reading. Up comes a word you do not know. Instead of writing the word on a list, you keep the whole sentence, with that word inside it, and that becomes the card.

The reason it works is that the sentence carries everything the word list throws away. Which sense was live. What register it was in. What it sat next to. A word on its own is trivia you will fail to recognise in the wild.

The hard part was never the idea. It is that collecting the sentence, by hand, thirty times a day, is more work than anyone keeps up.

Yomitan and Language Reactor live in the browser

Yomitan is very good, and it is a browser extension aimed at Japanese. Language Reactor is very good, and it reads subtitle tracks on specific streaming sites. Both do their job well inside their box.

Outside the box there is a lot of reading. A PDF in Preview or Skim. A paper in Zotero. An ebook in a desktop reader. A film in IINA with the subtitles burned into the picture. A lecture recording. A screenshot a classmate sent you.

None of that has a page source for an extension to read.

Box the word, keep the line

Press ⌥A and drag a box around the word. macOS text recognition reads that patch of screen, and the line the word sits in comes along with it. The explanation you get back is for the sense this sentence is using, not a list of every meaning the dictionary knows.

Both go into your library together. That takes about a second, which is the whole point, because a second is cheap enough to do thirty times without noticing.

The cards at the end

When you want a deck, select the words, choose which fields land on the card and which of the layouts to use, look at the preview, and export. Anki imports the file the normal way.

If you would rather not leave the app at all, Leafy has its own review built in, and it shows the word inside the sentence you saved it from with the word blanked out. Same idea as a cloze card, without the export step.

The mistake almost everyone makes first

Mining feels productive, so people mine everything. Two weeks later there are four hundred cards waiting and the whole thing gets abandoned.

Read material where you already understand most of the sentence and meet roughly one unknown word in it. Mine that word. Skip words you are only interested in because they sound impressive, because you will not meet them again and the card will haunt you for a year.

Ten to twenty new cards a day is plenty. Reviews settle at somewhere around ten times your daily new count, so twenty new means two hundred reviews a day once it matures. Decide whether you want that before you decide you want more.

Where the text goes

Your library lives on your Mac. No account, no sync server, no ads. The picture is read on your machine and never leaves it.

The text you scan does go out to an AI so it can explain the word as your sentence uses it, with nothing attached that says who you are. The privacy page spells out exactly what is sent and what is kept.

Mine the sentence, not the word list.

Free while Leafy is in beta. One download, no account, macOS 14 and up.

Also worth a look: building the deck, mining from subtitles, or why Anki eats your evening.