How to Extract Vocabulary From EPUB Books Without Getting a Useless Deck
Pulling vocabulary from an EPUB sounds easy until you realize most extracted word lists are bloated and practically unreviewable.
Short enough to finish in one sitting.
Clear chunks to keep momentum up.
Structured for uninterrupted reading, not skimming.
Designed to be finished, not skimmed.
Use the progress helper while reading. Once you reach the end, the next section will hand off to closely related posts instead of dropping you back into the full archive.
Extracting vocabulary from an EPUB book sounds like a solved problem.
Take the text, identify the hard words, translate them, export a deck. Done.
Except most outputs are garbage.
The output looks smart right until you try to review it
The first problem is that raw extraction gives you way too much. Proper nouns. One-off adjectives. Weird technical terms that only matter in a single paragraph. If you throw all of that into a deck, you get something that looks impressive and feels terrible.
I've seen this happen over and over. A learner runs one book through a vocab tool, gets hundreds of cards back, reviews them once, and never touches the deck again. The tool technically worked. The result was still useless.
The real challenge is not extraction.
It's filtering.
I had exactly the same reaction the first time I generated a giant deck from a book. The list looked impressive for about 30 seconds. Then I scrolled it. Page after page of words I clearly wasn't going to review. That was the moment it clicked that extraction is the easy part. The hard part is deciding what deserves to survive.
A book is full of words that should stay in the book
Useful vocab from books has to meet a higher bar. The word should either block understanding, repeat often enough to matter, or feel like the kind of vocabulary you will actually run into again outside that single page. If it fails all three tests, it probably doesn't belong in the deck.
That sounds obvious, but most tools are built to collect, not to refuse. They treat every unknown token as equally valuable. Readers know that is nonsense.
Some words should stay in the chapter and die there.
That is fine.
Context is what turns a word into a usable card
The sentence matters too. A naked word list is weak. A word plus the original sentence is much stronger. When you review later, you aren't just seeing an item. You are seeing where it lived. That makes recall easier and the whole thing less abstract.
That is the difference between a generic deck and a book-based deck that still feels alive a week later.
If the word came from a scene you actually remember, your brain has something to grab onto. If it is just a detached translation pair, it is much easier to ignore during review.
A useful EPUB deck is smaller than you want
If you're extracting vocabulary from EPUB books, aim for a deck you might actually finish reviewing. Not the biggest possible list. Not the most complete list. Just one that still feels reasonable when you're tired on a Wednesday night.
That usually means:
- keep the original sentence
- cut one-off rare words aggressively
- delete anything you already half-know
- stop before the deck turns into a guilt machine
That standard cuts out a lot.
Good.
You made it through the full piece.
This is where most blogs lose the reader. Instead of sending you back to a noisy list, we surface the next few posts that stay on the same learning thread.