Reading for vocabulary
Reviews consolidate words. They do not supply them. Almost everything worth putting in a deck arrives from somewhere you were reading anyway.
Meeting a word is not learning it
A single encounter with an unfamiliar word in context leaves very little behind — usually a vague sense of having seen it, occasionally a wrong guess about its meaning that then has to be unlearned. Estimates of how many encounters incidental acquisition actually requires vary a great deal depending on how you measure, but they are consistently in the range of many rather than several: something like a dozen meetings, spread out, before a word is reliably available.
This is not an argument against reading. It is an argument against expecting reading alone to do the job on any useful timescale, and it is the entire case for keeping a deck. Reading supplies the encounters and the context; the deck compresses a dozen accidental meetings into half a dozen deliberate ones.
Choosing what to read
The variable that matters is the proportion of words you already know. A text where you know almost everything gives you fluency practice and few new words. A text where you know four words in five is not really readable at all — you can decode it with a dictionary, but you are translating, not reading, and almost nothing sticks because there is no comprehended context for it to stick to.
The productive band is narrow and higher than most people assume. As a rough working test: if you meet more than about one unknown word per two or three lines, the text is too hard for extensive reading, whatever its literary merits. Put it aside and come back in three months. It will be a different book.
| Density | What it is good for |
|---|---|
| under 1 in 100 | Fluency and speed. Very few new words, but the known ones become automatic. |
| about 1 in 50 | The sweet spot. Comprehension holds, guessing from context usually works, harvest is steady. |
| about 1 in 20 | Intensive reading only. Short passages, dictionary open, expect it to be slow work. |
| worse than that | Decoding. Enjoyable for some people; not an efficient way to acquire vocabulary. |
Read narrowly
The most effective adjustment I made was also the least intuitive: read several things on the same subject, or several books by the same author, back to back rather than ranging widely.
Vocabulary clusters by topic and by writer. Stay inside a cluster and the same unfamiliar words keep reappearing, which supplies exactly the repeated encounters that a single text cannot. The first article on a new subject is hard; the fourth is comfortable, and the words that carried you there arrived free. Ranging across unrelated subjects means starting that climb again each time, and meeting most new words exactly once.
Five articles about shipping will teach you more usable vocabulary than five articles about five different things, and take less effort to get through.
Harvesting without ruining the reading
Stopping to look up every unknown word destroys comprehension, which is the thing that makes the reading valuable in the first place. Two passes solve this.
First pass — read
Do not look anything up. Mark unknown words with a pencil line or a highlight and keep going. Guess from context as you pass; the guess is useful even when it is wrong, because checking it later produces a much stronger correction than a first look-up would have.
Second pass — harvest
Afterwards — end of the chapter, end of the day — go back to the marks. Most of them will no longer need looking up; the meaning arrived on its own from later context. That self-resolution rate is a good signal, incidentally: if almost nothing resolves, the text was too hard.
Of what remains, only a fraction is worth a card. My filter is roughly:
- Add it if the word blocked the sentence, or if I have now seen it twice.
- Note it if it was interesting but the sentence survived without it.
- Skip it if it is a proper noun, a technical term I will never need, or a word so rare that its own author probably reached for a dictionary.
Copy the whole sentence, not the word. It takes four seconds and it is the difference between a card that works and a card that becomes a leech — see building a word list.
Listening counts, with one caveat
Audio supplies encounters as readily as text and fixes the pronunciation half of word form, which reading leaves undone and which quietly limits how much of your vocabulary is available in conversation. The caveat is that you cannot mark a word you cannot spell. In practice this means listening to things that have transcripts, or accepting that listening consolidates words already in the deck rather than adding new ones. Both are worth having; they are not the same activity.