Do I Need to Learn Words? Vocabulary and Incidental Acquisition

There is a moment, a few weeks into reading seriously in a foreign language, when the question becomes hard to avoid. You have finished four stories. You have met, on any reasonable estimate, several hundred words you did not know before. And you could not, if somebody asked you now, write down thirty of them. Either something is happening that you cannot see, or you have spent a month enjoying yourself and the hours would have been better given to a list.

Both halves of that suspicion contain something true. Reading does build vocabulary, and it does it in a way that is genuinely invisible from the inside. It also has a shape — things it does quickly, things it does slowly, and things it never does at all. Knowing that shape is the difference between trusting the process and trusting it blindly.

What it means to know a word

The question “do you know this word?” sounds like it has a yes or no answer, and almost never does. Paul Nation’s standard breakdown splits word knowledge into three groups: form — how it sounds, how it is written, what parts it is built from; meaning — what concept it points to, what it brings with it, what it is associated with; and use — what grammar it takes, what it goes with, where it belongs and where it does not.

These arrive at different speeds, from different sources, and mostly not together. You can recognise a word instantly on the page and have no idea how it is pronounced. You can know exactly what it means and be unable to produce it in a sentence. The cleanest division to hold on to is between receptive knowledge — you meet it and understand — and productive knowledge — you need it and it arrives. Reading builds the first generously and the second hardly at all.

So the honest version of the opening question is not “am I learning words?” but “which parts of which words is this giving me?”

The size of a single encounter

The mechanism has a name — incidental vocabulary acquisition — and its central finding is that it is real, and small. Reading a word once, in a sentence you understood, while thinking about the story rather than the word, moves you a short distance towards knowing it. Estimates of how short vary with the study and the test, but the order of magnitude is consistent: something in the region of five to ten per cent of the words you meet once are retained in any measurable way, and what is retained is usually partial — a sense of the general area, a familiarity with the shape, not a definition.

Studies that gave learners a whole graded reader and tested them afterwards found gains of roughly a fifth of the unknown words in the book, with recognition of the written form holding up much better than recall of the meaning, and a good deal of that leaking away over the following months. This is the result that makes incidental acquisition sound like a poor investment when it is quoted on its own.

It is not a poor investment, because the unit is wrong. Nobody meets a word once.

How many encounters it takes

The usual working figure is that a word needs somewhere around eight to ten meetings in comprehensible context before it is reliably recognised, and rather more — twenty is often suggested — before it is secure enough to survive a break and turn up when you need it. The exact number depends on the word: concrete nouns in a scene that shows them are cheap, abstract nouns are dear.

Two details matter more than the number. The first is that the encounters must be spread. Eight occurrences on one page teach you about as much as two, because after the second the word is being read out of short-term memory rather than retrieved. Eight occurrences across four weeks are eight occurrences. The second is that each meeting has to be comprehensible: a word met in a sentence you could not parse is not an encounter, it is a shape you saw.

The arithmetic this produces

Put the two figures together — a few per cent per encounter, eight to ten encounters — and you get the argument for volume, in a form precise enough to plan with.

Vocabulary in any language is distributed brutally unevenly. The commonest few hundred words make up half of any page; by the second thousand you are covering most of a general text; and after that each additional thousand words buys less than the one before. Turn that round and it describes how often you meet things. A word among the commonest two thousand will appear several times in a single story, and you will acquire it whether you intend to or not. A word around the fifth thousand turns up perhaps once in ten thousand running words, so ten encounters need something like a hundred thousand words of text — four or five short volumes, or a couple of months of a half-hour habit. That is entirely achievable, and it is why mid-frequency vocabulary is the part of the language that reading is best at.

Past the eighth thousand the same sum turns unfriendly. A word you meet once in fifty thousand words needs half a million words of reading to be met ten times, and there are tens of thousands of such words. This is not a flaw in the method. It is what the tail of a vocabulary is: rare words are rare, and no approach makes them frequent. It does mean that the last stretch of a large vocabulary is built over years of reading rather than months, by anyone, by any means.

What reading teaches better than anything else

Set against that, incidental acquisition delivers several things that no deliberate method delivers at all.

• Which words go with which. Collocation is not derivable from meaning and cannot be studied at scale. It is learned by having met the combination.
• Register and connotation. Whether a word belongs in a letter, a court or a kitchen is carried by the company it keeps, and the company is only visible in real text.
• The other meanings. Frequent words are polysemous; the more frequent, the more meanings. Reading meets them one at a time, in contexts that distinguish them, which is the only sane way to acquire a word with nine senses.
• The words you personally need. Your reading is a frequency list built from the language you actually consume, weighted exactly as your own future use will weight it.
• Fluent access. A word retrieved in a hundred sentences is available at reading speed. A word retrieved in a hundred drills is available when you are thinking about vocabulary.

What it teaches slowly, or never

The blind spots are specific enough to be worth naming, because they are exactly where a small amount of deliberate work pays.

• Words the context does not explain. Real prose is not written to teach: authors are under no obligation to make a word inferable, and often the sentence gives you only that something happened. Abstract nouns, the connectives that organise an argument, and near-synonyms sitting where either would fit are the standard casualties.
• Wrong guesses. An inference made once tends to persist, and nothing in silent reading contradicts it. A private lexicon that is approximately right in eleven places is a real cost of the method, and the only cure is to check occasionally.
• Formal detail. Gender, stress, the aspect a verb belongs to, an irregular plural, the case or preposition a verb governs — all of it is physically present on the page and almost none of it is salient while you are following a plot. This is the largest gap, and in a heavily inflected language like Russian it is the difference between recognising a word and being able to use it.
• Pronunciation. Words met only in print acquire an invented sound that is often wrong and surprisingly durable.
• Production. Recognition is cheap and recall is expensive, and reading trains only the cheap one. A reader with a large receptive vocabulary who cannot produce a third of it is not failing; that is the expected result.

So: do I need to learn words?

Yes, but far less than the question implies, and not the words most people choose.

Deliberate vocabulary work earns its place when it is aimed at the list above and nowhere else: the word that has resisted three encounters, the word carrying the plot of the book in your hand, the formal detail the page does not make salient. Everything else — the ordinary run of unfamiliar words in a text you can follow — is being handled, slowly and invisibly, by the reading itself, and taking it out of the reading to study it separately trades a good process for a worse one.

There are two obvious follow-up questions, and each has an article of its own. One is whether it is worth attacking the first two thousand words directly instead of waiting for them: the case for and against frequency lists. The other is what the deliberate part should actually look like in practice, which is a question about spaced repetition and how to keep it small.

How many of the words on a page you need to know before any of this works at all is a separate and surprisingly exact question: the 50/70/90/98 problem. And the fastest way to multiply encounters without buying another book is to read the same one again.