Knowledge and comprehension

Why do two children who decode words equally well understand the same passage so differently?

Sit two nine-year-olds side by side and have them read the same page aloud. Both get every word right, at the same pace, with the same expression. Then ask what happened in it. One retells the passage; the other manages a sentence and stops. If the machinery of turning print into speech is equally good in both, what exactly is the second child missing — and why does the school so often answer "comprehension skills", as though comprehension were a separate faculty you could practise the way you practise scales?

Begin with what a sentence actually delivers. Take: he took his glove and headed for the mound. Nothing in that sentence says a game is starting, that the glove is not for warmth, or that a mound is a raised patch of dirt in a particular sport. A reader who understands it has supplied all three, silently, from somewhere other than the page. So the first observation is that text is radically underspecified. Writers omit whatever they expect the reader already holds, because a text that specified everything would be unreadable.

If that is right, comprehension cannot be extraction. It has to be construction: the reader builds a model of the situation and the words are the smaller half of the input. Which raises the useful question — where does the other half come from? It cannot come from the passage, and it cannot be conjured by effort. It has to already be in memory before the eyes reach the sentence.

Now put that to a test we can imagine failing. If knowledge is doing the work, then a weak reader who happens to know the topic should out-comprehend a strong reader who does not. That prediction is uncomfortable — it says reading ability can be beaten by knowing about baseball. Recht and Leslie ran essentially that experiment in 1988, crossing reading ability against baseball knowledge in middle-school children and giving all of them a passage describing a half-inning. Knowledge won. Poor readers who knew the game reconstructed the passage better than good readers who did not. Notice what the design rules out: the poor readers did not become better decoders for the length of one page.

There is a cleaner version still, where the only variable is knowledge availability. Bransford and Johnson in 1972 gave adults a paragraph about sorting things into groups, being careful not to overdo any one batch, and the tedium of repetition. Read cold it is nearly meaningless. Told beforehand that it describes washing clothes, readers rated it comprehensible and recalled roughly twice as much — same words, same reader, same eyes. The knowledge was not new; it was merely made retrievable in time.

So what is the mechanism, precisely? Three things stack. First, inference: every gap the writer left is bridged from memory, and a bridge you cannot build leaves the situation model with a hole in it. Second, working memory: material you recognise is held as one chunk rather than several, so a knowledgeable reader spends capacity on the argument while an unknowledgeable one spends it on holding the nouns. Third, vocabulary itself, which is largely acquired from reading — meaning that today's knowledge sets tomorrow's rate of acquiring more.

Why, then, does the school reach for "comprehension skills"? Because some of them are real. Summarising, questioning, monitoring whether you have understood — these help. But look at their profile: they are learnable in a handful of sessions, and further practice adds little. That is the signature of a trick being acquired, not a capacity being built. It is a poor explanation for a difference that widens over years.

Let me try to falsify my own account. If comprehension were purely knowledge, general comprehension scores should not be stable across topics — and they are quite stable. But that is what one would expect anyway: a child who knows a lot about one thing usually knows a lot about many, so a general test measures a general stock of knowledge rather than a general skill. The uncomfortable case is the child with narrow deep expertise, who should score badly on a broad test while comprehending brilliantly on their topic — and that is roughly what the baseball result shows.

Two refinements connect the steps. First, this does not demote decoding. The standard framing — reading as decoding multiplied by language comprehension — is a product, so either factor near zero drives the outcome near zero. Our two children have made decoding non-limiting; that is exactly why the second factor becomes visible.

Second, "knowledge" here means specific, topic-relevant knowledge, not general worldliness or intelligence. The claim is about a passage, not about a person.

An honest caveat: the direction of causation is entangled. Knowledge aids comprehension, and comprehension is how most knowledge arrives, so the two grow together and no single study cleanly separates them. Experiments that manipulate knowledge and measure comprehension immediately, as Bransford and Johnson's did, are the strongest evidence that the arrow runs the way I have drawn it, but they show the effect over minutes, not years.

• Why vocabulary breadth and knowledge breadth are so hard to tell apart in any measure of comprehension.
• What a curriculum looks like if knowledge, not strategy, is the thing being sequenced.

Every term the collection defines is gathered in the glossary.