Reading acquisition

Why is learning to read so much harder than learning to speak?

A child surrounded by talk learns to talk. Nobody sequences the lessons, nobody drills the vowels, and yet by four the child has a grammar no linguist can fully write down. The same child, five years later, may still be labouring over a page — with a trained teacher, a curriculum, and hours a day of practice.

The obvious explanation is that reading is harder. But harder in what sense? Both involve the same words and the same meanings. So what exactly did the second task ask for that the first did not?

Start with the difference in age of the two skills — not in the child, but in the species. Speech is old enough that every human population has it, and children acquire it on a schedule that looks biological. Writing is an invention, first appearing in Mesopotamia around five thousand years ago, with alphabetic writing later still. Most humans who ever lived could not read. Whatever machinery reading uses, evolution cannot have built it for reading, because for almost all of our history there was nothing to read.

So the brain must be borrowing. Borrowing what? The natural candidate is the visual system, which already had the job of recognising shapes reliably despite changes in size, position and lighting. Dehaene's neuronal recycling proposal says literacy colonises part of this apparatus: a patch of left ventral occipitotemporal cortex, now usually called the visual word form area, comes to respond selectively to letter strings. Its location is strikingly consistent across readers of very different scripts — the fingerprint of a borrowed region rather than a designed one, since it sits where object recognition already was. And the borrowing has costs: scans of adults who learned to read late show that region's responses to other categories, faces among them, shifting as literacy takes hold.

That explains why reading needs building. It does not yet explain why the building is painful. For that, ask what an alphabet asks a child to believe. It asks that a spoken word is made of a small number of separable sounds, and that each of those sounds is what a letter stands for. Is that true of speech as heard?

Say "bag" and try to find the boundary between the b and the a. There is not one. The mouth begins shaping the vowel while the lips are still closed, so the acoustic signal carries the two simultaneously; look at a spectrogram and there is no silence, no seam, nothing to cut. Phonemes are not units of sound. They are units of the system that produces sound, and their separateness has to be inferred.

Which predicts something testable: if the segments are not audible, then people who have never been taught an alphabet should be poor at manipulating them. Morais and colleagues tested exactly this with Portuguese adults who had not learned to read, and found them markedly worse at tasks like deleting the first sound of a word than matched adults who had learned late in life. So phonemic awareness — the capacity to hear a word as a sequence of phonemes — is not simply a prerequisite that ripens on its own. Alphabetic literacy substantially creates it, and the two build each other.

Notice how this reframes the difficulty. The child is not failing to see the letters; the child is being asked to notice a decomposition of speech that speech itself conceals, and that they have never had any reason to notice, because for all of speaking and listening the word was the unit.

One more source of difficulty, and it is not universal. How reliably do letters map to phonemes? In Finnish, close to perfectly; in English, notoriously not, thanks to layered borrowings and spellings frozen while pronunciation moved — compare through, though, tough, thought. Seymour and colleagues compared children across more than a dozen European orthographies and found those learning shallow ones near ceiling on basic decoding by the end of the first school year, while English-speaking children took considerably longer. So a substantial part of "English children read late" is a property of English spelling, not of English children.

Two corrections. First, "reading is unnatural" does not mean reading is arbitrary or that any method will do. Because the difficulty is specifically about mapping letters onto sounds that are hard to isolate, explicit and systematic instruction in that mapping helps — the accumulated trial evidence favours it. But decoding is only one half: once a word is recognised, comprehension runs on the language ability the child already had, so a fluent decoder with thin vocabulary still understands little.

Second, the picture is less a pipeline than a loop. Phonemic awareness supports decoding, decoding sharpens phonemic awareness, and reading itself feeds vocabulary, which makes further reading easier — so early advantages compound. And there is genuine dispute at the edges: how much of dyslexia is a phonological deficit versus other causes is still argued, and the strength of neuronal recycling as an account of what the visual word form area is doing is debated rather than settled.

• How Chinese character learning differs, given that its units are not phonemes.
• Why fluent reading eventually stops feeling like decoding at all.

Every term the collection defines is gathered in the glossary.