Decoding is not comprehension

A dyslexic reader listens to a chapter and understands it. A non-dyslexic reader reads the same chapter and understands it. Both arrive at the same place, holding the same argument, ready to disagree with it in the same ways. Whatever went wrong in between was not thinking. It sat in a specific, narrow, well-described place, and being precise about where changes everything that follows.

Dyslexia is a specific learning difficulty affecting accurate and fluent word recognition, decoding and spelling. That is the substance of it, and the omissions are the point. It says nothing about intelligence. It says nothing about vocabulary, about reasoning, about the ability to follow a complicated argument or to spot the hole in it. Prevalence estimates commonly quoted run somewhere around five to ten per cent of the population, with wide variation depending on where the definition is drawn and which language it is drawn in.

The consequence is a mismatch that dyslexic readers spend a great deal of their lives explaining to other people. Understanding of spoken language is typically intact, and often more than intact. Put the chapter through the ears and it lands. Put it through the page and something expensive happens between the letters and the meaning — and by the time the meaning arrives, it has been paid for twice: once in time, and once in the attention that was supposed to be holding the sentence together.

Reader Inc. is not a treatment, a therapy, an intervention, a diagnosis or a remediation, and it will not be described as one here or anywhere else. It is a way of putting words on a screen. Several of the things it does happen to sit on the seam this article is about — not because we set out to build assistive software, but because removing the decoding bottleneck was the original design problem. That coincidence deserves an honest accounting, mechanism by mechanism, with the gaps left visible.

The phonological account

The dominant research account puts the difficulty at the level of the sound structure of language rather than at the eye. Reading an alphabetic script requires you to hold the sounds of spoken words apart from one another, map letters and letter groups onto them, and assemble the result quickly enough that a word is recognised rather than constructed. Where that phonological machinery is less precise, decoding stays effortful long after it has become automatic for everyone else in the room. Dominant is not unanimous — the literature argues productively about visual attention and about naming speed — but this is the account most of the evidence stands behind.

It also explains why the folk image is wrong in a way that matters. Dyslexia is not letters rearranging themselves on the page, and treating it as a problem of the eye sends people towards the wrong tools with great confidence. If the difficulty lives at the level of sound and the mapping onto sound, then changing how text is displayed is not touching the difficulty at all. It is changing the conditions the difficulty operates under. Those are different things, and this article will keep them apart even when it would be more flattering not to.

English is a hard language to have this in

Orthographic depth is the technical name for how far a writing system strays from one letter, one sound. English strays a long way. Through, though, thorough, tough and thought share five letters and agree on almost nothing. A reader assembling words from their parts is working against a system that keeps changing the rules mid-word, and the assembly stays slow because the shortcut — a reliable mapping you could automate — is not fully there to be learnt.

Transparent orthographies behave differently. In Italian, Finnish or Spanish, where letters and sounds line up closely, the same underlying difficulty tends to show a different face: accuracy is less conspicuously affected, and what remains is slowness and effort. The difficulty did not go anywhere. The writing system stopped amplifying it. That comparison is one of the clearer results in the field, and it is a useful corrective for anyone inclined to treat reading difficulty as a fixed property of a person rather than of a person meeting a particular script under particular conditions.

The working-memory bill

Here is the mechanism that matters most, and the one most often skipped. Reading a sentence is not a sequence of independent word recognitions. You have to hold the beginning of the clause while the end of it arrives, keep a pronoun open until it resolves, carry a negation across the thing it negates. That holding costs capacity, and capacity is finite. If decoding is automatic it costs almost nothing and the whole budget goes to the sentence. If decoding is effortful it takes its share first, and the sentence gets the remainder.

So comprehension suffers — downstream of a problem that was never about comprehension. A reader who could follow the argument perfectly well if someone read it aloud finishes the paragraph with the words intact and the structure gone. This is why fluency, not just accuracy, is the thing specialists watch. And it is why the experience is so corrosive to confidence: the failure presents itself as not understanding, which looks from the inside exactly like not being clever enough, when the actual event was a budget overrun three levels below.

The words were decoded. There was simply nothing left over to hold the sentence together with.

The mechanism that matters most

Mechanism by mechanism, with the gaps left in

What follows is not a list of benefits. It is a list of things that demonstrably change when text is presented one word at a time at a fixed point, paired with what can and cannot be said about each. Some of these are mechanical facts about the display and are not in dispute. Some are reasonable inferences from good evidence. At least one is our own untested reasoning, and it is labelled as such, because a table that hides the difference is worse than no table.

One setting deserves separating out, because it is the one people misread. Pace here is a floor, not a ceiling. Nothing about this app requires speed, and 120–180 wpm — comfortably below an ordinary adult rate of roughly 200 to 300 — is a completely legitimate configuration that plenty of readers use deliberately. They are not there for the words per minute. They are there for the fixed point, the single word and the absence of a page. Running it slow is not a failure mode or a beginner setting. For some readers it is the entire reason the thing is useful.

The things that actually change outcomes

Structured literacy instruction — explicit, systematic, cumulative teaching of the sound-to-letter system, delivered by someone trained to do it — is the thing with the evidence behind it. Speech and language therapy is the thing with the evidence behind it. Specialist assessment, which gets a reader the right description of their own difficulty and the accommodations that follow from it, is the thing with the evidence behind it. These are not the polite paragraph before the sales pitch. They are the actual answer to the actual question, and anyone choosing between them and a browser tab should choose them without hesitating.

A presentation setting belongs somewhere much smaller and much later: alongside larger type, generous spacing, an audiobook, a reader pen, a quiet room. Those things do not teach anyone to decode. They change what a given amount of decoding ability is able to get through in an afternoon, which is a real and underrated kind of usefulness, and it is entirely compatible with also getting proper instruction. The mistake is only ever in the labelling — a tool that changes conditions being sold as a tool that changes the reader.

Ten minutes, on your own material

Individual variation is not a hedge at the bottom of this article. It is the finding. Some readers try single-word presentation and describe it as the first time text has held still for them. Others find it worse than a page — the vanishing word removes the ability to look again, and looking again is how a lot of readers repair a sentence that came apart. Both of those are ordinary results. Neither tells you anything about the other person, and no average across a group tells you anything much about either.

Which leaves one recommendation. Take something real — the coursework, the contract, the chapter you have started four times — and run it through the app at a slow setting with the voice on, then answer the recall check. It parses on your machine and uploads nothing, so the material can be anything at all. Ten minutes is enough to know whether the fixed point does anything for you. If it does not, you have gained a true fact about your own reading, which is more than most claims here offer. If it does, the reasoning is on the science page, graded by how much we actually know.