A student writes "the paper discussed about three approaches." Their supervisor reads it and understands it completely. So would everyone else in the seminar. So do we.
Our app underlines it.
It is worth being exact about what that underline achieved. It did not make the meaning clearer, because the meaning was never in doubt. It did not repair a failure of communication, because no communication failed. What it did was move the sentence closer to a convention that British academic writing happens to use. The word for that is conformity. We would rather use the word than dress it in something friendlier, because everyone who reads this page can see what happened: a person was understood perfectly well, and the software told them to write it differently anyway.
We think we should carry on doing it.
What the student cannot get out of
The reason has nothing to do with clarity, and pretending otherwise is where most language products start lying. The reason is that the convention has consequences the student cannot get out of. They are going to be read by academics who expect British academic English in a British department, and our job is to get them ready for that. We do not want to leave a student unprepared for the things that will get marked anyway. Someone who has learned Indian academic English comes here and has to learn English academic English. That is kind of just the way it is, and saying so describes an arrangement without approving of it.
One honest limit on that reasoning. We do not know what "discussed about" costs in marks. We have not measured it, we have not seen a study that measures it, and we would be inventing a number if we gave one. What we have is the ordinary expectation that a marker who notices it will treat it as an error, which is enough to make the feature worth teaching and not enough to make the penalty defensible. Those are separate questions and the product only answers the first.
So both things are true at once. Correcting a feature that did not impede understanding is about conformity rather than communication. Teaching the dominant convention is still a service, because the convention is enforced by people the student cannot avoid. We have seen enough language marketing to know that the usual move is to hold only the second and hope nobody asks about the first.
Same edit, different story
What follows from holding both is a question about how the correction is presented, not about whether to make it. There are two stories the same underline can tell. The first says the student's sentence was wrong, or unclear, or in some way worse than the alternative. That story is false, and a student who grew up writing "discussed about" among people who all understood each other may already know it is false, which costs the product something every time it repeats it.
The second story says the form is standard where the student learned it and that British academic writing does it differently. The person marking the work will expect the British form. Both stories produce exactly the same edit. Only one of them is true.
Saying it is a convention while still teaching it
What makes both halves holdable at once is a distinction of Michael Young's (2008). Knowledge of the powerful is whatever elites happen to hold, and cataloguing it is a sociological exercise. Powerful knowledge is knowledge that lets somebody do things they otherwise could not, including argue back, and withholding it liberates nobody. Wheelahan (2010) carried the same argument into vocational education: epistemic access is the precondition for social access, and a curriculum that criticises the dominant form without giving students command of it leaves them standing outside it. We are not in a position to adjudicate how that argument runs inside applied linguistics, and neither of them was writing about "discussed about". What we take is the distinction, and the distinction is what this feature rests on.
Bisong (1995), writing from Nigeria against Phillipson, made the neighbouring point that people who choose English are choosing it from inside a repertoire that already holds other languages, and that reading the choice as false consciousness is its own kind of condescension. A student who wants the convention that will be marked is not confused about what they want.
It also matters that the convention is imposed on people who were born here. No-one's first language is academic English. A British undergraduate from a house with no books has to learn this register too, and Lea and Street (1998), studying writing in two British universities, found what counted as good writing shifting between departments and between individual markers, which is not the shape of a single skill a student arrives already holding. That does not make the convention neutral. It makes it a convention, which is the thing we want to be able to say out loud while still teaching it.
The correction we have not built
That distinction has a consequence for the product. We have not acted on it.
Our correction currently says that something is wrong. It could instead say what kind of thing it is: a form that is standard in Indian English and different in British academic writing, with the British form expected by whoever marks the work. The edit is identical and the information is greater. Nothing in that sentence claims the student's English was worse, only that it belongs somewhere else, which is what it means to treat somebody as holding a repertoire being extended instead of carrying a deficit.
We have not built this. It is not in the app today, it is not scheduled, and if this page is read as a description of a shipped feature then we have written it badly. It is what we think the correction should say.
The standard is fixed, the explanation language is chosen
What we have built is smaller and worth stating accurately. The standard the app applies is British academic English, and the student does not choose it, because the product exists to prepare people for UK academic assessment and a standard that moved with a menu setting would not do that. What the student does choose is the language the explanation arrives in. A correction can be explained in their own first language, or in plainer English, or in academic English, and none of those choices touches the standard being applied to the writing.
Two things follow, and neither is a breakthrough. The first is that the standard is named. A corrector with no stated standard is still applying one, and what it applies is whatever its training data mostly held, which for English means American convention arriving by default and never announced. Naming British academic English, and saying why, is a claim about honesty rather than about quality.
The second is that separating the standard from the explanation language is itself a position. It says the thing being taught and the language it is taught in do not have to be the same, which is a refusal of the monolingual assumption in the design rather than in the marketing. A settings menu is not a theory of language and we are not going to present it as one. We have not audited the market, so the strongest thing we will say about it is that we have not yet seen a competitor draw the distinction.
We have no line between an error and a difference
Now the part that costs us something.
We do not have a line between an error and a difference. The best we have managed is that the app is just doing English English. That is an accurate description of the build and it is not an answer to the question. We have also tried a version where accent is a difference and word choice is an error, on the grounds that an accent everyone understands is fine and the wrong words are the wrong words. That version does not survive contact with the literature.
"Discussed about" is not an individual slip. It is systematic in its variety, codified, taught, and used by people writing carefully, which is the ground Kachru's World Englishes work stands on: Indian, Nigerian and Singaporean English treated as codified varieties and not as defective attempts at British English. The split between accent and words puts a systematic feature on the error side of a line it was supposed to sit on the other side of.
The only line we can draw is assessment
The nearest thing we have to a working criterion is whether the feature is assessed. Accent is not assessed in written work and word choice is, so the app corrects one and leaves the other. That is consistent with everything else we do, and it makes the rule one about British assessment practice and not about language. It borrows its authority entirely from the marker.
And to label a feature honestly, we would need a description of the variety we are labelling. Saying "standard in Indian English" is a factual claim about Indian English, and getting it wrong in public would be worse than the current silence. We do not hold a reference for those varieties, and we have not costed the work of acquiring one. So the honest statement of where we are is that the app applies one standard, treats everything outside that standard as an error, and that we have described a better design without having earned the right to say it exists.
The judgement is made in the listening
The argument we have no answer to is Flores and Rosa (2015). Their claim is that racialised speakers are heard as deficient however closely they conform, because the judgement is made in the listening and not in the speaking. If that is right, then a product that teaches a student to produce the expected convention has not delivered what it implies, and it has relocated the responsibility for somebody else's prejudice onto the person experiencing it. We think the claim is substantially true.
We are not going to build for the listener, because it is not a language app's job to change how markers hear students, and because we would be bad at it. The work that gets nearest to the listener happens inside departments. Wingate (2015) argues that academic literacy should be taught within the disciplines, by subject lecturers working alongside language specialists, rather than handled elsewhere as a generic add-on, and that model at least puts the expectations in front of the people who set and mark the work. We do not know whether it touches the listening problem or is aimed at something else. It reaches nearer than a language app can. That is a division of labour, not a refutation.
Which leaves an open question we cannot settle from here. We do not know whether telling a student that a convention is a convention changes anything for the student. They still have to make the edit. The marker still expects what the marker expects. It is possible that the honest label is mostly a comfort to us, a way of feeling better about running a conformity engine, and that the person on the other end experiences it as the same underline with a longer sentence attached. We would want to know whether students who are told what kind of thing they are being corrected on write differently afterwards, or notice at all. We have not built the feature, so we have not been able to ask them.
