Asked what a piece of software will eventually be able to do in language learning, we answered with a list. Video lessons. A tutor that interrupts the moment something goes wrong. Coaching on pronunciation. Coaching on a piece of writing. We said we thought most of that was buildable on something like a five year view, and that the exception would be real social situations, because those are hard to replicate. The same question came back four more times in different clothes over the following weeks, and each time we gave a version of the same answer without noticing we had given it before. So it is a position rather than a remark, and it is better written down and examined than left to surface in a meeting.
The part worth examining is what is actually on the list, because we did not look at it at the time. Every item is a delivery function. Something gets presented and something gets corrected. Taken together they describe a teacher as a channel: the thing that carries material from where it currently is to where the student is. We arrived at that picture the way anybody arrives at an assumption they have never been asked about, which is by not arriving at it at all. It is also the picture that Basil Bernstein spent a career arguing against, and we would rather concede that in writing than have it pointed out.
Somebody has to select what gets taught
The word doing the work is recontextualisation, and it is Bernstein's. Knowledge does not travel from a discipline into a classroom unchanged. Somebody selects what gets taken out of the discipline and decides how it is transformed for a particular group of students on a particular course. Bernstein's claim is that the selection is never neutral, because somebody has decided what counts and what gets left out. That somebody is the practitioner, and the judgment doing the selecting is built over years and is specific to a context.
Bernstein called the machinery of that selection the pedagogic device, and it works in stages. Knowledge is produced in one place, then recast as a curriculum somewhere else, then recast again as whatever actually happens in a room. Judgment is required at every stage, and a unit where none of that judgment is possible is easy enough to describe. Staff are directed to a generalist textbook that makes no reference to disciplinary differences and work through it and their prior training and nothing else. Where an institution stops paying for scholarship time, the people teaching stop producing pedagogic material and only consume it. That is what happens to expertise when nobody funds it.
Our list has nothing to say about any of that. It covers the carrying and is silent on the selecting. Karl Maton's name for the habit our list displays is knowledge blindness: talking about teaching and learning at length while never examining the knowledge being taught (Knowledge and Knowers, Routledge, 2014).
Modelling the easy half
If the difficulty in teaching sits where Bernstein locates it, then we have modelled the easy half and described it as the whole thing. We think there is a real chance that is exactly what we have done. What we do not have is a way of telling from the inside. We are a maths student and an engineering student. Neither of us has spent ten years working out what a particular group of students in a particular department needs next, so we are badly placed to notice the part of that work that never shows up in a feature list.
The version of this that would be convenient for us is that the selecting is a small residue on top of the delivery. We have no evidence for that and an obvious commercial reason to believe it, which is the position a reader should assume we are arguing from unless we say otherwise.
It gets more uncomfortable when the argument is pointed at the shape of the product rather than at the theory behind it. The standing objection to bought-in generic materials is that nobody working inside the local context shaped them, and software sold to several universities at once is the purest available example of that. A textbook at least gets chosen by somebody who knows the cohort.
One part of our product does start from the student's own materials, which is the vocabulary extraction, and it draws its word lists from whatever the student uploads rather than from a general academic list. That is the only part. It does not currently feed the voice tutor or the writing tutor, which are both generic. Paul Ashwin's formulation is that skills without knowledge is no skill at all, and a tutor with no access to what a student is studying is working on the skills half alone. We can build the connection and we have not built it, and until we do, the answer to "what happens when this meets a discipline" is: in one corner of the product, something, and elsewhere, nothing.
Something else follows from the point about scholarship time, and it runs against our interests rather than for them. A university that bought this and treated it as a reason to fund fewer hours of skilled work would have produced the unit described above, and would have used us to do it. That is the use we would argue against in a procurement conversation, and writing it down costs us a sales argument we could otherwise make. The product is designed to run alongside a pre-sessional programme and we have not built it to stand in for one. What it is for is the hours when nobody is available, and there are a great many of those hours.
What a student actually has at eleven at night
So it is worth being plain about the ceiling. On our own description, the best a tool like ours reaches is something close to a mediocre classroom. Not a good one. A classroom where the explanations are adequate and nothing in it was designed for the person sitting there. We do not think that is a reason to stop, and the reason has to do with availability rather than quality. A mediocre classroom that is open at eleven at night is open to somebody who cannot get into the good one, and it does not get tired of the same question asked six times in a row. The comparison that decides whether this is worth building is with what a student actually has at eleven at night, which for most of them is nothing at all.
That argument is about availability, and availability is not evidence of learning. Nobody has used UniFluent for a hundred hours. There is no efficacy data, and the sentence above is a claim about when a thing is open rather than a claim about what it produces. We would rather state it that way round than have somebody else do it for us.
What the roleplay does when a student stalls
There is one place where the product does something other than carry material, and it is worth naming, because it is the only honest answer we have to the question of what theory of learning sits underneath any of this. In the roleplay, when a student stalls, the system offers the least explicit hint that might work and escalates only as far as it has to. Poehner and Lantolf call that dynamic assessment: mediation and assessment fused, with how much help somebody needed treated as the measure of where they are.
We did not build it out of the literature. We built it because handing over the answer immediately felt like it defeated the point, and we found out afterwards that the thing we had built has a name and a tradition behind it. We are not certain whether that is reassuring or the opposite.
Where preparation stops
What we are trying to help with is a room. At some point a student has to say something out loud, in front of people whose opinion they care about, in a language they are still assembling as they go. Everything that makes that hard is in the room and not in the material. You can know the vocabulary and still not say it. Preparation is a different event from doing it, and no amount of preparation converts into having done it.
Which makes the honest description of what we sell fairly narrow. Our job is to get them ready for those social interactions. Confidence is not ours to hand over, and we do not think it is the sort of thing that can be handed over: it comes from other people, in real contact, over time, and mostly from having survived the room once already. What we are trying to do is lower what the first attempt costs and make it possible to be wrong in private beforehand. We cannot be in the room, and a tool that claimed it could would be lying about the one part of this that is easy to check.
That is a modest claim and it is the best statement of purpose we have.
The story this is built on, and what we cannot check
There is a loop in this that we did not design and only noticed while writing these essays. We were told about a seminar where a student's classmates pointed at her to answer on the group's behalf, because she was the one person in the room who had grown up speaking English. Nobody spoke. Somebody pointed. That is a group reorganising itself around a language problem, silently and by consensus, and it is the observation this product keeps returning to. The one capability we are confident no software acquires, the one thing that stayed on the exceptions list every time the question was put to us, is being a person in that room. The tool exists to prepare somebody for the only place it cannot go.
We should be careful about what that story is evidence of. It reached us from the person being pointed at. We have not heard the same seminar described by anybody who was doing the pointing, and neither of us has ever been the person in the room who could not follow. So the experience the product is built on is one we have assembled from the outside, which is a poor position from which to be confident about anything.
We are also not sure that the preparation works, and the reason we are unsure is structural rather than a matter of not having run the study yet. The roleplay takes the stakes out on purpose. Being wrong in front of a piece of software at eleven at night costs nothing, and that is the whole of the design: it is meant to be repeatable at no cost. The seminar is difficult because the stakes are real and cannot be removed.
So the question underneath our entire product is whether practice with the stakes taken out prepares anybody for a situation whose defining feature is the stakes. We do not know. We are not certain what evidence would settle it, and the people most likely to have a view are the ones who have watched cohort after cohort go through it and formed an opinion we have not heard.
