WordsThatStay.

Many people have a language-learning streak of several hundred days and still cannot order a coffee. I find this one of the sadder achievements of modern software.

They did what they were asked to do. They came back every day. They collected the points and kept the flame alive. What they learned, in a very thorough way, was how to play that application. The product was optimised for the return visit, and it succeeded. Speaking a language was a separate matter.

I do not want to be unkind about this. Keeping people motivated is hard, and games are one honest way to do it. But I think a mistake has been made at the level of purpose. A word learned for points is held differently from a word learned because you need it. The mind is quite good at telling the difference.

You can keep a streak for years, and all you will have learned is how to keep a streak.

VAJ begins from the opposite position. Its name reaches back to an old root in the languages of my region, close to the idea of a word and a voice. Its full name is VAJ by oOh!, the atelier edition, and its philosophy is written into its first line. It wants to turn language learning from screen fatigue and gamified distraction into something closer to archival stationery: cold-pressed paper, good ink, and a serious typeface.

That sounds like an aesthetic preference, and it is one. But there is a deeper claim inside it, and it is the subject of this essay. The way an instrument feels teaches you how to treat what it holds. If words arrive to you with confetti, they are trivial. If they arrive on a quiet page, set with care, they are worth something.

The working application is built on three ideas: that memory has a mathematics, that a word lives in the voice as well as the eye, and that calm is a method. I will take them in turn.

Forgetting is not random. It has been studied for more than a century, and it follows a curve. A new memory fades quickly. If you recall it just as it begins to slip, the memory returns stronger, and it fades more slowly the next time. This is one of the most reliable findings in the psychology of learning. The practical consequence is simple: the best moment to review something is shortly before you would have forgotten it.

Most people study at the wrong times. They cram, reviewing what they already know and neglecting what is slipping. A good system does the timing for you.

VAJ uses FSRS, a modern open scheduling algorithm that models how stable each memory is and how difficult each word is for you personally. It decides when each word should come back. The gaps start short and grow, from hours to days to months. A word you know well may not return for a long time. A word that keeps escaping you comes back soon.

The most beautiful detail in the product follows from this. In the reading view, words fade upon mastery. As a word becomes yours, it literally recedes from the page. Most learning products add things as you progress: badges, levels, trophies. VAJ takes something away. Your progress is the disappearance of your need for help.

Most apps reward you by adding. VAJ rewards you by taking something away.

It has one lovely way of showing the whole picture. The core lexicon, the Oxford 5000, is laid out as a star map. Each word is a star, and it glows according to how stable it has become in your memory. Over the months a dark sky fills with light. It is a far better image of learning than a progress bar. It suggests that your vocabulary is a territory, and that you are slowly illuminating it.

Words also arrive in sentences, never alone. A word in a list is only a label. A word in a sentence has neighbours, rhythm and a job to do. Context is how a child learns, and it remains the way a mind holds meaning. A long press on any word opens a small inspector with its meaning in context, its translation, its sound, its register, its common companions and its history. Etymology is included on purpose. When you learn where a word came from, you have one more hook to hang it on, and the language starts to feel like a place with a past.

Here is the failure that gamified apps rarely admit. You can recognise ten thousand words on a screen and still be unable to say a single sentence with confidence. Recognition and production are different skills. A word is not really yours until it can leave your mouth, at the speed of conversation, with the right music.

This is why VAJ takes speech so seriously. The technique at its centre is shadowing, a method used by interpreters: you listen to a native sentence and speak along with it, a fraction behind, imitating not only the sounds but the rise and fall. In VAJ, your voice appears as a live waveform, and the application tracks the pitch of your speech and shows it against the model.

That matters because of something most learners are never told. Every language has a melody. English moves up and down in places that Persian and Kurdish do not. You can pronounce every sound correctly and still sound foreign, or even be misunderstood, because your sentence has the wrong tune. Seeing that tune as a line changes what you can notice, and you cannot correct what you cannot perceive.

Every language has a melody. You can say every sound correctly and still sing the wrong song.

I work in Kurdish, Persian, Arabic and English, and in my region most people learn languages the same way: out of necessity. In a place like Kurdistan, a language is not a hobby. It is a door to a scholarship, a job, a border crossing or a conversation with someone who can help you. So I have little patience for a tool that treats it as entertainment. People are often learning under real pressure, and they deserve an instrument that takes them seriously.

Then comes the last step in the sequence: recall. The word is hidden, and you have to produce it from nothing. It is a little uncomfortable, and it is meant to be. Psychologists call this retrieval practice, and it is among the strongest known ways of building memory. Rereading feels productive. Being forced to retrieve is what actually works. Good learning is slightly difficult, and any app that removes all difficulty has probably removed the learning too.

Now to the third idea, which most engineers would treat as decoration.

The interface of VAJ is paper and ink. The ground is a warm archival off-white. The text is carbon black, and it is set in a serif at a generous size, with plenty of space between the lines. Motion is slow and intentional. A faint grain sits over everything. Selecting text makes the ink seem to spread into the paper. Red appears once, as a small wax seal.

Why bother? Because attention is not a fixed quantity. An environment shapes it. A loud, flashing, rewarding screen trains the brain to expect stimulation and to leave when it stops. A quiet page does the opposite. It tells you that nothing here will grab you, and that you may settle. Deep learning needs that state, and it is getting rarer. Phones are designed against it.

Look at the matrix. The serious drill has a strong method and nobody opens it. The game is opened all the time, and its method is weak. The nice notebook is beautiful and empty. The space worth occupying is the top right: real science, on a page that a person actually wants to return to. I would put it as a rule that applies far beyond language. Rigour needs beauty to get used, and beauty needs rigour to be worth using.

Rigour needs beauty to get used. Beauty needs rigour to be worth using.

This also explains the monogram, a veiled V drawn with fine hairlines at a precise angle. It explains the typeface choices and the strict colour rules. A person who has spent time on such details is making an implicit promise: that the same care went into the parts you cannot see. A luxury watch does not tell time better than a cheap one. But it tells you that someone cared, and you treat it differently. I want people to treat their own learning that way.

There is a real risk in this approach, and I admit it. Calm does not go viral. Nobody shares a screenshot of a quiet page. The market's most successful products got there through streaks and guilt, and they have enormous distribution. VAJ is unlikely to beat them at that game, and it should not try. It is built for a smaller and more serious group of people: those who have already found out that the game did not work.

There is an awkward question that any language product must now answer. Real-time translation is improving quickly. Earbuds can already interpret a conversation. Why spend years learning words when a machine will do it for you?

I think the answer has several layers, and the first is practical. Translation works for transactions. It will get you a hotel room. It will not get you a friend, a joke, a negotiation or a grandmother's trust. Nobody has ever fallen in love through a subtitle. A language is the way into how a people thinks, and that part does not transfer through a device.

The second layer is about the mind. Learning a language is one of the hardest and most rewarding things a brain can do. It changes what you are able to notice. Someone who has learned a second language has learned that their first one was not reality, only one way of cutting it up. I cannot think of a more useful thing to know.

The third is about dignity. To speak to someone in their own language, even badly, is a form of respect that no machine can perform on your behalf. I have seen faces change when a foreigner tries a few words of Kurdish. For the people on the receiving end of history, that effort is not a small matter.

Nobody has ever fallen in love through a subtitle.

AI is inside VAJ too, and I see no contradiction. It powers the inspector, which explains a word in its context, and it helps generate material. A machine is a superb tutor for the parts that can be explained. It cannot do the part where a word enters your memory. That can only happen in you, through repeated effort, and no one can do it for you. In a world of instant answers, that may be the last honest kind of work: changing what is inside your own head.

The app works today, offline-first, with your words kept in a local vault. I should say that it has not launched publicly, and that I can claim nothing yet about how well it teaches anyone at scale. It rests on well-established science, but that is not proof that the product itself works. Only its users will decide that.

In the end I return to the name, and to what it points at: a word, a voice, the thing that comes out of a person. I want the words that a learner gains to do what the tagline promises. Not points. Not a streak. Words that stay.