What English sounds like in an Indian home.

There is a moment in a lot of Indian homes that you have to be looking for to see.

A six-year-old has been given a storybook in English by her parents — a Roald Dahl, a Julia Donaldson, maybe an old hand-me-down with a sticker inside saying it belonged to her older cousin. She is reading it aloud, the way her teacher has asked her to. The book is sitting in her lap. The TV is muted. Her grandmother is in the next room, keeping the kind of half-attention all Indian grandmothers seem to have for the language their grandchild is being asked to read.

The child gets to a word she does not know.

She pauses. She tries it. She tries it differently. The room waits, in the small uncomfortable way rooms do when a child is reading aloud and gets stuck. Eventually someone — her mother, her father, the grandmother in the next room — offers a pronunciation. It is usually almost right. The child copies it. She keeps reading.

What just happened in that pause is the texture of childhood for a very large number of children. It is also the thing that nearly all consumer AI is, at this point, badly equipped to help with.

Three languages in the air.

Most Indian children grow up between languages. The exact mix is different in every home, but the structure is almost always the same: one language at home (sometimes two, if the grandparents speak a different mother tongue than the parents), one language at school, and increasingly a third language that the child learns from screens.

English is the school language in most middle-class Indian homes. It is the language her textbooks are in, the language her tests will eventually be in, the language her parents have decided will be most useful to her in the country they have raised her into. It is also, almost always, a second or third language for everyone in the room.

This means that when a six-year-old in Pune or Hyderabad or Lucknow is reading aloud in English, what you are hearing is not what an American child reading aloud in English sounds like. It is a phonology rearranging itself in real time. The child is asking her mouth to do something her family language did not require it to do. She is reaching for sounds that her father, who is fully proficient in English in a board meeting, still cannot quite catch correctly from a children's book at bedtime.

She is doing genuinely hard cognitive work. And she is being graded on it.

What goes wrong.

There is no shortage of children's English-learning apps in the world. The number that work well for a child whose first language is Tamil is roughly zero.

The standard reading app was trained on American children. Its pronunciation reference is American. Its word-frequency assumptions are American. When the Indian six-year-old says "twenty" with the gentle t of her family language, or hesitates on the th in "thirteen" because there is no equivalent sound in Marathi, the software marks her wrong.

It is not wrong. It is the very predictable cognitive accommodation of a child who is learning a language her parents are also still learning. But the software does not know that, because the software was not built for her.

The result, over the course of a hundred sessions, is that a child whose English is actually progressing nicely starts to be told — by the small device in her hand — that she is doing it wrong. Slowly, without anyone really noticing, she becomes less willing to read aloud. Her parents conclude that the app isn't working. They install another one. The new app does the same thing.

The child is being failed by a category of product that was never asked to think about her.

What we try to do instead.

ANA's first product, ZigZu, is a reading coach built specifically for Indian children learning to read English. It listens for the kinds of pronunciations a child whose home language is Tamil or Hindi or Gujarati or Bengali would naturally produce, and it does not flag them as mistakes. It pauses where the child pauses. It encourages where the child hesitates. It is patient about the things that deserve patience.

This sounds obvious when you read it as a sentence. It is, in practice, the difference between a child gaining confidence and a child quietly losing it.

I do not want to make this essay an argument for ZigZu specifically. The point is broader: the children we are designing for already do an enormous amount of cognitive work that the AI industry, at the moment, does not see clearly. Software that sees them clearly is not a feature. It is a starting condition. It is the very first thing a product needs to get right before it is allowed to do anything else.

A small moment, at the end.

Some evenings, when I am thinking about this work, I come back to that small scene I started with — the child, the storybook, the grandmother in the next room. There is something quietly extraordinary about a six-year-old being asked to read aloud in a language nobody in her house grew up reading. She is doing something her own parents could not do at her age. She is, in a real sense, a generation ahead of them, just by sitting on that sofa with a book in her lap.

The software she is reaching for does not get to fail her.

That is the thing we are trying to keep in mind.