The Speaking Index

Why this exists

Every European language has a shared scale for what a learner can do. Heritage Hindi has none. A parent in Singapore cannot say where their child stands, a weekend school cannot say what a year of Saturdays achieved, and the two cannot compare notes. The Speaking Index is our attempt at that shared scale, for the one skill families care about most: can the child speak.

We publish it as a draft. It will be revised as teachers rate recordings against it, and each revision will be numbered.

Built for the heritage profile

Every established scale, from CEFR to ACTFL, carries a quiet assumption: that a learner's skills rise together. Someone who understands at level B1 speaks at roughly B1. That is true of an adult learning a foreign language from zero. It is false for a heritage child.

A heritage child understands Nani perfectly and answers her in English. Comprehension can sit three levels above production for years. Put that child on a scale that averages the skills, and she reads as a solid intermediate who somehow cannot order dinner in Hindi. The scale has described nobody.

So the Speaking Index does two things differently.

It reports speaking. Comprehension is not assumed: it is rated separately, on its own 1 to 5 scale, after every conversation, and when it is low it is the first thing Kiki works on, with shorter sentences and one question at a time, before asking for more Hindi back. The Index itself reports production, because that is the skill the heritage profile lacks. Its first level, Listener, names the profile: understands, answers in English. On most scales that child has no level at all.

And it does not average. Within speaking itself, the five measures do not rise together either. A child may produce plenty of Hindi in three-word answers, or long answers with one grammatical form fossilised in every sentence. Rather than blend these into a score, the Index finds the weakest of them that is holding the next level shut, names it in the parent's report, and that single thing becomes what Kiki works on for the week. A child moves up when the weakest measure moves, not when the average does.

That is the whole design: find the one weakness holding the next level shut, work on it, measure again.

The five levels

Each level is what a child at that stage typically can do with the Hindi speakers in their life. Levels are earned from a run of conversations, never from one, and a level once earned is never taken away.

1. 5

   Conversationalist

   Typically can hold a five-minute conversation with a Hindi-only relative and argue their point.

2. 4

   Storyteller

   Typically can tell Nani a story about their week on a call without switching to English.

3. 3

   Speaker

   Typically can tell you what happened at school today in Hindi sentences, if you ask and wait.

4. 2

   Responder

   Typically can tell the canteen aunty or the auto uncle what they want, in Hindi words.

5. 1

   Listener

   Typically can follow Nani on a video call and answer her in English.

Most children who arrive at Speak with Kiki are Listeners or Responders. The scale is built for the distance between Responder and Storyteller, because that is the distance most diaspora children never cross.

What the Index measures

Five things, from the child's own turns in conversation.

• Output. How much of what the child says is Hindi. Mixed sentences count. A child who says half a sentence in Hindi is producing Hindi, and the scale says so.
• Length. How long a typical answer runs. Three words is a reply. Ten words is a story.
• Accuracy. How often a Hindi sentence needs a correction. Only judged when there is enough Hindi to judge.
• Independence. Whether the child answers on their own or after a parent supplies the words. A coaching parent is never scored as the child.
• Range. How many different Hindi words the child reaches for.

No large language model rates the child. Each measure is a count over the transcript, so the same conversation gives the same result on any day, and a teacher with the recording can check it by hand.

Where children stand today

As of 21 September 2026, drawn from more than 4,700 sentences spoken by children across 704 conversations. Counted for children with at least four scored conversations.

Of the 18 children with four or more weeks of history, 5 have moved up a level. The rest hold their level and are working through the gates of the next one. This table updates weekly.

One child, 10 weeks, Speaker to Conversationalist. Each point is the last five conversations pooled, so one quiet session cannot move it.

For teachers and researchers

The rating manual, with the thresholds behind each level, is shared with schools and academic reviewers on request. If you teach heritage Hindi and would rate ten short recordings against this scale, write to us. Your placements are what turn a draft into a framework.

For parents

Every weekly report shows your child's level, the one thing standing between them and the next level, and one question to try at dinner. Your answer to that question is how we check that the scale reads true at home.

How to cite

The Speaking Index is published under a Creative Commons Attribution 4.0 licence. Use it, adapt it, put it in your own term reports. Name the source.

Speak with Kiki (2026). The Speaking Index: a five-level scale for the spoken Hindi of heritage children, version 2. https://www.speakwithkiki.com/hindi/speaking-index

@misc{hinditutor2026speakingindex, title={The Speaking Index: a five-level scale for the spoken Hindi of heritage children}, author={{Speak with Kiki}}, year={2026}, note={Version 2}, url={https://www.speakwithkiki.com/hindi/speaking-index}}

A DOI for each numbered version will be listed here.