Skip to main content

Heidi: Understanding Zurich German Comes First

We are building Heidi, a place to learn Züritüütsch. Before writing code we read the research, and it overturned the obvious plan: exposure beats rules, listening beats speaking, and in Switzerland understanding is a complete goal on its own.

September 10, 20262 min readCato

Most people who move to Zurich learn German and then discover it does not help at the lunch table. Everyone is speaking dialect, they understand almost none of it, and the Swiss politely switch to Hochdeutsch or English the moment they notice. The learner is denied the input precisely because they are a learner.

That is the problem Heidi is for. Before building anything, we spent a day reading what the research actually says. Three findings overturned the plan we would otherwise have shipped.

The obvious curriculum has a published null result

Zurich German maps onto Standard German by regular sound correspondences: Kind becomes Chind, Haus becomes Huus, ist becomes isch. The obvious course teaches those rules first. That was tested, on Dutch speakers learning Frisian, with a lesson on exactly this kind of correspondence, and it produced no significant improvement in comprehension (Bergsma, Swarte & Gooskens, 2014). In the largest study of how people understand closely related languages, exposure alone explained most of the variance; linguistic distance explained far less. Exposure is the mechanism. Rules are scaffolding inside practice, not a lecture before it.

Understanding is the goal, not the consolation prize

Second-dialect acquisition research is consistent: adults rarely reach native production in a second dialect. In Switzerland that matters less than anywhere else, because passive competence is a socially complete endpoint. Enormous numbers of well-integrated residents understand dialect fluently and reply in Standard German, and that is normal and respected. So Heidi puts listening first, texting second, and speaking last.

The audio has to be recorded, not downloaded

There is no purchasable Swiss German text-to-speech, the best open speech models translate dialect into Standard German rather than transcribing it, and SRF's archives are licensed for private use only. Every second of Zurich German a learner hears in Heidi will come from a real person. That is a cost, and it is also the moat.

What is live today

A first page at heidi.orangecat.ch, built through FleetCrown and deployed to our own box. The product will be measured rather than gamified: the number we want to show you is your comprehension of unfamiliar Zurich speakers, before and after, not a streak.

If you live in Zurich and would let us record you saying ordinary things in your own dialect, we would like to hear from you.

Part of our commitment to building in public

More from OrangeCat

Discover more insights about Bitcoin, security, and building in public.