Open your phone or your local government’s website in Ladin, Romansh or Friulian and you’ll quickly run into the same wall: the language isn’t in the list, the speech tool doesn’t recognize it, the system simply wasn’t built with you in mind.
Ladin, Romansh and Friulian aren’t three random minority languages thrown together for a grant application — they’re the three surviving branches of Rhaeto-Romance, close enough in structure and history that a tool built for one has a real shot at working for another.
That’s not just a nice story; it’s the actual bet behind DIGI-RLF, an INTERREG Italy-Switzerland project bringing together governmental and scientific organizations across all three. This talk is a fairly honest account of how that bet is playing out.
There are three strands to the work:
1) The Bureaucratic Plumbing: getting these languages properly registered in CLDR (the Common Locale Data Repository), which sounds bureaucratic but is the actual reason your phone doesn’t offer them as an option today.
2) The Data Challenge: building speech and text resources for them — not easy when the three languages together add up to around half a million speakers, not millions.
3) The Part People Often Skip: figuring out the governance and licensing so that what gets built with public money doesn’t quietly disappear once the project officially ends.
On the speech side, we’re not starting from zero. Eurac Research’s linguaXlab already built AUGUSTA, a model that transcribes South Tyrolean dialectal German speech into Standard German, and with DIGI-RLF we’re now working through what it takes to adapt that approach for Ladin — a real test of whether the family resemblance between these languages actually pays off in practice.
We’ll also touch on Tradutur Ladin, a sister project on translation at the University of Innsbruck, because the same lesson keeps showing up there: without shared datasets and open standards, every small language community ends up reinventing the same tools from scratch, badly, and on a budget too thin to stretch across all of them.
None of these three languages could pull this off alone. Romansh has a real head start in automatic translation thanks to years of work with the University of Zurich; Friulian just got its SuperDizionari online, pulling three major dictionaries into a single platform; Ladin is still out in the Dolomites recording speakers for its very first speech to text data. DIGI-RLF is built on the bet that putting these efforts side by side — across a Swiss-Italian border, between three languages close enough to actually learn from each other — gets everyone further than going it alone.
Come hear what actually carries over from one to another, and what doesn’t.
