Voice assistants and Luxembourg's languages: why it's complicated

Try it: ask a voice assistant for the weather in Luxembourgish. Best case, it answers something vaguely related. Worst case, it offers to call someone. In a country where people juggle several languages before breakfast, voice assistants always seem one step behind. And there's a real reason why.

The small-language problem

Speech recognition learns by listening to huge amounts of recorded audio. The more a language is spoken, written and recorded, the better machines get at it. Luxembourgish is very much alive out loud, but it circulates less in writing and in audio archives than its big neighbours. Less data means clumsier machines: the logic is brutal, and it automatically puts smaller language communities at a disadvantage.

Then add the local speciality: switching languages mid-sentence. A word of Luxembourgish, a French expression, an English term, and everyone follows along effortlessly. Machines, on the other hand, love tidy boxes, one language at a time. Everyday speech in Luxembourg is exactly the kind of blend that throws them off.

So you end up adapting to the machine

Watch people around you: to talk to their devices, many switch to French, German or English. Funny reversal, isn't it? The machine is supposed to serve you, yet you're the one changing languages to keep it happy. Not to mention the country's place names, which text-to-speech sometimes pronounces with truly baffling creativity.

Why there's hope

Technology for so-called low-resource languages is improving: researchers and enthusiasts are collecting recordings, and newer models learn from less data than before. Above all, every conversation, every text, every song in Luxembourgish enriches the soil that future tools will grow from. Speaking a language also makes it visible to tomorrow's machines.

In the meantime, take it with a smile: your everyday multilingualism, the one that feels completely ordinary to you, is still a feat the most advanced assistants can't pull off. Next time a smart speaker answers you with nonsense, remind yourself you just beat artificial intelligence at something. In three languages.