Culture

Spotify Tagged Welsh Jolene as Irish Because Engineers Never Asked

Dash·Monday, October 26, 2026 Edition

When Systems Replicate Blindness

Spotify tagged Dolly Parton's posthumous Welsh-language version of "Jolene" as Irish — not a close call, not a debatable edge case. Irish when the language is Welsh, when Parton's heritage was Welsh, when the original recording happened in Welsh.

The streaming giant did not issue a correction for weeks. No public acknowledgment that the metadata was wrong until journalists started asking questions nobody should have needed to ask.

This is not a typo — this is infrastructure failing at scale. Spotify's backend taxonomy routes Welsh and Irish through the same cataloging pipeline because both are minority European languages using the same character sets in the same administrative bucket, built by engineers with no native speakers of either language anywhere near the decision loop.

How systems replicate their creators' blindness

The pattern repeats everywhere metadata gets automated. In 2019, Microsoft's image-recognition system labeled darker-skinned faces with lower confidence than lighter ones, not because anyone programmed racism in. Because the training data came from a world where those disparities already existed. The system wasn't biased. The system was replicating the blindness of its creators. Platforms treat minority-language categorization the way they treat dark mode settings. They view it as optional, low-stakes, and something you sort out once and never think about again.

But for actual speakers of those languages, metadata is infrastructure. It's how you find your own culture on the system you pay for. How the algorithm decides whether you see your own language at all. The moment you stop thinking about Welsh and Irish as linguistic problems and start thinking about them as infrastructure problems, you realize you probably have the same blindness somewhere in your own systems. If you're building anything that touches multiple communities, your taxonomies are almost certainly written by people who don't belong to most of them. The question isn't whether you've got gaps but which gaps you're willing to see.

Key Facts

*Spotify's system routes Welsh and Irish through the same pipeline because engineers saw them as interchangeable minority languages with identical character sets.

*Microsoft's 2019 image-recognition system labeled darker-skinned faces with lower confidence, not from programmed bias but from training data reflecting existing real-world disparities.

*Platforms treat minority-language metadata like dark mode settings—optional low-stakes features, not infrastructure that determines whether speakers can find their own culture.