YouTube’s auto-caption algorithm mistook English for Dutch — the results are hilarious!

Clearly, YouTube’s auto-captions algorithm has some bias when it comes to accents.

My colleague Mariya Voytyuk recently posted an Epidemiology 101 video on Science Showcase. Her voice over is in crisp, clear English, but with a European accent — which YouTube’s auto-captions algorithm decided to decode as Dutch.

Not speaking Dutch, I used YouTube’s auto-translate function to auto-translate the Dutch auto-captions back to English.

The results are quite bizarre! Clearly YouTube’s algorithm was trying to find meaning in the narration beyond literal phonetic translation. Which says something about the sophistication of the natural language processing they’re deploying here.

But it also suggests a rather deep flaw with YouTube’s AI when it comes to jumping to algorithmic conclusions about where someone’s from, and what language they’re speaking!


The original video can be watched at Science Showcase.