I agree that seems presumptive. But of course twitter conveniently didn’t share the source code for how that factor is calculated. Turns out it’s hard to evaluate a big complex distributed system by looking at just one slice and no data.
There's also a place in Twitter's settings to show languages that Twitter has inferred that you know (e.g. following a German-language account and interacting with it almost certainly means that you know some German).