That seems an edge use case: using a keyboard for a language you don't speak. Fingerprinting with language setting is minimal for a typical use with max 2-3 languages and millions of people in the world with the same fingerprint. Lowering priority to 3.

> That seems an edge use case: using a keyboard for a language you don't speak.

I would expect that use case to be pretty common in NBU where English keyboards may be the ones most readily available and/or where it may be a question of prestige to use an English keyboard.

Regarding fingerprinting, I've used Panopticlick [1] to estimate entropy for HTTP ACCEPT [2] headers: {en-US} (4.75 bits), {en-US,de} (10.86 bits), {en-US,en-UK} (12.12 bits) and {en-US,de,en-UK} (16.96). Taking en-US as a baseline and subtracting that value to get rid of the contributions of non-language ACCEPT headers:

 0.0 bits: en-US
 6.1 bits: en-US, de
 7.4 bits: en-US, en-UK
12.2 bits: en-US, de, en-UK

Compared to the total entropy that is sufficient for identification of approximately 30 bits these are very sizeable numbers and I'd kindly ask you to be mindful of this when designing language features and to not consume more entropy than absolutely necessary.

[1] https://panopticlick.eff.org/

re: "I would expect that use case to be pretty common in NBU where English keyboards may be the ones most readily available and/or where it may be a question of prestige to use an English keyboard." Those users mostly have their UI on English for the same reasons and thus have English in their accept language header by default.

Thanks for the entropy numbers. We'll keep that in mind.

> Those users mostly have their UI on English for the same reasons and thus
> have English in their accept language header by default.

Which seems less than ideal. But I understand that better heuristics/ML are in the works?

Yes (happy to chat offline if you're curious) though they wouldn't change anything in this case. 

Having thought about this more, the convenience to most users of combined settings makes it unlikely that we'd ever decouple the too. Marking as won't fix.