There really needs to be some way for language preferences to include quality in some way. No, I probably don't want an English dub of something Asian when there's also a German dub. But I certainly also don't want some machine or lowest bidder translation that only exists to serve people with no foreign language skill at all (or that's not even remotely as good as the company that commissioned the translation thinks.
In theory, there is. It's called content negotiation. With your browser request you already send a list of preferred languages, like "Accept-Language: da, en-gb;q=0.8, en;q=0.7". RFC9110 states that such a quality/weight is assigned to different representations of a resource.
So all the server needs to do is multiply their quality with your preference and return the one with the highest value.
So the technology for that is already in place. It just isn't done.
See: https://www.w3.org/International/questions/qa-lang-prioritie...
AFAIK, there's no standard for allowing client-side media-playback frameworks to probe language preferences with quality values out of the browser they're running in. The navigator.languages property (https://developer.mozilla.org/en-US/docs/Web/API/Navigator/l...) doesn't expose qvalues for some reason.
And even if there was, there's no way, in any media container format I know of, of annotating subtitle tracks with quality values — so there's no way of saying that a given track really sucks and should be deprioritized relative to the user's preference ordering, but still used if there's no other choice.
It turned out that a supplier of a critical subcomponent in some of the systems we were designing and building, had terrific German-language documentation.
The English docs? Not quite as great, packed with ambiguities which simply weren't there in the originals.
Seeing as I was the only one on the team who could read German almost fluently, I simply pulled the 'Who's going to read the docs if I am let go?' card during every downsizing process for years. It worked until someone realized we hadn't supplied any systems with that particular component in it for years and I, alas, had typed up excellent documentation for our service department, covering their needs.
> Link Zeitcode Generierung
be best understood as something like “Link time-code generation,” rather than, “Link-time code generation,” or something like that?
Just for clarification, this says "Link Timecode Generation" where it's supposed to be "Link-time Code-generation"?
I always assumed English-German MT must be somehow serviceable, I wonder if that had been true but limited to its syntax and did not apply to vocabulary, or if German MT is just as useless but German computer users are just patient.
Can't we "just" train a model with a big corpus of English, a big corpus of German, and a big corpus of human-translated Rosetta-Stone-like "annotated with translation" texts — and out will pop a "bilingual" language model that can do translation as fluently as a human could?
Do we just not have the right training data available to make this happen?
And instead of this butchering of the fine German engineering language, just switching to English is indeed the better alternative (and less painful).
My favourite trivia btw is that the 'else' in 'if-else' is a "bad" translation from German into English ;) (https://github.com/e-n-f/if-then-else/blob/master/if-then-el...)