Re: [Ltru] my technical position on extlang

Mark Davis scripsit:

> Just to make sure that we're talking about content negotiation, eg using
> Accept-Language to get the language for a website, where "zh" means "zh" or
> anything starting with it. 

Check.

> Here's an example. Let's suppose that content tagged with "fr", "gsw", and
> "yue" (or "zh-yue") are available, and nothing else. Let's suppose also
> that I understand German, French, and Mandarin, but not [anything else].

> Both models
> 
>    - If I specify "de fr", I'll get "fr" under either model. Fine,
>    I can understand "fr".
>    - If I wanted to also request Swiss German as well as German,
>    I would use "gsw de fr", and get "gsw".
>    - I only get "gsw" if I request it. That's good, because I don't
>    understand "gsw".

Check.

> No Extlang Model
> 
>    - If I specify "zh cmn fr", I'll get "fr" under either model. Fine,
>    I can understand "fr".
>    - If I wanted to also request Cantonese as well as Mardarin,
>    I would use "yue cmn zh fr" and get "yue"

That would only work if the Cantonese content was labeled "yue", which is
not possible until 4646bis goes into effect.  So for practical purposes
you'd need to say "yue zh-yue cmn zh fr", since "zh-yue" is currently
the *only* way to tag Cantonese.

>    - I only get "yue" if I request it. That's good, because I don't
>    understand "yue".

That's literally true, but you *will* get Cantonese content tagged "zh-yue".

>    - I need to include "zh", and will for the indefinite future because
>    the vast majority of Mandarin content is tagged that way.

Check.

>    - I might get content tagged "zh" but which is actually "yue" but the
>    chances of that are extremely remote.

Not so.  If the Cantonese content is tagged "zh-yue" (and again, that is the
*only* way to tag Cantonese today), you will get it.

> Extlang Model
> 
>    - If I specify "zh zh-cmn fr", I would get "zh-yue". Broken, I don't
>    understand Cantonese!

Check.

>    - In order to specify that I want "zh" or "cmn" but no other languages, I
>    have to use "zh-cjy;q=0, zh-cpx;q=0, zh-czh;q=0, zh-czo;q=0, zh-gan;q=0,
>    zh-hak;q=0, zh-hsn;q=0, zh-mnp;q=0, zh-nan;q=0, zh-wuu;q=0, zh-yue;q=0 zh
>    zh-cmn fr".

Check.  And yet:

If most users can handle most of the encompassed languages, then the
no-extlang model is painful (you have to specify too many positive-q
tags).  If most users can only handle a few encompassed languages, then
the extlang model is painful (you have to specify too many zero-q tags).
The situation is symmetrical.

>    - Moreover, as you said, if another encompassed language shows up
>    for zh, I could get that inadvertently unless I add this to my list.

Again, it's symmetrical: if another encompassed language shows up, I
could miss that inadvertently (in the no-extlang model) unless I add
it to my list.

>    - Again, I need to include "zh", and will for the indefinite future
>    because the vast majority of Mandarin content is tagged that way.

Check.

-- 
You know, you haven't stopped talking           John Cowan
since I came here. You must have been           http://www.ccil.org/~cowan
vaccinated with a phonograph needle.            cowan@ccil.org
        --Rufus T. Firefly
_______________________________________________
Ltru mailing list
Ltru@ietf.org
https://www.ietf.org/mailman/listinfo/ltru