Message-Id: <6.2.3.4.2.20050827195628.04e712d0@mail.afrac.org>
Date: Sat, 27 Aug 2005 20:11:17 +0200
To: "Addison Phillips" <addison.phillips@quest.com>,
	"Harald Tveit Alvestrand" <harald@alvestrand.no>,
	"Mark Davis" <mark.davis@icu-project.org>
From: r&d afrac <rd@afrac.org>
Subject: RE: Nonstandard distance metrics (Re: [Ltru] Re: matching for 
  non-RFC3066bis langtags)
Cc: "Doug Ewell" <dewell@adelphia.net>,
	"LTRU Working Group" <ltru@ietf.org>
In-Reply-To: <634978A7DF025A40BFEF33EB191E13BC0C9A61C3@irvmbxw01.quest.c
 om>
References: <634978A7DF025A40BFEF33EB191E13BC0C9A61C3@irvmbxw01.quest.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed

Dear Addison,
I was not refering to the mechanism which is really interesting, and 
I will review in detail when you have stabilised it. To see if one 
could establish local matrices, may be matching our investigated DNS 
matrices for emergency services for example (when you are traveling 
and get an accident or an hart failure, you want http://hart.sos to 
resolve at the nearest hospital [DNS root matrix], you also want to 
be able to talk with them). This is not that complex once there is 
the proper architecture. This is for the time being what I focus on.

I was refering to the nice formulation about that "interesting 
special case" Mark told Harald "where the user wanted "xx-Aaaa" and 
"xx-Bbbb" to match with a score of +INFINITY, so that when the user 
asked for "xx-Aaaa", "en" would match better than "xx-Bbbb".

Most Gov, Industry and Commerce people I discussed with around the 
world about your langtags tend to think that Harald's, Mark's, 
Peter's, Randy's and your's motivation is that "special" case. No idea why?

But may be should you get familliar with that confusion?
Cheers.
jfc




At 18:08 27/08/2005, Addison Phillips wrote:
> > This has a known name: "shaping the world".
> > It took time before someone eventually acknowledges it.
>
>No, this is called "language negotiation according to user 
>preferences". It is what the "q" weights in Accept-Language are for, 
>for example.
>
>There are some different examples of how this could be useful.
>
>Case #1: Consider a user who speaks both Slovenian and Italian who 
>subscribes to an RSS news feed. RSS indicates the language of an 
>article using a language tag. If the user instructs her newsreader 
>to filter the articles in the news feed, she might use this 
>Accept-Language setting:
>
>sl;q=1.0,it;q=0.8,sl-rozaj;q=0.0
>
>The lowest distance metric value for each article (when comparing to 
>each range in succession) is the weight for the article as a whole. 
>That weight can be divided by the q weight associated with that 
>range to produce the "final score". When the articles are ordered by 
>score, this produces a list in which Slovenian articles always 
>appear before Italian articles on each "level" of matching, but in 
>which articles in the Resian dialect ("sl-rozaj") never appear 
>(score is "infinity" essentially, since divide by zero is an error) 
>or at least appear at the very end.
>
>Case #2: Consider a news feed from Japan. A friend tells me that an 
>interesting English language article appeared in this particular 
>news feed, but can't remember exactly when or what it was called. I 
>can set "ja;q=0.0" to filter out all the Japanese (presumably the 
>majority of the news feed), making the article easier to find.
>
>Addison
>
>Addison P. Phillips
>Globalization Architect, Quest Software
>Chair, W3C Internationalization Core Working Group
>
>Internationalization is not a feature.
>It is an architecture.