🚀 go-pugleaf

flaps@dgp.toronto.edu (Alan J Rosenthal) writes:

>unrza3@cd4680fs.rrze.uni-erlangen.de (Markus Kuhn) writes:
>>Believe me, German users who know what they are talking about won't
>>complan if "a, "o and "u are not sorted as ae, oe and ue.

>I believe you.  I'm sure that the analogous statement is true for French.  But
>I'm almost as sure that Swedish readers won't want "a-circle" to be anywhere
>near "a".  I think your flexibility on this is a language-relative phenomenon.

This would be really bad news, could someone from Sweden coment on this?

My idea is, that there is one latin alphabet, that is known by everyone
using a latin script:

  abcdefghijklmnopqrstuvwxyz

And the following rule may be understood by every user within less then
10 seconds:

  sort latin characters with diacritics (e.g. "a, 'a, ^a, ...) near their
  pure latin version a.

The time needed to understand this rule is independend of which latin
alphabet based language the user uses normally.

Then you still have to insert special characters (e.g. Icelandic Thorn)
at suitable places THAT ONLY NEED TO BE REMEMBERED BY PEOPLES INTERESTED
IN THIS LANGUAGE! That's really simple to understand. Ok, perhaps users
should not be forced to use this standard multilingual sorting, but if it
is offered as one possible 'international locale', I bet many people will
like it very quickly.

My algorithm is a kind of generalized upcase conversion: All a letters
(a, A, A-ring, a-ring, "a, "A, ... perhaps even greek alpha and Alpha)
form one group. Without the huge far east character sets, there might
be about 50-80 groups. Each ISO 10646 character is assigned a group
number and members of the same group have the same group numbers.
Then we sort according to group numbers, not according to character
codes. If the comparison between two words fails, because their group
string doesn't differ, than let's compare by the positions within a group.
Each character is also given a group position number, that provides a total
order on all characters within a group. E.g. we might define in the standard,
that ring-above comes always after acute-accent etc. These definitions are
only necessary to offer a total ordering and need NOT be known by 99% of
the users, because they will be significant very rarely (e.g. comes
cooperation before or after coöperation [ö="o]?). In addition,
there is a punctuation group, that is ignored in the first pass
and a space group. Not to forget 10 digit groups, where all the
different digit versions in Unicode will be summarized, etc.

I'd like to discuss here, where the cyrillic, greek etc. letters
should be included. The positions that correspond to the latin letters
used in international transcription systems  might be a good starting
point, if this is possible while preserving the ordering of e.g. the
greek alphabet. If this is absolutely impossible (e.g. with han characters),
than of course completely seperate groups have to be used.

I have started to write a full algebraic specification of the algorithm,
but it is really a pain to do this in ASCII. :-) Perhaps a 20-line C
function will be a better specification.

It would be possible to encode group number and position within a group
in the character code, but this would increase the number of bits needed,
because there would have to be big gaps in the code space. I prefer
16-bit ISO 10646 together with a table that gives me a 2x2 byte code 
(group x position in group) for each character. Clever programmers
will store this table efficiently in much less then 128kBytes.

>>Word lists produced by my algorithm are pretty easy to scan for human eyes.

>is language-relative as well.  If people are used to thinking of a-circle as
>being at the end of the alphabet, it may become a different letter.  For all I
>know they think of it as a circle with an "a" as a diacritical mark rather than
>the other way around.

If we want ONE alphabethical order for an international locale, than we
have to use what is easily understood by the majority of the world
population. No one should be forced to use the international sorting,
but for many people, a systematic approach will be very useful.

Are there any experts in the Unicode Consortium, that believe, that specifying
the details of this algorithm together with a well designed group table
would be worth the effort? This would be a very nice next standard submitted
by the Unicode Consortium to ISO ... :-) With out it, there will terrible
ISO 10646 sorting methods based on the Unicode code number of each character
become common practice!!! The definition of these groups might perhaps also be
useful for case-and-diacritic-invariant searching, because case-only-invariant
searching is only half a solution in Unicode that was ok with ASCII.

Markus

-- 
Markus Kuhn, Computer Science student -=-=- University of Erlangen, Germany
Internet: mskuhn@immd4.informatik.uni-erlangen.de  |  X.500 entry available
--- Wer, wie, was? Wieso, weshalb, warum? Wer nichts fragt bleibt dumm. ---