Matras combining char's?

[Michael Everson]

Lloyd Anderson of Ecological Linguistics asked me to forward this to
the INDOLOGY list.
 
Date: 01 Oct 93 01:21 GMT
From: ECOLING at AppleLink.Apple.COM (Ecological Linguistics,Anderson,PRT)
 
A note in Unicode 1.1 suggests the inclusion of Indic matras as
"combining characters" should be considered a "defect" in 10646 (and
therefore should be changed, presumably).
 
This note to the Indology list is to give my best understanding of the
situation, since I have not been able to get a technical justification
for the position referred to.
 
The original motivations for "combining characters" included at least
that they should not be isolated from their base characters at line
breaks.  I think, though am not absolutely certain, that some Unicoders
want a class of characters which have the combination of properties
[combining; non-spacing; overlapping only to the left (in left-to-right
scripts) or to the right (in right-to-left scripts)] and perhaps some
other properties as well.
 
As contrasted with the prototypical Latin combining marks, Indic matras
have two differences: (a) they are not all non-spacing, and (b) in
surface rendering, they sometimes connect with something later in the
text stream, rather than only with something earlier in the text
stream.  These two differences are however not matters of
"combiningness", they are separate features.  The rule that "combining"
characters may not be separated from their "base" characters still
holds.  So new character attributes can be added, "zero-width" and
"combines-to-the-right", but the old feature "combining" should not be
deleted.   Possibly it should be reinterpreted as "combines to the
left" unless an additional feature handles the left vs. right linking,
but since the "combines to the right" affects primarily renderings
among the standard Indic vowel matras, this may not matter.  The matra
is linked to the same base letter (conjunct) in either case, before or
after matra-reordering to the rendering form.
 
The Unicode 1.1 treatment of the diacritics over two letters, such as
long and short marks over double "oo" for phonetic dictionary entries
for English, places the code for the diacritic between the codes for
the two base letters. The approach considers such diacritics to be
"combining to the left", but "overlapping" to the right, whereas they
are really equally combining in both directions, so "combines to the
left" and "combines to the right" would both be attributes of these
double-base diacritics.
 
Does anyone have further insight into what might make Indic matras
different or similar in any relevant attributes?  The chief cases I can
locate for combining-tothe-right, by the way, are Devanagari  repha
(itself a result of ligaturing "r" with virama non-finally), and
Burmese "eng" which sits on top of the consonant which logically
follows the "eng" (itself a result of ligaturing "ng" with a marker for
conjuncting.
 
Lloyd Anderson, Ecological Linguistics.
==========
 
Michael Everson
School of Architecture, UCD; Richview, Clonskeagh; Dublin 14; E/ire
Phone: +353 1 706-2745  Fax: +353 1 283-8908  Home: +353 1 478-2597