Re: Berber/Tifinagh (was: Swahili & Banthu)

From: Doug Ewell (
Date: Sun Nov 09 2003 - 23:29:37 EST

  • Next message: Chris Jacobs: "Re: Transliterating font"

    Michael Everson <everson at evertype dot com> wrote:

    > This has nothing to do with encoding. You are harkening back to the
    > hideous world of 8-bit font hacks of twenty years ago.

    and Philippe Verdy <verdy underscore p at wanadoo dot fr> responded:

    > In fact that's exactly the opposite which may be possible: the
    > Tifinagh code point could have two graphical representations: the
    > classical one with its exclusive glyphs, or the Latin one. Such
    > tolerance is not allowed when representing code points assigned to
    > Latin letters, so Berber texts that are already coded with Latin code
    > points (using many more or less successful but incompatible
    > conventions) will need an explicit transliterator to convert them in
    > the new common script...

    "Such tolerance" would not be allowed for Tifinagh code points any more
    than it is for Latin code points. Michael is right: the character/glyph
    model says that you are not supposed to render characters in such a way
    as to alter their basic identity. You can render LATIN SMALL LETTER A
    using the one-loop (Comic Sans MS) or two-loop (Times New Roman) glyph,
    as a wedding-script or Fraktur, or Sütterlin letter, but you cannot
    render it as a "b" or an Arabic or Tifinagh letter, or you have
    essentially lied about what character it really is.

    Likewise, let's suppose for illustration that TIFINAGH LETTER YEB ends
    up being encoded at U+08A1, which seems reasonable given the current
    roadmap and discussion document. This letter apparently has at least
    three reasonable glyph variants, which look roughly like U+25EB, U+29B6,
    and U+2296 respectively. (Not that I am suggesting unifying this letter
    with any of these shapes!) If we assume that all of these glyphs turn
    out to be acceptable for TIFINAGH LETTER YEB, then any of them (or all,
    if the OpenType table has such a mechanism) could be used in a font and
    mapped to U+08A1, because they would all reflect the basic identity of
    the character. But it would not be appropriate to render Tifinagh yeb
    as a Latin b or a Greek beta or a Cyrillic be, because these are four
    different letters, and to render them otherwise is to lie about
    character identities.

    > But the new script would gain an immediate representation with
    > existing Latin fonts (for example a Times New Roman or Arial font,
    > updated to support the Tifinagh code points), until new fonts that
    > support the classical glyphs become available to users. Of course, to
    > not break this evolution, there's a need of some consensus about which
    > alternate Latin (or Greek) glyph can safely represent the Tifinagh
    > code point. Then it's up to the user of that script to select which
    > font best represents the script and it becomes a matter of style...

    Again, Michael is right, and so is John Hudson. If you want to
    transcribe between Tifinagh and Latin representations of the same text,
    the font is not the place to do it. You don't have to wait for Tifinagh
    to be encoded to try this sort of experiment; you could do it now with
    (e.g.) Katakana, mapping U+30CE to both the correct Katakana glyph and
    the Latin letters "no". But that's all it should be, an experiment, not
    a standard practice. Create two Web pages for Berber text, one in Tfng
    and one in Latn, and give your users a choice that way.

    -Doug Ewell
     Fullerton, California

    This archive was generated by hypermail 2.1.5 : Mon Nov 10 2003 - 00:10:00 EST