🚀 go-pugleaf

>>>>> "Stefan" == Stefan M�hlbacher <e&s@kabsi.at> writes:

    Stefan> A computer can easily find out the identity when the data
    Stefan> are really equal.  Therefore the pictograms should be
    Stefan> exactly identical pixel by pixel.

Pixel by pixel?  Are you kidding?


It  is  possible to  DEFINE  the  pictograms  in terms  of  high-level
components (circles,  lines, B-spline  curves) instead of  pixels.  To
compare  2 pictograms  then involves  comparing these  components (and
their  parameters)  instead of  comparing  pixels.   This approach  is
techically    superior.     Consider     a    pictogram    which    is
Circle(100,100,50), which is  a circle of radius 50  units centered at
coordinate  (100,100).   Comparing  it  with another  pictogram  means
comparing the component "Circle" and parameters (100,100,50).  This is
much more precise and efficient.   Pixel by pixel comparison will fail
miserably here: different people (or machine) may pixelize a circle in
slightly different ways.  Some people may draw it rounder than others.
So, the  generated pixels are  slightly different.  As a  result, your
pixel-by-pixel  comparison algorithm  must tolerate  such differences,
and this would inevitably introduce lots of inaccuracies.


    Stefan> Now let us assume we describe the symbol fish by a
    Stefan> pictogram. It is very easy, I think we will find a good
    Stefan> one.  But the next task is much harder: Now we want to
    Stefan> find a symbol for each sort of fish in the world: I don't
    Stefan> know how many there are, but I think more than 10 000
    Stefan> maybe more than 100 000. ---> We will need a good
    Stefan> resolution for our pictograms.

Answer: Derivation

Most fish names  in Chinese are characters derived  from the character
"fish".  The character  for a fish name can  be structurally broken up
into 2  parts.  The  first part, the  signific part, is  the character
"fish", which  is used as a  radical in the fish  name character.  The
second  part, the  phonetic  part, is  a  _hint_ (i.e.  NOT a  precise
transcriptio) to the pronunciation of the fish name.

So, when I encounter an  unknown character (e.g.  those created by the
Japanese themselves) having  this structure, I can know  it's the name
of  a kind  fish.  I  don't even  need to  know how  the  character is
pronounced.  (Some  of these fish-name characters are  invented by the
Japanese,  and  hence  have  no  Chinese pronunciation,  and  I  ain't
interested in knowing how it  is pronounced in Japanese.)  I know it's
a fish and that often suffices.


    Stefan> But when we use a good resolution then its very hard to
    Stefan> write down a symbol, you have to define pixel by pixel in
    Stefan> an image editor.  

It's silly to do it pixel by pixel.

Nowadays,  professional  computer  typesetting  do not  represent  the
printouts pixel  by pixel (how many pixel  are there on an  A4 page at
2400dpi?  Astronomical figure!).  Instead, the printout is *described*
in terms  of high-level componnents: lines,  circles, B-spline curves,
and  characters.   (But characters  in  quality  fonts are  themselved
defined in ters of lines and  curves.)  This not only reduces the data
volume, but also makes the description resolution/device independent.

I can't see any advantage  in defining pictograms pixel by pixel.  Why
not give the shapes of them precise mathematical definitions, in terms
of parameterized lines and curves?  

Chinese characters are basically defined in terms of stroke shapes and
their *relative* position and  intersection with one another.  This is
practical and efficient.  Pixel by pixel definition is impractical and
silly.




    Stefan> It is also hard to tell, if a pictogram
    Stefan> is the original pictogram or if there are just a few pixel
    Stefan> different.  

That's why pixel-based definition fails.  Why not define pictograms by
high-level  components, which  are  more natural  to  human beings  to
understand?  Define them in terms of lines and curves!


    Stefan> My theory for this is: at the end it turns out, that each
    Stefan> language can describe the same knowledge, the same
    Stefan> facts. But there is a different way to look at the same
    Stefan> thing:

That's a "very very different" way.


    Stefan> For example: Let us assume languageA knows only a symbol
    Stefan> for the circumference of a circle, not the radius.
    Stefan> LanguageB knows only a symbol for the radius of a circle,
    Stefan> not for the circumference.

    Stefan> Now both languages can define a circle. They can also
    Stefan> define the same circle.  But a translator cannot translate
    Stefan> the size of the circle unless he knows the relation:
    Stefan> circumference = 2*r*PI

Another  example: How is  French "j'ai  faim" represented?   Would the
representation lead to the English translation "I have hunger" instead
of the more natural "I'm hungry"?











-- 
Lee Sau Dan                     $(0,X)wAV(B(Big5)                    ~{@nJX6X~}(HZ) 
.----------------------------------------------------------------------------.
| http://www.cs.hku.hk/~sdlee                        e-mail: sdlee@cs.hku.hk |
`----------------------------------------------------------------------------'