signal and noise in phonologies and scripts

Muke Tever wrote:
> An optimal language will have a balanced signal-to-noise ratio....
This was my gripe about Janet Kagan's HELLSPARK.  The central character
is from a trading culture, whose language deliberately uses all the
phonemes of all known human languages, so that the traders can speak all
languages without `accent' (phonological contamination).  The noise
problem is never mentioned.


In related news, I've invented a script.
I don't know how others go about that.  My approach was formal.  I
wanted to build all the characters out of a small set of features, to
ensure a family resemblance.

I chose to use connected sets of 6 segments out of these 12:
         __ __
        |__|__|
        |__|__|

This gives about three hundred figures (I've misplaced my notes).

But I'm looking for an efficient mathematical way to soften the angles.
I want to write a program to approximate a given piecewise-linear figure
such as

        |__|
           |__|

with analytic functions of

         \______
         /
        /

in the complex plane.  (I was going to use least squares, but found that
cumbersome; I want something that lends itself naturally to successive
approximation, like a Taylor series or continued fraction.  One special
case gives a neat Fourier series, shown on my `doodles' page.)

After my program draws a sequence of such approximations,
I'll decide which degree looks most like real lettering.


--
Anton Sherwood  --  http://www.ogre.nu/