Create a formal process for justifying phonemes

[thetommoore]

As it is, it is difficult to determine whether phonemes should be added or removed because the guidelines are largely qualitative. It would be better to use tools like spectral analysis to determine if two sounds are sufficiently different to be classified as different phonemes.

[trans]

That would be very cool and logical. But how to do it? I did a little research and found some interesting things:

• http://www.aclweb.org/anthology/W06-3201
• http://www-01.sil.org/computing/pa/
• http://search.cpan.org/~jaspax/Lingua-Phonology-0.3503/Phonology.pm

Even so, I am not sure exactly how we would go about such an endeavor. The best I can come up with as a simple model that we can perhaps work out manually is a tree. Start with all possible phones, then divide them into groups based on a clear distinction. Continue to do so for each subgroup. Perhaps it could even be a binary tree all the way down. This might not be quite as scientific as we would prefer, but it might work well enough.

For a more precise approach we would have to find a way to enumerate the character of each phone such that it accurately represented the psycho-acoustic distinction, and such that each of these measures was part of a geometry so the distance between phones could be calculated. If we could do that, then we should be a able to reverse engineer it, so to speak, to give N number of phones optimally distant from one another (given a single point-of-origin phone).