Re: Sound changes

On Sat, 23 Apr 2005 10:00:59 -0400, Geoff Horswood
<geoffhorswood@...> wrote:

> How do you generate a reasonably logical, realistic set of sound change
> rules?
Guesswork, whim and a shot of randomness. That seems to be how it works in
the real world.

> How rigorously do you need to apply these rules? (are there exceptions in
> historic sound change systems?)
Sound change operates completely regularly, without exception. Any
apparent exceptions have rules about when and how the exception occurs,
that themselves apply absolutely regularly -- except for their (entirely
regular) exceptions. And so on.They're also semantically blind -- I don't
think there's a sound change that only works on (e.g.) nouns, or whatever.
As such, sound changes are relatively simple to apply computationally
(once you have a sufficiently advanced context-sensitive search and
replace function).

Working in the opposite direction is a process called "analogy". Analogy
takes the horrible mess that sound changes can sometimes create, and
regularizes it. It is also *not* semantically blind. An analogical rule
can work based on the part of speech or semantic domain of a word. As
such, it's much harder to model computationally.

To sum up: Sound change works regularly to create irregularity, and
analogy works irregularly to create regularity.





Paul