intervocalic devoicing in Welsh (?)

[bwald]

----------------------------Original message----------------------------
In agreement with Alan King's last message on Welsh intervocalic devoicing,
I hope that some Welsh experts will indeed enlighten us further on the
situation.  Meanwhile, the possibility of the phenomenon raised a question
of a more general nature in my mind.  I embedded it in my last message, but
I would like to restate it more elegantly.  It concerns the phonological
conditions on sound change in the following way.
 
Suppose, for the sake of argument, that one or more intervocalic voiced
stops geminate.  Let's express this as:  [+voiced] > [+geminate] / V
[+stop,__] V
 
Maybe that's already odd, if [+geminate] implies an INTERRUPTION of voicing
ONLY in intervocalic position.  Presumably, however, this change is
attested in Northern Welsh.  Thus, odd or not, it becomes an interesting
fact about possible sound change under some conditions.
 
It's really the next stage that raises the issue that I'm most interested in.
 
Suppose that next, the voiced geminate devoices (somehow avoiding merger
with a previously existing voiceless geminate -- that raises all kinds of
questions, but that's not my main concern here).  My main concern is that
since (or if) geminates only occur in intervocalic position, then we can
express the sound change simply as:  [+geminate] > [-voiced]
The point is that there is no need for a condition.  This could be a trick
of economical notation, OR it could be a claim that if there WERE
non-intervocalic geminates they would also devoice.  I suggest that the
sound change is the latter (it's not a notational trick).  In that case, it
is not "intervocalic" devoicing, where "intervocalic" implies a
conditioning factor.  That is, intervocalic position has nothing to do with
the sound change, even though the change only has the OPPORTUNITY to occur
in intervocalic position.
 
There are plenty other cases of such a sequence of sound changes, abstractly:
 
1.  X > Y /W__Z         conditioned
2.  Y > Q                    unconditioned
 
But what other cases are there of sound changes where
 
a.      Y only exists in context W__Z    AND
b.      X > Q / W__Z  is an UNEXPECTED sound change.