At present, segments smaller than <w> (e.g. <c>, <m>) and those larger (eg <phr>)
are all members of model.segLike. This means that (inter alia) the following are
all valid 
(i) <phr> <c/> </phr>
(ii) <phr> <w> <c/> </w> <c/> </phr>
(iii) <c> <w> <phr/></w></c>
While we can all agree that (iii) is barking mad and should be stopped,
it's less clear what to do about (ii). On the one hand, we do now have
a nice new <pc> element for punctuation, which could be defined as a
sibling for <w>, so that (ii) could be replaced by
(ii*) <phr> <w> <c/> </w> <pc/> </phr>
On the other hand, in the absence till now of <pc>, there are
literally millions of words of corpus out there which follow the
pattern of (ii). Do we really want them all to become broken?
I think not, but maybe (as one of the perpetrators) I'm biassed.