Quoting John Vertical <[log in to unmask]>:

> Roger Mills wrote:
> >This is the old Bi-uniqueness Problem of classical Phonemics:  1) given a
> >phonemic form, the phonetic form must be predictable, and 2) given a
> >phonetic form, the phonemic form must be predictable. Thus the problem with
> >e.g. German [bUnt] =? bunt 'colored' or Bund 'association'; and oddities
> >like Engl. house ~houses [haUz@z], vs. most (all?) other words with a /-s/
> >that doesn't voice in the plural.
>
> I actually recall reading that the final devoicing of stops in German is only
> a
> near-merger - that there's some slight phonetic cue that's basically
> impossible
> to hear, but appears regularly when recordings are examined rigorously. A
> similar scenario might apply here, too. Frex the [v] in [niva] might be
> slightly
> palatalized in comparision to [neva], and we could then declare that
> there's /v/ = [v v_j], and /f/ = [f v]. [v] would however still be the
> realization
> of two different phonemes then. (Andreas just posted last week about a
> similar issue in Swedish, and suggested preferring to consider the same
> phones
> also the same phonemes.)

I do not think the issue of 'light sj' in my 'lect is particularly comparable;
here we have genuine [f]~[v] alternation in stems - the lack of [x]~[S]
alternation is one of the two principal reasons I would suggest rejecting the
identification of light and dark sj as the same phoneme (the other being the
acoustic identity and complimentary distribution of light sj with tj (aka
/s\/)).

                                              Andreas