MR = moi
CG = Christophe

MR> Hearing a distinction doesn't make it phonemic.  Show me a minimal pair. :)

CG> This kind of arguments leads directly to the "heng" phoneme, i.e. that [h]
CG> and [N] are a single phoneme because there is no minimal pair distinguished
CG> by only those sounds.

How about "see her" vs. "singer"?   (In my 'lect, at least, the 'i'
in "-ing" is [i], not [I]).

CG> The speakers intuition is a *very* big part in recognising what is a phoneme
CG> and what is not in a language.

A native speaker's intuition. specifically.

[Excellent French example snipped]

CG> The lexicon of a language is never infinite. So it's normal that you can
CG> find phonemes that don't have any contrastive minimal pair. So that
CG> argument is actually a weak one when you search for phonemes. It's
CG> sufficient, but certainly *not* necessary. On the other hand, the speaker's
CG> intuition is a *very* good argument when looking for phonemic distinctions,
CG> and unless you have *very* good reasons to doubt it (and the absence of a
CG> minimal pair in the current lexicon of the language is *not* a good
CG> argument), it is proof enough.

Ok, I'll buy that the absence of a minimal pair is not sufficient
to disprove phonemicity.  I don't, however, accept that a native
speaker's intuition is sufficient to prove it.  Sometimes native
speakers recognize allophonic differences (even if they can't
describe exactly what the difference is), but that doesn't make
such allophonic pairs distinct phonemes.