No Proto-Celtic?

[Thomas McFadden]

>> don't.  and in most cases where multiple variants exist, there is an
>> identifiable difference, even if its hard to find.  usually it's not a
>> difference in the truth-conditional semantics, but there's perhaps a
>> difference in what is implied, or in the conditions where you would say it.

> The last is quite frequent.  Often one way to get at this sort of thing is
> by asking what question each of the variants might be an answer to.

yes, a very good point.  it's generally very difficult for people to
explain their intuitions about this sort of variation (like if you asked
somebody 'when would you say it like this and when would you say it like
that').  the best way to investigate this sort of stuff seems to be by
looking at corpora.  when you get 100 examples the one way and 100 the
other of how people actually use various constructions, the patterns, that
are hard for a speaker to describe, often become extremely clear.