My own opinion is that there was a single Proto-World language which
was spoken by the first Homo sapiens to evolve in Africa. But that was
so very long ago that any recognizable surviving features of it that
might help us to reconstruct it, have been "lost in the noise" of
accidental resemblances. We can reconstruct IE quite well, and modern
linguists have written at least one reasonably long connected text in
it, namely Schleichter's fable about the sheep and the horses. Of
Nostratic we can reconstruct only oddments. As for reconstructing
any part of Proto-World :: forget it. Someone once compared Hindi and
modern English, and between those two he found more accidental
resemblances of vocabulary than resemblances which were genuine
cognates.
If you don't mind a rather fanciful metaphor: every so often in this
or that line of scientific discussion the persistent UFO called "The
Flight of Fancy" seems to seems to turn up, and otherwise sensible
people are tempted to board it, and flying about in it see, or think
they have seen, all sorts of strange irrelevant things in the far
beyond, instead of keeping their feet safely on the ground of reality
and practicality. (See
SHADO@yahoogroups.com for ideas how to get it
shot down :-) )
Accidental resemblances can happen in grammar as well as in
vocabulary. E.g. putting a pronoun prefix on subject as well as on
verb :: for "I, Peter, sing", French has "Moi Pierre je chante", and
Nahuatl has "niPetoloh nicuia", exactly parallel, and completely
accidental. So, e.g. how much reliance is there in such things
as looking for areas of ergative languages?