I have read a copy of the original article (somebody on another forum
sent it to me.)  The bulk of it is about finding particular
co-occurrence patterns (which don't seem that spectacular to me, but
the work done on it is solid) in baby babble, and showing that they
are also prevalent in real language.

He presents as a suggestion the idea that these patterns are very
basic to language and therefore could be the substance of a
"proto-language" (not in the hist. ling. sense, but in the sense of an
intermediate evolutionary step towards modern language).

He then points out that they are statistically very common in Bergson
& Ruhlen's 27 "world etymologies."

Overall, the "proto-language" stuff mostly forms a speculative coda to
the main part of the paper, but in the Discover article it turned from
speculation into solid findings and from a speculative coda to the
meat of the paper.