And especially considering that the researchers "reduced all sounds to 34 distinct consonants and 7 vowels." (Consider that the more we reduce and simplify the vocalizations the more likely we should expect correlations. For example, if we simplified all sounds to grunts we should expect tremendous overlap.)

The correlation of the corpus of 100 words isn't particularly surprising given what we already know about language development. The null hypothesis was that there'd be no correlation whatsoever (i.e. the sounds would be completely random), which I don't think anybody could reasonably expect.