Some meta-analysis of the actual study, along with some examination of how the media has generally thoroughly misrepresented the study, is available at Language Log.
Thus Component 1 (23.6% of test variance) was significantly heritable — h2 = 0.538. The symbol h2 is used to denote "narrow-sense heritability", which is the ratio between the variance due to average effects of alleles, and the phenotypic variance as a whole:
$$h^2 = \frac{Var(A)}{Var(P)}$$
In other words, about half of the variance in a PCA component accounting for about a quarter of the variance in test results was accounted for by genetic variation.
Component 3 (10.8% of test variance) was also significantly heritable, with h2 = 0.335. Thus about a third of the variance in a PCA component accounting for about a tenth of the variance in test results was accounted for by genetic variation.
The genetic relationships of components 2 (11.7 of test-score variance) and 4 (8.2% of test-score variance) were not statistically significant.
A quarter plus a tenth of the test results were shown to be related at all (not in whole, but at all) to heritable traits. The grand total overall was just under 16% (a half of a quarter, plus a third of a tenth).
Now, I don't know about you, but I wouldn't describe 16% as "largely". I'd describe 16% as "partly", or "mildly", or "somewhat". But of course, reporters for Nat'l Geo and The Independent and the like aren't big on math.
It's still an interesting and intriguing study, of course, but so grossly misreported that it boggles the mind. We need a better grade of chimpanzee writing science articles for the general public! :D