But, how does that explain the decline from, say, 1900 to 1990? The decline appears fairly steady over the 120 year period.
I can put a chart of lexical similarity over a chart of the number of employed airplane mechanics and probably pontificate that by golly, we need to stop fixing airplanes for the sake of intellectual diversity!
And that would be about as well-reasoned.
"The diversity words are new (new-ish). Anything else newish would be inversely correlated with cosine distance too. For example, I repeated the same experiment with "tech" words like Google, YouTube, iPhone, browser, and so on. Those tech words correlated at -0.6 with cosine distance"