[wedn3sday]

My first thought was that this is just the output from the loss function for a genetic algorithm. Lots of people write grant applications, some of those get accepted, the next round of people writing applications then look at the success stories from the previous round and emulate them. Rinse and repeat. New members of the grant evaluation board want to conform to the standards of their predecessors, and so look at the accepted applications from yesteryear to see what they should accept, rinse and repeat. Its a pretty classic self reenforcing feedback loop, and as other commenters have pointed out constriction in lexical space doesnt mean a constriction in idea space. That said, I really enjoyed the article! Love this kind of data driven analysis.