The x-axis doesn't mean anything, and the years are jittered a little to make it easier to read. If I ever get this database running quickly enough to use, I'd love to color-code the words by machine-categorized field (science, religion, politics, etc.). Any other ideas for additional information this could show?
*Actually the year in which a fitted loess curve peaked. I turned up the sensitivity of the loess curve because it gave better results for a few words (knownothingism, in particular) that were intensely used for only a few years and then at a lower level of use later.