eginhard

No corpus is ever "clean". Depending on the type of corpus there might be many misspellings, so obviously they will occur in the graph alongside other low-frequency words.

Zipf's law [1,2] generally holds for a corpus in any natural language and can be applied to a lot of other things outside linguistics as well.

[1] Zipf’s word frequency law in natural language: http://colala.bcs.rochester.edu/papers/piantadosi2014zipfs.p... [2] https://en.wikipedia.org/wiki/Zipf%27s_law

Considering that it makes use of internal machine learning tools and probably refers to a lot of other internal technologies it wouldn't be of much value to anyone outside Google I guess.

It's the normal plural, but 3 and 4 take the nominative, while other numbers take the genitive plural.