Interestingly, the tall s seems to have replaced the short one in a relativvely shout timeframe. apart from the transition period arount 1800, this looks like a smooth curve: http://ngrams.googlelabs.com/graph?content=fuck,suck&yea...

Andrew West provides an excellent (and detailed) list of the rules for using "long s" (as well as some Ngram graphs) at http://babelstone.blogspot.com/2006/06/rules-for-long-s.html.

Just one example of why people need to be careful using this data. It's a very useful source, but I've been seeing a lot of uncritical uses cropping up. For example, some people are using it to track intellectual trends--- compare the graphs of Heidegger and Russell or so on. This can sometimes work, but depends heavily on: 1) uniqueness of names; and 2) the particular set of books included in Google's corpus (especially if comparing people not from the same exact area, like a scientist versus an artist).

Even with relatively unique names, it can be tricky. The case of completely or almost completely unique last names (like "Nietzsche") is easy, but with the available interface to the data, it's difficult to handle cases where First+Last is unique, but last alone isn't. You need to count things like "First Last" and "Last, First", plus variants like "First M. Last", without double-counting.

Multiple meanings for 'tits' are also a consideration. (Maybe my monitor or eyes suck, but I had to do a tits vs fuck only ngram search to see that the big spike is in fact tits.)

If a positive is a hit on one of the seven words, then aren't you saying there are a lot more "f"-"uck"s than there should be (via "suck")? I would have thought the reverse : than there would be more "f"-"hit"s - i.e. false negatives.