One can count occurrences of epithets in a few lines of a scripting language. Identifying hate speech is a matter of identifying the speaker's objectives--reasons for speaking. Natural language processing and pattern recognition can do this. If FB, Youtube and Twitter are really trying to identify hate speech by counting insults, then we're Actual Idiots for taking them seriously.

In Arlington County, Virginia of the mid-1970's, 7th and 8th grade math students were treated to a week-long exploration of BASIC programming, complete with access to a
HP 9830A, and an HP 7260A Optical Mark Reader. We used HP Educational BASIC Cards.
So, the drill was: write your program on paper, transcribe it to the cards with #2 pencil, then get in line to put your cards in the reader. Inevitably the Reader would choke on a card, and issue a "Pick Fail" error. That could be due to a damaged card or to the number of erasures and rewrites on a card. Pick Fails were always accompanied by three honks from an alarm inside the Reader. Moans from students waiting in line for card reading usually followed. The best you could hope for was one iteration of your program per day, but realistically you got 2 or 3 runs during programming week, what with all the Pick Fails.