Google's new product identifies whether a comment could be perceived as “toxic" to a discussion

Their sarcasm detection is excellent.

Like all natural-language processing and machine learning (totally not a buzzword) projects, this is something that linguists totally had a hand in, and is not the intellectual masturbation of some propellerhead computer geek. Sentiment analysis is totally based on reality, because single words and maybe digrams or trigrams tell you everything you need to know about whether or not a sentence is “good” or “bad”. Context be damned!