Even then sentiment analysis is comically broken. I was playing with a tool that analyzed Tweets and plugged in a variety of words and phrases. It thought the term “white male” generally had a pretty strong positive sentiment, but on drilling into the Tweets, every single one that it coded positive was actually very disparaging. However, those Tweets were often overtly sarcastic and/or used terms like “privilege” which S.A. models believe to be an indicator of positivity. Some S.A. models think that the term “white” is positive and “black” is negative (probably not a racist conspiracy, but rather because they are too simplistic to understand the racial connotations).
Yeah I've yet to see any NLP solution that comes close to being able to understand sarcasm. It's not an easy thing to do though it requires understanding the whole culture a statement comes from and the message as a whole because of how much earlier sentences change the actual meaning of words.
I doubt we'll have something that really grasps sarcasm until we get close to full human level AI given how much it can trip up even people when communicated by text.
