Is NLP almost at human-level accuracy in picking up on subtext?

Can it formulate a human-level response to sarcasm? Could it respond in kind?

Does it know when a question is rhetorical? :)