Even more problematic is the difficulty current algorithms have in distinguishing "engaging" in a certain class of speech acts from studying said class. This requires a subtlety of parsing humans are (sometimes) capable of, but machines still struggle with.