Software Recognizes Sarcastic Tweets 168
An anonymous reader writes "Even humans sometimes fail to recognize sarcasm and irony; can machines do better? An algorithm that identifies sarcastic tweets (PDF) on Twitter and sarcastic sentences in product reviews on Amazon will be presented next week in the International Conference for Weblogs and Social Media in Washington, DC, and in the Computational Natural Language Learning in Sweden in July. A team from the Hebrew University, Israel, has developed an algorithm that identifies sarcastic sentences by using a machine learning technique in which a small number of sarcastic sentences act as seeds for the software to learn and generalize upon. The algorithm can then identify sarcastic sentences that are nothing like the examples. The variety of recognized sarcastic sentences is impressive, though the results are not perfect. But again, we don't do it so well ourselves, do we?"
Finally a way to verify Smith's Law (Score:3, Interesting)
Re:This is great! (Score:5, Interesting)
Re:This is great! (Score:5, Interesting)
Many, though not all, Asperger's types actually have average to excellent parsing of written communications, or the strictly verbal component of other people's utterances(ie. the part that would get written down, if a transcriptionist were in the room). Odds are, most such people could easily outperform this algorithm(since, obviously, the purpose of the algorithm is to provide large volumes of adequate analysis for cheap, not to be human level).
The part of communication that is really difficult, though, is the nonverbal component, the stuff that doesn't show up in text. Tone of voice, expression, tiny muscular movements and reconfigurations around the eyes, that sort of thing. Since typical social standards of politeness and interaction actually discourage direct statement of things(ie. "Your story bores me." "Yes, I am interested." "No, go away") in favor of relying on subtle nonverbal communication of those message, this can be a real handicap. You care about what others around you are thinking, since you naturally want to be on good terms with them(or, even if you don't, you want to be on bad terms deliberately, not accidentally); but you just can't tell, unless somebody explicitly says something, which is rare, unless you've already really fucked up.
In fact, in my experience,(and yes, "my experience" = "N of 1" = "anecdote") I tend to find text-based communication comfortable for exactly these reasons. For normal people, strict text-based communication is harder, because they are denied the nonverbal cues that they normally take for granted. For me, I don't see the nonverbal cues that never mean much anyway, and we are both forced to rely on strict verbal expression, which is my best-practiced level.