Slashdot Mirror
Why Sarcasm Is Such a Problem In Artificial Intelligence (thestack.com)
An anonymous reader writes: A new paper from researchers in India and Australia, "Automatic Sarcasm Detection: A Survey," highlights one of the strangest and ironically most humorous facets of the problems in machine learning and humour. The paper outlines ten years of research efforts from groups interested in detecting sarcasm in online sources. It details the ways that academia has approached the sarcasm problem, including flagging authors and ring-fencing sarcastic data. However, the report concludes that the solution to the problem is not necessarily one of pattern recognition [PDF], but rather a more sophisticated matrix that has some ability to understand context.
. . but rather a more sophisticated matrix that has some ability to understand context.
Yeah, right.
I am Slashdot. Are you Slashdot as well?
Wake me up when we solve the problem of deterministically detecting sarcasm with human intelligence.
The reverse is also true. Because it's so hard to discern sarcasm in online text, it can also be difficult to express sincerity. One can write something truly sincere, only to have others interpret it as sarcastic, flippant, or derisive. Adding emphasis, such as "I'm being absolutely serious here" only worsen the problem, since over-emphasis is a hallmark of sarcasm.
There is also sometimes intended ambiguity; people write things that could be interpreted as either sincere or sarcastic/joking, and wait until after the fact to claim the intended meaning. For instance, one can send a text/email that is flirty, and decide (based on the recipient's response) whether to claim it was sincere or a joke.
In short, this is not just a hard problem for AI, it is a hard problem for intelligence more generally. Sarcasm is--quite intentionally--sitting right on the edge between credibility and exaggeration.