Software developed by a Dutch university in Groningen can now detect sarcastic facial expressions, according to NOS. Researchers Xiyuan Gao, Shekhar Nayak and Matt Coler of the University of Groningen used clips from the TV shows Friends and The Big Bang Theory to train their algorithm to spot sarcasm.
The researchers spotted a number of patterns when it comes to sarcasm in different languages. In English and German the pitch drops and varies less, while in Italian, French and Cantonese, the pitch actually rises, and the variation also increases.
The researchers concluded that a reliable sarcasm detector will therefore have to be trained per language. "The challenge is to not only look at the sound of the speech but also, for example, at the context in which a statement is made," speech technologist Matt Coler told NOS. “And facial expressions are also a very accurate predictor of sarcasm.”
The researchers also struggled with a lack of available data to train their algorithms. "That was a big problem for us," said Coler. In the end, the team used four hours of TV clips to train their algorithm.
The results? 74 percent of the time, the algorithm detected sarcasm correctly. The software was also able to indicate with approximately equal success clips that were not sarcasm.