This 'Sarcasm Detector' Trained on Friends, Big Bang Theory

'Multimodal algorithm' out of University of Groningen is quite accurate at identifying the snark
By Jenn Gidman,  Newser Staff
Posted May 21, 2024 8:10 AM CDT
Scientists Use Friends to Build a Sarcasm Detector
David Schwimmer, left, and Matthew Perry in a scene from the "Friends" reunion special.   (Terence Patrick/HBO Max via AP)

Is someone speaking earnestly to you, or are they overflowing with snark? Researchers out of the Netherlands decided they wanted to build a device that could offer some insight, resulting in an artificial intelligence-driven "sarcasm detector" that did quite well in sussing out what speech was sincere and what was more of the mocking or ironic variety. Per the Guardian, scientists at the University of Groningen's speech technology lab debuted their detector on Thursday at a joint meeting in Ottawa, Ontario, of the Acoustical Society of America and the Canadian Acoustical Association, explaining how they trained their algorithm using a previously compiled database called MUStARD.

Using footage from the database that included scenes from such sitcoms as Friends and The Big Bang Theory, researchers Xiyuan Gao, Shekhar Nayak, and Matt Coler created a "multimodal algorithm," per Interesting Engineering. They trained their new neural network on not only the text, but also the audio and emotional content from those clips; sentences pulled from those clips had previously been given annotated sarcasm labels by other scientists from the US and Singapore. Gao and team then ran unlabeled conversations through their sarcasm detector and found that it could pick up on sarcasm in about three-quarters of those exchanges.

That figure was even higher when the team used "synthetic data" to tweak the accuracy, though that research hasn't been published yet. Gao notes the scientists would like to expand on this experiment, possibly incorporating more gestures and expressions (smirking, raised eyebrows) into the algorithm, as well as exploring running more languages through the detector, per the Register. That outlet also notes that such technology could prove useful across a variety of fields, including AI-assisted health care. But even Gao concedes that it's doubtful their sarcasm detector could ever exactly approximate the wit of humans. "Are we going to have a machine that is 100% accurate?" she says, per the Guardian. "That's not something even humans can achieve." (More sarcasm stories.)

Get the news faster.
Tap to install our app.
Install the Newser News app
in two easy steps:
1. Tap in your navigation bar.
2. Tap to Add to Home Screen.