Sarcasm detector helps AI understand language

A research team from the University of Groningen in the Netherlands has presented a multi-stage algorithm that is designed to help artificial intelligence in particular understand the subtleties of human communication.

Most attempts to correctly classify sarcasm are based on just one parameter, which regularly leads to failure. However, a specially developed analysis of moods based on the text and emotions based on the audio recording should provide a better overall picture.

In this way, videos or audio recordings can be used to filter out the pitch, the speed of speech or the type of pronunciation between powerful and velvety. In addition, after transcription into text, an in-depth analysis is carried out. This can be used to determine emotions for individual text passages, which are then added as a second layer in the form of emoticons.

This expansion of an audio recording to include emotions and linguistic deviations can then be trained using machine learning. Although the accuracy rate is not perfect, the AI trained in this way can distinguish sarcasm from serious statements fairly reliably.

Other features that can be easily extracted from existing video files are also planned. For example, there are certain movements and gestures that can be observed when only half-serious content is being presented. This additional layer could make the algorithm even more reliable.

The system is also to be extended to a large number of languages. And according to the study, it is also conceivable that the algorithm could be used to detect other hidden messages beyond sarcasm or irony. Lies would certainly be a desirable example.

But there is one thing that AI cannot prevent: The fact that portals such as oe24 take over content that is definitely not meant seriously from websites that are definitely sarcastic, but take them for real themselves. That would be real-life comedy.