The detection and handling of verbal irony in human-computer interaction: a comparison between four chatbots

Abstract

This study investigates what AI-technique works best for detecting and handling verbal irony. The effects of linguistic irony markers have also been taken into consideration. The AI-techniques have been examined through statistical analysis of chatbot replies in ironical human-computer conversations. It has been found that UltraHAL (pattern matching) handles irony significantly better, and that Rose (NLP meaning extraction/ChatScript) handles irony significantly worse, than others.