Project: Modelling common-sense knowledge in automatic irony detection

2021-10-01 – 2025-09-30

Abstract

Despite the major advancements in the domain of sentiment analysis during the past few decades, automatically modelling implicit sentiment still remains a major hurdle in the field. While sentiment analysis systems perform well at identifying explicit sentiments, they have a hard time inferring implicit sentiments or identifying non-evaluative words that evoke a particular sentiment within a text or a sentence. The reason why is simple: automatic systems, compared to human beings, do not possess any common sense or world knowledge to decipher hidden meanings which renders them unable to create links with the world around them. Instead, computers can only rely on what they have learned from specific data. The presence of irony is a prime example of implicit information that an automatic system is unable to fathom. Transplanting this so-called human commonsense knowledge to a system and being able to allow it to finally grasp implicit sentiments would signify a real breakthrough in any field related to NLP. This project aims to address the challenge of modelling implicit or prototypical sentiment in the framework of automatic irony detection for Dutch.