Change My Mind: how Syntax-based Hate Speech Recognizer can Uncover Hidden Motivations based on Different Viewpoints

Hate speech recognizers may mislabel sentences by not considering the different opinions that society has on selected topics. In this paper, we show how explainable machine learning models based on syntax can help to understand the motivations that induce a sentence to be offensive to a certain demographic group. To explore this hypothesis, we use several syntax-based neural networks, which are equipped with syntax heat analysis trees used as a post-hoc explanation of the classifications and a dataset annotated by two different groups having dissimilar cultural backgrounds. Using particular contrasting trees, we compared the results and showed the differences. The results show how the keywords that make a sentence offensive depend on the cultural background of the annotators and how this differs in different fields. In addition, the syntactic activations show how even the sub-trees are very relevant in the classification phase.