Discorso d'odio e lessico connotato. Un'applicazione del modello VAD al corpus HaSpeeDe

The lexicon of natural languages includes both connoted and neutral terms. Connoted terms express the speaker's attitude towards the referent of the term. By contrast, neutral terms do not express any such attitude. Connotation can be positive or negative. Hate speech (HS) is understood as any message that expresses contempt or hatred towards an individual or a target group. Hence, a quite natural hypothesis would be that HS contains a high number of negatively connoted terms. Our work aims at verifying this hypothesis. To do this, we use the model developed by Montefinese et al. (2014), which classifies the affective connotation of 1121 Italian words based on three different parameters: valence, arousal, and dominance. We calculated the mean value of these three dimensions in an already annotated Italian HS corpus (HaSpeeDe 2020). The result is quite unexpected as there seems not to exist any meaningful correlation between HS and negatively connoted terms. Not only negatively connoted terms are not necessary to classify a message as HS, but they are not sufficient either. Consequently, HS detection software must take other dimensions into account.