Cholo, Indio, Mono: Ecuador and the Linguistic Context of Hate Speech on Twitter

Hate speech on social networks consists of the online publication of offensive language generally aimed at vulnerable groups in society. Previous studies have suggested a causal relationship between terrorist attacks and hate posts on social media. Faced with this problem, some governments have proposed legal action to require social media platforms to control content that may incite hatred and violence. In Ecuador, no government actions have been proposed to regulate content on social networks. On this basis, this study develops a first approach to the development of a theoretical-practical framework to identify the hate speech of users of the Twitter platform in Ecuador. Furthermore, regarding this use case, this research aims to generate empirical evidence for the debate and discussion of the laws that regulate the offensive content of the publications generated. For this, we assess the ability of three models based on NLP techniques to then classify Twitter posts as hate speech. The results showed that the word embedding techniques are better adapted to the linguistic idiosyncrasies of Ecuador; and therefore, they are more accurate in identifying Hate Speech with semantics typical of this geographical area. On the other hand, our findings show that the transfer learning technique of the pre-Trained BERT (Bidirectional Encoder Representations from Transformers) model is better adapted to identify Hate Speech in general linguistic contexts.