How Do You Speak about Immigrants? Taxonomy and StereoImmigrants Dataset for Identifying Stereotypes about Immigrants

Stereotype is a type of social bias massively present in texts that computational models use. There are stereotypes special difficulties because they do not rely on personal attributes. This the case about immigrants, category preferred target hate speech and discrimination. We propose new approach to detect immigrants focusing attributes assigned minority but frames, is, narrative scenarios, which group placed public speeches. have proposed fine-grained psychology grounded taxonomy with six categories capture different dimensions stereotype (positive vs. negative) annotated novel StereoImmigrants dataset sentences Spanish politicians stated Congress Deputies. aggregate these two supracategories: one Victims expresses positive other Threat negative stereotype. carried out preliminary experiments: first, evaluate automatic detection stereotypes; second, distinguish between supracategories immigrants’ stereotypes. In experiments, we employed state-of-the-art transformer (monolingual multilingual) four classical machine learning classifiers. achieve above 0.83 accuracy BETO model both showing transformers can high level accuracy.