This task is designed in a hierarchical fashion by chaining two subtasks and allowing participants to either model the simple binary scenario or complete the entire pipeline by identifying a second binary classification problem. That is, firstly, to detect whether a text contains a stereotype, and secondly, whether the stereotype is expressed explicitly or implicitly:

  • Subtask 1, stereotype Identification: This is a binary classification task the aim of which is to determine whether a comment or sentence contains at least one stereotype or none, considering the full distribution of labels provided by the annotators. This subtask follows the SemEval 2021 Task 12 (Uma et al., 2021) proposal about learning with disagreement, in which the authors state that there does not necessarily exist a single gold label for every sample in the dataset. This fact is particularly evident when multiple contradictory annotations arise at the data labeling stage due to “debatable, subjective, or linguistic ambiguity”. The actual gold label of this subtask is left as a proxy to determine the subset of comments that will be evaluated in the posterior subtask. Examples of instances with and without stereotype:

    Stereotype: Immigrants have more rights than us.

    No Stereotype: The new government will reform the immigration law.

  • Subtask 2 (Optional), Implicitness Identification: This subtask introduces a novel binary classification problem to determine whether the stereotype is manifested or latent within the text, that is, whether the stereotype is implicit or explicit. The added difficulty in this case is that implicit stereotypes are not directly expressed in the text, and a process of inference must be applied by the annotators. Moreover, there are different strategies in which an implicit stereotype can be coded, such as metaphors, irony and other figures of speech, evaluations of the in-group, and the overgeneralization of a social group from features of some of its members. For instance, an example of an implicit stereotype taken from our dataset is the sentence “We will end up being a minority in our own country”, in which the stereotype is expressed through an evaluation of the in-group. This subtask will be presented as a hierarchical binary classification problem. Examples of instances of explicit and implicit stereotypes:

    Explicit: Immigrants have more rights than us.

    Implicit: We will end up being a minority in our own country.


Uma, A., Fornaciari, T., Dumitrache, A., Miller, T., Chamberlain, J., Plank, B., Simpson, E. & Poesio, M. (2021). ‘SemEval-2021 Task 12: Learning with Disagreements’. In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021) (pp. 338-347). Association for Computational Linguistics. DOI: https://doi.org/10.18653/v1/2021.semeval-1.41