Despite advances in LGBTQIA+ rights, the Internet continues to be a hostile environment for LGBTQIA+ individuals.. The increasing frequency, severity, and complexity of online hate crimes are mirrored in the real world. In a recent ISTAT-UNAR survey on discrimination on work places suffered by LGBTQIA+ people, 43,9% of the participants has been target of insults, 61,8% suffered a micro-aggression (including hate speech), and 1.1% has been physically assaulted. In addition to this, anti-LGBTQIA+ hate crimes have risen drastically in the past three years. Natural language processing (NLP) is a key subject of research for combatting online hate speech since it can automate the process at scale while reducing online moderators' labor and mental stress [1]. Despite the NLP community’s interest in hate speech detection datasets and models, very few studies covered homotransphobia [2]. This is a concern, due to the target-oriented nature of hate speech: recent studies [3] have revealed that hate speech detection methods cannot be used to multiple sorts of hate speech targets.
The proposed shared task HODI will identify Italian homotransphobia on Twitter. This will allow us to investigate a phenomenon that has received little attention from the worldwide NLP community and has never been built for the Italian one. To encourage participation, we will make available a dataset including a binary task of being either homotransphobic or not.
It is critical to comprehend not only whether a text is hateful, but also why. Recent European legislation, General Data Protection Regulation (GDPR [4]), has introduced a “right to explanation”. This necessitates a paradigm change from performance-based models to interpretable models [5]. This shared task will also contribute towards this need by assessing the models' explanation abilities to recognize the terms relevant for hate speech. This will permit, in the future, to control for possible biases of models overfitting to specific terms [6].
The HODI shared task will focus on identifying homotransphobia in Italian tweets. HODI is organized according to two main subtasks:
1. Chaudhary, M., Saxena, C., & Meng, H. (2021). Countering online hate speech: An NLP perspective. arXiv preprint arXiv:2109.02941.
2. Chakravarthi, B. R., Priyadharshini, R., Ponnusamy, R., Kumaresan, P. K., Sampath, K., Thenmozhi, D., Thangasamy S., Nallathambi R., & McCrae, J. P. (2021). Dataset for identification of homophobia and transophobia in multilingual YouTube comments. arXiv preprint arXiv:2109.00227.
3. Nozza, D. (2021, August). Exposing the limits of zero-shot cross-lingual hate speech detection. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers) (pp. 907-914).
4. Eu regulation 2016/679 general data protection regulation (GDPR). https://eur-lex.europa.eu/eli/reg/2016/679/oj, 2016. Accessed:2022-10-09.
5. Mathew, B., Saha, P., Yimam, S. M., Biemann, C., Goyal, P., & Mukherjee, A. (2021, May). Hatexplain: A benchmark dataset for explainable hate speech detection. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 35, No. 17, pp. 14867-14875).
6. Attanasio, G., Nozza, D., Hovy, D., & Baralis, E. (2022, May). Entropy-based Attention Regularization Frees Unintended Bias Mitigation from Lists. In Findings of the Association for Computational Linguistics: ACL 2022 (pp. 1105-1119).