Date and time: 26/06/25, 5pm CET
Speaker: Simona Frenda, Computer Science Department, Heriot-Watt University, Edinburgh y
Aequa-tech, Turin
Presenter:
Abstract:
The dominant paradigm in natural language processing (and other areas of artificial intelligence) is to resolve disagreements observed in annotations and transform them into a single ground truth through aggregation or statistical means. Recently, however, the automated solution of subjective tasks (e.g., abuse detection) has highlighted the need to consider the multiple points of view emerging from data annotation as equally valid. Indeed, the harmonization of labels and the presupposition of a single ground truth silence minority voices.
In this talk, we will explore the reasons for the importance of disagreement in model creation, walk through the typical workflow in supervised machine learning for text classification, and discuss different approaches to creating and validating models informed by diverse voices, opinions, and perspectives.
Bio: Simona Frenda is a postdoctoral researcher in the Department of Computer Science at Heriot-Watt University and co-founder of the startup aequa-tech. She obtained her PhD in a co-supervised program between the University of Turin and the Polytechnic University of Valencia in Computer Science in 2022, with a thesis examining implicit manifestations of abusive language against women, migrants, and cultural minorities in a multilingual setting, paying particular attention to the role of humor. Her research focuses on the creation of models that include minority voices, especially in tasks such as irony, hate, and stereotype recognition, personalized counterspeech generation, and automatic text rewriting using inclusive language. This study received the award for the best doctoral thesis in NLP from the SEPLN. She has co-organized several shared tasks such as IronITA 2018, HaSpeeDe 2020 and EMit 2023 at EVALITA, DETESTS-Dis 2024 at IberLEF, and GFG 2024 at CALAMITA, and is co-organizer of several editions of the NLPerspectives workshop at ECAI 2023, LREC 2024 and EMNLP 2025.
Registration (mandatory).
Zoom link (mandatory registration): https://zoom.us/webinar/register/WN_U326nRbrTIe79oV_uhAk8A
