From November 4 to 9, 2025, we were in Suzhou, China, attending the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP 2025).
On Thursday, November 6, Guido Ivetta presented the paper “HESEIA: A Community Dataset to Evaluate Social Biases in Large Language Models”, together with Luciana Benotti, during the Q&A session. Our collaborative project brings together more than 45,000 sentences to identify intersectional and local biases, challenging the self-mitigation techniques of current language models. The presentation was highly praised!
The HESEIA experience emerged from the teacher training course developed by Vía Libre in 2024, which involved 370 teachers and more than 5,000 students from 189 schools across Latin America. Throughout the process, participants collaboratively built a dataset of 46,499 sentences to identify and analyze biases and stereotypes present in artificial intelligence models. Unlike other existing resources, HESEIA incorporates an intersectional and situated perspective, reflecting local contexts and the pedagogical experiences of educational communities. The poster presentation also took place during the conference.
As Guido stated during the conference:
“The collaborative construction of regional evaluation strategies for AI models paves the way toward more equitable, representative technologies that are aware of voices historically marginalized or made invisible.”
The poster presentation also took place during the conference. You can access the published paper at the following link.
On November 7, Luciana Benotti took part in the session “How Will Multimodal Language Models Shape Industry and Society?”, within the industry session panel. Based on questions proposed and voted on by the audience, the discussion addressed key topics about the future of #NLP, biases in LLMs, and the emerging role of small language models in the multimodal language processing ecosystem, including images.
In addition, on this occasion, Luciana Benotti received the Outstanding Senior Area Chair Award, granted by the ACL during the conference, in recognition of her outstanding work evaluating scientific papers for #EMNLP 2025 within the Human-Centered NLP area.
Proud to share this work at one of the most important conferences in natural language processing, representing a perspective from the Global South. 🌎💪
