Offensive language is pervasive in social media. One of the most widely-used strategies that social media platforms use for tackling this problem is to use computational methods to identify offensive content in user-generated content (e.g. posts, comments, microblogs, etc.). In this talk, I discuss some of the challenges and opportunities of using NLP models to recognize offensive content online.
I will discuss the lessons learned from creating new annotation taxonomies (e.g., OLID, TBO), organizing shared tasks (e.g., OffensEval at SemEval), and applying different computational models to this task (e.g., multi-task learning, federated learning). Finally, I also discuss recent work presented at ACL and EMNLP 2023 that aims to improve explainability of model predictions.
Marcos Zampieri is an assistant professor at the School of Computing at George Mason University. His research interests are in computational linguistics and natural language processing (NLP). His research deals with the collection and processing of large bodies of texts from various sources (e.g. social media, newspapers) with the goal of understanding how language and communication works, and how we can use this knowledge to build robust NLP systems.
He has published over 100 peer-reviewed papers on topics such as language and dialect identification, machine translation, educational NLP applications, and offensive language in social media. His publications appeared in journals and proceedings of conferences such as ACL, COLING, EMNLP, and NAACL. He has co-edited a dozen journal special issues, workshop proceedings, and edited volumes.