Roberto Dessì

Roberto Dessì is a researcher in Artificial Intelligence and Natural Language Processing, affiliated with Universitat Pompeu Fabra (Barcelona) and Facebook AI Research (FAIR) Paris. He holds a PhD in Language Sciences (2023) and advanced degrees in Cognitive Science (University of Trento, 2019), Cognitive Neuroscience (SISSA, 2019), and Computer Science Engineering (La Sapienza University of Rome, 2017).

His research focuses on emergent communication in multi-agent systems, cross-domain image captioning, language models, and compositional generalization in neural networks. Notable contributions include the EGG framework for studying language emergence and collaborative work on the Toolformer model. He actively reviews for top conferences like NeurIPS, ICML, and ACL, and organizes workshops on emergent communication.

Dessì has supervised multiple students (BSc/MSc theses and industry projects) and delivered talks at institutions including Chalmers University, IST Lisbon, and UC Berkeley. His work bridges theoretical AI research with practical applications in speech translation, tool use by language models, and ethical multi-agent interactions.