Andrea Bruera



Enrichment Student

Cohort year


Partner Institution


Andrea Bruera is a 3rd year Ph.D. student in the CogSci group at QMUL, under the supervision of Massimo Poesio. His Ph.D. project revolves around the semantics of individual entities and proper names, looking at how these are represented in brains and machines, and trying to bring together both cognitive and computational approaches. He was a visiting Ph.D. student at the Language, Learning and Reading lab in SISSA in Trieste, and at the COLT group in Universitat Pompeu Fabra in Barcelona. Also, while on a break from his Ph.D., he worked as a Junior Research Scientist at SAP Security Research in Mougins, France, on applying NLP techniques for de-identification and data anonymisation. He holds a BA in Philosophy from the University of Torino and a M.Sc. in Cognitive Science from the University of Trento.

Research interests

The topic of Andrea’s Ph.D. is the investigation of how the representations of individual entities - entities referred to by proper names, such as ’Jacinda Ardern’ - are shaped and structured, both in the brain and in artificial intelligence models of language. In both cases it has been shown that individual entities have peculiar features when compared with generic entities such as trees, streets, birds - mostly because they are extremely fine- grained in meaning and social in nature.

The methodology that he is applying lies therefore at the intersection of computational linguistics (Natural Language Processing - NLP) and cognitive neuroscience. The aim is making the most of both approaches, in order to provide relevant evidence for both fields. From a neuroscientific perspective, models of language from computational linguistics, especially when implemented as vectors as in distributional semantics models, are of particular interest. They provide a way to isolate very precise phenomena in the brain, by using machine learning to map from the models to the brain, or the other way around. From the point of view of artificial intelligence, brain data allows to directly quantify the cognitive plausibility of each model, by comparing the fit between models and brain data.

As a brain data recording technique, he is using mostly electroencephalography (EEG), which captures cognitive brain processing as reflected by electric signals recorded on the scalp, but also magnetic resonance imaging (fMRI).