Individual entities and proper names in cognitive neuroscience and natural language processing
Proper names refer to individual entities, also called unique entities or individual concepts, which are personally and socially relevant. In cognitive neuroscience, proper names of people and geographical places seem to most clearly involve entity-specific semantic processes in the brain, differing from those involved with generic (non-individual) entities. It is an open question, however, what drives the organization of the representations of people and places in the brain and what are the core components of such representations. To the aim of investigating such questions, we run three experiments using high-density EEG and a multivariate encoding/decoding approach, with both whole-brain and Searchlight-based approaches. We find that we can decode (1) categorical information, both at a coarse- (people vs places) and fine-grained (taxonomical hypernyms) level. We can also decode (2) familiarity-related information, with patterns that present striking commonalities with those found in the face recognition literature, suggesting they may reflect amodal processes. Then, we (3) employ models coming from computational linguistics and natural language processing and based on purely textual, distributional lexical information, that in recent years have shown to capture lexical knowledge to a surprising extent. We provide original evidence that recent contextualized models such as BERT and XLM best capture brain processing of individual entities, both personally familiar and famous. Finally, when (4) analyzing comparatively the spatio-temporal patterns of encoding/decoding via a searchlight approach, our results converge with previous works in showing that bilateral fronto-temporal regions are associated consistently with processing of individual entities, but also that differences among people and places emerge.
- Theses