Data from the DictEmbed dataset (Papers 1-3). Each point is a legal term definition embedded by nomic-embed-text into 768 dimensions, then projected to 3D.
The cross-jurisdictional terms (Pflichtteil, Sorgerecht, hadanah, etc.) represent legal concepts that exist in other legal systems but have no English equivalent. Their position in embedding space reveals how models handle concepts that fall outside their training distribution.