Skip to main content
Paper

BERT's Conceptual Cartography: Mapping the Landscapes of Meaning

Authors
  • Nina Haket (The University of Cambridge)
  • Ryan Daniels (The University of Cambridge)

Abstract

We present a method for analysing context-sensitive word meanings using BERT embeddings and Gaussian Mixture Models in the fields of lexical pragmatics and Conceptual Engineering. Our methodology generates visual conceptual landscapes that reveal how words cluster in different contexts, demonstrated through a case study examining the term \textsc{planet}. We provide quantitative metrics for meaning stability and contextual variation, useful for researchers studying lexical pragmatics and meaning change. We also provide an open-source tool which offers an accessible interface for generating visualisations and metrics, requiring minimal technical expertise. Results show that even seemingly straightforward terms exhibit complex meaning landscapes that resist simple definition, highlighting the importance of context-sensitive analyses, combining quantitative metrics and qualitative approaches. This work bridges theoretical pragmatics and computational linguistics, offering empirical grounding for studying how word meanings shift across contexts.

Keywords: lexical pragmatics, word embeddings, conceptual engineering, BERT, word meaning

How to Cite:

Haket, N. & Daniels, R., (2025) “BERT's Conceptual Cartography: Mapping the Landscapes of Meaning”, Society for Computation in Linguistics 8(1): 10. doi: https://doi.org/10.7275/scil.3145

Downloads:
Download PDF

44 Views

10 Downloads

Published on
2025-06-13

Peer Reviewed