IEEE VIS 2024 Content: Seeing the Shift: Keep an Eye on Semantic Changes in Times of LLMs

Seeing the Shift: Keep an Eye on Semantic Changes in Times of LLMs

Raphael Buchmüller - University of Konstanz, Konstanz, Germany

Friederike Körte - University of Konstanz, Konstanz, Germany

Daniel Keim - University of Konstanz, Konstanz, Germany

Download camera-ready PDF

Room: Bayshore I

2024-10-13T18:05:00ZGMT-0600Change your timezone on the schedule page
2024-10-13T18:05:00Z
Exemplar figure, described by caption below
Hi, and thanks for joining. In a nutshell, our research looks at how Large Language Models are reshaping the conceptual framework of our language. While language change has traditionally been driven by socio-linguistic factors like metaphorization, we introduce three new ideas: recontextualization, standardization, and what we call semantic dementia. Using visual analytics, we can track these shifts to preserve linguistic diversity and reduce bias. We review key methods, like embedding-based techniques, to detect and explain these changes. In the end, we call for new visualization tools to better understand how LLMs are impacting our language. Thanks for watching.
Fast forward
Abstract

This position paper discusses the profound impact of Large Language Models (LLMs) on semantic change, emphasizing the need for comprehensive monitoring and visualization techniques. Building on established concepts from linguistics, we examine the interdependency between mental and language models, discussing how LLMs influence and are influenced by human cognition and societal context. We introduce three primary theories to conceptualize such influences: Recontextualization, Standardization, and Semantic Dementia, illustrating how LLMs drive, standardize, and potentially degrade language semantics.Our subsequent review categorizes methods for visualizing semantic change into frequency-based, embedding-based, and context-based techniques, being first in assessing their effectiveness in capturing linguistic evolution: Embedding-based methods are highlighted as crucial for a detailed semantic analysis, reflecting both broad trends and specific linguistic changes. We underscore the need for novel visual, interactive tools to monitor and explain semantic changes induced by LLMs, ensuring the preservation of linguistic diversity and mitigating linguistic biases. This work provides essential insights for future research on semantic change visualization and the dynamic nature of language evolution in the times of LLMs.