Title: The Evolution of Computational Conceptual History in Science: Tracing the Path from Early Digital Techniques to Large Language Models

Abstract

This paper places large language models (LLMs) in the broader context of the historical development of computational methods used for concept analysis within the history, philosophy, and sociology of science (HPSS). It investigates the unique contributions of LLMs to established methodologies, identifies persistent issues they inherit from earlier approaches, and surveys contemporary case studies utilizing these technologies.

The first section reconstructs the landscape of computational conceptual history prior to the rise of LLMs by synthesizing three distinct research traditions: early digital techniques in HPSS, distributional methods originating from digital history and adjacent fields, and approaches to detecting lexical semantic change. This overview highlights key challenges and opportunities, with a particular emphasis on corpus development, operationalization and modeling decisions, as well as evaluation and interpretation strategies.

The second section shifts focus to the LLM era. After providing a brief introduction to LLMs, it reviews existing LLM-based research on lexical semantic change detection and relevant HPSS case studies. Finally, it revisits the methodological concerns raised earlier, demonstrating how challenges related to corpus construction, model selection and training data, operationalization trade-offs, and evaluation and interpretation manifest within LLM-driven workflows.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC