Evolution of anatomical concept usage over time: Mining 200 years of biodiversity literature

Abstract

The scientific literature contains an historic record of the changing ways in which we describe the world. Shifts in understanding of scientific concepts are reflected in the introduction of new terms and the changing usage and context of existing ones. We conducted an ontology-based temporal data mining analysis of biodiversity literature from the 1700s to 2000s to quantitatively measure how the context of usage for vertebrate anatomical concepts has changed over time. The corpus of literature was divided into nine non-overlapping time periods with comparable amounts of data and context vectors of anatomical concepts were compared to measure the magnitude of concept drift both between adjacent time periods and cumulatively relative to the initial state. Surprisingly, we found that while anatomical concept drift between adjacent time periods was substantial (55% to 68%), it was of the same magnitude as cumulative concept drift across multiple time periods. Such a process, bound by an overall mean drift, fits the expectations of a mean-reverting process.