Abstract
Moralization, the process by which concepts and practices gain moral attributes, plays a pivotal role in shaping individual and social behaviour. However, research on how moralization unfolds over time remains limited. We present HistMoral, an open-access computational framework based on human word association, historical text corpora, and graph neural networks that enables scalable, retrospective analysis of moral trajectories of many different concepts. We apply our framework to analyze the moral time courses of over 20,000 concepts within the Corpus of Historical American English over the past 150 years, as well as within the New York Times annotated corpus from 1987 to 2007. Our findings provide robust evidence of moralization across diverse categories, from diseases to world leaders, and identify moralization around economic-political shifts of recent decades.