The semantic space of good and evil in early modern English: a computational study of moral contrasts

Abstract This article introduces a novel approach to understanding cultural meaning through the geometric analysis of word vectors in early modern English. Using a seventeenth-century corpus Early English Books Online (EEBO-TCP), the study constructs semantic models that chart the moral and gendered associations of over 32,000 words. Building on methods from distributional concept analysis, the authors define heuristic axes—good-evil and feminine-masculine—and project words onto these axes to explore normative structures embedded in language use. Statistical metrics such as situation, spread, and slope reveal how concepts align with broader patterns of moralized gender. Notably, most terms cluster in the ‘feminine/evil’ quadrant, yet semantic neighborhoods often slope toward abstract, masculine, and positive domains. A new metric, the moral contrast score, enables unsupervised discovery of latent binaries, identifying the cultural dimensions that structure early modern discourse—from affective expression and trust to social refinement and territorial legitimacy. The result is a richly layered moral landscape, where contrasts are not static but overlapping and mobile. Far from mapping fixed binaries, the model reveals a dynamic, high-dimensional field in which language organizes—and reconfigures—cultural values. This method offers a scalable, interpretive framework for investigating the moral geometry of meaning in historical texts.