Describing Moral Concepts with Topic Modelling and Wikipedia

The topic of morality has occupied great minds for much of human history. One line of research fundamental to moral psychology is providing a description of the content of morality that is capable of explaining variations in moral priorities across individuals, cultures and time periods. Unfortunately, current descriptions of morality are often highly subjective, constructed based on a small team of researchers’ interpretation of a few prominent religious and / or academic sources. The primary aim of our project is to construct a quantitative, cross-cultural description of the moral domain, employing an alternative approach that addresses this issue. To this end, we used Wikipedia, one of the largest structured databases of human knowledge, to construct a corpus of morality-related texts. This corpus was constructed by identifying a set of unambiguously morality-related “seed pages” (e.g., the Wikipedia entries for “Ethics”), and programmatically retrieving these pages, along with any Wikipedia pages with links to or from these seed pages. This resulted in a corpus of > 14,000 articles, containing > 5,000,000 words. This corpus was then analyzed using topic modelling, a natural language processing method that summarizes corpora using latent variables (topics), identified based on patterns of word co-occurrences. As expected, the topical content covered a wide range of identifiable moral subject areas, including generic moral terms, specific classes of transgressions, and varieties of moral agents. Future work will extend this approach to languages other than English and validate the model as a tool for describing new moral texts.