centre.org

This paper presents a new corpus-driven approach applica-ble to the study of language patterns in social and political contexts, or Critical Discourse Analysis (CDA) using Distri-butional Semantic Models (DSMs). This approach considers changes in word semantics, both over time and between com-munities with differing viewpoints. The geometrical spaces constructed by DSMs or “word spaces ” offer an objective, robust exploratory analysis tool for revealing novel patterns and similarities between communities, as well as highlight-ing when these changes occur. To quantify differences be-tween word spaces built on different time periods and from different communities, we analyze the nearest neighboring words in the DSM, a process we relate to analyzing “concor-dance lines”. This makes the approach intuitive and inter-pretable to practitioners. We demonstrate the usefulness of the approach with two case studies, following groups with opposing political ideologies in the Scottish Independence Referendum, and the US Midterm Elections 2014. 1