The role of ethnicity in civil war is a time-honoured subject of peace and conflict studies. The question is whether and how ethnic diversity increases the likelihood of the outbreak of a civil war - or an ethnic conflict in that case. Despite valuable insights from both qualitative and quantitative studies on the issue, there remains a conceptual ambiguity about what an ethnic conflict actually is and how it comes about. How exactly does ethnicity contribute to the outbreak of civil wars?
Expanding on Cederman, Wimmer, and Min's work, this paper tries to further our understanding of the role of ethnicity in civil wars by integrating discourse theoretical notions from cultural studies (e.g., Stuart Hall) with machine-learning-based approaches to text analysis.
We argue that in order to understand the emergence of ethnic conflicts, we need to understand how ethnicity is discursively constructed as an antagonistic signifier. It is not enough to measure the ethnic composition of a region or even the demographic changes over time to understand how and when ethnic diversity leads to ethnic conflict. Instead, we need to analyse the discursive dynamics that change the role of ethnic signifiers from descriptive to antagonistic.
In contrast to current conceptions in peace and conflict studies, ethnicity should not be treated as an ascriptive category but rather as a constructed and flexible collective identity. Thus, the antagonistic potential of ethnicity lies not in the existence of ethnic difference, but in the possibility to discursively construct it as an antagonistic signifier in contentious political moments.
In our empirical analysis, we illustrate that it is possible to systematically trace the activation of ethnicities as antagonistic signifiers in public discourse. To that end, we analyse two most-similar cases that show comparable levels of ethnic diversity but different levels of ethnically motivated political violence. Using both cross-case and within-case variation, we develop a refined causal mechanism for the relation between ethnicity and civil wars.
The purpose of these case studies is mainly exploratory and geared toward theory-building. However, we also use them to show how machine-learning-based approaches to text analysis can facilitate a large-scale application and analysis of our theoretical proposition. Discourse theoretical scholarship and quantitative approaches to text analysis are often perceived as being at odds with each other. In this paper, we demonstrate that machine-learning-based approaches are ontologically and epistemologically compatible with most discourse theoretical approaches and are therefore promising for bridging epistemic divides.
