Political Speech as a Forecasting Tool for Democratic Backsliding

• In Event: Navigating Political Narratives: New Frontiers in Text Analysis, LLMs and NLP

Abstract

Is it possible to forecast democratic backsliding from the speeches of political leaders, before institutional decline becomes empirically observable? Traditional methodologies for evaluating democratic backsliding predominantly depend on expert surveys, which often involve extensive resources and suffer from a delay in reflecting real-time changes. Focusing on Turkey as a prime example of an incumbent-driven subversion of democracy, I match expert survey scores with an LLM-driven empirical framework that numerically evaluates the normative commitment to democracy in 18,000 speeches delivered by Erdoğan from 2001 to 2023. Changepoint analysis reveals a noticeable lag between LLM-powered data and expert opinions on Erdoğan, suggesting that extant time-varying explanations for the causes of democratic backsliding may be spurious. This study highlights the need to create alternative, data-driven methods for analyzing democratic trends worldwide, moving away from expert-centric approaches to data derived from local contexts.

Author

• Humeyra Biricik, Oxford University