Measuring Autocratic Originality: Parsing Genuine from Propaganda in Speeches

• In Event: Quantitative Study of Elite Communications

Abstract

Understanding the nuances of elite politics within authoritarian regimes necessitates innovative computational methodologies. This study introduces a novel approach employing advanced Natural Language Processing (NLP) techniques to unveil autocratic leaders' preferences and their influence on state policies.
At its core, the methodology utilizes Language Models (LLMs) for word embeddings, facilitating a comparative analysis of leaders' speeches against official statements. This analysis yields a Semantic Similarity matrix, quantifying semantic correspondence and revealing phrases of personal interest to leaders, along with deviations from prescribed narratives. Augmenting this, clustering techniques are applied to speeches. When combined with the similarity score, this creates a policy-specific similarity index for the autocrat, quantifying policy priorities and their subsequent impact on state decisions.
Introducing a diachronic perspective, the approach establishes two pivotal indices. The "Originality Index" gauges temporal deviations from established agendas, providing a quantification of the leader's innovation. In contrast, the "Dictation Index" evaluates alignment between leaders' speeches and forthcoming official statements, shedding light on the leader's influence in dictating state directives.
Importantly, this methodology relies solely on publicly available data, eliminating the requirement for insider information. The meticulously designed pipeline encompasses web scraping, text preprocessing, word embedding, Semantic Similarity Index computation, clustering, and index derivation. This framework systematically uncovers the intricate interplay of preferences and dominance within elite autocratic politics.
In conclusion, this paper presents a computational NLP methodology for deciphering the intricate dynamics of authoritarian elite politics. By quantifying leaders' policy preferences and their dictation power over state actions, this approach enables precise, systematic, and data-driven analysis of autocratic governance.

Author

• Jonathan Elkobi, Yale University