Extracting Causal Arguments from Political Text

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

Abstract

Narratives are the main means by which humans organize, process and communicate information. For politics, narratives shape beliefs and present versions of political reality. Analyzing narratives is usually limited to small-N studies and are subject to researchers’ interpretation. However, because they have an underlying structure, computational tools can be used for their systematic analysis. Previous research has experimented with language models to predict characters and plots to automate the extraction of narrative in text. We propose accessing narratives via their causal arguments. Leveraging recent advances in NLP, we test novel LLM fine-tuning and RAG techniques to extract and summarize causal arguments from political texts. We annotated the causal structure of 18,000 sentences to use as dataset for training and testing. For political communication, this method can automatize causal arguments, that can then be used for inferences about how politicians construct narratives.

Author

• Paulina Garcia Corral, Hertie School