One Language is Enough: Using Transfer Learning to Detect Populist Rhetoric Across 25 Languages Without Translation

• In Event: Text as Data

Brief Overview

We show that fine-tuning a multilingual language model on just one language can accurately detect populism across 25 languages—removing the need for translation or multi-language labels—enabling scalable cross-lingual political text analysis.

Authors

• Yuan (Cecilia) Sui, Washington University in St Louis

• SOYEON river JEON, Washington University in St Louis

• Christopher Lucas, Washington University in St Louis

• Jacob Michael Montgomery, Washington University in St Louis

• Margit Tavits, Washington University in St Louis