Using Transfer Learning to Detect Populist Rhetoric Across 25 Languages Without Translation

Christopher Lucas, associate professor in the Department of Political Science at the Washington University in St. Louis, will present on his approach to introduce multilingual language models to political science for cross-lingual text analysis without translation. He’s found that multilingual language models offer a practical, efficient solution for global comparative research, dramatically reducing the data requirements for cross-lingual political text analysis while still producing accurate, valid measures.

About the Seminar

Social scientists increasingly analyze text data across multiple languages, yet common methods rely on costly translation or require labeled training data for multiple languages. To address these limitations, Lucas introduced multilingual language models to political science for cross-lingual text analysis without translation.

Using a new dataset of populist rhetoric labeled in 25 languages, he found that fine-tuning a multilingual model on just one language matches the performance of models trained on all 25 languages, eliminating the need for translation or language-specific training data. He has validated this approach by measuring populism in politicians' Facebook posts worldwide through out-of-sample testing, expert comparisons, and an analysis of campaign rhetoric in Switzerland.

About the Speaker

Christopher Lucas is an associate professor in the Department of Political Science and a faculty affiliate with the Division of Computational & Data Sciences at Washington University in St. Louis. He studies methodology, political communication, and the media. His work appears in Science, PNAS, APSR, AJPS, and JOP, among other venues, and has received the Gosnell Prize for Excellence in Political Methodology twice. Ongoing work is supported by multiple grants from the National Science Foundation.

Contact

Contact Cheryl Patalano with questions about the event.