New Dataset Provides Multimodal Analysis of Russian Government Speeches Across Decades

A new computational linguistics research project has released a structured dataset documenting Russian government speeches spanning multiple decades, filling a significant gap in available resources for studying authoritarian political communication.

Dataset Scope and Structure#

The collection gathers official addresses from senior figures in the Kremlin and the Russian Ministry of Foreign Affairs, presenting each speech in dual-language format alongside visual materials. For every speech included, researchers provide both Russian and English texts, associated photographs and image captions where applicable, and standardized metadata recording speaker identity, delivery date, geographic context, and official government classifications. A system of unique identifiers enables researchers to track connections between images and their corresponding speeches, as well as to align the Russian and English versions of identical communications.

Enhanced Annotations for Research#

Beyond raw content, the dataset incorporates validated thematic tags for both the speeches themselves and the images within them. These annotations were generated using transformer-based multimodal topic modeling and then refined by subject matter experts specializing in Russian politics. This dual human-machine approach aims to ensure accuracy and consistency across the entire corpus.

Research Applications#

The linked, multilingual structure supports investigation across multiple dimensions—temporal trends in messaging, geographic patterns in communication emphasis, visual-textual alignment in political narratives, and comparative analysis between domestic and foreign policy rhetoric. The researchers position the resource as particularly valuable for social scientists examining how authoritarian regimes communicate internally and internationally, and as a testing ground for evaluating large language models when applied to politically sensitive content in non-English contexts.