When Words Divide: Diachronic Ideological Polarization in Political Discourse on Social Media

This study investigates the long-term evolution of ideological polarization in online political discourse. While previous research has often relied on cross-sectional "snapshots" of language, the authors perform a longitudinal analysis of Reddit discussions spanning 2008 to 2023. By constructing temporally aligned, community-specific word embeddings, the researchers measure how the semantic meaning of political concepts—such as "lockdown," "radicals," and "reparations"—diverges between left-leaning and right-leaning user communities over time.

The researchers identified a clear, directional trend of increasing ideological polarization. Out of 850 analyzed political terms, approximately 14% showed a statistically significant increase in semantic divergence, while only 0.6% showed a decrease. This suggests that as time progresses, the same political concepts are being framed through increasingly distinct and incompatible vocabularies by opposing political groups. The study further notes that this divergence is particularly concentrated in substantive policy and identity-related topics, providing empirical support for the theory that online discourse is becoming more ideologically fragmented.

Understanding how political language evolves is crucial for grasping the mechanics of societal polarization. By moving beyond simple affective measures (like hostility or sentiment) to analyze semantic framing, this research offers a more nuanced view of how political groups construct different realities. The methodology provides a scalable framework for future researchers to track these linguistic shifts across other platforms and time periods, helping to quantify the "language gap" that complicates cross-partisan communication.