Interpretative Variation: How Interpretations Diverge in the U.S. Congress

• In Event: 20268 - Culture and Computational Social Science (Co-sponsored by Section on Mathematical Sociology)

Abstract

This paper advances the study of cultural variation by theorizing about interpretative variation and introducing a methodological framework to measure it. We distinguish between two types: unstructured heterogeneity, marked by ambiguity and individual uncertainty, and structured divergence, where meanings cluster along social lines, reinforcing group boundaries.

To measure these concepts, we use contextualized word embeddings from a RoBERTa model. Unstructured variation is quantified using pairwise cosine distances to reflect general heterogeneity, while structured divergence is measured with silhouette scores to capture the clustering of interpretations along group lines. This framework systematically analyzes how meanings evolve and polarize across groups over time. We ensure robustness through sensitivity analyses, including alternative model specifications and external validations.

Applying this framework to U.S. Congressional speeches from 1951 to 2016, we observe an accelerating trend of interpretative divergence, especially from the mid-1980s onward. This divergence is driven by a broadening set of politically contested words, rather than a few hyper-polarized terms, highlighting an expanding partisan divide. Additionally, we identify a regime shift in the relationship between unstructured and structured interpretative variation: earlier divergence was linked to interpretative heterogeneity, but this connection has weakened over time. This suggests that contemporary partisan divides are increasingly shaped by solidified group-based meanings rather than cultural ambiguity.

Authors

• Miriam Hurtado Bodell, Stanford University

• Amir Goldberg, Stanford University