Rhetorical Structure Embeddings: Measuring Rhetorical Organization in Political Texts
Abstract: Rhetorical structures are politically consequential but hard to measure at scale. I introduce Rhetorical Structure Embeddings, a representation of textual organization derived from internal layer-wise transformations in open-source BERT models. With a simple pipeline, the approach detects recurring rhetorical structures in corpora, which is shown to outperform conventional approaches in a register-separation validation task. Applying the method to analyze U.S. Democratic and Republican national party platforms from 1980 to 2024, I find six distinctive rhetorical structures and systematic partisan differences in their usage. I place these styles on a spectrum from descriptive written language to engaging oral language using rhetorical structure embeddings, and find episodic spikes in the Democratic party’s use of a strongly speech-like style. The introduced framework supports descriptive typology discovery and enables downstream analysis that uses identified rhetorical structures as variables or directly applies the embeddings for cross-domain analysis.
Lonely Resistance: Differential Responses to Assimilationist Language Policy across Ethnic Minority Subgroups
Abstract: The long-term consequences of language policies on identity and inter-ethnic group relations have attracted significant scholarly attention, yet their immediate impacts remain less examined. This article enhances current understanding by studying the immediate behavioral responses to the differential subnational enforcement of language unification policies among subgroups of ethnic minorities. Focusing on China’s explicit announcement of its assimilationist language policy in Inner Mongolia in 2020, this research uses difference-in-differences models to analyze the effects. With two prefecture-level panel datasets, the study uncovers varied responses among ethnic minorities: in the wake of the policy announcement, ethnic Mongols—the directly targeted minority group - showed an increase in protest participation, whereas both Mongols and other ethnic minorities with distinct languages exhibited a decrease in formal political participation. These findings show the crucial role of language recognition, demonstrating that even partial derecognition can incite contentious movements. However, the swift suppression of the protests indicates that only the reduction in formal participation may last, suggesting that the often observed long-term marginalizing effects of language policies might emerge earlier than previously thought. Moreover, this study emphasize the need to study distinctive subgroups under the umbrella term “ethnic minorities” by revealing the heterogeneity in political behaviors among these groups.
© 2025 Yinxuan Wang. All rights reserved.