Text Mining Judicial Dissent: A Computational Analysis of the California Supreme Court

Runlong Xu
MASDS, 2025
WU, YINGNIAN
This thesis investigates how dissenting justices of the California Supreme Court articulate disagreement, and how the nature of that dissent has shifted across two major institutional eras. Through a computational examination of full-text judicial opinions from the Third Series (1969–1991) and Fourth Series (1991–2016), sourced from the Harvard Caselaw Access Project, the study analyzes changes in the rhetorical and emotional architecture of dissent. Drawing on contemporary Natural Language Processing techniques—including TF-IDF for lexical salience, NRC sentiment scoring for emotional profiling, and Latent Dirichlet Allocation for thematic discovery—the project develops a quantitative account of the minority voice within the court. These tools allow for systematic identification of linguistic trends that traditional doctrinal analysis often overlooks, revealing how dissenters construct authority, signal disagreement, and frame legal conflict. The empirical findings suggest a marked rhetorical pivot between the two eras. Dissents in the Cal.3d period are characterized by vocabulary rooted in procedural review and a comparatively optimistic emotional tone. By contrast, dissenting opinions in the Cal.4th period move toward a more statutory and governance-oriented register, accompanied by stronger expressions of negative affect—particularly fear and disgust. These shifts align with broader historical interpretations of the court’s transition from an era of expansive judicial reasoning to one more constrained by statutory interpretation and institutional caution. By offering a reproducible framework for measuring linguistic and emotional changes in judicial writing, this research contributes to the growing literature on computational legal studies. It demonstrates how large-scale text analysis can clarify doctrinal transformations, illuminate judicial identity, and uncover subtle ideological currents within appellate courts. Future work may incorporate transformer-based models for contextual sentiment, explore dissent–majority divergence more explicitly, or extend the methodology to comparative state court systems.
2025