"Borrowed Language": Using Word Embedding Methodologies to Detect Amicus Influence in United States Supreme Court Rulings
The First Amendment of the Constitution of the United States establishes a formal separation between church and state, meaning that churches (regardless of religious affiliation) exist as institutions independent from the institution of the state. While this institutional separation has been formally upheld throughout US history, a tangible relationship between church and state has persisted (as it would for any non-state institution and the state). One way in which the US judicial system invites interaction from non-state institutions is through the amicus curiae brief (hereafter, amicus brief). Amicus briefs offer non-state institutions the opportunity to weigh in on court proceedings, offering input from more diverse positionalities than may be represented on the bench. Religious organizations leverage this opportunity to great effect, utilizing briefs as opportunities to challenge the secular law of the state with the religious law they primarily abide by. While the popularity of amicus briefs as a potentially influential tactic of religious groups is well-documented in the literature, the true efficacy of briefs over the judicial process is more difficult to measure. The existing literature operationalizes amicus brief influence as a measure of linguistic similarity between brief and ruling, most commonly by using cosine similarities generated from word, document, and sentence embeddings. While some studies have explored influence depending on political ideology of amicus, the extent of the use of pathos in briefs, and brief quality, none have explicitly examined the influence of religiously-affiliated briefs. The stacking of the Supreme Court with conservative Christian justices, along with recent Supreme Court rulings that aligned with the interests of the Christian right -- such Dobbs v. Jackson in 2022 -- make this oversight particularly pertinent. The relevance of this topic to current events is also why I have chosen to limit the scope of “religious affiliation” to only briefs with Christian-affiliate amici: while other religious traditions thrive in the United States, the Christian tradition has embedded itself in politics in a way that is so efficacious that it borders on hegemony. Hegemonic control over the law by such groups has profound implications for the wellbeing of all citizens, regardless of religious affiliation, if the interests of the few and powerful dictate access to services and opportunities for the many. In this project, I propose an exploration of the extent of the influence of briefs penned by religious, compared to secular, amici on U.S. Supreme Court rulings. In the project, linguistic similarity is operationalized as pairs of sentences from amicus briefs and subsequent Supreme Court rulings that meet certain thresholds for cosine similarity (measuring semantic similarity) and Jaccard similarity (measuring lexical similarity). I will collect a corpus of Christian and non-Christian affiliated amicus briefs and their corresponding rulings (estimated n= 10,500 documents), fine-tune an Alibaba-NLP embedding model on legal text, and generate sentence level embeddings for the entire corpus.
The data in this repository comes from a pilot version of the project that I ran on a small dataset consisting of 95 Christian-affiliated amicus briefs and 39 corresponding rulings related to abortion. Due to the small data size and the limited scope, the findings projected in these figures are not likely to be indicative of the findings I will see when I run the project on my full dataset. Nevertheless, figures like these are the kind that I anticipate I will produce when I conduct the full project. Becuase the data was collected from a proprietary database, I cannot share the data in its entirety here. However, the following search criteria can be employed in Proquest's Supreme Court Insight to manually download and reproduce the dataset that I utilized here.