Niyati Bafna, Emily Chang, Nathaniel Romney Robinson, David R. Mortensen, Kenton Murray, David Yarowsky, Hale Sirin. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Mete Ismayilzada, Defne Circi, Jonne Sälevä, Hale Sirin, Abdullatif Köksal, Bhuwan Dhingra, Antoine Bosselut, Duygu Ataman, Lonneke Van Der Plas. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

In the United States, foreign nationals who fear persecution in their home country can apply for asylum under the Refugee Act of 1980. Over the past decade, legal scholarship has uncovered significant disparities in asylum adjudication by judge, by region of the United States in which the application is filed, and by the applicant's nationality. These disparities raise concerns about whether applicants are receiving equal treatment under the law. Using machine learning to predict judges' decisions, we document another concern that may violate our notions of justice: we are able to predict the final outcome of a case with 80% accuracy at the time the case opens using only information on the identity of the judge handling the case and the applicant's nationality. Moreover, there is significant variation in the degree of predictability of judges at the time the case is assigned to a judge. We show that highly predictable judges tend to hold fewer hearing sessions before making their decision, which raises the possibility that early predictability is due to judges deciding based on snap or predetermined judgments rather than taking into account the specifics of each case. Early prediction of a case with 80% accuracy could assist asylum seekers in their applications.