Language and Stereotypes in Capital Trials of Women

Cornell University, Ithaca NY

Investigators

Abstract

This project sheds light on the understudied phenomenon of gender bias in the cases of women on death row. The team’s preliminary analysis indicates that legal actors frequently invoke biased tropes in the criminal trials of women sentenced to death, emphasizing women’s perceived sexual deviance, deficient mothering, and emotional manipulation. This project expands these findings and trains computational linguistic tools to identify gender-biased language in trial transcripts on a large scale. Led by national experts in gender, the death penalty, linguistics, and information science, this project ascertains the extent to which prosecutors and defense attorneys invoke gender stereotypes in the cases of women who are on trial for their lives. While this research focuses on women sentenced to death, project outcomes contribute to ongoing research about how gender stereotypes and bias affect the experiences of women throughout the criminal justice system. This project uses innovative, interdisciplinary qualitative and computational methods to explore and identify gender-biased language deployed in women’s capital trials. First, a team of researchers qualitatively codes the dataset, which consists of the trial transcripts of women on death row. The qualitative coding is informed by Critical Discourse Analysis, a methodological model that helps reveal the ways that gender stereotypes and biased discourses are both produced and influenced by specific language use. Additionally, the project draws from various disciplinary perspectives to describe the multiple forms of bias experienced by women of color. After qualitative data coding for gender-biased language, the team uses an iterative process to train and apply computational tools from computational linguistics to automate detection of gender-biased discourse invoked by courtroom actors; the model-detected phrases are confirmed by human experts and compared qualitatively across cases. This project uses a range of technologies to compare such discourse between womens’ and mens’ court transcripts, from established methods such as topic models and word embeddings, to emerging pretrained large language models (e.g. BERT, T5, GPT-Neo). By employing a range of analytical tools, this project produces the first analysis of the extent to which gender-biased discourse permeates capital trials. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.