Sexual Abuse in the Catholic Church: Annotation and Machine Learning of Hidden Patterns in Abuse Reports

This paper presents a computational-theological analysis of sexual abuse reports published by Catholic dioceses in German-speaking regions, with a focus on the largely overlooked experiences of women survivors. We compiled a corpus of five representative reports and developed a hierarchical annotation scheme to capture discursive patterns related to victims, perpetrators, institutional responses, and descriptions of abuse. Quantitative analyses reveal a notable number of female victims and a consistent dominance of institutional voices, while survivor perspectives appear far less frequently. Descriptions of abuse often rely on unspecific terms, reflecting discursive tendencies that obscure the nature of violence and reinforce epistemic injustice. Using sentence-level binary classification with transformer-based models, we achieved promising results (balanced accuracy ~0.8) despite class imbalance. These models enable scalable detection of text segments concerning women. The study demonstrates how DH methods can uncover hidden narrative structures in church documents and support critical theological inquiry.