Biased Language Detection in Court Decisions

Abstract

AbstractThe Portuguese Commission for Citizenship and Gender Equality advocates that equality between men and women is a fundamental principle of the Portuguese Constitution. While court decisions should reflect this principle, a preliminary analysis in cases of gender violence reveals that this is not always the case. Based on the extensive literature on subjectivity, modality and bias in Linguistics and in tandem with AI and Natural Language Processing (NLP) techniques, the research proposed in this paper aims to study the linguistic formulations that convey bias in court decisions. The goal is to develop a linguistic model and, subsequently, a tool to automatically detect gender bias in this text genre. A corpus of a set of legal sentences on gender violence has been extracted from the public access database of the Portuguese Ministry of Justice (IGFEJ), which can be subject to a manual annotation process according to a typology of biased categories and structures. By exploiting the corpus in a supervised machine learning approach while following the most recent advances in NLP, we aim to deliver the aforementioned tool for biased language detection.KeywordsBiased languageCourt decisionsDiscourse analysisNatural Language Processing