Confronting gender bias in heritage catalogues: A natural language processing approach to revisiting descriptive metadata

As the volume of digitised heritage collections increases, and as the online discoverability of heritage data grows, there is a risk that the historical perspectives within cultural heritage collections, and those documented within their catalogues, will amplify and reinforce gender bias. Stereotyping, discrimination, and related consequences of biased heritage data and systems have been well-documented by scholars in the Gallery, Library, Archives, and Museum (GLAM) sector. While debiasing research in the fields of natural language processing, machine learning, and artificial intelligence aims to remove bias from technology systems, currently these efforts have only managed to reduce or hide bias. Moreover, approaches that reduce or remove bias in non-heritage information sources, such as social media platforms and Wikipedia, often do not suit heritage institutions. Bringing together debates from the cultural heritage and gender studies communities, and aligning these with recent work from natural language processing, this chapter provides an overview of proposed approaches to gender bias and their critiques when situated in the cultural heritage space. Drawing on a case study utilising archival catalogue data, this chapter describes challenges to addressing gender biased language in heritage contexts and proposes a path forward.