Center for African Studies

The Center for African Studies, The Artificial Intelligence Academic Initiative Center & The Center for the Humanities and Public Sphere Present

“Automating Serendipity: Using Advanced Machine Learning to Explore South Africa’s Truth and Reconciliation Commission Archive”

Stephen Davis, University of Kentucky

Date: Thursday, February 16, 2023

Time: 3.30 pm

Location: 404 Grinter Hall

          In 1995, South Africa’s Truth and Reconciliation Commission (TRC) began a five-year project to construct ‘as large a picture as possible’ of human rights violations that occurred during the last three decades of apartheid. The primary vehicle for constructing this picture was the collection of massive amounts of testimony provided by both victims and perpetrators of human rights violations. The architects of the TRC argued that their approach of leaving ‘no testimony behind’ provided the best guarantee of creating an inclusive, comprehensive, and unbiased archive that might serve as a foundation for future reconciliation. Although this approach had its merits, a paradox lies at the heart of the TRC’s project: collecting testimonies at this unprecedented scale inadvertently obscured patterns that the commission sought to foreground by submerging them under an archival high tide of mass acquisition. This problem was further compounded by the complexities of language in South Africa where witnesses often described similar experiences but used divergent vocabularies and idioms. These characteristics made it difficult to see the ‘bigger picture’ within a single frame or at scale using only conventional research methods such as keyword searches and Boolean expressions. New machine learning methods such as sentence embedding and UMAP dimension reduction however now allow for novel kinds of visibility and legibility based on semantic proximity rather than exact matches of keywords or combinations of keywords. This layered technique allows researchers to search for sensory experience, emotional expression, and abstract concepts. These capabilities also open the door for a new creative research practice that we term ‘query poetics.’

This presentation will describe the embedding process in brief, provide a detailed evaluation of the utility of sentence embedding for conducting research in the TRC archive and conclude with informed speculation about the broader applicability of our methods for other corpora. Stephen Davis is an Associate Professor in the History Department at the University of Kentucky. He is also the Principal Investigator of the Bitter Aloe Project, an interdisciplinary research group that applies advanced machine learning methods to materials produced by the Truth and Reconciliation Commission (South Africa). Lastly, he is author of The ANC’s War Against Apartheid: Umkhonto we Sizwe and the Liberation of South Africa (Indiana University Press, 2018) as well as several articles on war, politics and biography in southern Africa.