Digital technologies to analyze eyewitness accounts of mass violence
Photography: Elodie Burrillon | http://hucopix.com
In 2017, Susan Hogervorst (Open University of the Netherlands) received a grant and in-kind expertise from the eScience Center for her project EviDENce. Much of our historical knowledge is based on oral or written accounts of eyewitnesses, particularly in cases of war and mass violence, when regular ways of documentation and record keeping are often absent. EviDENce explores new ways of analyzing and contextualizing historical sources by applying event modelling and semantic web technologies.
A meta-perspective on history
“Historical culture refers to all the ways individuals and societies deal and engage with the past. I wrote my dissertation on memory cultures of concentration camp Ravensbrück in Europe, 1945–2010. I studied (mostly unofficial, private) archives of associations of Ravensbrück survivors from different countries, to examine what story about that camp they had developed and made public after 1945, and how this story related to other representations about Ravensbruck — of historians, artists, journalists, politicians, and so on. In a sense, it was about reconstructing the process of production, circulation and competition of knowledge about the past in this case Ravensbrück, that was either derived from personal experience, or from other sources. I continued studying historical culture, for example for a book I co-authored about the Rotterdam bombing raid in public memory, which was published in 2015 at the commemoration of the bombardment, then 75 years ago.”
“Many people are interested in history. And especially in case of contemporary history, many people have historical knowledge based on personal experience, family stories, or otherwise. I find it important to take on a meta-perspective, especially regarding contemporary history, in which much has not yet been crystalized, and let people look differently to what they (think they) already know.”
The cultural function of eyewitnesses
“During the last couple of years, I got more and more interested in the eyewitness, both as a source of historical knowledge, and as a cultural figure. Since the 1960s, Holocaust survivors emerged as witnesses and bearers of memory. They appeared in the media, in educational settings, and were important in political debates. Since the 1980s, public memory of WW2 has been characterized by a sense of loss, as well as urgency, regarding the upcoming disappearance of the eyewitness generations. It is important to reflect on the cultural functions of eyewitnesses, and the meanings that different audiences ascribe to them. What will be lost, exactly, and for whom? What kind of knowledge about the past do eyewitnesses produce? How did this change over time and under which conditions? And how have eyewitnesses of important historical events been perceived by the public, and by historians?”
Turning to digital humanities
“In a current project at Erasmus University Rotterdam, I examine the use of digital collections of video interviews with eyewitnesses of war, in museums, in education, and online. I recently turned to digital humanities to address these kinds of questions in a different way. In a CLARIAH Research Pilot project, I automatically investigate and compare the changing content of oral history interviews with WW2 eyewitnesses, newspaper articles with WW2 eyewitness interviews, and radio and TV programs with WW2 eyewitnesses since 1945.”
A new perspective on the fundamentals of historical research
“ Now, in our collaboration with the eScience Center and the KNAW Humanities Cluster that recently started, I have the chance to expand these questions in multiple ways: regarding topics, research period, as well as methodology. We will try to examine the representation of war and violence in Dutch ego-documents over a period of no less than 5 centuries. Ego-documents is the umbrella term for all writings about the self, like diaries, letters, autobiographies, and memoirs. We use digitized eco-documents from both Netwerk Oorlogsbronnen as well as from the Meertens Institute Nederlab infrastructure.”
“Apart from discovering new things, and working together with colleagues from different disciplines, for me the most exciting aspect of digital humanities research is to contribute to developing (and sharing) a new perspective on the fundamentals of historical research.”
Collaborating with eScience Research Engineers
“By collaborating with eScience Research Engineers we want to develop a method that automatically extracts concepts and context information from large historical text corpora. Concepts like “oppression” or “terror”, that describe how individuals experienced violent events. Currently, historical researchers find their sources by searching for specific keywords, or by filtering on document types or time periods. The software tools they use are mainly relying on literally recognizing (sequences of) words. But in many of these documents the concepts they are looking for are described using different wording, or are only implicitly mentioned in the text. We believe that enriching the documents with descriptions such as “violent” or “first-hand experience” would enable researchers to more efficiently find relevant sources, and process more material.”
Combining semantic web technology and machine learning
“eScience Research Engineer Martine de Vos has expertise in semantic web technologies. The semantic web (the web of linked data) contains lots of shared vocabularies with knowledge on persons, locations and events. Labeling the text in historical documents with concepts from these shared vocabularies — a process called annotation — adds context and meaning to it. This would help us to collect and analyze information from historical documents, automatically and in a meaningful way. Even more, annotation with vocabulary concepts enables linking of these documents with datasets elsewhere on the semantic web that are on the same concepts.”
“eSience Research Engineer Wouter Kouw is an expert in machine learning — teaching computer systems to perform tasks. Using annotated historical documents as examples, we should be able to train a system that automatically detects persons, locations and events in new documents, that have not been annotated. We could use additional information from the semantic web to improve the system’s performance. For example, since we are specifically focusing on violent events, we could cross-reference the concept ‘raid’ with the WO2 thesaurus to determine that this is “an act by a controlling force on the populace” and therefore constitutes a violent event.”
Three years from now
“Three years from now we hope we can make empirically sustained statements about the ways past eyewitnesses have described violent events, and how this has developed over time. On a methodological level, we hope to have a thorough use case about how event modeling can help detecting relevant indicators for historical phenomena or concepts in text corpora that can hardly be translated into search terms. If this succeeds, we might have a way to analyze large text corpora in a way that is meaningful to me and other humanities scholars.”