Greenberg Research Fellow Christopher J. Anderson presents on landscapes of Holocaust rescue

On October 24, 2024, Christopher J. Anderson (PhD candidate in Geographic Information Science, Texas State University), the 2024-2025 Margee and Douglas Greenberg Research Fellow at the USC Dornsife Center for Advanced Genocide Research, presented on the research he conducted about Holocaust rescue during his monthlong residency at the Center. In his research, he focuses on the Plateau Vivarais-Lignon in south-central France, which was a culturally and statistically significant place of rescue during the Holocaust.

Anderson opened his lecture with an evocative quote from Marcel Proust, arguing that while he uses a variety of digital tools and methods in his research, it is the lived experiences of the participants in rescue – and their memories of places, people, and landscapes – that are most important to him in his research. He then outlined his central research questions, starting first with the meaning of landscape for participants in rescue, the ways that the landscape influenced – or was influenced by – social networks in the region, and the spatial distinctiveness of the Plateau. He is interested in the emotional and semantic landscapes of the testimonies, memoirs, and diaries he uses as sources in his research, in addition to the physical landscape the survivors describe.

Acknowledging the diversity of background of the audience, since the talk was cosponsored by the USC Mellon Humanities in a Digital World Program, USC Francophone Research & Resource Center, and the USC Spatial Sciences Institute, Anderson defined a few terms central to his study. He argued that he is expanding Yad Vashem’s definition of “Holocaust rescue” to include Jews alongside non-Jews as potential rescuers. Anderson then situated and historicized the Plateau itself and the culture of rescue there during the Holocaust.

After describing how his corpus of testimonies tripled during his residency, he discussed the extensive work involved in cleaning up and preparing the transcripts he relies on for his research. He spent the rest of his lecture illustrating and analyzing the variety of methods and digital tools he is using in his research so far, including geoparsing, rules-based natural language processing, topic modeling, book-level natural language processing, and semantic analysis that he is conducting to discern sentiments in the testimonies.

As he went through his explanation and illustration of each tool, he discussed the promise, limitations, and challenges involved in each one. One example is that in geoparsing, when identifying locations, a result is deemed accurate as long as it is within 161 km/100 miles of the actual location. Anderson argued that when it comes to the Holocaust and to the sources he works with, this standard of accuracy is not adequate, especially when even just a single meter could be the difference between life and death during the Holocaust. Thus, Anderson repeatedly retrained his models and through this, improved its accuracy in correctly identifying locations within 5 km/3 miles from 38% to 85%. The types of locations he’s using geoparsing to identify in the testimony transcripts include villages, children’s homes, schools, Jewish and non-Jewish organizations, and natural features linked to rescue on the plateau.

Anderson argued that tools like the ones he is using, despite their limitations, can allow researchers to read texts without missing the elements they are interested in studying. Awareness of the tools’ limitations is the key to being able to use them effectively, he asserted. Furthermore, Anderson emphasized that these digital tools do not give the answer to the questions he is posing. Rather, researchers like him still have to interpret the data the tools provide.

He showed many exciting examples of the what these tools can reveal, from the use of book-level natural language processing to highlight historical actors and their agency (what do the verbs reveal about what they were doing) as well as their subjugation (what was being done to them) to aspect-based sentiment analysis, which scores the positivity, negativity, and subjectivity of statements within transcripts. Anderson explained how these could scale up to a larger source base, providing new insights to the field.

Anderson then detailed what is next for his research. He plans to complete emotion classification for the transcripts and other sources he is working with, testing and comparing results of large language models powered by AI with traditional natural language processing. He also plans to train ArcGIS’s Deep Learning Toolbox to support his geographic analysis. Anderson plans to do various types of visualization and analysis to identify the semantic and emotional landscapes of the survivors and rescuers’ experiences on the Plateau. He is also analyzing historic aerial photographs of the plateau and plans to compare and combine them to/with the sentiment-based analysis as a way of exploring the social significance of landscapes.

Anderson finished his lecture by highlighting what he expects the result of his study to be, including how he hopes it will represent participants’ personal experiences of rescue at their own ‘lived’ scale, as well how it will contribute not only to geography, but to Holocaust Studies as well. He hopes this study can provide a methodology to approach a larger corpus of testimonies within both fields. Anderson then fielded a variety of questions, including the relationship between rescue and resistance, whether the tools he uses can help correctly identify locations that are inaccurate in existing indexes, how he located the testimonies he uses in his research, and his thoughts on using AI to give voice back to those who were murdered.

Read more about Christopher J. Anderson here.

Read an interview with Christopher J. Anderson here.